Who Does Not Respond in the Household Expenditure Survey : An Exercise in Extended Gini Regressions

The aim of this paper is to suggest a new, nonparametric multiple regression method, based on the Extended Gini (EG) measures of dispersion, which enables the user to follow the curvature of the regression curve. The method is capable of estimating a series of linear approximations of the regression curve, allowing the investigator to stress different sections along the range of one independent variable while keeping the treatment of other independent variables intact. The method is based on the extended Gini family which depends on one parameter, ν. The choice of this parameter enables the user to produce infinite alternative estimators of the regression curve. The difference between them lies in the weighting schemes applied to the slopes of the regression curve. By investigating the patterns of changes in those regression coefficients, the curvature of the regression curve can be traced. As an application, we investigate nonresponse patterns in the survey of household expenditures in Israel. We are mainly interested in whether nonresponse increases or decreases with income, and the kind of functional relationship one can find between income and nonresponse. The empirical illustration shows that the higher the income, the larger the response rate, and the larger the household, the higher the response rate. Also, the Arab population tends to respond more than the Jewish one, while the ultra religious group tends to respond less than the rest of the population. Those last two results hold with and without adjustment for income and household size. The implications on the bias in the estimates are discussed