Causal Duration Analysis Based on Survival Probability Ratio

For causal effects of a binary treatment on a right-censored duration, the widely used proportional hazard contrasts are non-causal with unrealistic restrictions. This article proposes an alternative flexible causal approach, where we estimate the cumulative hazard, not the hazard itself, using an additive or “exponential-additive” specification with freely time-varying parameters. Our approach includes the proportional hazard as a highly special case that allows only monotonic survival probability ratios (SPR’s), while our approach allows any shape of SPR’s. An empirical analysis on recidivism using the duration until re-arrest after prison release on parole/probation is provided, where the SPR trajectory is not monotonic, but has an inverted-U shape over time.