The Court voted to rehear this appeal in banc to consider the harmless error analysis applicable on review of a collateral attack upon a conviction obtained after a jury charge that, in conformity with then prevailing law, see United States v. Caming, 968 F.2d 232, 238-41 (2d Cir.1992), did not include an aspect of the mental state necessary for conviction, an aspect subsequently required, see Ratzlaf v. United States, 510 U.S. 135, 148, 114 S. Ct. 655, 663, 126 L. Ed. 2d 615 (1994). The Supreme Court having clarified harmless error analysis in California v. Roy, --- U.S. ----, 117 S. Ct. 337, 136 L. Ed. 2d 266 (1996), the in banc court dissolves itself and leaves the appeal for reconsideration by the original panel in light of Roy.

I concur in the decision of the in banc Court to dissolve and return this appeal to the panel for further consideration in light of the Supreme Court's recent decision in California v. Roy, --- U.S. ----, 117 S. Ct. 337, 136 L. Ed. 2d 266 (1996), but add these views in an effort to identify and illuminate uncertainties that have been created by the way the Supreme Court has explicated its recent harmless error jurisprudence in the context of constitutional errors.

I. Harmless Error Jurisprudence Before California v. Roy

Formulating an approach for deciding criminal cases in which a constitutional error is claimed to be harmless requires consideration of two distinct issues: what standard is to be applied in determining whether an error is harmless, and how certain must a court be that an error is harmless, under the applicable standard, before it may leave a conviction undisturbed. Though these issues are logically distinct, courts have not always considered them separately. Prior to California v. Roy, three Supreme Court decisions had established the basic framework for considering claims that a constitutional error was harmless.

In Chapman v. California, 386 U.S. 18, 87 S. Ct. 824, 17 L. Ed. 2d 705 (1967), the Supreme Court considered both the issues of the standard of harmlessness and the degree of certainty required. As to the standard, the Court said the question is whether the error "contribute[d] to the verdict," id. at 24, 87 S.Ct. at 828, adopting the standard it had previously enunciated in Fahy v. Connecticut, 375 U.S. 85, 86-87, 84 S. Ct. 229, 230-31, 11 L. Ed. 2d 171 (1963). As to the degree of certainty, Chapman ruled that the prosecution bore a burden of proving that the error was harmless "beyond a reasonable doubt." Chapman, 386 U.S. at 24, 87 S. Ct. at 828.1

In Brecht v. Abrahamson, 507 U.S. 619, 113 S. Ct. 1710, 123 L. Ed. 2d 353 (1993), the Court ruled that for "constitutional error of the trial type," id. at 638, 113 S.Ct. at 1722, the Chapman test applies only to direct review and that on collateral attack the standard for harmless error, drawn from Kotteakos v. United States, 328 U.S. 750, 66 S. Ct. 1239, 90 L. Ed. 1557 (1946), is "whether the error 'had substantial and injurious effect or influence in determining the jury's verdict.' "2 Brecht, 507 U.S. at 637, 113 S.Ct. at 1721 (quoting Kotteakos, 328 U.S. at 776, 66 S.Ct. at 1253). Brecht did not precisely articulate the degree of certainty a reviewing court must have in determining whether the standard of "substantial and injurious effect or influence" had been met, but the opinion makes clear that the "beyond a reasonable doubt" formulation of Chapman is being rejected. Id. at 636-37, 113 S.Ct. at 1721-22.

In O'Neal v. McAninch, 513 U.S. 432, 115 S. Ct. 992, 130 L. Ed. 2d 947 (1995), the Court determined the degree of certainty a reviewing court must have in order to conclude that a constitutional error, raised in a habeas corpus petition, was harmless under the Brecht standard. A six-member majority ruled that if the reviewing court is in "grave doubt" as to whether the error is harmless, the habeas petitioner wins. Id. at ----, 115 S.Ct. at 994. The Court explained that by "grave doubt" it meant that the issue of harmlessness is "so evenly balanced" that the judge feels "in virtual equipoise" on the issue.3 Id.

