3 ms·
30 is not even statistically significant... Plus, my belief at least, this is the tip of the iceberg for some fields. (So don't worry there will be more.)
by eftychis 8y ago
30 is not even statistically significant... Plus, my belief at least, this is the tip of the iceberg for some fields. (So don't worry there will be more.)
- Dylan16807 8y ago30 vs. 0 is statistically significant for something that's theoretically a coin flip. It even beats a stringent five-sigma threshold.
- et2o 8y agoDefinitely! If I were being pedantic [I guess I am :-) ], I'd probably say the best prior would be p =(n male scientists/number all scientists) Overall I bet men are slightly more than 1/2, but even if it's an 80/20 split the p value for this would be many sigma.
- et2o 8y agoIt's extremely statistically significant, given any reasonable prior. Don't take my word for it, here's a couple lines of R code. # vector of prior probabilities, natural choice is proportion of men prop.men <- seq(0, 1, length.out=100) # compute prob of observation (30/30) given prior probability pvals <- dbinom(x=30, size=30, prob=prop.men) plot(prop.men, log10(pvals), type="l", col="blue", lwd=2, ylab="log10 pval for 30/30 offenders being men", xlab="Baseline male percentage in science" ) abline(h=log10(0.05), col="red", lwd=2, lty=2) # "significance" line text(x=0.6, y=-30, lab="Below red line\nsignificant at α=0.05") # since prob vector is 100 long, indices of entries returned here # correspond to % men needed for this to have α > 0.05 which(pvals>0.05) The actual figure for completeness: https://imgur.com/a/E47AfSW https://imgur.com/a/E47AfSW Of course this is a bit back-of-the-envelope done in 5 minutes, but you get the point.