Statistical Rethinking 2022 Complete Course is complete. 20 lectures from foundations of causal & Bayesian inference through DAGS, multilevel models & poststratified causal effects to Gaussian processes, Bayesian imputation & ODEs. Theatrical trailer:
Follow
Richard McElreath 
@rlmcelreath
Anthropologist, telling anyone who will listen that, if we are very careful and try very hard, we might not completely mislead ourselves.
Richard McElreath 🦔’s Tweets
I’ve signed so I guess it’s official: in March 2023 I’ll be the new ‘Lise Meitner’ Group Leader mpg.de/lise-meitner-e! Honoured to join the program. Immensely thankful to (and many others) for outrageous support 🙏🏻. Job ads coming soon!
27
13
215
At first I found this info-graphic confusing, but then I realized it is a DAG
3
1
29
Show this thread
Helping a cat is the purist form of altruism bc they are never thankful
Quote Tweet
3
5
96
Topics to follow
Sign up to get Tweets about the Topics you follow in your Home timeline.
Carousel
Cooking with Mischka: Chocolate chip cookies. He doesn’t much like cookies but he knows i bought turkey today
3
55
Show this thread
This is special case of general problem of showing uncertainty on maps. Gelman & Price have a thoughtful paper from 1999 on the topic. The title is well-chosen. PDF: stat.columbia.edu/~gelman/resear
4
37
Show this thread
Fascinating & the Leipzig map matches expectations, but has me wondering how we could show *variance*. Because as every Deutsche Bahn rider knows, the variance looms large in planning.
Quote Tweet
0:16
Isochrones 
! This map shows you how far a 5h train ride will take you, departing from any city in Europe.
chronotrains-eu.vercel.app
Inspired by Direkt Bahn Guru by @juliustens.
! This map shows you how far a 5h train ride will take you, departing from any city in Europe.
chronotrains-eu.vercel.app
Inspired by Direkt Bahn Guru by @juliustens.Show this thread
5
1
37
Show this thread
Isochrones 😍 ! This map shows you how far a 5h train ride will take you, departing from any city in Europe.
chronotrains-eu.vercel.app
Inspired by Direkt Bahn Guru by .
0:16
1.9M views
718
12.5K
56.7K
Show this thread
'Invincible Vietnam' — East German stamps issued between 1966-1979 as part of campaign titled 'Aid for Vietnam'.
4
58
311
We maybe need some clear terminology so we don't call both of these things "moderation"/"interaction", which are just vague statistical concepts?
6
25
Show this thread
In contrast, if predators are less dangerous when the temperature is low (because they are sluggish), then there is a *mechanistic* interaction between predation and temperature that goes beyond the basic measurement scale interaction.
1
1
13
Show this thread
Example. A tadpole might be killed by multiple causes: predator, starvation, temperature. If any one is sufficient, then changing the others doesn't matter. That's "moderation" but it's a side effect of the binary outcome. >
1
10
Show this thread
When we try to model interaction/moderation, there are subtle issues. In non-linear models, everything interactions to some extent just because of how measurement works. But there may also be *mechanistic* interactions that are more traditionally "causal". >
2
15
Show this thread
Since I answer this question at least once a week, imma retweet this clear statement below
Quote Tweet
Replying to @bmwiernik and @mattansb
I think this thread cleared this up for me. Moderation isn't represented in the DAG. Given a DAG with nodes X,A,Y and edges X->Y and A->Y, A may or may not be a moderator of the effect of X on Y. The DAG implies no linearity or functional form assumptions. twitter.com/alexkyllo/stat
1
8
69
Show this thread
Multiverse analyses have started to become more common in psychology—but they don't save you from making wrong or overconfident inferences. One well-specified model is better than 100 models with unclear assumptions re the data generating process 🤖
youtube.com/watch?v=KNPYUV
8
34
171
Show this thread
2
4
Support Europe’s bold vision for responsible research assessment: great editorial in this week’s on the new Agreement on Reforming Research Assessement (with a welcome nod also to & ’s work) nature.com/articles/d4158
1
63
121
Bias, such as a gender bias or a racial bias, is an important concept in the social sciences, but it is not always clearly defined. In this paper and I propose a definition of bias, based on structural causal models. arxiv.org/abs/2207.13665
7
69
207
Show this thread
crazy frog, france, 15th century
read image description
ALT
57
1,731
15.2K
CmdStan has an option for new optimizations which are not yet enabled by default. The most recent release fixed some issues, and I can now recommend testing it more widely to speed up sampling.
