Interpretation of a positiv Regression calculation with a negativ indipendent variable and a negativ moderator

[Finley Voigt]

Hello there,

as you can see down below i have a negative coef. for CEOnarcism (indipendet var) and a negative coef. for CEOgender (moderator).

My hypotheses is that: The negative effect of CEOnarcism on digital innovation power is less for female CEOs that for Male.

Is that correct?

[Clyde Schechter]

First, it is impossible for anybody but you to interpret these results because you have not disclosed whether the 1 category for gender represents males or females.

Second, this is one of those situations where common English usage leads inevitably to confusion and misunderstanding. Here is what you have: In the zero category of gender, the effect of narcissism on fcitations_w is negative. That is to say, a higher narcissism score is associated with a lower value of xb, lower by 0.149 (to 3 decimal places) per unit of narcissism score. If we look at those in the one category of gender, the effect of narcissism is now given by the coefficient of narcissism (-0.149) plus the interaction coefficient (+0.179). That adds up to +0.03. So in the one category of gender, xb increases by 0.03 per unit of narcissism score.

It is difficult to put that into simple English without creating ambiguity and confusion. The direction of the effect actually changes from negative to positive and there are no simple English words for that. I think that in presenting these results it is best to simply show the marginal effects themselves (including their signs). By the way, I recommend that you calculate the actual marginal effects on fcitations_w as rate ratios. The -margins GenderCEO_lag, dydx(CEOnarcissism_lag)- command will give you those results, along with standard errors and confidence intervals. It is better to present results in the metric of fcitations_w itself, rather than in the xb metric. In the fcitations_w metric, the 0-gender rate ratio will be less than 1, whereas in the 1-gender group it will be greater than 1, reflecting the reversal of direction of the effect.

[Finley Voigt]

Thanks a lot. This helped so much :D Thank you