(dissenting). I agree with the majority that the Court should reach the substantive issues in this case.1 I respectfully dissent from its conclusion regarding the validity of the rules promulgated by defendant that prohibit insurers from classifying insureds on the basis of their credit records (the OFIS rules). Also, of particular concern to me is the majority’s harmless-error analysis. It is seriously flawed and sets a dangerous precedent for the future.
THE STANDARD OF REVIEW
In Luttrell v Dep’t of Corrections,2 this Court adopted a three-pronged test for analyzing the validity of an administrative agency’s rules:
Where an agency is empowered to make rules, courts employ a three-fold test to determine the validity of the rules it promulgates: (1) whether the rule is within the matter covered by the enabling statute; (2) if so, whether it complies with the underlying legislative intent; and (3) if it *411meets the first two requirements, when [sic] it is neither arbitrary nor capricious.[3]
In In re Complaint of Rovas against SBC Michigan, we held that an administrative agency’s interpretation of statutes is entitled to “respectful consideration” and “should not be overruled without cogent reasons.” 4
THE MAJORITY’S CRITICAL ERRORS
In my view, the majority goes awry in at least five significant ways. First, it misapplies the applicable standards of review. Under the first prong of Luttrell, the proper inquiry is not whether “insurance scoring is permissible under the Insurance Code.”5 The Code says nothing about insurance scoring. The relevant inquiry is whether rules banning the use of insurance scoring in setting insurance rates are within the matters covered by MCL 500.210.6
Second, by considering and rejecting each argument offered in support of the OFIS rules, the majority *412improperly shifts the burden of proof to defendant.7 Third, the majority fails to give “respectful consideration” to defendant’s interpretation of the applicable statutes as In re Rovas Complaint requires.8
Fourth, the majority errs by not confining its review of the record to conform to its “harmless error” analysis. The majority holds that “even if the trial court erred by not limiting its review to the administrative record, the error was harmless because there is ample evidence in that record to support the trial court’s conclusion that insurance scoring is permissible under the Insurance Code.”9 Yet the majority subsequently expands its review by referring to evidence outside that record.
The majority has it backwards. If the circuit court erred by creating its own evidentiary record, its conclusion must be wholly supportable on appeal by evidence in the administrative record. For an error to be considered harmless, the conclusion reached in the case must be supportable notwithstanding the alleged error.10 If the majority deems it necessary, as evidenced by its analysis, to examine both the administrative record and the circuit court record to support its conclusion, the error cannot be “harmless.”* 11
*413After conducting a proper “harmless error” analysis, I reach the opposite conclusion from the majority. Any procedural error was harmless because the evidence in both the administrative record and the circuit court record failed to establish that the OFIS rules are invalid. Thus, my inquiry gives plaintiffs the benefit of every doubt and examines the evidence in both records. The result is that, if a procedural error occurred, it was harmless. By expanding the scope of its review, the majority fails to accord defendant the same treatment, thereby making its harmless-error analysis erroneous.
Fifth and finally, the majority’s overly broad review of the record goes beyond even the administrative and circuit court records. It relies on sources outside any record provided to this Court.12
These errors are crucial to the outcome of the case.13 As the discussion of the merits of plaintiffs’ claims demonstrates, much conflicting evidence exists on whether insurance scoring is predictive of loss. The majority appears willing to overrule defendant’s decision simply because it disagrees with it. However, when *414the proper level of deference is applied, it is irrelevant whether the majority would decide the issue differently. Rather, after examining the conflicting evidence, one can only conclude that defendant did not exceed her authority by promulgating the rules banning the practice.
HOME BUILDERS
Under Home Builders,14 the circuit court indisputably erred by creating and considering an evidentiary record outside of what was created during the rulemaking process. However, despite the fact that the circuit court impermissibly expanded the record in contravention of Home Builders, I do not believe that the error is outcome determinative. Under the proper standard of review, neither the circuit court record nor the administrative record supports the trial court’s conclusion that the OFIS rules are invalid.
