Jon Mellon
@jonmellon.bsky.social
Co-director British Election Study. Political Scientist and Data Scientist. Political science methods/political behavior/causal inference. Posts do not represent employer.
Pinned
Jon Mellon
@jonmellon.bsky.social
· Nov 14
Out now open access at
@ajpseditor.bsky.social. 194 potential exclusion-restriction violations for studies using weather as an instrumental variable onlinelibrary.wiley.com/doi/full/10....
@ajpseditor.bsky.social. 194 potential exclusion-restriction violations for studies using weather as an instrumental variable onlinelibrary.wiley.com/doi/full/10....
Reposted by Jon Mellon
I am on paternity leave, so not promoting this properly, but lots of Britons have stable but idiosyncratic views , as shown in this book doi.org/10.1093/9780...
Validate User
doi.org
February 3, 2026 at 4:22 PM
I am on paternity leave, so not promoting this properly, but lots of Britons have stable but idiosyncratic views , as shown in this book doi.org/10.1093/9780...
Reposted by Jon Mellon
Yes that’s definitely true. I have a lot of cites to my “only use this method if you run a validation for your use case” papers that read “Mellon (2014) shows that this method is valid for my use case”
January 30, 2026 at 2:36 PM
Yes that’s definitely true. I have a lot of cites to my “only use this method if you run a validation for your use case” papers that read “Mellon (2014) shows that this method is valid for my use case”
I think the LLM hallucinated citations problem is going away (because it’s fundamentally easy to check if a paper exists using an API) but the much harder problem (especially for faculty reviewing student work) will be hallucinating the content of citations
January 30, 2026 at 1:18 PM
I think the LLM hallucinated citations problem is going away (because it’s fundamentally easy to check if a paper exists using an API) but the much harder problem (especially for faculty reviewing student work) will be hallucinating the content of citations
It's a bit annoying that the default python setup for datascience (jupyter notebooks) is a bad fit for git version control (you end up with inconsequential chunk output merge issues). I wish they'd gone with a stricter separation of code and output like rmd.
January 28, 2026 at 9:28 PM
It's a bit annoying that the default python setup for datascience (jupyter notebooks) is a bad fit for git version control (you end up with inconsequential chunk output merge issues). I wish they'd gone with a stricter separation of code and output like rmd.
Reposted by Jon Mellon
If it makes you feel better seahorses are more closely related to horses than sharks
January 26, 2026 at 1:00 PM
If it makes you feel better seahorses are more closely related to horses than sharks
Reposted by Jon Mellon
Psychologists stop trying to do psychological science
no wait
no wait
Economists stop trying to do political science
Physicists stop trying to do economics
January 15, 2026 at 2:06 PM
Psychologists stop trying to do psychological science
no wait
no wait
Reposted by Jon Mellon
When you stop borrowing our methodologies
January 15, 2026 at 1:59 PM
When you stop borrowing our methodologies
Economists stop trying to do political science
Physicists stop trying to do economics
January 15, 2026 at 1:45 PM
Economists stop trying to do political science
Tim Harford has a great podcast on this where he shows that people didn’t lose that much money on the tulip bubble but lost their shirts investing in railroads (despite the fact that those were a transformative innovation) timharford.com/2023/01/caut...
January 14, 2026 at 2:37 PM
Tim Harford has a great podcast on this where he shows that people didn’t lose that much money on the tulip bubble but lost their shirts investing in railroads (despite the fact that those were a transformative innovation) timharford.com/2023/01/caut...
Reposted by Jon Mellon
By the way this is separate from whether you think AI companies stocks are overvalued. The technology at its present level is going to reshape academic research even if Sam Altman goes bankrupt
January 14, 2026 at 2:15 PM
By the way this is separate from whether you think AI companies stocks are overvalued. The technology at its present level is going to reshape academic research even if Sam Altman goes bankrupt
I know Bluesky is more LLM sceptical, but I definitely had the experience both that gpt-5.2 is superhuman at getting proofs to compile and that it’s not that well aligned (and is chasing compilation success not proving the substantive claims I wanted to)
Even in this context my experience was that the LLM repeatedly substituted in similar sounding but far stronger assumptions or related but weaker theorems than intended
January 14, 2026 at 2:06 PM
I know Bluesky is more LLM sceptical, but I definitely had the experience both that gpt-5.2 is superhuman at getting proofs to compile and that it’s not that well aligned (and is chasing compilation success not proving the substantive claims I wanted to)
I've been playing with Lean to prove consistency of a 2-stage estimator in a sociology paper. I have a question for people who do lean proofs.
If you understand the theorem and the assumptions it relies on, how much can you treat the an LLM lean proof as a black box
If you understand the theorem and the assumptions it relies on, how much can you treat the an LLM lean proof as a black box
January 14, 2026 at 1:59 PM
I've been playing with Lean to prove consistency of a 2-stage estimator in a sociology paper. I have a question for people who do lean proofs.
