concurring:
This case requires us to set forth the appropriate test for determining whether a sufficient showing has been made to allow discovery in. a selective prosecution claim. I thought we could all agree that this test should be the one that our court developed in United States v. Bourgeois, 964 F.2d 935 (9th Cir.) (Bourgeois), cert. denied, — U.S. -, 113 S.Ct. 290, 121 L.Ed.2d 215 (1992). There, we held that “to obtain discovery on a selective prosecution claim, a defendant must present specific facts, not mere allegations, which establish a colorable basis for the existence of both discriminatory application of a law and discriminatory intent on the part of government actors.” Id, at 939. ,What divides us is whether that standard should be rewritten and its application to this case. I believe that the result the majority reaches is correct, and to that extent I concur. I write separately, however, to discuss why I do so.
I
The majority correctly points out that the dismissal of the indictment as a sanction for noncomplianee with the discovery order creates the appealable order that gives us jurisdiction over this appeal. However, this does not mean that the dismissal must be interpreted to be with prejudice. For a dismissal to be "with prejudice, “the prosecution must be forewarned” of that fact. United States v. Loud Hawk, 628 F.2d 1139, 1151 (9th Cir. 1979) (en banc), cert. denied, 445 U.S. 917, 100 S.Ct. 1279, 63 L.Ed.2d 602 (1980). While the government agreed to the sanction in this case in order to effectuate the appeal, it does not appear in the record before us that the government was forewarned of any sanction that would forbid further proceedings against the defendants.
II
Turning to the merits, the majority, in an attempt to “clarify” the Bourgeois standard, goes too far. Bourgeois, held that- “the discovery threshold should not be so high as to require establishment of a prima facie case.” Bourgeois, 964 F.2d at 939. Thus, according to Bourgeois, even though the defendant must make some showing that the elements of a selective prosecution claim can be met, that showing is less than what is ultimately required to prove selective prosecution. Bourgeois, as written, is sufficient for us to decide the case before us.
Undeterred, however, the majority attempts to “clarify” this test by stating that the “ ‘high threshold’ language in Bourgeois appeared to set an artificially onerous burden” on discovery. While the formulation of the colorable basis test shows that the test is certainly not as “high” as that required to prove a prima facie case, neither is it as “low” as a nonfrivolous showing. The dissent’s discussion to the effect that á high threshold “is appropriate because courts are ill equipped to assess a prosecutor’s charging *1521decisions, and oversight of prosecutorial decisions could undermine effective law enforcement” seems to me correct. To the extent that the majority opinion undermines this idea, I disagree, and- align myself with the analysis of the dissent.
On the other hand, my agreement with the majority’s result is supported by a very similar observation- about our role as a federal appellate court. -Just as the district court should be careful not to undermine effective law enforcement by constantly evaluating the prosecutor’s charging decisions, so too should we, as an appellate court, not second-guess the ruling of a district court in a close case. We are able to carry out this function by adhering to our standards of review. ■
When reviewing the district court’s discovery ruling on a selective prosecution claim, we ask whether the district court abused its discretion. Id. at 937; see also United States v. Bryan, 868 F.2d 1032, 1035 (9th Cir.), cert. denied, 493 U.S. 858, 110 S.Ct. 167, 107 L.Ed.2d 124 (1989). What this means is that “[ajbsent a definite and firm conviction that the district court committed a clear error of judgment, we will not disturb the district court’s decision.” Allen v. City of Beverly Hills, 911 F.2d 367, 373 (9th Cir. 1990).
In this case, the defendant provided: (1) evidence that all 24 cocaine base cases closed in 1991 by the Federal Public Defender involved black defendants; (2) two additional affidavits stating that there are Caucasian cocaine base users and that the number of such users prosecuted in state court is greater than the number prosecuted in federal' court; and (3) a Los Angeles Times article stating that federal cocaine base laws carry-tougher sentences. The district court, applying the Bourgeois standard, found that a colorable basis had been met, and ordered limited discovery. Because the abuse of dis: cretion standard permits the district court flexibility, I conclude we should not overturn that order in this case.
The dissent argues that, as a matter of law, the colorable basis test has not been met. This conclusion is driven largely by the ’ fact that there has been no showing here that “others similarly situated” have . not been prosecuted. I agree with the dissent that to establish a prima facie case, hard data about others similarly situated is necessary. Proving a prima facie ease of selective prosecution requires proof: (1) that others similarly situated have not .been prosecuted, and (2) that the prosecution is based on an impermissible motive (discriminatory purpose or intent). United States v. Gutierrez, 990 F.2d 472, 475 (9th Cir.1993). To the extent the majority disagrees with this formulation, and discusses how the statistical evidence here might be sufficient to prove both effect and intent, I do not concur. But this is all beside the point at the discovery stage, where only some evidence,' tending to show selective prosecution, is required. Where there is evidence of a large enough number of prosecutions directed at a single race over a sufficiently long period of time, eventually there becomes a point where that evidence is sufficient to establish a colorable basis of selective prosecution.
The language of the colorable basis standard does indicate that specific facts must exist to establish the basis for believing that “discriminatory application of law” and “discriminatory intent” are present. Put without discovery, the contention that “others similarly situated” have not been prosecuted (a-claim essential to the prima facie case) may be impossible tó show. The evidence produced by the defense in this .case presents a close question, and, had I been the district judge, I might well have concluded that it was insufficient to make out a colorable basis. Indeed, this en banc case might have resulted in a different outcome were we to review the district judge’s decision de novo. I am unable, however, to conclude that the district judge abused her discretion in allowing limited discovery.
I therefore concur in the result reached by the majority.
RYMER, Circuit Judge, with whom Circuit Judges LEAVY, T.G. NELSON, and KLEINFELD, join, dissenting:
For the first time in this circuit or any other, the en banc court has held that a raw number of prosecutions, without reference to *1522a comparison group and without evidence that others, similarly situated .except for their race, have not been prosecuted, provides a colorable basis for the existence- of both discriminatory effect and discriminatory intent sufficient to order diseoyery from the government in connection with a criminal defendant’s selective prosecution defense. Also, though’both the Supreme Court and this circuit have made clear that discriminatory effect and discriminatory intent are two different elements, each of which must exist, the majority’s opinion effectively collapses intent into effect by holding that both may be shown by the same, insubstantial statistic. Additionally the opinion has formally removed the “high threshold” that, until now, we explicitly (and other circuits implicitly) have required to be met before discovery relating to a selective prosecution claim can be ordered. In so doing, the majority opinion radically, and unnecessarily, rewrites the law of selective prosecution. I therefore dissent.
I
The government’s appeal requires us to decide whether five defendants indicted in 1992 for conspiring to distribute cocaine base (in part through úsing an armed guard) in violation of 21 U.S.C. §§ 846 and 841(a)(1), four of whom are also charged with using a firearm in connection with the drug offenses under. 18 U.S.C. § 924(c), are entitled to discovery on their claimed defense of selective prosecution on account of race.1 I would hold that they, are not because Armstrong has faded to show a colorable basis for the existence of either of the elements required for selective prosecution, discriminatory effect and discriminatory intent.
Armstrong seeks to show that this prosecution was undertaken not because of the crimes committed, but rather because the United States Attorney’s Office prosecuted these defendants as part of a systematic policy intentionally to discriminate against blacks in its prosecutorial decisions. The discovery request was based on evidence that of the 24 defendants represented by the Federal Public Defender’s Office (FPD) in § 846 and § 841(a)(1) cocaine base cases closed by the FPD in 1991, all were black. The filing stamp numbers on the closed cases in this study range from “88” to “91,” indicating that the defendants whose cases were closed by the FPD in 1991 were prosecuted over a four-year period.
