12 Logistic regression

Question 1

What happens to the distribution of variance when you fit a linear model to a binary DV?

Answer 1

You get heteroscedasticity. Assuming a strong relationship between the variables, you would get small SD at extremes, and large in middle.

Question 2

What does a risk ratio of .5 for treatment against placebo mean?

Answer 2

That those in the treatment group have 50% less risk than those in placebo.

Question 3

What does a risk ratio of 1 mean?

Answer 3

No effect. No difference in risk between groups.

Question 4

What is the formula for odds?

Answer 4

p(happens)/p(does not happen)

or

p / 1-p

Question 5

What is Exp(B) in a logistic regression output?

Answer 5

The odds ratio. The change in odds for a 1-unit increase in X –or if it’s a categorical variable, the odds of group A (of X) over the odds of group B (of X).

Question 6

What is an odds ratio?

Answer 6

The odds of X happening in one condition over the odds of X happening in another condition.

Question 7

If you exponentiate log odds, what do you get?

Answer 7

Just the odds.

Question 8

In logistic regression output, a negative Beta will give you exp(B) of what?

Answer 8

Less than one.

Question 9

What should you NOT mention when talking about odds?

Answer 9

RISK or PROBABILITY.

Question 10

How can you get risk from log odds? 2 steps

Answer 10

1. Exponentiate the log odds to get the odds.

2. Divide odds by (1+odds) to get risk.