1
$\begingroup$

I'm working with a truncated poisson (also known as positive poisson) and I want to interpret the coefficient interaction between two variables.

From this question, we see that the interpretation of coefficients is not direct, as it is with the typical Poisson. In the link we only see how a beta of a variable is interpreted, but not of any interaction.

Although with a little algebra I came to the conclusion that betas of different variables can be compared with each other to say who has the greatest effect on the dependent variable (despite the transformation that the beta goes through), it is not so direct the analysis of an interaction.

How do you usually interpret the term of interaction between two variables? It is the same as a normal value? Do you verificate cases? And if so, how you do it?

Thnks!

Improve this question
$\endgroup$
2
  • $\begingroup$ What does "verificate" mean? $\endgroup$
    – Peter Flom
    Commented Jan 18 at 13:23
  • $\begingroup$ @peter sorry, I meant to test cases, like var1 = 0 and var2 = 0, then var1= 0 and var2=1, etc. in the case both variables are binary. With this, I will check how the present effects change, but I had my doubts because, as the question I quote says, its analysis is not direct, since the expected value is not equal to lambda. $\endgroup$
    – ISIDORA
    Commented Jan 18 at 14:52

1 Answer 1

Reset to default
1
$\begingroup$

What I like to do with interactions in any kind of regression model is look at predicted outcomes at different levels of the independent variables that are involved in the interaction.

How exactly I do this depend on whether the IVs are categorical or continuous. I would always include some graphs. Then, for two categorical variables, I'd look at every combination and make both tables and graphs. For two continuous IVs I'd look at various quantiles of each (maybe three quartiles). For one of each, a combination.

For higher level interactions, I'd adapt the above.

For a count model, I might look not just at the predicted mean value, but at the probability of different values. This would depend in part on the range of the count.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.