5.

# Results

# Univariate regression results

Table 4 shows the marginal effects from univariate regressions of the two chronic diseases on labour force participation and the effect of risk factors on the prevalence of diabetes and cardiovascular disease for men and women aged over 25 years. Age and obesity stand out as significant factors associated with diabetes for men and women but obesity is a significant factor for cardiovascular disease only in men. Having a parent with diabetes increases the risk of diabetes. Smoking is significant for women in the prevalence of cardiovascular disease while lipid and hypertension treatment are associated with cardiovascular disease for both men and women. Diabetes is associated with cardiovascular disease (at least for women).

# Multivariate system results

The system of multivariate probit equations for males and females was significant on a Wald test (p Æ 0). The correlation between the labour force participation and diabetes equation residuals was moderate and significant for males and for women the correlation is moderate and statistically significant between the two disease equations (see lower part of Table 5). All other correlations were insignificant but a joint test rejects the null of no correlation across equations (p Æ 0). This suggests that that there may be advantages in accuracy in using the multivariate system to estimate the effects of chronic disease prevalence and risk. Calculating the marginal effects from three separate univariate probit regressions will give misleading results.

Table 5 shows the results of the multivariate probit analyses of the probability of labour force participation for males and females over the age of 25. It shows the marginal effect of having diabetes, cardiovascular disease or both on labour force participation. For males the effect of cardiovascular disease is to reduce labour force participation by 0.106 and from diabetes by 0.058. For females the effect is 0.088 and 0.124 respectively. The effect from cardiovascular disease is insignificant for females and from diabetes insignificant for men. It is worth noting that for males the marginal effects of diabetes and cardiovascular disease on participation estimated in the system of equations are considerably lower than the estimates in the univariate regressions. For women cardiovascular disease is insignificant in both estimates, while diabetes has a larger effect in the system of equations than in the univariate regression (-0.124 compared to -0.109). Taking account of the endogeneity of chronic disease in part through the common risk factors does alter the estimate of the treatment effect of chronic disease, but it does so differently for men and women. This is plausible as there are well known differences in men and women in the prevalence of these diseases, their risk factors and their complications in addition to differences in labour supply behaviour.

Chronic disease and labour force participation in Australia: an endogenous multivariate probit analysis of clinical prevalence data

11