The opinion of the Court was delivered by
PORITZ, C.J.*262 TABLE OF CONTENTS
I. The Origin and History of Proportionality Review.....266
A. From Furman to Pulley........................266
B. Proportionality Review in Other States...........268
C. Proportionality Review in New Jersey............274
II. The Constitutionality of N.J.S.A. 2C:ll~3e as Amended........................................277
A. Pending Appeals...............................277
B. The Supreme Court’s Appellate Review Function ........................................279
C. Application of Proportionality Review.............285
1. The Scope of the Statistical Universe of Comparison Cases........................287
2. Individual Proportionality Review............291
a. Frequency Analysis.....................291
i. Salient-Factors Test...............292
ii. Numerical-Preponderance-of-Aggravating-and-Mitigating-Factors Test....................294
iii. Index-of-Outcomes Test...........295
b. Precedent-Seeking Review..............296
3. Systemic Proportionality Review and Possible Racial Disparity in the Imposition of the Death Penalty......................298
a. Race as a Predictor of Outcome...........299
b. Review of the Statistical Models..........302
4. Proportionality Review and Its Status as a Separate Proceeding in Death Penalty Appeals.................................316
III. Application of the Methods of Individual Proportionality Review to Loftin.........................317
A. Facts.........................................317
B. Focus of Review...............................321
1. The Universe of Cases......................323
2. Method of Classifying Cases.................323
C. Comparison of Cases...........................324
1. Comparison Group..........................325
2. Frequency Analysis.........................327
a. Salient-Factors Test....................328
b. Index-of-Outeomes Test.................330
*263c. Frequency-Analysis Conclusion..........335
3. Precedent-Seeking Review..................335
a. Assessment of Defendant’s
Culpability...........................336
i. Moral Blameworthiness............336
ii. Degree of Victimization............338
iii. Character of Defendant............338
b. Comparison of Defendant’s Case to the B Cases............'..............339
D. Other Arguments..............................345
IV. Conclusion........................................345
Appendix A........................................346
Appendix B........................................347
Appendix C........................................348
Comparison Case Summaries
I. Prior Murderers with Two Additional Aggravating Factors or Particular Violence/Terror: B(l)...........................348
A) George Booker (1 and 2)........348
B) John Fauntenberry............350
C) Richard Feaster (2)............351
D) James Koedatich (1 A)..........353
E) James Koedatich (IB)..........354
II. Prior Murderers with One Additional Aggravating Factor or Particular Violence/Terror: B(2)...........................355
A) Marko Bey (2B)...............355
B) Richard Biegenwald (1A) .......356
C) Richard Biegenwald (IB) .......358
D) Richard Biegenwald (1C) .......358
E) Bryan Coyle (1A)..............358
F) Bryan Coyle (IB)..............359
G) Samuel Erazo (1 A).............359
H) Samuel Erazo (IB).............360
I) William Godette...............361
J) Frank Pennington.............362
K) Frank Pennington (IB).........364
L) Braynard Purnell (1 A)..........364
M) Braynard Purnell (IB)..........365
*264N) Thomas Ramseur..............365
O) Carlos Vasquez................366
III. Prior Murderers with No Other Aggravating Circumstances or Particular Violence/Terror: B(3)...........................367
A) Richard Biegenwald (2).........367
B) Jihad Muhammed..............368
C) Alberto Nieves................370
D) Thomas Williams..............371
In State v. Loftin, 146 N.J. 295, 680 A.2d 677 (1996) (Loftin I), we affirmed defendant Donald Loftin’s conviction and sentence of death for the murder of Gary Marsh. We also acknowledged defendant’s request for proportionality review of his death sentence pursuant to N.J.S.A. 2C:ll~3e. Id. at 397, 680 A.2d 677. This appeal requires us first to consider the constitutionality of an amendment to N.J.S.A 2C:ll-3e that limits proportionality review to a specific group of similar cases in which a jury has sentenced the defendant to death and, then, to conduct defendant’s review.
Proportionality review, although statutory in origin, is carried out by this Court in the exercise of its general authority as an appellate tribunal, N.J. Const, art. VI, § 2, ¶2, and its specific exclusive jurisdiction over capital causes, id. at § 5, ¶ 1(c). This authority is central to our primary function as the Court of last resort in the state judicial system and carries with it the power to determine the scope and content of appellate review. The integrity of appellate review by the Court is critical to the judiciary as an independent and coequal branch of government, and to the separation of powers among the executive, legislative and judicial branches. In exercising its authority in this case, the Court must ultimately decide whether the limitation on the proportionality review universe imposed by the Legislature precludes meaningful appellate review.
Prior to this case, we established the size of the proportionality review universe to include both death-eligible defendants and *265defendants who proceed to a penalty trial. State v. Marshall, 130 N.J. 109, 134, 137, 613 A.2d 1059 (1992) (Marshall II), cert. denied, 507 U.S. 929, 113 S.Ct. 1306, 122 L.Ed.2d 694 (1993). We anticipated that such broad categories would provide the most useful information about how decisions are made in the capital sentencing system by prosecutors and by juries. Id. at 132-37, 613 A.2d 1059. We also decided to conduct our review in two parts: a statistical comparison we call “frequency review,” and a descriptive analytic comparison of like cases described as “precedent-seeking review.” Id. at 152-59, 613 A.2d 1059. Always, we have sought a principled, careful approach “to ensure that the death penalty is being administered in a rational, non-arbitrary, and evenhanded manner, fairly and with reasonable consistency.” Id. at 131, 613 A.2d 1059.
Despite these efforts, many questions have been raised by our early eases, the parties, the Special Master appointed to consider Loftin’s allegations of racial disparity, and the Administrative Office of the Courts (AOC) about our present system of proportionality review. These questions cover a broad range of factual issues that must be resolved before we can rule on whether the statutory limitation unduly restricts appellate review by this Court. To this end, we are remanding these issues to a Special Master, appointed to hear and take testimony and to report to the Court. His general charge is to examine the proportionality review methodology used by the Court since Marshall II was decided over six years ago, and to test the assumptions on which the current system is based. Through a remand, a record can be developed that will enable us to consider the effect of the statute on our review function, including our ability to address any future claims that New Jersey’s system of capital punishment operates in an invidiously discriminatory manner.
We are cognizant of the Legislature’s clearly expressed intent to limit the boundaries of proportionality review and would not lightly reject its views. Our consideration of those boundaries must, however, await the findings and recommendations of the *266Special Master. Until we have had the benefit of his report, due pursuant to our Order on May 14,1999, we will continue, with one exception described below, see infra at 294-295, 724 A.2d at 149-150, to carry out proportionality review as before. Based on our proportionality review herein, we hold that defendant has not shown his death sentence to be disproportionate to the penalty imposed in similar cases.
I
The Origin and History of Proportionality Review
In Marshall II, we observed that “[t]he best way to understand the concept of proportionality review is to understand its origin.” 130 N.J. at 124, 613 A.2d 1059. That observation remains true today. It is helpful, also, to consider the history of proportionality review both in this state and in our sister states to gain perspective on the role of this form of review in New Jersey’s death penalty scheme.
A. From Furman to Pulley
Proportionality review arose in response to the United States ' Supreme Court’s decision in Furman v. Georgia, 408 U.S. 238, 92 S.Ct. 2726, 33 L.Ed.2d 346 (1972), wherein the Court held that a Georgia statute permitting defendants to be sentenced to death at the unfettered discretion of the judge or jury violated the Eighth Amendment prohibition on cruel and unusual punishment. Id. at 239-40, 92 S.Ct. at 2727, 33 L.Ed.2d at 350; see also U.S. Const. amend. VIII (“Excessive bail shall not be required, nor excessive fines imposed, nor cruel and unusual punishments inflicted”). Justice Stewart, in his concurring opinion, described “legal systems that permit ... [the death] penalty to be so wantonly and so freakishly imposed” as “cruel and unusual in the same way that being struck by lightening is cruel and unusual.” Furman, supra, 408 U.S. at 309-10, 92 S.Ct. at 2762-63, 33 L.Ed.2d at 390 (Stewart, J., concurring).
*267Four years later, the United States Supreme Court upheld statutes passed in response to Furman by Georgia, Texas and Florida, finding that the procedural safeguards provided by those statutes would prevent the death penalty from being imposed “capriciously or in a freakish manner.” Gregg v. Georgia, 428 U.S. 153, 195, 96 S.Ct. 2909, 2935, 49 L. Ed.2d 859, 886-87 (1976); see also Jurek v. Texas, 428 U.S. 262, 96 S.Ct. 2950, 49 L.Ed.2d 929 (1976); Proffitt v. Florida, 428 U.S. 242, 96 S.Ct. 2960, 49 L.Ed.2d 913 (1976).
The Georgia statute sustained by the Court in Gregg bifurcated capital proceedings into separate guilt-phase and penalty-phase trials, and provided that during the penalty phase the judge or jury would hear evidence of mitigating and aggravating factors. 428 U.S. at 163-64, 96 S.Ct. at 2920-21, 49 L.Ed.2d at 869-70. The defendant could be sentenced to death only if the judge or jury found that at least one of the statutory aggravating factors was present and outweighed the mitigating factors. Id. at 165-66, 96 S.Ct. at 2921-22, 49 L.Ed.2d at 870. The Georgia statute also provided for direct appeal to the state supreme court which, among other things, was to conduct a proportionality review to determine “ ‘[wjhether the sentence of death is excessive or disproportionate to the penalty imposed in similar cases, considering both the crime and the defendant’.” Id. at 166-67, 96 S.Ct. at 2922, 49 L.Ed.2d at 871 (quoting Ga.Code Ann. § 27-2537 (Supp. 1975)).
States seeking to enact constitutional death penalty statutes followed the statute upheld in Gregg like a recipe, careful to include provisions for appellate proportionality review. See infra at 268, 724 A.2d at 136. Six years later, however, in Pulley v. Harris, 465 U.S. 37, 104 S.Ct. 871, 79 L.Ed.2d 29 (1984), the United States Supreme Court held that proportionality review was not “indispensable” to a constitutionally acceptable capital punishment statute. Id. at 45, 104 S.Ct. at 876, 79 L.Ed.2d at 37.
*268B. Proportionality Review in Other States
Gregg did “not intend to suggest that only ... procedures [similar to the Georgia procedures] would be permissible under Furman or that any sentencing system constructed along ... [such] general lines would inevitably satisfy the concerns of Furman." 428 U.S. at 195, 96 S.Ct. at 2935, 49 L.Ed.2d at 887. Nonetheless, in the 1970s and early 1980s, twenty-five states enacted capital punishment statutes that required appellate proportionality review in all capital cases. See Ala.Code § 13A-5-53(b)(3) (enacted 1981); Conn. GenStat. § 53a-46b (enacted 1980); Del.Code Ann. tit. 11 § 4209(g)(2)(a) (enacted 1977); Ga. Code. Ann. § 17-10-35(c)(3) (enacted 1973); Idaho Code § 19-2827(c)(3) (enacted 1977); Ky.Rev.Stat. Ann. § 532.075(3)(c) (enacted 1976); LaCode Grim. Proc. Ann. art. 905.9 (enacted 1976); Md. Ann.Code, art. 27, § 414(e)(4) (enacted 1978); Mass. Gen. Laws Ann. eh. 279, § 71 (enacted 1982); Miss.Code Ann. § 99-19-105(3)(c) (enacted 1977); Mo. Ann. Stat. § 565.035 (enacted 1983); MontCode. Ann. § 46-18-310(l)(c) (enacted 1977); Neb. Rev.Stat. § 29-2521.03 (enacted 1978); Nev.Rev.Stat. § 177.055(2)(d) (enacted 1977); N.M. Stat. Ann. § 31-20A-4(C)(4) (enacted 1979); N.C. GenStat. § 15A-2000(d)(2) (enacted 1977); Ohio Rev.Code. Ann. § 2929.05(A) (enacted 1981); Oída. Stat. Ann. tit. 21, § 701.13(C)(3) (enacted 1976); 42 Pa. Cons.Stat. Ann. § 9711 (h)(3)(iii) (enacted 1974); S.C.Code Ann. § 16-3-25(C)(3) (enacted 1977); S.D. Codified Laws § 23A-27A-12(3) (enacted 1979); Tenn.Code Ann. § 39-13-206(c)(l)(D) (enacted 1977); Va. Code. Ann. § 17-110.1(0(2) (enacted 1977); Wash. Rev.Code Ann. § 10.95.130(2)(b) (enacted 1981); Wyo. Stat. Ann. § 6 — 2—103(d)(iii) (enacted 1982). These statutes were passed because of the then widely-held perception that the Supreme Court would not uphold state capital punishment legislation that lacked such provisions. See Leigh B. Bienen, The Proportionality Review of Capital Cases by State High Courts After Gregg: Only “The Appearance of Justice”?, 87 J.Crim. L. & Criminology 130,140 (1996).
*269After Pulley, nine states repealed their proportionality review requirements. See 1995 Conn. Legis. Serv. P.A. 95-16(West); 1994 Idaho Sess. Laws eh. 127; 1992 Md. Laws ch. 331; 1985 Nev. Stat. ch. 527 § 1; 1985 OklaSess. Laws ch. 265 (West), § 1; 1997 Pa. Legis. Serv. Act 1997-28 (West); 1992 Tenn. Pub. Acts ch. 952; 1998 Va. Acts ch. 872; 1989 Wyo. Sess. Laws ch. 171, § 2. A substantial number of jurisdictions, however, conduct proportionality review today pursuant to express statutory authority. In addition to the twenty-five states that enacted proportionality review provisions in the 1970s and early 1980s, three states passed similar statutes subsequent to Pulley, including Tennessee after having repealed its earlier statute, see N.H.Rev.Stat. Ann. § 630:5(XI) (enacted 1986); N.Y.Crim. Proc. Law § 470.30 (enacted 1995); Tenn.Code Ann. § 39-13-206(c)(l)(D) (enacted 1992). Today, twenty states, including New Jersey, conduct statutory proportionality review. In one state, Florida, the state supreme court has declared the court’s intention to conduct comparative review on its own initiative. See Sinclair v. Florida, 657 So.2d 1138, 1142 (Fla.1995).
As might be expected, the absence of any uniform requirements enforceable under the federal Constitution has led, over the years, to variation in the conduct of proportionality review. Thus, for example, the scope of the pool or “universe” of comparison cases used for proportionality review varies among the states. New York and Washington have defined broad universes encompassing some homicide cases that were not capitally prosecuted. See N.Y. Jud. Law § 211-a; N.Y. Ct. Rules § 510.18 (authorizing collection of case data for every criminal action in which defendant is indicted for first-degree murder); Wash. Rev.Code Ann. §§ 10.195.120, 10.195.130(2)(b) (authorizing collection of case data for every criminal action in which defendant is convicted of aggravated first-degree murder, regardless of whether defendant is capitally prosecuted).
Some states have limited the universe of comparison cases to those cases advancing to a penalty-phase trial. See, e.g., South *270Dakota v. Rhines, 548 N.W.2d 415, 455-56 (S.D.), cert. denied, 519 U.S. 1013, 117 S.Ct. 522, 136 L.Ed.2d 410 (1996) (finding that “ ‘[b]ecause the aim of proportionality review is to ascertain what other capital sentencing authorities have done with similar capital murder offenses, the only cases that could be deemed similar ... are those in which imposition of the death penalty was properly before the sentencing authority for determination’”) (quoting Tichnell v. Maryland, 297 Md. 432, 468 A.2d 1, 15-16 (Md.1983), cert. denied, 466 U.S. 993, 104 S.Ct. 2374, 80 L.Ed.2d 846 (1984), and citing Flamer v. Delaware, 490 A.2d 104, 139 (Del.), cert. denied, 464 U.S. 865, 104 S.Ct. 198, 78 L.Ed.2d 173 (1983), and cert. denied, 474 U.S. 865, 106 S.Ct. 185, 88 L. Ed.2d 154 (1985)); Flamer, supra, 490 A.2d at 139 (declaring “it inherently fair, logical and necessary to prevent disproportionate sentencing that this Court compare the sentence below to the facts and circum stances of cases in which a capital sentencing proceeding was actually conducted, whether the murderers have been sentenced to life imprisonment or death”); Missouri v. Bolder, 635 S.W.2d 673, 685 (Mo.1982) (finding court’s “inquiry would be unduly slanted were [the court] to compare only those cases in which the death penalty has been imposed” and determining “as similar [t]hose cases in which both death and life imprisonment were submitted to the jury”) (internal quotation marks omitted) (alteration in original), cert. denied, 459 U.S. 1137, 103 S.Ct. 770, 74 L.Ed.2d 983 (1983). Other states have defined the universe of comparison cases to include only those cases in which a death sentence was imposed. See, e.g., Sanborn v. Kentucky, 892 S.W.2d 542, 556 (Ky.1994) (considering all cases in which death penalty was imposed, as required by statute), cert. denied, 516 U.S. 854, 116 S.Ct. 154, 133 L.Ed.2d 98 (1995); Nebraska v. Palmer, 224 Neb. 282, 399 N.W.2d 706, 737 (1986) (finding universe of death-sentenced cases to be “a threshold requirement for comparative study”), cert. denied, 484 U.S. 872, 108 S.Ct. 206, 98 L.Ed.2d 157 (1987); South Carolina v. Copeland, 278 S.C. 572, 300 S.E.2d 63, 74 (S.C.1982) (relying only on death-sentenced cases because “[f]act findings of the trial court ... provide a fundamental line of demarcation” and *271because larger universe would cause court to “enter a realm of pure conjecture” and to engage in “intolerable speculation”), cert. denied, 460 U.S. 1103, 103 S.Ct. 1802, 76 L.Ed.2d 367 (1983).
Similarly, there is considerable variation among the states in respect' of the methods used both to select cases for comparison purposes and to make factual comparisons among selected eases. By way of illustration, the Tennessee Supreme Court has identified at least seventeen separate factors to be used in selecting and comparing similar cases including, among others, the means of death, the manner of death, the motivation for the killing, the absence or presence of premeditation, the defendant’s prior criminal record or prior criminal activity, the defendant’s cooperation with authorities, and the defendant’s remorse. Tennessee v. Bland, 958 S.W.2d 651, 667 (Tenn.1997), cert. denied, — U.S. -, 118 S.Ct. 1536, 140 L.Ed.2d 686 (1998). The Washington Supreme Court considers four factors: “(1) the nature of the crime, (2) the aggravating circumstances, (3) the defendant’s criminal history and (4) the defendant’s personal history.” Washington v. Brown, 132 Wash.2d 529, 940 P.2d 546, 562 (Wash.1997), cert. denied, — U.S. -, 118 S.Ct. 1192, 140 L.Ed.2d 322 (1998). In contrast, many courts have issued proportionality determinations without a particularized statement describing the comparative process. See, e.g., DeYoung v. Georgia, 268 Ga. 780, 493 S.E.2d 157, 168 (Ga.1997) (referring without discussion to appendix listing similar cases where death penalty upheld), cert. denied, — U.S. -, 118 S.Ct. 1848, 140 L.Ed.2d 1097 (1998); Sanborn, supra, 892 S.W.2d at 556-57 (incorporating by reference list of cases cited in previous decisions and referring to list of five additional cases); Davis v. Mississippi, 660 So.2d 1228, 1261-62 (Miss.1995) (referring without discussion to appendix with list of capital cases court previously affirmed), cert. denied, 517 U.S. 1192, 116 S.Ct. 1684, 134 L.Ed.2d 785 (1996); Pennsylvania v. Uderra, 550 Pa. 389, 706 A.2d 334, 342 (Pa.1998) (making passing reference to statistical data without mention of similar cases).
*272Although eleven state supreme courts have vacated death sentences as disproportionate, most have done so rarely. See, e.g., Hall v. Georgia, 241 Ga. 252, 244 S.E.2d 833, 839 (Ga.1978) (only one aggravating factor and co-defendant received life sentence); Idaho v. Pratt, 125 Idaho 546, 873 P.2d 800, 806 (Idaho 1993) (no prior criminal record); Missouri v. McIlvoy, 629 S.W.2d 333, 341 (Mo.1982) (minimal juvenile criminal record, limited education, limited intelligence, substantial alcohol problems, weak follower of co-defendant who received life sentence, and promptly turned himself in to police). The Florida Supreme Court, however, has vacated over thirty death sentences based on proportionality review, see, e.g., Williams v. Florida, 707 So.2d 683 (Fla.1998); Jones v. Florida, 705 So.2d 1364 (Fla.1998); Voorhees v. Florida, 699 So.2d 602 (Fla.1997); Curtis v. Florida, 685 So.2d 1234 (Fla.1996), cert. denied, — U.S. —, 117 S.Ct. 2521, 138 L.Ed.2d 1022 (1997); Sinclair, supra, and the North Carolina Supreme Court has vacated at least seven death sentences on proportionality grounds, see, e.g., North Carolina v. Benson, 323 N.C. 318, 372 S.E.2d 517 (N.C.1988); North Carolina v. Stokes, 319 N.C. 1, 352 S.E.2d 653 (N.C.1987); North Carolina v. Rogers, 316 N.C. 203, 341 S.E.2d 713 (N.C.1986). It is perhaps significant that Florida and North Carolina have relatively large death-row populations compared to other states, see Bienen, supra, 87 J.Crim. L. & Criminology at 169; more important, in those states proportionality review functions as a check against the arbitrary imposition of the death penalty.
The experience of other states is instructive, if only because it demonstrates the diverse responses to questions about the conduct of proportionality review. For example, the propriety of courts utilizing quantitative methods has been vigorously debated. Compare Washington v. Pirtle, 127 Wash.2d 628, 904 P.2d 245, 277 (Wash.1995) (noting that quantitative approach — comparing number of aggravating circumstances, victims and prior convictions in similar cases — “can point to areas of concern” and help court “to be as objective as possible”), cert. denied, 518 U.S. 1026, 116 S.Ct. 2568, 135 L.Ed.2d 1084 (1996), and Governor’s Memorandum of *273Approval of 1995 N.Y. Laws, c. 1, reprinted in N.Y. Correct L. § 650 (approving consideration of statistical evidence in conducting proportionality review to determine whether race is “having a significant impact upon the imposition of the death penalty”), with Connecticut v. Webb, 238 Conn. 389, 680 A.2d 147, 209 (Conn.1996) (rejecting New Jersey Supreme Court’s statistical methods as unworkable attempt to “quantify the unquantifiable”),1 and Bland, supra, 958 S.W.2d at 665 (criticizing New Jersey Supreme Court’s use of statistics as departure from jurisprudence of “individualized consideration”). Despite the concerns expressed by some courts about these methods, several states remain committed to a form of quantitative proportionality review to detect possible racial bias. See N.Y.Crim. Proc. Law § 470.30 (requiring proportionality review if request based on race of defendant or victim); Connecticut v. Cobb, 234 Conn. 735, 663 A.2d 948, 961-62 (Conn.1995) (recognizing statutory basis for statistical claim of racial disparity in imposition of death penalty);2 Washington v. Gentry, 125 Wash.2d 570, 888 P.2d 1105, 1154 (Wash.) (utilizing proportionality review to examine patterns in death sentencing based on race), cert. denied, 516 U.S. 843, 116 S.Ct. 131, 133 L.Ed.2d 79 (1995). Yet, statistical claims of racial bias in the administration of the death penalty present legal and methodological issues of exceptional complexity. We keep in mind the dialogue engendered by these difficult questions when considering how we might answer them ourselves.
*274We turn now to our own experience in applying proportionality review in capital cases.
C. Proportionality Review in New Jersey
When the New Jersey Legislature reintroduced the death penalty in 1982, it too substantially incorporated the procedural safeguards in the Georgia law sustained by the United States Supreme Court in Gregg, including the provision for proportionality review. See State v. Ramseur, 106 N.J. 123, 202-03, 524 A.2d 188 (1987). The New Jersey Capital Punishment Act, L. 1982, c. Ill (codified at N.J.SA 2C:ll-3), like the Georgia statute, called for a determination by this Court on “[e]very judgment of conviction which results in a sentence of death ____ whether the sentence is disproportionate to the penalty imposed in similar cases, considering both the crime and the defendant,” N.J.SA 2C:ll-3e. Later, the Senate Judiciary Committee explained that, at the time the Act was passed by the State Legislature, “it was thought that the United States Supreme Court would not uphold a capital punishment law that did not contain such a ‘proportionality review”’ provision. Senate Judiciary Committee, Statement to Senate Bill No. 950 (L. 1985, c. 178).