After declining to review the administrative record and instead constructing its own record, the circuit court concluded that the OFIS rules were invalid. It did so on the basis of its independent factual conclusion that insurance scores accurately reflect differences in risk. I believe that, by reaching its own factual conclusions and failing to consider the administrative record at all, the circuit court erred.
We long ago held that “courts accord due deference to administrative expertise and [may] not invade the province of exclusive administrative fact-finding by displacing an agency’s choice between two reasonably differing views.”15 Although this holding arose in the context of judicial review of quasi-judicial administrative decisions, I see no basis for limiting it to such cases. Judicial *415review of agency actions implicates significant questions about the separation of powers.16 The Court of Appeals in Home Builders cited ample authority in summarizing this point:
The federal courts generally limit judicial review to the administrative record already in existence, rather than permitting either review de novo or trial de novo. Florida Power & Light Co v Lorion, 470 US 729, 743-744; 105 S Ct 1598; 84 L Ed 2d 643 (1985); Camp v Pitts, 411 US 138, 142; 93 S Ct 1241; 36 L Ed 2d 106 (1973) (“[T]he focal point for judicial review should be the administrative record already in existence, not some new record made initially in the reviewing court.”); Nat’l Treasury Employees Union v Horner, 272 US App DC 81, 89; 854 F2d 490 (1988) (“Stated most simply, our task is to determine . .. whether [the agency] considered the relevant factors and explained the facts and policy concerns on which it relied, and whether those facts have some basis in the record.”); Norwich Eaton Pharmaceuticals, Inc v Bowen, 808 F2d 486, 489 (CA 6, 1987). For example, in Florida Power, supra at 744, the United States Supreme Court stated:
“If the record before the agency does not support the agency action, if the agency has not considered all relevant factors, or if the reviewing court simply cannot evaluate the challenged agency action on the basis of the record before it, the proper course, except in rare circumstances, is to remand to the agency for additional investigation or explanation. The reviewing court is not generally empowered to conduct a de novo inquiry into the matter being reviewed and to reach its own conclusions based on such an inquiry.”[17]
Moreover, agency actions taken in a judicial or quasi-judicial capacity, as contrasted with those taken in a *416quasi-legislative capacity, are subject to a heightened standard of review.18 Thus, quasi-legislative agency actions are afforded greater deference.
Accordingly, the factual findings on which an administrative agency’s rule is based certainly must be considered in reviewing the validity of that rule. To allow a court to make factual findings based solely on a record made in the court would allow the judiciary to substitute its own judgment for that of the agency. Moreover, it would allow the courts to usurp the authority that the Legislature granted to administrative agencies.
THE VALIDITY OF THE OFIS RULES
The first prong of the Luttrell analysis requires plaintiffs to show that the OFIS rules banning insur*417anee scoring are not “within the matter covered” by the Insurance Code.19 In that way, plaintiffs are given the burden of showing that a total ban on insurance scoring is inconsistent with the Code. Because I conclude that the OFIS rules are within the matter covered by the Insurance Code, I conclude that plaintiffs have not met their burden. Plaintiffs cannot demonstrate either that the rules are incompatible with the underlying legislative intent or that they are arbitrary and capricious.
