Concept
Harmless Constitutional Error
Harmless error
Harmless Constitutional Error
Rule 52(a) directs a federal court to disregard any error, defect, irregularity, or variance that does not affect substantial rights.1 The previous section showed how a court decides whether an ordinary preserved error crosses that line. A harder question arises when the error is a violation of the Constitution itself. The rule's breadth suggests that a constitutional violation might be forgiven too, yet guarantees like the privilege against self-incrimination exist to restrain the government, and it is not obvious that a conviction obtained in breach of one should stand on the theory that the breach probably made no difference. For a time it was thought that constitutional error in a criminal case called for automatic reversal, with no inquiry into whether the error affected the result, and Chapman v. California is the decision that repudiated that view.2
The facts of Chapman show why the question mattered. The defendants' trial featured repeated references to their failure to testify, which violated the Fifth and Fourteenth Amendments.3 They asked the Court to hold that every federal constitutional error must always be deemed harmful, which would have required automatic reversal and ended the discussion.4 The Court declined. Harmless-error rules, whether state statutes or the federal statute behind Rule 52(a), serve the useful purpose of blocking reversals for small errors that had little likelihood of changing the result, and none of them on its face separates constitutional errors from errors of state law or ordinary federal rules.5
Chapman did not, however, hand constitutional error over to the ordinary standard. The Court recognized that harmless-error rules can produce unfair results when highly persuasive forbidden evidence or argument reaches a trial in which guilt is close, and it concluded that a constitutional error of that kind casts a burden on someone other than the person it prejudiced to show that it was harmless.6 The reasoning follows from the structure of the problem. A reviewing court cannot look inside the jury room, so some party must bear the risk that the error's effect cannot be known. The prosecution is the party that benefited from a violation of the accused's rights, and placing the risk on the defendant would require proof of the effect of an error the government itself caused. The Court also described this allocation as a return to the original common-law practice, which put the burden on the beneficiary of the error to show there was no injury or to lose the judgment it had erroneously obtained.7
The rest of this section follows the three steps that Chapman and the decisions built on it supply. The first step is that constitutional error can be harmless at all, and that federal law rather than a State's own harmless-error rule decides the question when a federal right has been violated.8 The second step is the test itself, which demands a level of confidence from the reviewing court that exceeds the fair-assurance inquiry used for ordinary error.9 The third step is a sorting line, because the Court has since confirmed that most constitutional errors can be harmless while reserving a short list of errors that can never be treated that way.10 The first step comes first because the other two depend on it, and it turns on what Chapman refused to do when the defendants asked for automatic reversal.
Constitutional Error Can Be Harmless
Constitutional guarantees exist to restrain the government, so a conviction obtained in breach of one looks like a conviction that ought to fall, whatever the breach did to the outcome. The petitioners in Chapman v. California built their request for automatic reversal on that intuition, and its logic is tidy: if every federal constitutional error is deemed harmful, no court ever has to measure an error's effect.11 The Court declined to adopt any such rule. It held instead that some constitutional errors, in the setting of a particular case, are "so unimportant and insignificant" that they may be deemed harmless without reversing the conviction.12 The rule that follows is no automatic reversal for constitutional error: the violation opens the inquiry, and the effect of the violation on the verdict decides the outcome.