As enacted in 1982, the Capital Punishment Act required proportionality review, but did not describe the manner in which it was to be conducted or limit the “universe” of similar cases to be used for comparison purposes. In response to Pulley, however, the New Jersey Legislature amended N.J.SA 2C:ll-3e to abolish mandatory proportionality review and to require instead that defendants request such review by this Court. L. 1985, c. 178. Then, in 1992, the proportionality review provision of N.J.SA 2C:ll-3e was amended yet again tó define the universe of “similar” cases to be compared to defendant’s case as those “in which a sentence of death has been imposed.” L. 1992, c. 5. This amendment became effective on May 12, 1992, L. 1992, c. 5, and will, if ultimately determined to be valid, limit the scope of *275proportionality review undertaken by this Court since Marshall II.
In Ramseur, we explained that the development of “a procedure of review ... [would] be an evolving process,” requiring the advice of “criminal justice experts ... [and] experts from disciplines outside the law.” 106 N.J. at 328, 524 A.2d 188. Our view of the fundamental purpose of proportionality review would guide this process:
Proportionality review has a function entirely unique among the review proceedings in a capital proceeding. Proportionality review, in the context of a capital sentencing scheme, is not appellate review to ensure that the aggravating factors outweigh beyond a reasonable doubt all the mitigating factors, L. 1985, c. 178, or to determine if the death sentence is disproportionate to the crime in violation of the ban against cruel and unusual punishment. That death is not disproportionate in the sense of being a cruel and unusual punishment is presumed by the nature of the review. Pulley v. Harris, 465 U.S. at 43, 104 S.Ct. at 875, 79 L.Ed.2d at 36. Rather, the purpose of review here is “of a different sort____ It purports to inquire instead whether the penalty is nonetheless unacceptable in a particular case because disproportionate to the punishment imposed on others convicted of the same crime.”
[Id. at 326, 524 A.2d 188 (quoting Pulley, supra, 465 U.S. at 43, 104 S.Ct. at 875, 79 L.Ed.2d at 36).]
Therefore, when we conduct a proportionality review, we ask whether the “punishment fits the criminal,” Marshall II, supra, 130 N.J. at 129, 613 A.2d 1059, so as “to ensure that the death penalty is being administered in a rational, non-arbitrary, and evenhanded manner, fairly and with reasonable consistency,” id. at 131, 613 A.2d 1059.
Ramseur also explained that proportionality review provides a mechanism by which death sentences may be monitored, “to prevent any impermissible discrimination in imposing the death penalty.” 106 N.J. at 327, 524 A.2d 188. In Marshall II, we spoke of the “unique” commitment of the people of New Jersey “to the elimination of racial discrimination.” 130 N.J. at 207, 613 A.2d 1059. Today, as then, we believe that “[t]o countenance racial discrimination in capital sentencing would mock that tradition and our own constitutional guarantee of equal protection of the laws under New Jersey Constitution Article I, paragraph 1.” Ibid.. Comparison of like cases presents an opportunity for the *276Court to monitor whether impermissible factors are present in the capital sentencing system.
Marshall II describes in detail the creation of a database and the sorting processes we have used as a basis for comparison of similar cases. Id. at 141-45, 613 A.2d 1059. We have substantially relied on the Final Report of our first Special Master, David C. Baldus, Death Penalty Proportionality Review Project, Final Report to the New Jersey Supreme Court (Sept. 24, 1991) (Final Report). As outlined in Marshall II, supra, 130 N.J. at 152-59, 613 A.2d 1059, and followed in our subsequent proportionality review cases, we conduct a frequency analysis using statistical methods and a precedent-seeking review consisting of case-by-case analyses of comparative culpability. State v. Bey, 137 N.J. 334, 350, 645 A.2d 685 (1994) (Bey IV), cert. denied, 513 U.S. 1164, 115 S.Ct. 1131,130 L.Ed.2d 1093 (1995); State v. Martini, 139 N.J. 3, 28, 651 A.2d 949 (1994) (Martini II), cert. denied, 516 U.S. 875, 116 S.Ct. 203, 133 L.Ed.2d 137 (1995); State v. DiFrisco, 142 N.J. 148, 165-66, 662 A.2d 442 (1995) (DiFrisco III), cert. denied, 516 U.S. 1129, 116 S.Ct. 949, 133 L.Ed.2d 873 (1996). These methods compare the case at bar to other cases that have been found to have either similar fact patterns or similar levels of culpability and are used to determine whether the sentence imposed on the defendant in the case at bar is disproportionate to the sentences imposed in those similar cases.
More specifically, frequency analysis is a statistical approach that determines in three ways which cases have similar levels of culpability: (1) the salient-factors test; (2) the numerical-preponderance-of-aggravating-and-mitigating-factors test; and (3) the index-of-outcomes test. Bey IV, supra, 137 N.J. at 350-51, 645 A.2d 685. The salient-factors test “defines ‘similar cases in terms of factual comparability,’ ” while the numerieal-preponderance-of-aggravating-and-mitigating-factors test compares the raw number of those factors in each case. Marshall II, supra, 130 N.J. at 146-47, 613 A.2d 1059 (citations omitted). The index-of-outeomes test is a regression analysis that determines the culpability levels of defendants “ ‘as measured by the presence or absence in the cases *277of [statutory and non-statutory aggravating and mitigating] factors that appear to influence prosecutorial and jury decision-making.’ ” Id. at 147-48, 613 A.2d 1059 (citation omitted). Precedent-seeking review, on the other hand, compares all relevant statutory and nonstatutory aggravating and mitigating factors present in factually similar eases in order to determine defendant’s criminal culpability, or relative degree of deathworthiness. See DiFrisco III, supra, 142 N.J. at 184-85, 662 A.2d 442.
From the beginning, there has been extensive critical commentary by the Public Defender and Attorney General, along with alternative suggested approaches, on the frequency analysis methodologies accepted by the Court. We have considered that commentary and have chosen in each case to discuss the results of our frequency tests despite their acknowledged deficiencies, with the caveat that ultimately our judgments about the proportionality of death sentences are unquantiflable. Frequency review has always functioned as an adjunct to the detailed comparison of like cases that we undertake in precedent-seeking review and that, as judges, we are by training and experience best equipped to .do.3
II
The Constitutionality of N.J.S.A. 2C:ll-3e As Amended
A. Pending Appeals
In Marshall II, we determined that N.J.S.A. 2C:ll-3e, as amended, would not apply to defendant’s ease because his appeal *278was pending at the time the amendment took effect on May 12, 1992. 130 N.J. at 119, 613 A.2d 1059. Under the pre-1992 proportionality review provision of N.J.S.A. 2C:ll-3e, we held that the appropriate “universe” from which to choose similar cases included not only all cases in which the death penalty had been imposed, or had been sought, but also “clearly death eligible homicides in which the prosecutor elected not to seek the death penalty.” Id. at 137, 613 A.2d 1059. We chose the larger universe in order to consider the possibility that a jury had imposed a disproportionate sentence, id. at 133-34, 613 A.2d 1059, or that a prosecutor had misused his or her discretion, id. at 134, 613 A.2d 1059, and, in order to prevent “‘any impermissible discrimination in imposing the death penalty,’” id. at 135, 613 A.2d 1059 (quoting Ramseur, supra, 106 N.J. at 327, 524 A.2d 188).
In the three cases following Marshall II, we continued to make use of a universe of death-eligible homicides. DiFrisco III, supra, 142 N.J. at 162-63, 662 A.2d 442; Martini II, supra, 139 N.J. at 23, 651 A.2d 949; Bey IV, supra, 137 N.J. at 343-44, 645 A.2d 685. In each of these cases, the 1992 statutory amendment limiting our review to death-sentenced cases was inapplicable because the appeals in these cases were pending as of the effective date of the amendment. DiFrisco III, supra, 142 N.J. at 163, 662 A.2d 442;4 Martini II, supra, 139 N.J. at 23, 651 A.2d 949; Bey IV, supra, 137 N.J. at 343-44, 645 A.2d 685. In this case, defendant was not convicted until July 8, 1994, more than two years after the *279effective date of the 1992 statutory amendment.5 For that reason, we now consider whether we must conform our proportionality review methodology to the 1992 amendment thereby limiting proportionality review “to a comparison of similar cases in which a sentence-of death has been imposed.” N.J.S.A. 2C:ll-3e.
B. The Supreme Court’s Appellate Review Function
Comparing a death-sentenced defendant’s case to other similar cases enables us to consider whether the death penalty has been imposed arbitrarily on the defendant. Moreover, proportionality review may be the only mechanism that permits system-wide evaluation of both prosecutorial and jury decision making so as to determine whether there has been racial or other impermissible discrimination. These considerations weigh heavily when a life is at stake. It is a mark of our humanity that, no matter how heinous the crime, we focus, finally, on individual defendants, their acts and their lives. We seek to administer the most extreme penalty in a fair and consistent manner by comparing the defendant to others who have committed like crimes, thereby to decide whether there is a societal consensus that the defendant before us should be put to death. In our constitutional system, this Court is entrusted with that ultimate decision.
The source of the Supreme Court’s appellate review authority can be found in Article VI, Section 2, Paragraph 2 of the 1947 Constitution: “The Supreme Court shall exercise appellate jurisdiction in the last resort in all causes provided in this Constitution.” As defined by Article VI, Section 5, Paragraph 1, the Court hears appeals:
*280(a) In causes determined by the appellate division of the Superior Court involving a question arising under the Constitution of the United States or this State;
(b) In causes where there is a dissent in the Appellate Division of the Superior Court;
(c) In capital causes;
(d) On certification by the Supreme Court to the Superior Court and, where provided by rules of the Supreme Court, to the inferior courts; and
(e) In such causes as may be provided by law.
mm
We deríve from this grant of power, in both its discretionary and mandatory forms, our core function as this State’s highest Court— the power of appellate review.
In State v. Laws, 51 N.J. 494, 242 A.2d 333 (1968), Justice Jacobs described “the scope of this Court’s reviewing authority” under the 1947 Constitution:
That organic document purposefully modernized and greatly strengthened our judicial system. In the process it vested this Court with wide judicial power, perhaps more sweeping than that granted to any other court of last resort, and all to the end that it would be in a fail position to ensure that justice is truly and equally done.
[Id. at 500, 242 A.2d 333.]
See also Hager v. Weber, 7 N.J. 201, 205, 81 A.2d 155 (1951) (holding appellate review by Supreme Court “is a remedial procedure secured against legislative interference” by various provisions of State Constitution).
Justice Jacobs’s understanding of this Court’s broad appellate power echoes Governor Driscoll’s speech to the Committee on the Judiciary at the 1947 Constitutional Convention. Governor Driscoll had a vision of an integrated court system administered through a centralized authority and capable of “achieving the degree of uniformity [in dispensing justice] that is so highly desirable in a republic.” Alfred E. Driscoll, Address to Committee on the Judiciary, in 4 State of New Jersey Constitutional Convention of 19/J7 427, 435 (Sidney Goldmann & Herman Crystal eds., 1951) (Driscoll Address). He told the Committee on the Judiciary:
*281It is, as you know, the courts that have traditionally been the guardians of our constitutions, to whom the meanest citizen may appeal for protection against a wayward executive or a capricious legislature. Without independent courts, the whole republican system must surely fail. Our primary, our basic purpose in the drafting of a new Constitution is to secure beyond any question a strong, competent, easily functioning, but always independent, judiciary, and, therefore, in a position to curb any tendency on the part of the other two branches of government to exceed their constitutional authority.
It was Hamilton who quoted Montesquieu: “There is no liberty if the powers of judging be not separate from the legislative and executive powers.” “The complete independence of the courts of justice is peculiarly essential in a limited constitution,” Hamilton added.
[Id. at 428-29.]
Governor Driscoll’s call did not go unheeded. Students of New Jersey constitutional history, and particularly of the Judicial Article, often point to the extraordinary accomplishments of the 1947 Constitution in recasting an antiquated collection of multiple overlapping courts into a modern efficient judiciary. Leon S. Milmed, Introduction to the New Jersey Constitution of 1947 1, 12 (1954). The Report of the Committee on the Judiciary, submitted to the Convention on July 24, 1947, recommended simplifying and streamlining the court system, including centralizing administrative responsibility, merging the Courts of Law and Equity, and reducing the multiple functions of appellate court judges as well as multiple appeals at the intermediate appellate court level. See Committee on the Judiciary, Report and Proposal (Aug. 26,1947), in 2 State of New Jersey Constitutional Convention of 1947, swpra, 1180,1182-83 (Committee Report). Although the thrust of the Committee’s recommendations was administrative in nature, we must not forget that the purpose was to create an efficient vehicle in which judges could carry out their most basic function— deciding cases — and in which all New Jersey citizens would, in Governor Driscoll’s words, be “equal[ ] before the law.” Driscoll Address, supra, at 434.
Broad powers were granted to the Supreme Court at the same time that the Court’s as-of-right jurisdiction was limited. Committee Report, supra, at 1184. The Committee explained its decision:
*282There was some difference of opinion as to whether the jurisdiction of [the Supreme Cjourt should be selective and limited to important cases, including constitutional questions and capital offenses, or whether it should take appeals comprehensively, as does the present Court of Errors and Appeals____ By making the new Supreme Court’s appellate jurisdiction selective, that court is assured of an adequate opportunity to hear, consider and decide every case which comes before it.
[M at 1183-84.]
The framers of the 1947 Constitution clearly intended that the Supreme Court carry out its appellate review function by undertaking a thorough and comprehensive consideration of each case. To this end, the Court was also empowered to “exercise such original jurisdiction as may be necessary to the complete determination of any cause on review.” N.J. Const, art. VI, § 5, ¶ 3. “This grant of original jurisdiction ... [permits the Court to] review ... matters of fact as well as of law, in accordance with the historic function of an ‘appeal.’ ” Hager, supra, 7 N.J. at 211, 81 A.2d 155. How such comprehensive power was to be effectuated by the Court was left to the Court. At issue in this ease is the process by which the Court fulfills its substantive constitutional responsibility to review matters on appeal and, specifically, to review capital causes.
In State v. Laws, we held that our authority to undertake appellate review necessarily encompasses the exercise of discretion in modifying a death sentence to a life sentence “whenever the interests of justice so require.” 51 N.J. at 509-10, 242 A.2d 333. The constitutional dimension of this exercise of discretion is underscored by our exclusive appellate jurisdiction over capital causes. N.J. Const, art. VI, § 5, ¶ 1(c); see Joseph H. Rodriguez et al., Proportionality Review in New Jersey: An Indispensable Safeguard in the Capital Sentencing Process, 15 Rutgers L.J. 399, 422 (1984) (observing that framers of 1947 Constitution intended to vest Supreme Court with “the power to review fully all aspects of capital cases”). We cannot properly exercise this discretion unless we have the means to ensure that the death penalty is being administered in an evenhanded and nondiscriminatory man*283ner. Proportionality review is one way to achieve those objectives.
The Florida Supreme Court has held that its authority to conduct proportionality review rests on several provisions of its state constitution, including the court’s “mandatory, exclusive jurisdiction ... over death appeals.” Tillman v. Florida, 591 So.2d 167, 169 (Fla.1991) (citing Fla. Const, art. I, § 9). The court observed:
The obvious purpose of this special grant of jurisdiction is to ensure the uniformity of death-penalty law by preventing the disagreement over controlling points of law that may arise when the district courts of appeal are the only appellate courts with mandatory appellate jurisdiction. Thus, proportionality review is a unique and highly serious function of this Court, the purpose of which is to foster uniformity in death-penalty law.
[Ibid, (citation omitted).]
Our own jurisprudence reflects a continuing concern with proportionality and fairness in both noncapital and capital cases. See State v. Roth, 95 N.J. 334, 342, 471 A.2d 370 (1984) (noting that “ ‘the defendant’s right to appeal from his [noncapital] sentence as manifestly excessive has become firmly established in our State’ ”) (citation omitted); State v. Bess, 53 N.J. 10, 18, 247 A.2d 669 (1968) (holding that appellate court has “the power to revise a prison sentence where it is manifestly excessive, even though within statutory limits”); Laws, supra, 51 N.J. at 509-10, 242 A.2d 333 (asserting appellate power to modify death sentence to life sentence “whenever the interests of justice so require”). In Ramseur, we recognized that the interests of justice are heightened when a life is at stake. 106 N.J. at 326, 524 A.2d 188. The scope and conduct of proportionality review are therefore questions that require our most careful determination.
It is in this context that we must grapple with a statute in which a coequal branch of government has set limits on our review.6 See N.J.S.A 2C:ll-3e. By its terms, the State Constitu*284tion prohibits any one branch of government from exercising powers assigned to a coordinate branch. N.J. Const, art. Ill, ¶ 1. Nonetheless, we have long recognized that “[t]he compartmentalization of governmental powers ... has never been watertight,” In re Salaries for Probation Officers of Bergen County, 58 N.J. 422, 425, 278 A.2d 417 (1971); accord Communications Workers of America v. Florio, 130 N.J. 439, 449, 617 A.2d 223 (1992), and have applied a flexible approach to the separation of powers issues that have been brought to the Court, Knight v. Margate, 86 N.J. 374, 389, 431 A.2d 833 (1981).
More than forty-eight years ago, we held that appellate review is an unqualified and exclusive function of the judiciary “secured against legislative interference,” Hager, supra, 7 N.J. at 205, 81 A.2d 155, and we continue to adhere to that bedrock principle today. That appellate review is an exclusive power of the judiciary does not foreclose our “cooperative accommodation” of the views of the Legislature. Marie L. Garibaldi, The New Jersey Experience: Accommodating the Separation Between the *285Legislature and the Judiciary, 23 Seton Hall L.Rev. 3, 5 (1992). Where “legislative arrangements ... have not in any way interfered with this Court’s constitutional obligation,” we have accepted such arrangements in the interests “of comity and respect for other branches of government.” Passaic County Probation Officers’ Assoc. v. County of Passaic, 73 N.J. 247, 255, 374 A.2d 449 (1977). Our constitutional obligation has rarely dictated “a result different from that ordained by the Legislature.” CWA Local 1044 v. Chief Justice, 118 N.J. 495, 501, 572 A.2d 613 (1990).
When a legislative enactment impedes our ability to fulfill that obligation, however, we have “examine[d] the terms of the legislative enactment, its importance, the extent of its interference with sound judicial administration, and the significance of the • issue to the judiciary, ultimately striking a balance between the interests served by comity and those served by the administration of justice.” Ibid.; see also Garibaldi, supra, 23 Seton Hall L.Rev. at 9 (“The constitutional validity of a coordinate branch’s action turns on the legitimacy of the action’s underlying purpose and the nature and extent of its encroachment on judicial prerogatives and interests.”). In conducting this evaluation, we remain mindful of our obligation to uphold a legislative enactment absent constitutional repugnance. See Smith v. Penta, 81 N.J. 65, 74-75, 405 A.2d 350 (1979).
C. Application of Proportionality Review
To determine the validity of the Legislature’s limitation on the proportionality review universe to death-sentenced defendants, we must examine whether the limitation prevents meaningful appellate review. Ramseur, supra, 106 N.J. at 186 n. 18, 524 A.2d 188 (observing state death penalty statute’s provision for appellate review recognizes Court’s constitutional obligation to ensure “meaningful” review). The appropriate size of the universe is, however, but one of many questions that have been raised about the conduct of proportionality review and cannot be considered apart from those other questions. At the start, we characterized *286our effort to define the scope and conduct of proportionality review as an “evolving process.” Id. at 328, 524 A.2d 188. We observed that proportionality issues are “difficult and sensitive ... and hence review, reflection, and modification of the analysis ... may be required as more information is gathered.” Ibid
In the past six years, we have gained considerable experience in applying the methodologies developed by Professor Baldus and generally adopted by the Court in Marshall II, supra, 130 N.J. at 131-66, 613 A.2d 1059. See DiFrisco III, supra, 142 N.J. 148, 662 A.2d 442; Martini II, supra, 139 N.J. 3, 651 A.2d 949; Bey IV, supra, 137 N.J. 334, 645 A.2d 685; Marshall II, supra, 130 N.J. 109, 613 A.2d 1059. Our experience teaches us that the proportionality review methodologies we use are not without substantial shortcomings and, accordingly, warrant careful reconsideration. Our goal is to retain those elements of the present system that provide useful information, to refine and improve that which we retain, if appropriate, and to reject methods that have proved unhelpful. We seek a practical approach that ensures every defendant before us a rigorous and complete review of his or her sentence of death.
Our reconsideration extends to four discrete areas of concern: the size of the universe of comparison cases; particular issues in respect of individual proportionality review; questions relating to the statistical models used in both individual and systemic proportionality review; and the status of proportionality review as a separate proceeding in death penalty appeals.7 Because these issues, with one exception, cannot be resolved on the *287record before us, we are appointing a Special Master to conduct additional fact-finding and make recommendations to the Court. On our receipt of his report, we will be in a position to determine whether the statutory limitation on the proportionality review universe prevents meaningful appellate review. Until that time, we will continue to use the full universe of death-eligible cases as described below. Today we decide, based on our experience and on the record before us, that the numerical-preponderance-of-aggravating-and-mitigating-factors test should be abandoned.
We turn now to a discussion of each area of concern.
1. The Scope of the Statistical Universe of Comparison Cases
We have observed that “[t]he first step in any proportionality undertaking is to establish the ‘universe’ of cases that the Court will consider.” Marshall II, supra, 130 N.J. at 131, 613 A.2d 1059. In Marshall II, we examined three possible statistical universes from which to draw comparison cases: a universe consisting of only those cases in which a death sentence whs imposed, a universe consisting of all penalty-trial cases, and a universe consisting of all penalty-trial cases plus clearly death-eligible homicides in which the prosecutor elected not to seek the death penalty. Id. at 131-37, 613 A.2d 1059. After careful evaluation, we selected the third option as best serving the “purposes to be achieved by proportionality review.” Id. at 137, 613 A.2d 1059. We rejected a universe consisting of only death-sentenced cases as “inadequate,” id. at 133, 613 A.2d 1059, and offered in support of our determination this simple example:
On the assumption that 100 robbery-felony-murder cases are prosecuted as capital crimes, all defendants are convicted and one defendant is sentenced to death, a comparison of the death-sentenced defendant’s punishment with a punishment imposed only on other death-sentenced defendants would exclude from the proportionality-review process the ninety-nine robbery-felony-murder defendants that juries did not sentence to death. Indisputably, the determination whether that single death sentence is disproportionate can be made only by comparing it with the life sentences imposed on the ninety-nine defendants convicted of the same crime.
*288[[Id. at 133-34, 613 A.2d 1059.]
We concluded that it was “self-evident that the universe for proportionality review must, at a minimum, include all penalty-trial cases.” Id. at 134, 613 A.2d 1059.
We next considered whether the statistical universe also should encompass death-eligible homicides not prosecuted as capital crimes. Recognizing that disproportionality could originate in decisions made by prosecutors as well as decisions made by juries, we determined that our mandate to prevent arbitrariness in death sentencing should extend to review of prosecutorial decisions whether to seek the death penalty. Id. at 134-35, 613 A.2d 1059. We returned to our prior example of 100 robbery-felony-murder defendants, only one of whom was sentenced to death:
Were we to assume that the remaining ninety-nine defendants were prosecuted and convicted of non-capital murder because of prosecutorial decisions not to seek the death penalty, the disproportionality of the single defendant’s death sentence would arise not because of a disproportionate jury determination but because the prosecutorial decision to seek the death penalty was unique. That type of disproportionate death sentence could not be identified by a proportionality-review process that was limited to capital cases tried to a penalty phase; it could be identified, however, by a universe that included clearly death-eligible homicides that were not prosecuted as capital cases.
[Ibid.]
We further noted that prosecutorial decisions not to seek the death penalty were often influenced by prosecutors’ predictions of whether a jury would impose the death penalty after a penalty trial. In this sense, the prosecutorial decision may be understood as a calculation of “deathworthiness,” and this calculation, the Court concluded, should be considered in determining whether a particular death sentence was disproportionate. Id. at 135, 613 A.2d 1059.
In concluding that the proper scope of the statistical universe encompasses all death-eligible homicides, the Court in Marshall II observed that proportionality review is premised not only on the prevention of arbitrariness, but also on the prevention of impermissible discrimination in the administration of the death penalty. Ibid, (citing Ramseur, supra, 106 N.J. at 327, 524 A.2d 188). This *289purpose would be best served by including within the statistical universe those death-eligible homicides that did not advance to penalty trial. Id. at 136, 613 A.2d 1059. As in individual proportionality review, the larger universe would enable the Court to consider possible discrimination in prosecutors’ charging decisions as well as in juries’ sentencing decisions.
