alex hayes
@alexpghayes.com
postdoc @ stanford econ + incoming assistant prof @ oregon state statistics. networks, causal inference, contagion, measurement error, #rstats. he/him
https://www.alexpghayes.com
https://www.alexpghayes.com
In a shocking turns of events, making a small change to my simulations did in fact fix everything
January 13, 2026 at 12:20 AM
In a shocking turns of events, making a small change to my simulations did in fact fix everything
Also it's not simple! I think cognitive aspects of learning the new semi-parametric methods is a fairly serious constraint for applied use
January 1, 2026 at 7:06 PM
Also it's not simple! I think cognitive aspects of learning the new semi-parametric methods is a fairly serious constraint for applied use
Yeah I thought epi used sequential G things? Also, I'm not sure I think of front door as all that morally different from IV
January 1, 2026 at 7:04 PM
Yeah I thought epi used sequential G things? Also, I'm not sure I think of front door as all that morally different from IV
You get a free license through UW
December 31, 2025 at 6:15 PM
You get a free license through UW
@karlrohe.bsky.social both this and the mathematica mcp bit higher up the thread
December 31, 2025 at 5:43 AM
@karlrohe.bsky.social both this and the mathematica mcp bit higher up the thread
I've pulled out of several projects but far before any manuscript writing began
December 23, 2025 at 5:36 AM
I've pulled out of several projects but far before any manuscript writing began
I fairly agnostic about whether or not service is a good thing; think there are many reasonable choices. I'm a lot more interested in how statisticians talk about that choice, and other disciplines respond. The choice has consequences, whatever it is
December 16, 2025 at 12:02 AM
I fairly agnostic about whether or not service is a good thing; think there are many reasonable choices. I'm a lot more interested in how statisticians talk about that choice, and other disciplines respond. The choice has consequences, whatever it is
I just finished @gelliottmorris.com's delightful Strength in Numbers and I'm excited to read more about opinion research and polling
Let me know if you have recommendations for followup reading!
Let me know if you have recommendations for followup reading!
December 8, 2025 at 10:11 PM
I just finished @gelliottmorris.com's delightful Strength in Numbers and I'm excited to read more about opinion research and polling
Let me know if you have recommendations for followup reading!
Let me know if you have recommendations for followup reading!
Ah I must have been thinking of www.degruyterbrill.com/document/doi... without recalling the paper
AUC-Maximizing Ensembles through Metalearning
Area Under the ROC Curve (AUC) is often used to measure the performance of an estimator in binary classification problems. An AUC-maximizing classifier can have significant advantages in cases where r...
www.degruyterbrill.com
December 3, 2025 at 2:57 AM
Ah I must have been thinking of www.degruyterbrill.com/document/doi... without recalling the paper
I wonder if there's a way to develop a targeted estimator for ranking observations by P(Y=1|X) via the TMLE machinery. @sherrirose.bsky.social do you know of any work like this?
December 3, 2025 at 2:10 AM
I wonder if there's a way to develop a targeted estimator for ranking observations by P(Y=1|X) via the TMLE machinery. @sherrirose.bsky.social do you know of any work like this?
I suppose in neural net land there's also directly minimizing the (smoothed) AUCROC, but that's less generic since you can't always modify your loss function
December 3, 2025 at 1:56 AM
I suppose in neural net land there's also directly minimizing the (smoothed) AUCROC, but that's less generic since you can't always modify your loss function
I've been seeing his name around the R community for years, and then just the other day I had an issue with {ivreg} and he very kindly helped me out. I'm saddened to learn of his passing.
December 3, 2025 at 1:54 AM
I've been seeing his name around the R community for years, and then just the other day I had an issue with {ivreg} and he very kindly helped me out. I'm saddened to learn of his passing.
If you were going to handle a ranking problem, what kinds of tools would you reach for? Under/over-sampling or something else?
December 3, 2025 at 1:45 AM
If you were going to handle a ranking problem, what kinds of tools would you reach for? Under/over-sampling or something else?
