partial effect plots

[William Dupont]

A colleague is looking for an ado file that creates partial effect plots. Does anyone knows of a program that is similar to the one described below? If so, I would be most grateful to learn its name and location. Frank writes

I'd like to renew a call for a fairly general Stata function for getting partial effect plots for the students in our current course. This is probably done in most generality by adding test points to a dataset, with Y=missing, and easily subsetting to the test points for plotting curves and dot charts.

Users could specify which 1-2 variables to vary and what constants to set the other variables to. For any variables for which constant settings are not given, automatically setting to median (for continuous X) and mode or reference level (for categorical X) would be nice.

Ultimately we need to be able to pair predicted values to use subtraction to get effect estimates as a function of interacting factors, with confidence bands. But for now I'd be happy with getting absolute predictions and confidence bands.

The function needs to handle predictors being expanded into polynomials/splines, including the case where spline functions interaction with linear or categorical predictors. Later allowing 2 splines to interact would be a useful addition.

[wbuchanan]

Isn't much of this functionality already provided by the marginsplot command? The users would specify the marginal effects to estimate after fitting their model to the data which allows specifying different values and/or ranges of values over which the marginal effects are to be estimated. I'm not sure how well, if at all, it handles splines, but do know that it is a flexible tool for exploring interaction effects and polynomials (provided they are specified as interactions with the variable itself).