Your are viewing a read-only archive of the old DiS boards. Please hit the Community button above to engage with the DiS !
I need some help :(
until you get the answer you want.
I need help with my model. I have count data with a variable component of time, and matching between cases and controls of 1:4. The data I'm investigating has excessive zeros and is mostly over-dispersed. Now I need to what is the best model and whether the model has a coefficient for time.
My choices are:
Conditional Poisson Regression
Negative Binomial Regression
Zero-inflated Negative Binomial Regression
What to do, what to do?
Despite the off the cuff remark which gave the impression I know more about stats than I do.
It's been about 8 years since I've done them and that was just one module.
but what are you measuring?
if my memory serves me right, i'd probably go for the Poisson
but that's definitely not gospel
No stats buddies?!
or surgery consultations. The problem I have is I can't find a method that takes into account different time of follow-up AND allows for matching. I'm sure there is one though.
It's been a long time since I've done this, but what are you measuring?
from the sounds of it, I'd choose the Poisson cos it's got time, and from what you're saying about your data, it also accounts for rare events.
but that is by no means gospel
we all have access to Wikipedia. Therefore 100% of all DiS users are experts in every subject ever.
my replies are all over the place!