Predicting Criminal Recidivism Using "Split Population" Survival Time Models

Abstract

In this paper we develop a survival time model in which the probability of eventual failure is less than one1 and in which both the probability of eventual failure and the timing of failure depend (separately) on individual characteristics. We apply this model to data on the timing of return to prison for a sample of prison releasees4 and we use it to make predictions of whether or not individuals return to prison. Our predictions are more accurate than previous predictions of criminal recidivism. The model we develop has potential applications in economics; for example it could be used to model the probability of default and the timing of default on loans,

User-Generated Open Source Products: Founder's Social Capital and Time to Product Release

“…However, this probability rarely equals 1 for all observations (Schmidt and Witte 1989), because some open source projects may never release a product file. To account for two subpopulations (events that occur and those that do not), we modify the basic hazard specification with a split-hazard formulation (e.g., Dekimpe et al 1998, Sinha andChandrashekaran 1992); the probability of eventual product release i is a function of covariate vectors, including founders' social capital variables X DU i , characteristics of the interplay between communities X EU i , and OSS project and product characteristics X C i , as…”

“…The fit criteria indicated that the model with a log-logistic distribution (model M2: BIC M2 = 3 958 92, CAIC M2 = 3 991 92) 2 fit the duration data better than the model with the log-normal distribution (M1: BIC M1 = 4 375 70, CAIC M1 = 4 408 70) or the gamma distribution (M3: BIC M3 = 3 983 37, CAIC M3 = 4 017 37). Second, in the log-logistic model, a comparison of a probit (M4) versus a logit (M5) specification for the probability of eventual product release in the split-hazard specification (Schmidt andWitte 1989, Sinha andChandrashekaran 1992) shows that the logit specification fits the data better (M5: BIC M5 = 3 187 46, CAIC M5 = 3 942 46; M4: BIC M4 = 3 797 78, CAIC M4 = 3,861.78). We also find a statistically significant improvement in this logistic distribution specification for the hazard with logit link function specifications for the probability when we account for unobserved heterogeneity in the model (M6).…”

How do different entitlements to unemployment benefits affect the transitions from unemployment into employment?

“…Up to now we have neglected the existence of long-term survivors (or infinite durations). This approach has been used in the econometric literature in the context a split-population framework for a single risk (Schmidt and Witte, 1989) while Addison and Portugal (2003) offer a generalization of the split-population model to independent competing risks. 3…”

Proportional hazards models with discrete frailty

“…Lifetime models with this feature have been used widely [8]. They are also known as limited failure models [9] and as split-population models [10].…”