In addition to the Chapman and Brecht/O'Neal approaches to harmless error, the Court introduced a third approach in Carella v. California, 491 U.S. 263, 109 S. Ct. 2419, 105 L. Ed. 2d 218 (1989). Carella, which also involved collateral review, concerned a jury charge that informed the jurors of two mandatory presumptions applicable in a criminal case.4 There was no objection to the charge. The Supreme Court ruled that the charge was constitutionally erroneous, and reversed and remanded for consideration of whether the error was harmless. The error would be harmless, the majority opinion ruled, if "no rational jury could find the predicate acts [underlying the presumption] but fail to find the fact presumed." Id. at 267, 109 S.Ct. at 2421. Justice Scalia wrote separately, for himself and three other Justices, to make clear "that the harmless-error analysis applicable in assessing a mandatory conclusive presumption is wholly unlike the typical form of such analysis." Id. (Scalia, J., with whom Brennan, Marshall, and Blackmun, JJ., join, concurring in the judgment). In the "usual case," he wrote, where the inquiry is whether erroneous admission of evidence requires a reviewing court to reverse a conviction, the court considers the trial record as a whole to decide "whether the fact supported by the improperly admitted evidence was in any event overwhelmingly established by other evidence." Id. But with a conclusive presumption, he continued, a reviewing court may not make its own guess from the record as a whole as to whether a properly instructed jury would have found the presumed fact. Instead, the task for the reviewing court is to determine whether "the predicate facts relied upon in the instruction, or other facts necessarily found by the jury, are so closely related to the ultimate fact to be presumed that no rational jury could find those facts without also finding that ultimate fact." Id. at 271, 109 S.Ct. at 2423-24. In other words, the analysis concerns not what a properly instructed jury would have found, but what the jury actually found. In Justice Scalia's terms, whenever the test he enunciated is met, the findings the jurors did make "is functionally equivalent to finding the element required to be presumed." Id.5

The Court's approach in Carella, especially as explicated in Justice Scalia's concurrence, can be viewed as either a third standard for determining harmless error or a different way of determining whether the Brecht standard of harmlessness has been met. If a court considers Carella to set forth a standard, it would ask whether it concludes, upon examination of the facts the jury did find, that no reasonable jury could fail to find the presumed fact. If a court considers Carella to set forth a different way of meeting the Brecht standard, it would ask whether the erroneous charge had a substantial and injurious effect or influence on the verdict, and it would answer that question by inquiring whether it concludes, upon examination of the facts the jury did find, that no reasonable jury could fail to find the presumed fact; if so, there was no injurious effect upon the verdict.

The initial uncertainty arising from the articulation of these differing standards, especially the two approaches applicable to collateral review (Brecht/O'Neal and Carella/O'Neal ),6 concerned the types of errors to which these approaches applied. Some clarification as to the scope of the Brecht/O'Neal approach had begun to emerge when the Supreme Court decided Arizona v. Fulminante, 499 U.S. 279, 111 S. Ct. 1246, 113 L. Ed. 2d 302 (1991). The issue was whether harmless error analysis of any sort applied to the erroneous admission of a coerced confession. A five-member majority held that it did. Id. at 302, 111 S.Ct. at 1260 (part II of opinion of Rehnquist, C.J., with whom O'Connor, Scalia, Kennedy, and Souter, JJ., join). The Chief Justice drew a distinction between "trial error," which he described as "error which occurred during the presentation of the case to the jury," id. at 307, 111 S.Ct. at 1263, and "structural defects," which he described as a defect "affecting the framework within which the trial proceeds, rather than simply an error in the trial process itself." Id. at 310, 111 S.Ct. at 1265. "Trial errors," the majority ruled, were subject to harmless error analysis; "structural defects" were not, since they "defy analysis by 'harmless-error' standards." Id. at 309, 111 S.Ct. at 1265. As illustrations of "structural defects," the Chief Justice cited cases involving "total deprivation of the right to counsel," id. (citing Gideon v. Wainwright, 372 U.S. 335, 83 S. Ct. 792, 9 L. Ed. 2d 799 (1963)), "a judge who was not impartial," id. (citing Tumey v. Ohio, 273 U.S. 510, 47 S. Ct. 437, 71 L. Ed. 749 (1927)), "unlawful exclusion of members of the defendant's race from a grand jury," id. at 310, 111 S. Ct. at 1265 (citing Vasquez v. Hillery, 474 U.S. 254, 106 S. Ct. 617, 88 L. Ed. 2d 598 (1986)), "the right to self-representation," id. (citing McKaskle v. Wiggins, 465 U.S. 168, 177-78, n. 8, 104 S. Ct. 944, 950-51 n. 8, 79 L. Ed. 2d 122 (1984)), and "the right to public trial," id. (citing Waller v. Georgia, 467 U.S. 39, 49 n. 8, 104 S. Ct. 2210, 2217 n. 8, 81 L. Ed. 2d 31 (1984)).

