#ReadWithClutch: Convicting the Innocent by Brandon L. Garrett

The Book and Its Method

Brandon L. Garrett is a law professor now at Duke, previously at the University of Virginia, and before that a civil rights lawyer working alongside the founders of the Innocence Project. Convicting the Innocent: Where Criminal Prosecutions Go Wrong was published by Harvard University Press in 2011.

The method is the reason to read it. Garrett took the first 250 people exonerated by post-conviction DNA testing in the United States and went to the underlying trial records. Not the exoneration press releases. Not the appellate opinions, which describe evidence at a level of abstraction that hides how it was actually presented. The transcripts. What the eyewitness said on the stand, in what words. What the forensic analyst claimed about probability. What the confession contained and when those contents first appeared in the record.

That is an enormous amount of unglamorous work, and it converts a body of individual tragedies into a dataset. Once you have a dataset you can ask questions that anecdote cannot answer, which is whether the failures repeat.

They repeat.

What the Transcripts Showed

Eyewitness Identification

Eyewitness misidentification appeared in roughly three-quarters of the first 250 cases, making it the most common single factor. The important finding is not the frequency, which was already suspected. It is what the transcripts revealed about the identifications themselves.

In a substantial share of cases, the witness had initially failed to identify the defendant, or had expressed uncertainty, or had identified someone else, and that initial response never reached the jury. By trial the identification had hardened into confident certainty. Confidence at trial, Garrett found, bore little relationship to accuracy, and it had frequently been manufactured by the identification procedure itself: suggestive lineups, confirming feedback from investigators, repeated exposure to the suspect’s image.

Nearly half of the eyewitness cases involved cross-racial identification, a condition the research literature had already flagged as elevating error rates.

Forensic Testimony

Forensic evidence appeared in a comparable share of the cases, and in a large portion of those the testimony was invalid on its face: analysts stating probabilities with no empirical basis, describing techniques as more discriminating than the underlying science supported, or presenting exclusionary results as inconclusive. In some cases the analysis actually excluded the defendant and the jury never heard it.

Readers of the July review of Fabricant’s Junk Science will recognize the pattern. Garrett supplies the denominators that Fabricant supplies the narrative for.

False Confessions

This is the finding that should disturb anyone who has ever assumed a confession settles a question. Forty of the 250 exonerees had confessed to crimes they did not commit. Garrett then examined what those confessions contained, and found that nearly all of them included specific, non-public details about the crime scene: how the victim was positioned, what weapon was used, what was taken.

Those details were then presented at trial as corroboration, on the theory that only the perpetrator could have known them. Since the confessors were innocent, the details came from somewhere else. They came from the interrogation room. The evidence that made the confessions persuasive was evidence the interrogators had introduced.

The Appellate System Did Not Catch Any of This

The chapter on post-conviction review is the one that matters most for accountability work, and it is the least discussed.

Every one of these 250 people was factually innocent. Nearly all of them appealed. Among those whose cases produced published written decisions, only about fourteen percent obtained any vacatur before DNA testing intervened. The appellate system, reviewing the records of 250 innocent people, correctly identified the problem in a small minority of them.

The reasons are structural rather than mysterious. Appellate review is deferential to factual findings by design. Claims are procedurally defaulted if not raised at the right time in the right form, which punishes defendants with inadequate trial counsel twice. Harmless error analysis absorbs identified misconduct without consequence. And innocence itself is not, standing alone, a cognizable claim in most procedural postures.

Garrett’s framing here is precise and worth stealing: the appellate system was not designed to detect wrongful convictions and should not be evaluated as though it were. It was designed to police procedure. When procedure was followed and the outcome was still wrong, the system has no lever.

Why This Is the Methodological Model

I return to this book more than any other for one reason that has nothing to do with wrongful convictions specifically. It is the best demonstration in the accountability literature of the difference between a story and a finding.

A story about a wrongful conviction is moving and forgettable, and it is answerable with the response every institution reaches for first, which is that the case was an aberration. A dataset of 250 wrongful convictions coded from primary records is not answerable that way. When the same failure appears in unrelated cases across unrelated jurisdictions with unrelated actors, aberration stops being an available explanation.

That is the entire logic of the Clutch Justice Judicial Misconduct Database, and of the FitBench framework, and of every piece on this site that treats one county’s docket as evidence about a system rather than evidence about a county. Garrett got there first and did it more rigorously.

The Limits

Garrett names the biggest limitation himself, and it is the one that should keep people up at night. DNA exonerations are only possible in cases with testable biological evidence, which is a small fraction of criminal cases. Most convictions involve no DNA at all. The 250 cases in this study are therefore not a sample of wrongful convictions. They are a sample of the wrongful convictions that happened to be falsifiable.

Whatever the error rate looks like in this dataset, there is no principled reason to think the same practices produced better outcomes in the vastly larger universe of cases where no test could ever be run. That is the uncomfortable implication and Garrett does not soften it.

The other limitation is readability. This is academic writing, and it reads like it. The prose is clear but it is dense, and readers coming from narrative nonfiction will find the pace slower. That is a fair trade for what you get, which is numbers you can cite and a method you can copy.

Assessment

If you are building a case that an institution has a pattern rather than an incident, this is the book that shows you how the argument is constructed. Read Godsey for the psychology, Davis for the structural power analysis, and Garrett for the proof standard. Together they are a curriculum.

Garrett also maintains a public database extending the original research well past the first 250 cases, which means the underlying evidence is checkable. That is not common in this genre and it is the correct model. A finding you cannot audit is an assertion.