Long-term follow-up studies typically follow up individuals convicted of a particular target offence repeatedly and regularly, updating their conviction status. This is especially valuable for specialist forms of offence such as sexual offending, where the risk of reoffending appears to stay relatively high even after a long conviction free period. Conventional approaches to such data would use logistic regression, assuming that covariates collected at the target offence still have predictive power many years later. However, it is likely that the effect of many covariates will fade over time. This talk presents methods for operationalising time-varying effects in hazard rate analysis for covariates measured at a single point in time, using as illustration a sample of middle-class offenders who were released in the early 1970s and followed up at 10 years, 20 years and 35 years. We show that while previous criminal history variables still have predictive power many years after release, other demographic variables which are important in short term reconviction appear to fade with time