Author of Probably Overthinking It, Think Python, and Think Bayes. Emeritus Prof at Olin College, consultant with PyMC Labs.
Boston, MA
Joined April 2013
- Tweets 8,063
- Following 1,021
- Followers 17,253
- Likes 13,718
Pinned Tweet
My copy of Probably Overthinking It has arrived!
It should be shipping soon. If you would like to pre-order, you can get 30% from University of Chicago Press. Use the code UCPNEW.
press.uchicago.edu/ucp/books…
“I’m way closer to LeBron James than you are to me.”
-- Brian Scalabrine
He's probably right, because in a lognormal distribution of ability, it's levels to this …
allendowney.com/blog/2025/11…
This is really good.
I would rather buy a new 1999 Toyota Camry than any car made after 2010.
piped.video/HauQtcj7UTM
The newest chapter of Think Linear Algebra up now!
It is about least squares regression, QR decomposition, and orthogonality:
allendowney.github.io/ThinkL…
4. And if 5-year survival increases over time, it is tempting to conclude that treatment has improved.
In fact, none of these inferences are correct.
allendowney.com/blog/2025/10…
2. Looking at the difference in survival between early and late detection, it is tempting to conclude that more screening would save lives.
3. Looking at the difference in survival between early and late detection, it is tempting to conclude that more screening would save lives.
Five-year survival might be the most misleading statistic in medicine.
Even smart people can make incorrect inferences. Here are the top four:
1. If a patient is diagnosed early, it is tempting to think the probability is 91% that they will survive five years after diagnosis.
This is very good.
piped.video/watch?v=S-UqYUfv…
The original sin of the engineering curriculum is the Foundation Fallacy: math (especially calculus) and science (especially physics) are (1) the foundations of engineering, and therefore (2) the prerequisites of engineering education.
allendowney.com/blog/2025/10…
I just posted a new chapter of Think Linear Algebra.
It's about projection, rejection, rotation, and pool!
allendowney.github.io/ThinkL…
It looks like they did -- if I include missing data in two states, and treat it as zeros, that explains the flat line fit when we combine the groups (incorrectly).
I just gets worse.
Before I posted this, I didn't notice the zeros in the lower left. Were they included in the analysis? I hope not.
I love a good Simpson's paradox. Sadly, this is not one of them: allendowney.com/blog/2025/10…
In fact, I think the whole paper is nonsense.
Published in Nature, too.
This video intro to network science is excellent --
and if you like it, you might like Think Complexity, too
Chap 3 Watt Strogatz networks
Chap 4 Barabasi Albert networks
Chap 12 evolution of cooperation
piped.video/watch?v=CYlon2tv…
Hey, on Sunday morning I heard about 180 million people scream "Yes! Yes you are!" at the same time.
Does anyone know what happened?
Sometimes we can use Bayesian methods to infer the effect of selection bias and produce an unbiased estimate.
Here's an example that uses PyMC to solve a classic probability puzzle (the image shows what I think is the original version from 1958).
allendowney.com/blog/2025/09…
