Marginal effects plots crossing zero, negative binomial regression

[Christina Faegri]

Hi,

I was recently attacked for presenting "insignificant" results based on marginal effects plots. The observer is correct in observing that the plot crosses zero, but is this a problem for the model? We used a negative binomial regression with fixed effects.

I would really appreciate your feedback!

Thank you very much in advance.

All the best,
Christina

[Ariel Karlinsky]

I'm not really sure I understand the question.
If the CI crosses 0 it indeed means that the result is not statistically significant.
It does not mean that there are problems with the model! those are 2 completely different issues.

[Phil Bromiley]

As always, you're more likely to get a useful answer if you follow the faq on asking questions - provide Stata code in code delimiters, Stata output, and sample data using dataex.

I can't tell what your marginal effects plots plotted and what your model included. I assume you're doing margins with the dydx option and have a model that includes interactions (I hope specified with factor notation). Suppose you include an interaction of x1 and x2 in the model. It is quite possible for the influence of x1 to have values of x2 where it is statistically significant and values of x2 where it is not statistically significant. It is also possible to have values of x2 where x1 has a positive statistically significant influence and values of x2 where x1's influence is negative and statistically significant (and some in the middle where it is not statistically significant). Generally, you should explicitly report where the influence is statistically significant and where it is not statistically significant.

Alternatively, if you are doing predictive margins, then whether a prediction differs significantly from zero is not relevant to whether the variable is significant.