My impression from brief stinks in industry was that folks primarily up/down-sampled because they wanted a system good at ordering samples by P(Y=1|X), and that beyond ranking having a calibrated classifier didn't matter much
December 3, 2025 at 1:45 AM
My impression from brief stinks in industry was that folks primarily up/down-sampled because they wanted a system good at ordering samples by P(Y=1|X), and that beyond ranking having a calibrated classifier didn't matter much
Reposted by alex hayes
which is the worse statistical sin, multiplying regression coefficients (mediation) or dividing regression coefficients (instrumental variables)
December 1, 2025 at 8:53 AM
which is the worse statistical sin, multiplying regression coefficients (mediation) or dividing regression coefficients (instrumental variables)
The LW shade made me cackle
November 29, 2025 at 9:51 PM
The LW shade made me cackle
Keith and I strengthened our lower bounds for randomized experiments in the linear-in-means model to a general minimax result
Updated results available at arxiv.org/abs/2410.10772
Updated results available at arxiv.org/abs/2410.10772
November 25, 2025 at 9:55 PM
Keith and I strengthened our lower bounds for randomized experiments in the linear-in-means model to a general minimax result
Updated results available at arxiv.org/abs/2410.10772
Updated results available at arxiv.org/abs/2410.10772
You might also enjoy www.science.org/doi/10.1126/...
Gerlach, Martin, Tiago P. Peixoto, and Eduardo G. Altmann. “A Network Approach to Topic Models.” Science Advances 4, no. 7 (2018): 1–11. doi.org/10.1126/scia....
Gerlach, Martin, Tiago P. Peixoto, and Eduardo G. Altmann. “A Network Approach to Topic Models.” Science Advances 4, no. 7 (2018): 1–11. doi.org/10.1126/scia....
A network approach to topic models
A new approach to topic models finds topics through community detection in word-document networks.
www.science.org
November 23, 2025 at 1:38 AM
You might also enjoy www.science.org/doi/10.1126/...
Gerlach, Martin, Tiago P. Peixoto, and Eduardo G. Altmann. “A Network Approach to Topic Models.” Science Advances 4, no. 7 (2018): 1–11. doi.org/10.1126/scia....
Gerlach, Martin, Tiago P. Peixoto, and Eduardo G. Altmann. “A Network Approach to Topic Models.” Science Advances 4, no. 7 (2018): 1–11. doi.org/10.1126/scia....
Even if you don't find the main figure concerning, do you find the contrast with pre-registered RCT z-scores concerning? It's hard to imagine a world in which this contrast is benign
November 15, 2025 at 11:58 PM
Even if you don't find the main figure concerning, do you find the contrast with pre-registered RCT z-scores concerning? It's hard to imagine a world in which this contrast is benign
Reposted by alex hayes
A new release of the mgcv #RStats 📦 is out on CRAN and Simon Wood (U Edinburgh) has added some significant new features despite the small bump in version number:
🌟 scasm() for estimating GAMs with shape constrained smooths. Can be used with any family & smoothness selection is via the EFS method
🌟 scasm() for estimating GAMs with shape constrained smooths. Can be used with any family & smoothness selection is via the EFS method
November 12, 2025 at 11:28 AM
A new release of the mgcv #RStats 📦 is out on CRAN and Simon Wood (U Edinburgh) has added some significant new features despite the small bump in version number:
🌟 scasm() for estimating GAMs with shape constrained smooths. Can be used with any family & smoothness selection is via the EFS method
🌟 scasm() for estimating GAMs with shape constrained smooths. Can be used with any family & smoothness selection is via the EFS method
Do the RStudio Check/Test/Load all keybinds work in Positron?
November 6, 2025 at 9:50 PM
Do the RStudio Check/Test/Load all keybinds work in Positron?
there's some recently theory on optimal tests in arxiv.org/abs/2107.07575, with accompanying code in achambaz.github.io/mediation.te...
tl;dr it's a complicated hypothesis to test
tl;dr it's a complicated hypothesis to test
Optimal tests of the composite null hypothesis arising in mediation analysis
The indirect effect of an exposure on an outcome through an intermediate variable can be identified by a product of two regression coefficients under certain causal and regression modeling assumptions...
arxiv.org
November 5, 2025 at 9:02 PM
there's some recently theory on optimal tests in arxiv.org/abs/2107.07575, with accompanying code in achambaz.github.io/mediation.te...
tl;dr it's a complicated hypothesis to test
tl;dr it's a complicated hypothesis to test
Just read this -- really enjoy the differencing out of the calibration curve to find questions that folks are most differentially miscalibrated on
November 5, 2025 at 7:59 PM
Just read this -- really enjoy the differencing out of the calibration curve to find questions that folks are most differentially miscalibrated on
Just ordered a copy! Thanks so much for all these resources :)
November 4, 2025 at 7:27 PM
Just ordered a copy! Thanks so much for all these resources :)