1. I have multiple models representing distinct hypotheses; which is best supported by the data? |
Different models represent different hypotheses. |
2. I have a set of predictors that are all hypothesized to be important for the response. Which are supported by the data? What is their relative importance? Should all be included in a single model, or should a smaller model of ‘significant’ predictors be used? How should such a smaller model be chosen? |
Different predictors represent different hypotheses (but a particular model could include one or more predictors) |
3. I have a large number of predictors that may or may not be important, and I want to do an exploratory analysis to see which are best supported by the
data. How do I construct model(s) to choose among them and quantify their importance? |
Not a hypothesis! Exploratory analysis / data mining / data dredging (not comparing a priori hypotheses, sifting for relationships). |