Interpretation of coefficients with interaction term with odds ratios (logit regression)

[Natan Sklair]

Hi, I want to clarify the interpretation the effect of a variable when an interaction term is included, using odds ratios.

Looking at the effect of debt on depression (all binary), I have:

depressed(time t)= B1depressed(t-1) + B2debt(t-1) + B3debt(t-1)xdepressed(t-1) +Xi

What is the interpretation of these when using odds ratios (assuming B1 to B3 are already exponentiated)?


Current odds ratios are:

Depressed(t-1)	0.236***
	(0.0283)
Hasdebt(t-1)	1.346***
	(0.138)
Depressed x debt (t-1)	0.339***

Would I multiply them to get the final odds ratio?

Any guidance would be greatly appreciated.

[Clyde Schechter]

Would I multiply them to get the final odds ratio?

In principle, yes. That will not, however, get you standard errors (or confidence intervals or tests). So let Stata do the hard work: use -lincom, or-. If you are not familiar with how -lincom- works, read -help lincom-.

[Natan Sklair]

In principle, yes. That will not, however, get you standard errors (or confidence intervals or tests). So let Stata do the hard work: use -lincom, or-. If you are not familiar with how -lincom- works, read -help lincom-.

Clyde Schechter Thanks Clyde. It works great. Just clarifying. Does this mean removing my interaction term in my original regression and using lincom only?

[Clyde Schechter]

No, no. Use the original regression, including the interaction term. Then use -lincom-.

[Phil Bromiley]

You may also find it helpful to use margins after the estimation.