Weighing Error Against the Record

Weighing Error Against the Record

An appellate court almost never meets an error alone. Viewed in isolation, nearly any evidentiary mistake might have changed the result: evidence admitted in error may have been the reason the jury ruled against the appellant on a critical issue, evidence wrongly excluded may have tipped an issue the jury decided the other way, and a cross-examination cut short, or an improper attack on a witness, may have changed how the jury weighed that witness's testimony.¹ The earlier sections supplied the yardsticks for deciding whether an error matters, which are the fair-assurance inquiry of Kotteakos for ordinary error and the beyond-a-reasonable-doubt test of Chapman for constitutional error.² Those yardsticks need something to measure against, and what they measure against is the rest of the trial record. The appellant's last task, persuading the court that the error mattered, is therefore usually decided by the record around the error.

The party that won below will not let the error stand alone. It points to the rest of the record and argues that the mistake made no difference, and three such arguments recur often enough to have hardened into recognized doctrines: that other evidence established the same point, that something done at trial cured the error, and that the properly admitted evidence was overwhelming.³ The law lets these circumstances count because of what a reversal costs. A reversal means a brand new trial, which is justified for errors serious enough to have mattered and wasteful for errors that did not.⁴ Whether an error mattered can be judged only by looking at what the jury had besides the error. Evidence that merely repeats what the jury already heard adds little, an instruction can change how the jury is allowed to use what it heard, and proof that is overwhelming makes a different verdict unlikely. Each doctrine therefore answers the same question, which is whether the error made a difference to the verdict this jury returned.

All three doctrines carry one limit. Students lose this issue when they ask whether enough other evidence supported the result, because that is the question the Court rejected in Kotteakos. The reviewing court asks whether the error itself had substantial influence on the verdict, and if it cannot say with fair assurance that the judgment was not substantially swayed, or is left in grave doubt, the conviction cannot stand.⁵ The same limit governs the cumulative-evidence doctrine: an error that probably affected the outcome calls for corrective action even when enough other evidence exists to support the conclusion the jury reached.⁶

The limit has a functional reason and an incentive reason. Sufficiency asks whether a jury could have convicted on the remaining evidence, while harmlessness asks what this jury did, and a jury that heard the erroneous evidence may have leaned on it even though the rest would have sufficed. The incentive problem sits with the reviewing judge, who has the cold record and may be convinced of the defendant's guilt, and who is therefore tempted to treat any error as immaterial. That temptation replaces the jury's reaction with the judge's own, so Kotteakos instructs reviewing courts to gauge the jury's reactions with allowance for how others might respond and to resist measuring them by their own sense of guilt.⁷

The same record-weighing governs constitutional error, and the Court has said what to look at. In Delaware v. Van Arsdall the Court held that denying a defendant the chance to impeach a witness for bias is subject to harmless-error analysis, and it identified five factors that bear on harmlessness in that setting.⁹ They are these:

• the importance of the witness's testimony in the prosecution's case;
• whether that testimony was cumulative;
• whether other evidence corroborated or contradicted the witness on material points;
• how much cross-examination was otherwise permitted; and
• the overall strength of the prosecution's case.

Applying those factors in Olden v. Kentucky, the Court found it impossible to conclude beyond a reasonable doubt that restricting cross-examination was harmless, because the witness's story was central to the prosecution, only a largely derivative witness corroborated it, and the jury's verdicts suggested that the State's case was far from overwhelming.¹⁰ Two of the factors preview the doctrines that follow, since cumulativeness corresponds to the cumulative-evidence doctrine and the overall strength of the case corresponds to the overwhelming-evidence doctrine.

A different idea shares part of a name with the first doctrine and should be kept apart from it. Cumulative-error analysis asks whether several errors, each harmless when taken alone, together prejudiced the outcome, and if they did, the errors combined may amount to reversible error.¹¹ The cumulative-evidence doctrine, by contrast, concerns a single error and the other proof on the same point. The first adds errors together and the second reduces the weight of one error by pointing to duplicate evidence, so a court can use both on one record without inconsistency.

