Stuart Buck
@stuartbuck1
Followers
4,870
Following
775
Media
1,364
Statuses
19,803
Executive Director, Good Science Project Senior Advisor, Social Science Research Council "He's an idiot," Nassim Taleb
Texas
Joined June 2011
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
TLP Finished Shooting
• 691502 Tweets
#ปิ่นภักดิ์ปิดกล้องแล้วจ้า
• 682805 Tweets
Apple
• 674765 Tweets
OpenAI
• 123314 Tweets
Nodal
• 91234 Tweets
Cazzu
• 66108 Tweets
水分補給
• 55374 Tweets
#ปาฏิหาริย์รักร้อยปี
• 55174 Tweets
Ángela Aguilar
• 54443 Tweets
रजत शर्मा
• 54110 Tweets
HAPPY 3RD BINIVERSARY
• 41800 Tweets
Kevin Hart
• 40642 Tweets
Druski
• 39057 Tweets
Oilers
• 38854 Tweets
Panthers
• 37598 Tweets
エアコン
• 27965 Tweets
Edmonton
• 24101 Tweets
Marisol
• 16431 Tweets
Victoria Monet
• 15856 Tweets
Enchanting
• 12909 Tweets
モンスト
• 11710 Tweets
愛の不時着
• 10286 Tweets
Mariners
• 10254 Tweets
How the vast majority of people and organizations think about evidence:
61
1K
4K
This Harvard Business Review chart seems to have been made by someone who didn't know what most of the terms meant:
343
1K
4K
Boss: We need a name for our film studio
Me: Let him go first, he'll copy my idea
1-up Karl: No I promise I won't
Me: Ok my idea is 19th Century Fox
1-up Karl: *looks at camera*
3
262
2K
3
81
3K
Driver: My God… that weasel…
Onlooker: He just went… “pop”…
Weasel’s family: *sobbing*
Ice-cream man: I’ve got an idea for a song y’all.
2
81
645
2
43
2K
It's easy to romanticize the past . . .
My grandfather was born in this farmhouse in Arkansas. Life was so hard that he couldn't afford new shoes--he had to repair old shoes with tire rubber.
He hoped I would grow up to "get an indoors job."
20
57
1K
I've been seeing this story a lot on Twitter today:
Headline: Florida Surgeon General shows that there is an "84% increase in the relative incidence of cardiac-related death among males 18-39 years old within 28 days following mRNA vaccination"
44
242
776
So much misinformation out there. The only way to determine causality is with so-called "p values."
* If the p value is under 0.05, the relationship is causal.
*If over 0.05, you're looking at correlation not causation.
This is a common misconception I see a lot in my intro econometrics class. To detect causality in regressions you actually need to look at the *adjusted* R-squared, since the regular R-squared always increases with more controls. Hope this helps!
48
23
450
40
30
673
@realchrisrufo
@soncharm
Seems very doubtful that 1st grade teachers are telling kids that all white people are born evil. At most, what might have happened is that the teacher said something about how "white people enslaved black people centuries ago" (true), and the 6-year-old extrapolated from that.
96
25
626
@SarahTheHaider
Maybe it's not strictly a fallacy in and of itself, but if people start making a slippery slope argument, they will be more likely to make worse and worse arguments and before you know it, they will be guilty of innumerable logical fallacies.
17
7
490
I'm not sure how it's the case that London is an expensive place to live, yet every job like this seems to pay half or a third of an equivalent job in the US.
The head of pandemic preparedness in the UK will literally make less than most random 24 year olds I met when I was in NYC lol
18
68
1K
33
12
460
Great, just great, now we need to stay 6 feet away from tigers too.
"Tiger at Bronx Zoo in New York City tests positive for coronavirus" via
@abc7
23
110
434
It’s time to publicly announce something I’ve been working on for quite some time: the Good Science Project.
16
81
441
@the_dugster
@KelseyTuoc
@teapottery_
Indeed (see screenshot).
About 10 years ago at the Arnold Foundation, I suggested to a colleague that we sponsor the biggest/best RCT ever as to fingerprints. [cont]
2
15
435
Given what apparently triggered a key dispute on the OpenAI board, it's interesting to recall that none other than Larry Summers once told Elizabeth Warren about a key distinction: Insider power vs. outsider power.
From her memoir:
16
39
368
@MJRosenbergDad
@mattyglesias
He didn't take issue with those beliefs per se; he took issue with the idea that making everyone write out such anodyne statements actually makes any difference in the real world.