“WITHIN THE MATTER COVERED BY THE ENABLING STATUTE”
When courts apply the first prong of the Luttrell test, the most relevant authorities are the enabling statutes of the Insurance Code. MCL 500.210 defines the scope of the Insurance Commissioner’s regulatory powers. Section 210 provides that
[t]he commissioner shall promulgate rules and regulations in addition to those now specifically provided for by statute as he may deem necessary to effectuate the purposes and to execute and enforce the provisions of the insurance laws of this state in accordance with the provisions of Act No. 88 of the Public Acts of 1943, as amended, being sections 24.71 to 24.80 of the Compiled Laws of 1948, and subject to Act No. 197 of the Public Acts of 1952, as amended, being sections 24.101 to 24.110 of the Compiled Laws of 1948.[20]
MCL 500.210 delegates broad discretionary authority to the Commissioner to promulgate rules “as he may deem necessary” to enforce insurance laws and “effectuate the purposes” of the Insurance Code. Even Judge ZAHRA’s opinion, which would have held the OFIS rules invalid, conceded that “[i]n the broadest sense, the rules *418under review do not offend the first prong of the Luttrell standard.”21 Moreover, as Judge WHITE noted,22 the OFIS rules were promulgated in compliance with the prescribed procedures in the Administrative Procedures Act (APA)23
I agree that, on their face, the OFIS rules are within the broad discretionary authority that the Legislature bestowed on defendant to “effectuate the purposes” of the Insurance Code. The majority reaches the opposite conclusion because, as noted previously, it merely asks whether insurance scoring is “permissible” under the Insurance Code.24
COMPLIANCE WITH THE UNDERLYING LEGISLATIVE INTENT
The title of the Insurance Code provides that the purpose of the Code is “to provide for the continued availability and affordability of automobile insurance and homeowners insurance in this state and to facilitate the purchase of that insurance by all residents of this state at fair and reasonable rates... .”25 Thus, the OFIS rules cannot satisfy this prong of the Luttrell standard if they are contrary to that intent. Ascertaining legislative intent necessitates a close examination of the statutory language of the applicable statutes. I would hold that the OFIS rules do not comply with the Legislature’s intent if Chapters 21, 24, and 26 of the Insurance Code authorize insurance scoring.
CHAPTER 21
It is undisputed that insurance scoring is not included within the enumerated permissible rating fac*419tors in MCL 500.2111. Therefore, using insurance scoring to set insurance rates is only permissible under Chapter 21 if it is a permissible “premium discount plan” under MCL 500.2110a. Affording the required “respectful consideration” to defendant’s interpretation of MCL 500.2110a, I agree that setting rates based on insurance scoring does not constitute a “premium discount plan.”
At the public hearings held before the OFIS rules were adopted, insurers conceded that eliminating insurance scoring would not change the total premiums that they collect from their insureds.26 It follows that the proposed “discount plan” based on insurance scoring does not reflect a belief on the part of the insurers that insurance scoring will reduce its overall losses. If it did, the insurers would surely be forced to increase their premiums to reflect the expected increase in losses that would be incurred once their “discount plans” had been disallowed.
Rather, setting premium rates on the basis of insurance scoring simply reallocates the amount each insured pays on the basis of its insurance score. Defendant uses an example in which an insurer must collect $900 in premiums to pay for its expected losses and expenses. Dividing these losses evenly across Class A, Class B, and Class C, each class of insureds would be charged $300. However, using insurance scoring to predict losses, the insurer charges its highest scoring insureds (Class A) a $200 premium. The insurer charges its less favored policyholders a $300 premium, and its lowest scoring policyholders a $400 premium. *420The insurer still collects $900 in premiums.27 I agree with Judge WHITE that this classification scheme constitutes an unapproved rating factor rather than a discount.28 Therefore, I see no cogent reasons to overrule defendant’s interpretation of MCL 500.2110a.29
CHAPTER 24 AND CHAPTER 26
Chapter 24 and Chapter 26 of the Insurance Code allow insurers greater authority in setting premiums than does Chapter 21.30 MCL 500.2403(1)31 and *421MCL 500.2603(1)32 provide the limitation on setting rates at issue here.
Plaintiffs argue that under MCL 500.2426 and MCL 500.2626, defendant may not disapprove rates that are based on insurance scores because insurance scoring is actuarially sound. Defendant counters that insurance scoring is not a “reasonable classification system” under MCL 500.2403(l)(d) and MCL 500.2603(l)(d). A “reasonable classification system” is defined as
a system designed to group individuals or risks with similar characteristics into rating classifications which are likely to identify significant differences in mean anticipated losses *422or expenses, or both, between the groups, as determined by sound actuarial principles and by actual and credible loss and expense statistics or, in the case of new coverages or classifications, by reasonably anticipated loss and expense experience.[33]
Thus, defendant claims that rates set based on insurance scores are “unfairly discriminatory” because they are not “likely to identify significant differences in mean anticipated losses or expenses.”