Two layers of deference make these arguments hard for an appellant to win. Harmless-error analysis makes reversal on an evidentiary ruling difficult, and most evidentiary rulings are reviewed for abuse of discretion, which means the appellate court will not substitute its own judgment even if it would have ruled differently.¹² The degree of probability a court demands before calling an error prejudicial also varies by jurisdiction and by whether the case is criminal, civil, or constitutional, ranging from a possibility that the error affected the verdict to a probability that it did.¹³ Because deference protects any reasonable ruling, a court can affirm one decision admitting a type of evidence and another excluding the same type in substantially the same circumstances, and the apparent inconsistency reflects that deference and not a change in the governing rule.¹⁴

Step three is where the three doctrines enter, and step four is where answers go wrong, because the pull is to answer it with a sufficiency inquiry. That pull is strongest with overwhelming evidence, since in practice the approach that asks whether the error contributed to the verdict has been eroded by an approach that asks whether the record shows overwhelming evidence of guilt.¹⁷ The limit stated above therefore has to be held onto when that doctrine is applied, and the later section on overwhelming evidence returns to it. The first of the three circumstances, other evidence on the same point, comes next, because it is the most common basis for finding that an error did not matter.

Cumulative Evidence

An error is easiest to excuse when the jury already had whatever the error added, or would have had whatever the error kept out. That idea is the cumulative evidence doctrine, which supports affirmance despite errors in both admitting and excluding evidence: the trial judge did err, but so much other proper evidence supported the same point that the jury would likely have decided against the appellant even if the ruling had gone the other way.¹⁹ The doctrine is one way of answering the general question whether an error probably affected the result, since a reviewing court reverses only for error serious enough to require a brand new trial.²⁰ Its reason lies in what a retrial would accomplish. A jury that has heard a point established several times over has little to learn from one more item on it, so removing a wrongly admitted item, or restoring a wrongly excluded one, leaves what the jury could rationally conclude about that point where it already stood, and a second trial would probably end in the same verdict.

The admission case is the familiar one. The appellant complains that the jury heard something it should not have, and the party that won below answers that the jury heard the same thing properly from other sources.²¹ Whether that answer succeeds depends on how fully the other evidence covers the point, as the contrast below shows.

The doctrine runs equally in the other direction, which students overlook when they picture errors only as improper admissions. In an illustration built on an employee named Trivedi, who lost at trial, the judge admitted five witnesses and twenty employment records showing that coworkers had been promoted while Trivedi was not, but wrongly excluded three more promotion records on hearsay grounds over a timely objection; the three records added nothing to a point already well proved, so the exclusion probably did not affect substantial rights.²³ Change what was excluded to the testimony of Trivedi's own psychiatrist that his mental state was normal though stressed, and the answer flips, because the defendants had pressed a delusional disorder through two doctors and the psychiatrist's testimony would have been powerful rebuttal on that point.²⁴ A reported criminal case shows the same logic at work. In the prosecution of a defendant named Knight, the district court wrongly excluded an eyewitness's opinion that a shooting was accidental, yet the reviewing court found the error harmless because the eyewitness had been allowed to describe fully the circumstances behind the opinion, Knight himself testified that the discharge was accidental, and the prosecutor all but conceded the point in closing.²⁵ The two excluded-evidence examples differ in the variable that matters, which is whether the rest of the record already gave the jury the substance of what was kept from it.

Because the doctrine depends on what the record actually contains, the other evidence has to be shown rather than assumed. In Crane v. Kentucky the state courts excluded testimony about the physical circumstances under which police obtained the defendant's confession, and the State argued that the error was harmless because the excluded evidence ultimately reached the jury through other witnesses. The Supreme Court agreed that the erroneous ruling was subject to harmless-error analysis but concluded that the State's argument should be directed in the first instance to the state court.²⁶ Evidence admitted earlier without objection can also supply the cumulative effect. A party who failed to object to the first item of similar evidence has not by that alone waived an objection to the next, since lawyers commonly hold their objections for evidence that does real damage, but overruling the later objection may still be harmless error because the earlier evidence had already put the fact before the jury.²⁷

The doctrine has a boundary, and it deserves as much attention as the doctrine itself. Cumulative evidence does not justify affirmance merely because other evidence was sufficient to sustain the verdict the jury reached.²⁸ The question remains whether the erroneously admitted evidence probably affected the outcome, or whether the erroneously excluded evidence probably would have, and an affirmative answer calls for corrective action even when enough other evidence exists to support the jury's conclusion, which is the distinction Kotteakos drew between support for the result and an error that itself had substantial influence.²⁹ Two features of the doctrine's wording explain why sufficiency cannot stand in for it.

