Various readers generously sent me copies of the Hartley article I mentioned here (Hartley, R.D., et. al., Do you get what you pay for? Type of counsel and its effect on criminal court outcomes, Journal of Criminal Justice (2010)), and I’m puzzling my way through it, giving myself a crash course in statistics on the way.
That crash course is not simplified by the fact that, throughout the article, Hartley and his colleagues are (I finally figured out) using “more. . . than” when he should be using “as. . . as,” as in:
Property and drug offenders are over 3 and 6 times more likely to be released ROR than violent offenders.
What the statistics show is that Exp(ß) for property offenders (as opposed to violent offenders) being released on their own recognizance is 3.20. That is, property offenders are 3.20 times as likely to be released ROR as violent offenders are, or more than two times more likely to be released ROR than violent offenders.
(ß, also b in Hartley’s notation, is the natural logarithm of the mean of results in case 1 divided by the mean of results in case 0. So if the mean of results in case 1 is .32 and the mean of results in case 0 is .1, ß is 1.163 and Exp(ß) is 3.20. Exp(ß) can never be negative, so if an outcome is as likely in case 0 as in case 1 it equals 1—not “one time more likely than” but “as likely as.”)
The first quote in my first post on the subject should say:
Black defendants who retain a private attorney are almost twice as likely to have the primary charge reduced as black defendants who are represented by a public defender.
. . . which is very different than the proposition as Hartley phrased it.
There are other errors in the article. Without picking the nits, Hartley’s first dependent variable, “Pretrial Status (ROR)” is poorly defined. In Table 1, case 1 is defined as “released” and case 2 is defined as “detained or had bail set.” But in the text, Hartley and his colleagues write that “a majority of the defendants (71.3%) had bail set in their case.” Could the authors be that unfamiliar with their own data?
It’s to take such distractingly sloppy writing seriously, but I’ll take a run at it, and write next about the conclusions that Hartley’s data support.