The taxonomy of constitutional errors outlined in Fulminante appeared likely to place most, and perhaps all, errors in jury instructions into the category of "trial type" errors, to which some form of harmless error analysis would apply, but left it uncertain whether the same harmless error approach applicable to erroneous evidentiary rulings would apply to constitutional errors in jury charges.

Brecht itself did not resolve this uncertainty, since the error there concerned the prosecutor's improper use of the defendant's post-Miranda silence for impeachment purposes, an error, like the error in Fulminante, relating to evidence.

Six weeks after Brecht, the Court ruled in Sullivan v. Louisiana, 508 U.S. 275, 113 S. Ct. 2078, 124 L. Ed. 2d 182 (1993), that at least one type of error in a jury instruction is a "structural" error, not subject to any form of harmless error analysis. The issue was whether a constitutionally deficient instruction on reasonable doubt could be harmless error. A unanimous Court ruled that it could not. Justice Scalia, the author of the Carella concurrence, wrote the Court's opinion, in which all the Justices joined. The Chief Justice, the author of Brecht, wrote a separate concurring opinion.

Justice Scalia supported his ruling with two modes of analysis. The first built upon his concurring opinion in Carella. As he had said of the jury charge error concerning presumptions in Carella, Justice Scalia ruled that the jury charge error concerning reasonable doubt in Sullivan could not be assessed under traditional harmless error analysis, i.e., inquiring on the basis of the entire record what the jury would have found in the absence of the error. Sullivan, he noted, was a weaker case for harmless error analysis than Carella. At least in Carella, he said in Sullivan, the reviewing court could examine the predicate facts that the jury did find in order to apply the conclusive presumption and uphold the conviction if those facts are so closely related to the presumed fact that no rational jury could find the former without finding the latter. 508 U.S. at 280-81, 113 S.Ct. at 2082-83. In Sullivan, however, where the jury was improperly charged as to reasonable doubt, no facts have been constitutionally found, and there "is no object, so to speak, upon which harmless error scrutiny can operate." Id. at 280, 113 S.Ct. at 2082 (emphasis in original). It was therefore improper to speculate upon what verdict a properly instructed jury would have returned.

After concluding his discussion about the absence of any jury fact-finding that could support even a Carella analysis of harmless error, Justice Scalia then turned to what he called "[a]nother mode of analysis." Id. at 281, 113 S.Ct. at 2082. In one brief paragraph, he categorized the erroneous reasonable doubt instruction as a "structural error," as to which Fulminante renders harmless error analysis inapplicable.

The Chief Justice's concurring opinion suggested that even a constitutionally deficient reasonable doubt instruction might be subject to harmless error analysis, but ultimately agreed with the Court that it was not. Notably, the Chief Justice contrasted the defective reasonable doubt instruction in Sullivan with the conclusive presumption instruction in Carella, which was subject to harmless error analysis. 508 U.S. at 284, 113 S.Ct. at 2084.

Sullivan made clear that at least one type of constitutional error in a jury charge was not subject to any harmless error analysis. It did not resolve, however, whether other instruction errors were also not amenable to harmless error analysis, nor whether instruction errors that were amenable to harmless error analysis were to be assessed on the record as a whole to determine what a properly instructed jury would have found (Brecht ) or by examining what facts the jury did find in order to determine whether the facts found were "functionally equivalent" to the element incorrectly charged (Carella concurrence).