The rule has two supports, and each has its own logic. The first is functional. Rule 52(a) already tells courts to disregard errors that leave substantial rights untouched,13 and harmless-error rules, state and federal, serve to block the setting aside of convictions for small errors that had little likelihood of changing the result. None of them on its face separates constitutional errors from errors of state law or ordinary federal rules.14 A retrial costs witnesses, victims, and courts the same effort whether the underlying error was constitutional or not, and for an error that could not have changed the verdict it purchases only a second jury hearing the same case. The second support concerns how much constitutional errors differ. A passing remark and the wholesale loss of a defense can each be constitutional errors, yet they differ enormously in what they did to the jury's decision. The word "constitutional" identifies the right that was infringed, while harmlessness asks what the infringement did to the verdict, so a single rule of reversal would treat the trivial and the decisive alike. The permission is also conditional, because a reviewing court must be able to declare the error harmless beyond a reasonable doubt, a demanding standard taken up in the next section.15
Before reaching harmlessness, Chapman had to settle whose rule supplies the answer. A state's own harmless-error rule governs when the error involves only state procedure or state law. A denial of rights guaranteed by the Fifth and Fourteenth Amendments raises a federal question, because whether a conviction should stand after a state has failed to honor a federally guaranteed right is as much a federal question as what the right guarantees.16 The reason is uniformity of protection. The right comes from the Federal Constitution, and if each state formulated its own rules about whether a violation of that right matters, the protection would vary across state lines and would be weakest where a state's rule was most forgiving. The Court accordingly said it could not leave to the states the formulation of the rules and remedies for infractions of federally guaranteed rights.17 In Chapman itself the state conviction was tainted by comment on the defendants' silence, so federal law decided whether the error could stand.18
Chapman did not remain an abstract permission. In the decades after it, the Court applied harmless-error analysis to a long list of constitutional errors, including improper comment on a defendant's silence, restrictions on cross-examining a witness for bias, jury instructions that misstate an element or contain an erroneous presumption, and denial of counsel at a preliminary hearing, and it recognized that most constitutional errors can be harmless.19 A concurring opinion in United States v. Gaudin put the point as a "strong presumption" that a constitutional violation will be subject to harmless-error analysis.20 The reach extends even to a coerced confession. Until 1991 the admission of a confession found coerced under the Fourteenth Amendment could not be deemed harmless, but in Arizona v. Fulminante a five-to-four majority held that admitting an involuntary confession is trial error, similar in degree and kind to the admission of other improper evidence, which the reviewing court tests by examining the remaining evidence to see whether the admission was harmless beyond a reasonable doubt.21 A confession is among the most persuasive proof a prosecution can offer, so if the harmless-error inquiry reaches it, the inquiry reaches the ordinary run of constitutional error.
Because harmlessness depends on the whole record, the question often arises only after a reviewing court has found a violation, and when the lower courts never addressed it, the usual response is a remand rather than a new trial. In Coleman v. Alabama the accused had gone through a preliminary hearing without counsel, and the Court did not order reversal. It vacated the convictions and sent the case back so the Alabama courts could decide, under Chapman, whether the denial of counsel was harmless, and therefore whether the convictions should be reinstated or a new trial ordered.22 In Coy v. Iowa the Court found that the defendant's right to face-to-face confrontation had been violated, saw no reason to treat that denial differently from other Confrontation Clause violations, which are subject to harmless-error analysis, and left the question to the court below.23 California v. Green is a lighter illustration. The Court held that the Confrontation Clause did not bar the prior testimony at issue and then remarked that a separate statement to an officer "may pose a harmless-error question" better resolved first by the California courts, so Green decides nothing about harmlessness itself and shows only where the question sits in the sequence.24 Coleman's preliminary-hearing setting also carries a limit: the total denial of counsel at trial is among the errors that escape harmless-error review, and this chapter draws that line below.25
On an exam stem that describes a constitutional violation at trial, the first line of the analysis therefore cannot be "reverse." The trap is to treat a proven constitutional violation as a proven reversal. The violation opens the harmlessness inquiry, and the answer depends on what the error did to this verdict, subject to the short list of structural errors discussed later. Federal law supplies the standard whenever a federal right was infringed, even in a state prosecution, and the state's own rule governs only errors of purely state procedure or state law.
