Statistics for Business: Decision Making and Analysis, 3rd Edition

Related Algorithms

Stepwise regression may be run in various ways. We recommend forward stepwise regression that incrementally adds variables. The alternative version called backward stepwise regression starts with a model having many explanatory variables (usually the saturated regression) and at each step removes the explanatory variable that adds the least to R².We prefer a forward search because it avoids conflicts caused by redundant variables (Manager and Plant in this example) and does not require the saturated regression. In some applications, we have more explanatory variables than observations. In that situation, we do not have enough cases to fit the saturated model. A hybrid search combines these two. Mixed stepwise regression runs ...