@stephenjwild @mikedecr Equal variance + independence + usual t-test assumptions (https://t.co/QCf6MJ54jN). IMO you gain the ability to do a back-of-the-envelope mean difference comparison, but lose the ability to compare uncertainty across papers and also
@penthousedranks @BronskiJoseph @vituperativeerb @aw_dionysian Overlapping confidence intervals do not imply that the difference is not statistically significant. https://t.co/MpzYyOmzKP
@zyudhishthu That said, non-overlapping 95% confidence intervals around the mean are a very stringent test of difference in means. Quite likely that the old trick of 83.4% confidence intervals would show many state-to-state differences are "significant" h
RT @janzilinsky: Readers shouldn’t be making judgments based on “overlap of confidence intervals” And researchers should use the direct di…
Readers shouldn’t be making judgments based on “overlap of confidence intervals” And researchers should use the direct difference of means between groups for confidence interval estimation to reduce type II errors https://t.co/h8wFgqCWmv https://t.co/8P
@data_depot I know the error bars are overlapping a bit, but that doesn't prove that the difference is statistically insignificant. https://t.co/wRxUtVJQNQ
@benpreis Now I’m deep in the rabbit hole. Looks like it might be a thing: “To arrive at a type 1 error probability of 0.05, 83.4% confidence intervals should be calculated…if the variance is equal and the effect estimates are independent” https://t.co/ZV
The (mis)use of overlap of confidence intervals to assess effect modification https://t.co/CalFeGs6Cp
@ken_rothman Mirjam Knol has a nice paper about problems with interpreting overlapping confidence intervals, illustrated in the context of effect measure modification: https://t.co/SYlyczm59r https://t.co/0tmPKJVfpB
@societyforepi The following article addresses the misuse of overlapping confidence limits for assessing effect modification, which may be useful: https://t.co/DnblqoyT5j
@davidshor 83.4% works well most of the time: https://t.co/YoK0oqbry6
The case for 83.4% confidence intervals instead of 95% CIs http://t.co/pgmLFCTJmc