An examination of both the circuit court record and the administrative record reveals the reasonableness of defendant’s conclusion that rates based on insurance scores are unfairly discriminatory. First, the accuracy of credit reports, on which insurance scores are based, is unclear. The GAO report cited by the majority concluded that “a comprehensive assessment of overall credit report accuracy using currently available information is not possible.”34 As defendant noted, the evidence on this point is inconclusive.35
The majority appears to concede that the reliability of credit reports is subject to question. Yet it proceeds by effectively requiring defendant, rather than plaintiffs, to show that “the unreliability [in credit scores] would have to result in a ‘differential between the rates’ that ‘is not reasonably justified ....’ ”36
*423I would conclude the contrary and hold that the uncertainty surrounding the accuracy of credit reports is evidence per se that a classification system based on those reports is unreasonable. It should be plaintiffs’ burden to rebut this conclusion by producing evidence that such a classification is reasonable. To do so, plaintiffs would need to demonstrate that classifying persons on the basis of insurance scores is “likely to identify significant differences in mean anticipated losses or expenses.”37
The record simply does not establish that credit scores correlate with the risk of loss in a way that makes insurance scoring a “reasonable classification system” under MCL 500.2403(l)(d) and MCL 500.2603(l)(d). Defendant reasonably rejected some of the studies submitted at the public hearings in opposition to the OFIS rules because they used “univariate analysis”38 and analyzed data from states other than Michigan.39 The only study not conducted by plaintiffs that included data on Michigan automobile policies, which plaintiffs cited often as supporting their position, showed “a total *424lack of correlation.”40 Moreover, defendant.noted that “the agency is not aware of any study at all. . . that *425includes data on Michigan home policies.”41 The majority entirely ignores these findings and picks and chooses from among the available data to independently consider whether a classification system based on credit scores is reasonable.42
However, the circuit court record provides little that undermines defendant’s factual findings made at the public hearings. Plaintiffs continued to rely heavily on the studies that defendant reasonably rejected. The new evidence introduced in the circuit court consisted primarily of affidavits from various insurance industry executives. These cite statistics that purportedly show a correlation between credit scores and risk. While generally supporting plaintiffs’ position, the affidavits are *426insufficient to rebut defendant’s conclusion that the use of insurance scoring to set rates is not a “reasonable classification system.” The statistical data in the affidavits, like the studies in the administrative record, are based on a univariate analysis. For reasons cited previously, it was not unreasonable for defendant to reject this analysis.43 Finally, I do not address the majority’s discussion of sources outside the administrative and circuit court records because the majority improperly relies on them.44 Reference to statutes that are not applicable to this case may be appropriate when discerning the proper interpretation of a statute; however, it is not warranted simply as a means of bolstering the evidence that is on the record.
As with Chapter 21, defendant’s interpretation of the applicable statutory provisions in Chapters 24 and 26 is entitled to “respectful consideration” under In re Rovas Complaint. Because setting rates using insurance scoring is not clearly permissible under any chapter, I conclude that the OFIS rules do not violate the legislative intent behind the Insurance Code.
ARBITRARY AND CAPRICIOUS
The majority concludes that it need not decide whether the OFIS rules are arbitrary and capricious because “they are not ‘within the matter covered by the enabling statute.’ ”45 Given that I disagree with the majority’s conclusion on the latter point, I am com*427pelled to also address the former issue regarding whether the OFIS rules are arbitrary and capricious.
“A rule is not arbitrary or capricious if it is rationally related to the purpose of the enabling act.”46 For the reasons stated previously, I conclude that the OFIS rules are rationally related to the purpose of the Insurance Code: to provide for continued availability and affordability of insurance in this state and to facilitate the purchase of that insurance by all residents at fair and reasonable rates.
