“…outsourcing the entire writing process to LLMs may deprive us of the opportunity to reflect on our field and engage in the creative, essential task of shaping research findings into a compelling narrative — a skill that is certainly important beyond scholarly writing and publishing.”
7 reasons to use Bayesian inference!
statmodeling.stat.columbia.edu/2025/10/11/7...
Chatbots — LLMs — do not know facts and are not designed to be able to accurately answer factual questions. They are designed to find and mimic patterns of words, probabilistically. When they’re “right” it’s because correct things are often written down, so those patterns are frequent. That’s all.
Drawing of gnat
Drawing of silk
Drawing of a needle's point
Robert Hooke's drawings of objects under the microscope were so beautiful. These are from "Micrographia" in 1665
Persistent minimizing of the COVID death toll hits me especially hard in the #demography feels. To be clear:
➡️ Over one million Americans died of COVID-19.
➡️ Official COVID deaths were likely undercounted, not overcounted.
jenndowd.substack.com/p/how-many-p...
The psych job market may not be dead... but it is gravely injured 😬 So far it's looking like the Trump administration's attacks on higher ed/research are going to have more than 2x the impact on the job market as the covid-19 pandemic. #psychjobs #neurojobs #academicjobs
The AI and Education discourse is maddening. "Does AI improve learning?" is a dumb question. What form of "AI" and in what context? What do you mean by learning? Improving on a test or knowledge for use?
If you're asking that question, you don't actually care about a real answer.
Screenshot of title page of article published in the Journal of Applied Research in Memory and Cognition titled "Expert Thinking With Generative Chatbots."
Great article with one of the best brief layperson introductions to AI v. LLMs that I’ve seen. And I love the exploration of whether and how LLMs’ are useful depends on the user’s level of expertise. #PsychSciSky #AcademicSky #EduSky
doi.org/10.1037/mac0...
Models as Prediction Machines: How to Convert Confusing Coefficients into Clear Quantities Abstract Psychological researchers usually make sense of regression models by interpreting coefficient estimates directly. This works well enough for simple linear models, but is more challenging for more complex models with, for example, categorical variables, interactions, non-linearities, and hierarchical structures. Here, we introduce an alternative approach to making sense of statistical models. The central idea is to abstract away from the mechanics of estimation, and to treat models as “counterfactual prediction machines,” which are subsequently queried to estimate quantities and conduct tests that matter substantively. This workflow is model-agnostic; it can be applied in a consistent fashion to draw causal or descriptive inference from a wide range of models. We illustrate how to implement this workflow with the marginaleffects package, which supports over 100 different classes of models in R and Python, and present two worked examples. These examples show how the workflow can be applied across designs (e.g., observational study, randomized experiment) to answer different research questions (e.g., associations, causal effects, effect heterogeneity) while facing various challenges (e.g., controlling for confounders in a flexible manner, modelling ordinal outcomes, and interpreting non-linear models).
Figure illustrating model predictions. On the X-axis the predictor, annual gross income in Euro. On the Y-axis the outcome, predicted life satisfaction. A solid line marks the curve of predictions on which individual data points are marked as model-implied outcomes at incomes of interest. Comparing two such predictions gives us a comparison. We can also fit a tangent to the line of predictions, which illustrates the slope at any given point of the curve.
A figure illustrating various ways to include age as a predictor in a model. On the x-axis age (predictor), on the y-axis the outcome (model-implied importance of friends, including confidence intervals). Illustrated are 1. age as a categorical predictor, resultings in the predictions bouncing around a lot with wide confidence intervals 2. age as a linear predictor, which forces a straight line through the data points that has a very tight confidence band and 3. age splines, which lies somewhere in between as it smoothly follows the data but has more uncertainty than the straight line.
Ever stared at a table of regression coefficients & wondered what you're doing with your life?
Very excited to share this gentle introduction to another way of making sense of statistical models (w @vincentab.bsky.social)
Preprint: doi.org/10.31234/osf...
Website: j-rohrer.github.io/marginal-psy...
It's so weird how LLMs know so much about things I don't know anything about, and yet make fundamental and basic errors about things I do know anything about. Oh well, I'm sure that's a coincidence.
Large Language Models Do Not Simulate Human Psychology
arxiv.org/pdf/2508.06950
All of Statistics is secret linear algebra.
We often don't do the linear algebra or don't know we are doing it because of assumptions.
Here is a more accessible resource for what you're really doing when you analyze human subjects data.
