CK020 - Logistic Regression Flashcards
1
Q
Why can’t you use linear regression for a binary outcome?
A
- Predictions may be out of range
- Variance is not constant
- Errors do not have a normal distribution
- Linear regression is not reasonable, because we want a S-shaped curve
2
Q
What is the key to a logistic regression model?
A
A ‘Logit link’
3
Q
How can you transform the odds from a logistic regression back to probability?
A
By using the ‘Inverse logit function’
4
Q
What are the assumptions of a logistic regression model?
A
- Outcome is a binary variable
- Observations are independent
- Log(odds) is a linear function of the covariates
- Large enough number of observations
5
Q
What is the interpretation of a coefficient of a dichotomous covariate?
A
The difference in log odds between X=1 & X=0, where the Odds Ratio = exp(b1)
6
Q
What is the interpretation of a coefficient of a continuous covariate?
A
The increase in log odds for a unit increase in X, where the Odds Ratio = exp(b1)
