daniel saunders
@danielsaunders.bsky.social
writing about methods models and stats in evolutionary social sciences.
Reposted by daniel saunders
It’s become fashionable in some circles to reject decision theory (and other basic statistical ideas) for vaguely political reasons.
There are valid critiques—but also some worth being wary of.
Some thoughts:
statmodeling.stat.columbia.edu/2025/10/17/s...
There are valid critiques—but also some worth being wary of.
Some thoughts:
statmodeling.stat.columbia.edu/2025/10/17/s...
Separating the whack from the chaff in critiques of decision theory | Statistical Modeling, Causal Inference, and Social Science
statmodeling.stat.columbia.edu
October 17, 2025 at 4:21 PM
It’s become fashionable in some circles to reject decision theory (and other basic statistical ideas) for vaguely political reasons.
There are valid critiques—but also some worth being wary of.
Some thoughts:
statmodeling.stat.columbia.edu/2025/10/17/s...
There are valid critiques—but also some worth being wary of.
Some thoughts:
statmodeling.stat.columbia.edu/2025/10/17/s...
Reposted by daniel saunders
Are you a phd student? Are you in the vicinity of one and want to help them with their bad life choices? Then this is for you! #philsci
✨Don't forget✨ PSA Office Hours are back! Join us this Thursday, October 9 at 12 PM EST with S. Andrew Schroeder. Sign up at the link below to save your spot!
www.philsci.org/psa_...
www.philsci.org/psa_...
October 6, 2025 at 7:57 PM
Are you a phd student? Are you in the vicinity of one and want to help them with their bad life choices? Then this is for you! #philsci
It’s an amazing package. The website also has a bunch of neat strategies to diagnose hmc samplers, ones I haven’t seen discussed elsewhere.
🥧 nutpie got a website now! pymc-devs.github.io/nutpie/
If you're doing Bayesian inference with PyMC or Stan, this might be worth checking out. Nutpie can sample PyMC and Stan model, and typically twice as fast.
#BayesianStats #PyMC #Stan
If you're doing Bayesian inference with PyMC or Stan, this might be worth checking out. Nutpie can sample PyMC and Stan model, and typically twice as fast.
#BayesianStats #PyMC #Stan
Nutpie
pymc-devs.github.io
June 4, 2025 at 4:33 AM
It’s an amazing package. The website also has a bunch of neat strategies to diagnose hmc samplers, ones I haven’t seen discussed elsewhere.
Reposted by daniel saunders
“We didn’t ever hide that that’s what it was. People were mad because we were calling them effects,” she says. “Then they say to us, but they’re just associations with 20 covariates. But the point is we said that from the beginning. They’re associations with 20 covariates.”
May 22, 2025 at 6:18 PM
“We didn’t ever hide that that’s what it was. People were mad because we were calling them effects,” she says. “Then they say to us, but they’re just associations with 20 covariates. But the point is we said that from the beginning. They’re associations with 20 covariates.”
Reposted by daniel saunders
Friend telling me about how few statisticians are actually Bayesians. It's a real shame how far we have fallen from God's light.
May 22, 2025 at 7:07 AM
Friend telling me about how few statisticians are actually Bayesians. It's a real shame how far we have fallen from God's light.
