For a data analysis course, I am looking for multivariable (multiple-predictors) real datasets (no MC simulations), for which regression models are a reasonable description and on which model selection methods (all subsets AIC or cross-validated R^2, stepwise ANOVA selection) can be compared.
Ideally, the datasets should have the following properties:
- No (or omittable) categorial predictors, because I would like to discuss the encoding schemes of nominal features and approaches for dealing with them at some later point.
- Not too many predictors (less than eleven), so that all-subset model selection is feasible.
- No missing values, because I would like to discuss imputation methods at some later point.
Ideally, I am looking for at least two datasets each for ordinary regression and logistic regression.