9
1
354
But it's one of the worst studies I've ever seen. It doesn't show what you might think, not even close.
Here's why.
6
55
350
@the_dugster
@KelseyTuoc
@teapottery_
Give a bunch of examiners an incomplete print, and then randomize them to compare it to 1 of 2 databases (1 that contains the real print and 1 that doesn't).
Overall goal is to see whether they differ in rates of finding a "match."
But then also randomize into further groups.
4
9
297
Unintended consequences strike again:
18
67
282
Not very plausible that Harvard would investigate a superstar researcher, write a 1000÷ page report, and suspend that person, just because of internet outrage over problematic data. Universities err in the opposite direction all the time, and a superstar well-funded researcher
Regardless of the underlying cases,
@lessig
makes good points about universities caving to internet outrage here.
16
0
8
5
17
281
@the_dugster
@KelseyTuoc
@teapottery_
In one subgroup, examiners are fed false information that there is additional confirmation that the guy really did the crime. Goal: See whether additional info like that can bias the fingerprint examiner towards finding a "match."
1
4
270
@GaryMarcus
@timnitGebru
Many in the field of "AI ethics" are discrediting themselves on a daily basis, unfortunately.
3
8
261
@the_dugster
@KelseyTuoc
@teapottery_
You can imagine a number of other subgroups in which examiners are tested for bias in one direction or another.
So, a huge multi-stage RCT that tests for false positives and for many possible biases. Ought to happen.
2
4
247
In other words, they *only* looked at people who literally died within 25 weeks of a Covid vaccine.
This means that "rate of cardiac death" really means something like, "of the few people that died, what proportion of deaths were cardiac-related?"
1
23
243
1) Terrible mismatched chart from Case/Deaton
2) Roughly the same chart with scale fixed for white/black deaths
17
151
217
Just released today: a short piece I wrote for Nature.
5
51
212
@Acyn
Longest-serving professor in the history of MIT?
He was a professor from 1936 to 1973--some 37 years.
Gilbert Strang spent 61 years on the MIT faculty.
Heather Lechtman has been at MIT since 1974:
19
31
214
Found all the talk of neural nets impenetrable, but on looking into the actual math, it looks like computer scientists managed to reinvent (multinomial) logistic regression (+optimizing the loss function) and renamed all the traditional statistics terms for no reason?
19
19
199
There's a tendency in the effective altruism movement (most egregiously with FTX) to give lots of money, power or responsibility to youngish people who are really smart and well-meaning, and to worry less about who actually has experience with boring issues like public
16
24
205
A dismal success rate at arguably the single most important task that police have:
Some details from my first investigation with
@CBSNews
:
Cops in the US only solve about HALF of murders in the US. It's the lowest rate ever recorded.
1K
5K
22K
18
56
200
Rings true for every parent, but note that this unbelievable level of self-determination may help explain why humans learn so many skills in the first few years of life.
I like having fellow toddler moms as friends because I’ll text them like “sorry, we are going to be 25 minutes late to the park, we saw an ant.” And they’re like “no problem we’re gonna be 30 bc he had to put on his own shoes.”
130
4K
94K
4
7
202
The latest "plagiarism" case is absurd. All of the passages involve routine descriptions of how a survey was constructed.
Are scholars supposed to invent new wording when they quote the survey questions? Are they supposed to come up with new ways to say "Methods and Data"?
13
28
199
Imagine that before vaccination, 100 people die in 25 weeks, and 20% are cardiac. After vaccination, 40 people die in 25 weeks, and 25% are cardiac. Even though cardiac deaths went down, this sort of study would say that cardiac risk increased.
2
13
197
Another mischievous idea:
Give nutritional epidemiologists a dataset that is real except for 2 variables that are random noise.
One is labeled "blueberries" and the other "red meat."
Ask them to estimate the effect on mortality, disease, etc.
Then sit back and watch.
8
16
190
I was going to write up a response to this PNAS piece: Redish et al., “Reproducibility failures are essential to scientific inquiry.”
5
101
179
@Chelsea_Fagan
Yep. Looks like Elon is just going out of his way to have as many children as possible?
5
0
176
In any event, if you select just a group of people (however small) who got vaccinated and then died in the next 6 months, you can't say anything whatsoever about their risk of death compared to the overall population.
1
15
169
@Meelsie143
@JackTierney30
@RapnCap
@bitchardthe3rd
@uncoolboyfriend
"Bon Iver! Rock on, man, I love 'Living on a Prayer.'"
1
5
163
Which I did.
