Model Selection in Complex Classes of Models

• Towards general principles for model selection. • ‘Give up your small ambitions’ • Check your assumptions – – of the models – of the model selection theories Model Choice in 1977 That’s both 25 years ago and when I started to learn about this. • The set of models one could consider was severely limited by computational constraints, although packages such as GLIM 3.77 were becoming available. • Stepwise selection was the main formal tool, using hypothesis tests between a pair of nested models, e.g. F tests for regressions. No one did enough tests to worry much about multiple comparisons issues. • Residual plots were used, but they were crude plots and limited t