However, the four Justices concurring in Carella had left no doubt that in their view the approach applied by the Court in Carella and explicated by the concurrence should be applied to jury charge errors beyond those concerning conclusive presumptions. Justice Scalia said that his analysis applies not only to conclusive presumption instructions but also to an instruction containing a "misdescription of an element of the offense"; both, he said, "deprive[ ] the jury of its factfinding role, and must be analyzed similarly." Carella, 491 U.S. at 270, 109 S.Ct. at 2423. Presumably, Justice Scalia would apply his analysis to an instruction that omitted an element of an offense, a circumstance that also "deprives the jury of its factfinding role."

The uncertainties as to whether jury charge errors concerning an omission or a misdescription of an element were "trial type" errors subject to harmless error analysis and, if so, whether at least some of these errors were to be analyzed under the approach of the Carella concurrence remained until the decision in California v. Roy.

Against this background, the Supreme Court decided California v. Roy, --- U.S. ----, 117 S. Ct. 337, 136 L. Ed. 2d 266 (1996). Roy concerned a federal collateral attack upon a state court conviction for first-degree murder and robbery. The prosecution's theory, to the extent viable in light of the jury's verdict, see Roy v. Gomez, 81 F.3d 863, 864-65 (9th Cir.1996) (in banc), was that Roy had helped his confederate commit robbery and thereby aided commission of a felony murder. The jury charge required for an aiding and abetting conviction knowledge of the confederate's unlawful purpose and some action to aid commission of the crime. Id. at 865. After Roy's trial, the California Supreme Court ruled that an aiding and abetting instruction required the jury to find not only (a) knowledge of the confederate's unlawful purpose and (b) action to assist its accomplishment, but also (c) intent to encourage or facilitate the unlawful purpose. See People v. Beeman, 35 Cal. 3d 547, 561, 199 Cal. Rptr. 60, 68, 674 P.2d 1318, 1326 (1984).

Since no objection to the charge on intent had been made at the trial, which occurred before the decision in Beeman, the issue for the federal habeas court was whether the trial judge's failure to charge that an aider must intend to help his confederate was harmless. That issue, it bears repeating, comprehended two subsidiary issues: (a) what standard does a federal habeas court apply in determining whether an error of this sort is harmless, and (b) how sure must the federal habeas court be that the error is harmless before it may deny habeas relief.

A majority of the Ninth Circuit in banc court ruled that the appropriate standard was that set forth by the four-Justice plurality opinion in Carella. In the Ninth Circuit's view, "the omission [of the required intent instruction] is harmless only if review of the facts found by the jury establishes that the jury necessarily found the omitted element." Roy v. Gomez, 81 F.3d at 867 (emphasis in original). Following the Carella concurrence, which had concerned harmless error analysis in the context of a presumption, the Ninth Circuit "treated the omitted element as the 'presumed fact' and considered whether a rational jury could have found the remaining elements of the offense without also finding the omitted element." Id. at 866.

The Ninth Circuit also noted that the error in the jury charge could be described as either the omission of an element, which the Court labeled "specific intent," or the misdescription of an element, which the Court labeled "intent." Id. at 867 n. 4. Whether considered an omission or a misdescription, the Court felt obliged to apply the analysis of the Carella concurrence, id., since Justice Scalia had written that "misdescription of an element of the offense ... deprives the jury of its factfinding role, and must be analyzed similarly [to a conclusive presumption]." Carella, 491 U.S. at 270, 109 S.Ct. at 2423.

The Supreme Court reversed and ruled that the Brecht standard, not the standard of the Carella concurrence, applied to determination of the harmlessness of the error in the Roy jury charge. The Supreme Court's brief per curiam opinion first noted that the Carellaconcurrence "set out the views of several Justices about the proper way to determine whether an error in respect to the use of a presumption was 'harmless.' " Roy, --- U.S. at ----, 117 S. Ct. at 338.7 The Court then noted that, after Carella, the Kotteakos standard had been held appropriate for "trial errors." Id. (citing Brecht ). The Court concluded that the error in Roy was a trial error, subject to the Brecht standard of harmlessness. Id. at ----, 117 S.Ct. at 339. The Court noted, in agreement with the Ninth Circuit, that the jury charge error could be characterized as either an omission or a misdescription of an element and labeled the error as "an error in the instruction that defined the crime." Id.

