Professor Garrett’s impressive empirical analysis of the first 200 postconviction DNA exonerations in the United States (“Garrett Study”) has the potential to affect contemporary debates surrounding our nation’s criminal justice system. This Response explores this potential by harnessing the Study’s data in support of arguments for and against a contested doctrinal proposition—that guilt-based harmless error rules should never apply in death penalty appeals.
My analysis starts with the premise that the Study’s real world impact will necessarily depend on how jurists, politicians, and scholars extrapolate the explanatory power of the data beyond the 200 cases themselves. While critics of contemporary criminal justice policies will likely see Professor Garrett’s data as revealing the tip of an iceberg of deeper structural flaws, defenders of the status quo will predictably resist generalizations from this closed data set to any larger picture of criminal justice administration. Much therefore rides on the perceived inductive reach of these 200 cases.
Perhaps wisely, Professor Garrett declines to engage in a specific evaluation of his study’s implications in terms of the number of innocents still in prison, except to assert cautiously that known innocence cases “represent the tip of an iceberg.”1 Curious readers might nonetheless persist in asking: just how much larger is the phenomenon that these cases represent? I suggest that this line of inquiry is not a particularly useful one because the empirical question it asks is essentially unanswerable. Given that conclusive forensic proof of guilt or innocence remains unavailable in the overwhelming majority of convictions, any attempt to precisely quantify the data’s representativeness inevitably will rest on guesses (educated or not) concerning either overall system error rates or the raw numbers of undiscovered wrongful convictions.
Rather than attempt such quasiempirical triangulations, I propose instead to test the explanatory power of the Garrett Study via a rhetorical experiment. Specifically, I hope to gauge the practical reach of the Study by examining its persuasive impact when called into service to support a particular and contested proposition—that guilt-based harmless error rules should never apply in death penalty appeals.2 The experiment is designed to test what persuasive “truths” might emerge from the Study when it is subjected to the crucible of an imagined adversarial process. Though the actual proposition chosen for debate matters, the assumption animating the experiment is that dialectical process and rhetorical dynamics would cause similar “truths” to emerge even when contesting different propositions.
The crux of the rhetorical experiment has three stages, mirroring the stages in a typical litigation briefing. Reasoning from the Garrett Study, I first advance imagined prima facie arguments in support of the proposition that guilt-based harmless error rules should never apply in death penalty appeals. Second, I respond to the prima facie case with counterarguments focusing on potential and perceived weaknesses in the Study. Third, I offer a reply to rehabilitate the original proposition and the relevance of the Study. After the arguments are submitted, I conclude the experiment by attempting to assume a neutral stance and analyzing the broader rhetorical and practical implications of the imagined exchange.
Prima Facie Case: Guilt-Based Harmless Error Rules Should Never Apply in Death Penalty Appeals
Guilt-based harmless error rules empower courts to deny relief to criminal appellants despite finding legal error in the proceedings that led to conviction. As Professor Garrett observes, these guilt-based rules surface in various doctrinal inquiries such as the Chapman test (which excuses constitutional error at trial if the state shows beyond a reasonable doubt that the error did not contribute to the guilty verdict) or the Strickland test (which precludes a finding of constitutional error when ineffective assistance of counsel failed to prejudice the outcome due to the evidence of the client’s guilt).3 Harmless error rules exist to limit or prevent reversals of otherwise reliable convictions or sentences on technicalities.
A critical assumption that must underlie the creation and application of guilt-based harmless error rules is that reviewing courts can effectively judge the reliability of a conviction independent of any legal error that may have occurred in the proceedings below. By definition, harmless error rules only come into play when a fairness-implicating error has occurred. Ultimately, then, a conclusion that a conviction is reliable represents a judgment on the substantive correctness of the guilty verdict or guilty plea. It then follows that guilt-based harmless error rules must assume that courts can effectively judge the guilt or innocence of an appellant.
Professor Garrett’s study of the first 200 postconviction DNA exonerations in the United States undermines this assumption. At the most basic level, all 200 cases represent discrete instances where the courts failed to detect innocence. Every one of these 200 individuals had his or her conviction sustained on appeal and was not exonerated until after postconviction DNA testing yielded exculpatory results and/or a “hit” to a third party in a DNA database. Without question, courts have confidently but mistakenly judged actually innocent appellants to be guilty.