After the district court ordered discovery, the government moved for reconsideration and presented evidence that at least four non-black defendants were prosecuted on crack charges during the period when the cases in the FPD’s study were opened, two of whom-were represented by the FPD, and at least seven were charged in 1992, the year the Armstrong defendants, were indicted. In addition, the government submitted declarations indicating that in total, without breakdown by type of drug, approximately 2400 persons were charged with violations of § 841, and 1700 with violations of § 846, in the past three years; that race played no part in the investigation which resulted in the Armstrong defendants’ arrests; that the county district attorney’s offices prosecute many black cocaine base offenders; that so-cio-economic factors account for the prevalence of drugs and trafficking patterns in different communities; and that all charging decisions made by the United States Attorney’s Office, including in this case, are made on the basis of whether a federal offense meeting the Office’s guidelines has occurred, the overall strength of .the evidence, the deterrence value and federal interest associated with the particular case, the criminal history of the suspects, and other race-neutral criteria.. Specifically as to this case, the declaration of the then-Chief of the Criminal Complaints Section explained why the Armstrong defendants were prosecuted and stated that the decision to charge them was consistent with the guidelines: there were over 100 grams of cocaine base involved — more than twice the quantity necessary to trigger a ten year mandatory minimum sentence; there were multiple sales involving multiple defendants, thereby indicating a substantial crack cocaine ring; the case was jointly investigated with a federal Bureau of Alcohol, Tobacco and Firearms agent; there were multiple *1523federal firearms violations intertwined with the narcotics trafficking; the overall evidence was extremely strong, including audio and videotapes of the defendants; threats had been made to arresting officers by Armstrong; and several of the defendants had criminal histories including narcotics and firearms violations.
Armstrong countered with a declaration by defense counsel stating that she had spoken with a halfway house intake coordinator who told her that in his experience in treating cocaine base addiction, the number of Caucasian and minority users and dealers is the same. Armstrong also offered a declaration by another defense attorney averring that of the defendants facing crack charges whom he has represented in federal court, all were black, and that he has never heard of non-blacks being prosecuted in federal court, whereas based on what he has heard, he believes that the state prosecutes many non-black cocaine base offenders in state court.' Finally, Armstrong proffered an article from the Los Angeles Times, which indicates that blacks disproportionately commit cocaine base offenses.
The district court found that Armstrong had made a sufficient showing to require the government to provide to the defense:
(1) A list of all eases from the prior three years in which the government charged both cocaine base offenses and firearms offenses;
(2) The race of the defendant(s) in each of those cases;
(3) Whether each case was investigated by federal, state or joint law enforcement authorities; and
(4) An explanation of the criteria used by the United States Attorney’s Office for deciding whether to bring cocaine base eases federally.
The court’s initial ruling concluded:
In this case we have a fairly general charge — one that we see regularly in this courthouse — and whether it’s coincidental or not, that out of the group that the public defender — that Ms. O’Connor provides us information on — all of them happen to be of the same racial group.
I think the number is adequate that would at least require the Government to provide some explanation. The time period is such that would require some explanation. The charges are the same or similar, and the race is the same in each case.
In’ denying the motion to reconsider, the court further concluded:
The statistical data provided by the Defendant raises a question about the motivation of the Government which could be satisfied by the Government disclosing its criteria, if there is. any criteria, for bringing this case and others like it in Federal court.
Without the criteria, the statistical data is evidence and does suggest that the decision to prosecute in Federal court could be motivated by race.
Without expert testimony, this Court cannot conclude that the Defendants’ evidence is explained by social phenomena.
The executive branch has a responsibility to dissuade public opinion that its decisions to prosecute or where to prosecute are not motivated by improper reasons such as race, and the evidence of who was prosecuted and why those persons were prosecuted and where those persons were prosecuted is within the peculiar knowledge of the Government and therefore, as the Court indicated at the previous hearing, it would be this Court’s position that it is the Government that would have to provide that evidence so that Defendants could analyze it and decide if there is any basis for filing a motion to dismiss.
The government declined to comply with the order, and the court granted Armstrong’s motion to dismiss. This appeal followed.
The panel, relying on United States v. Bourgeois, 964 F.2d 935 (9th Cir.), cert. denied, — U.S. -, 113 S.Ct. 290, 121 L.Ed.2d 215 (1992), reversed on the ground that a defendant must supply a colorable basis for' believing that others similar to him — except that they are not in his protected class — were not prosecuted, and that other evidence adduced by Armstrong was too flimsy to be accorded any weight. United *1524States v. Armstrong, 21 F.3d 1431 (9th Cir. 1994).2
We went en banc in part to resolve the tension between Bourgeois (“colorable basis” founded on specific facts required for discovery of whether prosecution is discriminatory) and United States v. Redondo-Lemos, 955 F.2d 1296 (9th Cir.1992) (“prima facie showing” required to make prosecution answer for a particular charging or plea bargaining decision satisfied by district court’s sua sponte suspicion of unconstitutional conduct or a defendant’s evidence demonstrating reasonable inference of invidious discrimination), which the panel construed as being limited to discovery orders based on the judge’s own experience. Without commenting one way or the other on a court’s sua sponte powers to initiate a selective prosecution inquiry because that issue is not before us,3 I agree with the majority that discovery should be governed by a single standard.
However, I disagree with the majority’s opinion. Although it appears to embrace the “colorable basis” standard we adopted in Bourgeois, the opinion actually guts it by holding that when a selective prosecution claim is based on race, evidence “tending to show that only members of racial or ethnic minority groups have been prosecuted” will suffice. Op. at 1517, n. 6. Instead, I would reaffirm the Bourgeois standard, and hold that the district court erred as a matter of law in finding that Armstrong made an adequate showing in the absence of any evidence that.others, similarly situated except for being non-black, were not prosecuted. By the same token, the court should not have ordered discovery based on its conclusion that “the statistical data provided by the Defendant raises a question about the motivation of the Government”; the proper legal standard is not whether the defendant’s evidence “raises a question,” but whether it provides a colorable basis for the existence of discriminatory effect and discriminatory intent. In the absence of any evidence that the government purposefully selected these defendants for prosecution on account of their race, there is no colorable basis for the existence of discriminatory intent as a matter of law. The district court also went astray in founding its order on a “fairly general charge” instead of on specific facts as to these defendants. Further, at the discovery stage, it is the defendant’s burden to provide facts that, if fleshed out by the discovery sought, would tend to prove both elements of selective prosecution. Thus, the district court’s order fails in two additional respects: first, it shifts to the government the “responsibility to dissuade public opinion,” but ignores the government’s submissions; and second, the discovery ordered does not advance the ball game because racial identity and criteria for federally charging crack and firearms violations will still only show who was prosecuted, not who wasn’t. In any event, the FPD’s study relates to crack eases closed by that office and says nothing about any pool of crack and firearms offenders. As the discovery that was ordered targets prosecutions for both crack and firearms violations, it lacks any basis at all. I would, therefore, reverse.
II
Armstrong contends that the en banc court should adopt the “reasonable inference” test articulated in Redondo-Lemos instead of the “colorable basis” standard set out in Bourgeois. Short of this, he suggests that Bourgeois and Redondo-Lemos could be reconciled by defining “colorable basis” as the Seventh Circuit did in United States v. Hei-*1525decke, 900 F.2d 1155 (7th Cir.1990), to mean that a defendant need only produce “some evidence to show the essential elements of the claim.” Id. at 1159.
