interpretation of log-linear model vid interaction term

[Klaudia Lennerling]

Hello!

I'm having a hard time interpreting my results from my regression which is:

Ln(sales) = beta0 + beta1G + beta2Finance + beta3G*Finance

where:
G is 1 if it is a woman and 0 if it is a man
Finance is a categorical variable going from 0-4

What does a unit change in Finance correspond to a change in sales?

best,
Klaudia

[Fei Wang]

One unit change in Finance is associated with 100*(exp(beta2+beta3)-1) percent change in sales for women, and is associated with 100*(exp(beta2)-1) percent change in sales for men.

But if 0-4 are simply categorical identifiers for Finance, I would recommend breaking it down to several binary (0-1) variables rather than directly controlling for its original scale.

[Carlo Lazzaro]

Klaudia:
interested listers wil have even harder times in helping you out if you do not share what you typed and what Stata gave you back (as per FAQ). Thanks.

[Klaudia Lennerling]

Thank you, Fei Wang!

Is the method of grouping this variable not possible? Do you recommend that if we break it down that we incorporated four different interaction regressors for each categorical identifier for finance?

[Fei Wang]

Below is what I would recommend.

reg lnsales c.G##Finance

I think it's acceptable to use the original scale of Finance if 0-4 represent something in order, like "very bad", "bad", "hard to say", "good", and "very good". But if they are simply identifiers without meaningful implications, the solution above is definitely needed.

[Klaudia Lennerling]

Thank you very much!

They represent measures from no obstacle to very severe obstacle. Therefore I assume that my first approach is justified and that I will interpret a change associated to finance as you stated in the first response.
Thank you so much for your help!