The cover of Bayesian Data Analysis book
The cover of Regression and Other Stories book
The cover of Active Statistics book
All three books I've co-authored are freely available online for non-commercial use:
- #Bayesian Data Analysis, 3rd ed (aka BDA3) at stat.columbia.edu/~gelman/book/
- #Regression and Other Stories at avehtari.github.io/ROS-Examples/
- Active Statistics at avehtari.github.io/ActiveStatis...
Reminder that all three books I've co-authored are freely available online for non-commercial use (and the fourth will be, too)
Today I’ll be sharing some of my favorite Quarto learning resources 💻✨
Most of what I know, I learned thanks to the amazing work of others. So this thread is a small tribute to open educational content and the people behind it.
Let’s go! 👇
#QuartoPub #RLadies #RStats
At this point, I might as well --
Here's an infographic showing different ways to include age as a predictor. The top shows two extremes, just as a plain old numerical predictor (imposes linear trajectory) vs. categorical predictor (imposes nothing whatsoever). And then three solutions in between!
#statstab #386 {bayestestR} Evaluating Evidence and Making Decisions using Bayesian Statistics by @mattansb.msbstats.info
Thoughts: Want to start using Bayesian stats? Here is a quick but comprehensive guide in #R
#bayesian #bayes #mcmc #easystats #guide
mattansb.github.io/bayesian-evi...
Sometimes authors summarize their responses to 3-6 major points raised by the reviewers in their resub cover letters or at the top of their rebuttal letters. This is so super helpful to editors, because it brings us right up to speed after not having read the paper/reviews for months. THANKS!!
Hey #rstats,
What's your rule for splitting R scripts that form part of a wider analysis pipeline / project?
I usually write a single script which includes sections for each step from data cleaning to the final results, but it can become unwieldy when the script becomes long.
...
This article does a nice job describing the dos & do-not-dos of media comparison research. People wishing to compare e.g., ChatGPT v. traditional instruction would be wise to review the "Background on Technology in Education" section. #PsychSciSky #AcademicSky #EduSky doi.org/10.1007/s106...
Experimentology cover: title and curves for distributions.
Experimentology is out today!!! A group of us wrote a free online textbook for experimental methods, available at experimentology.io - the idea was to integrate open science into all aspects of the experimental workflow from planning to design, analysis, and writing.
Positron with a chat panel open with a question about fixing a ggplot plot
Positron now supports a chat panel with Claude and it's pretty neat
I read *a lot* of scientific abstracts that are missing key elements.
Here are the 5 things an abstract needs:
1. Introduce the topic,
2. State the unknown,
3. Outline the method used to answer the question,
4. Preview the findings, and
5. Tell us what your work teaches us.
"Even when explicitly prompted for accuracy, most LLMs produced broader generalizations of scientific results than those in the original texts."
Say it after me: Chat GPT is not a search engine. It does not scan the web for information, it just generates statistically likely sentences. You cannot use it a search engine, or as a substitute for searching.
Now. Please never use an LLM for information searches ever again.
"The most troubling implication of the learning styles myth is that it invites low expectations. Labelling a student a “kinesthetic learner” or a “visual learner” easily becomes shorthand for what they can’t do."
Three box plot/density plot/rainfall plots showing body weight for three penguin species, with Gentoo significantly higher. Colours also show density, and species. Background is light with dark text.
Three box plot/density plot/rainfall plots showing body weight for three penguin species, with Gentoo significantly higher. Colours also show density, and species. Background is dark with light text.
A quick plot for #TidyTuesday this week - celebrating the addition of the penguins data into base R 🐧 I used {ggdist} to plot the distribution of body weights with four different chart types in one! 📊
I couldn't quite decide whether I prefer the light or dark version💡
#RStats #DataViz #ggplot2
Increasing the playback speed of video lectures is popular amongst students as a time saving strategy, but does this negatively impact test performance? Here, we conducted a meta-analysis to examine the effect of increasing video lecture playback speed on content test performance. A meta-regression with robust variance estimation was used to aggregate data from 110 effect sizes, stemming from 24 studies of learning from lecture videos. The results demonstrated that increasing the playback speed of lectures can negatively impact content test performance, but this cost is small (and often non-significant) for speeds 1.5 x and slower. In addition, we found no evidence of moderation of this cost by a number of theoretically important variables (e.g., test type, lecture duration). These results contribute important insights into a popular study strategy and one that is likely to be a mainstay in educational settings for years to come.
Extremely useful and timely meta-analysis on the impact of increasing the speed of lecture recordings - 1.25x and 1.5x no/low cost that is probably balanced by increased engagement but 2x and 2.5x impairs learning.
#AcademicSky
link.springer.com/article/10.1...
#statsmeme
