- you possibly experienced an overdispersion problem in performing a Poisson
regression and decided to switch to a negative binomial model;
- as other Statalisters have already pointed out, you have a "denominator
problem", since incidence rate ratio needs a measure of time (eg patient
follow-up in clinical trials) in its denominator to be calculated.
Ronan Conroy proposed to assume that all subjects are observed for the same
span of time, provided that this hypothesis can be soundly defended in your
research field. Unfortunately, is really rare that all the subjects
contribute to the same extent in terms of patient-years, as this would mean
that no one is lost during the observation period.
Kind Regards,
Carlo
-----Messaggio originale-----
Da: [email protected]
[mailto:[email protected]] Per conto di Muhammad Riaz
Inviato: mercoledì 2 settembre 2009 13.42
A: [email protected]
Oggetto: st: Fw: Negative binomial regression and Incident rate ratio(IRR)
Hi,
I am resending this query if somebody could help with, I sent it yesterday
but did not get any response.
Adding further my query given below, my dependent variable is count (0-9)
based on certain scale, i mean the dependent variable measure counts of 9
items for each individual, these counts are given in the table as:
d_cap |
| Freq. Percent Cum.
------------+-----------------------------------
0 | 449 38.71 38.71
1 | 214 18.45 57.16
2 | 163 14.05 71.21
3 | 112 9.66 80.86
4 | 89 7.67 88.53
5 | 56 4.83 93.36
6 | 41 3.53 96.90
7 | 18 1.55 98.45
8 | 12 1.03 99.48
9 | 6 0.52 100.00
------------+-----------------------------------
Total | 1,160 100.00
Can we fit a regression model with negative binomial distribution?
How can we compute the Incident rate ratio?
--- On Tue, 9/1/09, Muhammad Riaz <[email protected]> wrote:
> From: Muhammad Riaz <[email protected]>
> Subject: Negative binomial regression and Incident rate ratio(IRR)
> To: [email protected]
> Date: Tuesday, September 1, 2009, 3:22 PM
> Dear All,
> I am trying to fit a model for count data using Negative
> binomial regression, it gives incident rate ratios for the
> variables in the model, one of the options for this type of
> model is specifying the offset variable while fitting the
> model in stata using glm. This offset should be the person
> years (if i am correct). however I do not have data on this
> variable. My question is:
>
> Is it correct to fit the model without an offset and would
> the Incident rate ratio be still correct to report If I do
> not specify the offset variable in the model?
>
> I will really appreciate any help.
>
> Thank you,
>
> M Riaz
>
>
>
>
>
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
