Convicting the innocent, and freeing them

A man just got out of prison in Texas after doing 27 years for rape he didn’t commit. And we’re still using the demonstrably error-prone identification process that sent Michael Green away. Inexcusable.

The Harris County (Houston) District Attorney’s office has a unit devoted specifically to identifying innocent people in prison and letting them out. That leaves about 2900 D.A.’s offices that don’t.

One way innocent people get to prison is the “line-up” or “photo spread” in which a victim or other witness is asked to identify the perpetrator from a group of six people, or six photographs. That creates a strong impression that the perpetrator is somewhere in the group, and there’s overwhelming evidence that someone – whoever looks most like the actual perp – is likely to be selected. Once that happens, everything pushes the witness toward more and more certainty about the identification, no matter how spurious.

It turns out that there’s a different way to do the identification process: give the witness a set of photos, or a group of people, to look at one-by-one, asking in each case, “Is this the guy?” In experiments, this approach is less likely to lead to a positive identification, which is why police and prosecutors don’t like it. But the different between the two techniques consists entirely of false IDs. Yet such is the muscle of law enforcement in the political system that most states still allow, and most police departments still use, the “six-pack” process, with its predictably high false-positive rate.

So innocent people continue to lose large chunks of their lives to false verdicts based on eyewitness ID. How many? If the rate of false conviction were 3% – which I would regard as an insanely optimistic number – that would be 50,000 people doing hard time for stuff they didn’t do. It’s hard not to get angry.

Author: Mark Kleiman

Professor of Public Policy at the NYU Marron Institute for Urban Management and editor of the Journal of Drug Policy Analysis. Teaches about the methods of policy analysis about drug abuse control and crime control policy, working out the implications of two principles: that swift and certain sanctions don't have to be severe to be effective, and that well-designed threats usually don't have to be carried out. Books: Drugs and Drug Policy: What Everyone Needs to Know (with Jonathan Caulkins and Angela Hawken) When Brute Force Fails: How to Have Less Crime and Less Punishment (Princeton, 2009; named one of the "books of the year" by The Economist Against Excess: Drug Policy for Results (Basic, 1993) Marijuana: Costs of Abuse, Costs of Control (Greenwood, 1989) UCLA Homepage Curriculum Vitae Contact:

20 thoughts on “Convicting the innocent, and freeing them”

  1. The link goes to a story about high-stakes testing in education, not false convictions. What justification do police departments offer for the lineup method? Leeway to generate admissible pseudo-evidence against suspects they have inadmissible evidence against?

  2. I don't think you have the science right on this — not by a long shot:

    1) Simultaneous arrays don't make it "likely" (i.e., >50%) that someone will be identified even if the true person is not included. I believe that's a misstatement of the relevant studies.

    2) Sequential arrays don't merely reduce false positives without also affecting false negatives. There is indeed a rise in false negatives — in otherwords, a decline in true identifications.

    3) While this isn't my field of expertise, I'm not aware of studies addressing line-ups per se (as opposed to photo arrays). Are you sure that the results are transferrable? And is there any comparison of overall accuracy of in-person line-ups compared to arrays? (It's possible that simultaneous lineups would actually be more accurate than simultaneous OR sequential photo arrays — it's easier to recognize people in person.)

  3. Are most public defenders aware of the lineup-leads-to-false-ID issue? If not, is there no easy way to get this information–if the evidence is strong enough–into their hands? This is just like the infamous field drug screen kits (that have tons of false positives) that aren't meant as substitutes for lab testing, but apparently are used as such, at least enough to pressure someone into a plea.

    Is it a matter of better informing defenders or is the problem that the CJS would grind to a halt if they brought this stuff up whenever appropriate?

  4. There have been studies on the unreliability of witness IDs going on 30 years. These studies have even identified the factors that are likely to make subsequent recall less accurate. These include: high stress (e.g., in an armed robbery one is much more likely to focus on a weapon or a path to safety than the perpetrator's face), stranger or cross-racial identifications, lack of light, speed of the crime, distance from the perp, and so on. The only thing that makes for a really reliable identification is something like a tattoo or a scar or a very unusual voice or facial configuration. These are things that can be reliably retained.

    Some courts actually allow expert testimony to undermine a witness's Id based on this research, but it makes the trial hugely more expensive and time consuming.

  5. "Are most public defenders aware of the lineup-leads-to-false-ID issue?"

    Yes. The problem is that the courts make an "independent" reliability determination that is, ironically, based in part on the subjective certainty of the identifying witness. Happily, the courts also strongly disfavor the presentation of expert testimony on the (un)reliability of eyewitness identification. So it goes…

  6. I'm glad to know that Harris County has its own Innocence Project. It's something every DA's office in the country should have.