2
12
94
Show this thread
But overall and so there is no misunderstanding, showing that most of the these nudges have small (or zero) effects is important. It's a common intuition that subtle cues can have powerful effects. Showing they don't is progress. Papa Karl smiling down on us:
4
2
60
Show this thread
And last time I looked, still evidence that some "default" designs work. But maybe I missed something.
1
19
Show this thread
Anyway I think it's a good thing that we are seeing comprehensive debunking like this. We know much of science is wrong bc most of science has been wrong historically. (Pessimistic induction) If we aren't finding wrong science, we aren't looking.
2
8
61
Show this thread
The 2021 meta paper on nudging had this funnel plot. I remember laughing out loud when I saw it. Seems obviously a mix of publication bias and fraud. This is a field that won a Nobel prize in 2017. doi.org/10.1073/pnas.2
Commentary quoted below unpacks it transparently
Quote Tweet
The death knell for nudging?
“When this publication bias is appropriately corrected for, no evidence for the effectiveness of nudges remain.”
pnas.org/doi/10.1073/pn
Show this thread
7
93
335
Show this thread
these chopped vegetables compel you to pause scrolling and take care of yourselves
read image description
ALT
6
10
196
Not since the Battle of Teutoburg Forest in 9 AD have the Germans so demoralized the Romans
Quote Tweet
20
87
830
Dear funders, publishers, research instit. & researchers
We waste 82-89% of conducted ecological research
Ignorance of the problem is expensive, yet we seem to be successfully ignorant for decades
More in our paper , preprint & this 🧵
links in later tweets
1/x
read image description
ALT
7
94
210
Show this thread
So there's a lot going on but my book is 20% off right now, so if you've been wanting the high-quality dead-tree version of the 2nd edition, now's an opportune moment. Sale link: routledge.com/Statistical-Re
3
39
151
👇 we share our thoughts on how GitHub can accelerate collaborative and reproducible research in eco evo
this preprint resulted from discussions about (missed) opportunities, at the 2021 conference
Quote Tweet
Interested in using #GitHub in your research? Our preprint (osf.io/preprints/meta) includes:
12 ways #researchers in #ecoevo can start using GitHub in their work
10 tips for learning the basics of GitHub
and *much* more!Show this thread
8
28
The Department of Primate Behavior and Evolution would like to continue growing! and I are looking to hire a technician to support molecular genetics and cell culture-related research. Please spread the word!
56
47
I mean this is a person who understands the power of a well-timed triangle
1
9
Show this thread
Sure the internet seems like a mistake. But counterpoint (really enjoy the percussionist's vibe here):
1
4
27
Show this thread
Artyom Tolstuhin, Ukrainian painter
26
124
Thanks! Need data minding and data confession here rss.onlinelibrary.wiley.com/doi/abs/10.111, not much data mining or AI. Sampling rate 0.00005% could be ok. The million -- or billion -- dollar question is whether there was a strict quality control of the sampling & human review processes ...
Quote Tweet
Criticism of random sampling here and various comments bring to mind @XiaoLiMeng1’s paper.
#StatsTwitter
projecteuclid.org/journals/annal
1
9
52
Quote Tweet
Thanks! Need data minding and data confession here rss.onlinelibrary.wiley.com/doi/abs/10.111, not much data mining or AI. Sampling rate 0.00005% could be ok. The million -- or billion -- dollar question is whether there was a strict quality control of the sampling & human review processes ...
3
Show this thread
COVID bonus round, biggest vaccine surveys had the largest errors. Full paper: nature.com/articles/s4158
1
7
34
Show this thread