Reposted by daniel saunders
Mitzi Morris's new case study mc-stan.org/learn-stan/c... illustrates with hierarchical and spatial models the better efficiency of the new sum_to_zero_vector constrained parameter introduced in Stan 2.36 (2024-12). Mitzi used CmdStanPy, but the Stan code is the same with all interfaces
The Sum-to-Zero Constraint in Stan
mc-stan.org
May 5, 2025 at 4:17 PM
Mitzi Morris's new case study mc-stan.org/learn-stan/c... illustrates with hierarchical and spatial models the better efficiency of the new sum_to_zero_vector constrained parameter introduced in Stan 2.36 (2024-12). Mitzi used CmdStanPy, but the Stan code is the same with all interfaces
A brief history of tech disruptions:
2010: We are going to disrupt that horrible corporation, Blockbuster
2012: We are going to disrupt those awful cab monopolies
2014: We are going to disrupt the record labels
2022: We are going to disrupt reading and writing
2010: We are going to disrupt that horrible corporation, Blockbuster
2012: We are going to disrupt those awful cab monopolies
2014: We are going to disrupt the record labels
2022: We are going to disrupt reading and writing
January 15, 2025 at 10:40 PM
A brief history of tech disruptions:
2010: We are going to disrupt that horrible corporation, Blockbuster
2012: We are going to disrupt those awful cab monopolies
2014: We are going to disrupt the record labels
2022: We are going to disrupt reading and writing
2010: We are going to disrupt that horrible corporation, Blockbuster
2012: We are going to disrupt those awful cab monopolies
2014: We are going to disrupt the record labels
2022: We are going to disrupt reading and writing
Feels extremely true, watching how business leadership reacts to the long search process involved in finding a good model. The bayesian workflow literature largely assumes academic contexts where research papers are expected to be in development for a year or more.
PPLs have struggled to gain traction in industry. Conventional wisdom blames scaling. I argue that PPLs' challenges aren't about scaling at all. They're about learning. And sometimes, to go faster, we need to slow down.
heresy.ai/a-better-ppl/
#bayesian #machinelearning
heresy.ai/a-better-ppl/
#bayesian #machinelearning
Making PPLs More Useful With Two New Operators | Heresy
Probabilistic programming is a counter play to black box machine learning. Probabilistic programming practitioners seek to build interpretable models of phenomena and to captu…
heresy.ai
December 30, 2024 at 12:16 AM
Feels extremely true, watching how business leadership reacts to the long search process involved in finding a good model. The bayesian workflow literature largely assumes academic contexts where research papers are expected to be in development for a year or more.
Reposted by daniel saunders
🎄✨ 𝐌𝐞𝐫𝐫𝐲 𝐂𝐡𝐫𝐢𝐬𝐭𝐦𝐚𝐬 𝐚𝐧𝐝 𝐇𝐚𝐩𝐩𝐲 𝐇𝐨𝐥𝐢𝐝𝐚𝐲𝐬 𝐟𝐫𝐨𝐦 𝐏𝐲𝐌𝐂 𝐋𝐚𝐛𝐬!
🎁 This holiday season, we want to thank everyone in our community for your support and enthusiasm. We’re grateful to see so many of you using PyMC-Marketing and CausalPy
#MerryChristmas #HappyNewYear #PyMCMarketing #CausalPy #Gratitude
🎁 This holiday season, we want to thank everyone in our community for your support and enthusiasm. We’re grateful to see so many of you using PyMC-Marketing and CausalPy
#MerryChristmas #HappyNewYear #PyMCMarketing #CausalPy #Gratitude
December 24, 2024 at 4:23 PM
🎄✨ 𝐌𝐞𝐫𝐫𝐲 𝐂𝐡𝐫𝐢𝐬𝐭𝐦𝐚𝐬 𝐚𝐧𝐝 𝐇𝐚𝐩𝐩𝐲 𝐇𝐨𝐥𝐢𝐝𝐚𝐲𝐬 𝐟𝐫𝐨𝐦 𝐏𝐲𝐌𝐂 𝐋𝐚𝐛𝐬!
🎁 This holiday season, we want to thank everyone in our community for your support and enthusiasm. We’re grateful to see so many of you using PyMC-Marketing and CausalPy
#MerryChristmas #HappyNewYear #PyMCMarketing #CausalPy #Gratitude
🎁 This holiday season, we want to thank everyone in our community for your support and enthusiasm. We’re grateful to see so many of you using PyMC-Marketing and CausalPy
#MerryChristmas #HappyNewYear #PyMCMarketing #CausalPy #Gratitude
This piece deftly puts words to my frustration with the way we talk about AI. It’s about education but it feels apt in business, software, etc
mail.cyberneticforests.com/how-does-ope...
mail.cyberneticforests.com/how-does-ope...
How Does OpenAI Imagine K-12 Education?
