Richard McElreath 🐈‍⬛

@rmcelreath

Anthropologist - Bayesian modeling - science reform - cat and cooking content too - Director @ MPI for evolutionary anthropology

1.411 Folgt18.262 Follower1.624 Posts

Bluesky

Ist das dein Profil?Übernimm es — dann gehört dir rmcelreath.dirty.social.Profil übernehmen

Richard McElreath 🐈‍⬛@rmcelreath· 1d

Working on my book, and as always getting distracted in side quests reading old primary lit. This week it seems to be this 1988 paper by Zellner advancing an optimality argument for bayes theorem (rather than usual coherence arguments). I might try to make a geometric version of this argument...

section 3 of Zellner's 1988 paper "Optimal Information Processing and Bayes's Theorem"

Richard McElreath 🐈‍⬛@rmcelreath· 6d

When I was a kid, I thought mac'n'cheese was Scottish. An honest mistake. My version: Béchamel base, lots of grated cheese, some pasta with a hole, a splash of paprika to make it dandy.

home made macaroni and cheese in a bowl, with some paprika on top for color

Richard McElreath 🐈‍⬛@rmcelreath· 8d

lol

The Order of Things, Michel Foucault

I tried but I can’t with continental philosophy.

Richard McElreath 🐈‍⬛@rmcelreath· 9d

smooth hackery, using the logistic function like a crowbar

$In practice, we require a non-zero probability for every temperature $T$. Which requires for $T \geq m + n$ that the function return some tiny non-zero probability. One way to do this is to use a logistic function to smoothly interpolate between $f$ and zero in the region of the thermal maximum $m+n$. Like this: \begin{align} g = f \, ( 1+\exp(\beta(T-(m+n)) )^{-1} \end{align} where $\beta$ is an arbitrary slope that governs the rate of transition. Unlike the raw function $f$, which goes to $-\infty$ as $T \rightarrow \infty$, the new function $g$ goes to zero, for $\beta>0$.$

Richard McElreath 🐈‍⬛@rmcelreath· 11d

imma take this rare Patti Smith in the news to promote her 2010 memoire "Just Kids" about her life with Robert Mapplethorpe. It's in turns incredibly romantic, tragic, and fulfilling. Read it!

Richard McElreath 🐈‍⬛@rmcelreath· 13d

I don't care what the haters say, tikz is nice, can have a tiny fish for a treat

LaTeX page showing a Tikz plot for a rescaled function of exp(x)(1-x), also with a small fish silhouette for fun

$code to generate the tikz plot: \begin{tikzpicture} \begin{axis}[axis lines = left, xlabel = $T$, ylabel = $f$, width=7cm, height=4cm, tick label style={font=\tiny}, label style={font=\small} ] \addplot[domain=-3:27.5, red, thick, samples=70] {exp( (x-25)/(0.5*(30-25)) )*( 1 - (x-25)/(0.5*(30-25)) )}; \node at (100,500) {\includegraphics[width=1.25cm]{fish.jpg}}; \end{axis} \end{tikzpicture}$

Richard McElreath 🐈‍⬛@rmcelreath· 13d

For examples of the priorsense package in use, see e.g. chapter 17 case study ("sleep study") from the Bayesian Workflow website: avehtari.github.io/Bayesian-Wor...

figure 10 from the linked page, showing prior sensitivity checks for three parameters

Richard McElreath 🐈‍⬛@rmcelreath· 15d

ok I looked up the different editions and I am pretty sure this is the cover of the used paperback I had. so like 1985. you young people have no idea how cool this book was

Richard McElreath 🐈‍⬛@rmcelreath· 15d

Statistical Rethinking chapter 7 is Odyssey-coded, but so far the movie adaptation has not boosted sales

