Comparing two incident rate ratios with post estimation

[Bob Sanders]

The output below is from the Poisson example at https://stats.oarc.ucla.edu/stata/da...on-regression/.
The incident rate for program 2 is 2.96 times the incident rate for program 1 (the reference group) (p < .01).

If I wanted to compare program 2 vs. program 3, would this be the recommended approach:

lincom 2.program+3.program, irr

This will produce an IRR, SE, and p value, etc. Is the IRR listed showing the IRR for the first value in the lincom command (in this case, program 2) compared to the second value (program 3)?

Any suggestions? Thank you!

[Clyde Schechter]

Not quite. It would be

lincom 2.program - 3.program, irr

[Bob Sanders]

Ahh ... thank you! Would be nice if the STATA manual showed an example of using the minus instead of plus.

As for interpreting the one IRR the output shows: Is this the IRR of the first level in the equation compared to the second level? For example, if that code showed an IRR of 1.663 (p = 0.000), would that mean the rate for program 2 is 1.7 times higher than the rate for program 3, controlling for other covariates in the model?

[Clyde Schechter]

As for interpreting the one IRR the output shows: Is this the IRR of the first level in the equation compared to the second level? For example, if that code showed an IRR of 1.663 (p = 0.000), would that mean the rate for program 2 is 1.7 times higher than the rate for program 3, controlling for other covariates in the model?

I'm a little unclear just what you are referring to in this question, the output shown in #1, or hypothetical output from the code proposed in #2.

Anyway, the rate ratios in the output for #1 are all relative to the omitted reference category of prog (which, I guess, is 1). So they are rate conditional on 2:rate conditional on 1 and rate conditional on 3:rate conditional on 1. If you ran the -lincom- command proposed in #2 and got IRR = 1.663, that would mean that the ratio of the awards rate in program 2 is 1.7 times as high as (not higher than)the awards rate in program 3.

[Bob Sanders]

Thank you. I was asking about the hypothetical output from the code proposed in #2, and your response makes sense.