Friday, January 11, 2013


It is prisons Friday and I decided a look at pretrial detention might be in order.  Surprise, surprise, seems that like pretty much everything else in the US of A, race seems to play a role.  A large body of research finds that Black, Native Americans, and Latino defendants are much more likely to be detained prior to trial then are white defendants, and that legal factors do not account for the disparity.

Obviously, class also plays a huge role in being actually released on bail. If bail is granted, they you either have the bucks are you don't.  

Of course, numerous previous studies have shown that those who don't make bail are those who are most often then convicted.  Other studies show that those locked up prior to trial are far more likely to plead out than those free on bond.  Still other studies have found that pre trial detention makes a more severe sentence all the more likely for the person locked up then the person out on bond.  

Thus, whether a person can get out on bail impacts even more then just being locked up before trial.

I could go on, but you get the drift.

...and I won't even get into the whole sleazy business of bail bonding....

I found this thing at the Social Science Research Center web site.

Testing for Racial Discrimination in Bail Setting Using Nonparametric Estimation of a Parametric Model

Jonah B. Gelbach 

Yale Law School; Yale University - Department of Economics Program in Applied Economics and Policy

Shawn D. Bushway 

University at Albany

August 20, 2011

Black defendants are assigned greater bail levels than whites accused of similar offenses. To investigate whether this difference can be explained without taste-baseddiscrimination, we construct a simple model of optimal bail setting. We develop a two-step econometric method that allows us to estimate the model while holding constant defendant heterogeneity that judges can observe, even when we do not. In return for making the behavioral model’s relatively weak parametric assumptions, we are able to allow an arbitrary conditional distribution of such heterogeneity. We estimate the model using publicly available administrative data on felony defendants for five counties in 2000 and 2002. Point estimates suggest discriminatory bail levels in at least one, and possibly two, of these counties, where estimates suggest that judges set bail as if the value of blacks’ lost freedom is less than two-thirds the value of whites’ lost freedom. This result translates into a substantial black-white gap in the value of lost freedom — at least $64 per day.


...Table 1 reports some basic summary statistics for the data we use in the analysis below. The first two columns of the table report descriptive statistics for blacks and whites charged with a violent crime. Average bail for blacks charged with violent crimes is roughly $36,000, about $7,000 greater than the average bail assigned to whites charged with these crimes. One consequence of the higher average bail facing blacks charged with violent offenses is that only 40% of them made bail, by comparison to 53% of whites charged with violent offenses.

A naive approach to testing for discrimination would take the $7,000 difference in bail for defendants charged with violent offenses as evidence of anti-black bias. However, blacks and whites accused of violent crimes differ in many ways. For example, the “Any prior FTA” row shows that 32% of the blacks in the sample have previously failed to appear in court, by comparison to only 21% of whites. Additionally, 26% of blacks in the violent-offense category could not have previously failed to appear, meaning that they were not previously ordered to appear in court; among whites, this figure is 33%. Blacks charged with violent crimes also were more likely than whites to have been arrested previously (74% versus 67%), and more of them had been arrested at least ten times (22% versus 16%). They were also more likely to have previously served time in prison (24% versus 18%, though roughly the same share of blacks and whites had previously served time in jail—45% versus 44%), and they were also twice as likely to have an active criminal justice status at the time of charging. In addition, blacks facing a violent-offense charge were less likely to be represented by a private attorney, with black-white differences in the share represented by public defenders accounting for most of this difference...8


As with violent offenses, blacks charged in the other three offense-type categories also face higher bail levels than do whites. These differences are especially large for drug crimes (about $13,000) and public order crimes (more than $10,000). Blacks charged in these three categories are also much less likely than whites to make bail. As with those charged with violent crimes, blacks in the other three categories uniformly have more extensive criminal- justice backgrounds than whites, as measured by the variables considered in Table 1...


We consider a model of optimal bail setting in which judges choose bail levels to trade off various costs. On one side are the costs of jailing defendants who do not make bail. These costs include both the pecuniary costs of running jails and the like, and the value judges place on freedom lost by (potentially innocent) defendants. On the other side are the social costs expected to be imposed by those defendants who make bail and are released pending trial. The optimal bail level is an implicit function of the defendant’s characteristics, including x, which both we and the judge observe, and u, which only the judge observes. The presence of heterogeneity that we cannot observe invalidates conventional cross-race comparisons of average bail levels, even conditional on x.

We develop an approach to testing for discrimination based on estimation of the judge’s first-order condition in an auxiliary regression. There are two key steps in making this estimation method practical. First, when there is a one-to-one mapping between optimal bail and the unobserved heterogeneity term u, conditional on x, one need not know a defendant’s u value to consistently estimate his conditional choice probabilities at the optimal bail level, given (b, x, u). This result allows us to leave the distribution of u entirely unrestricted, even conditional on x.

Second, because partial derivatives of defendants’ conditional choice probabilities can be written in terms of the levels of these probabilities under a multinomial logit structure, we do not need to directly estimate the partial derivatives of conditional choice probabilities with respect to bail. Instead, we are able to estimate these partial derivatives using nonparametric estimates of the levels of the conditional probabilities. We then recover the key parameter of interest, Rv, by estimating the first-order condition after re-writing it in terms of auxiliary variables that depend only on the level of bail and the conditional choice probabilities. 

Our point estimates suggest that judges value freedom significantly less for blacks than whites in Harris county, and in our preferred specification, Dallas county as well. Back-of-the- envelope calculations suggest that judges in these counties value lost freedom substantially less for blacks than whites, with $64 per day being a lower bound. For a defendant held until trial for a typical period of nearly three months, this range amounts to several thousand dollars’ difference.

These findings suggest the possibility of substantial bias against blacks in bail setting. This result is important for multiple reasons. First, it is evidence of substantial judicial bias against blacks, which is of per se concern. Second, bail affects defendants’ utility directly by affecting the probability that a defendant will lose his freedom for the potentially lengthy pre-trial period. Discriminatorily higher bails thus will make black defendants worse off in this sense. Finally, being held over might also affect the probability of conviction. Our results are therefore of substantial legal policy interest....

No comments: