I am running a Logsitic regression, But the non events (0's) is just 1 and Events are 49(1's), Missing is 53, I fee this doesnt satisfy binomial distribution to run the MLR analysis? Always I get scenarios with rare events but not with rare non events.. Anyone can give suggestions would be great!
Another same model with a different outcomes has common events (yes and no are 54 and 49), this MLR is not converging, So I used Poisson regression with robust error variance (still the link function is Log) using GEE approach (Proc Genmod), The model got converged. Any thoughts on this would be helpful.
I do not think that you have enough information to estimate a reliable model and the non-convergence is telling you that. It does not matter whether the 0's or 1's are the rare events.
I do not think any any analysis will improve the situation, but changing the design could. If you have a rare outcome ( few 1's) consider a case-control design where you select on the outcome say to have an equal number of 1 an 0's and then find out how they differ in terms of X - their exposure. Rather than allow the 1's and 0's to occur naturally - you then use the conditional logit to do the analysis.
For background see Analysis of Case-control Studies: Logistic Regression
http://www.oxfordjournals.org/our_journals/tropej/online/ma_chap11.pdf
In SAS see
https://support.sas.com/documentation/cdl/en/statug/63033/HTML/default/viewer.htm#statug_logistic_sect062.htm
Hi Kris,
Regarding the first model, it would be useful to know what you are estimating. What is the dependent variable? What does it mean if it has a missing value? What are the independent variables? It is hard to imagine that you would be able to estimate any coefficients if you have almost no variation in the outcome.
As for the second model, I can't say for sure but your model may not be converging because you have included some independent variables that do not vary at all with respect to the dependent variable. So, for example, if you are controlling for respondent sex and the dependent variable is always equal to 0 when the respondent is female, you may have problems with convergence.
Hope this helps!
I do not think that you have enough information to estimate a reliable model and the non-convergence is telling you that. It does not matter whether the 0's or 1's are the rare events.
I do not think any any analysis will improve the situation, but changing the design could. If you have a rare outcome ( few 1's) consider a case-control design where you select on the outcome say to have an equal number of 1 an 0's and then find out how they differ in terms of X - their exposure. Rather than allow the 1's and 0's to occur naturally - you then use the conditional logit to do the analysis.
For background see Analysis of Case-control Studies: Logistic Regression
http://www.oxfordjournals.org/our_journals/tropej/online/ma_chap11.pdf
In SAS see
https://support.sas.com/documentation/cdl/en/statug/63033/HTML/default/viewer.htm#statug_logistic_sect062.htm
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis