Using the Logistic Regression in Analysis of Results from Statistical Observations

Abstract

The article is focused on investigating the problems arising in the statistical analysis with use of logistic models of the ordered multiple choice, which are constructed by the results of statistical observations involving the existence of a categorical dependent variable. This group of models should be used when a discrete dependent variable takes several alternative values. The examples include assessment of student performance (perfect, good, satisfactory, unsatisfactory). These models’ parameters are estimated using the algorithms based on elements of the probability theory. The purpose of constructing the multiple choice model is to determine the factors with impact on the probability of the occurrence of a particular event and the choice of an alternative, as well as the strength of this impact. A detailed description of the algorithms for estimating logit models of binary and multiple choice is given, with demonstrating the model application in solving a particular problem (statistical analysis of the results of self-assessment of health status by household members) by use of SPSS package. It should be noted that statistical packages like SPSS, STATISTICA or STATA contain the modules for constructing logit and probit models.

The assessment of population’s health status includes the objective assessment of their health status by the official statistics data on the prevalence of deceases and the cumulative subjective assessment of the individual health status by the results of sociological studies  It is important to know to what extent the objective assessment of the population’s health status complies with the subjective perception of the health status by individuals. Because the primary files of the sample survey of households are confidential, the multiple choice model was constructed by the author using the proxy data with characteristics close to actual values. Variables such as residence place, gender, age, assessment of health status, sports practicing, smoking and income were reported in the process of the sample survey. In constructing the model, the variable “health” was used as a dependent variable; “gender” and “education” were used as categorical variables; “age” and “income” were used as covariates. Once the model was constructed and its identification capacity (i. e. the correctness of the predicted dependent variable) estimated, its specification was saved in a special file for the subsequent rebuilding.