The Beyond-a-Reasonable-Doubt Test
Once a reviewing court accepts that a constitutional error can be harmless, the next question is how certain the court must be before it lets the conviction stand. Chapman v. California supplied the answer: before a federal constitutional error can be held harmless, the court must be able to declare a belief that it was harmless beyond a reasonable doubt.27 The phrase borrows the standard a jury applies to guilt and points it at the error, so the court asks whether it can honestly say the error played no part in the verdict. The standard also settles who must persuade the court. Constitutional error that lets highly prejudicial evidence or comment reach the jury casts on someone other than the person it prejudiced the burden of showing that it was harmless, and on a criminal appeal that someone is the prosecution, the beneficiary of the error.28
The test grew out of an earlier question. In Fahy v. Connecticut the Court had asked whether there is a "reasonable possibility that the evidence complained of might have contributed to the conviction."29 Chapman saw little if any difference between that question and requiring the beneficiary of a constitutional error to prove beyond a reasonable doubt that the error did not contribute to the verdict, and it presented its holding as adherence to Fahy rather than a break from it.30 The two versions are mirror images. The Fahy question looks at the error from the defendant's side and asks whether a reasonable possibility of contribution exists, while the Chapman question looks from the prosecution's side and asks whether it can exclude that possibility beyond a reasonable doubt. A prosecution that cannot rule out a reasonable possibility of contribution has therefore not carried its burden. Chapman restated the test in reasonable-doubt terms because that standard is familiar to every court and promised to be more workable, while reaching the same result.31 In Gilbert v. California the Court confirmed the equivalence, holding that a state court's inquiry into whether there was no reasonable possibility that the error might have contributed to the conviction satisfied the standard as Chapman defined it.32 Chapman also observed that California courts had leaned, perhaps too heavily, on their own view of overwhelming evidence, and it preferred the Fahy question to that emphasis; a later section returns to what the strength of the other evidence can and cannot accomplish.33
The previous section explained why the burden falls on the prosecution; what matters here is how that allocation behaves when the record is unclear. A party that carries the burden loses when the court is left in doubt, so an error whose effect is hard to gauge works against the prosecution. A dissenting opinion in United States v. Gonzalez-Lopez, written to argue that harmless-error review should stay available even for an erroneous disqualification of counsel, described the standard in just these terms: when the effect of an error is hard to gauge, the prosecution will be unable to meet its burden.34 The instinct to require the defendant to show that the error mattered is the mistake to avoid, because on direct review of preserved constitutional error the prosecution carries the burden of showing that it did not.35 The same allocation governs a constitutional error at sentencing, which requires relief unless the government demonstrates beyond a reasonable doubt that the error did not affect the sentence.36 The allocation also removes a temptation. A rule that left the defendant to prove harm would let a prosecutor gamble that an error would go unremedied, while a rule that puts the burden on the prosecutor makes the gamble costly, a point made in a dissenting opinion in United States v. Bagley.37
The reasonable-doubt standard sets constitutional error apart from ordinary error. For nonconstitutional error on direct appeal, Kotteakos v. United States asks whether the court can say with fair assurance that the error did not affect the result, and the error warrants reversal only if it had a substantial and injurious effect or influence in determining the verdict.38 Fair assurance demands less confidence than proof beyond a reasonable doubt, and the gap tracks what is at stake. A constitutional guarantee limits what the government may do to an accused, so a conviction obtained in breach of one should stand only if the court is truly convinced that the breach made no difference. On federal habeas review of a state conviction the Kotteakos formulation applies instead, as Brecht v. Abrahamson held, which means the Chapman test described here is the direct-review standard.39
Chapman itself shows the standard failing. The prosecutor in that trial commented repeatedly on the defendants' decision not to testify, and the Court needed only to look at the comments compiled from the record to conclude, with no doubt, that the error was not harmless.40 Argument that treats silence as a sign of guilt speaks directly to the question the jury must decide, and when it recurs through a trial, no one can show that the jury set it aside. The prosecutor's comments and the judge's instructions about them therefore could not be shown beyond a reasonable doubt to have had no part in the verdict, and the convictions were reversed.41
Arizona v. Fulminante reaches the same result by a different route. The error there was a coerced confession, a single item of evidence rather than argument that pervaded the trial. The Court agreed with the state supreme court that the State had failed to establish harmlessness beyond a reasonable doubt, because the prosecution had relied substantially on the confessions and, without them, a conviction was unlikely.42 The cases differ in how the error mattered: in Chapman it saturated the trial, while in Fulminante it supplied proof the rest of the record lacked. The common thread is what the test asks of the court, which is to take the error out of the record and see what remains.
Step three keeps the inquiry tied to the record, where the prosecution's showing can be tested; a court free to imagine how witnesses or jurors would have behaved without the error could reach either conclusion, and the burden allocation would mean nothing. The factors in step four were stated for a Confrontation Clause error, and they carry over to other trial errors because each asks how heavily the jury could have leaned on the tainted material. Step one points ahead to the next section, which separates the trial errors that this test can measure from the structural defects that it cannot.
Trial Error Versus Structural Error
The beyond-a-reasonable-doubt test asks a reviewing court to take the error out of the record and see what remains, and that instruction assumes an error that can be taken out. Some constitutional errors cannot be isolated that way, so the first filter in any harmlessness problem is a classification, made before any weighing of evidence begins. The Court sorts constitutional error into two groups. Trial error is error that occurs during the presentation of the case to the jury, and a reviewing court can assess it quantitatively against the other evidence to decide whether it was harmless beyond a reasonable doubt.48 A structural defect is a flaw in the constitution of the trial mechanism itself, one that affects the framework within which the trial proceeds rather than a single event inside it, and such a defect defies analysis by harmless-error standards.49
The reason for the division follows from how the test operates. A court can decide whether an item contributed to a verdict only by subtracting the item and comparing what is left, and subtraction works when the error is confined to a discrete part of the trial. When the defect lies in the setup of the trial, nothing can be subtracted, and any finding of harmlessness would rest on speculation about a trial that never took place. In Holloway v. Arkansas the Court contrasted the ordinary harmless-error case, in which the error occurs at trial and its scope is readily identifiable, with joint representation of conflicting interests, where the harm lies in what the lawyer was compelled to refrain from doing, such as exploring plea negotiations, challenging evidence, and arguing at sentencing.50 A court asked to measure those omissions would be engaged in unguided speculation, so the Court held that reversal is automatic whenever a trial court improperly requires joint representation over timely objection.51
The line between the two groups has moved over time. Chapman v. California itself, in a footnote, listed a coerced confession, the denial of counsel, and a trial before a biased judge as examples of rights so basic to a fair trial that their infraction can never be treated as harmless.52 In Arizona v. Fulminante the Court took the first item off that list, holding that admission of a coerced confession is trial error similar to the erroneous admission of other evidence and does not necessarily render a trial fundamentally unfair.53 The other two items stayed where they were. The Court described the total deprivation of the right to counsel at trial and trial before a judge who is not impartial as structural defects, because the absence of counsel or the presence of a partial judge affects the entire conduct of the trial from beginning to end.54 A confession can leave the never-harmless list because its effect can be weighed: the reviewing court sets it aside and reviews the remaining evidence. The trial-error label also opens the inquiry without deciding it, since the Court in that case went on to find that the prosecution had relied substantially on the confessions and so had not carried its burden.55
The Court has given reasons why an error may be classified as structural, and the reasons explain which errors land there. In Weaver v. Massachusetts the Court identified at least three rationales. When the right protects an interest other than avoiding an erroneous conviction, such as the right to conduct one's own defense, harm is irrelevant to the basis of the right. When the effects of an error are too hard to measure, requiring a showing of harmlessness is impractical. When an error always produces fundamental unfairness, trying to show harmlessness would be futile.56 These categories are not rigid, and an error can be structural even if it does not produce fundamental unfairness in every case.57 The measurement rationale matches the logic of the beyond-a-reasonable-doubt test most directly. It helps explain why the erroneous disqualification of the defendant's chosen counsel has been treated as structural, since the difference a different lawyer would have made cannot be quantified.58
The recognized structural defects form a short list. Alongside total denial of counsel at trial and a biased judge, the Court has added unlawful exclusion of members of the defendant's race from a grand jury, denial of the right of self-representation, and denial of the right to a public trial.59 The grand jury item shows the category at work. In Rose v. Mitchell the Court declined to treat discrimination in grand jury selection as harmless merely because a properly constituted trial jury later found the defendant guilty beyond a reasonable doubt, and it adhered to its longstanding practice of granting relief for that defect.60 The later guilty verdict cured nothing. The grand jury section below returns to this contrast between ordinary error, which is tested for its effect, and the rare defect that is not.
