Using "Recycled Predictions" for Computing Marginal Effects
Source: University of California
A conventional way to estimate the marginal effect of a risk factor is to omit the risk factor (e.g., gender, treatment) from the multivariate model and then use the omitted risk factor as class/group variable to compare observed and predicted outcomes. This can have an undesirable impact on the model since the predicted outcomes could be incorrect if the omitted variable is a significant risk factor. An alternative approach is to use the recycled prediction method to estimate and compare marginal effects without removing the risk factor from the model. While STATA (StataCorp, 2005) and SUDAAN (Research Triangle Institute, 2004) have provided sub-routines for the recycled prediction method, SAS does not have a dedicated procedure for this purpose.