Mikołaj Kopernik (also known as Nicolaus Copernicus, 1473–1543): Polish astronomer,
ecclesiastical lawyer, and blasphemer. Famous for his heliocentric model of the solar sys-
tem, Kopernik argued for replacing the geocentric model, because the heliocentric model
was more “harmonious.” This position eventually lead (decades later) to Galileo’s famous
disharmony with, and trial by, the Church.
This story has become a fable of science’s triumph over ideology and superstition. But
Kopernik’s justification looks poor to us now, ideology aside. There are two problems: The
model was neither particularly harmonious nor more accurate than the geocentric model.
The Copernican model was very complicated. In fact, it had similar epicycle clutter as the
Ptolemaic model (Figure 7.1). Kopernik had moved the Sun to the center, but since he still
used perfect circles for orbits, he still needed epicycles. And so “harmony” doesn’t quite
describe the model’s appearance. Just like the Ptolemaic model, the Kopernikan model was
effectively a Fourier series, a means of approximating periodic functions. This leads to the
second problem: The heliocentric model made exactly the same predictions as the geocentric
model. Equivalent approximations can be constructed whether the Earth is stationary or
rather moving. So there was no reason to prefer it on the basis of accuracy alone.
Kopernik didn’t appeal just to some vague harmony, though. He also argued for the
superiority of his model on the basis of needing fewer causes: “We thus follow Nature, who
producing nothing in vain or superfluous often prefers to endow one cause with many ef-
fects.”98 And it was true that a heliocentric model required fewer circles and epicycles to
make the same predictions as a geocentric model. In this sense, it was simpler.
Scholars often prefer simpler theories. This preference is sometimes vague—a kind of
aesthetic preference. Other times we retreat to pragmatism, preferring simpler theories be-
cause their si…

Richard McElreath 🐈‍⬛@rmcelreath· 15d

I see now we missed an opportunity by not entitling our book "Bayesian Odyssey" instead of "Bayesian Workflow". The priors are polytropic! Still all the case studies are online avehtari.github.io/Bayesian-Wor... (and python translations arviz-devs.github.io/bayesian-wor...) for your adventure

screenshot of case study chapter 22, black cat adoptions

Richard McElreath 🐈‍⬛@rmcelreath· 16d

I know we are going to get birds bc everyone loves the birds. I wonder though why these birds? None are particularly European. Which is maybe the point? I couldn't find anything online explaining how these species were chosen. (I do not like the reverse design with EU gov buildings. Puke.)

grid of the proposed euro banknotes with birds on one side and various disgusting EU buildings on the reverse

Richard McElreath 🐈‍⬛@rmcelreath· 20d

Mischka cat is in fact the size of a small dog and insists on flopping on my lap. He has not read the Odyssey

Richard McElreath 🐈‍⬛@rmcelreath· 21d

When I am writing, I end up looking up lots of weird details. Today, I looked up who is responsible for the statistical independence symbol: Finnish mathematician Gustav Elfving in 1950s. He used a double orthogonal ⊥⊥, bc independence is stronger than orthogonality. Dawid's 1979 paper squished it.

Gustave Elfving, on right looking smoking hot, thinking about geometry no doubt.

Richard McElreath 🐈‍⬛@rmcelreath· 22d

graph! okay I need to get some work done. this topic triggered my stats lecture mode. it won't happen agian.

curve(1/sqrt(x),from=1,to=100,xlab="n",ylab="1/sqrt(n)",lwd=2)

Richard McElreath 🐈‍⬛@rmcelreath· 22d

Hah yeah. I can see a reason to control for unobserved study effects at times. Would still use pooling though, with Mundlak. Looking at figure in blog, the weights are fine. The tau is just big, because estimates are all over. I worry these estimates not transported to same demography though.

Richard McElreath 🐈‍⬛@rmcelreath· 22d

Yeah I read the blog and the section on random effects doesn't give me confidence the author understands random effects. He'd prefer a fixed-effect meta-analysis, which is certainly a choice. Anyway, I am not confident I understand his point. But agree it's not clear how to pool variable studies.

Let me emphasize: the difficulty of changing the outcomes of the prazosin meta-analyses is an artifact of their statistics, and not because of the strength of evidence for prazosin. The prazosin meta-analyses include mostly small trials with large effects. The standard random-effects meta-analytic approach, in essence, overweighs small studies by presuming that each study reflects a distinct effect in that individual trial.7 In the Zhang et al. (2019) meta-analysis, for instance, the 304-subejct PACT trial was weighted as 15.6% for the nightmare outcome, while the 13-subject cross-over trial (Taylor et al., 2008) was weighted at 13.6%. The approaches of the other meta-analyses are very similar. No one seriously believes that a 13 person cross-over trial should count as much as the PACT trial, and yet, that’s what the meta-analyses argue.

