I used SAS SURVEYSELECT procedure to identify 100 simple random samples from a larger data set. Now I want to show some sort of average odds ratios (with 95% confidence intervals) for a common outcome.
What calculations are statistically appropriate?
What SAS procedures will produce these statistics?
But does your "larger data set" not allow for a direct calculation of the odds ratios and accompanied confidence intervals? I do not see the reason for bootstrapping. I find bootstrap (and other resampling methods) useful when the sample distribution of the statistic of interest is intractable to derive analytically. However, the odds ratio does not provide a challenge - results for asymptotic sample distribution is available.
I agree with the first responder. If you stil want to compare bootstrap results with standard results, take the mean of the 100 odds ratios from the 100 bootstrap samples as an estimate of the OR. The simplest confidence interval is to take the 5th and 95th of your ordered 100 estimate to get a 90% confidence interval, however, 100 may be too small a number for reliable estimates. SAS PROC UNIVARIATE will do that for you. More sophisticated methods are described in the book of Efron and Tibshirani.
- Badges
- Science topic
- Similar topics
- Force