The structural label carries a procedural boundary. When a structural error was objected to at trial and is raised on direct appeal, the defendant is generally entitled to automatic reversal regardless of the error's actual effect on the outcome.61 The result differs when the error was not preserved and surfaces later in an ineffective-assistance claim. Prejudice is not then shown automatically, and the defendant must show the usual reasonable probability of a different outcome or, on the assumption the Court made in Weaver, that the violation rendered the trial fundamentally unfair.62 A public-trial violation illustrates why: the right also protects the press and the public, and a closure can occur in a trial that remained fair from the defendant's standpoint.63 The role of preservation returns in the closing section on plain error.
The outer edge of the structural category is contested. A dissenting opinion in United States v. Gonzalez-Lopez argued that the touchstone of structural error is fundamental unfairness and unreliability, that trial error and structural defect were not meant as the only two categories, and that automatic reversal is strong medicine reserved for errors that always or necessarily produce such unfairness.64 The Court has also declined to enlarge the category in some settings. In Rivera v. Illinois it rejected a claim that an improper refusal to excuse a juror on a peremptory challenge is structural error.65 An exam answer that places a new error in the structural category therefore needs a reason tied to measurability or fundamental unfairness, and an assertion that the right is important does not supply one.
Jury instructions are where the classification is hardest to predict. A strong presumption applies that a constitutional violation will be subject to harmless-error analysis, and the Court has applied that analysis to instructions that misstate an element or contain an erroneous presumption, while an erroneous instruction on the burden of proof has been treated differently.66 A defective reasonable-doubt instruction is the instruction error placed on the structural side, in Sullivan v. Louisiana.67 The Court reached its present position gradually. In Sandstrom v. Montana it declined to decide, as an initial matter, whether an unconstitutional instruction on an element of the crime could ever be harmless, and it left the question to the state court on remand.68 In Neder v. United States the majority later treated the omission of an element from the jury's consideration as an error that can be harmless, over the position of three Justices who would have held that depriving a defendant of the jury's determination of every element can never be harmless.69
The same constitutional right can fall on either side of the line, depending on what went wrong. The denial of counsel at a preliminary hearing appears among the constitutional errors the Court has subjected to harmless-error analysis, while the total deprivation of counsel at trial is the standard structural defect.70 Two facts move the answer: the stage of the proceeding and the totality of the deprivation. A missing lawyer at one preliminary event leaves a trial record against which the effect can be weighed, and a missing lawyer through the whole trial leaves nothing to compare.
One further item calls for caution. One account of the exceptions places the failure to disclose exculpatory evidence beside the denial of counsel at trial and the denial of an impartial factfinder.71 Another counts a prosecutor's failure to turn over exculpatory evidence among the constitutional errors that may be deemed harmless in an individual case.72 Because the sources disagree, the core of the structural list (total denial of counsel at trial and a biased judge) can be stated with confidence, while nondisclosure is an item to analyze for measurable effect and not to assume.
Students lose this issue when they treat the seriousness of the violation as the test. A coerced confession is about as damaging as evidence gets, and it is still trial error, because damage that arrives as an item of evidence can be weighed against everything else the jury heard.74 The question that sorts the cases is whether the court can measure what the error did. Because the classification decides whether the weighing is reached at all, it is best run first, in four moves.