Question

1)     You are reviewing a ridge regression model that your business partner is building. In the model, he has applied the shrinkage penalty to all terms (intercept included), leading to a massive reduction in variance. Is everything okay with this implementation? (Choose the MOST CORRECT answer.)

a) Yes, the ridge regression model is implemented properly.

b) No, there should be a massive reduction in bias (not variance).

c) No, the shrinkage penalty term does not apply to the intercept.

d) No, variance should only reduce slightly.

e) None of the above

---------------------------------------------------------------------------------------------------------------------

---------------------------------------------------------------------------------------------------------------------

2) Which of the following statements about Lasso and ridge regression is correct over all datasets? (select all applicable)

a) Because Lasso regression produces simpler models, it is better than ridge regression.

b) Because ridge regression coerces insignificant data downwards, it is better than Lasso regression.

c) Because the shrinkage penalty in ridge regression leads to a slight increase in bias, lasso regression is better.

d) Because lasso regression's shrinkage penalty has to be tuned separately, ridge regression is better.

e) None of the above.

---------------------------------------------------------------------------------------------------------------------

---------------------------------------------------------------------------------------------------------------------

3) How many dummy variables ought to be set in a model if the numerical variable in question has 10 values?

a) 8

b) 9

c) 10

d) None of the above

e) Cannot tell without further analysis

---------------------------------------------------------------------------------------------------------------------

---------------------------------------------------------------------------------------------------------------------

4) If a variable has X number of values, why is the number of dummy variables set to X-1? (Select the most precise/accurate answer.)

a) To prevent perfect multicollinearity

b) To prevent confusion in the model parameters

c) To decrease variance

d) To avoid heteroscedasticity

e) To decrease the number of variables used in the model

Answer 1

1. b) No, there should be a massive reduction in bias (not variance).

2. d) Because lasso regression's shrinkage penalty has to be tuned separately, ridge regression is better.

3. c) 10

4. c) To decrease variance.