There are several variations of stepwise regression. The version I just described is called forward selection since it starts from scratch and starts including variables. The alternative, backward elimination, starts by including all 1,600 variables and excludes those that are statistically insignificant, one at a time.