Our reasons for choosing the larger universe remain valid today. Yet, the reliability of the data underlying a prosecutor’s decision to seek or not to seek the death penalty continues to concern the Court. In each case, a prosecutor must decide what to do based on a complex of factors including, among other things, the strength of the State’s case, the availability and credibility of witnesses, and the resources of the prosecutor’s office. Id. at 225, 613 A.2d 1059 (Garibaldi, J., concurring in part and dissenting in part) (citing State v. Koedatich, 112 N.J. 225, 256, 548 A.2d 939 (1988)). Such a decision does not necessarily reflect a determination of deathworthiness. As a result, we are uncertain whether under our current approach the addition of death-eligible but noncapitally prosecuted homicides to the statistical universe of comparable cases is helpful to the proportionality review process. As a matter of general principle, the broadest possible statistical database should provide the most useful information; we are concerned when the additional quantum of data may be unreliable.
Even if we were fully convinced that a universe of death-eligible cases was otherwise warranted, our understanding of the practical difficulties attendant to data collection and analysis of noncapital cases requires further review of this issue. Data collection and analysis, which is performed by staff of the AOC, entail difficult and highly subjective determinations of whether a- prosecutor could have proven a case and whether a case is death-eligible. In a case that went to trial but was not capitally prosecuted, the AOC staff must make an assessment of aggravating and mitigating factors based on a review of the trial record. In a noncapital homicide, the sentencing of a defendant to a term of imprisonment is guided by statutory aggravating and mitigating factors. *290N.J.S.A 2C:44-1. Those factors, however, are different from the aggravating and mitigating factors to be applied by a penalty-phase jury in determining whether to return a verdict of death or life imprisonment. See N.J.S.A. 2C:ll-3c(4), c(5). Thus, with respect to a homicide case that was not capitally prosecuted, the AOC must make its own assessment of the applicable capital statutory aggravating and mitigating factors.
In a case in which the defendant pled guilty, the AOC must make a determination of deathworthiness without the benefit of a trial and based on a record of the offense, the offender, and the surrounding circumstances that may not be complete. We understand that the initial determination, both for noncapital trials and for pleas, is communicated to the parties and that they may object based on information available from their files. After the AOC conducts a meeting with the parties to address their objections, a final coding decision is made. Yet, it is not clear that this process can effectively deal with the potential for errors or omissions in particular case summaries. In turn, the relatively small size of the proportionality review database suggests that the cumulative effect of individual case errors could undermine the reliability of the results produced by the statistical models. We note, however, that the impact of such errors on the precedent-seeking review process, which uses the AOC case summaries, may not be substantial.
For these reasons, on remand to the Special Master, there should be additional fact-finding concerning the proper scope of the proportionality review universe. The Special Master should make an independent evaluation of the deathworthiness of a sample of cases previously classified by the AOC as either death-eligible or death-ineligible. The “provability” of the selected cases and the presence or absence of aggravating and mitigating factors should be considered, and the results compared to the data-coding decisions made by the AOC. If there is a variance between the survey results and the AOC data-coding decisions, possible causes of the variance should be identified along with recommendations *291for improved data-coding procedures. Perhaps, a questionnaire, such as that used in the State of Delaware, could be filled out by the judge in each case and used to improve both the data-collection and the data-coding process. Alternatively, if the Special Master determines that the intrinsic difficulties and ambiguities of data-coding death-eligible eases cannot be overcome, he should consider the impact of anticipated coding errors on the AOC models.
We are not aware of particular difficulties associated with the cases that proceed to capital prosecution but reserve decision on the composition of the proportionality review universe generally until we receive the Special Master’s report.
2. Individual Proportionality Review
a. Frequency Analysis
Although we have recognized the analytic potential of frequency analysis, we have also acknowledged, in each of our cases, that frequency analysis has so far fallen short of this potential. See DiFrisco III, supra, 142 N.J. at 171, 662 A.2d 442; Martini II, supra, 139 N.J. at 28-29, 651 A.2d 949; Bey IV, supra, 137 N.J. at 350-51, 645 A.2d 685; Marshall II, supra, 130 N.J. at 173-74, 613 A.2d 1059. Much of the difficulty has arisen because of the small size of the pool of eases used for comparison purposes. See Martini II, supra, 139 N.J. at 28-29, 651 A.2d 949. Because frequency analysis is statistically based, and because conclusions drawn from small sample sizes are inherently unreliable, we have not had confidence in the results produced by the models. See DiFrisco III, supra, 142 N.J. at 171, 662 A.2d 442; Martini II, supra, 139 N.J. at 28-29, 651 A.2d 949; Bey IV, supra, 137 N.J. at 350-51, 645 A.2d 685; Marshall II, supra, 130 N.J. at 173-74, 613 A.2d 1059. Still, we remained convinced that as the size of the pool of comparison cases increased over time, we would reach a level of statistical reliability that would enable us to place greater weight on the models. See Bey IV, supra, 137 N.J. at 350, 645 A.2d 685.
*292Our review of the records maintained by the AOC discloses that the number of penalty trials and the number of death sentences imposed by penalty-phase juries have remained small. Chew/Cooper/Harvey Report, tbl.2. In each year since 1995, fewer than ten penalty trials have been conducted and fewer than five death sentences have been handed down. Ibid. In contrast, the number of penalty trials conducted and death sentences imposed in the late 1980s was considerably higher. For example, in 1987 alone twenty-one penalty trials were held and nine death sentences were handed down. Ibid. We recite these numbers not to draw conclusions about the decisions made by prosecutors and juries in respect of the death penalty, but rather to point out that key components of our statistical database are not growing at the rate that earlier had been expected. Review and reflection suggest that a level of statistical reliability may not be attained in the near future. It may be possible to determine, based on projections about the size of the database over time, how long it will actually take before frequency review results can be useful to the Court. We expect the Special Master to make findings on this issue.
We turn now to a discussion of the problems specific to the individual frequency analysis tests.
i. Salient-Factors Test
The salient-factors test employs thirteen major comparison groups or categories of cases ranked generally in descending order of blameworthiness and derived from the statutory aggravating factors. Final Report, supra, at 81-82. Each of these thirteen major categories contains two to seven subcategories based on additional statutory and nonstatutory aggravating factors. A defendant is assigned to only one of the thirteen categories and, then, to only one subcategory. DiFrisco III, supra, 142 N.J. at 166-67, 662 A.2d 442. Most important here, the model contains a total of forty-seven groups or categories of cases. Final Report, supra, at 83-84, tbl.7.
Although we regard the salient-factors test as the “most persuasive” of our frequency tests, DiFrisco III, supra, 142 N.J. at 173, *293662 A.2d 442; Martini II, supra, 139 N.J. at 33, 651 A.2d 949; Bey IV, supra, 137 N.J. at 353, 645 A.2d 685; Marshall II, supra, 130 N.J. at 168, 613 A.2d 1059, we have said repeatedly that the small size of the comparison groups “preclude[s] us from investing great weight in ... [the] results.” DiFrisco III, supra, 142 N.J. at 174, 662 A.2d 442; see also Martini II, supra, 139 N.J. at 37-38, 651 A.2d 949. These small samples are, in part, a consequence of the relatively small universe of comparison cases, but also stem from the decision made by Special Master Baldus and adopted by this Court to establish a large number of comparison groups. Marshall II, supra, 130 N.J. at 146, 613 A.2d 1059 (citing Final Report, supra, at 80). In designing the salient-factors test, Professor Baldus sought to achieve “factual comparability” of.similar cases by narrowly and precisely delineating the salient-factors classifications, Final Report, supra, at 80, with the result that there are forty-seven separate categories in this test.
As a matter of general principle, we continue to accept the rationale of Professor Baldus that the salient-factors test should be “sensitiv[e] to [the] nuances of the cases which appear to be statistically and practically important.” Id. at 83. Our experience in conducting the test since Marshall II indicates, however, that a narrow and precise delineation of case classifications may not be optimal when the comparison group of eases produced by the classification scheme is insufficient to support reliable results. See DiFrisco III, supra, 142 N.J. at 174, 662 A.2d 442; Martini II, supra, 139 N.J. at 37-38, 651 A.2d 949. It may well be that realigned compact groupings would prove more useful to the Court. We note, by way of illustration, that the Tennessee Supreme Court employs seventeen separate factors in identifying and analyzing comparable cases and that the Washington Supreme Court employs only four. See Bland, supra, 958 S.W.2d at 667; Brown, supra, 940 P.2d at 562.
Special Master Baldus anticipated that the salient-factors test would “evolve in light of experience.” Final Report, supra, at 80. We agree, and conclude that experience requires us to reexamine *294the test’s present case classification scheme, and to ask the Special Master to consider in particular whether some reduction in the number of case classifications is possible without compromising the overarching principle that only similar cases be compared. The Special Master should also examine alternate case sorting approaches which account for mitigating factors in order to provide a more complete picture for comparison purposes.
ii. Numerical-Preponderance-of-Aggravating- and-Mitigating-Factors Test
In the numerical-preponderance test, a defendant’s case is compared to other eases having the same number of aggravating and mitigating factors. When adopted, the test was believed to “shed[ ] important light on how, over the long run, juries are likely to sentence defendants with any given number of aggravating and mitigating circumstances.” Id. at 89. The test was nonetheless recognized to have inherent limitations in “that it assumes an equal weight for all aggravating and mitigating circumstances.” Id. at 92.
In Marshall II, the Court expressed its reservations about the quantitative nature of the test:
This test is more problematic than either the salient-factors or the index-of-outcomes analyses. Abstractly, we sense that a quantitative rather than a qualitative analysis will be rather unproductive. No matter how many aggravating factors and how few mitigating factors are presented, the jury’s decision is intensely qualitative____ We have repeatedly emphasized that juries are to make a qualitative, and not a quantitative, analysis of aggravating and mitigating factors.
[130 N.J. at. 171, 613 A.2d 1059.]
Later proportionality review opinions have echoed these concerns. See DiFrisco III, supra, 142 N.J. at 175, 662 A.2d 442 (“[T]his test, too, is fraught with uncertainty____ [I]t assumes that juries weigh each of the aggravating and mitigating factors equally, an assumption that fails to account for the qualitative nature of jury deliberations.”) (citations omitted); Martini II, supra, 139 N.J. at 38, 651 A.2d 949 (acknowledging same concern). Nevertheless, we adopted and have continued to apply the numerical-preponderance test because of its potential to provide larger samplé sizes, and *295possibly more reliable results, particularly in comparison to the salient-factors test. DiFrisco III, supra, 142 N.J. at 175, 662 A.2d 442. In fact, the results of the numerical-preponderance test have been consistently discounted, not only because the test fails “to account for the qualitative character of jury deliberations,” Martini II, supra, 139 N.J. at 38, 651 A.2d 949, but also due to persistently small sample sizes, see, e.g., DiFrisco III, supra, 142 N.J. at 178, 662 A.2d 442.
In our experience the numerical-preponderance test has not contributed to the Court’s proportionality reviews and, in light of its inherent flaws, cannot be expected to do so in the future. We have determined, therefore, that this test should be abandoned.
iii. Index-of-Outcomes Test
In the index-of-outeomes test, the Court “compare[s] the blameworthiness of different defendants by statistically-relevant measures of culpability.” Martini II, supra, 139 N.J. at 42, 651 A.2d 949. The test uses multiple-regression analyses 8 employing formulas containing as many as thirty-two factors or variables. By this method, cases are ranked “according to overall defendant culpability, as measured by the presence or absence ... of factors that appear to influence prosecutorial and jury decision-making.” Final Report, supra, at 93. Yet, as early as Marshall II, and most recently in DiFrisco III, the Court has recognized that “the small sample size of eases with similar levels of blameworthiness and the great ranges in the confidence intervals” prevent us from *296relying on the results. 142 N.J. at 182, 662 A.2d 442.9 In essence, although we have assigned a predicted probability of receiving a death sentence to each of the defendants before us, we have been unable to rely on that prediction because of the instability of the multiple-regression models. Most important, for reasons we have discussed, supra at 291-292, 724 A.2d at 148-149, we are uncertain whether we will soon reach a sample size capable of supporting reliable results in these models.
We will ask the Special Master to undertake a review of both the strengths and the weaknesses of this test and to make recommendations whether the models can be modified and improved, or whether the index-of-outeomes test should also be eliminated.
b. Precedent-Seeking Review
Precedent-seeking review requires us to engage in a traditional case-by-case analysis of similar death-eligible cases. Martini II, supra, 139 N.J. at 46, 651 A.2d 949. We have consistently placed our reliance on this form of review because of the analytic difficulties we have encountered in applying frequency analysis. See DiFrisco III, supra, 142 N.J. at 171, 662 A.2d 442; Martini II, supra, 139 N.J. at 28-29, 651 A.2d 949; Bey IV, supra, 137 N.J. at 350, 645 A.2d 685; Marshall II, supra, 130 N.J. at 173-74, 613 A.2d 1059. Yet, our concerns with respect to precedent-seeking review overlap with our concerns about the underlying reliability of the data.
We have observed that data collection in cases that were not capitally prosecuted entails difficult and highly subjective determinations of whether a prosecutor could have proven the case and whether the case is death-eligible. Ibid. Because precedent-seeking review requires that we do a careful analysis of the facts developed by the AOC, and because the pool of cases that were not capitally prosecuted constitutes over fifty-percent of the total *297universe of cases, the reliability and sufficiency of this data is especially critical to the precedent-seeking inquiry.
Moreover, precedent-seeking review is directly affected by the case-classification scheme employed in the salient-factors test because that test and precedent-seeking review rely on the same initial classifications. See Bey IV, supra, 137 N.J. at 366, 645 A.2d 685. We have determined that an independent assessment is warranted on the question whether some reduction in the number of case classifications, presently forty-seven in number, is possible without compromising the principle that only similar cases be compared. As we have pointed out, the Tennessee Supreme Court does a precedent-seeking review based on seventeen separate groups of comparable cases. See Bland, supra, 958 S.W.2d at 667. Its approach may serve as an example for the Special Master in his review.
We are aware that a reduction in case classifications will result in more cases in each group. Larger groups present practical difficulties if we continue our present practice of giving separate consideration to each case in the classification to which the defendant’s case has been assigned. Even if we retain our present classification scheme, we would be unable to describe and review each comparable case because of the slow but continuing growth in the number of eases. Therefore, as part of our reconsideration of precedent-seeking review, we will ask the Special Master to examine appropriate methods by which to select a representative number of cases within the group of similar cases for consideration and comparison to the defendant’s case.
Precedent-seeking review has been and continues to be our most effective means of reviewing the proportionality of a defendant’s sentence of death. For this reason, the database and case groups should be constructed with great care. We will rely on this form of review even as we anticipate recommendations for improvements from the Special Master.
*2983. Systemic Proportionality Review and Possible Racial Disparity in the Imposition of the Death Penalty
In Marshall II, we said that one purpose of proportionality review is “the prevention of ‘any impermissible discrimination in imposing the death penalty.’ ” 130 N.J. at 135, 613 A.2d 1059 (quoting Ramseur, supra, 106 N.J. at 327, 524 A.2d 188). We rejected the view of the United States Supreme Court that “ ‘[ajpparent disparities in sentencing are an inevitable part of our criminal justice system,’” id. at 207, 613 A.2d 1059 (quoting McCleskey v. Kemp, 481 U.S. 279, 312, 107 S.Ct. 1756, 1778, 95 L.Ed.2d 262, 291 (1987)), and turned for guidance to the history and traditions of our state to “provide a basis for the independent application of [our] constitution,” State v. Hunt, 91 N.J. 338, 366, 450 A.2d 952 (1982) (Handler, J., concurring). We found in our past, reflected in New Jersey’s statutory and decisional law, the commitment of this state’s citizens “to the elimination of racial discrimination” in all of its invidious forms. Marshall II, supra, 130 N.J. at 207, 613 A.2d 1059. To accept racial disparity in capital sentencing would be to abandon that commitment and “our own constitutional guarantee of equal protection of the laws under New Jersey Constitution Article I, paragraph 1.” Ibid. We cannot follow such a course. “[W]ere we to believe that the race of the victim and race of the defendant played a significant part in capital-sentencing decisions in New Jersey, we would seek corrective measures, and if that failed we would not, consistent with our State’s policy, tolerate discrimination that threatened the foundation of our system of law.” Id. at 209, 613 A.2d 1059.
Robert Marshall was the first defendant to present a statistical- ' ly based claim purporting to show systemic racial disparity in capital sentencing. Id. at 210, 613 A.2d 1059. His claim was based on an apparent disparity in the rate at which black and nonblack defendants were sentenced to death at the middle range of culpability. Id. at 210-11, 613 A.2d 1059. We were, however, unable to draw any conclusions from the data because there ““were too few penalty-trial death-verdict cases involving black *299and non-blaek defendants with comparable levels of culpability to support any finding at all.’ ” Id. at 213, 613 A.2d 1059 (quoting David C. Baldus, Special Master, State v. Robert Marshall: Death Penalty Proportionality Review Project, A Report to the New Jersey Supreme Court 103 (Sept. 24, 1991) (Marshall Report)), We also considered statistical data showing an apparent disparity in the rate at which prosecutors seek the death penalty in black-victim cases as distinct from white-victim cases. Id. at 211, 613 A.2d 1059. Here too, we were unable to rely on the data, noting that the Special Master characterized the statistical results as “ ‘strictly preliminary.’ ” Ibid, (quoting Marshall Report, supra, at 101). We concluded that “evidence of eonstitutionally-signifieant race-based disparities in sentencing” had not been found. Id. at 210, 613 A.2d 1059.
In Bey IV, the defendant presented new statistical data purporting to show evidence of racial disparity in the imposition of the death penalty. 137 N.J. at 392, 645 A.2d 685. As in Marshall II, we rejected Bey’s claim because the statistical database did not contain “a sufficient number of cases to determine whether a significant statistical disparity exists between death-sentenc[ed] black and non-black defendants.” Id. at 393, 645 A.2d 685. We observed:
Without a sufficient number of similar cases, we cannot hold that race impermissibly influences the imposition of the death penalty. As vexing as waiting for more data may be, we have no alternative but to wait.
[Id. at 394, 645 A.2d 685.]
For similar reasons, we rejected claims of racial disparity in the two proportionality review cases following Bey IV. See Martini II, supra, 139 N.J. at 80, 651 A.2d 949; DiFrisco III, supra, 142 N.J. at 210, 662 A.2d 442.
a. Race as a Predictor of Outcome
In this case, the number of death-sentenced cases in the database has risen to forty-seven, Loftin Report, tbl.l, an increase of almost twenty-five percent over the number of cases in the Court’s previous proportionality review decision in DiFrisco III, supra, *300142 N.J. at 163, 662 A.2d 442; see also DiFrisco Report, tbl.1. Similarly, there are now 369 death-eligible cases, Loftin Report, tbl.l, an increase of nearly twenty percent over DiFrisco III, supra, 142 N.J. at 163, 662 A.2d 442; see also DiFrisco Report, tbl.l. Because of the larger database, the AOC used logistic regression procedures to model defendant’s racial disparity claim instead of the less reliable discriminate regression procedures used in all prior proportionality review cases. Discriminate analysis assumes that the independent variables used in the model are of the “interval” variety, ie., the variables represent factors that are capable of appearing across a broad scale. Memorandum from David Weisburd, AOC statistical consultant, to John P. McCarthy, Jr., Assistant Director, AOC 1 (Dec. 16,1995) (reprinted in Loftin Report, technical app. 9). Examples of interval variables include age and income. Ibid. Many of the variables considered in the AOC models, however, are of the “nominal” variety, ie., the variables stand for simple “yes/no” choices. Ibid. For example, all of the statutory aggravating and mitigating factors are of the nominal variety, ie., they are either present (yes), or not present (no). Ibid. In contrast to discriminate analysis, logistic analysis “allows a mix of variables with different levels of measurement.”10
The AOC employs three different statistical models to measure the race effect in capital sentencing — Schedules 2, 5 and 8. See Loftin Report, technical app. 10, at 16, 20. In each of the models, white victims (“whitvic”) and black defendants (“blackd”) are included as suspect variables. Schedules 2 and 5 attempt to measure the race effect of decisions made by penalty-trial juries. Schedule 2 includes twenty statutory aggravating and mitigating factors in addition to the suspect race variables. Schedule 5 includes all of the variables in Schedule 2 and adds eight nonstatu*301tory variables. Schedule 8 attempts to measure the race effect in all death-eligible cases, and contains variables identical to those found in Schedule 2.
The results of the AOC’s logistic regression analyses are summarized as follows:
Observed * Variable Schedule Coefficient Significance
whitvic 1.3555 .0412
blackd 1.4522 .0171
whitvic .5157 .5535
blackd 2.3796 .0066
whitvic 1.0699 .0240
blackd .9520 .0418
* Odds that chance alone would account for the result.
[Id. at 16, 18, 20.]
These outcomes show statistical significance in five of the six categories, ie., five categories have a p-value or p-level of .05 or less.11 On this basis, the AOC informed us that the outcomes “suggest! ] a possible ‘race effect’ in capital jury decision making.” See Memorandum from John P. McCarthy, Jr., Assistant Director, AOC, to Stephen W. Townsend, Clerk of the Supreme Court 1 (Nov. 25, 1996) (McCarthy Memorandum) (reprinted in Loftin Report).
*302According to defendant, Schedules 2 and 5 provide strong evidence that race is a predictor of the outcome of a penalty trial. With respect to Schedule 2, defendant points out that the variable “blackd” achieves a higher level of statistical significance than such statutory aggravating factors as murder of a law enforcement officer, N.J.S.A. 2C:ll-3c(4)(h), contract principal, N.J.S.A 2C:11-3c(4)(e) or felony murder, N.J.S.A 2C:ll-3c(4)(g). See Loftin Report, technical app. 10, at 16. Schedule 5 evidences a similar level of statistical significance for the variable “blackd” compared to other statutory aggravating factors. Id. at 18. As to Schedule 8, which encompasses all death-eligible cases, defendant notes that the “whitevic” variable achieves statistical significance, id. at 20, and suggests that this outcome shows the presence of a white victim has a pernicious effect on the capital prosecution and sentencing system as a whole.
The State vigorously disputes the accuracy and reliability of the statistical models. Relying on its own statistical expert, Dr. John E. Rolph, the State argues that the models contain numerous methodological flaws, including under-inclusion of variables relevant to jury decision making, an insufficient number of cases in the database, and the failure to account for the interaction effect that certain variables may have on one another. The State also questions the accuracy of the database and, in particular, certain data coding decisions made by the AOC. The claim is that errors in the coding of eases may infect the entire database, thereby rendering suspect any conclusions drawn from the models,
b. Review of the Statistical Models
The outcomes produced by the models, and the sharply contested interpretations concerning their significance, caused the Court to seek assistance in resolving the highly technical issues raised by the parties. On October 22, 1996, the Court appointed retired Superior Court Judge Richard S. Cohen as Special Master, and directed that he “conduct a review, perform analyses, and make findings and recommendations relating to defendants’ race as a possible factor in the decision of juries to impose the death *303penalty.” State v. Loftin, No. A-86-96, slip op. at 3 (Oct. 22, 1996). The Court directed the Special Master to perform the following tasks:
(1) ... examine and issue findings on the reliability of the data obtained by the Administrative Office of the Courts as it relates to the issue of racial discrimination in the proportionality-review proceedings [in this case and in State v. Harris ] ...;
(2) ... independently assess and rank by culpability, sentence and race of defendant all penalty-phase cases in the Administrative Office of the Courts’ proportionality review universe____;
(3) ... attempt independently to corroborate the appropriateness of the culpability rankings ...; and
(4) ... undertake or cause to be undertaken a precedent^seeking review of those cases within the mid-range of culpability____
[Id. at 3-4.]
Dr. John Tukey, Professor Emeritus of Statistics at Princeton University, served as a technical consultant to the Special Master.
Because of the Court’s calendar requirements, Judge Cohen was afforded only three months in which to perform his assigned tasks and file his report. Within this short period of time, he was able to define the universe of cases to be considered, analyze the raw numbers produced by the AOC, make findings concerning the reliability of the culpability rankings, and carry out an independent assessment of those rankings by conducting a survey in which fifty judges were asked to order groups of penalty-trial cases by culpability. The Special Master did not have sufficient time, however, to undertake a precedent-seeking review of those cases falling in the mid-range of culpability.
Judge Cohen presented his findings and recommendations in a report dated January 27, 1997. Richard S. Cohen,-Report to the Supreme Court of New Jersey (Jan. 27, 1997) {Special Master Report). In his report, he concluded that “the statistical evidence does not prove the defendant’s assertions of racial bias in penalty-trial verdicts, [and] neither does it prove that the system operates without bias.” Id. at 44-45. The Special Master found some indications of bias that “seem to trend, to a degree not revealed by the statistical evidence, toward showing a race effect.” Id. at 45. *304Nonetheless, in his view, “the Court is no closer than it was in the past to statistical evidence of race effect.” Ibid.