The Garrett Study further allows for a nuanced analysis of the precise manner in which the courts went astray. In 133 of 200 cases, courts issued nonsummary written opinions explaining their reasons for granting or denying relief. A key finding of the Study is that in 18 of these 133 cases—14%—courts actually did reverse and vacate the innocent person’s conviction. Although these 18 reversals were all sustained on appeal, they do not represent exoneration by the courts. In 12 of these 18 cases, the person was retried and reconvicted, and only later exonerated by DNA testing. In the remaining 6 cases, exculpatory DNA results were obtained while the person awaited his retrial.4 Nonetheless, the 14% reversal rate represents a baseline in the data set where the court system “got it right” at least once.
Obviously, this 14% success rate pales in comparison to its complimentary 86% failure rate. Even more revealing, however, is that in 43 of the 133 cases with written decisions—32%—courts made adverse guilt-based harmless error rulings in the course of denying relief.5 Looking more closely, Professor Garrett reveals that in 23 of these 43 harmless error cases, the court actually found an underlying legal or constitutional error, but then denied relief on harmlessness or lack-of-prejudice grounds.6 In other words, application of guilt-based rules caused 23 innocent individuals (17% of the 133 cases) to languish in prison despite judicial recognition that the proceedings that led to their conviction were in fact flawed and unfair.
These findings should give pause to even the most ardent supporters of the death penalty. Guilt-based harmless error rules permit the state to execute an individual in the face of legal error. This represents a judgment that the evidence of guilt is so strong that it outweighs or excuses admitted unfairness or technical deficiency and therefore permits taking the life of the condemned. Yet Professor Garrett has shown that courts demonstrate a guilt-confirming bias even when reviewing cases of actually innocent defendants. More courts identified reversible error but excused the failing on (incorrect) harmlessness grounds than identified error and permitted reversal—even if reconviction ultimately ensued.
While it may be unrealistic (though not unreasonable) to demand zero errors and execution of only the guilty, it seems entirely reasonable to require that no execution proceed if a court has identified reversible error. After all, “technical” procedural protections exist only because of their presumed role in enhancing the truth-seeking process of trial and in preventing wrongful conviction. Professor Garrett has demonstrated that courts can be seriously wrong when they ignore the legal recognition of errors in the truth-seeking process and substitute their own subjective assessment of guilt.
Response: Juries Real Source of Error; Capital Review Adequate
Harmless error rules cannot be strictly equated with an affirmative appellate judgment of guilt or innocence. Rather, harmless error analysis recognizes that factfinders’ verdicts should not be disturbed for any or all procedural defects. In this way, harmless error rules actually represent deference to the judgment of juries more than confidence in the record-reviewing power of appellate courts. The real problem identified by Professor Garrett’s study is that juries can make mistakes. However, this problem cannot be solved by tinkering with guilt-based harmless error rules.
Although Professor Garrett devotes much attention to the path of eventually-exonerated defendants’ appeals, the bottom line is that 191 of the 200 wrongful convictions in his Study—96%—resulted from a factfinder’s verdict after trial; a mere 9 cases resulted from guilty pleas.7 In 18 of the 200 cases, appellate courts judged legal error to be serious enough to disturb the jury’s judgment. However, as already cited, in 12 of these 18 reversals the innocent defendant was again convicted at trial. In other words, even when appellate courts get it right, the case will likely be returned to a jury for another chance to get it wrong. Professor Garrett’s study thus underscores a simple truth—juries are human and therefore fallible. Nonetheless, the reality remains that juries are a constitutionally required centerpiece of our criminal justice system.
With respect to the specific question of harmless error in the death penalty context, Professor Garrett’s Study supports the proposition that appellate courts currently give capital cases the additional scrutiny they deserve. Although a 14% overall reversal rate has already been cited, Professor Garrett himself is careful to distinguish between innocent appellants’ reversal rates in capital versus noncapital cases. As it turns out, 12 of the 133 cases with written decisions in the data set were capital cases, and 7 of the 18 reversals with written decisions were also in capital cases, resulting in a capital reversal rate of 58%.8 This 58% attrition rate is similar to the 68% capital attrition rate found in all capital appeals from 1973 through 1995 in the landmark Liebman, Fagan, and West study.9 This high reversal rate therefore seems reliable and generalizable beyond Professor Garrett’s own narrow data set to the capital system at large.
