Analysis of number of days, count data?

[Randolph Evans]

My data consists of the number of days in various components of a surgical procedure: days from initial appointment to consultation, consultation to surgery, surgery to discharge etc; and the independent variables are a mix of continuous: age, categorical: sex, surgeon etc. and ordered: severity of disease.

I considered the number of days to be a count and used negative binomial regression because the days are over dispersed and the results of the analysis make clinical sense.
However a referee has commented that since days are 'time' they are not count and the use of nbreg is inappropriate; and I should use streg with a Weibull or lognormal distribution.

My queries are:
1. Why cannot I treat the number of days as a count variable, there is an example at http://www.ats.ucla.edu/stat/stata/dae/nbreg.htm.
2. Is there any advantage to using streg?

I am using Stata 14.

Thank you.
Randolph

[Carlo Lazzaro]

Randolph:
most depends on your research goal.
The referee is possibly right in suspecting that your depvar is not focused on the same "kind" of days (I would be more confident with a poisson regression approach if you were dealing with explaining the variation in days of hospitalization via a set of predictors).
Using -streg- will take you down to the Stata survival analysis suite of commands. However, as -streg- implies parametric survival regression, you should choose the distribution that fits your data. There are some example on that topic in Stata .pdf manual, that you can access via -help streg-.