The first feature is the word same: other evidence counts only if it bears on the very point the erroneous ruling concerned, because proof on a different point leaves the jury's view of this point untouched, and students lose the issue when they credit other evidence of guilt on a different point as though it made the challenged item cumulative. The second feature is that the inquiry concerns the effect on the jury that actually sat. A duplicate item can still carry weight when that jury had to choose between two accounts, since an additional voice on a disputed point may supply the corroboration that settles the choice, and the item may carry an implication of its own that the duplicated proof lacked.

Krulewitch v. United States shows the limit at work. There the Government defended a conspiracy conviction by arguing that a co-conspirator's hearsay statement, admitted over the defendant's repeated objection, was merely cumulative and that the case without it was so strong that the error should be held harmless. The Court refused, because applying Kotteakos it was left in grave doubt whether the statement had substantial influence on the verdict.³⁰ The statement amounted to an unsworn declaration of the defendant's guilt; the corroboration did not reach the vital phases of the case, so the jury had to choose between believing the defendant and believing the complaining witness; and the procedural history, in which a Florida grand jury failed to indict and four trials in New York produced two mistrials and two convictions, the last with a recommendation of leniency, showed a difficult case. The Court could not say that the statement had not been the weight that tipped the scales against the defendant.³¹ Justice Burton, dissenting, agreed that the statement was inadmissible but would have affirmed, because an 800-page record held evidence that was "cumulative, repetitive and corroborated" and the verdict could not have been appreciably affected by one reported snatch of conversation.³² The majority and the dissent asked the same question, which was the statement's effect on the jury, and they parted over what a long and closely contested record showed about it.

Hawkins v. United States reached the same result by a similar path. A wife had been wrongly permitted to testify against her husband in a prosecution for interstate transportation for prostitution, and the court of appeals had held in the alternative that her testimony was harmless. The Supreme Court examined the record and could not say the testimony lacked substantial influence on the jury, because the only contested factual issue was the defendant's dominant purpose in making the trip, the jury could have resolved it either way depending on whom it believed, and the wife's evidence, "though in part cumulative," might have tipped the scales.³³ What her testimony added went beyond repetition. The mere sight of a wife testifying against her husband would likely impress jurors adversely, and her sworn statement that she had been a prostitute both before and after the marriage suggested that the defendant was the kind of man to whom the alleged purpose would come naturally.³⁴

The contrast with Knight and the three extra promotion records shows which fact does the work. In those matters the excluded item went to a point the rest of the record already settled: five witnesses and twenty records had proved the promotions, and in Knight the prosecution barely disputed that the shooting was accidental, so a duplicate item changed nothing the jury had to decide.³⁵ In Krulewitch and Hawkins the point was the contested center of the case, the other proof did not settle it, and the challenged evidence brought something beyond repetition, an implied admission of guilt in one and an adverse impression of the accused in the other.³⁶ Constitutional cases ask the same thing in different words, since whether testimony was cumulative is one of the factors in the Van Arsdall list, alongside the importance of the witness and the overall strength of the case, and in Olden the Court found it impossible to conclude beyond a reasonable doubt that restricting cross-examination was harmless where the witness's account was central and corroborated only by largely derivative testimony.³⁷

Curative Instructions

A preserved error does not always lead to reversal, because the trial itself sometimes supplies the repair. The curative instruction doctrine covers the cases in which the judge's own mistake on an evidence point is corrected in front of the jury, so that an appellate court can treat the error as harmless.⁴⁰ The doctrine takes three common forms. A limiting instruction confines evidence to the purpose for which, or the party against whom, it is admissible, which is the instruction Federal Rule of Evidence 105 provides for. An instruction to disregard removes evidence that should never have come in. And the verdict itself can cure the error when the jury finds for the appellant on the only issues the error touched.⁴¹

The reason an instruction can do this work is the assumption on which jury trial rests. In a footnote in Francis v. Franklin the Court explained that it presumes jurors attend closely to the trial court's instructions and strive to understand and follow them.⁴² If that presumption holds for a particular instruction, the jury decided the case as though the improper item, or the improper use of an item, were absent. The error then never reached the verdict, and a retrial under the same instruction would end the same way. A court that reversed despite an effective instruction would spend a second trial repairing a harm that no longer existed.

The limiting instruction is the easiest form to see. Evidence admitted against one party or on one point carries a risk that jurors will treat it as proof against another party or on another point. An instruction to use it only as admitted is usually viewed as effective, which disposes of an appellate contention that the evidence was misused.⁴³

The same logic covers evidence that should not have been admitted at all. A judge who allows testimony, then orders it stricken and admonishes the jury to ignore it, has corrected the error in the courtroom, and the error is generally treated as harmless.⁴⁵ Occasionally an instruction cures even an error in excluding evidence, as when the judge implies in the instructions that the issue has been, or should be, resolved for the party who offered the evidence. Sometimes the verdict does the curing, as when hearsay bearing on premeditation was admitted but the jury found no premeditation.⁴⁶

Each of these cures rests on the presumption that jurors follow instructions, and the presumption weakens as the risk grows. An instruction asks jurors to hold a piece of evidence in a separate compartment of the mind. The more damaging the evidence, and the more closely it bears on the issue the jury must decide, the less realistic that request becomes. The Supreme Court has recognized that the risk of prejudice can be so great that even a limiting instruction cannot adequately protect a criminal defendant, and it has pointed to Bruton v. United States and Jackson v. Denno as examples.⁴⁷ The reviewing court's question is therefore whether the instruction was equal to the danger, and students lose this issue when they treat the giving of an instruction as the end of the analysis.

The criminal cases show where the line has been drawn. An instruction to disregard an involuntary confession does not cure the harm of admitting it. A codefendant's confession that is admissible against its maker but also implicates the defendant ordinarily cannot be introduced in a joint trial on the strength of an instruction to use it only against the maker. And an instruction to disregard evidence of a prior conviction obtained without counsel did not make its introduction harmless.⁴⁸ In Jackson v. Denno the jury was told it must disregard a confession it found involuntary, yet the procedure was held defective because the jury could weigh the confession's truth in deciding voluntariness.⁴⁹ The Court added that a conviction resting in part on a coerced confession cannot be saved by showing that other evidence would have sufficed.⁵⁰ That last point is the sufficiency limit from earlier in this chapter applied to a confession: a court asks whether the error itself mattered, and the strength of the remaining evidence does not answer that question.

The codefendant-confession problem also shows how the answer can change. Before Bruton, Delli Paoli v. United States had held that appropriate jury instructions would suffice to prevent prejudice from references to a defendant in a codefendant's statement. In Gilbert v. California the Court noted that California, applying its own Aranda rule, regarded cautionary instructions as inadequate to cure that prejudice, and it declined to decide whether Delli Paoli should be reconsidered.⁵¹ Bruton, decided the next year, reversed Delli Paoli, as the dissent recorded in objecting to that result. That dissent suggested that effective deletion of every part of a confession implicating a codefendant would be the practical substitute for an instruction, and it added that a failure to follow such procedures would be relevant to whether the error was harmless.⁵²