For present purposes, the germane bottom line from these two studies is that one can reasonably expect approximately 58–68% of all capital convictions to be reversed. The logical counterpart of this conclusion is that guilt-based harmless error rules will ultimately be ignored or overcome in a comfortable majority of all capital cases. Indeed, it is not clear in the Study how many capital cases, if any, were among the 23 cases where harmless error applied to defeat a meritorious claim. In short, the issue of guilt-based harmless error rules blocking relief for innocent appellants does not seem to be a significant problem in the capital context.
Finally, the harmless error question needs to be understood in the context of courts granting relief for the actually guilty. In addition to a 58% capital reversal rate, Professor Garrett found a 9% reversal rate for noncapital cases in the data set.10 Of course, nothing in Professor Garrett’s study directly suggests that a higher percentage of capital defendants are actually innocent than noncapital defendants. And nothing in the study suggests that anywhere near 58%–68% of people on death row are actually innocent, or that 9% of convicted rape, murder, and rape-murder defendants in our system are actually innocent. Thus, it seems apparent that more defendants will obtain relief than are actually innocent. Put another way, despite the presence of the contested harmless error system, it appears that significant numbers of actually guilty defendants will obtain reversals. This suggests that harmless error rules may in fact be too lenient, and that the problem of actually guilty people unjustifiably earning relief is far more pronounced in the capital context.
In sum, it appears that courts may already be disturbing jury judgments too frequently, and that harmless error rules should not be cast aside.
Reply: Process Matters; Death Is Different
Regardless of whether guilt-based harmless error rules are characterized as affirmative judgments or deference-based judgments, they still focus the court’s attention on the reliability of the verdict instead of the fairness of the process. While it may be true that juries bear the most responsibility for wrongful convictions, this cannot be a reason to defer to jury judgment in the face of procedural error. The point is that courts can competently judge process, and that fair process has an underappreciated instrumental value in truth seeking. Abolishing guilt-based error rules in capital appeals may not solve the problem of jury mistake, but it would prevent courts from potentially compounding such mistakes.
It may well be that the unreliable application of harmless error rules is a more pronounced problem outside the capital context than within it. But this does not mean that the capital system is working properly. Indeed, a 58–68% reversal rate suggests that capital trials frequently suffer from intense flaws, and that appellate confidence in the system as a whole is not great.11 Moreover, a higher reversal rate in capital cases does not challenge Professor Garrett’s basic finding that appellate courts frequently mistook innocent appellants for guilty ones. All 133 cases with written decisions in the Study were serious crimes—rapes, murders, and rape-murders—and in 67 of those cases (50%), courts (incorrectly) referred to appellant’s guilt.12 Disturbingly, in 13 of these cases (10%), the court actually characterized the evidence of guilt as “overwhelming.”13 This suggests a powerful guilt-confirming bias.
Regardless of how many capital defendants had adverse harmless error rulings made against them, the fact remains that at least 14 innocent men have walked off of death row only because postconviction DNA testing proved innocence that juries and courts failed to detect. This raw number is significant as it represents 14 concrete instances where the state came close to killing the wrong man. While it may be possible that truly guilty individuals have earned some kind of relief because their trials were unfair, Professor Garrett’s study should put an end to the horribly naïve conception that our death penalty system could never shed innocent blood. Since executing an innocent can never be harmless, guilt-based harmless error rules should never apply in death penalty appeals.
Conclusion: Experimental Observations
Although the particular arguments advanced for and against the proposition might easily have varied, this exchange evidences many rhetorical characteristics one would expect when the proposition debated calls upon empirical support for a question that is not strictly empirical. Doctrinal or moral premises behind guilt-based harmless error rules immediately become contested, and empirical interpretations of innocence data vary depending on prior doctrinal perspective. A 58% capital reversal rate can mean the system is catching errors or it can mean the system is constantly breaking, depending on whether one views the system as broken or functioning.
A particular and revealing characteristic of the debate regarding this data set is the unknown (and unknowable) rate of wrongful conviction. Whether more actually guilty defendants win release than actually innocent people suffer incarceration or execution cannot be answered. The inability to answer this question naturally does not deter the use of statistics to buttress an argument that the rate is likely acceptably low or unacceptably high. Yet a rhetorical void is created without an empirical answer to the rate of wrongful conviction question, and this void leaves a more fundamental question unasked: What is an acceptable error rate in the administration of the death penalty? As a society, can we agree that it is acceptable that one innocent be killed for every hundred guilty men executed? Can we agree on one for one thousand?
