# Thread: Rate ratio interpretation for large estimates

1. ## Rate ratio interpretation for large estimates

Hi all,

Just a quick question - I have many rate ratios and the 95% CIs, most of which have narrow CIs and reasonable RRs (e.g., 1.23). I've been interpreting the RRs in terms of % increase, so for example, when my RR says 2.33, then it is a 133% increase over my reference category. However, I have some very high RRs and I know that it has to do with sample size issues because the CIs are very wide....but I'd still like to know how this translates into % difference. For example, I have a few RRs that are like 13 and one that is 48. I just can't seem to wrap my brain around how to translate that into a percentage - would it be 1300% or 4800%? Thanks!

2. ## Re: Rate ratio interpretation for large estimates

Your latter descriptions are correct, though typically you would switch over and describe them as times not percentages. So, 2.3 times, 13 times, and 48 times greater risk.

3. ## Re: Rate ratio interpretation for large estimates

Excellent, thanks! I was trying to see how I could possibly present the RRs in a graph, whether it would be % increase or the RR, but for the graph that would have the 13 and 48 estimates it just dwarfs the other estimates, which are like 1.7 or so. Maybe that's not the best way to go anyway!

4. ## Re: Rate ratio interpretation for large estimates

Are the confidence intervals really wide?

5. ## Re: Rate ratio interpretation for large estimates

For most of the RR they aren't very wide, but definitely for the 13 and 48 RRs (CIs are like 1.63-108.79 and 6.2-373).

6. ## Re: Rate ratio interpretation for large estimates

Sounds like a case of data sparsity. In particular, you have low cell counts in some sub-cell groups, causing CIs to blow-up. This gets listed under lack of separation or near quasi-separation in maximum likelihood modeling. There is some relevant information under the term positivity. Likely graphs won't work well for your results.

7. ## Re: Rate ratio interpretation for large estimates

Originally Posted by hlsmith
Sounds like a case of data sparsity. In particular, you have low cell counts in some sub-cell groups, causing CIs to blow-up. This gets listed under lack of separation or near quasi-separation in maximum likelihood modeling. There is some relevant information under the term positivity. Likely graphs won't work well for your results.
Thanks, yeah, I was thinking it was likely due to sample size issues!

 Tweet

#### Posting Permissions

• You may not post new threads
• You may not post replies
• You may not post attachments
• You may not edit your posts