Other independent variables

Abstract

Other than physical activity, there are other factors influencing demand for hospital services. For instance, individuals' health sta-tus, social and economic environments, and other health behav-iours affect the demand. There are systematic health inequalities among individuals from different socio-economic backgrounds, which cannot be solely explained by the differences in individual health behaviours. People with low income or education suffer more disease and premature mortality and their low socio-economic background also affects their health behaviour. 21,22 To control for socio-economic differences and variations in health sta-tus, economic and social factors, household composition and size, smoke-free environment, work-related factors, mental and general health status, and number of chronic conditions are included in the analysis. Individuals' behaviour related to LTPAs and health care use can also be affected by other health behaviours. To account for them, alcohol and tobacco consumption and diet are included. Active individuals may utilize more health care services due to sports-related injuries and they may stay inactive during the injuries. 23 To control for these factors, dummy variables for injuries related to exercise are included. In this study, physical activity is measured using LTPAs. How-ever, individuals can be active during work or other daily activities. To account for these factors, dummy variables for walking or bik-ing to work or while doing errands, and physical effort required at work or daily activities are also included. Tables 1 and 2 present a complete list of all variables. Count data regressions Following the literature, I used count data models to estimate the demand for hospital services, measured by the number of hospital stays. 24,25 A number of different count data models are developed to deal with potential problems that exist in count data sets. For instance, non-zero counts are typically observed for a small share of the population that increases skewness in the data, and the variance of the dependent variable can be higher than the mean (overdisper-sion problem). As opposed to other models, a zero inflated negative binomial (ZINB) model provides a solution to these problems. 26