    But there are two matters that concern me. The first is this: why aren't DA's offices being more careful about prosecutions to begin with? My (admittedly anecdotal) reading suggests that DAs tend to file the charges with the highest penalties they think they might be able to get a conviction. In a negotiation it makes sense to ask for everything you might want, and then bargain away to the key issue. But this isn't supposed to be a negotiation: the DA (in most cases it's really the ADA) who determines the charges is the first step in the criminal legal system. She's making decisions that markedly affect people's lives and ought to file the charges she believes fit the facts of the case as she knows them.

    The second troubling thing is the reliance on DNA markers for identification. People think of DNA as a fingerprint: no two are alike (except for identical twins). If we were to sequence the entire genome, that appears to be plausible. Everyone (save for identical twins) is different. But that isn't what we do. Instead, what we do is to look for the presence or absence of various markers. So, DNA identification is better thought of as sophisticated blood typing (another method that looks for the presence or absence of markers). It can rule a person out, (if a marker is present that the suspect doesn't have or if the suspect has a marker not present in the sample) but it can't say with certainty that a suspect left the markers. Anyone with that array of markers could have been at the scene.

    At best, Bayes' Theorem can be used to get the posterior probability of a suspect's presence at the scene. That posterior probability is sensitive to the frequencies of marker occurrence in the population. And that's the current problem with DNA identification: we don't have any good data on marker frequencies. States are increasingly requiring convicted felons (in some States felony suspects) to supply DNA samples for typing. But that isn't the proper population for comparison: the proper population is the entire population. Convicted felons are disproportionately members of minority ethnic groups, and marker frequency is strongly affected by ethnicity.

    Basically, when the DA has an FBI expert on the stand who says the (posterior) probability that the suspect was at the scene is one in a bazillion, it's bullshit. The probability is one in a bazillion relative to an unstated population. In order for that posterior probability to be useful, the population it's computed relative to must be stated. If you compute the probability relative to the population of felony convicts, most of whom are behind bars and couldn't have done the crime you have a GIGO computation.

  7. This happened to me when I was the victim of a crime. The perpetrator was an African-American woman, and the cops showed me a photo lineup of five white women and one African-American woman. I refused to make a positive ID, because I really wasn't sure. It really pissed me off. If they had shown me six African-American women, and one of them was actually the perpetrator, I probably could have picked her out easily. And I'm sure there are more clueless and/or less scrupulous people out there who would just pick the African-American woman whether it was the perpetrator or not.

  8. Before going to law school, I heard Barry Scheck speak and read his book on the Innocence Project. Having come out of Navy nuclear engineering, I was steeped in the doctrine of finding and fixing the root causes of problems — once you knew something was an issue, you couldn't rest until you had no only fixed it, but had done so in a way that would prevent it from recurring.

    I naively assumed that the law, especially public officials, would want that same approach used in criminal law, especially around preventing convictions of innocents. Ha! What I have learned since is that both sides in criminal trials have huge incentives to allow errors to perpetuate. Prosecutors, defined by careerism and the body count mentality, don't want anything to get in the way of convictions, and since all convicted people are guilty, they have no problem convicting anyone and everyone that falls in their path to advancement, and it's the job of defenders to worry about errors, not theirs. Defenders, outspent and outmatched in 99.999999% of cases, know that the presumption of innocence is a joke, and live for errors, which can be chips in plea bargains or, in the few cases tried, lead to successful appeals.

    With technology today, it would be trivial to design a truly unbiased facial identification process: take a database of several millions of faces with all faces coded for various attributes (M/F/indeterminate), colors (lightest to darkest), hair coverage (none to lots in various varieties), shape (round to square). Insert the suspect's photo into the database and the program would sort the photo into database next to its closest matches along the various attributes — just between the photos of people darker and lighter, for example of just one dimension.

    Supposed witness sits down. If they are sure of the sex of the suspect, they can select to be shown only pictures of the same sex. They are shown a large centered picture and, surrounding that, a circle of smaller pictures showing all the surrounding photos in the database (the sorted photos for all the other attributes) — if the suspect looks more like the photo on the left, go that way, if more like the one on the right, select that one, more like the guy on the left, select that one. Every time a new selection is made, the selected photo becomes the big center shot, with its surrounding photos circling it.

    And, of course, software would keep track of how many and which photos were used to find "the guy" (and how long the supposed witness looked at each one, and whether they jumped back and forth to compare photos etc.)

    An ID that nails the suspect after such a process seems likely to have a lot of credibility, and little chance for any bias that the supposed witness didn't bring to the party in the first place, which is a different problem.