Close Reading OpenAI's training module for educators
If you’re taking a free online training, it's helpful to understand who wrote that lesson plan and why. ChatGPT Foundations for Educators is a cou...
mail.cyberneticforests.com
November 29, 2024 at 2:45 AM
This piece deftly puts words to my frustration with the way we talk about AI. It’s about education but it feels apt in business, software, etc
mail.cyberneticforests.com/how-does-ope...
mail.cyberneticforests.com/how-does-ope...
briefly looking up empirical papers, giving up and resorting to nature documentaries is just ... extremely philosophy.
October 21, 2024 at 12:23 AM
briefly looking up empirical papers, giving up and resorting to nature documentaries is just ... extremely philosophy.
Reposted by daniel saunders
Evolution of Similarity-Biased Social Learning (by me + Alejandro Pérez Velilla). Now in press at Evolutionary Human Sciences.
Formalizes the long-standing idea that, for better or worse, it may be adaptive to ignore or down-weight information from outgroup sources. osf.io/preprints/so...
Formalizes the long-standing idea that, for better or worse, it may be adaptive to ignore or down-weight information from outgroup sources. osf.io/preprints/so...
Evolution of Similarity-Biased Social Learning.
New preprint with Alejandro Perez Velilla. A long time in the making. Feedback welcome!
Here’s a short summary thread.
osf.io/preprints/so...
New preprint with Alejandro Perez Velilla. A long time in the making. Feedback welcome!
Here’s a short summary thread.
osf.io/preprints/so...
October 2, 2024 at 7:59 PM
Evolution of Similarity-Biased Social Learning (by me + Alejandro Pérez Velilla). Now in press at Evolutionary Human Sciences.
Formalizes the long-standing idea that, for better or worse, it may be adaptive to ignore or down-weight information from outgroup sources. osf.io/preprints/so...
Formalizes the long-standing idea that, for better or worse, it may be adaptive to ignore or down-weight information from outgroup sources. osf.io/preprints/so...
It seems Ludwig Boltzmann had a bit of a drinking problem.
September 6, 2024 at 11:22 AM
It seems Ludwig Boltzmann had a bit of a drinking problem.
Reposted by daniel saunders
New Deep Dive on Splines and Hierarchical Splines for modelling Insurance Loss curves with Bambi/PyMC.
The focus is on the contrast between interpolation, extrapolation and how including extra hierarchical structure aids generalisation.
nathanielf.github.io/posts/post-w...
The focus is on the contrast between interpolation, extrapolation and how including extra hierarchical structure aids generalisation.
nathanielf.github.io/posts/post-w...
June 13, 2024 at 4:23 PM
New Deep Dive on Splines and Hierarchical Splines for modelling Insurance Loss curves with Bambi/PyMC.
The focus is on the contrast between interpolation, extrapolation and how including extra hierarchical structure aids generalisation.
nathanielf.github.io/posts/post-w...
The focus is on the contrast between interpolation, extrapolation and how including extra hierarchical structure aids generalisation.
nathanielf.github.io/posts/post-w...
Reposted by daniel saunders
This is an excellent (very short!) discussion of how to decide which methods to use.
(How can there be so many snappy and highly relevant pieces by Gelman et al. that I haven’t read?!)
(How can there be so many snappy and highly relevant pieces by Gelman et al. that I haven’t read?!)
April 14, 2024 at 3:41 AM
This is an excellent (very short!) discussion of how to decide which methods to use.
(How can there be so many snappy and highly relevant pieces by Gelman et al. that I haven’t read?!)
(How can there be so many snappy and highly relevant pieces by Gelman et al. that I haven’t read?!)
I picked up the dialectical biologist at a used bookstore a couple weeks ago. It slaps. Includes a whole chapter about pranks they played on EO Wilson along with this strategy for riches and professional success.
March 28, 2024 at 8:06 PM
I picked up the dialectical biologist at a used bookstore a couple weeks ago. It slaps. Includes a whole chapter about pranks they played on EO Wilson along with this strategy for riches and professional success.
I've been teaching a version of this class for a couple years. The hard tradeoff is how much of the class should be spent on practical skills vs explaining why common practice in journals is bananas. Students like the class more when there are more dunks obviously!
