nrbreg coefficient interpretation

[Ludovic Van Cau]

Hi

I'm a bit confused about how to interpret coefficients in a nbreg as I don't often use it and I'm quite new to the econometrics and Stata world.

I have my count depvar teamsize and the indepvar internet which is a dummy variable.

These are my results:

-----------------------------------------------------------------------------------
                  |               Robust
         teamsize |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
------------------+----------------------------------------------------------------
    internetdummy |  -.0138526   .0022496    -6.16   0.000    -.0182618   -.0094434
invt_network_size |   .0094635   .0001072    88.32   0.000     .0092535    .0096735

Intercept=0.948

What is the right interpretation of this -0.0138?

Thanks
Ludo

[Marcos Almeida]

You can read about the coefficient interpretation here.

That being said, I recommend to use the incidence rate ratio instead.

[Jeff Wooldridge]

It means that, holding other things fixed, the team size is on average about 1.4 percent lower for those with high-speed internet access. This magnitude is small enough that using the incident rate ratio (irr) will give you essentially the same answer.

BTW, this interpretation has entirely to do with the exponential functional form of the mean -- just like Poisson regression. The NegBin distribution plays no role. In fact, you should check your findings by doing Poisson regression with robust standard errors. Poisson regression is more robust than NegBin for estimating the conditional mean parameters.

[Ludovic Van Cau]

A poisson regression with robust se gives almost the same results. When running a nbreg the alpa is equal to .048, quite low. So I assume there is not much overdispersion and should stick with the poisson?

[Tim Tolkmitt]

Hello,

I'm also unsure with the interpretation of coefficients / IRR of the negative binomial regression. I thought, that the following equation was true Beta = ln(DV(x+1)/DV(x)), holding all other variables fixed. Following that equation in the case of Ludovic's example, would mean that high-speed internet access leads to a decrease of team size by a factor of e^-0.0138526=0,986. Is that interpretation valid? In that case it is the same result like Jeff proposed. Is the interpretation also valid for continuous IVs? In my case I have standardized values of a 5 point likert scale rating.

Thanks a lot in advance!

Tim