The parties were afforded the opportunity to respond to the findings and recommendations of the Special Master. Following the submission of briefs and oral argument, the Court asked the Special Master and Dr. Tukey to address certain issues raised by the Public Defender and several questions discussed at oral argument. Accordingly, Judge Cohen and Dr. Tukey filed a supplemental report dated May 4, 1997. Richard S. Cohen & Dr. John W. Tukey, A Study of Statistical Evidence of Race Bias in Penalty Trials (May 4,1997) {Supplemental Report).
The most significant aspect of the Supplemental Report is the modification of a principal conclusion set forth in the first report. On further consideration and in response to comments in Dr. Tukey’s supplemental report, Judge Cohen rejects his earlier statement that “‘it is troubling that so much of [the statistical evidence] tends in the same direction [toward a showing of racial bias].’ ” Id. at 10 (quoting Special Master Report, supra, at 43). The Special Master adopts the view of his consultant that the common leaning of the statistical evidence is of no significance when, as here, the separate analyses of the evidence are “ ‘substantially overlapping’” and have much in common with one another. Ibid, (quoting Dr. John W. Tukey, Report to the Special Master 2 (Jan. 27,1997) {Tukey Report)).
Moreover, in his earlier report, the Special Master had pointed to certain raw data as independent sources indicating some evidence of racial bias. In particular, he had noted the disparity in death-sentencing rates between black-defendant-white-victim cases and nonblack-defendant-nonwhite-victim cases. Special Master Report, supra, at 10, 43. He had acknowledged the difficulties in drawing conclusions from a small number of transracial cases,12 but nevertheless believed the disparity in the transra*305cial data supported his observation that “ ‘so much of [the data] tends in the same direction.’ ” Id. at 43. In his Supplemental Report, however, Judge Cohen found inconsistencies in coding transracial eases, ie., black defendants included black hispanics, and nonblack defendants included Asians and “others”; whereas white victims “excluded all hispanics, Asians, and others” who were instead included in the nonwhite-victim category. Supplemental Report, supra, at 10. When the coding was revised to realign the racial categories there were substantial shifts in the number of cases in various of the realigned categories. See Appendix A. As Judge Cohen pointed out, “sorting choices affect outcomes, sometimes a lot and sometimes a little, and the outcomes cannot be evaluated without keeping in mind the underlying choices that produced them.” Letter from Richard S. Cohen to the Supreme Court 2 (Feb. 7, 1997). The “ambivalence of the coding decisions” was another reason the Special Master rejected his earlier conclusion that apparent trends in the data suggest racial bias. Supplemental Report, supra, at 10.
Our dissenting colleague makes much of the Special Master’s initial concern about the tendency of the data to show discrimination, post at 381, 410, 724 A.2d at 194, 209, and downplays the modifications in the Supplemental Report. In our view, the Supplemental Report highlights the dangers inherent in the improper use of statistics. When separate analyses operate on the same data, it is not surprising that the results trend in the same direction. It is also not surprising that the results do not provide “mutually supportive[] evidence of race effect.” Supplemental Report, supra, at 10.
Judge Cohen does not, however, retreat from his earlier position that further inquiry on the question of racial bias is warranted. In the concluding section of his first report, he recommended *306continued efforts to develop and adopt more reliable statistical models of the race effect in capital sentencing. Special Master Report, supra, at 45. He further suggested that the Court consider establishing a permanent group of experienced judges to conduct assessments of penalty trial outcomes. Ibid. His conclusions were based on three discrete modes of inquiry: an examination of raw data, interpretations of statistical models prepared by the AOC, and the results of the judges’ survey on the comparative culpability of penalty-phase defendants.
We turn first to the raw data. Of the 362 death-eligible cases under study, fifty-six percent (203 cases) involved black defendants and forty-four percent (159 cases) involved nonblack defendants. Prosecutors capitally tried thirty-six percent Q%s cases) of the black defendants and forty-seven percent (74/159 eases) of the nonblack defendants. Thus, nonblack defendants were capitally tried thirty percent more frequently than blacks. Juries imposed death sentences on thirty-eight percent (28/73 cases) of the black defendants who were capitally tried and thirty percent (22/74 eases) of the nonblack defendants who were capitally tried, with the result that juries imposed the death sentence on black defendants twenty-seven percent more frequently than on nonblack defendants.
Considering the cumulative effect of these frequencies, the Special Master found that 13.79 percent (^oa cases) of death-eligible black defendants were sentenced to death and 13.84 percent (22/159 cases) of the nonblack defendants were sentenced to death. Id. at 8-9. Although he observed that the cumulative data are “race-neutral,” id. at 9, contrary to the dissent’s claim, the Special Master did not imply “that capital-murder prosecution is ‘race neutral.’ ” Post at 381, 724 A.2d at 195. Rather, Judge Cohen specifically stated:
The limited scope and significance of these findings should be made clear. They do not rule out race bias as a factor upstream in the process of investigation, arrest and indictment in homicide cases, and they do not deal at all with the fairness or accuracy of the guilhdetermining process. Also, they say nothing about the significance (if any) or the causes of an overage of capital prosecutions against *307nonblacks or an overage of death sentences against blacks. All they say is that nonblacks were 30% brought to penalty trial more frequently than blacks, and black penalty trial defendants received death sentences 27% more often than nonblack defendants.
[Special Master Report, supra, at 9 (footnote omitted).]
In the end, the Special Master determined that none of these results could be considered statistically significant including, most importantly, the apparent racial disparity in penalty-phase verdicts. Id. at 44 (finding the penalty-trial results are “well within one standard error ... and so are not near statistical significance”).
One final point in respect of the raw data deserves mention. To explain why the disparity between black defendants who were sentenced to death and nonblack defendants who were similarly sentenced is not statistically significant, the Special Master suggested an analogy to the random picking of balls out of a sack that contains an equal number of green and red balls. Id. at 14-15. He explained that even if there are 74 green balls and 74 red balls in the sack, a random picking of 50 balls will not necessarily produce 25 of each color. Id. at 14. The Special Master used this analogy to demonstrate that the disparity in the number of black defendants (28), as opposed to nonblack defendants (22), sentenced to death is not in and of itself statistically significant and “cannot be taken to prove a racial difference in results.” Ibid.
The dissent criticizes the Special Master’s analogy, stating that he “fail[ed] to account for culpability, the factor that should dictate a jury’s decision to sentence someone to death.” See post at 384, 724 A.2d at 196. Again, the dissent misreads Judge Cohen’s point: the Special Master understood the importance of the culpability rankings but found them to be unsound. The data simply does not “reveal tangible and frightening disparities____ adversely affect[ing] black defendants, killers of white victims, and black defendants who kill white victims.” Id. at 385, 724 A.2d at 197. The dissent’s rhetoric notwithstanding, the data cannot now inform us whether there is or there is not discrimination at any level in the death penalty system.
*308The Special Master counsels against reliance on the culpability rankings produced by the statistical models for many reasons, including numerous ambiguities in coding the eases, Special Master Report, supra, at 30-42, the small size of the database, id. at 20, and the manifest unsuitability of the models as a means to test racial bias, id. at 25-30. Notably, the small size of the database means that small changes in data coding can produce anomalous results; for example, if one defendant shifts one culpability level up or down, the statistical outcome may change substantially. These concerns ied the Special Master to conclude that the models are simply not “reliable bases for consequential decisions.” Id. at 44.
More specifically, the Special Master pointed to one of the culpability rankings assigned to defendant as evidence of the unreliability of the statistical models. Supplemental Report, supra, at 5. Defendant, a prior murderer, falls into culpability level one, the lowest level of culpability, in one of the three logistic regression models. Loftin Report, tbl.22 (reporting culpability scores among penalty-trial cases applying statutory and nonstatutory factors). The Special Master observed that penalty juries have returned death verdicts against a majority of prior murderers in a relatively large number of cases. Supplemental Report, supra, at 5. Thus, in his view, the AOC methodology “underrate[sj an aggravating factor which sound intuition and courthouse experience rank high.” Id. at 6. In so doing, the methodology “proceeds to give unwarranted support to the thesis that Loftin was condemned to death for his color.” Ibid.
Because this Court’s order to the Special Master “contemplated independent assessment of the culpability rankings and ratings of the penalty trial cases,” Special Master Report, supra, at 21, Judge Cohen asked experienced penalty-trial judges to participate in a culpability ranking survey that could then be compared to the AOC statistical results. In the survey, fifty judges were given short narrative statements describing each ease in a group of penalty-phase cases and were told to rank the individual cases in *309order of deathworthiness. The statements “included no information from which the judges could ascertain the race of either the defendant or the victim, or the outcome of the penalty trial.” Id. at 22. The judges were told to place each defendant into one of five culpability levels and to rank each'case in order of deathworthiness, with one being the least deathworthy and five being the most deathworthy. Id. at 23.
The judges’ culpability scores were averaged for each defendant and the results analyzed. The average rating of all penalty-trial defendants was 3.34, with 3.30 for black defendants and 3.38 for nonblaek defendants. Ibid. The difference, 2.4 percent, was characterized by the Special Master as “insignificant.” Ibid. The average culpability rating of defendants who received a death sentence was 3.72, with 3.68 for blacks and 3.77 for nonblacks. Ibid. The Special Master found this difference to be without meaning, id. at 23-24, and concluded that the judges “gave culpability ratings that were essentially colorblind.” Id. at 23. Judge Cohen believed these results to be important because
the survey judges did not know defendants’ race; the juries did____ [T]he judges’ race-blind ratings turned out to be race-neutral in culpability levels. They did not rate nonblack defendants ... more deathworthy than black defendants.
[Id. at 24-25.]
Judge Cohen concluded that the results of the judges’ survey challenge the accuracy and reliability of the statistical models indicating a race effect. Id. at 25.
The dissent believes that the Special Master’s decision to average the results of the judges’ culpability survey “may not reveal the true level of disparity in culpability between the races.” See post at 393, 724 A.2d at 201. By way of example, the dissent points out that averaging might be misleading because a large number of nonblack defendants could be at the extreme levels of culpability and the black defendants could be evenly divided among the five culpability levels. Id. at 393-394 n. 13, 724 A.2d at 201 n. 13 (assuming that there are nine nonblacks with culpability levels of 1, 1, 1, 3, 3, 3, 5, 5, and 5, and nine blacks all with *310culpability levels of 3). The fear is that the “true level of disparity” is hidden by averaging.
In fact, the judges’ culpability ratings essentially “produced a classic normal distribution” for both black and nonblack defendants. Special Master Report, supra, at 23; see also Appendix B. As the Special Master explained,
There is always a risk in averaging judgments that an idiosyncratic response or two will distort the results. If there were a large number of responses, so that an ample number remained after discarding outliers, that course of action might make sense. In our survey, however, we ordinarily had five judges rating each case, and we felt it unwise to discard any outlier, and leave only three or four. In addition, it appeared to us that the incidence of very disparate responses was low, and therefore outliers were not a significant problem. Furthermore, there is no reason to believe that outliers leaned more in one direction than the other, and therefore no indication that the course we took had any effect on the accuracy of the results.
[Supplemental Report, supra, at 8 (emphasis added).]
We agree with the Special Master’s conclusion “that the essentially race-blind results of the survey seriously undercut the notion that non-black penalty-trial defendants are noticeably more death-worthy than their black counterparts.” Id. at 7. We observe that the different results produced by the AOC data and the judges’ survey in respect of the relative culpability of nonblack defendants and black defendants may be explained by the small size of the database. It is possible that nonblack defendants are disproportionately represented in certain categories and that the small number of cases in each of these categories could distort the culpability estimates, resulting in an exaggeration of the death-worthiness of nonblack defendants.
There are, however, more fundamental problems with the statistical models. The methods we use were developed by Professor David Baldus for use by this Court in conducting individual proportionality reviews, not for the purpose of assessing systemic discrimination. See Marshall II, supra, 130 N.J. at 152-55, 172-74, 613 A.2d 1059. Special Master Cohen considered whether the models are capable of providing information about racial disparity in death sentencing and found that “the methodology reveals serious flaws when diverted to the task of providing reliable *311evidence relating to the race question.” Special Master Report, supra, at 42. In the view of the Special Master and Dr. Tukey, a statistical model designed to test racial bias should employ between five and ten parameters or. variables, whereas the AOC models employ as many as thirty-two.13 Id. at 27-28. A statistical model with a relatively small number of well-crafted parameters is known as a parsimonious model. Id. at 28; Tukey Report, supra, at 5-7.
Judge Cohen and Dr. Tukey further advise that a reasonably parsimonious model is necessary to achieve a minimal degree of statistical reliability when, as here, the size of the database is small and the purpose of the analysis is to “generalize” from the data “and make judgments about their significance.” Supplemental Report, supra, at 2. Dr. Tukey, in an effort to illustrate the parsimony principle, created three variants of parsimonious models to look at possible race effects. He then employed two methods for assessing the standard of error for the black defendant variable: (1) SAS, the off-the-shelf program used by the AOC; and (2) an alternative, known as the “jackknife technique.” Tukey Report, supra, at 9. Although Model 3 showed a statistically significant race effect when the SAS program was used, the results of the jackknife did not achieve statistical significance. Id. at 10. The dissent relies heavily on the SAS program results, again ignoring contrary outcomes, to conclude that there is racial bias against black defendants. The conflicting results derived from the same data set actually compel a conclusion that racial disparity has not been demonstrated.
*312Equally important, the Tukey models were not the result of independent research, but were based on the Baldus-methodology culpability ratings and operated on the same small body of data. They were done in an attempt to “alleviate some of the pathology that infected the models the AOC employed,” Supplemental Report, supra, at 14, and not as the most effective means of studying possible racial bias. In this context, the dissent’s suggestion that Model 3 is the “most methodologically sound” of the Tukey models, post at 388, 724 A.2d at 198, has little meaning. Given the limited usefulness of the Tukey models generally, and the mixed results from the third model specifically, the dissent’s reliance on Model 3 is misplaced.
Many of the Special Master’s findings are vigorously disputed by defendant. He asserts that the Special Master erred in discounting the reliability of the AOC culpability models and presents his own expert, Dr. Paul Allison, who states that the AOC “data compare favorably with some of the best research work in the social sciences and cannot be casually dismissed as ‘unreliable’.” Memorandum from Dr. Paul Allison, defendant’s statistical expert, to Mordecai Garelick and Claudia Van Wyck, defendant’s counsel 1 (May 22, 1997) (on file with Court). Defendant also offers alternate statistical models prepared by Dr. Allison. In an attempt to refine Dr. Tukey’s work, Dr. Allison added the white-victim variable to the three Tukey models. Two of the reconfigured models showed a statistically significant race effect.
Like the Tukey models, however, Dr. Allison’s models are “infected” by the suspect culpability ratings and the small size of the database. See supra at 310-312, 724 A.2d at 158. Moreover, Dr. Allison apparently relied on transracial data in which the black-nonblack coding of defendants differed from the white-nonwhite coding of victims. Supplemental Report, supra, at 13-14. As discussed earlier, supra at 305, 724 A.2d at 155, Judge Cohen found that after the data were recoded, there were dramatic changes in the number of cases in certain categories, Supple*313mental Report, supra at 13, indicating that the effect of coding decisions on the statistical results can be substantial. Even setting aside the coding and database problems, the models represent but. preliminary efforts to develop adequate parsimonious models. Indeed, both the Allison and the Tukey versions have not been subject to the type of study and analysis that would permit the Court to make an informed decision about their usefulness.
Finally, we disagree with our dissenting colleague’s suggestion that, despite the statistical models’ “imperfections,” the Court “should not hesitate” to find evidence of a race effect. Post at 400, 724 A.2d at 204. The dissent relies on our toxic-tort cases for the proposition that strict scientific standards may be relaxed where “ ‘the scientific method fails ... to address or accommodate the needs and goals’ to be achieved by a judicial determination.” Id. at 400, 724 A.2d at 204 (quoting Rubanick v. Witco Chemical Corp., 125 N.J. 421, 436-37, 593 A.2d 733 (1991)). .In Rubanick v. Witco Chemical Corp., we held that “a scientific theory of causation that has not yet reached general acceptance may be found to be sufficiently reliable if it is based on a sound, adequately-founded scientific methodology involving data and information of the type reasonably relied on by experts in the scientific field.” Rubanick, supra, 125 N.J. at 449, 593 A.2d 733; see also Landrigan v. Celotex, 127 N.J. 404, 414, 605 A.2d 1079 (1992) (stating that “the key to admission of the opinion is the validity of the expert’s reasoning and methodology”). The difficulty with the dissent’s argument, however, is that here the statistical models we use for proportionality review do not even meet the more lenient standard accepted by the Court in toxic-tort cases.
Our dissenting colleague further suggests that the risk of discrimination is sufficient to invalidate our capital punishment system. See post at 387, 724 A.2d at 198 (declaring that the Court cannot “continue to ignore the risk that racial discrimination is infecting the State’s imposition of the death penalty”). He eloquently details this country’s terrible history of slavery and con*314tinuing racial discrimination, id. at 404-410, 724 A.2d at 206-209, and finds that the “legacy of racism confirm[s] the statistical evidence that racial discrimination influences the imposition of capital punishment,” id. at 405, 724 A.2d at 207 (citing McCleskey, supra, 481 U.S. at 328-30, 107 S.Ct. at 1786-87, 95 L.Ed.2d at 302 (Brennan, J., dissenting)). But “[t]he question ‘is at what point that risk [of discrimination] becomes constitutionally unacceptable.’ ” McCleskey, supra, 481 U.S. at 308-09, 107 S.Ct. at 1776, 95 L.Ed.2d at 289 (quoting Turner v. Murray, 476 U.S. 28, 36 n. 8, 106 S.Ct. 1683, 1688 n. 8, 90 L.Ed.2d 27 (1986)). Past discrimination cannot validate flawed statistical data and prove present discrimination in the administration of the death penalty.
Under our system of proportionality review, the burden rests with the defendant to establish disproportionality. Bey IV, supra, 137 N.J. at 343, 645 A.2d 685. We conduct systemic proportionality review because of the risk of discrimination, but it is the defendant who must show that the risk is “constitutionally significant.” McCleskey, supra, 481 U.S. at 313, 107 S.Ct. at 1778, 95 L.Ed.2d at 292. Judge Cohen stressed the inherent limitations of statistical analysis when applied to the task of recognizing racial bias in the administration of the death penalty. He said that statistical approaches such as the AOC models are “incapable of disproving race bias.” Supplemental Report, supra, at 11 (emphasis added). At best, statistical approaches can furnish evidence that the administration of the death penalty is “inconsistent with raceblindness.” Id. at 12. In Judge Cohen’s view, a defendant seeking to invoke statistical analysis to show racial bias in the administration of the death penalty should be subject to “a burden of proof consistent with the nature of the evidence and conforming to the science he invokes.” Ibid. On the statistical evidence before him, Judge Cohen concluded that he did not find “relentless documentation or even a preponderance in the direction of the existence of any race bias.” Ibid.; see Marshall II, supra, 130 N.J. at 213, 613 A.2d 1059 (concluding that, where “‘statistical evidence ... relentlessly documents the risk that [death] sentence *315was influenced by racial considerations,’ ” the sentence would be invalidated) (quoting McCleskey, supra, 481 U.S. at 328, 107 S.Ct. at 1786, 95 L.Ed.2d at 301, Brennan, J. dissenting). Our consideration of the record in this case convinces us that defendant has not “relentlessly doeument[ed] the risk” of racial disparity in the imposition of the death penalty.
We make one other point in response to the dissent’s claim that we have “turned the convenient measure of ‘statistical significance’ into a talismanic rule.” Post at 401, 724 A.2d at 205. We agree that “[t]here is nothing magic about ____05” as a measure of statistical significance. Id. at 402, 724 A.2d at 205 (quoting Joseph Sanders, From Science to Evidence: The Testimony on Causation in the Benediction Cases, 46 Stan. L.Rev. 1, 15 (1993)). This standard is generally used by scientists “[b]ecause [they] want to be satisfied that they are not claiming false positives-causal relationships that do not exist____” Sanders, supra, 46 Stan. L.Rev. at 15. We, also, expect the link between racial bias and the death penalty to be clearly established before we find that our system of capital punishment is so seriously flawed as to require intervention by this Court. We do not seek “punctilious and pristine statistical analyses,” post at 398, 724 A.2d at 203, but, rather, a sound methodology that is capable of producing reliable results.
This Court is committed to a course of review that is capable of discerning possible racial discrimination in our capital sentencing system. Judge Cohen recommended that we continue our efforts to “develop and adopt more parsimonious models for more reliable regression studies of the race effect.” Special Master Report, supra, at 45. We adopt his recommendation and will ask the Special Master to undertake that effort.14 In addition, the survey of judges conducted by Judge Cohen provides encouraging but *316provisional evidence of nondiscrimination. At the least, the survey suggests that the culpability ratings derived from the statistical models should be subject to a form of independent verification. See id. at 25, 45. For this reason, Judge Cohen recommended that we appoint a panel of judges to perform periodic assessments of penalty-trial outcomes and thereby “provide another source of useful culpability ratings to guide not only tests for race effect but also [individual] proportionality review generally.” Id. at 45. This recommendation should be considered by the Special Master along with the composition and mandate of an independent judicial panel if appropriate.
Our review of the record in this case convinces us that defendant has not demonstrated racial disparity in the imposition of the death penalty.
4. Proportionality Review and> its Status as a Separate Proceeding in Death Penalty Appeals
Proportionality review has been conducted as a separate proceeding following a defendant’s unsuccessful direct appeal. See DiFrisco III, 142 N.J. at 156, 662 A.2d 442; Martini II, 139 N.J. at 15, 651 A.2d 949; Bey IV, 137 N.J. at 339, 645 A.2d 685; Marshall II, 130 N.J. at 117, 613 A.2d 1059. The Court expected that a separate proceeding would conserve resources because a proportionality review would not occur if the defendant’s direct appeal was successful. We recognize, however, that this practice also exacts a cost by drawing out the appeals process when a death sentence is affirmed.
The Court is sensitive to the need to achieve finality in the appeals process within a reasonable period of time consistent with a full and fair hearing of all of defendant’s claims on appeal. We will ask the Special Master to develop a factual record and issue findings concerning the desirability of maintaining proportionality review as a separate proceeding or, alternately, conducting proportionality review in connection with a capital defendant’s direct appeal.
*317III
Application of the Methods of Individual Proportionality Review to Loftin
We have decided that, until our review of the Special Master’s findings and recommendations is complete, we will not apply the 1992 statutory amendment limiting proportionality review to similar cases in which the defendant has been sentenced to death, but rather will continue to compare all death-eligible homicides with the ease before us. See supra at 286, 724 A.2d at 145. In addition, while we await the Special Master’s recommendations vis-a-vis proportionality review generally, we will continue to analyze defendants’ cases according to the methodologies and procedures previously utilized, except that we will no longer conduct the numerical-preponderance test previously used as part of our frequency approach.
A. Facts
The facts of this case are set forth in detail in Loftin I, supra, 146 N.J. at 318-33, 680 A.2d 677. We repeat here only those facts that are relevant to our proportionality review.
Sometime between 4:10 a.m. and 6:10 a.m. on the morning of May 5, 1992, Gary Marsh was shot in the head with a single bullet from a .380 caliber pistol during a robbery. He was working the night shift at a gas station in Lawrenceville, New Jersey. A coworker who arrived at 6:10 a.m. found Marsh lying on the floor in the office of the station, unconscious and struggling for breath, with his head in a puddle of blood. The office keys were in the door, the cash drawer was found empty on the counter, and there was loose change on the floor. Approximately ninety dollars was missing, including one fifty-dollar bill. There were no signs of a struggle in the small, narrow office. Marsh never regained consciousness and died approximately nine-and-one-half hours after he was discovered. An autopsy revealed no defensive wounds that would have indicated Marsh struggled with his assailant.
*318Defendant, twenty-six-years old at the time, was arrested four days after the murder when he attempted to purchase a computer using a Sears credit card belonging to Marsh. At the time of his arrest, defendant possessed Marsh’s identification, a Washington State gun permit, a receipt for the purchase of a .380 caliber pistol, and a fifty-dollar bill. The murder weapon was recovered from defendant’s car, as was a plastic mask.
On July 8, 1994, defendant was found guilty of purposeful or knowing murder by his own conduct. The guilt phase jury found specifically “that defendant’s intent was to kill [Marsh] ... as opposed to causing him serious bodily injury.” Id. at 324, 680 A.2d 677. A separate penalty-phase jury heard much of the same evidence the State had presented during the guilt phase. In mitigation, defendant called, among others, Carmeta Albarus, an expert in compiling psychosocial histories, and Dr. Edward J. Dougherty, a defense psychologist.