DUE PROCESS
Plaintiffs also argue that the OFIS rules are invalid because they deprive them of due process. They argue that the rules invalidate existing insurance rates without a contested case hearing and an opportunity for judicial review. I disagree.
The rules do not invalidate existing insurance rates. They are prospective only and apply solely to new and renewal policies issued after their effective date. Moreover, the rules are not self-enforcing; they do not invalidate rates. Defendant acknowledges that, after the rules take effect, insurers are entitled to notice and an opportunity for a hearing before rates may be invalidated. If an insurer’s rate filing uses insurance scoring, that rate filing will be disapproved as a violation of the OFIS rules. Plaintiffs’ argument conflates their right to a contested case hearing before a rate filing may be invalidated into a right to such a hearing before new rules may be promulgated. To create such a right would cripple an agency’s authority to promulgate *428rules and be duplicative of the procedural protections already present in the APA.
Finally, plaintiffs contend that any rate hearing will be a meaningless exercise because the outcome will be predetermined and the filing will be disapproved. This argument is disingenuous because plaintiffs chose to file this action for declaratory judgment attacking the facial validity of the rules. To accept plaintiffs’ due process argument would be to ignore that plaintiffs chose this forum, rather than individual contested case hearings, to challenge the OFIS rules. Moreover, this argument could just as easily be raised by an insurer that sets rates on the basis of impermissible factors such as race or gender; however, it is inconceivable that such rates would be allowed simply because the result of the contested case hearing was predetermined.
CONCLUSION
I agree with the majority’s decision to reach the substantive issues in this case. However, I dissent from its conclusion that the Insurance Commissioner exceeded her rulemaking authority under Luttrell v Dep’t of Corrections.
I would hold that the OFIS rules are valid and enforceable. Therefore, I would affirm the Court of Appeals judgment vacating the circuit court’s order granting a permanent injunction and declaring defendant’s rules illegal, unenforceable, and void.
Cavanagh and Hathaway, JJ., concurred with Kelly, C.J.I disagree with the majority’s waiver analysis. I reach the substantive issues because, like the majority, I have determined that, if there are procedural errors, they do not affect my conclusion on the substantive issues.
Luttrell v Dep’t of Corrections, 421 Mich 93; 365 NW2d 74 (1984).
Id. at 100 (citation and quotation marks omitted).
In re Complaint of Rovas against SBC Michigan, 482 Mich 90, 108; 754 NW2d 259 (2008).
Ante at 389; MCL 500.100 et seq.
Unlike the majority, I do see a difference between this inquiry and its phrasing of the question. Under MCL 500.210, the Commissioner has the authority to promulgate rules and regulations to effectuate the purposes of the Insurance Code. In my view, the power to enact rules to “effectuate the purposes” of the Insurance Code provides a broader grant of authority than the power simply to inquire whether the Code permits a particular practice. Because the majority hinges its conclusion on this prong of the Luttrell test, its precise application is imperative.
As the majority observes, I do question whether the Insurance Code authorizes insurance scoring. However, I do so as part of the second prong of my Luttrell analysis, ascertaining whether the OFIS rules comply with legislative intent.
For example, the majority offers no authority to support its conclusion that the Commissioner exceeded her authority by “enacting a total ban on a practice that the Insurance Code permits.” Ante at 407.
In re Rovas Complaint, 482 Mich at 108.
Ante at 384 (emphasis added).
For an error to be “harmless,” it cannot “affect a party’s substantive rights or the case’s outcome.” Black’s Law Dictionary (8th ed), p 582.
The majority seems to miss this point. Ante at 384 n 16. It can be paraphrased as follows: The majority says it can reach its result based on X (the administrative record), even if Y (the circuit court record) was erroneously admitted. Yet the majority refuses to rely solely on X, despite its assertion that X is sufficient to support its conclusion. Instead, it relies — and relies primarily — on Y, the record that it admits may have *413been erroneously admitted. If the majority refuses to rely solely on X to reach its conclusion, on what basis can it logically assert that X provides sufficient evidence for its conclusion?
