A question arose on #lesswrong as to whether female LWers might be more likely to find LW through MoR than not. There is an imbalance in MoR referrals by gender, but it’s not sufficiently extreme to hit significance in the limited survey dataset (need moar women):
R> lw <- read.csv("2012.csv")
R> lwm <- subset(lw, as.character(lw$Gender) == "M (cisgender)")
R> lwf <- subset(lw, as.character(lw$Gender) == "F (cisgender)")
R> summary(lwm$HPMOR)
" " "No" "Started it but haven't finished"
78 171 140
"Yes, all of it"
542
R> summary(lwf$HPMOR)
" " "No" "Started it but haven't finished"
6 18 20
"Yes, all of it"
49
R> 18 / (49+18)
[1] 0.2687
R> 171 / (542+171)
[1] 0.2398
R> # so, very similar percentages don't read MoR
R> nrow(subset(lwf, as.character(lwf$Referrals)=="Referred by Harry Potter and the Methods of Rationality"))
[1] 29
R> nrow(subset(lwm, as.character(lwm$Referrals)=="Referred by Harry Potter and the Methods of Rationality"))
[1] 206
R> c(29/93, 177/838)
[1] 0.3118 0.2112
R> # 10% difference? investigate with a chi-squared test
R>
R> N <- as.table(rbind(c(93, 29), c(838, 177)))
R> dimnames(N) <- list(morReferral=c("yes", "no"), gender=c("M", "F"))
R> chisq.test(N, simulate.p.value = TRUE, B = 10000000)
Pearson's Chi-squared test with simulated p-value (based on 1e+07 replicates)
data: N
X-squared = 2.943, df = NA, p-value = 0.1048
Doesn’t need to hit an arbitrary (if historically established) 0.05 to be significant. 0.1048 still means a (EDIT:) higher probability that you’ve found something than not.
A question arose on
#lesswrong
as to whether female LWers might be more likely to find LW through MoR than not. There is an imbalance in MoR referrals by gender, but it’s not sufficiently extreme to hit significance in the limited survey dataset (need moar women):Doesn’t need to hit an arbitrary (if historically established) 0.05 to be significant. 0.1048 still means a (EDIT:) higher probability that you’ve found something than not.
(Thanks for the correction.)
That is not what p-values mean.