Kevin E Staub, Rainer Winkelmann. Consistent estimation of zero-inflated count models

Kevin E Staub, Rainer Winkelmann. Consistent estimation of zero-inflated count models


The so-called problem of “excess zeros” plagues a majority of count data applications in health economics and other social sciences: The proportion of observations with zero counts in the sample is often much larger than that predicted by standard count models. By far the most popular explanation for the high proportion of zeros is that in addition to the standard count data process a second process produces extra zeros. For instance, consider the demand for health services as measured by the number of physician visits. A person might have had zero physician visits in a given time period because (i) she is healthy and does not require visiting physicians, or because (ii) despite requiring physician services regularly, no visit was observed in the time period. The zeros product of (i) –sometimes called ‘structural’ or ‘strategic’ zeros– stem from a binary process, in this case suffering from a health condition. The zeros product of (ii) – sometimes called ‘incidental zeros’– correspond to realizations of a count process to which only the ‘population at risk’ is subjected; in this example, individuals afflicted by a health condition. These models allowing for two separate types of zeros are known as zero-inflated count models (Mullahy, 1986, Lambert, 1992), the most prominently represented being the zero-inflated Poisson and zero-inflated negative binomial models.


Просмотреть документ

 (no votes)

Keywords: Economic Modelling

Printable version

Go back

Go to top