Resumo

SIFUENTES-AMAYA, Rigoberto  e  RAMIREZ-VALVERDE, Gustavo. Effects of specifying an incorrect model for logistic regression, with two independent correlated variables. Agrociencia [online]. 2010, vol.44, n.2, pp.197-207. ISSN 2521-9766.

Analysis of logistic regression is used to study the association between a binary response variable and a set of independent variables. When correlation is high between two independent variables, variances of the parameter estimators are large. However, if the linear model is poorly specified, the variances can decrease when the correlation between the independent variables increases. In this paper, using a simulation study, we evaluated the effect of the correlation between independent variables on the logistic regression model when the model has an incorrect specification. When a relevant variable was omitted, the bias increased and did not disappear even when n was augmented. An effect of the correlation was detected in the power of the hypothesis test on the estimated parameter, and the test size of the hypothesis was close to the nominal size. When an irrelevant variable was included, there was no effect on the bias, the mean square error showed evidence of consistency and power of the hypothesis test diminished when the correlation between the independent variables increased.

Palavras-chave : correlation; specification; variable omission and inclusion; logistic regression.

        · resumo em Espanhol     · texto em Espanhol     · Espanhol ( pdf )