Having seen this plot, are you more or less confident in the statement that IBS is associated with pet ownership?
-
Show this thread
-
See, the thing is, that top study appears to be contributing the ENTIRE association. Every other study found no result at all, but one single study has caused the entire relationship to become statistically significantpic.twitter.com/pcW30Bzkhm
1 reply 0 retweets 2 likesShow this thread -
Being the nerd I am, I decided to rerun the meta analysis on their sample using the metan command in Stata This is a bit quick and dirty, but using a random-effects model with an inverse variance, I get these resultspic.twitter.com/s95D5VYPQ7
2 replies 0 retweets 2 likesShow this thread -
For the epi nerds, when I run it with a fixed-effects model my results are the same as those reported in the paper, but my random-effects model CI crosses 0
2 replies 0 retweets 5 likesShow this thread -
Replying to @GidMK
what method did the paper use? fixed-effects (common-effects) seems rarely appropriate for most situations
1 reply 0 retweets 1 like -
Replying to @dailyzad
That's the thing - they report using a random effects model, but I can't replicate that in Stata. I'd probably need to see their code to get the precise effect estimate they generatedpic.twitter.com/2fH3xll8Tc
1 reply 0 retweets 0 likes -
(I'm also using the DerSimonian and Laird variance estimation, but my guess is that the procedure in Stata produces very slightly different results than other statistical software)
1 reply 0 retweets 0 likes -
Replying to @GidMK
Hmm interesting. I don't think they actually used the DL method... I calculated the standard errors by taking the natural logs ln(UL/LL) and tried to reproduce it with *metan logOR logSE, fixed eform* I got much wider interval estimates (as expected with random effects)pic.twitter.com/y7kXtgU7YU
2 replies 0 retweets 0 likes -
I got the reported results with *metan logOR logSE, fixed eform*. Double checked on R with metafor and got similar results. Any case... I don't think it actually matters whether the interval includes the null or not. All of included studies are regarding nonrandom exposurespic.twitter.com/Ng2PJYtv9A
2 replies 0 retweets 1 like -
Replying to @dailyzad
Yes I saw exactly the same. I reckon they misreported the model they used in the abstract I do agree about the nonrandom exposures, but I think it's worth noting that, done properly, even their own analysis doesn't show an effect
1 reply 0 retweets 0 likes
Also, it's really not ideal practice to not at least do a sensitivity analysis excluding the single study that's driving the entire effect, but to be fair that's what I'd expect from a conference abstract which is why this shouldn't have been a news story in the first place!
-
-
Just noticed if you exclude that study the I^2 drops from 27% to 0% so I reckon there's definitely a rationale for using the sensitivity analysis as the most robust estimate
1 reply 0 retweets 0 likes -
Replying to @GidMK
The contribution it has to the summary effect is interesting, but not sure if the analysis excluding it is more robust. Heterogeneity variance is a part of life and estimators need to account for it, would've been better to use REML or HKSJ rather than DL given the # of studies
1 reply 0 retweets 1 like - Show replies
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.