Tom Carpenter, PhD
@tcarpenter.bsky.social
🧪 Data science, survey science, social science
💻 Director of Data Science @ Microsoft Garage
[Posts do not represent my employer]
🧮 Stats, R, python
📝 Science, Research: measurement, social biases, emotion. Ex-academic but scientist at heart
💻 Director of Data Science @ Microsoft Garage
[Posts do not represent my employer]
🧮 Stats, R, python
📝 Science, Research: measurement, social biases, emotion. Ex-academic but scientist at heart
Reposted by Tom Carpenter, PhD
We have progressed from data collection to data analysis.
November 1, 2025 at 12:31 AM
We have progressed from data collection to data analysis.
Reposted by Tom Carpenter, PhD
ladies and gentlemen...we got him
October 30, 2025 at 7:10 PM
ladies and gentlemen...we got him
Reposted by Tom Carpenter, PhD
* hoppening
Things are happening at Portland's ICE facility tonight.
Read more of our protest coverage here: www.oregonlive.com/crime/2025/1...
Read more of our protest coverage here: www.oregonlive.com/crime/2025/1...
October 10, 2025 at 4:20 AM
* hoppening
Reposted by Tom Carpenter, PhD
this slide is from a colleague's introductory stats course, I think it fits many statisticians' experiences
August 18, 2025 at 12:31 AM
this slide is from a colleague's introductory stats course, I think it fits many statisticians' experiences
Reposted by Tom Carpenter, PhD
45. Academia doesn't reward building useful tools nearly as much as it should
July 28, 2025 at 12:44 AM
45. Academia doesn't reward building useful tools nearly as much as it should
Reposted by Tom Carpenter, PhD
14. We mostly evaluate latent variable models with the equivalent of Rorschach tests
July 27, 2025 at 4:20 PM
14. We mostly evaluate latent variable models with the equivalent of Rorschach tests
Reposted by Tom Carpenter, PhD
5. You should use a precision-recall curve for a binary classifier, not an ROC curve
July 27, 2025 at 1:42 PM
5. You should use a precision-recall curve for a binary classifier, not an ROC curve
Reposted by Tom Carpenter, PhD
Wow. Scientists have edited mosquito DNA to prevent the spread of malaria to humans "while supporting essential physiological functions... and negligible fitness costs" to the mosquito population.
Potentially ending the mosquito-born spread of malaria to humans.
www.nature.com/articles/s41...
Potentially ending the mosquito-born spread of malaria to humans.
www.nature.com/articles/s41...
July 25, 2025 at 3:23 PM
Wow. Scientists have edited mosquito DNA to prevent the spread of malaria to humans "while supporting essential physiological functions... and negligible fitness costs" to the mosquito population.
Potentially ending the mosquito-born spread of malaria to humans.
www.nature.com/articles/s41...
Potentially ending the mosquito-born spread of malaria to humans.
www.nature.com/articles/s41...
Curious how this compares to the cost of living per state
I hate when I see Ds accept the premise that Ds left the working class behind.
Culturally? Sure
Economically?
Pretty clear who has the back of the working class
Culturally? Sure
Economically?
Pretty clear who has the back of the working class
July 19, 2025 at 8:27 PM
Curious how this compares to the cost of living per state
Reposted by Tom Carpenter, PhD
Let's say you want to include age as a predictor in your model. How do you do that?
Here's an illustration of three options -- it's for a paper I'm working on (so if you feel like anything could be tweaked...).
Here's an illustration of three options -- it's for a paper I'm working on (so if you feel like anything could be tweaked...).
July 8, 2025 at 12:03 PM
Let's say you want to include age as a predictor in your model. How do you do that?
Here's an illustration of three options -- it's for a paper I'm working on (so if you feel like anything could be tweaked...).
Here's an illustration of three options -- it's for a paper I'm working on (so if you feel like anything could be tweaked...).
Reposted by Tom Carpenter, PhD
There should be a corner at Home Depot where a guy with a table saw will slice you off custom lengths of hot dog from an infinite hot dog coming out of the wall
July 5, 2025 at 12:15 AM
There should be a corner at Home Depot where a guy with a table saw will slice you off custom lengths of hot dog from an infinite hot dog coming out of the wall
Reposted by Tom Carpenter, PhD
There were two girls at Wawa just now talking about funny movies and one said, “Have you ever seen the movie Office Space? It’s an old people movie but it’s funny”
June 25, 2025 at 1:10 AM
There were two girls at Wawa just now talking about funny movies and one said, “Have you ever seen the movie Office Space? It’s an old people movie but it’s funny”
Reposted by Tom Carpenter, PhD
🤖Thrilled to share our latest work☄️
Have you ever wondered what LLMs know but they are not saying?
