Was Harold Zurcher myopic after all?: replicating Rust's engine replacement estimates

Rust (1987) studies the dynamic decision making under uncertainty made by Harold Zurcher to replace bus engines. In the decades since, the model has been applied, extended, and used as an example multiple times. This paper resolves some discrepancies in how data were transformed in the original and subsequent archives. Using a package that standardizes computation of estimated dynamic programming, it replicates the 12 original maximum likelihood estimates and the six main hypothesis tests of whether Zurcher's decisions were myopic or not. The discrepancy in the data processing results in modest differences in estimates and log‐likelihoods, but the p ‐values are essentially the same because the differences are very similar across values of the discount factor in the tests.