Stephen Wild
LightNews LightNews
stephenjwild.bsky.social
Stephen Wild
@stephenjwild.bsky.social
3.5K followers 970 following 5.2K posts
I try to put straight lines through things but usually fail. Try to be Bayesian when I can. Views my own. RT/like != endorsement.
Posts Replies Media Videos
stephenjwild.bsky.social
Meme of two astronauts looking at earth. One is pointing a gun at the other. First astronaught: "it's all mystical incantations?" Second astronaut, who is pointing a gun at the first asteonaught: "Always has been."
November 29, 2025 at 5:35 PM
stephenjwild.bsky.social
Oh, no!
R code below. They use a glm with family = binomial), but include national weights. modelES<- glm (support ~ treat*imm_1, data=spain, weight=nationalweight, family="binomial") summary(modelES, robust=TRUE) spain$predictINT2<-predict(modelES, spain, type="response")
November 28, 2025 at 5:42 PM
stephenjwild.bsky.social
Everyone should use {marginaleffects} because it includes s-values
A screenshot showing the output of the marginaleffects R package. Included is the s-value.
November 25, 2025 at 3:39 PM
stephenjwild.bsky.social
With respect.
Homer and Marge Simpson in bed. Homer is surrounded by crows. Marge: I don't like seeing all these subtweets. Homer: It's a "murder," honey. A group of subtweets is called a "murder
November 24, 2025 at 11:55 PM
stephenjwild.bsky.social
In meme form (with apologies):
Homer and Marge Simpson in bed. Homer is surrounded by crows. Marge: I don't like seeing all these subtweets. Homer: It's a "murder," honey. A group of subtweets is called a "murder."
November 24, 2025 at 11:51 PM
stephenjwild.bsky.social
This is a very good paper (focused on infectious diseases and DiD) that looks at how DiD handles different specifications.

www.medrxiv.org/content/10.1...
We vary the choice of outcome variable across rows and the underlying data-generating process across columns. We simulate data from an SIR model and average outcomes over 1000 draws. Lines indicate the average outcome value for two units. Assume unit 0 is graphed in blue, and unit 1 is graphed in red. There are no treatment effects in any scenario; therefore plots display untreated potential outcomes. If trends are parallel (checkmark), then DiD would estimate unbiased treatment effects with the corresponding model specification. Baseline parameters are N = 108, I0 = 50, 1γ = 10, βt = 0.105 ∀t. (a) “All parameters match” matches all parameters between units for all t. (b) “Different initial infections” changes the initial number of infections for the unit graphed in red to I1,0 = 100. (c) “Different transmission” sets β1,t = 0.2. (d) “Different time-varying transmission” sets β1,t = 0.2 for the treated unit and doubles both β0,t and β1,t for time t ≥ 45. (e) “Non-trivial susceptible depletion” sets a smaller population size N = 104 for both units and sets β1,t = 0.160
November 24, 2025 at 2:30 AM
stephenjwild.bsky.social
The text says .03
In a later analysis, Rosenthal (1990) calculated that the correlation of .03 between taking aspirin after a heart attack and prevention of future heart attacks implied the prevention of 85 attacks in a sample of 10,845 individuals.
November 23, 2025 at 9:35 PM
stephenjwild.bsky.social
The children have suffered enough with this monstrosity
Woody the talking Christmas tree. It is terrifying.
November 22, 2025 at 11:29 PM
stephenjwild.bsky.social
My view remains the same:
Personally, I think there is probably a small negative direct effect of social media on mental health. I am agnostic about the indirect effects: I have no idea if they are positive, negative, or how they interact. Things that appear simple at the individual level can lead to complex, unpredictable outcomes at the network level.
November 22, 2025 at 7:36 PM
stephenjwild.bsky.social
ridgewallet.ca/products/alu...
An image of a wallet. It is made of aluminum and designed to hold credit cards.
November 20, 2025 at 5:35 PM
stephenjwild.bsky.social
German and English really are close:
A post that says the author has marked a post as being written in German. The post is in English
November 13, 2025 at 2:27 PM
stephenjwild.bsky.social
That's why you need things lkle this (this comes from a paper that reported positive results. Some of them are clearly driven by the prior)
A plot of effect estimates. While most are positive, a few are only positive *because of the prior*
November 8, 2025 at 5:43 PM
stephenjwild.bsky.social
Image of a man with a couple of coffee sitting st a table. The sign in the table says, "November is a fall month. Change my mind."