Am I happier than him? Hard to say -- incommensurability of personal experiences, hedonic adaptation, etc.
But with my allergies, I suspect I'd feel like him if I had to spend all day in the dust, wind, and sun, cleaning out barns with animal excrement, etc.
6
3
160
Enjoyed Katalin Kariko's memoir, especially when this surprise popped up towards the end!
5
25
161
The way the headlines are written, you might think someone measured a baseline rate of cardiac deaths (let's say, prior to Covid or vaccines), and that this rate spikes up by 84% for young males after an mRNA vaccine, as compared to unvaccinated people.
1
9
152
Second, they exclude anyone who had Covid or died from Covid!
"Individuals were excluded if they (1) had a documented COVID-19 infection, (2) experienced a COVID-19 associated death..."
1
11
154
A number of studies have shown that smarter and more well-informed people are *more* skillful at coming up with reasons to ignore evidence they don't like.
4
28
140
@MommaMonkey5
@BradMossEsq
@RadioFreeTom
That's nothing. In Katy (where I lived), voters enacted a tax for a $70 million high school football stadium.
18
18
130
That is not what happened *at all.*
If you read the actual study, such as it is, it seems they did something very different.
4
11
132
@Meelsie143
@JackTierney30
@RapnCap
@bitchardthe3rd
@uncoolboyfriend
Just say you get Bon Iver confused with Bon Jovi sometimes.
1
0
129
PS: when my grandmother was growing up in a sharecropping family, she spent some time living in the corner of a chicken house. She loved Christmas because she would get an orange. Had to fetch water from a mile away. She agreed with him: "Get an education and an indoor job."
3
3
130
Note that even if the overall death rate is *lower* after vaccination and even if *cardiac* deaths are lower, the cardiac portion of the actual deaths could go up.
1
7
130
No more small crappy trials! My favorite moment so far from the NIH Advisory Committee to the Director Meeting from June 2023 (around 47 minutes):
Larry Tabak: "We'll have to have the discipline to stop supporting underpowered boutique studies, which quite frankly, many
3
27
128
@BenShindel
@SimoneHCollins
Exactly. Imagine reading that article and thinking that weightlifting reduces life expectancy by 3 times the amount from being a lifelong smoker.
4
1
119
It is a testament to how little we know about metascience that two of the most prominent suggestions are:
1) Fund the person without regard for what the project is;
2) Fund the project without regard for who the person is (that is, anonymize grant applications).
6
8
122
@zucchero
@SaveYourSons
One way you can tell this person has never been anywhere near a farm is the idea that after a day working 100 acres, anyone would have the time, energy, or need for resistance training.
3
3
119
"First analysis of ‘pre-registered’ studies shows sharp rise in null findings"
3
51
117
This is the kind of person academia should hire, not people who suck up to frauds.
Yes... I had to censor my work without which my committee wouldn't sign it... But the great thing with Internet is that you can post the original version and even add a disclaimer explaining why it differs from the published version…
37
351
2K
2
16
113
Proud to say that we're giving $2 million to COVID research via .
Hat tip to
@patrickc
for reaching out.
0
15
110
It is enormously difficult to learn basic motor skills (like tying one's shoes or putting on a coat) that we later take for granted (robots can't remotely come close).
We benefit from an innate drive to figure out how to do everything for our 3-year-old selves.
1
5
111
Plus, there was no "internet outrage" in the first place until *after* Harvard suspended her and Data Colada finally released their posts. When Harvard launched and conducted the investigation, exactly zero people on the Internet knew Gino was under suspicion.
1
1
107
Even if they were measuring the cardiac death risk correctly, it could be possible that a reduction in Covid deaths outweighs cardiac deaths. But if you exclude people who were infected with or died from Covid, it is impossible to make that comparison.
2
7
108
I think the single most important criterion for whether a scientific field is meaningfully rigorous is whether null/negative results are regularly published:
2
31
105
First, note who is in the study:
"For the primary analysis, Florida residents aged 18 years or older who died within 25-weeks of COVID-19 vaccination since the start of the vaccination roll-out (December 15, 2020) were included."
1
9
106
I actually hope President Trump spends at least the next year (full-time) trying every possible tactic in order to purchase Greenland.
7
13
102
This study is comically bad. They split the sample into 24 groups and, as expected, find one that has a p<.05 positive effect. Even dumber is the method: there's no control group. They just measure whether deaths within 25 weeks after vax are concentrated in the first 28 days.
