topic Re: Bootstrap confidence interval for Risk Ratio in Statistical Procedures

Bootstrap confidence interval for Risk Ratio

Sina — Tue, 20 Jun 2017 20:05:21 GMT

Hi. I'm fairly new to SAS and I need help with a bootstrap procedure. I have an epidemiologic 2x2 table that looks like this:

Now with this table, I can calculate risk ratio (defined as a/N0 / b/N1). However, since the numbers are not integers and are weighted quantities, I cannot use parametric methods to calculate CI. How can I calculate a bootstrap 95% CI for this sample? If there is not a direct way, would you recommend rounding the weighted quantities to the nearest integers and then create a hypothetical sample of say 2630 individuals and then use that sample for bootstrapping? Any other suggestions?

thanks!

Re: Bootstrap confidence interval for Risk Ratio

Rick_SAS — Tue, 20 Jun 2017 20:10:15 GMT

If the cells are not counts, what are they? A bootstrap resampling should usually mimic the data collection process. Please explain what these numbers are, what the underlying data look like, and how the underlying data were sampled.

Re: Bootstrap confidence interval for Risk Ratio

Sina — Tue, 20 Jun 2017 20:16:09 GMT

the cells are 'counts x stabilized weights'. The weights are in the inverse of the probability of being sampled and therefore can be viewed as sampling weights. So basically, we can deconstruct the data to rows with each individual having an exposure status, a disease status, and a weight.

Re: Bootstrap confidence interval for Risk Ratio

Rick_SAS — Wed, 21 Jun 2017 00:37:13 GMT

It sounds like this is survey data. If so, you can use PROC SURVEYLOGISTIC, which supports sampling weights. You can specify the CLODDS option on the MODEL statement to get confidence limits for odds ratios.Check out the Getting Started example and see if that helps.