Richard McElreath 🐈‍⬛@rmcelreath· 23d

I spontaneously wrote the sentence "The history of causal inference mirrors the history of the laser" so now I am reading the history of lasers. Just to make sure I haven't hallucinated anything

First paragraph of Albert Einstein's 1917 paper Zur Quantentheorie der Strahlung

Richard McElreath 🐈‍⬛@rmcelreath· 28d

Now that the Bayesian Workflow book is physically in the world, I am spending more time again on 3rd edition of Stat Rethinking. Lots of workflow stuff to weave into it. (Yeah I am still just using latex, you quarto kids get off my lawn.)

Richard McElreath 🐈‍⬛@rmcelreath· 29d

If Haaland wanted a raccoon, he should have come to Leipzig. We have I think 1.2 million of them.

Postcard for the city of Leipzig, showing photos of invasive raccoons and nutria that live and thrive in the city.

Richard McElreath 🐈‍⬛@rmcelreath· Jul 10

sad data trombone

Excerpt from a published paper:

Availability of data and materials

Data will be posted publicly on osf.io upon acceptance of the publication. If editor would like the data as part of review, please let us know.

Richard McElreath 🐈‍⬛@rmcelreath· Jul 9

well not much about stats is intuitive. i guess my post is the scoring rule version of this Taleb cartoon about infinite bread sticks

Richard McElreath 🐈‍⬛@rmcelreath· Jul 8

Cooking with Mischka: Halal chicken and rice. Marinated chicken (yogurt lemon salt garlic), turmeric rice, lettuce and tomato, garlic/mayo dressing. Mischka inspected the chicken personally.

Richard McElreath 🐈‍⬛@rmcelreath· Jul 8

This puzzle is in my first book (the one no one reads), and I know Claude ate my books (bc I got part of that copyright settlement against Anthropic), so here's Claude's answer, from my book. The whole chapter is about this puzzle basically, builds answer using covariance genetics.

3.4.5 What is r?
We’vecomefarenoughnowthatwecanspelloutmoreclearly
what the r in Hamilton’s rule is and is not.
r is not Pr(A|A). It is also easy to see from Equations 3.7
and3.8thatr ismostdefinitelynot thechancethattherecip-
ient has a given allele, given that the donor has it. This com-
mon mis-definition of r arises from a basic misunderstanding.
The recipient and donor can be alike simply because the al-
lele is common in the population (p is large). Instead, r is
the increased similarly of the donor and recipient relative to
the population mean. Only in extreme cases, as when p ≈0
and pi = 1, will r = E(pj|pi).
r is the extent to which actor’s genotype predicts re-
cipients’ genotypes. We showed a fairly standard deriva-
tion of r as a regression coeﬃcient, as Hamilton presented it.
The slopes of the lines in Figure 3.3 are measures of r. In
calculus-speak, r is the partial derivative of pj with respect to
pi, ∂pj/∂pi: the rate of change in recipients’ genotypes as a
function of actors’ genotypes. One important consequence of
this interpretation is that negative relatedness is possible. If
being an altruist predicts that the recipient is less likely than
chancetohavethealtruismallele, theslopeofthebest-fitline
will be negative. You might think this is a weird and unlikely
circumstance, but it might be important in some cases.

Figure 3.3: How positive assortment relates to the regression of
recipient’s genotype (pj) on actor’s genotype (pi). The size of the
white circles represents the number of pairs at each combination
of pi and pj. The black circles in each plot represent the mean
genotype of actors and recipients, which is p in each case. The
dashed lines are the regressions of pj on pi. The slopes of these
lines are the beta coeﬃcients important in Expression 3.6.

Richard McElreath 🐈‍⬛@rmcelreath· Jul 7

Alex Humboldt enters the chat www.friedrich-wilhelm-stollen.de

Richard McElreath 🐈‍⬛@rmcelreath· Jul 4

Food goes in, fur comes out

cat laying on an ottoman with a pile of fur and a brush

Richard McElreath 🐈‍⬛@rmcelreath· Jul 2

I don’t always eat crisps but when

Richard McElreath 🐈‍⬛@rmcelreath· Jun 29

Richard McElreath 🐈‍⬛@rmcelreath· Jun 28

Mischka always finds the place that is a degree cooler than the rest of the house

Richard McElreath 🐈‍⬛@rmcelreath· Jun 27

Siberian tomcat says Здесь слишком жарко

Richard McElreath 🐈‍⬛@rmcelreath· Jun 24

running some experiments now with priors on (ersatz-)R^2. Can really see the global shrinkage pattern, comparing prior on R^2 (left) to conventional independent priors (right). Only 2 "true" effects in this sim, easy to spot them