32
317
2K
4
8
104
Third, and perhaps most weirdly, they follow people for 25 weeks after a Covid vaccination, and say that they're interested in a risk period of 28 days after the vaccine.
Why follow people for the other 21 weeks? Because that is the BASELINE PERIOD. (!!!!)
1
6
100
@NAChristakis
@jflier
Please don't say "trigger warning" without advance notice.
The word "trigger" reminds me of guns, and that is very upsetting.
5
5
107
The best performers (sports, running, chess, music, even rock climbing) usually have trainers/coaches who specialize in optimizing their work, thinking of new ways to train, etc.
Question: Would it make sense to have the same sort of trainers/coaches for top scientists?
15
9
94
A new piece I published today:
5
25
95
The overall death rate (and the cardiac death rate) could all be lowered compared to BEFORE vaccination (the only sensible baseline period), but if the rate was higher for weeks 1 to 4 than in weeks 5 to 25, this method would make it look like vaccination was increasing risk.
1
6
91
The irony is that most people think that they themselves fairly assess the evidence, while it's only other people that do the above.
3
3
83
One thing about philanthropy that has bothered me for about a decade is how inefficient, bespoke, and ad hoc it is, especially when it comes to high-net-worth individuals and family offices.
Lots of people who have more money than they could ever spend on themselves, and who
5
16
88
An announcement:
After 9 years, I’m leaving Arnold Ventures this month (although I'll still do some consulting for them).
Excited for several opportunities, but if you know of anything else I should be considering, let me know soon! My personal email is stuartbuck
@gmail
.com
3
4
87
@Andercot
@DanielleFong
What explains the extraordinarily unusual accusation of dishonesty? Even in actual cases of dishonesty (e.g., embezzling), corporate boards usually opt to be discreet and circumspect for legal reasons.
2
2
84
Here's a piece I just published on scientific innovation and reproducibility:
Via the new online magazine Works in Progress, which is well worth reading.
7
25
83
How on earth is the BASELINE PERIOD really just weeks 5 to 25 after vaccination?
1
4
83
@LHSummers
@tylercowen
Why? The whole point of redoing LaGuardia was that everyone was sick of it being absurdly crowded...no place to sit, nothing to eat except deep fried pretzels. So this seems like asking, "why, in seeking to make it bigger and less crowded, did they make the airport bigger?"
3
2
83
Just came across a brilliant explanation of why (nearly) everything in government goes wrong over time:
6
19
83
2) Uncovering this kind of fraud is tedious and thankless work. Right now, only a handful of folks have the independent means to do so.
Funders (gov. and philanthropy) should fund a permanent organization of "Data Thugs."
3
15
82
Hot take: If Robert Moses had been a true visionary as to public works, he would have built a dam and created land in *at least* all the areas shown below, so as to avoid the need for as many bridges and tunnels
6
2
81
Assuming the timestamps here mean what I think they mean, one of the most viral tweets about today's Nobel prize was pretty obviously plagiarized.
7:55 am on the left; 7:59 am on the right.
7
13
77
Medical journals are odd to me. Either they publish rigorous randomized trials or else simplistic correlation-is-causation stuff, but not as much that's in-between (such as regression discontinuity).
10
7
81
@GregoryFaletto
Yes, and also, if the p value is under .05, that means there is a 95% chance that everything in your article is absolute truth.
2
0
77
Even recipe videos are having a reproducibility crisis:
I tried out some "fake bakes" that have clocked up millions of views on YouTube, but don't actually work! featuring
@howtocookthat
291
38K
51K
1
35
77
1
2
72
It shouldn't be "contrarian" to point out the fact that throwing more money at a problem isn't the same as solving the problem, and that we ought to put more thought into the best ways of spending that money.
4
6
76
@GoBlueChung
@ronawang
Right, it's hard to know whether this anecdote is an example of obnoxious mansplaining, or a case of the stars aligning over shared interests.
3
1
77
Lots of interesting facts in the New Yorker article on Francesca Gino:
2
20
76
Love these thoughts from Michael Stonebraker (the famous computer scientist), suggesting that academia start limiting the number of papers that count towards jobs/tenure:
5
9
75
My friend Vinay points to a major problem for the idea that AI is going to improve medicine or science anytime soon:
Too many papers are low quality, irreproducible, overestimates (esp in cancer), lacking in detail, and/or sometimes outright fraudulent.
You want your oncologist to read 5000 papers without control arms, with selection bias, with no ability to infer causality. Because you want your oncologist to be a total moron, just like people in tech who don't know anything about oncology telling you what you want.
25
79
676
9
12
74