S. R. Guimarães Martins, M. C. Iglesias Pérez, J. de Uña Álvarez
In this work we consider logistic regression when both the response and the predictor variables may be missing. We review the existing approaches and perform a comparative simulation. The simulations study was inspired by a real data set, where these methods are also compared. With these data referring to children from schools in the municipality of Viana do Castelo, in the north of Portugal, it was intended to relate the obesity indicator International Obesity Task Force (IOTF) with the provision of physical tests and, in this way, screen the risk of childhood obesity. The prediction of obesity levels through physical examinations can be a very useful tool to eliminate situations of overweight and obesity. Furthermore, in this work the impact of missing data on estimation and testing for significance is discussed and practical recommendations are given. It should be noted that all explanatory variables are significant and the maximum likelihood methodology presents the best results.
Keywords: Missing values, complete case analyses, inverse probability weighting, multiple imputation and maximum likelihood
Scheduled
Applications of Statistics I
June 10, 2022 10:10 AM
A15
