--- Dennis Claxton <ddclaxton at earthlink.net> wrote:
> and, just for grins, there's more statistical
> evidence of that sort here;
I am unimpressed - as I said, small sample which is a particularly bad problem given non-probability nature of it, and the controls they use are rather pedestrian, to say the least - not to mention that the whole piece reads like an advocacy piece where the data fit the already made conclusions, instead of demonstrating 'effects' of race in a scientifically accpeted manner.
What really kills this type of research from a statistical point of view is that first of all the sample is rather small due to the very nature of the issue - there not that many capital cases to beign with. As a result, a few "non-representative" cases can substantially skew the distribution and lead to erroneous results.
BTW, a decent sample size would be about 900 or more if you want to split to introduce controls, but there are usually fewer than a hundred capital cases per state, unless one lumps cases over a ridiculously long time period (over thirty years in the report you cite). That per se is not necessarily bad, but it requires addtional and quite sophisticated controls to compensate for varius changes over time, which in turn requires even larger sample sizes.
This is already a serious weakness that is further substantially aggravated by the fact that capital cases are self-selected by various factors: prosecutorial discretion (which in turn is affected by factors that may or may not be related to race, so they need to be separately controlled), gravity of the offence, quality of the evidence, public pressure - to name just a few - which results in the situation that certain cases have a much greater probablity of being tried as capital cases that other cases involving the same type of crime. That by defintion is a non-probablity or biased sample, which cannot be easily compensated by the usual methods (weights) because probabilities of selection are not known.
So it is quite possible that cases involving, say, white victims are more likely to be prosecuted as capital offences than other cases due to any of the factors I listed above - or their combination, which seriously biases the sample due to it small size. And there isn't really any way to control for that. The appearance that murderers of white victims are more likely to get death penalty than other murderers in such a situation is really only a reflection of a sample bias.
That is why using statistical inference in such cases is really shaky and basically pseudo-science. But then again, who needs science when the conclusions are already made - as for example suggested by your comment about courts trying to cover their own bias by not admitting statistical evidence. If one already knows that the courts are biased, why bother using science to "prove" it, especially if it is of questionable quality anyway?
Wojtek _______________________ DISCLAIMER: Opinions posted by this writer to this forum are solely forms of literary criticism exercised as the First Amendment right, and do not necessarily reflect the author's views or attitudes toward real-life people, including other writers posting to this forum, groups of people, institutions, or events to which the critiqued texts may refer, either explicitly or implicitly. Any statement asserting or implying such views or attitudes on the basis of this writer's opinions posted to this forum is thus unfounded, and may be libelous. ________________________
__________________________________________________ Do You Yahoo!? Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com