Shashank Singh

We conduct a survey of economists and a representative sample of Americans to infer the reduction in the perceived value of a paper when its authors have conflicts of interest (CoI), i.e., they have financial, professional, or ideological stakes in the outcome of the results. On average, a CoI decreases trust in the conclusions of an economics paper by 30%. This reduction in trust reflects a combination of the frequency of conflicted papers and the bias of papers when they are conflicted. To isolate the second term, we introduce a key construct: the CoI Discount, which measures the reduction in the value of a conflicted paper relative to a nonconflicted one. We show that, on average, conflicted papers are worth less than half of non-conflicted ones, though this effect varies significantly depending on the nature of the conflict. The discount is more pronounced when the conflict involves the interest of a private rather than a public entity. Restricted data access also leads to a substantial discount. We validate our survey based estimates by comparing them to actual biases observed in conflicted papers within the economics and medical literature.

AI-assisted methods can analyse large-scale legal datasets with approximately 70% accuracy, enabling more effective monitoring of environmental litigation outcomes. Large language models (LLMs) like ChatGPT-4 can assess whether court rulings have a positive environmental impact with accuracy rates that approach human expert analysis. Analysis of 12,615 environmental court cases in India reveals that approximately 35% of rulings are intended to be favourable to the environment, with significant variations across courts and case types.

Can judges enforce environmental justice? Though citizens increasingly rely on the judiciary to enforce environmental regulations, there is little empirical evidence on the effectiveness of judicial policies in improving environmental outcomes. We report the first estimates of the causal effects of judicial orders on water pollution and infant mortality in India. We construct a comprehensive dataset spanning four decades that includes court cases, judicial decisions, pollution indices, and infant mortality rates. We leverage the quasi-random assignment of cases to judges in India’s courts, as well as the writing style of judges in prior cases. We find that ’green’ cases are temporally associated with reductions in peak toxicity levels, but have almost no impact on infant mortality rates in subsequent months. Several years post-decision pollution and mortality rates exceed pre-decision levels. This analysis highlights the potentially limited effects of judicial environmental policies in high pollution settings such as India.

Collecting the available universe of High Court decisions in Kenya, we leverage the random assignment of cases to judges to evaluate the extent and drivers of judicial bias along gender and ethnic lines. We find that defendants are 4 or 5 percentage points more likely to win if they share the judge’s gender or ethnicity, respectively, but there is no in-group bias towards plaintiffs. We show that this effect is driven by mild bias among a large group of judges. We also find that judges displaying stereotypical or negative gender attitudes in their written judgments are more likely to display gender bias in their decisions. These results provide evidence that judicial bias is rooted in prejudiced attitudes, widespread but relatively mild, and conditional on whom the bias is targeting. This evidence sheds new light on the sources of, and potential solutions to, judicial bias.

Under what conditions do judges favor their own group? Collecting the available universe of Superior Court decisions in Kenya, we leverage the random assignment of cases to judges to evaluate the extent of judicial in-group bias along gender and ethnic lines. We find that defendants are 4 or 6 percentage points more likely to win if they share the judge’s gender or ethnicity, respectively, and that judges are significantly more biased in favor of defendants than plaintiffs. Our findings highlight the need to re-examine the emerging consensus that judges uniformly favor their own group and continue investigating the mechanisms driving judicial bias. We propose that the uneven application of bias can be explained by a framework of social identity and loss aversion, and we support this claim with data on the amount of damages in each case, which we extract using a large langauge model. Finally, we argue that our framework could serve as a more complete lens through which to decipher the cognitive underpinnings of judicial biases.