Deep IV in Law
Appellate Decisions and Texts Impact Sentencing in Trial Courts

This Element analyses real-world trial courts data with state-of-the-art machine learning and causal inference methods.

Zhe Huang (Author), Xinyue Zhang (Author), Ruofan Wang (Author), Daniel L. Chen (Author)

9781009296373, Cambridge University Press

Paperback / softback, published 25 August 2022

75 pages
22.8 x 15.1 x 0.4 cm, 0.09 kg

Do US Circuit Courts' decisions on criminal appeals influence sentence lengths imposed by US District Courts? This Element explores the use of high-dimensional instrumental variables to estimate this causal relationship. Using judge characteristics as instruments, this Element implements two-stage models on court sentencing data for the years 1991 through 2013. This Element finds that Democratic, Jewish judges tend to favor criminal defendants, while Catholic judges tend to rule against them. This Element also finds from experiments that prosecutors backlash to Circuit Court rulings while District Court judges comply. Methodologically, this Element demonstrates the applicability of deep instrumental variables to legal data.

1. Introduction
2. Theoretical framework
3. Dataset
4. Empirical model
5. Results
6. On the practical use of Deep IV for law and economics
7. Limitations from a computer science perspective
8. Limitations from an economics perspective
9. Potential future work
Appendix
References.

Subject Areas: Machine learning [UYQM], Natural language & machine translation [UYQL], Legal system: general [LNA]