In Richardson v. Marsh the justices divided over where the line fell, and the dissent's account shows which facts drive the answer. The codefendant's confession described a conversation in a car on the way to the victims' house, in which his companion said he would have to take the victims out after the robbery. The State's principal witness had testified that the companion and Clarissa Marsh arrived at the house together, so the dissent reasoned that the jury was certain to infer that Marsh had been in the car and overheard the plan. On that reading the confession was the only evidence directly linking her to the intent to kill, which was the closely disputed issue.⁵³ The dissent acknowledged that the statement did not name Marsh and so did not incriminate her on its face, but argued that it became incriminating when joined with evidence already before the jury. It concluded that the same unacceptable risk remained, that the jury would rely on impermissible evidence on a critical issue.⁵⁴

A second kind of case defeats instructions through volume, where Richardson turned on a single powerful item. In Kotteakos v. United States the indictment charged one conspiracy but the proof showed eight separate ones, involving thirty-two defendants at the outset, nineteen when the trial began, and thirteen when the cases went to the jury.⁵⁵ The Court refused to follow the earlier Berger decision, where only two conspiracies involving four persons had been proved, because the burden on a defendant tied to one or a few of so many transactions includes guarding against evidence affecting the others. Part of that guard is preventing its transference by psychological effect, in spite of instructions to keep separate transactions separate.⁵⁶ Kotteakos was decided under the harmless error statute on a variance between charge and proof, so it reaches instructions only indirectly. Its reasoning explains the failure, though, because no juror can realistically keep so many distinct transactions apart. Later courts have read it as an unwillingness to apply harmless error mechanically where a codefendant's out-of-court statements are improperly admitted or a mass of evidence implicates others.⁵⁷ The difference between the two cases is the source of the danger: in Richardson the dissent saw one item too damaging, and too tied to a contested issue, to be set aside, and in Kotteakos the danger was the number of transactions.

Overwhelming Evidence

The third circumstance that can turn an apparently serious evidentiary error into a harmless one is the strength of everything else in the record. Under the overwhelming evidence doctrine, a reviewing court that concludes the properly admitted proof supports the judgment overwhelmingly generally affirms, even in the face of errors in admitting or excluding evidence that might otherwise be considered serious.⁶⁰ The opinions applying it read as though the untainted evidence would have invited a directed verdict.⁶¹ The reason is practical: if the case against the appellant was that strong without the challenged item, the error probably changed nothing, and a reversal would send the parties back for a second trial that ends the same way. The doctrine resembles the cumulative-evidence doctrine taught earlier but works at a larger scale, since that doctrine compares the challenged item with other proof of the same point, while this one asks whether the whole body of untainted proof leaves no realistic room for the error to have mattered.

A standard illustration from federal civil practice shows the doctrine at its cleanest. In a diversity action over a defective product, the plaintiff offers no proof of a defect, the defendant's first expert testifies properly that the product was not defective, and the trial court also admits a second expert's affidavit to the same effect, which is inadmissible hearsay. The jury finds for the defendant. If the appeals court concludes that, even without the affidavit, the defendant offered so much unrebutted proof that the product was sound that a rational jury could not have found for the plaintiff, it affirms, because the plaintiff was extremely unlikely to win even had the affidavit been excluded.⁶² Criminal appeals follow the same reasoning. After Skilling v. United States confined honest-services fraud to bribes and kickbacks, the Third Circuit in United States v. Andrews treated a reference to honest services in the final instructions as harmless, because the government had presented overwhelming evidence of tangible wire fraud.⁶³

Federal constitutional error follows the same pattern under a stricter measure. If other evidence of guilt is so overwhelming that the defendant would have been convicted even without the constitutional violation, the violation is harmless.⁶⁴ The court must nonetheless be able to declare its belief that the error was harmless beyond a reasonable doubt, and the burden of showing that falls on the party who benefited from the error.⁶⁵ Overwhelming evidence is thus one route to satisfying that standard, and the standard itself still has to be satisfied. For nonconstitutional federal error the corresponding question is whether the court can say with fair assurance that the error did not substantially sway the judgment, and the strength of the rest of the record informs that answer.⁶⁶