Í do not believe that it is necessary to do' more than resolve the conflict between Bourgeois and Redondo-Lemos, because the evidence in this ease does not pass muster under either the Bourgeois or Heidecke standard. See United States v. Kerley, 787 F.2d 1147, 1150 (7th Cir.1986) (applying Heidecke test suggested by Armstrong but reversing discovery order because the defendant presented no evidence that he was singled out for prosecution while others were not prosecuted, though they similarly were in violation of the law). Nevertheless, the majority opinion uses this en banc opportunity to look afresh at what the standard for discovery on a selective prosecution defense should be.
There is no question that selective prosecution claims of the sort we address today are deeply troubling. Such claims bring powerful interests into conflict: the nation’s commitment to rid itself of drugs, on the one hand, and the individual’s right not to be singled out for prosecution on account of race, on the other. They reflect concerns which are legitimate and widespread — and which I share — that mandatory minimum sentences in general, and those for crack offenses iii particular, fall heavily on young black males.4 Such things are, however, up to the Congress and the United States Sentencing Commission. Our task is far more discrete, having to do only with the specific showing made by the defendants in this particular case, and being constrained by the difference between our role as judges and the role of the government as prosecutor, which the Supreme Court has admonished us to remember in the context of selective prosecution claims:
In our criminal justice system, the Government retains “broad discretion” as to whom to prosecute. “[S]o long as the prosecutor has probable cause to believe that the accused committed an offense defined by statute, the decision whether or not to prosecute, and what charge to file or bring before a grand jury, generally rests entirely in his discretion.” This broad discretion rests largely on the recognition that the decision to prosecute is particularly ill-suited to judicial review. Such fac-r tors as the strength of the case, the prosecution’s general deterrence value, the Government’s enforcement priorities, and the case’s relationship to the Government’s overall enforcement plan are not readily susceptible to the kind of analysis the courts are competent to undertake. Judicial supervision in this area, moreover, entails systemic costs of particular concern. Examining the basis of a prosecution delays the criminal proceeding, threatens to chill law enforcement by subjecting the prosecutor’s motives and decisionmaking to outside inquiry, and may undermine prosecutorial effectiveness by revealing the Government’s enforcement policy. All these are substantial concerns that make the courts properly hesitant to examine the decision whether to prosecute.
United States v. Wayte, 470 U.S. 598, 607, 105 S.Ct. 1524, 1530, 84 L.Ed.2d 547 (1985) (citations omitted).
In that light, I disagree with the majority that there is any call to revisit Bourgeois. There we held that
to obtain discovery on a selective prosecution claim, a defendant must present specific facts, not mere allegations, which establish a colorable basis for the existence of both discriminatory application of a law and discriminatory intent on the part of government actors. This is a high threshold. As has been true historically, it will be the rare defendant who presents a sufficiently strong ease of selective prosecution to merit discovery of government documents.
964 F.2d at 939. Bourgeois is in the mainstream of the law,5 and appropriately bal-*1526anees the interests of prosecutorial discretion and freedom from invidious discrimination. It correctly acknowledges that the threshold for discovery is high because the showing required on the merits of a selective prosecution defense is high.
A “high threshold” is appropriate because courts are ill equipped to assess a prosecutor’s charging decisions, and oversight of prosecutorial decisions could undermine effective law enforcement. Bourgeois; 964 F.2d at 939-40; Wayte, 470 U.S. at 607-08, 105 S.Ct. at 1530-31. As we explained in Bourgeois:
We believe that this high threshold will discourage fishing expeditions, protect legitimate prosecutorial- discretion, safeguard government investigative records, and yet still allow meritorious claims to proceed. This threshold also draws on the effectively high hurdle all the circuits have adopted, regardless of the phraseology of their standards. Our research of circuit court opinions uncovered only a handful of instances in the past few decades in which a defendant obtained discovery or dismissal of charges based on a selective prosecution claim. In those cases, the defendants presented solid, credible evidence in support of their claims.
964 F.2d at 940.
The majority opinion nevertheless declares that the meaning of the Bourgeois standard is “elusive,!’ and uses that excuse to rewrite the law. However, “colorable basis” as a standard is no more elusive than any other threshold standard. It is the standard most other circuits apply. Although there may be no way precisely to define it, I doubt that many judges would disagree that “colorable basis” is a lower level of specificity than a prima facie showing, but greater than an allegation. As Bourgeois indicates, it simply requires some specific facts — not allegations, speculation, conclusions, or questions — that are solid and credible enough to indicate that both elements of a selective prosecution defense exist.
While the majority opinion purports to clarify what “colorable basis” means, it actually crafts a new standard which is far more elusive than Bourgeois. It is variously described as “some evidence tending to show the essential elements of the claim,” op. at 1520, “more than frivolous and based on more than conclusory allegations,” id., more than “frivolous and unwarranted allegations,” id. at 1515, “some evidence that tends to show that the United States Attorney may be engaging in discrimination,” id. at 1517, and “evidence that reasonably permits one to find a colorable basis for concluding that race-based prosecutorial selection may have occurred.” Id. at 1519. To permit discovery on a showing that discrimination “may” have *1527occurred in the absence of any facts tending to show that the defendant has been singled out for prosecution on account of race puts us out of line with all other circuits!.6 To the extent that the new threshold is “more than frivolous and unwarranted allegations,” it may well be “more aldn” to that favored by two circuits and the dissent in Wayte, but is out of step with the others.7 Bourgeois is in line with the majority view, is faithful to the Supreme Court’s admonitions in Wayte, and appropriately balances the interests at stake. We should, therefore, stick with it.8
III
We previously held in Bourgeois, 964 F.2d at 937, and the majority continues to hold, that a district court’s decision to order or not order discovery is reviewed for abuse of discretion. While we should not substitute our judgment for that of the district court, we do reverse even under this deferential standard when the court either does not apply the correct law, or rests its decision on a clearly erroneous finding of material fact. Mar-chando v. Mercy Medical Center, 22 F.3d 933, 936 (9th Cir.1994).
IV
I would hold that the district court erred as a matter of law in concluding that- Armstrong made a sufficient showing in the absence of any evidence that similarly situated non-black offenders were not prosecuted, and in the absence of any finding that the statistical pattern presented could be so stark, if fleshed out by discovery, as to be inexplicable on grounds other than intentional discrimination on the basis of race. The Supreme Court has made clear that unless there is an-overtly discriminatory classification, which Armstrong does not claim in this ease, a defendant is required to show both discriminatory effect and discriminatory intent in order to make out a -defense of selection prosecution. Wayte, 470 U.S. at 608, 105 S.Ct. at. 1531. Without any evidence of similarly situated offenders who share everything in common with the defendant except for the constitutionally impermissible factor, there can be no colorable basis for the existence of discriminatory selection or effect. Nor is there a colorable basis for the existence of discriminatory intent because intent, to discriminate against black defendants cannot possibly be inferred from only a statistic that has no comparison pool, is statistically insufficient to suggest a pattern inexplicable save for race, and is in fact explained on grounds having nothing to do with race: that these defendants dealt in substantial quantities of crack, many times, .using firearms; made threats; had prior drug and firearms convictions; and were caught in the act on tape.
