Reader Comments

Post a new comment on this article

Type I error when analysing 18 conditions?

Posted by gkaneto on 07 Jul 2014 at 16:59 GMT

Hi! Have you heard about something like this?

"Maintains that a Type I error becomes inflated beyond conventional acceptable levels when a researcher performs individual univariate statistics (such as t tests or ANOVAs) on each of several dependent variables within a single research project."
http://psycnet.apa.org/ps...

Does that apply to your analysis?

(Also, I simulated a random population divided in 4 groups (same sizes as yours) with different conditions, and it's not very difficult to get 3 significant results...)

No competing interests declared.

RE: Type I error when analysing 18 conditions?

gkaneto replied to gkaneto on 07 Jul 2014 at 17:29 GMT

just found this too: http://en.wikipedia.org/w...

No competing interests declared.

RE: Type I error when analysing 18 conditions?

dr_ewan_cameron replied to gkaneto on 06 Feb 2016 at 14:29 GMT

Of course it does: this mistake should have been identified at review. A conservative solution to handle multiple testing is to apply the Bonferonni correction, which in this case would replace the p=0.05 threshold with a p=0.05/18 threshold, meaning the null hypothesis would only be rejected for the allergies association but not for urinary infections, cancer, and mental health. Rather changes the results don't you think?

No competing interests declared.