Use the Q7 dataframe. You want a ‘good’ model predicting Wins from all variables (without
interactions). Perform the ‘all possible’ approach, which produces a list of models with
whose AICs are within 4 of the overall lowest AIC of all considered models.
Identify the predictors in the model with the FEWEST predictors (it
could be the case your list has only one model). Redo
the ‘all possible’ approach considering just those predictors ALONG WITH all two-way
interactions between them to produce yet another list of models (your list might only have one).
Report the AIC of the model at the top of this list, i.e., the one with the lowest AIC.
Note: do not add ANY extra arguments (e.g., nbest or nvmax) to any of the commands.