In regression models, an interaction term captures the joint effect of two variables that is not captured in the modeling of the two terms individually. If the inclusion of an interaction term is warranted, the fit of the regression (e.g. as measured by R-squared) will improve when it is included.
