Using data from the 2017-2018 Household Budget Survey, the work investigates the demographic, geographic and socioeconomic determinants of Brazilian families' expenditure on acquiring health insurance plans for their members. A distinctive characteristic of the data is that most Brazilian families do not incur expenditure on health plans. Furthermore, the size of the expenditure, even when strictly positive, does not depend solely on factors directly determining the risk class of individuals in terms of its health conditions. The empirical evidence, therefore, goes against the theoretical prediction according to which, in a context of competitive market and complete information, every individual acquires an actuarially fair health plan that offers full coverage against financial shocks arising from health problems, so that the premium paid for this plan is higher the higher its risk class (in terms of the probability of health problems). Different behavioral and institutional hypotheses could explain the conflicting behavior of the data: adverse selection resulting from individuals having superior information about their risk class when compared to operators, the existence of a universal and free public health service alternative in the area of residence of the family, the heterogeneity of individuals in relation to schooling, degree of aversion to risk and insufficiency of family income. The above hypotheses open up channels through which a wide range of potential explanatory variables could influence family spending on health plan. The effect of these variables is estimated and tested in the work, which encompass personal characteristics of the head of the family, such as age, gender, color, education, occupation and marital status, and characteristics of the family and its residence, such as per capita income, area of residence, morbidity conditions, evaluation of health services and composition by age, gender and family relationship. To this end, the empirical strategy of the work is based on the estimation of a sample selection model that allows separating the determinants of the decisions regarding the acquisition of a health plan (i.e. whether or not to incur strictly positive expenditure) and the size of the expenditure, in addition to correcting the inconsistency arising from a possible correlation of the unobserved factors of those two decisions.
