Estimation of Households Income from Bracketed Income Survey Data

As far as data on personal income are highly confidential and sensible, it is a common practice to collect such information by asking people to classify their own earnings along a discrete scale of income “brackets”. This procedure provides an unbiased estimation of average income, under fairly general conditions, but it is well known that standard error of estimates increases with brackets size. On the other hand, people tend to underreport income, and this bias is likely to increase as brackets width gets smaller. Thus, an optimal bracket size can be generally identified, that insures a reduction of underreporting without increasing estimate variance too much. The paper presents an evaluation of brackets size effect on various procedures for estimating Italian households’ income. The first result is that the most reliable and robust procedures are those based on the extrapolation of income distribution in the upper open class by means of very simple functions. Secondly, reducing of the number of income brackets from the actual 22 to 5-7 seems to improve the accuracy of indicators for every procedure.