A COMPARISON OF THE “SICKLE FUNCTION” WITH ALTERNATIVE STOCHASTIC MODELS OF DIVORCE RATES

This study explains cohort-specific divorce distributions using a hypothesis with fuller information about the process generating mechanism such as life cycle effects. Using marriage cohort data for Austria (1959-1974 1960-1974) and the US (1949-1964) 6 hypotheses of competing stochastic models are compared: 1) the Poisson model which assumes a constant risk of divorce not influenced by a marriages duration; 2) the Weibull model which is a generalization of the Poisson model; 3) the log-logistic model which adds a damping factor to the Weibull model yielding a (inversely) U-shaped hazard function; 4) the model of cumulative inertia which has an exponentially declining hazard rate; 5) the mover-stayer model which postulates a fraction of a priori immune cases and an a priori fraction of non-immune cases exposed to the Poisson process; and 6) the sickle model whereby an empirically adequate hazard function increases after the wedding reaches a maximal value declines thereafter and tends to zero asymptotically. The sickle hypothesis works best because: 1) the inverted U-shape of the risk function fits the observed risk path well and 2) the sickle model is able to explain even a large fraction of "immune" cases. This sickle model shows how the divorce patterns in the 2 countries differ in 3 ways: 1) the maximal frequency of divorce is reached earlier in the US than in Austria; 2) in the US frequency of divorce in the first year after marriage is more than proportionate; and 3) the propensity to divorce is higher in the US than in Austria. The extensive conformity of this life cycle hypothesis with observed cohort-specific data shows very low period effects. Yet there are intercultural and cohort effects evident from the strong variation of the model parameters by cohorts and countries. Parametrization of the cohort specific model parameters is needed to identify regularities of second order. Given the number of couples in a marriage cohort the model then can be used to make divorce frequency forecasts.