Comparação de dados de aquisição domiciliar e consumo alimentar individual: Pesquisa de Orçamentos Familiares, 2008-2009.

Abstract

Household Budget Surveys (POF) are periodically conducted in Brazil to provide information on living conditions and household s budget composition. The 2008-2009 POF was the first and only survey to record representative information of individual food consumption in Brazil. Until then, the knowledge of food consumption s trends in Brazilian population and subpopulations was inferred from food acquisition data. There might be differences between these two methods: several items are predominantly consumed outside the house environment; others can be used as ingredients in preparations or wasted; besides that, the sampled population and food items differ from both methods. Thus, the objectives of this study are to compare the means of food from acquisition and consumption information, and to propose predictive equations for the average food consumption from acquisition food data. The data used in this study came from 55.970 households sampled at POF 2008-2009. The household members registered food acquisition information in the Collective Acquisition Booklet throughout 7 days period; and only 10 years old or higher subjects registered food consumption in two non-consecutive days by the food record method. Food items was grouped, as possible, into specific food groups; The averages of food groups, for each geographic stratum, were calculated by dividing the total food amount reported per the total population number sampled in the stratum. Due to zero inflated and asymmetric distribution of food variables, a two-part model was used to predict food consumption. The first part models the consumption probability by a logistic model, and the second part models the amount of food consumed by a gamma regression. Household acquisition overestimates cereal consumption (258g and 289g, acquisition and consumption respectively), dairy products (137g and 62g) and total sweets (66g and 34g); While underestimates mainly for vegetables (66g and 187g), FLV (118g and 152g), pasta (27g and 45g) and meat (82g and 155g). When the comparison considered the percentage of caloric contribution, the differences became less important. The equations were able to predict the consumption mean for most items with good model fit quality.