Empirical Legal Studies

Exploiting Facebook and Twitter Data

With Facebook (and Cambridge Analytica) and Twitter persistently featured in the news (and current events) of late, and as we're beginning to see an uptick in scholarship seeking to exploit data from Facebook and Twitter platforms, I thought the following two Stata Blog links might interest anyone seeking to import Facebook and Twitter data into Stata for analysis.

Race and Bankruptcy "Steering" or "Selection"?

Empirical bankruptcy scholars continue to explore reasons for the comparative "over-abundance" of African-Americans pursing Chapter 13 over Chapter 7 protection from creditors. Some scholars seeking to explain this anomaly point to a "saving your home" impulse while others hypothesize that bankruptcy lawyers' racial discrimination "steering" helps account for African-American preferencing Chapter 13 over Chapter 7.

The latest contribution to the literature, Race and Bankruptcy, by Ed Morrison (Columbia), Belisa Pang (Columbia), and Antoine Uettwiller (Imperial College Business School), tests an alternative hypothesis that pivots on a selection effect: "Financially distressed African Americans may be more likely to benefit from Chapter 13 than other consumers." The paper specifically notes the influence of the automobile as helping to account for a preference among African-Americans for Chapter 13. Drawing from data from Atlanta, Chicago, and Memphis, and exploiting a recent city-wide program in Chicago that resulted in a far more aggressive stance on collection of parking debts and automobile seizures, the paper finds that Chicago's city-wide program disproportionately affected African-Americans and, as a result, provided upward pressure on African-Americans' share of Chapter 13 filings. An excerpted abstract follows.

“Among consumers who file for bankruptcy, African Americans file Chapter 13 petitions at substantially higher rates than other racial groups. Some have hypothesized that the difference is attributable to discrimination by attorneys. We show that the difference may be attributable, in substantial part, to a selection effect: Among distressed consumers, African Americans have longer commutes to work, rely more heavily on cars for the commute, and therefore have greater demand for a bankruptcy process (Chapter 13) that allows them to retain their cars.… We show this using data from Atlanta, Chicago, and Memphis, each of which has been identified as a location with over-representation of African Americans in Chapter 13. We then test our hypothesis that African Americans’ reliance on automobiles is a cause of their substantially higher use of Chapter 13. We do this using data from Chicago, where the city recently implemented an aggressive program to collect parking debts by seizing the cars and suspending the licenses of consumers with large debts. We show that this city-wide program disproportionately affected African Americans and, as a result, their share of Chapter 13 filings increased substantially. Although we do not disprove the possibility of discrimination by attorneys, our data show that selection effects are potentially as important in explaining patterns in Chapter 13 cases.”

A Failure to "Replicate" or "Communicate"?

A recent post at Andrew Gelman's (Columbia--Statistics) blog (here), dives into fascinating, timely, and increasingly contested terrain involving "replication" studies. A quick summary of the salient background follows.

In a 2017 paper (here) Ballarini & Sloman claimed that it "failed to replicate" earlier findings reported by Dan Kahan (Yale), Ellen Peters (Ohio State) et al. (here). However, at various points in subsequent email exchanges among the various sets of authors, published in Gelman's blog post, Ballarini & Sloman appear to have either: a) "walked back" their published critique of Kahan, Peters et al., or, at the very least, b) noted that word choices in their published claim ("failed to replicate") may have been in apt or inexact. As the private email exchanges (and, perhaps, "clarifications") were insufficient to fully address Kahan's concerns, in the end Kahan & Peters (2017) elected to publish a formal "reply" paper of their own to make their case against Ballarini & Sloman's 2017 claims about "replication" (or, more specifically, the lack thereof).

Setting aside the particulars of this specific case, this incident reflects a broader point about the need to gain greater clarity and precision on what it means to "replicate" in the context of academic research. And this is especially true as various academic fields come to grips with a "replication crisis."

Equality By the Numbers? Algorithmic Tools and Fairness

Unsurprisingly, as attention to (and the use of) algorithmic tools increases a diverse set of complex issues emerges. This is especially true in the criminal law space. While the relevant technical empirical literature continues to grow and develop, a recent paper by Aziz Huq (Chicago), Racial Equity in Algorithmic Criminal Justice, helpfully "addresses the legal and moral dimensions of predictive algorithmic criminal justice tools that pertain to racial equity." The paper's abstract follows.

"Algorithmic tools for predicting violence and criminality are increasingly used in policing, bail, and sentencing contexts. Although some attention has been given to their procedural due process implications, how these instruments interact with the enduring and complex racial legacies of the criminal justice system is presently not well understood. This Article analyzes the questions of racial equity raised by these new predictive instruments using two lenses: constitutional doctrine and emerging technical standards of “algorithmic fairness.” I demonstrate that constitutional doctrine is poorly adapted to addressing the range of racial issues that potentially arise with algorithmic criminal justice. Instead, I demonstrate that the difficult questions of racial equity in this domain are best framed and evaluated though certain, but not all, emerging technical standards of algorithmic fairness."