While acknowledging that both discriminatory effect and discriminatory intent must be shown, the majority opinion effectively cob-lapses the two by holding that “inadequately explained evidence of a. significant statistical disparity in the race of those prosecuted suffices- to show the colorable basis of dis-criminátory intent and effect that warrants discovery on a selective prosecution claim.” Op. at 1514-15. In this it is incorrect: from the statistic in evidence, “statistical dispari*1528ty” cannot be inferred. Even if it could be, the statistical, disparity which could be inferred from 24 closed prosecutions against black defendants in one year and 11 opened cases involving non-black defendants over a four year period has no possibility of being “stark” enough to show discriminatory purpose when the government has offered legitimate, race-neutral reasons for prosecuting crack cases in general and the Armstrong defendants in particular.
A
We have held a number of times in a number of different cases that the first thing a defendant must show is that “others similarly situated have not been prosecuted....” United States v. Wayte, 710 F.2d 1385, 1387 (9th Cir.1983), aff'd, 470 U.S. 598, 105 S.Ct. 1524, 84 L.Ed.2d 547 (1985), quoted in Bourgeois, 964 F.2d at 938. Bourgeois failed to establish a colorable basis for the first element of his claim because he made “no attempt to show that, in any span of time, the government has declined to prosecute similarly situated, non-black felons illegally in possession of firearms.” Id. at 941. The majority opinion mistakenly distinguishes Bourgeois by emphasizing that it involved only one operation, over a single two^day period; in fact, Bourgeois’ focus .was on both the span of time and scope necessary to have-probative value and the lack of similarly situated prosecutions. In United States v. Wilson, 639 F.2d 500 (9th Cir.1981), the defendants claimed that they were singled out by the government because they were tax protesters. The evidence showed that out of 425 taxpayers who filed an “exempt” Wt4 form the only two who were prosecuted were tax protesters; that an IRS investigator was unaware of any.tax protestors who were not prosecuted for non-payment of taxes; and that others whose W-4 forms might bear-investigation had not been prosecuted. We held that the evidence did not suffice for a selective prosecution claim as the defendants
have not shown, as the Oaksm test requires, that other[9] similarly situated who have not exercised their rights have not been prosecuted. The statistics listed above show that there were other prosecutions and the Wilsons have not shown that those defendants were also tax protestors. This being so, we cannot find that the Wilsons carried their burden of proving that the decision to prosecute was made because they exercised their constitutional rights.
Id. at 504. In United States v. Aguilar, 883 F.2d 662 (9th Cir.1989), cert. denied, 498 U.S. 1046, 111 S.Ct. 751, 112 L.Ed.2d 771 (1991), we upheld the district court’s denial of defendants’ request for discovery and an evidentia-ry hearing on their motion for dismissal based upon selective prosecution because the defendants’ suggested definition of those similarly situated excluded the most relevant analogous class. Stating that “[a]bsent a similarly situated control group, the government’s prosecution of a defendant exercising his constitutional rights proves nothing,” id. at 706, we joined the D.C. Circuit in the understanding that “ ‘[discrimination cannot exist in a vacuum; it can be found only in the unequal treatment of people in similar circumstances.’ ” Id., quoting Attorney General of the United States v. Irish People, Inc., 684 F.2d 928, 946 (D.C.Cir.1982), cert. denied, 459 U.S. 1172, 103 S.Ct. 817, 74 L.Ed.2d 1015 (1983). Accordingly, we held that “if similarly situated persons are being prosecuted then appellants fail to make the required showing.” Id. Likewise in United States v. Gutierrez, 990 F.2d 472 (9th Cir. 1993),10 we affirmed the district court’s fail*1529ure to grant a motion to dismiss for selective prosecution on charges of distributing cocaine base within 1000 feet of a school because “[statistics showing a high percentage of certain groups being prosecuted, alone, are insufficient to establish that similarly situated persons are not being prosecuted.” Id. at 476. And in Ness, we considered a showing quite similar to Armstrong’s and held with respect to Ness’s showing:
While he showed that similarly situated members of his tax protest group had also been prosecuted, Ness did not show a single instance of a similarly situated but nonprotesting violator who had not been prosecuted. The fact that access to the Government’s files might be helpful to a defendant seeking to prove discriminatory prosecution does not relieve him of the burden of making an initial showing, nor does that fact, in itself, entitle every defendant raising such a claim to discovery.
652 F.2d at 892.
Thus it is well settled: to succeed on a claim of selective prosecution, a defendant has the burden of establishing that others similarly situated have not been prosecuted. The burden for purposes of discovery is substantively the same, only it is met by showing a colorable basis that others similarly situated have not been prosecuted.11 As the panel explained:
Requiring defendants to provide a color-able basis for believing that others similarly situated have not been prosecuted is a reasonable requirement. “Selective prosecution” implies that a selection has taken place. If a defendant is part of a protected class, that alone does not provide a colorable basis for believing that a selection has taken place; nor does evidence demonstrating that other members of the protected class were prosecuted provide a colorable basis for so believing. Rather, a defendant must supply a colorable basis for believing that others similar to him except that they are not in his protected class were not prosecuted. Without a col-orable basis to believe that others similarly situated were not prosecuted, the most reasonable conclusion is that the defendant was selected for prosecution because the government believed the defendant committed the offense; the fact that the defendant is a member of a protected class is coincidental.
Armstrong, 21 F.3d at 1436-37.12
Armstrong’s showing lacks any specific evidence that the government has failed to *1530prosecute similarly situated non-blacks, or that any similarly situated non-black cocaine base offender was not chosen for federal prosecution but was left to be prosecuted in state court. Evidence that all 24 crack cases closed by the Federal Public Defender in 1991 involved black defendants is a meaningless number without some pool against which to measure it. It shows only that blacks have been prosecuted, not that others similarly situated who are not black have not been prosecuted. The study says nothing about cases that were opened by the FPD during the period its closed cases were active.13 (Although that would come closer to a valid sample, there is no basis in the record for extrapolating from the Federal Public Defender’s study a figure representative of all relevant defendants prosecuted in the Central District.) Nor does the statistic speak to whether the Federal Public Defender had other cases involving non-black defendants that were opened during the same period that cases closed in 1991 were opened (1988-1991). Indeed, the government’s evidence that during this period at least two non-black defendants were prosecuted who were represented by the FPD, and two others were prosecuted who were not represented by the FPD, is uncontradicted. It is no answer to suggest, as Armstrong and the majority do, that the government’s evidence is irrelevant because it is outside the defendant’s study: the court must identify an appropriate comparison group, Aguilar, 883 F.2d at 706, and by definition, the appropriate population. In any event, the FPD study is only relevant to the extent that it accurately represents the universe of prosecutions.14 There can be no serious question that evidence of prosecutions which were actually brought is more on target than a study of a small sample of closed cases that is not random and is biased by the composition (indigent defendants only) of the FPD’s clients. Among other things, the government’s evidence shows that charges were filed against seven non-black defendants in 1992 — the same year that the Armstrong defendants were indicted. That is the only evidence in the record about what went on in the specific year of the prosecution in this case. Without putting too fine a point on it, that fact cannot be discounted as immaterial.
*1531In addition, there is no indication that any of the defendants in the 24 cases encompassed by the FPD study were similarly situated to the defendants in this case. All of the Armstrong defendants are charged with conspiring to distribute cocaine base using an armed guard, and four of the five are charged with a firearms violation. The population in the FPD study is crack offenders — not crack offenders who committed cocaine base offenses while possessing firearms. As the discovery ordered is of prosecutions for both crack and firearms offenses, there is no basis in the statistical evidence for finding that the discovery ordered is relevant. In addition, the study concerns only those defendants who were prosecuted in 1991 or before. Thus, the study population does not include Armstrong and cannot mean anything so far as these defendants are concerned. An inference of discrimination against these defendants can therefore not arise even if the statistic could imply discrimination against crack defendants whose cases were dosed by the FPD in 1991.
For these reasons, the only statistic in evidence affords no basis for inferring the existence of discriminatory selection, or effect. The mere fact that all the offenders share the same trait ii.e., are black) cannot prove that this trait was the motivation for selecting them for prosecution unless other similar offenders who do not possess that trait were not selected. The FPD study shows nothing more than the fact that all persons it represented in eases closed in 1991 shared the common trait of being black; it sheds no light on whether the government discriminated as there is no showing that those without the trait were, or would be, treated differently.
Armstrong also relies on the Los Angeles Times article and two declarations presented in connection with the government’s motion for reconsideration. One is counsel’s declaration that an Impact House employee told her that in his experience dealing with crack addiction, there are an equal number of Caucasian and black users and dealers; the other is another counsel’s declaration that he had never heard of non-blacks being prosecuted in federal court whereas they were in state court. It is unclear that the district court relied on anything other than the FPD study, as it made no reference to the declarations and said it found nothing of value in the newspaper article. Regardless, assuming that it was proper for the court to consider the declarations in connection with a discovery request even though both were hearsay, neither declaration sets out any specific fact or credible evidence of the existence of discriminatory effect or intent. Counsel’s declaration about not having heard of non-blacks being prosecuted in federal court is entitled to no weight in light of the government’s evidence that non-blacks were in fact prosecuted; his further statement that “many” crack cocaine sales cases prosecuted in state court involve racial groups other than blacks lacks any relevant force because it can’t be determined whether they are similarly situated in any respect to the Armstrong defendants. By the same token, Impact House is a residential drug treatment center; it houses addicts, not dealers unless they are also addicts, and nothing in the declaration suggests that the intake coordinator’s experience is with offenders who are similarly situated to the Armstrong defendants.
In sum, there is no indication that the government has not prosecuted similarly situated non-black offenders. However, even if an inference of disproportionate impact for purposes of selective prosecution could be drawn, Armstrong’s evidence does not establish a colorable basis that he was prosecuted because of his race.
B
“ ‘Discriminatory purpose’ ... implies more than ... intent as awareness of consequences. It implies that the decisionmaker ... selected or reaffirmed a particular course of action at least in part ‘because of,’ not merely ‘in spite of,’ its adverse effects upon an identifiable group.” Wayte, 470 U.S. at 610, 106 S.Ct. at 1532 (quoting Personnel Administrator of Massachusetts v. Feeney, 442 U.S. 256, 99 S.Ct. 2282, 60 L.Ed.2d 870 (1979)). Nothing in the record indicates that the government impermissibly targeted blacks for prosecution or acted improperly in bringing the Armstrong indictment; to the *1532contrary, there is evidence that the government made its charging decision in Armstrong based on the quantity of drugs involved, the number of sales and defendants, involvement of a federal agency concerned with firearms in the investigation, multiple firearms violations intertwined with the narcotics trafficking, strength of the evidence (including tapes), the making of threats, and criminal histories having to do with both drugs and firearms. These are the very factors identified by the Supreme Court in Wayte that courts must hesitate to examine because they are intrinsic to prosecutorial discretion. 470 U.S. at 607, 105 S.Ct. at 1530.
Armstrong argues, however, that a prima facie case of purposeful discrimination may be made by showing that the totality of the relevant facts gives rise to an inference of discriminatory purpose. He submits that intent may be found even though there is no evidence that the government is hostile to the group, and the government gives a race-neutral explanation for its action, Batson v. Kentucky, 476 U.S. 79, 106 S.Ct. 1712, 90 L.Ed.2d 69 (1986); that intent may be proved circumstantially, Village of Arlington Heights v. Metropolitan Housing Development Corp., 429 U.S. 252, 266, 97 S.Ct. 555, 563-64, 50 L.Ed.2d 450 (1977); and that circumstantial evidence of discriminatory intent may include proof of disparate impact, Washington v. Davis, 426 U.S. 229, 242, 96 S.Ct. 2040, 2048—49, 48 L.Ed.2d 597 (1976). I agree, in the abstract. In limited circumstances “a clear pattern, unexplainable on grounds other than race” may be “stark” enough to infer intent, but such cases are “rare.” Arlington Heights, 429 U.S. at 266, 97 S.Ct. at 563-64, McCleskey v. Kemp, 481 U.S. 279, 293 n. 12, 107 S.Ct. 1756, 1767 n. 12, 95 L.Ed.2d 262 (1987).15 Here, the record consists of the FPD study, which shows that in 24 cases prosecuted from 1988-1991 and closed by the Federal Public Defender in 1991 the defendant was black; the government’s proffer that 11 non-black defendants were prosecuted during the period these cases were opened and Armstrong was indicted; and evidence that the Armstrong defendants were charged because of the severity of their criminal activity. In no way is this a “stark” statistical pattern, unexplainable or unexplained on grounds other than race. The statistic alone cannot, therefore, support any inference of intent.
This is clear from comparing the two cases where the Supreme Court has seen impact alone as determinative, Gomillion v. Lightfoot, 364 U.S. 339, 81 S.Ct. 125, 5 L.Ed.2d 110 (1960), and Yick Wo v. Hopkins, 118 U.S. 356, 6 S.Ct. 1064, 30 L.Ed. 220 (1886). In each, the pattern was “stark,” Arlington Heights, 429 U.S. at 266, 97 S.Ct. at 563-64, and in each, the complaining parties demonstrated the existence of similarly situated persons who were treated differently. In Gomillion, when the state legislature redrew the boundaries of the City of Tuskegee from a square into a 28-sided figure, the effect was “to remove from the city all save for only four or five of its 400 Negro voters while not removing a single white voter or resident.” Gomillion, 364 U.S. at 341, 81 S.Ct. at 127. In Yick Wo, San Francisco had enacted an ordinance banning the use of wood buddings for laundries in the absence of consent by the Board of Supervisors. There were about 320 laundries in the city of which 310 were wood, 240 were owned by Chinese, and 80 were owned by non-Chinese. One hundred fifty Chinese operators were arrested for operating a laundry in violation of the ordinance, while no non-Chinese operators were arrested. Two hundred Chinese owners petitioned for consent from the Board, and all their petitions were denied; of the non-Chinese owners, all but one was granted. Thus, unlike this case, in Gomillion and Yick Wo there was an identifiable pool against which comparisons could be made, and that comparison showed such dramatic disparities in how the groups were treated that the “conclusion was irresistible, tantamount for all practical purposes to a mathematical demon*1533stration,” Gomillion, 364 U.S. at 341, 81 S.Ct. at 127, that the state acted with discriminatory intent.
Given the absence of any evidence about any comparison group in this case, and thus of any possible statistical disparity in how different classes are treated, there is no way under existing law that the evidence in this case establishes a colorable basis for the existence of discriminatory intent.