Every time I look into one of the standard stats textbooks for psychologists I get so angry I want to write one myself. It would be all talking about data-generating mechanisms, and also lots of meta-talk about common practices, how to make sense of them, where they go wrong, etc.
March 11, 2024 at 10:58 PM
I've been teaching a version of this class for a couple years. The hard tradeoff is how much of the class should be spent on practical skills vs explaining why common practice in journals is bananas. Students like the class more when there are more dunks obviously!
With blackjax, nutpie, numpyro, it compiles for 60 seconds. Then the progress bar goes from 0 to 100% in the blink of an eye. I'm telling you, people enjoy horse races more than formula 1 for a reason.
March 1, 2024 at 12:34 AM
With blackjax, nutpie, numpyro, it compiles for 60 seconds. Then the progress bar goes from 0 to 100% in the blink of an eye. I'm telling you, people enjoy horse races more than formula 1 for a reason.
I'm starting to worry that the advanced compiler-NUTS samplers that have come out in the last couple years have taken the thrill out of bayes. In pymc3, I'd stare at the progress bar with glean and horror. Sometimes, it would get abruptly faster for no reason. Sometimes, it would stop.
March 1, 2024 at 12:30 AM
I'm starting to worry that the advanced compiler-NUTS samplers that have come out in the last couple years have taken the thrill out of bayes. In pymc3, I'd stare at the progress bar with glean and horror. Sometimes, it would get abruptly faster for no reason. Sometimes, it would stop.
Dropping by a Bayesian ecology conference today to see what they are doing. Apparently, it is strapping an accelerometer onto wild sharks. Then fitting hidden markov models to classify the sharks into "resting", "methodical exploration", and "rapid exploration".
January 22, 2024 at 10:39 PM
Dropping by a Bayesian ecology conference today to see what they are doing. Apparently, it is strapping an accelerometer onto wild sharks. Then fitting hidden markov models to classify the sharks into "resting", "methodical exploration", and "rapid exploration".
Reposted by daniel saunders
Curious about how probability distributions transform? Ever wondered from where the term “marginal probability” comes? Confused by those “Jacobian” things? Have I got some writing for you.
January 15, 2024 at 3:02 PM
Curious about how probability distributions transform? Ever wondered from where the term “marginal probability” comes? Confused by those “Jacobian” things? Have I got some writing for you.
Nancy Cartwright is so impressive. She has a MasterClass now. Even more impressive its on a subject not in her AOS.
December 18, 2023 at 7:16 PM
Nancy Cartwright is so impressive. She has a MasterClass now. Even more impressive its on a subject not in her AOS.
Reposted by daniel saunders
When preferences change: www.enlightenmenteconomics.com/blog/index.p...
December 17, 2023 at 9:40 AM
When preferences change: www.enlightenmenteconomics.com/blog/index.p...
when u open the assignment to see those overleaf-leave-space-for-the-reader-to-draw-graphs-sized margins and the default-I-know-my-IQ-score-type font, u can just assign them an A and move onto the next one.
December 13, 2023 at 7:18 PM
when u open the assignment to see those overleaf-leave-space-for-the-reader-to-draw-graphs-sized margins and the default-I-know-my-IQ-score-type font, u can just assign them an A and move onto the next one.
Reposted by daniel saunders
A great-looking book on diff kinds of modelling of social behaviour. The mammoth task of translating these models into empirically solid causal claims receives frustratingly less attention than the process of building these models. Methodology of modelling is there, methodology of translation isn’t.
I wrote a post about modeling (and my book) for the Code Horizons blog.
codehorizons.com/what-are-mod...
codehorizons.com/what-are-mod...
What Are Models and Why Should We Use Them to Understand Social Behavior? | Code Horizons
Learn about instructor Paul Smaldino new book: Modeling Social Behavior: Mathematical and Agent-Based Models of Social Dynamics and Cultural Evolution.
codehorizons.com
December 8, 2023 at 5:00 PM
A great-looking book on diff kinds of modelling of social behaviour. The mammoth task of translating these models into empirically solid causal claims receives frustratingly less attention than the process of building these models. Methodology of modelling is there, methodology of translation isn’t.