The doctrine is real and easily misused. Overwhelming describes the state of the record, whereas the harmless-error question asks what the error did to the jury, and an opinion that stops at the first has not answered the second. In Kotteakos v. United States the Court put the limit this way: "The inquiry cannot be merely whether there was enough to support the result, apart from the phase affected by the error."⁶⁷ A test asking whether the verdict was clearly correct on the admissible evidence turns the appellate court into the trier of fact and overlooks the defendant's right to a fair trial even when guilt looks clear.⁶⁸ Kotteakos therefore directed attention to the effect the error had, or reasonably may be taken to have had, on the jury's decision, and it warned that the difference between judging the jury's reaction and one's own is easy to ignore when the sense of guilt comes strongly from the record.⁶⁹ That warning identifies the incentive problem: the more persuasive the record appears to the judges reading it, the greater their temptation to affirm on their own conviction of guilt, which is the very substitution of judgment the substantial-rights inquiry is meant to prevent. Students lose this issue when they treat enough other evidence to sustain the verdict as the end of the analysis.

Some formulations nonetheless lean toward the government. One treats an error in admitting evidence as harmless if enough other evidence existed on which a reasonable juror could have based the verdict, even when the erroneous item was probably the decisive one in the jury's mind.⁷⁰ That version sits uneasily with the Kotteakos statement just quoted, yet practice has drifted toward it, with the overwhelming-evidence approach partly eroding the older question whether the error contributed to the verdict.⁷¹ The drift has a cost that explains why appellants find these rulings frustrating: an appeals court that says the appellant could not have won in any event has wide latitude to call serious errors insignificant.⁷² A court's invocation of overwhelming evidence is therefore a conclusion that still has to be tied to the effect on the verdict.

In Chapman v. California the same record produced opposite readings. The prosecutor had repeatedly commented on the defendants' failure to testify, which violated the Fifth and Fourteenth Amendments.⁷³ Justice Stewart, joined by Justice White, would have affirmed, writing that the evidence of guilt was overwhelming, that the petitioners were apprehended near the scene with the victim's property in their possession, and that medical evidence showed rape, while Justice Harlan likewise thought the verdict amply supported.⁷⁴ The majority took a different path. It observed that the California courts had neutralized their "miscarriage of justice" standard by emphasis, and perhaps overemphasis, on overwhelming evidence, and it adopted a test under which the court must be able to declare the error harmless beyond a reasonable doubt.⁷⁵ Applying that test, the majority had no doubt the error was not harmless, pointing to the extent of the prosecutorial comments compiled from the record.⁷⁶ The record the dissenters read as showing overwhelming guilt therefore left the majority unable to declare, beyond a reasonable doubt, that the comments did not contribute to the verdict.

Overwhelming evidence also loses force when the issue the error touched was genuinely contested. In Francis v. Franklin the defendant's only defense was that he lacked intent to kill, the facts did not overwhelmingly preclude that defense, and an erroneous instruction on intent threatened to eliminate it, so the Court agreed the error could not be deemed harmless; the jury's request for reinstruction on malice and accident lent weight to that view.⁷⁷ The test drawn from Neder v. United States asks whether an omitted element was "uncontested and supported by overwhelming evidence," and a court should not find harmlessness where the defendant contested the element and raised evidence sufficient to support a contrary finding.⁷⁸ In United States v. Miller, a prosecution arising from the forcible cutting of Amish men's beards and women's hair, the majority held an erroneous instruction on whether the assaults were committed because of the victims' religion was not harmless, because motive was the key issue at trial and the defendants presented evidence of non-religious motives.⁷⁹ The dissent applied the same test and called the evidence of religious causation uncontested and overwhelming, so the two opinions differed over whether the record left a contested issue, and not over what the test requires.⁸⁰

The doctrine must also be kept apart from sufficiency review. Sufficiency asks whether, viewing the evidence in the light most favorable to the prosecution, any rational trier of fact could have found the essential elements beyond a reasonable doubt.⁸¹ That deferential question tests whether the verdict could stand on the record, while the harmless-error question tests whether this error affected this verdict, so a record can pass the first with ease and still fail the second. The Second Circuit in United States v. Newman treated them as separate steps: it held the instructional error not harmless because the defendants had contested their knowledge of any benefit to the tippers, and it then addressed insufficiency of the evidence under its own heading.⁸²