See, e.g., ante at 402 n 30.
A careful reading of the majority opinion reveals that, of all the evidence relied on, only a small fraction is part of the administrative record. To the extent that the majority does rely on the administrative record, it cites it primarily for conclusory statements by plaintiffs, not actual data. Ante at 391 (“ ‘[T]he use of credit information is the most powerful predictor of losses to be developed in the past 30 years.’ ”); ante at 392 (“[0]ur data shows that credit information is highly predictive of loss ....”). This stands in stark contrast to the short shrift the majority gives to even attempting to accurately summarize defendant’s arguments. See ante at 395 n 23 (“it appears that defendant’s repeated references to ‘overall’ premiums and ‘overall’ losses are to industry-wide premiums and losses.”).
Michigan Ass’n of Home Builders v Dep’t of Labor & Economic Growth Dir, 481 Mich 496; 750 NW2d 593 (2008).
Michigan Employment Relations Comm v Detroit Symphony Orchestra, Inc, 393 Mich 116, 124; 223 NW2d 283 (1974).
See, e.g., In re Rovas Complaint, 482 Mich at 97-99.
Michigan Ass’n of Home Builders v Dep’t of Labor & Economic Growth Dir, 276 Mich App 467, 476; 741 NW2d 531 (2007) (emphasis in original), vacated in part on other grounds in Home Builders, 481 Mich at 501.
Compare the deferential Luttrell standard for review of quasi-legislative administrative agency actions, supra at 410-411, with the standard of review for judicial and quasi-judicial actions. The latter must be “ ‘authorized by law’ ” and its factual findings “ ‘supported by competent, material and substantial evidence on the whole record.’ ” Viculin v Dep’t of Civil Serv, 386 Mich 375, 384; 192 NW2d 449 (1971), quoting Const 1963, art 6, § 28.
Moreover, MCL 24.306, which governs judicial review involving judicial and quasi-judicial actions, allows for reversal of an agency’s decision when a decision or order of the agency is “any of the following”:
(a) In violation of the constitution or a statute.
(b) In excess of the statutory authority or jurisdiction of the agency.
(c) Made upon unlawful procedure resulting in material prejudice to a party.
(d) Not supported by competent, material and substantial evidence on the whole record.
(e) Arbitrary, capricious or clearly an abuse or unwarranted exercise of discretion.
(0 Affected by other substantial and material error of law.
Luttrell, 421 Mich at 100, quoting Chesapeake & Ohio R Co v Pub Serv Comm, 59 Mich App 88, 98-99; 228 NW2d 843 (1975).
See also MCL 500.2484 (“The commissioner may make reasonable rules and regulations necessary to effect the purposes of this chapter.”); MCL 500.2674 (same).
Ins Institute, 280 Mich App at 375 (opinion of Zahra, J.).
Id. at 358 (opinion of WHITE, P.J.).
MCL 24.201 et seq.
See ante at 389.
1956 FA 218.
Despite this concession, the majority feels free to conclude that “insurers do, of course, anticipate reductions in their own losses or expenses to result from the use of premium discount plans using insurance scoring.” Ante at 395 (emphasis omitted).
The majority’s discussion of industry-wide losses and expenses is a red herring, as this illustration demonstrates. Defendant’s brief frequently refers to “overall” losses. However, nowhere does defendant specifically contend that a discount plan is permissible under MCL 500.2110a only if it reflects anticipated reductions in losses or expenses across the entire insurance industry, nor does my analysis impose such a requirement.
Rather, I interpret MCL 500.2110a as the majority does, as requiring that a discount plan reasonably anticipate a reduction in losses or expenses to each insurer. Initially, I note that an insurer cannot “reasonably” anticipate a reduction in losses or expenses based on a discount plan premised on insurance scoring if credit reports are unreliable. See infra at 16-20.