Albarus, who had constructed defendant’s social history through interviews and document review, described his life to the jury. She testified that defendant was one of seven children born to Fred and Ellen Loftin. Fred, though nurturing, was a poor provider and disappeared when defendant was five-years old leaving Ellen as the sole support for the family. Because she worked long hours, Ellen was rarely at home. In Albarus’s view, Mrs. Loftin did not relate well to her children and had little ability to nurture them.
When only six-years old, defendant set his mattress on fire and burned the family home to the ground. After that, he and his family moved to a one-bedroom hotel room for an extended period. Although defendant struggled with drugs in his early teens, he later obtained his GED and “expended significant time and effort trying to help his siblings get on or stay on the right path.” Id. at 329, 680 A.2d 677.
Defendant married when he was twenty-years old. Because his mother did not approve of his marriage, defendant abandoned his birth family in order to remain loyal to his wife. As a result, his *319mother did not attend the funeral of his stillborn son. In part because of his inability to provide for his new family, defendant’s marriage was conflict-ridden. After searching for a job he could hold, defendant joined the Navy and attained some success. However, on the day he was due to sail, his wife attempted suicide and he was discharged to take care of her. Although he then attended and completed school to become an auto mechanic, he could only find work washing ears. Defendant subsequently went to community college but became depressed and frustrated because of his poor grades. His frustrations manifested themselves in his mental, and perhaps physical, abuse of his wife and children.
Dr. Dougherty testified that defendant suffered from a “borderline personality disorder.” The doctor explained that a person is diagnosed with a borderline personality if he or she satisfies five or more of the nine criteria set forth in the DSM-IV manual for evaluating this disorder. According to Dr. Dougherty, defendant exhibited symptoms of six or seven of the nine criteria:
He demonstrated: (1) frantic efforts to avoid real or imagined abandonment ... [;] (2) a pattern of unstable and intense interpersonal relationships characterized by alternating between extremes of idealization and devaluation (overidealization of his wife and mother); (3) identity disturbance: markedly and persistently unstable self-image or sense of self; (6) affective instability due to a marked reactivity of mood ... [;] (7) chronic feelings of emptiness ... [;] (8) inappropriate, intense anger or difficulty controlling anger ... [;] and (9) transient, stress-related paranoid ideation or severe dissociative symptoms.
[Id. at 331, 680 A.2d 677.]
In rebuttal, Dr. Charles Martinson, a psychiatric expert for. the State, testified that defendant did not suffer from a borderline personality disorder at the time of the offense, but rather “exhibited narcissistic and antisocial personality traits.” Id. at 332, 680 A.2d 677. Dr. Martinson said that defendant “committed the murder for economic reasons,” id. at 331, 680 A.2d 677, and “that the only ‘remorse’ defendant exhibited occurred when he learned that Marsh held two jobs.” Id. at 332, 680 A.2d 677. The defendant’s substantial inability to feel remorse was further demonstrated in a sentence completion test administered by a defense psychiatrist when, after the words “‘My greatest mistake was *320... ’[,] defendant wrote, ‘become incarcerated.’ ” Ibid. On December 6, 1994, the penalty-phase jury sentenced defendant to death. The jury determined the State had proven beyond a reasonable doubt three statutory aggravating factors: N.J.S.A 2C:ll-3e(4)(a) (prior murder); N.J.S.A 2C:ll-3c(4)(f) (escape detection); and N.J.S.A 2C:ll-3c(4)(g) (felony murder). The c(4)(f) aggravating factor was based on: the defendant’s attempt to disguise his appearance by wearing a mask and fatigues; the distance defendant traveled from his Bristol, Pennsylvania home “passing hundreds of potential targets along the way, to a secluded station in Lawrenceville Township, New Jersey;” the fact that Marsh “might have been able to identify his assailant’s voice, height, weight, and overall build” and could have “described the color, make, and model of defendant’s ear, not to mention its Pennsylvania license plates;” the lack of a “forced entry or of physical struggle inside or outside the office;” and the fact that “defendant did not have to kill Marsh to effectuate the robbery.” Loftin I, supra, 146 N.J. at 376-78, 680 A.2d 677. The c(4)(a) aggravating factor was based on defendant’s March 28, 1992, knowing or purposeful murder of Sophia Fetter in which defendant was implicated after his arrest for the Marsh murder. He was convicted of the Fetter murder on September 22, 1993, and that conviction was affirmed on appeal. State v. Loftin, 287 N.J.Super. 76, 670 A.2d 557 (App.Div.), certif. denied, 144 N.J. 175, 675 A.2d 1123 (1996).
At least one juror found the existence of three statutory mitigating factors. One juror found the c(5)(a) mitigating factor (extreme mental or emotional disturbance); two jurors found the c(5)(c) mitigating factor (defendant’s age); and several jurors found the c(5)(h) mitigating factor (catch-all) for different reasons: defendant was traumatized by the loss of his father at an early age; he was traumatized by the fire he set; he was emotionally impoverished growing up; he was raised in poverty; he was deprived of a positive male role model during his youth; he was a considerate and loving son; he provided his siblings with a positive sense of direction; he had assumed the responsibility of being a family man *321before he was emotionally and financially prepared to do so; he was traumatized by the loss of his first son; he had served in the Navy, attended community college, and been a model prisoner; he led a crime-free existence for his first twenty-five years; he has the love and support of his family; he was under the influence of mental and emotional pressure at the time the crime was committed; and he had offered to plead guilty in exchange for a life sentence. No juror found that defendant had sincere and heartfelt remorse.
The jurors unanimously concluded that the three aggravating factors outweighed the mitigating factors and sentenced defendant to death. We affirmed defendant’s conviction and death sentence and acknowledged his request for proportionality review. Loftin I, supra, 146 N.J. at 397-98, 680 A.2d 677. Subsequently, the Clerk of the Court set a briefing schedule and directed the AOC to undertake its usual statistical study in connection with the proportionality of defendant’s sentence. The AOC issued its Loftin Report on November 25,1996.
B. Focus of Review
We conduct proportionality review to determine whether a particular defendant’s death sentence is disproportionate when compared to other defendants “with similar characteristics ... [who have] committed] factually similar offenses in the same jurisdiction.” Bey IV, supra, 137 N.J. at 343, 645 A.2d 685 (internal quotation marks and citations omitted). Our review is offender-oriented. We “presume[] that the death penalty is proportional to the offense,” and ask ‘“whether the punishment fits the criminal,’ ” DiFrisco III, supra, 142 N.J. at 161, 662 A.2d 442 (citing Marshall II, supra, 130 N.J. at 129, 613 A.2d 1059, additional citations omitted). “ ‘[A] death sentence is comparatively excessive if other defendants with similar characteristics generally receive sentences other than death for committing factually similar offenses----’” Marshall II, supra, 130 N.J. at 153-54, 613 A.2d 1059 (quoting Tichnell, supra, 468 A.2d at 17 n. 18).
*322Our dissenting colleague turns this concept on its head when he asserts that “[t]he standard measuring disproportionality ... must require ... that, for defendants similarly situated, the death sentence be received in a defined preponderance of cases.” Post at 425, 724 A.2d at 217. Because New Jersey jurors have been sparing in their imposition of the death sentence, it will never be the case that death would be “generally received” or “received in a defined preponderance of cases.” Because juries impose death infrequently, we have recognized that “death need not be normal or general to be a licit sentence.” Marshall II, supra, 130 N.J. at 153, 613 A.2d 1059. The dissent has missed the point. He would have us find that death is the normal sentence when that can never be so.
“Proportionality review seeks to determine only whether a particular death sentence is aberrational, not whether it compares perfectly with other sentences.” Bey IV, supra, 137 N.J. at 352, 645 A.2d 685 (citation omitted). For this reason, statistical disparity is permissible, indeed expected. For this reason also, the Court has repeatedly declined to set a threshold at which the imposition of the death penalty becomes disproportionate. Di-Frisco III, supra, 142 N.J. at 171-72, 662 A.2d 442; Martini II, supra, 139 N.J. at 30, 651 A.2d 949; Bey IV, supra, 137 N.J. at 351, 645 A.2d 685. Yet, the dissent again urges a numerical standard for determining proportionality. Post at 418, 724 A.2d at 214; see also DiFrisco III, supra, 142 N.J. at 233, 662 A.2d 442 (Handler, J., dissenting); Martini II, supra, 139 N.J. at 90-91, 651 A.2d 949 (Handler, J., dissenting); Bey IV, supra, 137 N.J. at 408, 645 A.2d 685 (Handler, J., dissenting). We continue to believe, however, that setting such a standard “would introduce unacceptable arbitrariness into proportionality review.” Martini II, supra, 139 N.J. at 20, 651 A.2d 949; see also Bey IV, supra, 137 N.J. at 351, 645 A.2d 685 (stating that “[a] general standard, although admittedly imprecise, is not necessarily arbitrary”). As we observed in Martini, an absolute numerical standard “suffers from an inherent failure to distinguish between defendants.” *323Martini II, supra, 139 N.J. at 32, 651 A.2d 949. Furthermore, application of a strictly quantitative approach may unduly “ ‘limit the legitimate exercise of judicial discretion’ ” and “ ‘inappropriately suggest that the complex judgments involved in proportionality determinations can be expressed with mathematical precision.’ ” Marshall II, supra, 130 N.J. at 153, 613 A.2d 1059 (quoting Final Report, supra, at 42-43). We have never believed that statistics can replace the process of judging; we do believe that statistical results, properly used and understood, can alert us to the need for increased vigilance in our quest for impartial justice.
1. The Universe of Cases
The Loftin Report includes all death-eligible cases collected by the AOC from 1983 to September 1, 1996. Loftin Report, tbl.l n. 1. There are 369 death-eligible cases, 154 of which went to penalty trial, producing a rate of forty-two percent. Id., tbl.3. Of the 154 penalty-trial cases, forty-seven, or thirty-one percent, resulted in a death sentence. Id., tbl.2. The overall death-sentencing rate is therefore thirteen percent (47/369). Id., tbl.l.
2. Method of Classifying Cases
After establishing the universe of cases, we next convert that universe into a database for comparison purposes. As in our previous proportionality cases, we use two approaches: an a priori approach and an empirical approach. See DiFrisco III, supra, 142 N.J. at 163-64, 662 A.2d 442; Martini II, supra, 139 N.J. at 24, 651 A.2d 949; Bey IV, supra, 137 N.J. at 345, 645 A.2d 685; Marshall II, supra, 130 N.J. at 141-43, 613 A.2d 1059. In the a priori method, we analyze cases based on those factors that experience has shown influenced whether a defendant was sentenced capitally. See DiFrisco III, supra, 142 N.J. at 164, 662 A.2d 442; Martini II, supra, 139 N.J. at 24, 651 A.2d 949; Bey IV, supra, 137 N.J. at 345, 645 A.2d 685; Marshall II, supra, 130 N.J. at 141-42, 613 A.2d 1059. In the empirical method, we review both defendants who were sentenced to death and those who were not to “identify those characteristics that determine the *324patterns of life sentencing versus death sentencing.” DiFrisco III, supra, 142 N.J. at 164, 662 A.2d 442; see also Martini II, supra, 139 N.J. at 24, 651 A.2d 949; Bey IV, supra, 137 N.J. at 345, 645 A.2d 685; Marshall II, supra, 130 N.J. at 142-43, 613 A.2d 1059.
' We decline once again to attempt to define in advance all characteristics of a murder ease that would capture the critical facts of a particular defendant’s case. DiFrisco III, supra, 142 N.J. at 164, 662 A.2d 442. Such an exercise “would fail [both] to distinguish between individual defendants” and to recognize the uniqueness of every ease. Martini II, supra, 139 N.J. at 24-25, 651 A.2d 949. Also, as in previous cases, we include in the death-sentenced universe those cases where the defendant’s death sentence was reversed on appeal and where the prosecutor chose not to proceed capitally on remand. DiFrisco III, supra, 142 N.J. at 164, 662 A.2d 442; Martini II, supra, 139 N.J. at 25-26, 651 A.2d 949; Bey IV, supra, 137 N.J. at 345-48, 645 A.2d 685; Marshall II, supra, 130 N.J. at 194 n. 10, 613 A.2d 1059. Although reversed for a variety of reasons, we have found that “the original penalty trials ... [still] refleet[ ] juror values of deathworthiness.” Marshall II, supra, 130 N.J. at 194 n. 10, 613 A.2d 1059.
We continue to analyze the data in two sets, one including defendant and one excluding him. “[Proportionality review is a search for community values, and the case under review is a partial reflection of ... those values.” Martini II, supra, 139 N.J. at 27-28, 651 A.2d 949. An analysis which utilizes both sets of data is therefore more complete but does not attempt to conceal its own limitations, i.e., “the bias produced by including defendant’s case.” Id. at 28, 651 A.2d 949.
C. Comparison of Cases
Once we establish the universe of cases on which we will rely and the criteria for coding those cases, we “next group those cases according to their comparative levels of blameworthiness.” Ibid. We measure blameworthiness according to statutory *325aggravating and mitigating factors as well as “nonstatutory factors based on ‘objectively-verified measures of blameworthiness.’” Bey IV, supra, 137 N.J. at 350, 645 A.2d 685 (quoting Marshall II, supra, 130 N.J. at 145, 613 A.2d 1059). As explained earlier, we evaluate these factors using two approaches: frequency analysis and precedent-seeking review, placing our reliance on precedent-seeking review in deciding disproportionality.
1. Comparison Group
The AOC characterizes the various death-eligible cases on the basis of the aggravating factors present in those cases. The similar cases to be examined in proportionality review are then selected on the basis of the aggravating factors present in defendant’s case. DiFrisco III, supra, 142 N.J. at 185, 662 A.2d 442 (citing Martini II, supra, 139 N.J. at 49, 651 A.2d 949). There are thirteen basic categories,15 each of which contains two to seven subcategories.
*326Here, the penalty-phase jury found that the State had proven beyond a reasonable doubt three statutory aggravating factors: N.J.S.A. 2C:ll-3e(4)(a) (prior murder); N.J.S.A 2C:H-3e(4)(f) (escape detection); and N.J.S.A. 2C:ll-3c(4)(g) (felony murder). The AOC therefore coded defendant’s case as a B, or a murder by a defendant with a prior murder conviction. Further, the AOC placed Loftin’s case in the B(l) subcategory16 because, in addition to the prior murder conviction, the case presented two other aggravating circumstances. Loftin Report, tbl.7A.
Defendant argues that his case should be compared not only to all death-eligible B(l) cases, but also to all other death-eligible B cases and to all death-eligible A, multiple-murder, cases in which the defendant killed during the course of a robbery. Defendant asserts that certain A cases are similar to his case in that all A defendants also killed more than one person and some committed murders in the course of a robbery. The State argues that defendant’s case should be compared only to other prior murder cases, as in the proportionality review in Bey IV, supra, 137 N.J. at 367, 645 A.2d 685, another case involving a prior murderer.
Defendant’s “essential attribute” is his prior murder conviction. Because of the “exceedingly small number of cases” in the B(l) subcategory, DiFrisco III, supra, 142 N.J. at 170, 662 A.2d 442, *327we will compare defendant’s case to all death-eligible cases in the B, or prior murder, category. We will not, however; compare defendant’s case, as he has proposed, to certain category A cases — murders involving multiple victims where the murders were committed in the course of a robbery. “We defer generally to the AOC’s expertise, and particularly to its unique assignment of defendants to only one comparison category.” Id. at 167, 662 A.2d 442. In this case, the AOC designated defendant’s case a B, or prior murder, case; we see no compelling reason to disagree with this designation or to deviate from it by including in defendant’s comparison group certain of the A, or multiple-murder, cases.
Loftin’s comparison group thus consists of sixteen defendants: George Booker, John Fauntenberry, Richard Feaster, James Koedatich, Marko Bey, Richard Biegenwald, Bryan Coyle, Samuel Erazo, William Godette, Frank Pennington, Braynard Purnell, Thomas Ramseur, Carlos Vasquez, Jihad Muhammed, Alberto Nieves, and Thomas Williams. We glean the facts of these comparison cases from opinions previously published by this Court and from the AOC’s Detailed Narrative Summaries.
2. Frequency Analysis
We have described frequency analysis earlier in this opinion. To summarize, this approach has heretofore consisted of three different statistical tests designed to gauge a defendant’s relative criminal culpability: the salient-factors test, the numerical-preponderanee-of-aggravating-and-mitigating-factors test, and the index-of-outcomes test. DiFrisco III, supra, 142 N.J. at 171, 662 A.2d 442; Martini II, supra, 139 N.J. at 29-30, 651 A.2d 949; Bey IV, supra, 137 N.J. at 350-51, 645 A.2d 685; Marshall II, supra, 130 N.J. at 154, 613 A.2d 1059. As discussed supra at 294-295, 724 A.2d at 149-150, based on our experience over the six-year period that we have conducted proportionality reviews, we have determined that the numerical-preponderance test does not assist our efforts and should be abandoned. We will confine frequency *328analysis in this case, and in cases to follow, to the salient-factors and index-of-outcomes tests.
a. Salient-Factors Test
In the salient-factors test, we consider the sentences imposed in faetually-similar cases to measure the relative frequency of the imposition of the death sentence. “Its purpose is to help us determine whether the death sentence is imposed-in a category of comparable cases often enough to create confidence in the existence of a societal consensus that death is the appropriate remedy.” Martini II, supra, 139 N.J. at 33, 651 A.2d 949. Cases are first categorized according to the presence of statutory aggravating factors and then placed into subcategories “according to circumstances that serve either to aggravate or to mitigate the blameworthiness of the defendants in those cases.” Ibid. This test has been consistently viewed as the most persuasive of the frequency tests. DiFrisco III, supra, 142 N.J. at 173, 662 A.2d 442; Martini II, supra, 139 N.J. at 33, 651 A.2d 949; Bey IV, supra, 137 N.J. at 353, 645 A.2d 685; Marshall II, supra, 130 N.J. at 168, 613 A.2d 1059.
In this case, the most salient factor is that Loftin had been convicted of a prior murder. Loftin, supra, 287 N.J.Super. at 89, 670 A.2d 557. Loftin is placed in category B for his prior murder, and in the B(l) subcategory because his ease presents two additional aggravating factors. Of the seven death-eligible cases in the B(l) subeategory, five went to penalty trial and, of those penalty-trial cases, two resulted in death sentences, including defendant’s case. Loftin Report, supra, tbl.TA. Thus, the death sentencing rate for all B(l) murderers is twenty-nine percent and, for those advancing to penalty trial, forty percent. Those figures are significantly higher than the overall death-sentencing rates of thirteen percent for all death-eligible cases and thirty-one percent for all penalty-phase cases. Id., tbls.l, 2. Moreover, B(l) murderers are much more likely to advance to the penalty phase than the general death-eligible murderer. Id., tbls.l, 2, 7. In tabular form, the figures are:
*329Death-Sentencing Rates for B(l)* Murderers and All Death-Eligible Murderers
Death-Sentencing Rate at Penalty Trial_ Death-Sentencing Rate for All Death-Eligible Cases Proportion of Cases Advancing to Penalty Trial
B-l inch D .40 (2/5) .29 (2/7) .71 (5/7)
B-l exel. D .25 (1/4) .17 (1/6) .67 (4/6)
All Ds .31 (47/154) .13 (47/369) .42 (154/369)
All Ds but D .30 (46/153) .125 (46/368) .42 (153/368)
*B(1) — prior murder, two additional aggravating factors.
Removing defendant’s case from the group lowers the rates somewhat, but still places him at 12.5% and 30% within the general overall range of thirteen percent for all death-eligible cases and thirty-one percent for all penalty-phase cases.
For the entire prior-murderer category, the figures are as follows:
Death-Sentencing Rates for All B Prior Murderers and All Death-Eligible Murderers
Death-Sentencing Rate at Penalty . Trial_ Death-Sentencing Rate for All Death-Eligible Cases Proportion of Cases Advancing to Penalty Trial
B inch D .59 (10/17) .38 (10/26) .65 (17/26)
B exel. D .56 (9/16) .36 (9/25) .64 (16/25)
All Ds .31 (47/154) .13 (47/369) .42 (154/ 369)
All Ds but D .30 (46/153) .125 (46/368) .42 (153/368)
Of the twenty-six death-eligible cases in category B, seventeen, or sixty-five percent, went to penalty trial and, of those penalty-trial cases, ten resulted in death sentences. Id., tbl.7. The death-sentencing rate for all death-eligible prior murderers is thirty-eight percent, while the rate is fifty-nine percent for those advancing to penalty trial. Ibid. Once again, these figures are significantly higher than the rates for other types of murderers. Ibid.
In sum, application of the salient-factors test appears to suggest that prior murderers receive the death penalty more frequently than most defendants in the death-eligible and penalty-trial universes and that they are considered by community consensus to be highly blameworthy. In fact, prior murderers are sentenced to death at a higher rate than any other category of murderers. See ibid. While we are mindful that the small sample sizes prevent us from relying on the results of this test, Martini II, supra, 139 *330N.J. at 38, 651 A.2d 949, they do support a finding of no disproportionality.
b. Index-of-Outcomes Test
As explained supra at 295, 724 A.2d at 150, in the index-of-outcomes test, “we compare cases that are factually dissimilar but that are nevertheless comparable from the perspective of the defendants’ blameworthiness.” Martini II, supra, 139 N.J. at 42, 651 A.2d 949. More specifically, we seek to identify those factors, both statutory and nonstatutory, that inform jury and prosecutorial determinations of blameworthiness. The AOC calculates the weight of each factor by performing a multiple-regression analysis on the cases within the proportionality review universe. DiFrisco III, supra, 142 N.J. at 178, 662 A.2d 442. This calculation permits us to “compare the blameworthiness of different defendants by the statistically-relevant measures of culpability found in the circumstances of their cases.”. Martini II, supra, 139 N.J. at 42, 651 A.2d 949 (citing Bey IV, supra, 137 N.J. at 362, 645 A.2d 685).
In the Loftin Report, the AOC has prepared tables listing death-sentencing rates that measure deathworthiness based on case characteristics deemed important by prosecutors and jurors. Loftin Report, technical app. 9, at 1. The tables sort the cases into five levels of culpability which reflect predicted probabilities of the return of a death sentence, “cut[ting] the cases at each 20-percentage points of increasing probability of a death sentence.” Id. at 5. “From those groups, we derive the actual probabilities of death sentencing for eases in the various levels of culpability.” Martini II, supra, 139 N.J. at 43, 651 A.2d 949.
The AOC once again urges us to exercise caution in relying on the results of its regression analyses because, given the small pool of cases and the large number of factors assessed, “the culpability estimate ... is often still too soft” to permit “substantive rebanee.” McCarthy Memorandum, supra, at 3-4. A significant difficulty is that the confidence intervals we have seen so far are extremely broad. When a quantity in a population is being *331estimated on the basis of a sample drawn from that population, the estimate is sometimes expressed in a range, called a confidence interval or probability range. Suppose for example that we wished to estimate the average height for the American adult male population on the basis of a random sample. Suppose that this results in a ninety-five percent confidence interval [range] of 69-75 inches. This means that we can be ninety-five percent certain that the true population average lies within that range. Conley & Peterson, supra, at 1223. In Loftin’s case, “we are ninety-five percent certain that all defendants with characteristics similar to [Loftin] will have a predicted probability of receiving a death sentence” within the range. DiFrisco III, supra, 142 N.J. at 180, 662 A.2d 442. The problem is that when a defendant’s probability of receiving a death sentence ranges from zero to ninety-five percent, as does Loftin’s predicted probability, we are ninety-five percent certain that he could have a zero percent, or a ninety-five percent, or an anywhere in-between probability of being sentenced to death. Coincidentally, this is the widest possible range for a confidence interval. Such numbers have little usefulness. We treat the results accordingly. Id. at 179, 662 A.2d 442; Martini II, supra, 139 N.J. at 43, 651 A.2d 949.
The model results indicate that, considering both statutory and nonstatutory factors in penalty-trial cases, Loftin has a predicted probability of fourteen percent of receiving a death penalty, with a probability range that has a lower limit of zero and an upper limit of ninety-five percent. Defendant’s case falls within culpability level one (0-<20% culpability), which contains seventy-four cases, and has an overall death-sentencing rate of seven percent (5/74). Loftin Report, tbls.21, 22.
Using the same tables, a comparison between Loftin and our prior proportionality review defendants reveals that Loftin’s rates are quite low, see Table, infra, at 113-14, but this alone does not compel a finding of disproportionality. As in all four of our prior cases, the expansive confidence interval in Loftin’s case, from zero to ninety-five percent, tells us that we can vest Loftin’s numbers *332with little significance. See DiFrisco III, supra, 142 N.J. at 180, 662 A.2d 442 (“[T]he smaller the confidence interval, the more reliable the predicted probability.”).
