Stability of Parameter Estimates in the Split Population Exponential Distribution

Maltz and McCleary have suggested a split-population exponential distribution to predict parolee recidivism where one portion of the population will never return and the remainder will return according to an exponential distribution. They have estimated parameters to fit the distribution to some sample data. A reanalysis of their data shows that the parameter estimates change systematically as more and more observation time is allowed in the computation. It is argued that this feature limits extrapolability. Local data concerning readmissions of discharged psychiatric inpatients are used to illustrate this point.