When modeling decisions
are based on the data under study, bias is introduced. The correct
answer is: (c) Be biased to more extreme values. When modeling
decisions are based on the data under study, we usually pick
up more extreme patterns than present in the underlying population.
Hence, p-values are estimated too small, and regression coefficients
too large (in absolute value). Regression to the mean may be
expected when the model is validated in new patients.