Analyzing the relationships between organ failure and recuperating ability of patients in intensive care units

Abstract

A research on logit, probit and linear probability models have been carried out. The dependent variable being used is binary while the independent variables used consisted of eight variables which were binary, categorical and continuous. Research done based on these models covered its discoveries, functions of models, assumption of residuals, probability distributions and estimation techniques. A case study using the three methods of modeling as mentioned above was carried out. The influential variables were selected using forward stepwise method, backward stepwise method and manual selection. Selected variables using manual selection were based on the principle of parsimonious model. For each model, the results showed two out of eight chosen variables were significant at a significance level of 0.05. Tests on residual, multicollinearity, hypothesis testing, goodness of fit and influences and outliers data diagnostics were applied to these three models. Interpretations on estimated parameters were also carried out. After comparing the three models, it was found that logit model appeared to be the best model having the highest probability of accuracy. Softwares like ‘SPSS for Windows Version 10’, 'Minitab Release 13.2’ and ‘Microsoft Excel 2000’ were used to analyze these data.