C
The majority opinion’s logic flows from its assumption that “statistical disparities alone may suffice to provide the evidence of discriminatory effect and intent that will establish a prima facie case of selective prosecution.” Op. at 1513. For this it relies on Redondo-Lemos, 955 F.2d at 1301-02. However, what Judge Kozinski actually wrote is that, in connection with the discriminatory effect prong, “[a]t a threshold level, whether or not there is a significant disparity in the treatment of classes of defendants can normally be determined on the basis of statistical evidence.” Id. at 1301. Redondo-Lemos does not say that any statistic is a statistical disparity, or that a statistical disparity, if it does exist, proves intent no matter what its significance. Nevertheless, the majority’s opinion makes a statistic — which is all there is in this case — into a “statistical disparity”; then makes a “statistical disparity” — no matter how significant — into a prima facie showing; then asserts that the standard for discovery is lower than a prima facie showing; then holds that “inadequately explained evidence of a significant statistical disparity” suffices to show a colorable basis of discriminatory intent and effect. Op. at 1513. This syllogism simply is not the law. Statistics have to be measured against something, otherwise there can be no “disparity” because there is nothing from which to deduce a difference.
Of the need to show that others similarly situated have not been prosecuted, the opinion first says the obvious: “a defendant is not required to demonstrate that the government has failed to prosecute others who are similarly situated. He need only provide a color-able basis for believing that other similarly situated persons have not been prosecuted.” Id. at 1516. Then the opinion comes full circle by concluding that the FPD study “clearly satisfies this requirement” because “[a]t a threshold level, whether or not there is a significant disparity in the treatment of classes of defendants can normally be determined on the basis of statistical evidence, without reference to the underlying facts of individual cases.” Id. This doesn’t follow as a matter of fact, or as a matter of law.16
*1534Perhaps recognizing this, the majority opinion reveals in a footnote what its real holding is: “no ‘comparison pool’ is necessary when the record contains statistical evidence tending to show that only members of racial or ethnic minority groups have been prosecuted. Were we to conclude otherwise, we would be accepting unwarranted racial stereotypes.” Op. at 1517, n. 6. For this there is absolutely no authority. The need to identify similarly situated persons who have not been prosecuted does not evaporate when race is advanced as the constitutionally impermissible factor; rather, some evidence of those who are similarly situated except for the constitutionally impermissible factor but were not prosecuted is necessary in order to show that the defendant, who has that trait, has been prosecuted selectively. The issue is not prosecution, but selective prosecution, and there can be no selective prosecution without selection of the particular defendant for prosecution because of the constitutionally impermissible factor.
The majority opinion then holds that the government must explain “evidence of a significant statistical disparity.” Op. at 1514. There are two problems with this step. First, the FPD study does not show “significant” statistical “disparity.” It is not a random sample in which every case has the same chance of being selected, nor is a sample of 24 cases indicted over four years but closed in a single year large enough, or representative enough, to say (or surmise) anything meaningful about the history of crack prosecutions in the Central District. The second, and more important problem lies in the fact that the opinion regards the government’s explanation in this case as unpersuasive, and in so doing, essentially holds that the defendant’s statistical showing cannot be explained without the full disclosure that the discovery requested would entail.17 While the government may offer an explanation and if it does, that explanation should be considered, there is no precedent that it must make any explanation or carry any burden at the discovery stage. In any event, imposing such a burden on the government provides the next building block for the opinion, which then says: “In the absence of a response undermining the study’s accuracy, the district judge appropriately gave credence to the defendants’ showing that a study of federal cocaine base eases closed by the federal public defender over a particular period of time showed that only black defendants were involved.” Op. at 1517. Yet the *1535study’s accuracy is not at issue; its significance is. If a district court may credit a showing that cases closed by the FPD over a “particular” period of time, involved only black defendants despite contrary evidence with respect to cases opened during the same period, there is no reason that a showing that cases closed in any “particular period of time” — one month, one week, or the two-days we rejected in Bourgeois — should not suffice, regardless of how many cases the attorney handles or how many others, similarly situated to the defendant but for race, were actually prosecuted during a reasonable period of time.
The majority opinion shifts the burden to the government in another respect that is unprecedented as well. It first assumes that “[gjiven the prevalence of all kinds of drugs throughout our community, at least some crack distributors are likely to be non-blacks.” Op. at 1514. This is probably correct, but is speculation nonetheless which has no basis in the record. From this premise, however, the opinion creates a presumption: “We must start with the presumption that people of all races commit all types of crimes — not with the premise that any type of crime is the exclusive province of any particular racial or ethnic group.” Id. Although “[tjhese presumptions and premises are, of course, not conclusive,” id., the opinion gives no clue as to what they are to prove, or how they are to be disproved. Because the opinion finds that the government’s identification of some non-black crack offenders who have been charged does not undermine the force of the FPD study, id. at 1517-18, apparently the presumption cannot be overcome by a showing that of “some crack distributors [who] are likely to be non-blacks,” some have been prosecuted. If that is so, it is unclear to me that the “presumption” is rebuttable at all. Nor does the opinion suggest why such a “presumption” is needed; the party who seeks to use statistics must show what the comparison pool looks like. Moreover, to the extent that this “presumption” is intended to suggest that the government has some obligation to charge equal numbers, or proportionate numbers, of black and non-black offenders — or explain why not, it turns the law of selective prosecution on its head. The government has the right to be selective about whom it prosecutes, so long as it does not select this defendant because of & constitutionally impermissible factor. Wayte, 470 U.S. at 610, 105 S.Ct. at 1532. Indeed, instead of flipping the burden as the opinion appears to do, we have previously recognized that the defendant’s showing has to be considered “in light of the presumption of prosecutorial propriety.” United States v. One 1985 Mercedes, 917 F.2d 415, 421 (9th Cir.1990) (denying discovery because defendant had failed to produce evidence of improper government motivation sufficient to justify discovery in light of the presumption of prosecutorial propriety and the two-prong requirement of vindictive prosecution, the same as for selective prosecution).
If the majority opinion remains the law, it will be at no small cost. In this case alone, locating more than 3000 files and figuring out which were crack and firearms prosecutions, the racial identity of each defendant (presumably from the Presentence Report, because the United States Attorney’s Office has no records of race), and the investigating authorities will be a time-consuming and expensive process. The opinion is, of course, license for the same fishing expedition in legions of other cases. If a raw number— the number of cases prosecuted — suffices for discovery, I would suppose that discovery could be called for in virtually every case. For example, 100% of antitrust convictions are of white males. United States Sentencing Commission, Annual Report 1991, tables 16 and 17. Of pomographers, 92.6% are Caucasian. Id., table 16. Men comprise 83.3% of all those convicted of ail federal crimes. Id., table 17. And so on.
Resources intended for controlling crime, one of the nation’s most pressing concerns, will be chasing statistics instead. With the prospect of having to fight discovery and justify every charging decision, it will not be surprising for crack prosecutions to wane. That will disserve those who suffer from what the Los Angeles Times has just called “the plague [that] destroyed dreams, lives and families.” Los Angeles Times, Dec. 21, 1994, at A1 (headline). The opinion will also *1536force the government to keep statistics on race, turning the focus from race-neutral to race-numbered prosecutions. That is a step backwards toward race-conscious decisions.
This is not what we, or any other circuit, have permitted until now. If there are specific facts which afford a colorable basis for finding that discriminatory effect and discriminatory intent exist, the costs are worth it. Race cannot determine whether one is prosecuted or not. But, sound policy as well as separation of powers constraints demand that there be a meaningful threshold before the courts may authorize an intrusion into the discretionary workings of our co-equal branch. For these reasons, I cannot join the majority opinion.
. I refer to all of the defendants collectively as "Armstrong."
. Hon. Harlington Wood, Jr., Senior United States Circuit Judge, Seventh Circuit Court of Appeals, sitting by designation, wrote for the court. I joined that opinion, and Judge Reinhardt dissented.