    That we don't use such a tool — at best relying on small photo books for sequential photo lineups — speaks to our refusal to do anything to interrupt the conveyor belt trial and conviction model that has led to the creation of the American gulag.

  9. In the spring of 1964, as I now recall, I went to a large store on the west side of Chicago for some household items. This store had a multi-level parking garage.

    As I began to get out of my car, two men ran up to my car, with guns, asking for money. I was still sitting in my car, and gave them all the paper money I had with me, something like $30.00 as I recall. That was enough, I guess, and I was not murdered for putting them at risk robbing me without their being adequately compensated for the risk they took. Evidently, $30.00 was worth the risk.

    Of course, I notified store and the Chicago Police. In due time, I was in a Chicago Police Station, looking at pictures. My method was to look for any feature which would rule out each person pictured. Alas, there was one person whose picture did not allow me to rule him out. There ensued an arranged encounter, face to face, in a police station.

    As the photograph was taken from a starkly different angle than the view I had of the person who robbed me during the robbery, there were facial features I saw during the robbery which could not have appeared in the police photograph. When I saw the person in the photograph in person, there was exactly one small facial feature I had noted during the robbery which the person who robbed me did not have and which said person could not have concealed.

    I informed the police that the person could not possibly have been the robber in question.

    Instead of doing what I have a hunch many people may do, in looking for similarities, I looked for at least one facial feature which would eliminate the person, and found one. Whereupon, I determined that eye witness identification is terribly likely to be wrong unless the eye witness previously knows the "perp."

  10. "Sequential arrays don’t merely reduce false positives without also affecting false negatives. There is indeed a rise in false negatives — in otherwords, a decline in true identifications."


  11. Dennis,

    Basically, when the DA has an FBI expert on the stand who says the (posterior) probability that the suspect was at the scene is one in a bazillion, it’s bullshit. The probability is one in a bazillion relative to an unstated population.

    Do you mean the expert says the probability someone other than the suspect left the DNA is one in a bazillion?

    BTW, I thought there were databases in use, at least for paternity cases, that contained marker distributions by ethnic groups. Is this true, or are they inaccurate or somehow unsuitable for forensic work?

  12. Dennis, most of the Innocence Project cases like that profiled in the NYT use DNA evidence to verify that the individual did not commit the crime. This evidence truly was not available at the time. So DNA evidence should help to avoid similar mistakes in certain kinds of crime, rape and violent murder in particular. Don't get me started on prosecutorial ethics. There is wide variation, but often, by the time the prosecutor enters the scene everyone is convinced, or has convinced themselves, of the person's guilt. Reexamination along the way is not as common as it probably should be.

  13. Don't worry, I won't get you started on prosecutorial ethics — other than the lady in Houston, I've never heard or seen any sign of them, anywhere.

    The only way we're going to see any prosecutorial ethics is if election to prosecutor meant that the person elected had to serve as a defense attorney in the same district for an equivalent term of years upon completion of the first term as a prosecutor. Lawyers shouldn't even be allowed to decide that they will only serve the state as a client; and no lawyer should EVER be allowed to serve the state as a prosecutor until they've done at least three years in defense work.

  14. Needtolookagain … i'm looking for material to build into a basic stats course. I was intrigued by your statements, and wonder if you have the citations for the studies you referred to in your commnent above. I'd like to read them for possible use in my class. Thanks … stats professor

  15. Bernard,

    Yes, I mis-stated. Someone other than the accused, etc.

    The last time I looked carefully at the issue was about 5 years ago. At that time, FBI experts were using the marker frequencies from their data base to compute posterior probabilities. I don't know what they're using now, I seriously doubt they're doing much differently. Barbara probably knows more about it than I.


    I am aware that in many of these cases DNA marker evidence was unavailable, and I support the work of The Innocence Project. But today, even the presence of DNA evidence ruling a suspect out doesn't lead to the immediate release. A severely retarded man in Albuquerque gave a coerced confession to the police. A DNA specimen was taken, compared to DNA found on the victim's clothing and three weeks later the suspect was ruled out. The man spent a total of 968 days in custody. Most of his time was in administrative segregation: he was accused of the rape and murder of an 11 year old girl.

    The police and DA claim the held him because he knew details of the murder, details he was fed in the interrogation. Penn and Teller have it right: our criminal justice system is bullshit.

    I've tried to link to the Albuquerque Journal's website, but it's not loading for some reason. The story is in today's (August 14, 2010) paper, under Scott Sandlin's by-line. The accused man's name is Robert Jacob Garcia.

Comments are closed.