We built an auditing framework to study information suppression in LLMs, and demonstrated it to quantify and characterize censorship in DeepSeek.
Read more:
arxiv.org/abs/2506.12349
Have you ever wondered what LLMs know but they are not saying?
We built an auditing framework to study information suppression in LLMs, and demonstrated it to quantify and characterize censorship in DeepSeek.
Read more:
arxiv.org/abs/2506.12349
Information Suppression in Large Language Models: Auditing, Quantifying, and Characterizing Censorship in DeepSeek
This study examines information suppression mechanisms in DeepSeek, an open-source large language model (LLM) developed in China. We propose an auditing framework and use it to analyze the model's res...
arxiv.org
June 22, 2025 at 11:52 PM
🤖Thrilled to share our latest work☄️
Have you ever wondered what LLMs know but they are not saying?
We built an auditing framework to study information suppression in LLMs, and demonstrated it to quantify and characterize censorship in DeepSeek.
Read more:
arxiv.org/abs/2506.12349
Have you ever wondered what LLMs know but they are not saying?
We built an auditing framework to study information suppression in LLMs, and demonstrated it to quantify and characterize censorship in DeepSeek.
Read more:
arxiv.org/abs/2506.12349
Reposted by Tom Carpenter, PhD
Pleased to share our ICML Spotlight with @eberleoliver.bsky.social, Thomas McGee, Hamza Giaffar, @taylorwwebb.bsky.social.
Position: We Need An Algorithmic Understanding of Generative AI
What algorithms do LLMs actually learn and use to solve problems?🧵1/n
openreview.net/forum?id=eax...
Position: We Need An Algorithmic Understanding of Generative AI
What algorithms do LLMs actually learn and use to solve problems?🧵1/n
openreview.net/forum?id=eax...
June 20, 2025 at 3:48 PM
Pleased to share our ICML Spotlight with @eberleoliver.bsky.social, Thomas McGee, Hamza Giaffar, @taylorwwebb.bsky.social.
Position: We Need An Algorithmic Understanding of Generative AI
What algorithms do LLMs actually learn and use to solve problems?🧵1/n
openreview.net/forum?id=eax...
Position: We Need An Algorithmic Understanding of Generative AI
What algorithms do LLMs actually learn and use to solve problems?🧵1/n
openreview.net/forum?id=eax...
Someone please explain why linear algebra isn’t taught more in high school in the United States? Seems like maybe if you’re lucky you get a few lectures on matrices and that’s it.
June 18, 2025 at 3:46 AM
Someone please explain why linear algebra isn’t taught more in high school in the United States? Seems like maybe if you’re lucky you get a few lectures on matrices and that’s it.
Is it 2015 or 2025?
Researchers tried to replicate a bunch of sports science studies.
They got "robust" replication of 28% of them.
Only one replication showed a larger effect size.
But here's what got me ...
link.springer.com/article/10.1...
They got "robust" replication of 28% of them.
Only one replication showed a larger effect size.
But here's what got me ...
link.springer.com/article/10.1...
June 17, 2025 at 3:26 PM
Is it 2015 or 2025?
“Data available upon request”
The researchers contacted the authors of the original studies -- 156 of them.
And you know how many were willing to share data?
14% of them.
If we're going to address replication issues, it has to start with sharing data.
And you know how many were willing to share data?
14% of them.
If we're going to address replication issues, it has to start with sharing data.
June 17, 2025 at 3:25 PM
“Data available upon request”
Reposted by Tom Carpenter, PhD
Under the Trump agenda, energy will cost more. And when energy costs more, everything costs more.
www.nytimes.com/2025/06/04/c...
www.nytimes.com/2025/06/04/c...
Electricity Prices Are Surging. The G.O.P. Megabill Could Push Them Higher.
www.nytimes.com
June 8, 2025 at 5:39 PM
Under the Trump agenda, energy will cost more. And when energy costs more, everything costs more.
www.nytimes.com/2025/06/04/c...
www.nytimes.com/2025/06/04/c...
Many academics are taught “do more” instead of “prioritize”. They are taught the academic superhero myth, that “truly great/smart” scholars can handle it. So they sabotage their own success in service to the cult of personality
#AcademicSky : “Why is it that smart early-career researchers consistently sabotage their own success by saying yes to everything?”
