I'm doing my master's thesis with R using the 2019 data from National Survey on Drug Use and Health. My purpose is to find out whether there are connections between the MDMA use and serious psychological distress measured by K6 scale.
I'm using the survey weights in the dataset.
I ran different binomial logistic regression models, with weights and without weights. I have rather many variables that I'm adjusting, like other substance-use, so in the end the p-values are mostly too big and the confidence intervals as well. Without weights I at least get some numbers, but when I run the model with weights, it only gives me NaN as CI and p-values.
What can I do about this? I already erased all NA's from the data, so that's not the cause.
There's over 40 000 respondents, but I don't know would it help if I had even more, which I could do by combining data from different years. Although 40 000 is a lot, I'm also adjusting the frequencies of the MDMA-use and within those groups the n drops quite a lot, like the group of the people who have used MDMA within a month have only 145 respondents. And only 37 respondents have serious psychological distress AND have used MDMA within a month.
Anyway, the odd ratios I'm getting now don't really tell anything if the CIs and p-values are just "NaN". So I have to do something about this. Any ideas?