Jeff Rouder

I think the less packages you use in R, the better. Right now, ggplot2 3.5.0 is incompatible with ggtern. Boo!

Today, it was announced that all UCI faculty will have to take a mandatory training course on ethics and conflict of interest. Today, it was reported Trump is demanding $230M from the Justice Dept. led by his crony appointees.

ok, i am back here. it’s been a while, but i miss academic twitter. Todays subject: Canvas is out. It is affecting my course as we had an assignment due last night and now I am dealing with many, many emails from students.

This is a good example where you want to include main effects in the model, however, because the lack of observed main effect is surely coincidental

Karl Marx studied calculus in his spare time: www.marxists.org/archive/marx...

“Stats are for losers.” — Wayne Gretsky, 2/12/25, TNT broadcast.

I am reading the most insightful, scholarly, fun text I have read in a long while. It is “The Myth of Statistical Inference” by Michael C. Acree. I cant put it down. Fascinating analysis; charming writing style. Wish I had the author’s email if he still gets email.

Anyone ever read, “Boswell, J. (1986). The awful truth about publishing: Why they always reject your manuscript— and what you can do about it. New York: Warner Books.” Sounds like my kind of book.

Q? Have you transitioned from base R or ggplot2 graphics to python Seaborn? Are you happier? Is it easier? Can ppl with experience in both comment? Thinking about going from R to python and Seaborn would be foundational in my workflow for visualization. Thx.

I am gratefully in the service of my students. I am gratefully in the service of my students. I am gratefully in the service of my students. If I repeat it enough, will i detest grading any less?

Q? Do you teach looking up values in t/F/z tables? Did you? If so, when did you stop? For me, it was around 2005. So, up to 20 years ago, I taught students how to use these tables. Does anyone still teach this?

Yes, I do wonder what percentage of our natural gas bill is for seasoning cast iron pans.

To that point, if I were doing ML, I might do full range optiminzation on log variance in my target function to avoid minimization to negative values. Why not?

I dont mean to sound like “Bayes is better, you suck” kind of guy. I see this as a constrained optimization problem and I am not sure why negative variances are allowed in optimization. It is assuredly my lack of understanding. But maybe u Bluesky’ers can help? Why not constrain solutions?

I am a Bayesian, we dont have these issues. I suspect ppl encounter this a lot and it affects what they do and conclude. This effect is kind of weird to me; why dont the programs just use properly constrained optimization?

Do you use lavaan? If so, do you run across negative variances or correlations out of range on occassion (Heywood cases). What do you do?

Hello All, First time here in a while. I used to love twitter, and I left when Elon bought it. I hope to see this platform become that vibrant exchange of ideas for me that twitter was.

Computer modern latex font. It's the best because it shows you use latex.

Damn, are you all as nervous as I am. Can't really concentrate on anything today. Just waiting for polls to close.

I am currently prepping my course for the week. There is some chance that come Wednesday I am so traumatized and intellectually paralyzed by the election results that the only words I can stammer in class are, "What The Fuck Is Wrong With This Country." At least I will have good slides then.

Nice find.

I cant tell you how disconcerting hallucination on a Google search is. By 2008, we had witnessed the end of bullshit. Whenever someone was just bullshitting, we had smart phones and could call them out. Now, the smart phones are full of bullshit.

Ok, we all know facts on chatGPT is not reliable. And, the vast majority of facts we find on Google, say "what is the annual rainfall in Tel Aviv," is reliable. Except, now, with new Google AI, it rains 57" a year in Tel Aviv. It became a rain forest. 1/2

You know, even after decades in the field and scores of papers an grants rejected, you would think it wouldn't bug me. Oh, quite the contrary, getting a paper rejected still upsets me greatly.

I love this!

I have been social media averse lately. But I miss my old twitter feed greatly. So, I am in the market for more friends and enemies, especially those that post and argue (thoughtfully preferred, but I am not too picky). There is a lot to complain about!

Picking a fight w/ bels and decibels. We are better off without them. Why, because there is no physical unit log(cm) or log(kg). We just say "log-multiplier" and "tenth-log-multiplier." When you say a multiplier, the multiple has no unit and the need for a physical referent is obvious. Thoughts?

still need any help?

I am traveling now, but when I get my laptop back at the end of camino pilgrimage I will write more. In the meantime rouder.mehrarz.2024 and rouder.haaf.ampps 2018? address related.

Distributions with full support of reals and a natural shift parameterization: normal, gumbel, ex-gaussian. Can you think of others?