If you understand the theorem and the assumptions it relies on, how much can you treat the an LLM lean proof as a black box
If you understand the theorem and the assumptions it relies on, how much can you treat the an LLM lean proof as a black box
I’ve had an idea floating around my head that there might be a niche audience that would be interested in watching me live stream researching and writing social science papers in realtime on twitch. Terrible idea or kind of fun?
January 10, 2026 at 3:22 PM
I’ve had an idea floating around my head that there might be a niche audience that would be interested in watching me live stream researching and writing social science papers in realtime on twitch. Terrible idea or kind of fun?
Reposted by Jon Mellon
I'm sure someone else has suggested this, but it's time for mandatory DOIs in reference lists. This will make it easier to identify hallucinated references
January 9, 2026 at 1:06 PM
I'm sure someone else has suggested this, but it's time for mandatory DOIs in reference lists. This will make it easier to identify hallucinated references
Does anyone see an issue with using parenthesis color highlighting when typesetting equations like this in latex? ChatGPT seems to think it’s uncouth but I think it would be a clear improvement.
January 4, 2026 at 8:23 PM
Does anyone see an issue with using parenthesis color highlighting when typesetting equations like this in latex? ChatGPT seems to think it’s uncouth but I think it would be a clear improvement.
Reposted by Jon Mellon
Yeah basically agree with that. Memorising syntax, I'm relatively sure, will become like punch cards
December 28, 2025 at 6:16 PM
Yeah basically agree with that. Memorising syntax, I'm relatively sure, will become like punch cards
Reposted by Jon Mellon
Yes but someone has to also evaluate the foundations of some set of competing approaches. Maybe it's fine and we can just trust LLMs to do that faultlessly. But I fear we're in an odd limbo land rn. Even SOTA models can still do some mad things
December 28, 2025 at 6:13 PM
Yes but someone has to also evaluate the foundations of some set of competing approaches. Maybe it's fine and we can just trust LLMs to do that faultlessly. But I fear we're in an odd limbo land rn. Even SOTA models can still do some mad things
Expanding on this. The rise of LLMs push the role of data scientists closer to systems engineers and away from coders. Systems engineering is about turning competing stakeholder demands into consensus criteria for judging a solution and coming up with ways to verify whether criteria have been met.
I’ve seen some junior data scientists really rise to the challenge of AI and have bigger impacts than you could have expected of someone with that experience level. I’m cautiously optimistic for then in the next year or two
December 28, 2025 at 5:19 PM
Expanding on this. The rise of LLMs push the role of data scientists closer to systems engineers and away from coders. Systems engineering is about turning competing stakeholder demands into consensus criteria for judging a solution and coming up with ways to verify whether criteria have been met.
Reposted by Jon Mellon
The industry data science and research jobs that academics have been (maybe) preparing students for over the last decade or so will be unrecognizable in the next 3 to 6 months because of AI. It's coming that quickly and the level of change in day-to-day work is unprecedented.
December 28, 2025 at 1:52 AM
The industry data science and research jobs that academics have been (maybe) preparing students for over the last decade or so will be unrecognizable in the next 3 to 6 months because of AI. It's coming that quickly and the level of change in day-to-day work is unprecedented.
Gemini3-pro has a weirdly condescending personality.
December 4, 2025 at 5:35 PM
Gemini3-pro has a weirdly condescending personality.
Does anyone else feel like there’s a lot more respiratory diseases going around this winter? Is there any data backing this up or is everyone I know unusually unlucky?
December 3, 2025 at 2:06 PM
Does anyone else feel like there’s a lot more respiratory diseases going around this winter? Is there any data backing this up or is everyone I know unusually unlucky?
I’ve been running a history book club for the last few months. What are everyone’s favorite history books?
November 30, 2025 at 6:03 PM
I’ve been running a history book club for the last few months. What are everyone’s favorite history books?
It’s a real job! But sadly it actually involves things making classifications of diseases and medicines
Just saw someone described as an "ontologist" and now I want Clinical Ontologist to be a job title.
Welcome to the Department of Clinical Ontology. Let's go over your chart. I see that you are. Are you still as you were or are you otherwise? For your case of being, I prescribe becoming.
Welcome to the Department of Clinical Ontology. Let's go over your chart. I see that you are. Are you still as you were or are you otherwise? For your case of being, I prescribe becoming.
November 22, 2025 at 12:35 PM
It’s a real job! But sadly it actually involves things making classifications of diseases and medicines
One day I want to create a wrapper package for lm called uniform_prior_lm that does nothing but relabel CIs and p-values as their Bayesian equivalents.
Confidence interval discussion time! The perfect opportunity to repost this blog post answering the question you haven’t dared to ask: www.the100.ci/2024/12/05/w...
November 19, 2025 at 1:40 PM
One day I want to create a wrapper package for lm called uniform_prior_lm that does nothing but relabel CIs and p-values as their Bayesian equivalents.
Reposted by Jon Mellon
