Uriel Cohen Priva
LightNews LightNews
Uriel Cohen Priva
@urielcohenpriva.bsky.social
75 followers 66 following 34 posts
Likely not here at all.

Diverse, new, and dangerous opinions.

www.urielcohenpriva.com
Posts Replies Media Videos
Linguists seem to believe that modeling an effect using random effects causes the estimation to be very different than using fixed effects, because random effects are assumed to be normally distributed. Usually, the estimates are (almost) identical, regardless of the actual distribution.
The R code shown is: g <- 1:200 |> as.character() g.x <- 2**rnorm(length(g)) n <- 10**4 y <- rnorm(n) + g.x d <- data.frame(y, g) m.lmer <- lmer(y ~ 0 + (1 | g), d) m.lm <- lm(y ~ 0 + g, d) cor(coefficients(m.lm), ranef(m.lmer)$g[,1]) and the result is 1 (the two perfectly match) The following R code: g <- 1:200 |> as.character() g.x <- runif(length(g)) n <- 10**4 y <- rnorm(n) + g.x d <- data.frame(y, g) m.lmer <- lmer(y ~ 0 + (1 | g), d) m.lm <- lm(y ~ 0 + g, d) cor(coefficients(m.lm), ranef(m.lmer)$g[,1]) The result is again 1 R code: g <- 1:200 |> as.character() g.x <- c(runif(length(g)/2, 1000, 1100), runif(length(g)/2, 0, 100)) n <- 10**4 y <- rnorm(n) + g.x d <- data.frame(y, g) m.lmer <- lmer(y ~ 0 + (1 | g), d) m.lm <- lm(y ~ 0 + g, d) cor(coefficients(m.lm), ranef(m.lmer)$g[,1]) Again, the correlation is 1, despite the very non-normal distribution
July 15, 2025 at 3:45 PM
xkcd.com/386/ 🙂
A density plot of schwa (second vowel in "adult") and the wedge vowel (the first vowel in "adult"), showing that the two are distinct, though of course there's some overlap.
March 16, 2024 at 1:37 PM
Israel news item (related to politics, but isn't really). The situation in Gaza *increased* Israeli Palestinians' (that is, Arab Israeli citizens) identification with the state of Israel, by a huge margin. I just hope Israel doesn't squander that sentiment as it always does. The alt text translates.
A figure that shows Jewish and Palestinian / Arab Israeli citizens over time. Identification is fairly solid for Jews, but shifted from around 85% to 94% between June and November. For Palestinian / Arab Israeli citizens it jumped from 40% in June to 70% in November, and has only been that high in 2005 (the year Israel pulled out of Gaza) but never since. Margin of error is 4%
November 10, 2023 at 2:32 AM