Forecasting accuracy of behavioural models for participation in the arts

In this paper, we assess the forecasting performance of count data models applied to arts attendance. We estimate participation models for two artistic activities that differ in their degree of popularity -museum and jazz concerts- with data derived from the 2002 release of the Survey of Public Participation in the Arts for the United States. We estimate a finite mixture model - a zero-inflated negative binomial model - that allows us to distinguish "true" non-attendants and "goers" and their respective behaviour regarding participation in the arts. We evaluate the predictive (in-sample) and forecasting (out-of-sample) accuracy of the estimated models using bootstrapping techniques to compute the Brier score. Overall, the results indicate good properties of the model in terms of forecasting. Finally, we derive some policy implications from the forecasting capacity of the models, which allows for identification of target populations.