Former federal judge, now law professor Paul Cassell links to the prosecutor's account of what happened in the Troy Davis case. You may know about his case: After being convicted of murdering a police officer, Davis has been pursuing a new trial ever since seven of nine non-police witnesses against him recanted their testimony, basically saying they were coerced into giving their statements by the cops.
There's a lot of hand waving going on here, but basically the reliability of the recantations comes down to whether or not you think it is more likely that witnesses are subject to intimidation by police investigators, or by groups like Amnesty International investigating potential wrongful convictions. My default stance is to be suspicious of unrecorded police questioning in Black on White crimes in Georgia. The affadavits Amnesty provides are compelling to me because of their consistency -- they all give very similar accounts of police harassment, and come back to a consistent theme that the police wouldn't let the witnesses leave until they said what the police wanted to hear.
For his part, the prosecutor, Spencer Lawton, actually uses the 80% recantation rate as evidence in favor of collusion and manipulation. He raises the specter of anti-death penalty forces coming in, long after the event in question, and basically guilting the witnesses with the specter that their testimony may be putting an innocent man in prison. Is it possible you don't remember clearly? Are you sure that Davis is the shooter?
The problem with this defense is that it doesn't match up with the affidavits Amnesty actually obtained. They're not, by and large, based off fuzzy memories or unclear recollection of the events in question. They, nearly without exception, are premised off police intimidation to get certain testimony -- something unlikely to be that fuzzy and not premised off any difficulty in remembering the events of the murder. Lawton notes that the witnesses were questioned about this at trial and did not say anything about intimidation. But at trial the police threats (if they so happened) were likely to be fresh in the witnesses mind, and going against one's sworn statement raises the prospect of a perjury investigation (something which, if I were a sleazy cop, would be sure to remind my cowed witnesses of after they signed the document).
The other part of Lawton's argument, which is a bit more subtextual, is that the affidavits were acquired by groups with an anti-law and order agenda in general, who are motivated simply by blanket opposition to the death penalty. The problem is that Amnesty appears to have worked quite hard to acquire statements implicating another man in the crime. Opposition to the death penalty hardly seems consistent with trying to get another person convicted of capital murder (although they could hope that the new suspect would not get a death sentence).
All of this really just points back to my current, long-standing intuition about the death penalty. Our system is so badly broken that there is really no way that it can imposed in a fair and just manner, where we can be sufficiently assured of the defendant's guilt. When, for example, there isn't that essential trust that the police won't try and coerce testimony -- particularly when the victim was another officer -- it is simply impossible to evaluate the reliability of trial testimony in the face of future recantations. In such an environment, the death penalty cannot be justly put into practice.