Logistic Regression

Question 1

Sample Size

Answer 1

1. at least n=10 per each parameter for each group
2. maximum likelihood estmimation requires larges sample size
3. (number of ind. variables x 10)/probability of least frequent outcome (0)
4. info is in 1s and not in 0s

Question 1

Objectives of Logistic Regression

Answer 1

1. Explanation - identifying ind. variables that impact group membership in dependent variable
2. Classification - classifying observations into correct groups

Question 2

Assumptions

Answer 2

1. independence of observations
2. linearity of logit - Box-Tidwell for each continuous ind. variable- if it’s significant - it’s non linear
3. little or no multicollinearity

Question 3

Predictive accuracy of actual outcome

Answer 3

1. Sensitivity - true positive rate - tp/tp+fn
2. Specificity - true negative rate - tn/tn+fp

ROC & AUC

0.9 excellent

above 0.5 - good

Question 4

odds vs logit values in confidence intervals

Answer 4

logit - shouldn’t contain 0
odds - shouldn’t contain 1