133230DHS618 - Logistic Regression Example

Module 3 Logistic Regression Example QUESTION Is the assumption of collinearity met? To test this assumption, you need to run this in linear regression (as this is not included under logistic regression). Select Analyze  Regression  Linear. Then select Statistics and check collinearity diagnostics. When you run linear regression, you should obtain the results below. Disregard all results except for collinearity statistics. Since the tolerance values are above 2 and the VIF is below 4, the assumption of collinearity is met.

Perform logistic regression for all independent variables Select analyze  regression  binary logistic to obtain screen below. Move dependent variable into box (i.e. pass). Move independent variables into box (i.e. score, exp). As the independent variables are not categorical, you do not need to select the categorical box. In order to obtain goodness-of-fit test results and 95% CI for odds ratios, select options and check the following: Hosmer-Lemeshow goodness-of-fit, Classification plots and CI for exp(B). Select continue.

4. The classification table describes the predictive accuracy of the model. Overall, the predictive accuracy of the model was 73.1% The model accurately predicted 60% of those who passed and 81.3% of those who did not pass. Perform logistic regression for one independent variable at a time Follow the same procedure above for performing logistic regression, except enter only one independent variable. When exp is entered as the only IV, your SPSS output should include the following: 1. When only one IV is included, the odds ratio in the table below is referred to as crude OR. In bivariate analysis, experience was statistically significant (OR=1.155, 95% CI =1.002- 1.331).

Variables in the Equation B S.E. Wald df Sig. Exp(B) 95% C.I.for EXP(B) Lower Upper Step 1 a exp .144 .072 3.958 1 .047 1.155 1.002 1.331 Constant -2.340 1.062 4.854 1 .028 .096 a. Variable(s) entered on step 1: exp. 2. In bivariate analysis, score was not statistically significant (OR=1.263, 95% CI =.744- 2.143 ). Variables in the Equation B S.E. Wald df Sig. Exp(B) 95% C.I.for EXP(B) Lower Upper Step 1 a score .234 .270 .750 1 .386 1.263 .744 2.143 Constant -1.132 .881 1.649 1 .199 .323 a. Variable(s) entered on step 1: score. Univariate Versus Multivariate When comparing crude to adjusted OR, there was no meaningful difference. Experience remained statistically significant in bivariate and multivariate analysis. Score remained non- significant. This suggested that there was no interaction or confounding among these variables. Variable Crude OR 95% CI Adjusted OR 95% CI Experience Score 1.155 1.263 1.002 – 1.331 .744 - 2.143 1.158 1.333 1.004 – 1.336 .739 – 2.404