Moreover, defendant’s example illustrates that insurance rates that are set on the basis of insurance scoring will not reduce the premiums that an individual insurer collects. Instead, such rates will reallocate the dollar amount of the premium paid by each insured. I fail to see how such rates can reflect “reasonably anticipated reductions in losses or expenses” if the insurer continues to collect the same amount in total premiums.
Ins Institute, 280 Mich App at 361.
In re Rovas Complaint, 482 Mich at 108.
MCL 500.2426 and MCL 500.2626 both state that no “rating plan” that is filed pursuant to the requirements of their respective chapters shall be disapproved if the rates thereby produced meet the requirements of the chapter.
MCL 500.2403(1) provides in part:
*421(c) Risks may be grouped by classifications for the establishment of rates and minimum premiums. Classification rates may be modified to produce rates for individual risks in accordance with rating plans that measure variations in hazards, expense provisions, or both. The rating plans may measure any differences among risks that may have a probable effect upon losses or expenses as provided for in subdivision (a).
(d) Rates shall not be excessive, inadequate, or unfairly discriminatory. A rate shall not be held to be excessive unless the rate is unreasonably high for the insurance coverage provided and a reasonable degree of competition does not exist with respect to the classification, kind, or type of risks to which the rate is applicable. ... A rate for a coverage is unfairly discriminatory in relation to another rate for the same coverage, if the differential between the rates is not reasonably justified by differences in losses, expenses, or both, or by differences in the uncertainty of loss for the individuals or risks to which the rates apply. A reasonable justification shall be supported by a reasonable classification system; by sound actuarial principles when applicable; and by actual and credible loss and expense statistics or, in the case of new coverages and classifications, by reasonably anticipated loss and expense experience. A rate is not unfairly discriminatory because the rate reflects differences in expenses for individuals or risks with similar anticipated losses, or because the rate reflects differences in losses for individuals or risks with similar expenses. Rates are not unfairly discriminatory if they are averaged broadly among persons insured on a group, franchise, blanket policy, or similar basis.
The pertinent parts of MCL 500.2603 are identical to the portions of MCL 500.2403 cited in note 31 of this opinion.
Mich Admin Code, R 500.1505(3).
United States General Accounting Office (GAO), Statement for the Record Before the Committee on Banking, Housing, and Urban Affairs, U.S. Senate, Consumer Credit, Limited Information Exists on Extent of Credit Report Errors and Their Implications for Consumers (July 31, 2003). Available at: <http://www.gao.gov/new.items/d031036t.pdf> (accessed June 24, 2010).
OPIS Report to JCAR (October 1, 2004), p 23 (observing the “wide divergence in opinion” regarding the accuracy of credit reports).
Ante at 405.
In my view, if insurance scores are based on credit reports containing inaccurate information, they cannot be “likely to identify significant differences in mean anticipated losses.” However, I do not further address this issue because I conclude that plaintiffs have not shown a correlation between credit scores and risk of loss.
A “univariate analysis” is an analysis that takes only one factor or variable into consideration. See Anmol’s Dictionary of Statistics (2005); Oxford Dictionary of Statistical Terms (2003).
Other authors have criticized the use of univariate analysis in some of the studies cited in the administrative record. See Cheng-Sheng Peter Wu and James Guszcza, Does Credit Score Really Explain Insurance Losses ? Multivariate Analysis from a Data Mining Point of View, <http://casualtyactuaries.com/pubs/forum/03wforum/03wfll3.pdf> (accessed June 24, 2010), p 9 (“Unfortunately, univariate statistical studies such as Tillinghast’s do not always tell the whole story.”).
OFIS Report to JCAR, supra at 20.
Id. The study that included data from Michigan was Michael Miller and Richard Smith, The Relationship of Credit-Based Insurance Scores to Private Passenger Automobile Insurance Loss Propensity, available at <http://www.progressive.com/shop/EPIC-CreditScores.pdf> (accessed June 24, 2010).
