About calculating absolute incidence rate in poisson model

[Zihan Dong]

Hi all,

I'd like to estimate absolute incidence rate using poisson regression.

I used following code:

poisson outcome i.ssc sex1 year2 region, irr exp(persondays)
margins ssc, predict(ir)

However, my main independent variable "ssc" is a binary count variable. And Stata warnings:
ssc: string variables may not be used as factor variables

Then I destringed "ssc" and re-runed the codes, it worked.

I was wondering is this destring appropriate? And could anyone explain what's the actual meaning of "exp(persondays)"?

Thank you in advance!

Best,
Z

[Paul Dickman]

As the message says, ssc is stored as a string variable. Even if it only contains numbers, you need to convert it to a numeric variable to use it as a factor variable. In short, the destring is not only appropriate but neccessary.


Your model is:

log(rate) = xb

log(events/persondays) = xb (where events is the total number of events (sum of outcome) for each covariate pattern)

log(events) = xb + log(persondays)

The outcome in Poisson regression is the number of events. When you are modelling rates, the log of person-time appears on the right-hand side. It is known as an offset. The syntax for the -poisson- command is to specify the offset using the exp() option. Other command use a different syntax (e.g., in -glm- you use the offset() or lnoffset() options).

[Zihan Dong]

As the message says, ssc is stored as a string variable. Even if it only contains numbers, you need to convert it to a numeric variable to use it as a factor variable. In short, the destring is not only appropriate but neccessary.


Your model is:

log(rate) = xb

log(events/persondays) = xb (where events is the total number of events (sum of outcome) for each covariate pattern)

log(events) = xb + log(persondays)

The outcome in Poisson regression is the number of events. When you are modelling rates, the log of person-time appears on the right-hand side. It is known as an offset. The syntax for the -poisson- command is to specify the offset using the exp() option. Other command use a different syntax (e.g., in -glm- you use the offset() or lnoffset() options).

Thank you very much Paul! The explanation is super helpful!