Online Program

Return to main conference page

All Times ET

Program is Subject to Change

Tuesday, June 15
Tue, Jun 15, 11:30 AM - 1:00 PM
TBD
Treatments for Nonresponse: Imputation, Weighting, and Other Models

An evaluation of some Imputation Methods on the Annual Survey of Commerce’s companies’ Revenue (308119)

*João Carlos Silva Rodrigues, Brazilian Institute of Geography and Statistics (IBGE) 

Keywords: Imputation on Economic Surveys, Economic Sample Surveys, Commerce Annual Survey’s revenue

The present work uses information from the Annual Survey of Commerce (PAC), one of the four economic structural surveys of the Brazilian Institute of Geography and Statistics (IBGE), to evaluate its current imputation model against some other available models in the literature. At PAC, there has been an increase in the impact of non-responses on its estimated totals. This is due to a combination of factors: population increase of some economic activities; small number of companies of some strata – leading to high sample weights; elevated number of small businesses closure; and asymmetry of the economic variables. These problems point out the need to study alternatives treatments for these non-responding companies. The work consists of an analysis of more than one thousand companies that had responded the survey selected randomly and assuming that they had not responded the survey. Subsequently, these companies were submitted to the twenty imputation models and the results were evaluated using Mean Square Error and the Percentage Variation between the estimated totals against the real values. For that, it was used the Revenue information of PAC from the years of 2014 and 2015. The models selected can be divided into four groups: average of the respondents; using a regression function with auxiliary variables of cadaster; average of the closest respondents using a distance function; and through a regression function of the closest respondents using a distance function.