November 1, 2025 at 3:21 PM
stephenjwild.bsky.social
Yet, I contend, some datasets on which the MLM has been estimated are non-probability samples for the MLM. If so, estimators are biased and the tools of inferential statistics (e.g., standard errors) are inapplicable, dissolving the foundation for findings from such studies. Further, this circumstance may not be rare; the processes transforming probability samples into problematic samples for the MLM may be inconspicuous but widespread. If this reasoning is correct, many published MLM study findings are likely wrong and, in any case, cannot be evaluated, meaning that, to the extent our knowledge depends on that research, our knowledge is compromised.
October 28, 2025 at 4:59 PM
stephenjwild.bsky.social
In fairness, he does say he is joking
A Twitter poster says he was joking.
October 28, 2025 at 12:45 AM
stephenjwild.bsky.social
Mr Burns pointing at a painting of his body. Text: "You know, I'm no art critic but I know what I hate, and I don't hate this."
October 26, 2025 at 10:29 PM
stephenjwild.bsky.social
Sigh...
Costs on Amazon. Kindle edition: $115.0. Hardcover: $260.00. Paperback: $119.39.
October 25, 2025 at 12:27 PM
stephenjwild.bsky.social
This is just funny tbh:
Answered by Demetri. Views change over time, and it's cool to see.
October 25, 2025 at 11:31 AM
stephenjwild.bsky.social
Was coming to suggest Cohen's book. He also say a bit earlier than the sample sizes would be so large as to be outside the resources of investigators (which was a serious constraint then, not so much for some areas now)
However, in the judgment of the author, for most behavioral science research (although admitting of many exceptions), power values as large as .90-.99 would demand sample sizes so large as to exceed an investigator's resources. Even when, with much effort or at much cost, these large n's can be attained, they are probably inefficient, given the nature of statistical inference and the sociology of science.
October 24, 2025 at 12:35 PM
stephenjwild.bsky.social
It contains on my favorite misdirection gags.
Homer running from dogs. President Ned Flanders is on the screen. Homer running away from angry dogs. He pulls some weiners out of his pocket and says, "Oh, no, they're gaining on me. Wait! I have an idea. These wieners will give me the energy I need to escape."
October 19, 2025 at 1:15 AM
stephenjwild.bsky.social
Obligatory Simpsons reference
Scene from The Simpsons. Homer and Marge are in bed. Homer is surrounded by crows. He says, " It's a "murder," honey. A group of crows is called a "murder." "
October 17, 2025 at 10:38 PM
stephenjwild.bsky.social
I have a soft spot for three:
A DAG showing a mediator: A -> M -> B
October 17, 2025 at 1:19 PM
stephenjwild.bsky.social
I would be a wee bit concerned about recruitment and selection effects here (and no, a "diverse sample" does not mitigate this)
Participants were recruited via posts that were made on social media platforms, leveraging the reach and audience of the University of Oregon and Google accounts. Also, in app-notifications were made in the Fitbit app inviting users to participate in the study. Furthermore, the study was limited to individuals who use an Android smartphone as their personal device, which could have introduced some demographic and other biases (41), although the large and diverse sample that was recruited does mitigate against this concern somewhat (Table 4)
October 16, 2025 at 6:56 PM
stephenjwild.bsky.social
Gonna anger everyone
A meme featuring a penguin, an elephant, and weird penguin-elephant hybrid in Noah's Ark. The elephant is labeled "Confidence intervals," the penguin is labeled as "credible intervals," and the weird hybrid as "confidence distributions." Noah gestures at the hy rid and says, "What the hell is this?"
October 8, 2025 at 9:40 PM
stephenjwild.bsky.social
Gonna have to tap the sign: bsky.app/profile/step...
In the present study, we explore the correlates of sexlessness in data from the UK Biobank which consists of ~500,000 middle and old age UK residents who have been genotyped and assessed with extensive testing and surveys
October 6, 2025 at 8:30 PM