Considering both statutory and nonstatutory factors in all death-eligible cases, the predicted probability of a death sentence in Loftin’s case is thirty-eight percent, with a probability range that has a lower limit of nine percent and an upper limit of eighty-one percent. Defendant’s predicted probability places him in culpability level two (20-<40% culpability), which contains thirty-seven cases, and has an overall death-sentencing rate of thirty-eight percent (14/37). Loftin Report, tbls.13,14.
Once again, a comparison to prior proportionality review defendants is helpful. See Table, infra, at 113-14. While Loftin’s rates are only slightly higher than Bey’s and DiFrisco’s, they are far higher than those of Martini and Marshall. Because the confidence intervals are still very large, we are also unable to rely on these figures; they do not, however, demonstrate disproportionality.
Restricting the analysis to statutory factors in penalty-trial cases, Loftin has a predicted probability of sixty-seven percent of receiving a death sentence, within a probability range extending from twenty-two to ninety-three percent. Defendant’s case falls within culpability level four (60-<80% culpability), which contains twenty cases. The overall death-sentencing rate for these cases is seventy percent (14/20). Loftin Report, tbl.23, 24.
, Turning once more to our prior proportionality review defendants, see Table, infra, at 113-14, we see that Loftin’s figures are slightly lower than Bey’s and greater than those of DiFrisco, Martini and Marshall. This portion of the index-of-outcomes test also does not demonstrate disproportionality.
Considering statutory factors in all death-eligible cases, the predicted probability of a death sentence in Loftin’s case is forty-seven percent, with a lower limit of fourteen percent and an upper limit of eighty-three percent. This places defendant in culpability *333level three, (40-<60% culpability) which includes fourteen cases and has a death-sentencing rate of fifty-seven percent (8/14). Loftin Report, tbls.23, 25.
In this final group, see Table, infra, at 113-14, Loftin’s figures exceed those generated for each of our prior proportionality ' review defendants and fail to demonstrate disproportionaEty.
Index-of-Outcomes Test Results (data from Loftin Report, tbls. 13-14,21-25)
Statutory & Nonstatutory Factors Statutory Factors Only
Penalty Trial Death-Eligible Penalty Trial Death-Eligible
Loftin
Predicted Probability of Death Sentence .14 .38 .67 .47
Range (.00 to .95) (.09 to .81) (.22 to .93) (.14 to .82)
Culpability Level 1 2 4 3
Death-Sentencing Rate .07 (5/74) .38 (14/37) .70 (14/20) .57 (8/14)
DiFrisco
Predicted Probability of Death Sentence .64 .22 .46 .20
Range (.19 to .93) (.02 to .79) (.08 to .89) (.01 to .84)
Culpability Level 4 2 3 1
Death-Sentencing Rate .53 (9/17) .38 (14/37) .64 (14/22) .05 (14/287)
Martini
Predicted Probability of Death Sentence .80 .09 .26 .12
Range (.34 to .97) (.02 to .34) (.08 to .57) (.04 to .32)
Culpability Level 5 1 2 1
Death-Sentencing Rate .81 (21/26) .05 (14/299) .24 (8/33) .05 (14/287)
Bey
Predicted Probability of Death Sentence .59 .37 .71 .36
Range (.17 to .91) (.11 to .74) (.37 to .91) (.15 to .65)
Culpability Level 3 2 4 2
Death-Sentencing Rate .54 (7/13) .38 (14/37) .70 (14/20) .26 (13/50)
*334Marshall
Predicted Probability of Death Sentence .65 .17 .31
Range (.16 to .95) (.01 to .89) (.11 to .93) (.04 to .84)
Culpability Level 4 2
Death-Sentencing (13/50) Rate .53 (9/17) .05 (14/299) .64 (14/22) .26
The index-of-outcomes results in defendant’s case are not consistently higher or lower than results generated for prior cases and “do not indicate disproportionality or any aberration” in Loftin’s death sentence. Martini II, supra, 139 N.J. at 45, 651 A.2d 949. As we observed in our review of the first table in this test and thereafter, however, “the small sample size of cases with similar levels of blameworthiness and the great ranges in the confidence intervals,” DiFrisco III, supra, 142 N.J. at 182, 662 A.2d 442, substantially limits the usefulness of those results. That the dissent finds a comparison of the defendant with prior proportionality review defendants unjustified does not surprise us.17 Post at 423, 724 A.2d at 216.
As before, we must continue to rely on precedent-seeking review. DiFrisco III, supra, 142 N.J. at 183, 662 A.2d 442.
*335c. Frequency Analysis Conclusion
Compared to other penalty-trial cases, Loftin’s case shows a death sentencing rate of forty percent under the salient-factors test and, under the index-of-outcomes test, predicted probabilities of receiving a death sentence of sixty-seven percent considering only statutory factors and fourteen percent considering statutory and nonstatutory factors. Loftin Report, tbls.7A, 22, 24. In comparison to all death-eligible cases, Loftin’s salient-factors rate is twenty-nine percent, and his index-of-outcomes probabilities are forty-seven percent considering only statutory factors and thirty-eight percent considering both statutory and nonstatutory factors. Id., tbls.TA, 14, 25. On balance, we find that defendant has not offered, through frequency analysis, evidence of disproportionality and “we do not find that for cases such as his a sentence other than death is generally imposed.” Martini II, supra, 139 N.J. at 46, 651 A.2d 949.
3. Precedent-Seekiny Review
The precedent-seeking approach, also referred to as comparative-culpability review, is the second component of proportionality review. DiFrisco III, supra, 142 N.J. at 183, 662 A.2d 442; Martini II, supra, 139 N.J. at 46, 651 A.2d 949. “Here we engage in traditional ease-by-case review in which we compare similar death-eligible cases, considering the cases individually.” DiFrisco III, supra, 142 N.J. at 183, 662 A.2d 442. In specific, “we examine defendant’s criminal culpability to determine whether it exceeds that of similar life-sentenced' defendants and whether it equals or exceeds that of other death-sentenced defendants.” Id. at 184, 662 A.2d 442. Our primary concern is whether “defendant’s culpability justifies the capital sentence” or whether it “requires a reduction of sentence to a life term.” Martini II, supra, 139 N.J. at 47, 651 A.2d 949. Although identical sentences are not required even in like cases, the defendant must not have been “singled out unfairly for capital punishment.” Ibid, (citing Marshall II, supra, 130 N.J. at 159, 181, 613 A.2d 1059).
*336a. Assessment of Defendant’s Culpability
The first step in comparing defendant’s case to the similar cases is an evaluation of defendant’s own culpability. This assessment requires us to examine three elements: (1) defendant’s moral blameworthiness; (2) the degree of victimization; and (3) defendant’s character. See DiFrisco III, supra, 142 N.J. at 203, 662 A.2d 442; Martini II, supra, 139 N.J. at 74-75, 651 A.2d 949; Bey IV, supra, 137 N.J. at 366, 645 A.2d 685. In determining defendant’s culpability, we consider objective criteria derived from both statutory and nonstatutory aggravating and mitigating factors. Marshall II, supra, 130 N.J. at 159, 613 A.2d 1059. These criteria are “rooted in traditional sentencing guidelines, were clearly presented to the sentencing jury, and are likely to influence a jury’s sentencing decision,” Bey IV, supra, 137 N.J. at 368, 645 A.2d 685. Also, when considering the catch-all mitigating factor, we recognize that defendant’s jury may have rejected a specific mitigating factor, but nevertheless been influenced by the evidence presented in support of that mitigating factor. DiFrisco III, supra, 142 N.J. at 185, 662 A.2d 442; Bey IV, supra, 137 N.J. at 368, 645 A.2d 685.
i. Moral Blameworthiness
To determine defendant’s moral blameworthiness, we examine such characteristics as motive, premeditation, justification or excuse, evidence of mental defect or disturbance, knowledge of helplessness of the victim, defendant’s age or maturity level, and defendant’s involvement in planning the murder. Marshall II, supra, 130 N.J. at 155, 613 A.2d 1059. Defendant’s motive for his murder of Gary Marsh is clear: to steal money from the gas station and, as the jury found, to escape detection and punishment for his crime. Whether the murder of Gary Marsh was premeditated for any length of time is less clear. Although the Public Defender admits that the robbery was carefully planned and premeditated, the fact that a face mask was found in defendant’s car and that defendant drove many miles from home to commit *337the crime may suggest that he originally intended to disguise his identity and thus escape detection in that way. There is no evidence of any justification or excuse for defendant’s crime, nor is there any evidence of provocation on the part of the victim. As we noted in Loftin I, “defendant presented no evidence of a struggle or an accidental discharge ____ [or] that Marsh might have done something to provoke defendant.” 146 N.J. at 353, 680 A.2d 677.
Evidence was presented by the defense psychologist that defendant had a “borderline personality disorder,” and one juror found that the c(5)(a) mitigating factor (extreme mental or emotional disturbance) was present. In addition, three jurors found the catch-all mitigating circumstance that defendant was under the influence of mental and emotional pressure at the time the crime was committed. However, the State’s expert concluded that defendant exhibited narcissistic and antisocial personality traits and not a borderline personality disorder.
Defendant was clearly aware of the helplessness of the victim. Defendant was armed, there was no indication Gary Marsh had any weapon, and they were in an isolated area very late at night. However, because defendant did not know Marsh, and there were no other persons present during the murder, defendant cannot be presumed to have specific knowledge of the effects of the murder on any nondeeedent victims. But see State v. Muhammad, 145 N.J. 23, 46, 678 A.2d 164 (1996) (“While a defendant might be unaware of the specific characteristics of his victim or of the particular survivors that the victim will leave behind, it is completely foreseeable that the killing will eliminate a unique person and destroy a web of familial relationships.”). Although defendant was only twenty-six at the time of the murder, and two jurors found that his age should be considered a mitigating factor, the State correctly points out that defendant lived as an adult with adult responsibilities, i.e., he was married with two children, owned a home, and had served in the United States Navy. Finally, it is clear that defendant was alone in planning the murder and *338that, whether or not the murder itself was premeditated, the underlying felony, the robbery, was meticulously planned ahead of time.
ii. Degree of Victimization
The characteristics to be examined in determining the degree of victimization include the violence and brutality of the murder and the injury to nondecedent victims. Marshall II, supra, 130 N.J. at 155, 613 A.2d 1059. In comparison to other murder cases we have examined, this was not a particularly violent or brutal killing. Aside from the gunshot wound that killed him, there is no evidence of any other type of assault on Marsh by defendant. Indeed, as we noted in Loftin I, the killing was performed almost “execution-style,” 146 N.J. at 387, 680 A.2d 677, in that Marsh was shot in the head “at close range,” id. at 353, 680 A.2d 677. Marsh was, however, undoubtedly terrorized by defendant and in fear for his life from the time he was confronted by defendant until the time he was shot.
iii. Character of Defendant
Finally, in determining the character of defendant we examine defendant’s prior record, whether defendant has committed other unrelated acts of violence, whether defendant cooperated with authorities, defendant’s remorse, and his capacity for rehabilitation. Although, as three jurors found, defendant led a crime-free existence for the first twenty-five years of his life, defendant had committed a prior murder only five weeks before the murder of Gary Marsh. Thus, the jury unanimously found that the c(4)(a) prior murder aggravating factor was present. As we have noted, “a two-time murderer is among the most blameworthy of defendants.” Bey IV, supra, 137 N.J. at 367, 645 A.2d 685.
In addition, the jury heard evidence that defendant had struck his wife and their young children. There is no evidence that defendant cooperated in any way with authorities, although the defense psychologist testified that defendant has the capacity for rehabilitation within a prison setting. Finally, although defendant *339expressed remorse for his crime when he spoke with the defense and the state psychologists, and when he addressed the jury during his allocution, he also wrote that his “mistake” was being incarcerated, not that he had killed two human beings. The jury unanimously rejected the catch-all mitigating circumstance that defendant “has sincere and heartfelt remorse.”
b. Comparison of Defendant’s Case to the B Cases18
Whether defendant’s sentence is disproportionate in comparison with the culpability levels of the comparison group must be determined by comparing the facts of those cases to the facts of defendant’s ease. As discussed above, supra at 324, 724 A.2d at 164, in conducting this comparison, “we identify in the comparison group all relevant aggravating and mitigating factors, both statutory and non-statutory, that are ‘rooted in traditional sentencing guidelines,’ ” DiFrisco III, supra, 142 N.J. at 184-85, 662 A.2d 442. In comparing defendant’s case to the other category B cases, we find that defendant’s culpability level is high, and that his death sentence is not disproportionate. Defendant contends that his “case is more like the life-sentenced cases than the death-sentenced cases.” We disagree.
We compare first the B(l) prior murderers. Each of the four cases in this category has the e(4)(a) (prior murder) aggravating factor, plus two or more additional aggravating factors. And, like Loftin, these four prior murderers have as an additional aggravating factor the c(4)(g) (felony murder) factor. Additionally, Booker, Fauntenberry and Koedatich have the c(4)(f) (escape detection) aggravating factor. In this respect, at least, Loftin shares with these B(l) defendants a similar level of culpability. Yet, of the four defendants in the B(l) comparison group, only Koedatich received a death sentence for his crime, a sentence which was later reduced to life on retrial of the penalty phase. This apparent disparity compels us to take a closer look at the factors which set defendant apart from these B(l) defendants.
*340The greatest distinguishing factor between Loftin and the B(l) defendants is that, with the exception of Fauntenberry, each B(l) defendant presented uncontroverted evidence that he suffered from some mental disease or defect. Bookers’s psychiatrist testified that Booker had an I.Q. of 80 and suspected brain damage, and that he was experiencing a drug-and alcohol-enhanced psychotic episode at the time he committed his murder. Feaster had encephalopathy, an injury to the left frontal lobe region making him more violence-prone. Koedatieh suffered untreated personality disorders as a child as well as untreated suicidal and homicidal tendencies as a teenager, and was diagnosed in prison as a sociopathic personality. In defendant’s case, however, the State disputed defense evidence that Loftin suffered from a borderline personality disorder and introduced evidence that Loftin had a narcissistic or antisocial personality that did not impair his ability to act purposefully. Therefore, even though one juror found the c(5)(a) mitigating factor to exist in Loftin’s case, uncontroverted evidence that other B(l) defendants suffered from a mental disease or defect “could explain why their juries did not deem them to be deathworthy.” Bey IV, supra, 137 N.J. at 384, 645 A.2d 685.
Loftin did present evidence in mitigation that he was deprived of a nurturing mother and a male role model, and that he was traumatized by the fire he set as a child that burned his family home to the ground. This evidence, however, is not as compelling as the evidence of chronic child abuse presented by the B(l) defendants. Booker was regularly beaten by his father and was forced to watch helplessly as his father beat his mother and ten siblings. Fauntenberry was mentally and physically abused by his many step-fathers and witnessed the physical abuse of his mother and the sexual abuse of his sister, acts that ultimately led to Fauntenberry’s own sexual abuse of his sister. Feaster’s parents were both alcoholics and his father was also abusive. Koedatieh suffered severe, physical harm from his parents and was abandoned by them to his grandparents who neglected him. The absence of such evidence in Loftin’s case, together with the fact *341that his claim of mental disease or defect was disputed, distinguishes his ease from the B(l) comparison cases.
Moreover, other considerations serve to distinguish Loftin from the life-sentenced B(l) defendants, Booker, Fauntenberry and Feaster.19 Booker and Fauntenberry confessed, a fact that very' well may have led to the jury’s decision to sentence Booker to life and the prosecutor’s decision to accept Fauntenberry’s plea to noncapital murder. Loftin did not confess; he offered a confession in exchange for the promise of a lesser sentence. Feaster, also, did not confess, but he was only twenty-two-years old at the time of his offense, living with his parents and unemployed. Although Loftin was only twenty-six at the time of his murder, and two jurors found the c(5)(c) mitigating factor to exist, because Loftin was living an adult life — he had trained to serve in the Navy, was married with two children, and owned a home — the jury may have reasonably chosen to hold Loftin to a higher, adult standard.
We turn to the B(2) and B(3) defendants. Of the nine defendants in the B(2) category, seven received a death sentence from their first penalty-phase jury or, in Biegenwald, (1), from his first two penalty-phase juries. Of these seven death sentences, only Bey’s was ultimately upheld and found to be proportionate. Two B(2) defendants, Godette and Vasquez, were permitted to plead to felony murder and received life sentences. The four B(3) defendants were all sentenced to life. The following discussion reveals that Loftin’s culpability equals or exceeds that of Bey, if not that of the B(2) defendants whose death sentences were later reduced to life, and exceeds that of the life-sentenced B(2) and B(3) defendants.
We note that the B(2) and B(3) defendants as a group are less culpable than the B(l) defendants because, in addition to having a *342prior murder conviction, their cases present one or no additional aggravating factor. For this reason alone, Loftin, a B(l) defendant himself, is more culpable than the B(2) and particularly the B(3) defendants.
We first consider Bey, the only defendant currently on death row for a B(2) murder. Bey’s murder of Carol Peniston was intensely violent, including sexual assault, beating, stomping and strangulation. In this respect, at least, his culpability exceeds that of Loftin whose only assault on Gary Marsh was a single gun shot to the head. In other respects, however, Loftin is more culpable than Bey. Bey testified that he was drunk and high on marijuana at the time of the Peniston murder; Loftin offered no justification or excuse for his crime. Bey offered uncontroverted evidence that he had an antisocial personality disorder and lacked the ability to control his anger; although Loftin claimed that he suffered from a borderline personality disorder, that claim was sharply contested by the State. And, finally, Bey offered evidence that his mother, an alcoholic, severely abused him and that his dependence on alcohol and drugs began as early as age nine; Loftin offered no comparable evidence of childhood or substance abuse. Based on these factors, we are convinced that Loftin’s culpability is equal to or greater than that of Bey.
We next consider the B(2) defendants who were initially sentenced to death: Biegenwald, Coyle, Erazo, Pennington, Purnell and Ramseur. Biegenwald’s “thrill” motive was horrific; Coyle, Erazo, Purnell and Ramseur committed violent murders in great excess of what was necessary to end life; and Pennington killed a woman in front of her daughter. When compared to these aspects of the B(2) defendants’ murders, Loftin’s pecuniary-motive, execution-style murder, which created no risk to third persons, appears less deathworthy. This limited comparison, however, fails to take into account the aspects of these defendants’ cases, absent from Loftin’s case, suggesting that Loftin’s death sentence was not an aberration. In particular, all but one of these cases involved *343provocation, to varying degrees, uncontroverted evidence of mental disease or defect, and/or evidence of severe childhood abuse.
Biegenwald was abused as a child by his father, who was an alcoholic mental patient, and was institutionalized at age five, when he was diagnosed as a child schizophrenic and given some twenty electro-convulsive shock treatments. Expert testimony also established that he had an antisocial personality with paranoid traits and that he lacked the emotional capacity to conform his behavior to law. The penalty-phase jury in Coyle’s case found that his victim had participated in the conduct which resulted in his death. Although the jury discarded the c(5)(a) mitigating factor, evidence was also presented that Coyle suffered from latent schizophrenia and substance-dependency, and was laboring under an extreme mental and emotional disturbance at the time of the murder. Both of Erazo’s penalty-phase juries found that he suffered from an extreme mental and emotional disturbance, that he was under unusual and substantial duress at the time of the murder, and that his victim participated in the conduct which resulted in her death. Moreover, evidence at his retrial established that he suffered from a fractured skull caused by child abuse, seizures, and chronic low-level depression. Pennington .produced compelling evidence that he was expelled from a school for emotionally disturbed children for being too disturbed. He was honorably discharged from his Vietnam duty with the Marines due to his mental condition and diagnosed with post-traumatic stress syndrome, subsequently acquitted of murder by reason, of insanity, and diagnosed with a personality disorder and schizophrenia. Ramseur presented evidence of alcoholism, as well as paranoia and psychomotor seizures that combined could lead to violence under provoking circumstances; this evidence no doubt persuaded the jury to find the c(5)(a) and c(5)(d) mitigating factors in his case. When compared to these aspects of the defendants’ cases, Loftin appears more deathworthy. In sum, Loftin is at least as culpable as this group of B(2) defendants who were initially sentenced to death.
*344This analysis, however, does not account for Purnell (1A and IB), who, like Loftin had a monetary motive. The fact that Purnell’s penalty-phase jury, unlike Loftin’s, failed to find the c(4)(f) (escape detection) aggravating factor differentiates Purnell from Loftin. More important, the jury in Purnell’s case found that Purnell’s victim solicited the conduct that resulted in his death — a mitigating factor clearly missing in Loftin’s case.
Turning to the life-sentenced B(2) and B(3) cases, unusual circumstances in the B(2) Godette and Vasquez cases help explain why their prosecutors accepted pleas. Godette can be distinguished from Loftin in that he volunteered a confession and was sentenced to life imprisonment with a minimum term of thirty years, to run concurrently with the term of life imprisonment plus forty years that he was already serving for murder in North Carolina. In Vasquez’s case, as pointed out by the State, the prosecutor reached an agreement to spare the thirteen-year-old victim’s family the trauma of returning from their native Puerto Rico to relive the incident. Further, Vasquez agreed that the sentences on the charges of felony murder and aggravated sexual assault would be the maximum allowed and would be consecutively served, and that any merger argument would be waived. As such, the plea assured that the forty-five-year-old Vasquez would not be eligible for parole until age eighty-five.
The B(3) defendants, all of whom were sentenced to life, are distinguishable for a variety of reasons. Unlike Loftin, none of the B(3) defendants killed for money but, rather, out of fear, anger or both. Biegenwald, Nieves and Williams each presented some evidence of provocation. Biegenwald broke up a struggle between his friend and the victim, Ward, who was clutching a gun, by shooting Ward. The penalty-phase jury in Nieves’s case found that the victim solicited the conduct which led to his death. And it appears that Williams killed his mother after an argument over money and his girlfriend. Loftin presented no such evidence of provocation. See Loftin I, supra, 146 N.J. at 353, 680 A.2d 677. Biegenwald, Muhammed and Nieves also presented uncontrovert*345ed testimony of mental disease or defect and/or substantial mitigating circumstances. Biegenwald suffered child abuse, childhood institutionalization with shock treatments, and schizophrenia as a child, and antisocial personality with paranoid traits as an adult. Muhummed was diagnosed as a sociopath. Nieves and his seventeen siblings grew up in abject poverty, he was beaten by his alcoholic father, one of his brothers was murdered, and another brother went to prison for avenging that murder. As discussed above, Loftin failed to produce uncontroverted evidence of mental disease or defect or comparable evidence of child abuse. These distinctions, in the aggregate, support the juries’ determinations, or the prosecutor’s in Muhummed’s case, that the B(3) defendants, as compared to Loftin, were not deathworthy.
As in Marshall II and Bey IV, “we do not [here] find a pattern of life sentencing or the taint of an invidious factor that would require us to reverse [defendant’s] death sentence.” Ibid.
D. Other Arguments
Defendant argues that because death sentences are not generally imposed for the crime of capital murder, imposing the death penalty upon him violates the Eighth and Fourteenth Amendments to the United States Constitution and Article I, Paragraphs 1 and 12 of the New Jersey Constitution, and that the geographic distribution of capital charging and sentencing decisions reveals inconsistent, unfair and arbitrary application of the death penalty, again in violation of the federal and state constitutions. We have previously considered and rejected these arguments and we continue to do so here. See DiFrisco III, supra, 142 N.J. at 210, 662 A.2d 442; Martini II, supra, 139 N.J. at 79-80, 651 A.2d 949; Bey IV, supra, 137 N.J. at 396, 645 A.2d 685; Marshall II, supra, 130 N.J. at 188-215, 613 A.2d 1059.
IV
Conclusion
We have in our prior proportionality review cases repeatedly concluded by acknowledging the limitations of our statistical com*346parisons. Today we have moved to reexamine the methods we use in order to improve and simplify the review process, and to better understand the effect of the legislative restriction on the proportionality review universe. We seek to carry out our appellate review function in capital causes in such manner that every defendant will be ensured a rigorous and complete review of his or her sentence of death.
We conclude that Loftin has not met his burden of proving that his death sentence is disproportionate or that race has operated as an impermissible factor in the imposition of the death penalty. Accordingly, defendant’s sentence of death is affirmed.
