International business (IB) theory often includes conditional hypotheses. A conditional theory reflects scholars’ recognition of the need to include a moderating variable in a proposed causeand-effect relationship. The output of such theorizing takes the form of a hypothesis in which the relationship between a dependent variable and a primary explanatory variable of interest varies across the level or existence of some other moderating variable. To test a conditional hypothesis, researchers typically specify a regression model that includes a multiplicative interaction term. However, despite the growing number of articles containing such terms, IB researchers rarely distinguish – either conceptually or statistically – between two very different questions in their analysis of moderated relationships (Aiken & West, 1991).

Commonly, researchers ask only the following question: is the estimated coefficient on the interaction term in the regression statistically significant? If yes, then they generally conclude that support exists for the conditional hypothesis. However, IB researchers seldom explore a second, equally important question identified in the literature as contributing to a more complete test of the conditional hypothesis (e.g. Brambor, Clarke, & Golder, 2006; Berry, Golder, & Milton, 2012; Spiller, Fitzsimmons, Lynch, & McClelland, 2013). This question asks: is the effect of a change in the primary explanatory variable on the dependent variable (or, more simply, the “marginal effect” or “regression slope”), for any specific value of the moderating variable, statistically different from zero? The answer to the latter question provides vitally important additional information about the support for a conditional hypothesis. Whereas the first question asks whether marginal effects differ from one another for any two values of a moderating variable, the second question asks whether a marginal effect differs from zero for any specific value of a moderating variable (Aiken & West, 1991).

Differentiating between the two questions is critical. As we demonstrate in this paper, it is entirely possible to find, simultaneously, that the estimated coefficient on an interaction term is statistically insignificant and that the effect of a change in the primary explanatory variable (i.e., the marginal effect) is statistically different from zero over some portion of the range of the moderating variable. Itis also possible for the researcher to find a statistically significant estimated interaction coefficient but that the effects of a change in the explanatory variable are significantlydifferent from zero for only some value(s) of the moderating variable.