Ai and Norton (2003) warn that interpreting continuous interaction coefficients
from a logit or probit is misleading, as an interaction might be significant even if not
indicated by a t-test. Following this warning and the more recent suggestion of
Brambor, Clark, and Golder (2006), multiplicative effects are not interpreted from
the models in Table 1. I instead rely on the Grinter utility developed by Boehmke