APPENDIX A
[[Image here]]
*347[[Image here]]
*348[[Image here]]
APPENDIX C
Comparison Case Summaries
I. Prior Murderers with Two Additional Aggravating Factors or Particular Violence/Terror: B(l)
A) George Booker (1 and 2)
On September 11, 1985, George Booker was asked to leave the home of friends, a married couple with whom he had been living. He had been having an affair with the wife. Booker went to the home of a female friend where he sexually assaulted her at knife point. He then took her car and ran down a male pedestrian, stopped to steal the ailing pedestrian’s wallet, and drove away. On September 13, 1985, police officers responded to a suspicious-*349person call and canvassed the neighborhood. They entered the home of an elderly woman when they discovered the door to the home ajar, and saw Booker heading down the stairs toward them with a knife in hand. After refusing to drop the knife and attempting to grab one of the officer’s guns, Booker was wrestled to the ground and arrested.
In response to questioning, Booker told police that he had killed two women, but that it might have been just a dream. On September 14, 1985, however,' the bodies of two women were discovered in their home. One victim was tied up and gagged, and had a cord around her neck. Her mouth and forehead had been bashed in. The cause of death was suffocation, strangulation, or head injuries. The other victim was also gagged and suffered knife cuts on her chest, abdomen, throat and over her left eye. There was a defense wound on her right hand. The cause of death was multiple stabbings, twice in the throat and once in the back. The State theorized that Booker had raped, sodomized and killed the first victim before the second victim got home. When the second victim arrived, Booker forced her to undress and lie next to the first victim’s body on the bed. He then handcuffed the second victim to the first victim’s body before stabbing the second victim to death.
Booker claimed diminished capacity at trial, asserting that he was drunk and high on numerous narcotics. At the time of his crime spree, Booker was thirty-six years old, an alcoholic and a high school drop-out. One of eleven children born to sharecropper parents, Booker’s father regularly beat him, his siblings and his mother. Although Booker never received psychiatric care, such care had been recommended for him in the past and a psychiatrist for the defense testified that Booker was experiencing a psychotic episode, exacerbated by his consumption of alcohol and drugs, at the time of the offenses. The psychiatrist also testified that Booker had an I.Q. of eighty and that mental tests suggested brain damage. His extensive criminal record includes a conviction *350for murder in 1972, two robberies, and terroristic threats, aggravated assault, and disorderly persons offenses.
The jury convicted Booker of two counts of knowing murder, aggravated sexual assault and possession of a weapon for an unlawful purpose. For the murder of the first victim, the jury found four aggravating factors: c(4)(a) (prior murder), c(4)(c) (extreme suffering/torture), c(4)(g) (felony murder). For the murder of the second victim, the jury found three aggravating factors: c(4)(a) (prior murder), c(4)(c) (extreme suffering), and c(4)(f) (escape detection). For both murders, at least one juror found the e(5)(a) (extreme mental and emotional disturbance) and e(5)(h) (catch-all) mitigating factors. Because the jury could not reach a unanimous decision regarding the weighing of the factors for either murder, Booker was sentenced to two life terms, each with a mandatory minimum of thirty years, to run consecutively. The aggravated sexual assault merged with the murders and, on the weapons offense, Booker was sentenced to a concurrent ten-year term.
B) John Fauntenberry
On February 1, 1991, John Fauntenberry found himself at a New Jersey truck stop without money. While trying to sell some of his personal belongings to finance his trip home to Ohio, he met a male truck driver who solicited him for sex. Fauntenberry accompanied the driver back to his truck, shot him once in the head with a .22 caliber pistol, stole some of his money and property, and left the truck stop. In March 1991, Fauntenberry was arrested in Juneau, Alaska for murder. While in custody, he confessed to five other murders: three in Oregon, one in Ohio, and the truck stop murder in New Jersey. In his confession, Fauntenberry admitted that he killed his victims so as not to leave witnesses.
Fauntenberry was twenty-seven years old at the time of the offense, had graduated from high school, and was living with his aunt in Ohio. He had been employed as a truck driver but was *351unemployed when he committed the murders. He had also previously served in the Navy and was discharged due to heavy drinking. In addition to alcohol, Fauntenberry admitted abusing cocaine and LSD. As a child, he was mentally and physically abused by a string of step-fathers, witnessed the physical abuse of his mother and sister, and witnessed the sexual abuse of his sister. Fauntenberry himself sexually abused his sister. Prior convictions included aggravated assault, carrying a concealed weapon and theft. At the time of the disposition in this case, Fauntenberry had prior murder convictions in Alaska, for which he received a life sentence, and Ohio, for which he was sentenced to death. Charges were also pending for two of the three murders he admitted committing in Oregon.
Fauntenberry was charged with capital murder, felony murder, robbery, possession of a weapon for an unlawful purpose and unlawful possession of a weapon. The state served a notice of factors, including three aggravating factors: c(4)(a) (prior murder), c(4)(f) (escape detection), and c(4)(g) (felony murder). Thereafter, on September 8, 1993, Fauntenberry pled guilty to noncapital murder, for which he was sentenced to life with a mandatory minimum of thirty years, and robbery, for which he received a concurrent twenty-year sentence, both terms to run consecutively to the sentence he was already serving in Alaska.
C) Richard Feaster (2)
On November 1, 1993, the telephone company received a call from a man claiming to have been stabbed. Central dispatch determined the number and location of the caller and notified the police. Police officers arrived at a local gas station to find a male, dead and lying in a pool of blood just outside the station’s office door. The victim’s front pocket had been turned inside out and blood-stained money was found on the floor. An autopsy later determined that the victim had been stabbed and slashed in excess of forty times about the head, neck and hands.
*352On November 3, 1993, a woman contacted the Prosecutor’s office and identified Richard Feaster as the person responsible for two area homicides. She claimed to have heard him confess to the first murder in a bar. A male informant also reported that on October 31, 1993, Feaster had asked the informant to accompany him to pick up his pay-check. While at an intersection, Feaster looked at a gas station and said, “Good, there’s only one.” Feast-er then parked the car and told the man to sit in the driver’s seat. About twenty minutes later, when Feaster returned, he was covered with blood, his hand was bleeding, and he was holding a bloody knife and some cash. The man drove to his home with Feaster where Feaster put his bloody clothes and boots in a plastic bag and threw the knife behind the house.
Based on this information, the police executed a search warrant at the informant’s home and recovered the bloody clothes and knife. The police also executed a warrant at Feaster’s home, where they recovered a butcher block knife set with one knife missing. Further investigation revealed that an area hospital treated Feaster for lacerations on his hand on November 1, 1993. A jailhouse informant later reported that Feaster told him that he had killed to feel the thrill.
At the time of the offense, Feaster was twenty-two years old and lived with his parents and older sister. He had graduated from high school and had previously worked in construction, but was unemployed at the time of the murder. Feaster had no physical, mental or drug dependency problems although he had prior convictions for possession of marijuana and simple assault.
Feaster was charged with capital murder, felony murder, robbery and possession of a weapon for an unlawful purpose. On April 1, 1996, he pled guilty to noncapital murder, for which he was later sentenced to life imprisonment with a minimum term of thirty years, and robbery, for which he received a consecutive twenty-year sentence with a minimum term of ten years, both terms to run consecutive to his sentence for the other murder. *353The AOC coded the following aggravating factors: c(4)(a) (prior murder) and c(4)(g) (felony murder); and mitigating factor c(5)(h) (catch-all).
Before the disposition in this case, Feaster was convicted of capital murder, felony murder, robbery, possession of a weapon for an unlawful purpose and possession of a sawed-off shotgun. In Feaster (1) he was sentenced to death for the October 6, 1993, murder of a gas station attendant, classified by the AOC as E(5). The following mitigating evidence was presented: both of Feast-er’s parents were alcoholics; his father was abusive; he had a history of alcohol and cocaine abuse; he was using alcohol before and after the murder; and he suffers from encephalopathy, an injury to the left frontal lobe region, making him more violent than people who do not have this condition.
D) James Koedatich (1A)
On November 23, 1982, James Koedatich kidnapped eighteen-year-old Amie Hoffman from the shopping mall where she was employed. He sexually assaulted and stabbed her several times. Defensive wounds were found on Hoffman’s hands and her left ear had been severed. Hoffman bled to death several hours after the abduction. Her body was found floating face down in a water-retention tank in a wooded and secluded area. Almost two months later, Koedatich was arrested. He was convicted when evidence was discovered linking him to the murder after he reported to the police that he had been stabbed by a person pretending to be a police officer.20
At the time of the offense, Koedatich was thirty-four years old and lived with his mother. Having dropped out of high school, he had achieved a GED while in a Florida prison and had held jobs as a superintendent and as a gas station attendant. In 1971, he was *354involved in an armed robbery in Florida and, before the case went to trial, he and a friend choked a man to death. For these two crimes, he was sentenced to serve concurrent twenty-year terms, of which he served approximately ten years. Psychological evaluations performed while Koedatich was in prison indicated that he had a sociopathic personality.
Koedatich was charged with capital murder, felony murder, kidnapping, aggravated sexual assault, unlawful possession of a weapon and possession of a weapon for an unlawful purpose. He was convicted on all counts. Although he did not permit his counsel to present any mitigating evidence at the penalty phase trial, the court submitted the c(5)(h) (catch-all) mitigating factor to the jury. The jury found two aggravating factors present: c(4)(a) (prior murder) and c(4)(g) (felony murder). No mitigating factors were found and the jury sentenced defendant to death. On the other counts, the court sentenced him to serve a consecutive, aggregate term of thirty years with a minimum term of fifteen years. The Court reversed Koedatich’s death sentence because of improper instructions. State v. Koedatich, 112 N.J. 225, 548 A.2d 939 (1988), cert. denied, 488 U.S. 1017, 109 S.Ct. 813, 102 L.Ed.2d 803 (1989).
E) James Koedatich (IB)
On retrial of the penalty phase, a psychologist for the defense testified that, as a child, Koedatich suffered severe physical abuse at the hands of his parents, was abandoned by his parents, and forced to sleep nightly in a car by his grandparents. He developed severe personality problems that were left untreated. As a teenager, he had suicidal and homicidal tendencies that were also untreated.
The jury found all four aggravating factors, c(4)(a) (prior murder), c(4)(c) (outrageously and wantonly vile), c(4)(f) (escape detection) and e(4)(g) (felony murder), as well as the c(5)(h) (catch-all) mitigating factor. Because the jury could not unanimously agree *355on the weighing of the factors, Koedatich was sentenced to life in prison.
II. Prior Murderers with One Additional Aggravating Factor or Particular Violence/Terror: B(2)
A) Marko Bey (2B)
On April 26, 1983, Marko Bey accosted Carol Peniston in front of her apartment building intending to rob her. When he heard someone else approaching, he pulled Peniston into a nearby shed, sexually assaulted her, beat her, stomped on her chest, and strangled her. Bey stole eight dollars and her car keys from Peniston. He crashed and abandoned Peniston’s car in his flight from the scene.
After his arrest, Bey confessed and was charged with murder, felony murder, kidnapping, aggravated assault, aggravated sexual assault, robbery and theft. At trial, he testified that he was drunk and high on marijuana at the time of the murder. He explained that he had killed Peniston because he had become scared when he saw her looking at him as he rifled through her poeketbook. Bey expressed some remorse by acknowledging that the murder should never have happened.
Bey was convicted and sentenced to death. The penalty phase jury found two aggravating factors, e(4)(c) (outrageously and wantonly vile) and c(4)(g) (felony murder), and no mitigating factors. This Court affirmed the conviction, but reversed the death sentence because the trial judge erred in charging the jury that mitigating factors must be found unanimously. State v. Bey, 112 N.J. 123, 548 A.2d 887 (1988) (Bey II).
On retrial of the penalty phase, the State alleged two aggravating factors: c(4)(a) (prior murder), and c(4)(g) (felony murder). As to the prior murder factor, defendant had been sentenced to death for the rape-murder of Cheryl Alston, which occurred approximately three weeks before the Peniston murder. The Court reversed his death sentence after concluding that the death-*356penalty statute does not permit the execution of minors. State v. Bey, 112 N.J. 45, 548 A.2d 846 (1988) (Bey I). Defendant had turned eighteen in the interim between the Alston and Peniston murders.
Bey alleged the existence of four mitigating factors: c(5)(a) (extreme mental or emotional disturbance), c(5)(c)(age), c(5)(d) (mental disease or defect or intoxication), and e(5)(h) (catch-all). In support of the mitigating factors, Bey produced fresh evidence. His mother drank excessively, and severely abused and neglected Bey and his siblings. His father rejected him. Bey began drinking at age nine and using drugs, particularly marijuana, at age eleven. He had been hospitalized twice for overdosing. Eighteen-years-old at the time of the murder, Bey had dropped out of school in junior high and was unemployed. Medical experts testified that Bey suffered from organic brain damage, a frontal lobe impairment caused by in útero exposure to alcohol, preadolescent consumption of drugs and alcohol, and head injuries. Bey also suffered from an organic personality disorder and lacked the ability to control his anger.
The jury found two aggravating factors, c(4)(a) (prior murder) and c(4)(g) (felony murder), and at least one juror found two mitigating factors, c(5)(a) (extreme mental or emotional disturbance) and c(5)(h) (catch-all). The jury unanimously determined that the aggravating factors outweighed the mitigating factors and sentenced Bey to death. The Court affirmed the death sentence, State v. Bey, 129 N.J. 557, 610 A.2d 814 (1992), cert. denied, 513 U.S. 1164, 115 S.Ct. 1131, 130 L.Ed.2d 1093 (1995) (Bey III), and found it to be proportionate, Bey IV, supra, 137 N.J. at 339, 645 A.2d 685.
B) Richard Biegenwald (1A)
On August 27, 1982, Richard Biegenwald approached eighteen-year-old Anna Olesiewicz on the Asbury Park Boardwalk and offered her marijuana. Olesiewicz accompanied Biegenwald to his house. Biegenwald intended to have his roommate and protégé, *357Theresa Smith, kill Olesiewiez so that she could understand the thrill of killing. However, when he was unable to rouse Smith, Biegenwald shot Olesiewiez four times in the head. The next day, after coaxing Smith to touch the victim’s body and tell him how it felt, Biegenwald and another friend dumped the body in a vacant lot behind a fast food restaurant. On January 14, 1983, Olesiewiez’s skeleton was discovered. One week later, Smith went to the police and implicated Biegenwald in the murder. Police subsequently recovered several guns, narcotics, and a ring belonging to Olesiewiez from Biegenwald’s house. Biegenwald was arrested and charged in four different murders, including the murder of Olesiewiez.
Biegenwald was verbally and physically abused as a child. He was institutionalized at the age of eight, diagnosed as schizophrenic, and given at least twenty electro-convulsive shock treatments. He later entered a state hospital after trying to set himself on fire. On his release, he was beaten again by his father, stole from his mother, and frequently ran away from home for days at a time. At age eighteen, Biegenwald was convicted of a murder committed while robbing a store, for which he served seventeen or eighteen years in prison. At the time of the Olesiewiez murder, he had been married for two years, had a five-month-old daughter and was employed as a construction worker. He denied abusing alcohol but admitted smoking marijuana.
A forensic psychiatrist for the defense testified that Biegenwald suffered from a severe personality disorder known as antisocial personality with paranoid traits. Although he did not find Biegenwald to be legally insane, the psychiatrist testified that Biegenwald lacked the emotional capacity to appreciate the wrongfulness of his act or to conform his behavior to the law.
The jury found two aggravating factors, c(4)(a) (prior murder) and c(4)(c) (outrageously and wantonly vile), and two mitigating factors, c(5)(d) (mental disease or defect) and e(5)(h) (catch-all). The jury determined that the mitigating factors did not outweigh *358the aggravating factors and sentenced Biegenwald to death. On appeal, the Court reversed the death sentence because of the inadequacy of the jury charge regarding the weighing of aggravating and mitigating factors. State v. Biegenwald, 106 N.J. 13, 524 A.2d 130 (1987) (Biegenwald I).
C) Richard Biegenwald (IB)
On retrial of the penalty phase of the above offense, a jury again found the c(4)(a) and c(4)(c) aggravating factors and the c(5)(d) and c(5)(h) mitigating factors. Defendant was again sentenced to death, and this Court reversed that sentence because of inadequate voir dire. State v. Biegenwald, 126 N.J. 1, 594 A.2d 172 (1991) (Biegenwald III).
D) Richard Biegenwald (1C)
On the second retrial of the penalty phase, Biegenwald presented additional testimony from his mother, a forensic social worker, and his prison doctor, who testified that Biegenwald had been treated for high cholesterol, diabetes and lung cancer. The jury found two aggravating factors, c(4)(a) (prior murder) and c(4)(c) (depravity), and three separate c(5)(h) mitigating factors. The jury determined that the aggravating factors did not outweigh the mitigating factors and sentenced Biegenwald to life imprisonment.
E) Bryan Coyle (1A)
Brian Coyle became friendly with his neighbors, Rhonda and Seth Lemberg, and began having an affair with Mrs. Lemberg. On July 28, 1993, as was customary, the three sat drinking on the Lemberg porch until the early hours of the morning. After Coyle returned home, the Lembergs got into an argument. Mrs. Lemberg went to Coyle’s house. Soon after, Mr. Lemberg went to Coyle’s house and broke the front window when no one answered the door. Coyle, who had taken mescaline prior to Mrs. Lemberg’s arrival, retrieved his gun and fired a warning shot into the floor. Mr. Lemberg fled and called the police.
Coyle and Mrs. Lemberg then attempted to drive off. Mr. Lemberg, who may have had a gun, blocked the way. Mrs. *359Lemberg got out of Coyle’s car and ran, and Mr. Lemberg chased her. After he caught up with her and after another argument, Mr. Lemberg went back into his house. However, he soon reappeared and started chasing his wife again. Coyle opened fire on Mr. Lemberg, firing several times as Mr. Lemberg ran away. Coyle chased Mr. Lemberg behind a tree and fired three more shots. Two hit Mr. Lemberg in the shoulder and head, killing him. Coyle fled to South Carolina but was later apprehended.
Coyle pled non vult to a murder charge in 1975, received a ten-to-fifteen-year sentence, and was paroled in 1983. He also had two prior convictions for robbery. Although he claimed to have no drug or alcohol problems, Coyle had been diagnosed in the past as being dependent on marijuana, barbiturates and alcohol. He was also diagnosed as being a latent schizophrenic and as having an extreme mental and emotional disturbance.
Coyle was charged with capital murder and convicted. The jury found two aggravating factors, c(4)(a) (prior murder) and c(4)(c) (outrageously or wantonly vile), to be present. Although the jury found the c(5)(b) (victim participated in conduct that resulted in his death) mitigating factor, it determined that the aggravating factors outweighed the mitigating factor and sentenced Coyle to death. The Court reversed defendant’s death sentence and conviction based on the absence of a jury instruction on Coyle’s intent to cause death, State v. Gerald, 113 N.J. 40, 69, 549 A.2d 792 (1988), deficiencies in the jury instructions, and improper admission of evidence in the guilt phase. State v. Coyle, 119 N.J. 194, 574 A.2d 951 (1990).
F) Bryan Coyle (IB)
On October 7, 1991, after the reversal of his conviction and death sentence, Coyle pleaded guilty to noncapital murder and was sentenced to life,. with a minimum term of thirty years.
G) Samuel Erazo (1A)
On July 20, 1986, Samuel Erazo stabbed his wife to death after an evening of drinking and quarreling. Erazo and the victim were *360married when he was incarcerated at Rahway State Prison for killing the daughter of the woman with whom he had been living in 1974. On the night of the murder, Erazo and his wife were both intoxicated; the victim’s blood alcohol level was found to be .195. At some point that evening, Erazo drove some friends home and his wife went out for a walk. When she returned home sometime after midnight, Erazo stabbed her eight times. The State theorized that Erazo’s wife had purposefully cut her hand during her walk and then threatened to call the police to tell them that Erazo had inflicted the wound and thereby violated his parole. Erazo maintained that he killed his wife in the heat of passion.
Erazo was charged with capital murder. At the penalty phase, Erazo offered testimony from his sister, and from a corrections officer who said that he was a “model inmate.” In addition, he offered into evidence a letter he had written in which he said that he had proven capable of leading a useful and productive life in prison. While previously imprisoned, Erazo acquired a GED, took vocational courses, completed a three-year correspondence course, and received recognition for his work on a toy drive for underprivileged children and on the prison representative committee.
The jury found the c(4)(a) (prior murder) and c(4)(c) (extreme suffering/torture) aggravating factors as well as the c(5)(a) (extreme mental or emotional disturbance), c(5)(b) (victim participated in offense), c(5)(d) (intoxication), and c(5)(e) (duress) mitigating factors. The jury found that the aggravating factors outweighed the mitigating factors and sentenced defendant to death. On appeal, the Court reversed defendant’s conviction and death sentence based on improper jury instructions at the guilt phase. State v. Erazo, 126 N.J. 112, 594 A.2d 232 (1991). The Court also held that the c(4)(c) factor could not be considered on remand. Id. at 137-39, 594 A.2d 232.
H) Samuel Erazo (IB)
On retrial, Erazo was again convicted of capital murder. The State served notice of the c(4)(a) (prior murder) aggravating factor *361and Erazo served notice of the following mitigating factors: c(5)(a) (extreme mental or emotional disturbance), c(5)(b) (victim solicited, participated in, or consented to conduct resulting in her death), c(5)(e) (duress), and thirteen separate factors falling under c(5)(h) (catch-all).
In addition to the mitigating evidence supplied during the first penalty trial, an expert testified that Erazo had a fractured skull due to child abuse by his grandmother. He had been treated for seizures and may have been brain damaged as a result of his multiple suicide attempts. Erazo also suffered from chronic low-level depression. The sum of these factors, the expert opined, led to Erazo’s loss of control at the time of the offense.
The jury found the c(4)(a) aggravating factor and each of the submitted mitigating factors, including twelve of the thirteen e(5)(h) factors, but was unable to reach a verdict on sentencing. The court sentenced Erazo to life imprisonment with a minimum term of thirty years.
I) William Godette
William Godette worked as a handyman for the victim, a seventy-nine-year-old man. On November 14,1990, when Godette went to the victim’s home and demanded payment for work he had done, the victim refused to pay him. That night, at approximately 6:30 p.m., Godette returned to the victim’s home, knocked on his door and demanded payment. When the victim again refused to pay, Godette forced his way into the home, knocking the victim to the ground. The victim then called Godette a name. According to Godette, the name-calling caused him to “panic.” Godette first tried to strangle the victim and then struck him six times in the head with a hammer. Godette took some money, jewelry and a VCR from the home and fled. The victim’s body was discovered the next day by fire department personnel who had been notified by coworkers concerned because the victim did not show up for work.
*362On August 8,1991, Godette was arrested in North Carolina and charged with robbery and the murder of his step-father. On February 14, 1992, he volunteered a confession detailing his involvement in that murder. Godette was later convicted of both robbery and murder in connection with his step-father’s death, and was sentenced to a term of life imprisonment plus forty years.
At the time of the murder of the elderly New Jersey man, Godette was thirty-seven-years old. He had dropped out of high school and had held a variety of unskilled jobs. He was separated from his wife and was the father of three children. He admitted using “all drugs” and claimed to have spent two weeks in a hospital detoxification program. His only other prior conviction was for the possession of controlled dangerous substances.
Godette was charged with capital murder, felony murder, unlawful possession of a weapon and possession of a weapon for an unlawful purpose. On February 11, 1993, he pled guilty to robbery and felony murder. The robbery conviction merged for sentencing purposes and he was sentenced to life imprisonment, with a minimum term of thirty years, to be served concurrent with the term he was serving in North Carolina. Although the AOC codes only the c(4)(a) (prior murder) aggravating factor as being present, the State served notice of the following additional aggravating factors: c(4)(c) (depravity), c(4)(f) (escape detection), and c(4)(g) (felony murder). The AOC also codes mitigating factors c(5)(a) (emotional disturbance), c(5)(b) (victim participation), and e(5)(h) (any other relevant factor).
J) Frank Pennington (1A)
Frank Pennington arrived at an East Rutherford bar at about 11:30 p.m. on September 2, 1986. About one-half hour later, Arlene Connors, the victim and owner of the bar, arrived to help her daughter Pam close up. When Arlene announced that it was closing time, Pennington asked for and was served a fourth beer before leaving. At this point, the parties’ versions vary. According to Pam, she heard her mother say to Pennington, “It’s the *363bewitching hour,” followed by Pennington’s response, “Bewitch this.” Thereafter, Pam heard a lot of commotion and breaking glass, and turned to see her mother leaning on the bar. In his sworn statement to the police, however, Pennington claimed that he pulled a gun out with the sole intention of robbing the bar. He ducked to avoid a glass thrown by Ms. Connors “and as he straightened up, he pulled the trigger of his gun.” It is undisputed that Pennington fired a single shot, striking Ms. Connors in the heart and killing her instantly.
