I have four competing models which I use to predict a binary outcome variable (say, employment status after graduating, 1 = employed, 0 = not-employed) for n subjects. A natural metric of model performance is hit rate which is the percentage of correct predictions for each one of the models.

It seems to me that I cannot use ANOVA in this setting as the data violates the assumptions underlying ANOVA. Is there an equivalent procedure I could use instead of ANOVA in the above setting to test for the hypothesis that all four models are equally effective?

**Contents**hide

#### Best Answer

Contingency table (chi-square). Also Logistic Regression is your friend – use dummy variables.

### Similar Posts:

- Solved – Interpreting Triple Difference coefficient with dumthe variable
- Solved – Interpreting Triple Difference coefficient with dumthe variable
- Solved – How to interpret a regression model with a variable that appears as an inverse
- Solved – Only binary predictors — ANOVA, regression, or other
- Solved – One-way ANOVA or two-way ANOVA for more than 10 independent variables and 1 dependent variable