a quote from the economics Nobel laureate Tjalling C. Koopmans from 1949 (published in Econometrica).  He gives a list of topics for future research on statistical inference.  Number 3 on his list:

"[I]f the choice of the model is influenced by the same data from which the structural parameters are estimated, the estimated sampling variances of these estimated parameters do not have direct relation to the reliability of the estimated parameters which they would have if the estimation were based on a model of which the validity is given a priori with certainty."