I. True/False/Uncertain - Briefly explain. No credit without an explanation (5 marks each).

1. GMSE is bigger with more Xs than the true model than with less Xs.

2. If the exit criterion sls in automatic search is elevated, then the model ends up with less Xs in it.

3. If = 0.01 in a logit model, then for each unit increase in X, the probability that Y=1 is on average higher by 1%, ceteris paribus.

4. A model with 98% correct classification rate is a useful model.

5. With the cumulative logit, the probability to be in the second highest category K-1 or higher is pK1 = FK1 1.

6. If the p-value=0.06 of testing Poisson vs. Negative Binomial model with the 2 1 distribution, one should use the Poisson model.

7. With all observations in the same cluster, RSQ = 1.

8. With 500 multiple imputation samples generating 500 predictions of 0 or 1 for the binary dependent variable for each observation, to combine the 500 predictions into one for each observation, one has to take the mean of all 500 predictions and use a cutoff of 50%.