The Michigan-specific data, which showed no correlation between insurance scores and frequency in filing of insurance claims, is Appendix Q of this study. It is available at <http://www.michigan.gov/documents/ Attachment_5_-_EPIC_Charts_-_MI_l 13194_7.pdf> (accessed June 24, 2010).
The majority correctly observes that, in the body of its report, the authors asserted that the “graphs for each state ... exhibit strikingly similar patterns of decreasing claim frequencies with increasing insurance scores to the pattern observed in the countrywide data.” Ante at 392 n 21 (emphasis omitted), quoting Plaintiffs’ Appendix in Docket No. 137400, at 33b. However, this assertion is undermined by the actual data, which show that claim frequency in Michigan based on insurance scoring ranged from only 0.5% to 0.8%. While the claimants with the highest insurance score did have the lowest rate of claims (0.5%), claimants with the third highest insurance score had one of the highest rates of claims (0.8%). Therefore, one is hard-pressed to square the actual data with the authors’ conclusion that the majority quotes.
The majority excuses this disparity by citing Michael Miller’s affidavit, in which Miller attempts to explain it away. Ante at 393 n 21. The affidavit claims that “the relatively few claims resulted in substantial random variations in the data, making the correlation between credit-based insurance scores and losses less obvious in the Michigan data for this coverage.” The existence of an excuse for why the Michigan data makes the connection between credit scoring and losses “less obvious” does nothing to justify the majority’s reliance on it.
Finally, Miller attached five graphs to his affidavit with Michigan-specific data purporting to buttress the EPIC study’s conclusion. Again, the majority takes the author’s stated conclusion at face value. However, as with the data from the EPIC study, most of the actual numbers do not show a strong correlation between credit scoring and propensity for loss. Instead, the portion of the graphs charting Michigan-specific data often deviate significantly from that pattern and do not demonstrate the “strong correlation” that the majority posits.
OFIS Report to JCAR, supra at p 20.
The majority’s response to this dissent on the substantive issues involved is unavailing because it presupposes the majority’s ultimate conclusion: that insurance scoring is predictive of loss. Ante at 397, 397 n 25. Thus, its conclusions that “[djiscounts for anti-lock brakes are offered because they reduce the risk of loss, and discounts for high insurance scores are offered because they reduce the risk of loss” do not advance its position. Ante at 397. Similarly, I see little value in speculating that offering discounts based on insurance scoring might lead sometime in the future to reductions in premiums. Ante at 397 n 25. Indeed, for defendant to rely on such speculation as a basis for formulating administrative rules is, in my view, erroneous.
Finally, the majority’s contention that “setting premium rates without considering insurance scoring also reallocates the amount insureds pay” is similarly unavailing. Ante at 397 n 25 (emphasis omitted). Plaintiffs contend that the use of insurance scoring is permissible because it constitutes a “discount plan” under MCL 500.2110a. However, as previously noted, if insurance scoring simply reallocates rates, rather than resulting in an overall reduction in premiums, it is an unapproved rating factor, not a discount plan.
By contrast, no party has contended that setting premium rates without considering insurance scoring constitutes a “discount plan” within the meaning of MCL 500.2110a. That system simply reallocates the amount insureds pay based on permissible rating factors laid out in the relevant statutes.
See note 38 of this opinion.
Ante at 402 & n 30 (citing sources outside the existing record, including a contract between the state and Credit Technologies, Inc, and references to credit scores in MCL 493.137(4)(b)(i) and MCL 493.163(l)(a)(¿¿) as “evidence” that credit reports are rehable).
Ante at 407 n 35.
Blank v Dep’t of Corrections, 462 Mich 103, 128; 611 NW2d 530 (2000) (opinion by Kelly, J.), citing Dykstra v Dep’t of Natural Resources, 198 Mich App 482, 491; 499 NW2d 367 (1993).