Abstract
Negative binomial model was applied to the reimbursement records of large companies in Japan. The data include about 540 thousand male/female employee and spouse reimbursement records for 2 years (1997-1998). The means and variances in health capital could not be directly observed from reimbursement records. However, the number of healthy days, defined as the number of the days individuals spend without health care, could be obtained by subtracting the number of inpatient and outpatient days in each month. In the demand equation for health care, the dependent variable is the number of treatment episodes, and the independent variables include mean healthy days per month in the preceding year, while standard deviation of healthy days by month in the preceding year, in addition to age, wages, etc. Mean and standard deviation of healthy days in the preceding year were used as proxies of the expected health and the degree of uncertainty in health. Estimated results were consistent with the results of theoretical models which modify the hypothesis on the distribution of health capital in Dardanoni and Wagstaff (1990). Mean healthy days has a negative effect on the demand for health care. A one day increase in mean healthy days reduces treatment episodes by 0.09. Standard deviation in healthy days has a positive effect. A one day increase in standard deviation in healthy days increases treatment episodes by 0.02. These results show that patient behavior is related not only to the expected health but also to the variance in health.