. The district judge in this case ruled on the basis of the defendants’ motion for discovery.
Although the majority opinion indicates' that it does not consider the question of whether or under what circumstances a judge may rely on facts within her own experience, op. at 1514-15, it cites Redondo-Lemos, 955 F.2d at 1298, 1302, with approval on the point that district judges are well situated to discern "patterns of discrimination.” Op. at 1512. To the extent this simply means that when exercising discretion within the law, reviewing courts should be deferential to the district judge, I could not agree more. However, to the extent that the opinion means to endorse discovery based on something outside the record, it is dicta and I disagree.
. Data from the United States Sentencing Commission, for example, indicate that 87.9 percent of all those convicted for cocaine base offenses nationwide are African-American. United States Sentencing Commission Annual Report 1993, table 62.
. The First Circuit test for an evidentiary hearing is sufficient facts tending to show that the defen*1526dant has been selectively prosecuted and raising a reasonable doubt about the propriety of the prosecutor's purpose. United States v. Penagaricano-Soler, 911 F.2d 833, 838 (1st Cir.1990). The Third Circuit's test is "colorable entitlement to the defense of discriminatory prosecution....” United States v. Berrigan, 482 F.2d 171, 181 (3rd Cir.1973). The Fifth Circuit's is "colorable claim of selective prosecution,” United States v. Kohl, 583 F.2d 1351, 1355 (5th Cir.1978). The Sixth Circuit allows discovery where there is “some evidence tending to show the existence of the essential elements of the defense.” United States v. Adams, 870 F.2d 1140, 1146 (6th Cir.1989). The Seventh Circuit’s standard is a "colorable basis for the claim,” which it defines as "some evidence tending to show the essential elements.” United States v. Heidecke, 900 F.2d 1155, 1158-59 (7th Cir.1990). The D.C. Circuit requires a "colorable showing” of both elements. Attorney General of the United States v. Irish People, Inc., 684 F.2d 928, 933 (D.C.Cir.1982), cert. denied, 459 U.S. 1172, 103 S.Ct. 817, 74 L.Ed.2d 1015 (1983).
The Fourth Circuit requires a nonfrivolous showing of both elements to merit discovery. At least a legitimate issue of improper governmental conduct must be raised, and in determining whether one has been, the district court may consider the government's explanation for its conduct. United States v. Greenwood, 796 F.2d 49, 52 (4th Cir.1986) (relying on Wayte dissent). The Eleventh Circuit's standard is "colorable entitlement” which will be met by evidence that is “past the frivolous state and raise[s] a reasonable doubt as to the prosecutor’s purpose.” United States v. Gordon, 817 F.2d 1538, 1540 (11th Cir.1987), vacated on other grounds, 836 F.2d 1312 (1988) (quoting United States v. Hazel, 696 F.2d 473, 475 (6th Cir.1983)).
The Second and Eighth Circuits require that a defendant make a prima facie showing before discovery can be ordered. St. German of Alaska Eastern Orthodox Catholic Church v. United States, 840 F.2d 1087, 1095 (2d Cir.1988); United States v. Parham, 16 F.3d 844, 846 (8th Cir. 1994) (stating that the defendant's burden is a "heavy one”).
. See n. 16, infra at 1533.
. See n. 5, supra at 1525-26.
. The majority opinion also appears to make the showing required — whatever its threshold — contingent on whether the facts that would establish a colorable basis are "easily obtainable” by' defendants. Op. at 1514. It does so by relying on the dissent in Wayte, 470 U.S. at 624, 105 S.Ct. at 1539. However, all that Justice Marshall said on this point was that a standard which requires a defendant to present "some evidence tending to show the existence of the essential elements of the defense” "recognizes that most of the relevant proof in selective prosecution cases will normally be in the .Government's hands.” Id. In no way did he suggest a moving target de-. pending on “what evidence is readily available,” as the majority opinion holds. Op. at 1514. In any event, we have previously held that the fact that information might be helpful does not get defendants over the discovery hurdle. See e.g., United States v. Ness, 652 F.2d 890, 892 (9th Cir.), cert. denied, 454 U.S. 1126, 102 S.Ct. 976, 71 L.Ed.2d 113 (1981). The showing required should be a colorable basis that discriminatory effect and intent exist; it should not vacillate. depending upon whether counsel has access to more or less information, or how readily available the evidence is. See, e.g., United States v. Steele, 461 F.2d 1148, 1151 (9th Cir.1972) (Steele himself located six other similarly situated persons who had not been prosecuted).
. The reference is to United States v. Oaks, 527 F.2d 937 (9th Cir.1975), cert. denied, 426 U.S. 952, 96 S.Ct. 3177, 49 L.Ed.2d 1191 (1976). Oaks held that "[t]o sustain a claim of selective . or discriminatory prosecution, a defendant bears the burden of proving first that 'others similarly situated generally have not been prosecuted’ for similar conduct, and second that ‘his selection was based on an impermissible ground such as race, religion or his exercise of his first amend-, ment right to free speech.' " Id. at 940, quoting United States v. Scott, 521 F.2d 1188 (9th Cir. . 1975).
. Although the majority purports to overrule Gutierrez to the extent that it is inconsistent with the opinion’s view that "statistical disparities alone may suffice to provide the evidence of discriminatory effect and intent that will establish a prima facie case of selective prosecution,” op. at 1513, n. 1, I believe Gutierrez remains good law in part because I wrote it, and in part *1529because I question the license of an en banc court considering a discovery issue to overrule the standard for prevailing on the merits. Discovery cannot be unhinged from the merits; retroactively to change the merits standard to facilitate discovery is to put the cart before the horse. It is all the more remarkable to do so with respect to a decision that was not taken en banc, and that is consistent with all prior authority in this circuit, as well as others, see, e.g., Irish People, 684 F.2d 928; United States v. Huff, 959 F.2d 731, 735 (8th Cir.) (statistics showing 87% of people arrested are African-Americans insufficient to establish prima facie case), cert. denied, - U.S. -, 113 S.Ct. 162, 121 L.Ed.2d 110 (1992).
. This view is in accord with all other circuits to have considered the question. See, e.g., Irish People, 684 F.2d at 946 (reversing discovery order; no evidence others weren’t prosecuted); Kohl, 583 F.2d at 1353 (discovery denied; no showing that the government didn’t prosecute others similarly situated); Parham, 16 F.3d at 847 (no threshold showing because of absence of evidence that other’s acts of ballot fraud were tolerated without prosecution); United States v. Larson, 612 F.2d 1301, 1305 (8th Cir.) (no prima facie case; no showing that others similarly situated have not been prosecuted for conduct similar to that for which he was prosecuted), cert. denied, 446 U.S. 936, 100 S.Ct. 2154, 64 L.Ed.2d 789 (1980); cf. Hazel, 696 F.2d at 475 (arguably asserting that 34 other members of tax revolt group had not been prosecuted suffices for first prong); Greenwood, 796 F.2d at 52 (discovery denied; contention that five similarly situated white agents were not prosecuted groundless).
. In Irish People, the D.C. Circuit similarly concluded that "it makes sense to require a color-able claim of both [parts of the two-element test for selective prosecution] before subjecting the Government to discovery,” explaining:
Defendant must make a colorable showing that he has been especially singled out, that there exist persons similarly situated who have not been prosecuted. Without such a showing, it is irrelevant that the British or Irish may also have wanted the suit brought.
