On Model Selection and Model Misspecification in Causal Inference