Why Early Career Researchers Can't Say No (And How To Fix It) ⮛
Why Early Career Researchers Can't Say No (And How To Fix It) ⮛
More strategic tips on how to say no in my FREE deep dive right here:  lennartnacke.com/why-early-ca...
June 5, 2025 at 2:00 PM
Many academics are taught “do more” instead of “prioritize”. They are taught the academic superhero myth, that “truly great/smart” scholars can handle it. So they sabotage their own success in service to the cult of personality
I love this technique because it gives a cool way to isolate change and stability components of within-person measurement using latent variables. The stable portion of IATs is far more predictive of other individual-difference measures than one would think given traditional scoring/analyses
Using latent trait-state models, we found that states explain 42% of variance in IATs & traits explain 62% of variance in those states. This finding shows the IAT is capturing traits AND states, but there’s a lot of measurement error.
June 4, 2025 at 11:00 PM
I love this technique because it gives a cool way to isolate change and stability components of within-person measurement using latent variables. The stable portion of IATs is far more predictive of other individual-difference measures than one would think given traditional scoring/analyses
I love this analysis technique because it gives a cool way to isolate change and stability components of within-person measurement using latent variables. The stable portion of IATs is far more predictive of other individual-difference measures than one would think given traditional scoring/anakysus
Using latent trait-state models, we found that states explain 42% of variance in IATs & traits explain 62% of variance in those states. This finding shows the IAT is capturing traits AND states, but there’s a lot of measurement error.
June 4, 2025 at 11:00 PM
I love this analysis technique because it gives a cool way to isolate change and stability components of within-person measurement using latent variables. The stable portion of IATs is far more predictive of other individual-difference measures than one would think given traditional scoring/anakysus
That moment when you had completely forgotten about a cool data app you’d made, only to find others posting about it ;)
Thanks @calvinklai.bsky.social!
Thanks @calvinklai.bsky.social!
FAQ
Q1: 4 IATs is a lot!?
A: You only need 2 IATs to reap the benefits of aggregating.
Q2: It's a lot of work to set up these models.
A: Here's a shiny app to make it easy: tcarpenter.shinyapps.io/trait_iat/
Q3: I don't want to learn CFA.
A: Averaging IATs together gets you some of the benefits.
Q1: 4 IATs is a lot!?
A: You only need 2 IATs to reap the benefits of aggregating.
Q2: It's a lot of work to set up these models.
A: Here's a shiny app to make it easy: tcarpenter.shinyapps.io/trait_iat/
Q3: I don't want to learn CFA.
A: Averaging IATs together gets you some of the benefits.
June 4, 2025 at 10:56 PM
That moment when you had completely forgotten about a cool data app you’d made, only to find others posting about it ;)
Thanks @calvinklai.bsky.social!
Thanks @calvinklai.bsky.social!
Reposted by Tom Carpenter, PhD
Paper w/ @tcarpenter.bsky.social & @alexgoedderz.bsky.social at PSPB!🚨 The standard IAT is only 5 min long. We found that making the IAT longer by taking it multiple times greatly improves predictive validity. 🧵below, with practical advice about how to run IATs!
LINK: osf.io/preprints/ps...
LINK: osf.io/preprints/ps...
June 4, 2025 at 4:49 PM
Paper w/ @tcarpenter.bsky.social & @alexgoedderz.bsky.social at PSPB!🚨 The standard IAT is only 5 min long. We found that making the IAT longer by taking it multiple times greatly improves predictive validity. 🧵below, with practical advice about how to run IATs!
LINK: osf.io/preprints/ps...
LINK: osf.io/preprints/ps...
Reposted by Tom Carpenter, PhD
BOTTOM LINE: Aggregating 2 IATs together and using CFA can lead to much higher predictive validity.
Finally, I wanted to thank Tom and Alex! This paper came out while they were moving out of academia & I was going on parental leave, so we hadn't publicized it. Better late than never! 😉
Finally, I wanted to thank Tom and Alex! This paper came out while they were moving out of academia & I was going on parental leave, so we hadn't publicized it. Better late than never! 😉
June 4, 2025 at 4:49 PM
BOTTOM LINE: Aggregating 2 IATs together and using CFA can lead to much higher predictive validity.
Finally, I wanted to thank Tom and Alex! This paper came out while they were moving out of academia & I was going on parental leave, so we hadn't publicized it. Better late than never! 😉
Finally, I wanted to thank Tom and Alex! This paper came out while they were moving out of academia & I was going on parental leave, so we hadn't publicized it. Better late than never! 😉