Identificação de subdeclarações de renda no cadastro único para programas sociais utilizando regressão logística

Abstract

Income transfer programs are very important to guarantee the basic human rights of housing and food. They are government initiatives through the Ministry of Development and Social Assistance, Family and Fight against Hunger, with the Bolsa Família Program being the most important and covering the largest number of families. One of the requirements that a family must meet to become a beneficiary of the program is not to have a per capita family income higher than the level established by the Federal Government. To verify the eligibility of families regarding this income criterion, the Federal Government uses the per capita family income, calculated based on the individual income of each family member, reported in the Single Registry. One problem with this approach is that, in order to benefit from the program, a person may declare having a lower individual income than they actually have or even omit their individual income, reducing their per capita family income. One of the mechanisms adopted by the Federal Government to try to identify this type of fraud is to cross-reference CadÚnico data with other databases, such as RAIS and CAGED. In addition to this mechanism, the use of mathematical/statistical models to identify families with under-declared income in CadÚnico has been proposed. This paper proposes a procedure, based on logistic regression, to identify families suspected of having fraudulently declared income below the level established by the Federal Government. Families are considered suspicious when they declare having an income below the level established by the Federal Government, but according to the adjusted logistic regression model, they have a “small” probability of meeting this requirement. The procedure was applied to CadÚnico data for the year 2018 and identified 33 suspicious families. Of these, 14 were beneficiaries of the program.