With the question reframed this way, it seems clear that the argument against harmless error rules in death penalty cases best rests on a nonempirical appeal to the value of an individual life and thus predicts the Reply’s closing focus on the 14 men released from death row.14 Conversely, this reframing suggests that the argument in support of harmless error rules derives its greatest strength from an appeal to the specter of guilty individuals escaping just punishment. In the end, this experiment suggests that the illuminating power of Professor Garrett’s study will depend on our ability to honestly struggle with the ultimate nonempirical questions it raises.
* Acting Assistant Professor of Lawyering, New York University Law School. Mr. Starger was a Staff Attorney at the Innocence Project of Cardozo Law School from 2003 to 2007, where he litigated postconviction DNA access claims and was counsel on four postconviction exonerations.
1. Brandon L. Garrett, Judging Innocence, 108 Colum. L. Rev. 55, 62 (2008).
2. My choice of proposition was inspired by recent remarks of Judge Carolyn Engel Temin, a Senior Judge in the Court of Common Pleas in Philadelphia. I participated in a panel with Judge Temin at a Constitution Project sponsored conference on the legacy of Strickland v. Washington held at the Library of Congress. At one point in her presentation, Judge Temin declared that she wanted to go “on the record” as believing that there can be no harmless error in the penalty phase of death penalty cases. Webcast of Judge Temin’s remarks available at http://www.loc.gov/law/news/webcasts.html. I chose to raise the stakes by shifting focus to the guilt phase and the implied question of innocence. Of course, the idea that different harmless error rules might apply in the capital context has previously received scholarly attention. See, e.g., David McCord, Is Death “Different” for Purposes of Harmless Error Analysis? Should It Be?: An Assessment of United States and Louisiana Supreme Court Case Law, 59 La. L. Rev. 1105 (1999).
3. See Garrett, supra note 1, at 107–09.
4. See id. at 98–99.
5. See id. at 108 tab. 8.
6. Id.
7. See id. at 74.
8. Id. at 99–100. The 6 reversals without written decisions resulted from DNA exonerations prior to retrial. Id.
9. See James S. Liebman et al., Capital Attrition: Error Rates in Capital Cases 1973–1995, 78 Tex. L. Rev. 1839, 1854 (2000). Professor Garrett also notes that his “reversal” rate is calculated slightly differently from the “attrition” rate calculation by Liebman et al. See Garrett, supra note 1, at 100 nn. 167–168.
10. This reversal rate also seems plausibly representative as it is statistically indistinguishable from the 8% reversal rate found in the professor’s own randomly matched comparison group of 121 noncapital cases. See Garrett at 99. Of course, this reversal rate would presumably describe a reversal rate for all rape, murder, and rape-murder convictions combined (the crimes comprising the data sets) rather that describe a more general noncapital reversal rate.
11. Furthermore, it needs to be pointed out that the high 58–68% capital reversal rate may well be lower since the passage of the Antiterrorism and Effective Death Penalty Act (AEDPA) in 2000. Professor Garrett acknowledges that AEDPA had little effect on his data set since almost all the habeas petitions in the innocence group were filed before AEDPA’s 1996 effective date. See Garrett at 101. Similarly, the Liebman, Fagan, and West study concerned only pre-AEDPA appeals. See Liebman et al., supra note 9. However, at least one recent study concludes that significantly fewer habeas writs are being granted post-AEDPA. See Nancy J. King et al., Final Technical Report: Habeas Litigation in U.S. District Courts 60–62, Nat’l Ctr. for State Courts, (2007) available at http://law.vanderbilt.edu/article-search/article-detail/download.aspx?id=1639.
12. See Garrett, supra note 1, at 108.
13. Id.
14. As of January 26, 2008, there have been 15 men released from death row because of postconviction DNA testing. On May 11, 2007, Curtis Edward McCarty was exonerated after serving over 21 years—including 16 on Oklahoma’s death row—for a crime he did not commit. See Cheryl Camp, Convicted Murderer is Freed in Wake of Tainted Evidence, N.Y. Times, May 22, 2007, at A16. I had the honor of representing Mr. McCarty, along with a talented team of lawyers from Oklahoma, during his exoneration.
