Sampling weights in the 2003 Brazilian Oral Health Survey

The 2003 Brazilian Oral Health Survey (BOHS) is the most comprehensive national oral health research yet performed in Brazil. Methods of probabilistic sampling was used, but sampling weights, essential to assure that data represent the selected age groups of the Brazilian population, were not estimated. Objective: The aim of this paper was to perform analyses using preliminary sampling weights and compares these results with the initial results achieved without the use of any sampling weights. Methods: Sampling weights correspond to the inverse of the probabilities of inclusion of the elementary units in a sample. This work accounts for cities, the first unit of selection; the other unitsx information is not yet available in the data base. Specially, the probability of inclusion of states' capitals was one. Results: Variation of sampling weights were from 17.8 (Center-West; population: 50-100,000 inhabitants, age: 12 years old) to 16,137.5 (Southeastern, >100,000 inhabitants, age: 34-44 years old). Despite the ample variation observed in each age group, estimates of variables with non-different distribution in the selection group did not present bias when the sampling weights were ignored. On the other hand, estimates of variables with different distributions in the selection group, such as `type of service used in the last consultationx showed some alterations. In the elders' group, for example, the analyses of this variable suggest the same ratio of 40.5% in both the public and private liberal services, when the sampling weights are ignored. However, when the sampling weights are incorporated, a difference of 45.6% was shown in favor of liberal private services over public services.

Conclusions: The preliminary results indicate differences between the sample estimates when the sampling weights are used and are not. This reveals potential problems in the disseminated results achieved without the sampling weights. Therefore, I argue that it is necessary to determinate and utilize sampling weights in this survey in order to produce results without bias.