Using Replicate Weights for Zero-inflated Negative Binomial in Stata


I am using the pooled Time Use data (2005-19) and conducting a series of zero-inflated negative binomial models in Stata.

However, when I applied the replicate weights to these models, Stata run these models pretty slow and even with some red ‘x.’ One solution I am considering is to only use WT06 without replicate weights. However, I still want to estimate more precise standard errors.

I wonder if anyone has a similar experience on this issue and how you fix the problem.



Advising on specific analytical models is beyond the role of the IPUMS User Support Team, though other Time Use scholars on the forum may have ideas. I will note that the data provider’s recommendation is to use the replicate weights for the most precise standard errors. I am linking the IPUMS CPS guidance on using replicate weights in case reference syntax (albeit for a different dataset) would be helpful. We often encourage people to troubleshoot their code before applying replicate weights as it can be much slower to run analyses with replicate weights. Finally, you might consult the relevant literature or try Statalist or Stats Stack Exchange for feedback on your model.