Left uncertain by the Roy per curiam was whether all errors in instructions that "define[ ] the crime" are subject to the Brecht standard, or whether that standard applies only to errors that concern fairly narrow departures from proper charge language. The charge in Roy had not omitted the element of intent altogether, but had omitted (or misdescribed) only an aspect of the required mental state--and an aspect that was somewhat similar to the ingredients that were correctly stated.

What did seem clear from the Roy per curiam, however, was that the harmless error analysis of the Carella concurrence was relegated at least primarily and perhaps exclusively to jury charge errors that concerned conclusive presumptions. But even that conclusion, and perhaps the entire meaning of the Roy per curiam, was placed in considerable doubt by the concurring opinion in Roy, written by Justice Scalia for himself and Justice Ginsburg.

Justice Scalia began by stating both what he understood the Court to have decided and what he understood had not been decided. He wrote:

I agree with what the Court decides in its per curiam opinion: that the Brecht-O'Neal standard for reversal of the conviction ("grave doubt as to the harmlessness of the error") rather than the more stringent Chapman standard (inability to find the error "harmless beyond a reasonable doubt") applies to the error in this case when it is presented, not on direct appeal, but as grounds for habeas corpus relief. The Ninth Circuit did not apply that more deferential standard, and I therefore concur in the remand.

I do not understand the opinion, however, to address the question of what constitutes the harmlessness to which this more deferential standard is applied--and on that point the Ninth Circuit was quite correct.

Id. at ----, 117 S.Ct. at 339 (Scalia, J., concurring).

Thus, for Justices Scalia and Ginsburg the Supreme Court's reversal in Roy concerned only the degree of certainty the reviewing court must have as to whether the error was harmless and did not even "address" the standard for assessing harmlessness when a jury charge omits (or misdescribes) an element of an offense. This characterization of Roy is somewhat surprising in view of the Ninth Circuit's in banc opinion, which the Supreme Court rejected. That opinion had explicitly noted that (a) the Chapman approach was not applicable on collateral review, (b) Brecht stated the correct standard (whether the error had substantial and injurious effect or influence in determining the jury's verdict), (c) O'Neal stated how sure the reviewing court must be that the Brecht standard has been met, and (d) the Carella concurrence stated the correct mode of analysis for assessing the harmlessness of the omission of an element from a jury charge. Roy v. Gomez, 81 F.3d at 868. The Ninth Circuit had thus combined the standards of Brecht and the Carella concurrence, and invoked the certainty test of O'Neal.

Undaunted by the Roy per curiam 's apparent relegation of the Carella concurrence to presumption cases, see Roy, --- U.S. at ----, 117 S. Ct. at 338, Justice Scalia stated that the analysis set forth in that concurrence applied to the error in Roy because, as in Carella, the jury had not rendered a verdict on each element of the crime.8 There could not be speculation, he continued, as to what a properly instructed jury would have found. Instead, "[t]he error in the present case can be harmless only if the jury verdict on other points effectively embraces this one [the uncharged aspect of intent] or if it is impossible, upon the evidence, to have found what the verdict did find without finding this point as well." Id. at ---- - ----, 117 S. Ct. at 339-40 (emphasis in original) (citing the Carella concurrence). He concurred, he said, "so that the Ninth Circuit may determine whether there is 'grave doubt' that this is so, rather than (what it did) determine whether it is impossible to 'be certain' that this is so." Id. at ----, 117 S.Ct. at 340 (citing Roy v. Gomez, 81 F.3d at 867).

In the absence of the Roy concurrence, I would have thought that the Roy per curiam required the Brecht standard for harmlessness, rather than the standard of the Carella concurrence, to be applied to the jury charge error in Roy. Even with the Roy concurrence, there is a strong argument that this is the view of the seven Justices who joined the Roy per curiam opinion. But when Justices Scalia and Ginsburg state that the Court has not even addressed the proper standard and has decided only how sure the reviewing court must be that the standard has been met (O'Neal rather than Chapman ), I am left in some uncertainty.

Even if Roy has decided that the Brecht standard applies to the charge error in that case, a further uncertainty would remain. It would not be clear whether the Brecht standard is applicable only because the words omitted from the jury charge are so closely related to the words that were included, as they were in Roy, or whether the harmlessness of any "error in the instruction that defined the crime," Roy, --- U.S. at ----, 117 S. Ct. at 339, is also to be assessed under the Brecht standard. In Roy, the charge obliged the jury to find one aspect of mens rea, knowledge of the confederate's purpose, but omitted (or misdescribed) another aspect, intent to aid. What if the charge had omitted an element altogether? May the reviewing court examine the evidence and predict that a properly instructed jury would have found the element beyond a reasonable doubt? If such an analysis may be made where an entire element is omitted (one not even related to what was included in the charge), is it applicable only to technical elements such as FDIC insurance in a bank robbery case or interstate commerce in a RICO case, or is it also applicable to omission of an element like mens rea?

A further uncertainty, inherent in most cases considering claims of harmless error, is whether the reviewing court is to consider the effect of the error on the jury or predict what verdict would have been rendered in the absence of the error. See Harry T. Edwards, To Err Is Human, But Not Always Harmless: When Should Legal Error Be Tolerated?, 70 N.Y.U.L.Rev. 1167 (1995) (Madison Lecture). For Justice Scalia, in those cases in which the Carella standard is applicable, i.e., where the jury has not made a required finding, the answer is clear: the reviewing court should consider the effect of the error and examine only the facts that the jury found. Whether other Justices agree is somewhat doubtful. Where the Carella standard is not applicable, it seems likely that most Justices would be willing to predict what verdict would have been rendered in the absence of the error, i.e., consider how strong was the evidence of guilt that was unaffected by the error.9

The principal uncertainty created by Roy (which standard is applicable) is heightened by the fact that the per curiam opinion does not explicitly reckon with and reject the analysis of the per curiam opinion set forth by Justices Scalia and Ginsburg. I recognize that those writing the principal opinion for a multi-judge court are not obliged to respond to everything expressed in an opinion by other members of the court, whether in dissent or in concurrence, and surely we cannot expect one Supreme Court decision to resolve more issues than the case presents. Nevertheless, Roy presents a situation where a clear resolution of at least the principal uncertainty created by the per curiam and concurring opinions seems warranted. Habeas corpus challenges to state court convictions are among the most frequently litigated matters in the federal district courts and courts of appeals, and assessment of harmlessness is probably the single most recurring issue presented in such cases. On such a vital matter as the appropriate mode of analysis for determining whether a constitutional error is harmless, we ought not to be left with the unresolved tension between the Roy per curiam and concurrence. See Joseph Goldstein, The Intelligible Constitution: The Supreme Court's Obligation to Maintain the Constitution as Something We the People Can Understand passim (1992); id. at 133 (urging the Court to develop "canons of comprehensibility"). Possibly some clarification might be forthcoming in the near future. See United States v. Johnson, 82 F.3d 429 (11th Cir.1996) (table), cert. granted, --- U.S. ----, 117 S. Ct. 451, 136 L. Ed. 2d 346 (1996) (raising, on direct review of federal conviction, issues concerning plain error analysis to be applied where element of materiality was not submitted to jury).

Without the benefit of Roy, there was considerable justification for our Court to rehear the pending appeal in banc to explore and hopefully resolve the differing views as to the harmlessness standard that were well articulated by the panel majority and dissenting opinions. See Peck v. United States, 73 F.3d 1220 (2d Cir.1995) (majority opinion of Judge Mahoney); id. at 1229 (dissenting opinion of Judge Walker). Now that Roy has been decided, it is appropriate to return the matter to the panel for a fresh consideration of the appeal in light of Roy. In all likelihood, the panel will be able to dispose of the appeal in a manner that will eliminate any issue warranting in banc rehearing. If not, the in banc Court may be reconvened. The ultimate solution, however, is for the Supreme Court to articulate, authoritatively and unambiguously, the mode of analysis applicable on habeas corpus review when jury instruction errors are assessed for harmless error.