talgalili
commented
2 months ago TODO (thoughts):
- Decide on a ratio of target/sample (it should probably also depend on the model matrix, might be worth finding some good role of thumb from the literature).
- Maybe propose a user to do the sampling for them as an argument? (with control of the seed)
- Should the warning be during the loading of the data, or when running adjust, or both?
Add a warning message to Balance when trying to run very large/imbalanced weights (say, anything more than like 100k cases and population frame that's 10x the sample). The thinking here is that, if the population is >10x the sample, then basically all the standard error in comparing (sample vs population) is coming from the sample rather than the population. It’s comparable to a 1-sample t-test rather than a 2-sample t-test.
Idea from: Ben Mainwaring