... It is clear ... that a demonstration of selection is indispensable for the defense and that the burden of so demonstrating lies squarely on the defendant.
... Where defendant cannot show anyone in a similar situation who was not prosecuted, it *1530has been held that he has not “met even the threshold point of the Yick Wo doctrine — ‘official discrimination ... between persons in similar circumstances, material to their rights....' ”
684 F.2d at 946-47. See also Kerley, 787 F.2d at 1150 (to discover government documents defendant must introduce some evidence tending to show the existence of the essential elements of the defense, including that he was singled out for prosecution while others similarly situated haven’t been).
. The majority opinion’s statement that "the factor used to choose the cases examined in the study — whether the case had been closed — was a factor independent of and not correlated with race,” op. at 1517, n. 7, is without any support in the record. It could as well be said — equally without any support in the record — that the reason closed cases instead of opened cases was chosen was because (as the government's evidence shows) non-blacks were prosecuted in cases that were opened during 1988-1991 but were closed in some year other than 1991.
. The majority opinion “fail[s] to see the difference" between cases closed and cases opened. Op. at 1517, n. 7. They are, of course, opposite sides of the same coin. Whether looked at through the lens of "opened” or "closed,” the number of black defendants prosecuted on crack charges, standing alone as it does here without reference to whether non-black crack dealers who sold large quantities and used firearms and had priors and were caught in the act on tape were not prosecuted, has no tendency to prove the elements of selective prosecution. However, on the theory adopted by the majority opinion— that the number prosecuted is alone sufficient — it does make a difference whether the focus is on cases opened or cases closed. The opinion fixes only on the closed aspect of the closed cases in the FPD study, whereas the government's evidence focuses, correctly, on the opened aspect of the closed cases as well as others which were prosecuted against non-black defendants who were either not represented by the FPD at all or at least not in cases that were closed in 1991. Cases closed can’t have any tendency to prove discrimination except to the extent that they show what cases were brought, or opened, and thus what was motivating the United States Attorney’s Office when it decided to file charges against Armstrong. The cases closed in 1991 were indicted in 1988, 1989, 1990, and 1991. The evidence shows that non-black defendants were also indicted during that period. Thus, contrary to the majority opinion, the government’s showing directly rebuts both the "accuracy” of the study (to the extent that accuracy is in issue, which it really isn’t) and its probative value on the elements of selective prosecution (which is in issue).
. Jury-selection cases, see, e.g., Castaneda v. Partida, 430 U.S. 482, 97 S.Ct. 1272, 51 L.Ed.2d 498 (1977), are somewhat of an exception because of the nature of the task, McCleskey, 481 U.S. at 293 n. 13, 107 S.Ct. at 1768 n. 13, but they are inapposite for the additional reason that the statistics in those cases show a disparity between the population and those summoned for duty. As there is no comparison group here, no disparity can be shown.
. There has been evidence of disparity (not just a study of numbers prosecuted) in each of the few cases in which a finding of selective prosecution or an order for discovery has been upheld on appeal. See, e.g., Steele, 461 F.2d at 1151 (conviction reversed; defendant located and presented evidence of six other persons who, like him, had refused on principle to complete census forms but who, unlike him, had not taken a public stand against the census and had not been prosecuted); Adams, 870 F.2d at 1146 (discovery ordered; record suggests that taxpayers who un-derreport income and voluntarily amend returns and pay deficiency, as defendants did, have not been prosecuted, though defendants were); Gordon, 817 F.2d at 1540 (colorable entitlement shown and discovery and evidentiary hearing affirmed; magistrate judge found that defendants presented some evidence of similar violations by other persons who have not been prosecuted); cf. United States v. Berrios, 501 F.2d 1207, 1211— 12 (2d Cir.1974) (declining to reverse order for hearing and production of government memorandum in camera based on counsel’s affidavit that hundreds of other union officials had prison records and only three had been prosecuted, but indicating that, to show a “colorable basis” for discovery "we would first require some evidence tending to show the existence of the essential elements of the defense” and characterizing existence of unprosecuted violations of statute defendant was prosecuted under as an "essential element”); United States v. Cammisano, 413 F.Supp. 886 (W.D.Mo.1976), vacated on other grounds, 546 F.2d 238 (defendants demonstrated many persons in apparently similar circumstances who violated Federal Meat Inspection Act, but who have not been prosecuted). In addition, in all but Berrios, there was direct evidence that the prosecution was brought for an impermissible reason. In Steele, the government admitted that it was concerned about census protestors and provided no non-discriminatoiy explanation for prosecuting Steele; in Gordon, a Department of Justice official said that the voting fraud investigation was brought about by "arrogance on the part of the blacks”; in Adams, a government official acknowledged that the defendant was prosecuted as “revenge" for her filing an EEOC claim; and in Cammisano, an FBI agent threatened defendant.
*1534By the same token, district court orders for discovery which have not been based on a finding that others similarly situated have not been prosecuted have been reversed. See, e.g., Wayte, 710 F.2d at 1387 (reversing discovery order on the footing that Wayte had not shown he was selected from the larger group because of his exercise of constitutional rights, but noting that the first element had been established because the district court had found that over 500,000 eligible men had not registered for the draft whereas only 12 others in addition to Wayte had been indicted for failure to register, all of whom were vocal nonregistrants); Kerley, 787 F.2d at 1150; Irish People, 684 F.2d at 946.
. In this connection, the majority opinion asserts that the government’s response "does not speak to the inference that blacks are more frequently prosecuted for federal crack offenses than even their allegedly disproportionate involvement with the drug would justify.” Op. at 1518. The assertion implies that the response somehow should have spoken to the issue of state and federal prosecutions and is deficient because it didn’t. This is incorrect because the defendants ’ showing says nothing about state and federal prosecutions of similarly situated offenders so there is nothing to respond to. More important, it slides into the balance of whether discovery should be ordered a consideration federal courts of appeal — including ours — have uniformly rejected as implicating equal protection: the fact that higher federal sentences for cocaine base offenses fall disproportionately on blacks. United States v. Singleterry, 29 F.3d 733 (1st Cir.), cert. denied, - U.S. -, 115 S.Ct. 647, 130 L.Ed.2d 552 (1994); United States v. Frazier, 981 F.2d 92 (3rd Cir.1992), cert. denied, - U.S. -, 113 S.Ct. 1661, 123 L.Ed.2d 279 (1993); United States v. D'Anjou, 16 F.3d 604 (4th Cir.), cert. denied, - U.S. -, 114 S.Ct. 2754, 129 L.Ed.2d 871 (1994); United States v. Galloway, 951 F.2d 64 (5th Cir.1992); United States v. Reece, 994 F.2d 277 (6th Cir. 1993); United States v. Chandler, 996 F.2d 917 (7th Cir. 1993); United States v. Lattimore, 974 F.2d 971 (8th Cir.1992), cert. denied, - U.S. -, 113 S.Ct. 1819, 123 L.Ed.2d 449 (1993); United States v. Coleman, 24 F.3d 37 (9th Cir.), cert. denied, — U.S. -, 115 S.Ct. 261, 130 L.Ed.2d 181 (1994); United States v. Easter, 981 F.2d 1549 (10th Cir.1992); United States v. King, 972 F.2d 1259 (11th Cir.1992), cert. denied, - U.S. -, 113 S.Ct. 2448, 124 L.Ed.2d 665 (1993); United States v. Thompson, 27 F.3d 671 (D.C.Cir.), cert. denied, - U.S. -, 115 S.Ct. 650, 130 L.Ed.2d 554 (1994).