22.4 Many Predictors

• Often we want to include a large number of predictors, e.g. to make ignorability more plausible.
• Challenge is to avoid overfitting leading to increased variance and uncertainty in the estimations. If you cannot reduce the parameters some options include:
  • Combine (related) predictors in a structured way
  • Regularization, e.g. informative priors, horseshoe (breifly mentioned in Chapter 11) and lasso.