Estimates based on household surveys often differ considerably from administrative records. In the National Household Sample Surveys (PNADs) of the Brazilian Statistics Department, the estimated number of beneficiaries of the Family Allowance (Bolsa Família) Program (PBF), and of beneficiaries of the Continuous Cash Benefit (Benefício de Prestação Continuada) Program (BPC) is always lower than official figures. This paper presents a simple methodology, based on the sampling design of household surveys, to explain these differences, by decomposing them into three terms: the representativeness bias (derived from the choice of places, census tracts or municipalities for the survey); the data collection bias (derived from data collection problems at the chosen sites); and the interaction between them. The application of this methodology to the present cases shows that the representativeness bias of the PBF accounts for 40% of the difference between official records and the PNAD. For the BPC, the representativeness bias tends to act in the opposite direction. That is, In other words, the data collection bias is entirely responsible for the observed difference. Also, the erroneous reporting of the BPC as a Social Security benefit on the PNAD seems to have occurred mostly in the years prior to 2004, and does not entirely explain the low number of beneficiaries identified in the PNADs.
Public policies; Sample-based surveys; Administrative data
