Abstract
Regression specifications in applied econometrics frequently employ regressors, which are defined as the product of two other regressors to form an interaction. Unfortunately, the interpretation of the results of these models is not as straight forward as in the linear case. In this article, we present a method for drawing inferences for interaction models by defining the partial influence (PI) function. We present an example that demonstrates how one may draw new inferences by constructing the confidence intervals for the PI functions based on the traditional published findings for regressions with interaction terms.
Acknowledgements
The research carried out in this article was partially funded by a grant from the Faculty of Economics and Commerce at The University of Melbourne. We wish to thank David Moreton for helpful comments.
Notes
1 Note that a similar analysis can be performed on .
2 When we use the case of the continuous regressor w, it can be shown that these results are the same for the discrete case.