"Bayesians v. Frequentists": A Case for the Former

While I have neither the time nor interest in wading into a "Bayesian v. Frequentist: Holy War" debate, I nonetheless wanted to pass along an informative and balanced description (click here) of leading reasons to consider Bayesian strategies from Andrew Gelman (Columbia--statistics). While Andrew certainly has a point of view (and various priors), his helpful post succeeds in its goal of identifying and briefly describing various leading reasons to consider Bayesian strategies.

Basic Primer on Stata Graphics

While the graphics functions in Stata remain among the more cumbersome and nettlesome, helpful external resources exist for those interested in a quick primer.

Among the more helpful remains Maarten Buis (Konstanz) workshop (click here). To get the most out of the site requires that users access (in Stata) the 1978 "training" auto data set. To do so, simply enter the following command to grab the data: sysuse auto

With the training auto data set, users can follow along and practice generating graphs ranging from the more basic (e.g., line, scatter, and bar graphs) to the more complex (e.g., twoway, stacking, and predicted probability graphs).

"Making Bail" and Race

Megan Stevenson (UVa) begins a recent paper, A Decomposition of Racial Disparities in Pretrial Detention, with the observation that "[U]nderstanding what causes racial disparity in criminal justice is an important step towards learning how to ameliorate it." And to this end, the paper dives deeply into bail data from Philadelphia courts. Specifically, the data set covers all Philadelphia arrests in which charges were filed between September, 2006 and February, 2013, and involves 86,469 white and 172,860 black defendants.

The paper's main--and intriguing--take-away is that "differences in ability/willingness to pay bail are a substantial contributor to the race gap in pretrial detention, particularly at the higher end of the bail distribution." The paper's abstract follows.

"Understanding what causes racial disparity in criminal justice is an important step towards learning how to ameliorate it. Previous research on racial disparities in pretrial detention have focused primarily on evaluating the racial bias of judges, but racial differences in ability-to-pay to monetary bail could also contribute. In this paper I conduct a Oaxaca decomposition of racial disparities in pretrial detention to determine how much is due to black defendants having higher bail and how much is due to black defendants being less likely to post, conditional on the bail amount. Using detailed court records from Philadelphia, I find that differences in the ability to post bail account for 1/3 of the race gap in detention. Differences in posting rates are particularly pronounced at the higher quantiles of the bail distribution: white defendants who must pay a bail deposit of $2500 to secure release are twice as likely to post as black defendants with that same bail. Racial differences in the bail distribution are substantial, but disappear entirely once legally relevant characteristics such as charge, criminal history, age and gender are accounted for. Finally, I discuss how racial differences in bail posting rates complicates tests of racial bias."

CELS 2025 Registration

At the request of the CELS 2025 conference co-organizers, I am delighted to pass along the following conference registration information:

 

Conference on Empirical Legal Studies

24-25 October 2025

Conference Registration

Conference registration information and link are found here.

Even Descriptive Statistics (and Tables) in Stata Got A Bit More Powerful (and Complicated)

While perhaps I shouldn't be surprised, but even some simple, quick-and-easy descriptive commands in Stata have become a bit more complex. The up-side, however, is that these commands became a bit more powerful and useful as well.

One recent example is that, and helpfully described and summarized in a tutorial video (here), the standard sum command may now be overshadowed by a more flexible and powerful dtable command. One quite nice feature is that the dtable command includes sub-commands (e.g., tests) that will automatically subject variables to appropriate tests to generate p-values. Again, in all, these new features, while not earth-shattering, involve more efficient coding.

Should Confidence Intervals Increase or Decrease Confidence?

As described in this helpful essay (here), a confidence interval is one of those "statistical inventions that hides its trick in plain sight" as it implies "a promise of certainty, when in fact they are a carefully hedged warranty." Thus, paradoxically, confidence intervals' highest use may well be, as Nasir Bashir notes, to underscore a key--yet under-appreciated--point that "statistics is a tool for dealing with uncertainty, not for producing certainty. Confidence intervals are a reminder of that."

Structuring Coding for "State-" and "District-Level" Controls

While this discussion on the StataList (click here) does not specifically speak to the legal context per se, the logic of what is being discussed carries over directly to law-related research that seeks to include, e.g., district and circuit court control variables in a single regression specification. The relevant issues discussed span from how (right-side) control variable ordering in the coding can matter to which level of clustering makes the most sense.