Pennington was born out of wedlock and had no relationship with his father. His mother was promiscuous, mentally-challenged, bad-tempered, and had married six times. One of his stepfathers was an abusive alcoholic. Pennington attended a school for emotionally-disturbed children, but was expelled after trying to kill a teacher who had slapped him. He committed ten juvenile offenses and was incarcerated at a juvenile institution at least once. At age nineteen, Pennington enlisted in the Marines and served in Vietnam disarming mines. In 1970, he was honorably discharged after having been diagnosed as an emotionally unstable personality with secondary depressed reaction. A year later, he was admitted to a psychiatric hospital and diagnosed as suffering from post-traumatic stress syndrome. Pennington was acquitted by reason of insanity for armed robbery in 1972 and pled guilty to first-degree murder in 1974. He has since attempted suicide, been diagnosed as suffering from a personality disorder, schizophrenia and depression, and has become an alcoholic.
Pennington was charged with capital murder, felony murder and a weapons offense, and convicted on all counts. The jury found two aggravating factors, c(4)(a) (prior murder) and c(4)(g) (felony murder), and the c(5)(d) (mental disease or defect) mitigating factor. The aggravating factors were found to outweigh the mitigating factors and Pennington was sentenced to death. The Court reversed the capital murder conviction and death sentence because the trial court failed to give a Gerald'charge to the jury, see Gerald, supra, 113 N.J. 40, 549 A.2d 792, but affirmed the *364felony murder conviction, State v. Pennington, 119 N.J. 547, 575 A.2d 816 (1990).
K) Frank Pennington (IB)
After the reversal, the State decided not to retry Pennington for capital murder and, on October 25, 1991, he was sentenced on the felony murder count to life imprisonment with a minimum term of thirty years.
L) Braynard Purnell (1A)
On August 26, 1988, Braynard Purnell killed Lawrence Talley during the course of an attempted drug transaction. Purnell and the State’s key witness, Marie Simmons, frequently used drugs together. On the day of the murder, Purnell went to Simmons’s house to buy a small amount of cocaine. Simmons went to a neighborhood playground and met Talley to negotiate the drug purchase. They could not agree on a price and Talley later sent a confederate, Jeffrey Davis, to Simmons’s house to sell a $20 bag of cocaine. Purnell refused to buy the drugs because he wanted a larger quantity. Purnell then went to the playground himself to talk with Talley. Purnell and Talley walked back to Purnell’s house where they argued about the drugs and eventually started to fight. Defendant stabbed the victim fifteen times in the neck, chest and abdomen, stole his drugs, and hid the body.
At the time of the offense, Purnell lived with his daughter from his first marriage, his fiancé and her two children. He had dropped out of high school and served in the army for one year. He was employed and had held multiple unskilled positions in the past. According to his mother, various other relatives, and his fiancé, he had substantially improved upon release from his first prison term. They testified that he was religious and active in the church, and that he not only avoided drugs, but took pains to warn others about the dangers of drug use. He had previous convictions for second-degree murder and shoplifting.
Defendant was ■ charged with capital murder, two counts of hindering apprehension or prosecution, possession of a weapon for *365an unlawful purpose, and perjury, and was convicted on all counts, save one count of hindering apprehension. The jury found two aggravating factors, c(4)(a) (prior murder) and c(4)(g) (felony murder), and two mitigating factors, c(5)(b) (victim participated in conduct that resulted in his death) and c(5)(h) (catch-all). The aggravating factors were found to outweigh the mitigating factors and Purnell was sentenced to death. On the remaining charges, he was sentenced to a consecutive, aggregate term of ten years with a minimum term of two years. This Court affirmed Purnell’s murder conviction, but reversed his death sentence because the trial court failed to instruct the jury on the possibility of convicting Purnell of the lesser charge of felony murder. State v. Purnell, 126 N.J. 518, 601 A.2d 175 (1992).
M) Braynard Purnell (IB)
The State chose not to retry the guilt phase, which would have been necessary in order to seek the death penalty on remand, and Purnell was sentenced to life imprisonment with a minimum term of thirty years.
N) Thomas Ramseur
Thomas Ramseur and Asaline Stokes, his former girlfriend, had a history of arguments and violent encounters. Stokes lived with her grandchild across the street from Ramseur’s aunt’s home. Ramseur physically attacked Stokes and frequently threatened to kill her, her children and her grandchildren. On August 24, 1982, Ramseur and Stokes had a heated argument, which ended with Ramseur threatening, ‘You’ll be sorry,” taking a knife from Stokes’s kitchen, and leaving.
The next day, Stokes was speaking with a mechanic who was working on a truck near her home when Ramseur, who had watched the exchange, came out of his aunt’s house. He approached Stokes, patted her on the shoulder, and then repeatedly stabbed her. He began to walk away but returned to inflict more wounds and to taunt the dying woman by saying, “If I see your kids again I’m going to kill them too.”
*366A police officer who was driving through the area came on the ' scene and arrested Ramseur. Stokes did not die immediately; she dropped into unconsciousness only after reaching the hospital and died shortly thereafter. She had major stab wounds in the face and chest, including several defensive wounds on her hands and two chest wounds about eight and one-half inches deep that pierced the lung.
At the time of the murder, Ramseur was forty-three-years old. He had graduated from high school and, although unemployed at the time of the offense, had worked as a laborer, assembly-line worker and welder. He was an alcoholic and admitted having had two drinks directly prior to killing Stokes. Ramseur had previously suffered four severe head injuries and had experienced memory loss, migraine headaches, and bleeding in the brain. He also suffered from paranoia and psychomotor seizures, a form of epilepsy, that combined could lead to violence under provoking circumstances. Ramseur had prior convictions for the 1966 murder of his wife, for which he served six-and-one-half years, and for weapon possession and shoplifting.
Ramseur was charged with capital murder, unlawful possession of a weapon and possession of a weapon for an unlawful purpose. He was convicted on all counts. At the penalty phase, the jury found two aggravating factors, c(4)(a) (prior murder) and c(4)(c) (outrageously and wantonly vile), and two mitigating factors, c(5)(a) (extreme mental or emotional disturbance) and c(5)(d) (mental disease or defect). The jury determined that the aggravating factors outweighed the mitigating factors and sentenced Ramseur to death. The Court reversed defendant’s death sentence and imposed a life sentence because of the trial court’s coercive instructions after the jurors had announced they were deadlocked. Ramseur, supra, 106 N.J. 123, 524 A.2d 188.
O) Carlos Vasquez
On June 3, 1988, at approximately 8:00 a.m., Carlos Vasquez enticed a thirteen-year-old girl into his apartment with a discus*367sion of religious matters. Once inside, Vasquez bound the girl’s hands, raped her, and asphyxiated her using a towel. The victim’s hands and feet were then tied together behind her back with electrical cord and clothesline. Vasquez placed the body into yellow plastic garbage bags and then into a cardboard box, and put the box out to be collected with the trash. The cause of death was asphyxia caused by gagging, ligature strangulation, and fracture of the cervical spine.
Vasquez had a 1975 conviction for murder in Puerto Rico for which he served seven years before being paroled in 1982. After moving to the United States from Puerto Rico in 1987, he was employed as a construction worker. He denied any physical or mental health problems or substance dependency.
Vasquez was charged with capital murder, felony murder, kidnapping, aggravated sexual assault and sexual assault, and pled guilty to felony murder and aggravated sexual assault. On the felony murder charge, he was sentenced to life imprisonment with a minimum term of thirty years and, on the sexual assault charge, he was sentenced to serve a consecutive term of twenty years with a minimum term of ten years. The AOC coded the case as having two aggravating factors, e(4)(a) (prior murder) and c(4)(g) (felony murder), and the c(5)(h) (catch-all) mitigating factor present.
III. Prior Murderers with No Other Aggravating Circumstances or Particular Violence/Terror: B(3)
A) Richard Biegenwald (2)
On September 21, 1982, Biegenwald and the State’s principal witness, Dherren Fitzgerald, met with William Ward to arrange the terms of a “hit” which Ward wanted Fitzgerald to perform for $25,000. According to Fitzgerald, he joined Ward in Ward’s car and Biegenwald followed them in Fitzgerald’s car to Fitzgerald’s home. Before Biegenwald arrived, Fitzgerald and Ward discussed the terms of the “hit.” Fitzgerald refused Ward’s request to be present for the murder and an argument ensued, during *368which Ward displayed his revolver. The men wrestled over the gun and the gun went off, striking Ward in either the shoulder or neck. Fitzgerald reached for his Luger .22 caliber pistol but was unable to cock the gun with one hand and instead struck Ward on the head with its barrel, causing Ward to bleed and rendering the gun inoperable. The struggle spilled onto the porch, ending with Fitzgerald on top of Ward, who was on his back and still clutching his gun. Then, Biegenwald appeared and shot Ward five times in the head with a .22 caliber Beretta.
Biegenwald and Fitzgerald loaded the body into Ward’s car, which Fitzgerald then drove to a nearby mall. Later, Biegenwald reported that the police had not come to the scene. Fitzgerald drove back to his home, where he stored Ward’s body until that night when he buried Ward at Mount Calvary Cemetery in Sea View Square. The two men later abandoned Ward’s car in Brooklyn.
Biegenwald was convicted of capital murder. The jury found one aggravating factor, c(4)(a) (prior murder), and two mitigating factors, e(5)(d) (mental disease of defect) and e(5)(h) (catch-all), but was unable to reach a verdict on sentencing. Therefore, the court sentenced Biegenwald to life with a minimum term of thirty years. He was also convicted of theft, possession of a weapon for an unlawful purpose, and unlicensed possession of a handgun, for which he was sentenced to serve a consecutive, aggregate term of five-and-one-half years. In an unreported opinion, the Appellate Division affirmed.
B) Jihad Muhammed
On August 3, 1984, at approximately 1:00 p.m., Jihad Mu-hammed, thirty-two-years old, approached an eighteen-year-old male and his girlfriend standing outside the girl’s home, and offered to sell them some speed. The girl went inside and told her mother, who instructed her to tell Muhammed that she was not interested. When the girl returned, Muhammed yelled to someone across the street, “Come on, this guy don’t want nobody *369messin’ around with his lady,” and left. He reappeared about twenty minutes later, with a friend, Forrest Boyer, and a handgun. Muhammed pointed the gun at the couple and warned the girl, “Don’t move — you move and I’ll shoot you first.” He then fired a shot into the ground. Muhammed tried to force the male to buy drugs from him but the male refused. Meanwhile, Boyer took the girl’s pocketbook, rifled through it, and emptied its contents into a car. When the girl’s boyfriend told Boyer to return the pocketbook, an argument ensued and Muhammed took out a sawed-off shotgun and shot the male. Having heard the shots, the girl’s father came outside and asked Muhammed why he had shot the boy. Muhammed responded, “I didn’t like his attitude. You’d better get in the house or I’ll shoot you too____ I want to shoot me a white boy.” More than three weeks later, the boy died in the hospital trauma unit from the gunshot wound.
At the time of the offense, Muhammed lived with his wife and had, for the previous four years, worked as a houseman at a hotel. He had only a fifth grade education and was institutionalized as a juvenile and diagnosed as a sociopath. At age ten he was convicted of larceny and put on probation. Since then, he was arrested six additional times and convicted four, including a 1971 conviction for murder and a 1975 murder charge that was pleaded out to atrocious assault.
Muhammed was charged with, among other things, murder and felony murder. He pled guilty to felony murder and was sentenced to life imprisonment with a minimum term of thirty-six years. He also pled guilty to possession of a weapon for an unlawful purpose, possession of a rifle or shotgun, and certain persons not to have weapons, for which he was sentenced to a consecutive, aggregate term of twelve years. The AOC coded the case as having the c(4)(a) (prior murder) aggravating factor and the c(5)(h) (catch-all) mitigating factor.
*370C) Alberto Nieves
On March 25, 1987, Alberto Nieves was exiting a grocery store when he heard Hector Rentas, Sr., sound his automobile horn at Nieves’s wife who was seated outside in the Nieves’s ear. Rentas stopped his ear to speak to Nieves, whereupon Nieves proceeded to his car, reached for a gun under his car seat, and approached Rentas. Nieves pointed the gun at Rentas’s head and told him that if he “wanted the girl to come in and get the girl and keep her.” He held the gun to Rentas’s head for about thirty seconds, lowered it, and returned to his car.
Three days later, on March 28,1987, Rentas drove his six-year-old son to the grocery store and was parking nearby when Nieves approached the automobile. Nieves again told Rentas to “Stop messing around with my girl,” whereupon Rentas replied, “I’m not messing with your girl.” Nieves then raised his gun, shot Rentas once in the head, and fled the scene. The bullet passed through Rentas’s head, killing him, and lodged in the seat between him and his son.
At the time of the offense, Nieves was twenty-six-years old. He was separated from his wife, had three children by three different mothers, had dropped out of school upon completing seventh grade, and had previously worked as a packer. One of eighteen children, he had grown up in poverty, without adequate food or clothing, and had been beaten by his alcoholic father. One of his siblings had been murdered and another was serving jail time for avenging that murder. Nieves was not substance-dependent or mentally ill. In 1980, he was convicted of murder, aggravated assault and possession of a weapon for an unlawful purpose, for which he served six years of a twenty-five-year sentence; he was on parole at the time of the Rentas murder.
Nieves was convicted of capital murder, possession of a handgun, possession of a weapon for an unlawful purpose, and hindering apprehension or prosecution. In the penalty phase, the jury found the c(4)(a) (prior murder) aggravating factor and two miti*371gating factors, c(5)(b) (provocation) and c(5)(h) (catch-all). The jury was unable to reach a decision regarding the weighing of the factors and the court sentenced Nieves to an aggregate term of life imprisonment with a minimum term of thirty-two-and-one-half years.
D) Thomas Williams
Thomas Williams and his mother got into an argument because Williams was living on her resources and wanted his girlfriend to spend the night. His mother may also have learned that he was renting a basement apartment to neighborhood prostitutes. During this argument or soon thereafter, Williams stabbed his mother nine times in the chest and heart and then left her body beneath a blanket in the basement of her apartment building. Some weeks later, on August 20, 1993, Williams notified the police that he had found his mother murdered. When the officers arrived, Williams was calm and unconcerned. The police soon discovered that Williams had attempted to cash some of his mother’s checks and that he had filed a false report claiming that his mother’s car had been stolen. He had actually given the car to a juvenile in exchange for a twenty-dollar bag of cocaine.
At the time of the offense, Williams was thirty-eight-years old and lived with his mother. He had a ninth-grade education and was self-employed. He admitted using cocaine and has since sought treatment. He claimed to be in otherwise good physical and mental health. Williams had prior convictions for murder, atrocious assault and battery, unlawful possession of a weapon and possession of a controlled dangerous substance. The prior murder conviction stemmed from an altercation with his girlfriend and her landlord in which Williams shot both, killing the landlord.
Williams was convicted of noncapital murder and sentenced to serve a life sentence, with a minimum term of thirty years.
The Connecticut Supreme Court continues to conduct proportionality review for capital cases that were pending at the time the state repealing statute became effective. See Webb, supra, 680 A.2d at 200 n. 71; 1995 Conn. Legis. Serv. P.A. 95-16 (West).
Although the Connecticut Supreme Court rejected statistical analysis as a means to identify an aberrant death sentence under the state's proportionality review statute, see Webb, supra, 680 A.2d at 209-10 (construing Cown. Gen.Stat. § 53a-46b(b)(3)), the court decided that statistical methods should be used to look for possible systemic racial bias in the imposition of the death penalty under a separate statutory provision, Conn. Gen.Stat. § 53a-46b(b)(l); Cobb, supra, 663 A.2d at 961-62. Unlike the section dealing with proportionality review, section 53a-46b(b)(l) has not been repealed.
In a 1996 luncheon address printed in Seton Hall Law Review, our first Special Master David Baldus spoke on this issue:
A word about complex statistical analyses. They may be helpful in some cases, but they are clearly not essential. Indeed, for the purpose of helping the community dialogue focus on the worst cases, quite rudimentary pictures of the system are sufficient.... Also, principled, well-focused reviews can be conducted strictly with narrative factual summaries of the death sentenced review case and the comparison cases.
[David Baldus, When Symbols Clash: Reflections on the Future of the Comparative Proportionality Review of Death Sentences, 26 Seton Hall L.Rev. 1582, 1604(1996).]
We declined to apply the 1992 statutory amendment in DiFrisco III for reasons that are specific to the procedural history of that case. DiFrisco was first sentenced to death in 1988. DiFrisco III, supra, 142 N.J. at 163, 662 A.2d 442. On direct appeal, this Court affirmed the conviction but reversed the death sentence and remanded for a penalty phase retrial. Ibid. DiFrisco was again sentenced to death in Februaiy 1993, after the 1992 statutory amendment took effect. Ibid. Because "the genesis of ... [the proportionality review] proceeding was defendant’s first conviction, which occurred long before the statute was amended," we held that the amendment would not be applied in DiFrisco III. Ibid.
Defendant committed the offense giving rise to his conviction on May 5, 1992, seven days before L. 1992, c. 5 took effect. Thus, defendant asserts that the law as applied to him violates the Ex Post Facto Clauses of the federal and state constitutions. See U.S. Const, art. I, § 10, cl. 1; N.J. Const, art. IV, § 7, ¶ 3. As we are not restricting the universe of similar cases in conducting defendant's proportionality review, we need not address his ex post facto claims.
This case may be distinguished from Winberry v. Salisbury, 5 N.J. 240, 74 A.2d 406, cert. denied, 340 U.S. 877, 71 S.Ct. 123, 95 L.Ed. 638 (1950), the *284seminal case construing the nature and extent of judicial rulemaking power under Art. VI, § 2, ¶ 3. In Winberry, the Court was faced with a statute and a conflicting court rule purporting to establish the time to appeal a judgment. Id. at 243, 74 A.2d 406. The court rule was adopted pursuant to the Court’s constitutional authority to "make rules governing ... the practice and procedure in all ... [state] courts,” an authority made "subject to the law.” Art. VI, § 2, ¶ 3. The Court interpreted the constitutional qualification "subject to the law” as relating exclusively to "substantive law.” Winberry, supra, 5 N.J. at 247, 74 A.2d 406. Because the challenged legislation purporting to govern the time to appeal was not "substantive,” the Court held that the court rule should prevail. Id. at 255, 74 A.2d 406. In essence, Winberry established that the Court’s rulemaking power, when applied to matters of practice and procedure, is “not subject to overriding legislation." Ibid.
In this case, Winberry's procedural/substantive distinction is not dispositive because the Court’s authority to undertake proportionality review does not arise from the rulemaking power granted to it by Article VI, § 2, ¶ 3. Rather, the Court's authority to undertake proportionality review arises from its general appellate review powers and from its specific power to hear capital causes. See supra at 279-284, 724 A.2d at 142-144; see also NJ. Const, art. VI, § 2, ¶ 2; NJ. Const, art. VI, § 5, ¶ 1; NJ. Const, art. VI, § 5, ¶ 3.
We use the term "individual proportionality review” to describe our efforts "to ensure that the death penalty is being administered in a rational, nonarbitraiy, and evenhanded manner, fairly and with reasonable consistency.” Marshall II, supra, 130 N.J. at 131, 613 A.2d 1059. We use the term "systemic proportionality review" to describe our efforts to monitor the death penalty for the purpose of “preventfing] discrimination on an impermissible basis, including, but not limited to, race and sex." Ramseur, supra, 106 N.J. at 330, 524 A.2d 188.
Multiple-regression analysis is a statistical tool used to describe the relationship between one or more independent variables (e.g., prior murder) and a dependent variable (e.g., the death penalty). See Final Report, supra, at 117. The tool is used in employment discrimination cases when it is claimed that a class of employees has been denied promotion or accorded differential compensation based on factors such as race or gender. Salary would be the dependent variable to be explained whereas gender and race would be the explanatory or independent variables. EEOC v. Sears Roebuck & Co., 839 F.2d 302, 325 (7th Cir.1988).
For a discussion of confidence intervals see infra at 330-331, 724 A.2d at 167-168.
Logistic regression analysis was first employed in the AOC report prepared for State v. Harris, No. A-3-96. Harris Report, tbls.l, 3. That case, however, was not decided by this Court because of defendant's unrelated death. See State v. Loftin, No. A-86-96, slip op. at 2 (Oct. 22, 1996).
A finding is said to be statistically significant if chance, acting alone, probably would not have caused it. In many situations, the probability of the chance occurrence of a relationship or difference ... of a particular magnitude can be calculated and expressed in terms of a p-value. In many research contexts, a relationship or difference will be considered statistically significant if p <.05; that is, if there is less than a one-in-twenty probability that a comparable finding would have emerged as a matter of chance alone.
[John M. Conley & David W. Peterson, The Science of Gatekeeping: The Federal Judicial Center’s New Reference Manual on Scientific Evidence, 74 N.C. L.Rev. 1183, 1209 n. 159 (1996).]
In a letter to the Court dated May 15, 1997, Judge Cohen focused on this problem. He said that because there are so few cases in the database, an *305increase of four new penalty trials can cause the "percentage race difference ... [to] drop by 23%,” a rather dramatic change. Letter from Richard S. Cohen to the Supreme Court 2 (May 15, 1997).
A model containing an excessive number of parameters in relation to the small number of cases studied can lead to a statistical phenomenon known as "overfitting." Special Master Report, supra, at 27. Overfitting "can result in the false attribution of effect to one or more of [the parameters]." Ibid. Judge Cohen opined that the AOC models were subject to overfitting, i.e., "there is such an excess of factors compared to the number of death verdicts as to cripple the system’s predictive capacities.” Id. at 42.
We ask also that the Special Master review the question whether purging, i.e., the removal of the indirect effects of race from variables that appear to be unrelated to race, produces results that are useful. In the opinion of Dr. Tukey, it does not. Tukey Report, supra, at 11.
The thirteen basic categories are as follows:
(A) Multiple-Victim Murder (aggravating factor c(4)(b));
(B) Prior Murder Conviction without (A) above (aggravating factor c(4)(a));
(C) Sexual Assault Murder without (A)-(B) above (aggravating factor c(4)(g));
(D) Victim a Public Servant without (A)-(C) above (aggravating factor c(4)(h));
(E) Robbery Murder without (A)-(D) above (aggravating factor c(4)(g));
(F) Arson Murder without (A)-(E) above (aggravating factor c(4)(g));
(G) Burglary Murder without (A)-(F) above (aggravating factor c(4)(g));
(H) Kidnapping Murder without (A)-(G) above (aggravating factor c(4)(g));
(I) Pecuniary Motive Murder without (A)-(H) above (aggravating factors c(4)(d) and c(4)(e));
(J) Torture/Aggravated Assault Murder without (A)-(I) above (aggravating factor c(4)(c));
(K) Depravity of Mind Murder without (A)-(J) above *326(aggravating factor c(4)(c));
(L) Grave Risk of Death to Another Person as primary statutory aggravating circumstance without (A)-(K) above (aggravating factors c(4)(b) and (c)(4)(g));
(M) Escape Detection Murder without (A)-(L) above (aggravating factor c(4)(f)).
[Loftin Report, tbIs.6-7.]
The subcategories of category (B), murder by a defendant with a prior murder conviction, are as follows:
1. With two or more additional aggravating circumstances or particular violence/terror.
2. With a single additional aggravating circumstance or particular violence/terror.
3. With no other aggravating circumstances or particular violence/terror.
[Loftin Report, tbl.6.]
The dissent also suggests that "[t]he death-sentencing rate statistic ... should primarily guide this Court's interpretation of the index-of-outcomes test results." Post at 423, 724 A.2d at 216. But unless it is independently verified that a defendant's case is actually similar to other cases in his culpability level, there is no reason to focus on the death-sentencing rate for that culpability level. Indeed, as the dissent recognizes, there are "deficiencies in the methodology behind the death-sentencing rate statistic; for instance, defendants in culpability level one with a culpability rating of .05 may differ substantially in death-sentencing rate, despite being in the same level, from those defendants on the upper end of culpability level one, who have culpability level ratings such as .19." Id. at 422-423, 724 A.2d at 216. Most important, the death sentencing rates are generated by the defendants' culpability ratings and suffer from the same infirmities found in the statistics predicting the probability of a death sentence. See supra at 330-331, 724 A.2d at 167-168.
Summaries of the B cases are provided in Appendix C.
Feaster, after pleading guilty to noncapital murder, received a life sentence in Feaster (2). See Appendix C, at 7a. One month earlier, he had been sentenced to death for a prior murder in Feaster (1).
Koedatich was also arrested and convicted for the murder of twenty-five-year-old Deirdre O’Brien for which he received a life sentence. That murder occurred shortly after the Hoffman murder.