Resumo

COA CLEMENTE, Ramiro. OVERESTIMATION OF THE INFANT MORTALITY RATEOBTAINED WITH THE INDIRECT METHOD. A MODEL FORITS CORRECTION. Revista Varianza [online]. 2023, n.22, pp. 1-16. ISSN 9876-6789.

ABSTRACT This research has two objectives: on the one hand, to check whether the indirect methodfor estimating infant mortality rates in Bolivia systematically leads to an overestimation ofthe indicator and, on the other hand, to correct this overestimation ifsuch aproblem isfound to exist. Assuming that the infant mortality rates obtained with the direct method are correct, it is confirmed that the indirect method systematically leads to overestimated rates. In addition, there is statistical evidence that greater overestimation occurs with any ofthe Coale and Demeny mortality modelfamilies usedfor its calculation. To correct the problem of overestimation in the infant mortality rate, the lasso regression model was selected, since the infant mortality rates predicted with this model were closer to the correct rates than the rates predicted with other models. It is noteworthy that the lasso model selected includes a single predictor variable, whose calculation is very simple and does not require the use of mortality models: the proportion ofdead children in women aged 25-29 years. Since countries such as Bolivia still require the indirect method mainly to estimate infant mortality rates for small areas, such as most municipalities, it is necessary to continue exploring statistical models that allow us to obtain infant mortality rates that are not overestimated.

Palavras-chave : Direct method; Machine learning; Mortality model tables; Lasso regression.