Will Lowe | Lightnews

Light up
your news

About Privacy Terms Help

Will Lowe

Will Lowe

@conjugateprior.org

3.4K followers 750 following 10K posts

Señor Research Scientist, NPC at the Hertie School in Berlin 🇩🇪 via Princeton, Mannheim, Edinburgh and a bunch of other ivory towers that will probably be billiard balls and decorative boxes by the end of the decade.

Rome Statute appreciator. .. more

Computer science 32%

Political science 28%

more

Posts Replies Media Videos

Pinned

Will Lowe @conjugateprior.org · Apr 15

For Monty Hall problem aficionados: a #causal DAG, with explanation in the alt text.

Who knew that the M in M-bias stood for Monty?

A DAG for the Monty Hall problem.

G, the door Monty opens, is a function of your first door choice D_F and of where the car is C.

Conditioning on G, which happens because it determines your door choices when you have the opportunity to switch, induces collider bias between C and D_F, such that they are now less likely to be equal. (Why? because G cannot be C and G cannot be D_F. These are in the same 'direction' so, conditional on G, D_F is less likely to be C.)

You decide to switch conditional on G, which you just observed, and D_F, because that defines what you can switch to.

Conditional on G you now know D_F is less likely to be equal to C. So you switch.

Note: In the original problem, W is a deterministic function of C and D_S so it's slightly cheeky to put it in a DAG, but I'm going to do so anyway. The argument doesn't require it, but if you want to think informally about opening paths to W from D_F via C, it's helpful.

That makes this problem a variation on M-bias (or a clipped wing butterfly bias, if you prefer insects to goats).

Will Lowe @conjugateprior.org · 1d

OBLIGATE CARNIVORE DISAPPOINT

Will Lowe @conjugateprior.org · 1d

c.f. the awkward moment when Neyman notices that Karl doesn't get the difference between independence and uncorrelatedness.

Will Lowe @conjugateprior.org · 1d

There is no universe where noisily measured causal effects are not monotonic over quartiles? TIL that I live in an impossible universe, which is pretty epic news to get just after breakfast, I reckon.

Will Lowe @conjugateprior.org · 1d

bwahahagodimsosorry

Will Lowe @conjugateprior.org · 2d

Weeder channeling Wittgenstein *again* 🙄

weeder @weeder.bsky.social · 2d

Don't Inner Monologue in Meme Formulas Challenge (impossible)

Reposted by Will Lowe

weeder @weeder.bsky.social · 2d

Don't Inner Monologue in Meme Formulas Challenge (impossible)

Will Lowe @conjugateprior.org · 2d

It's "mentally healthy" because that's how things actually work.

Reposted by Joanna Bryson

Will Lowe @conjugateprior.org · 2d

While this is a classic Good diss, the highlighted section is more important. The key point for subjective Bayesians is that Quine et al. are right: you don't automatically know your own mind just because you talk to yourself about it. You're estimating things, and that includes your credibilities.

Andy Seaton @aseaton.bsky.social · 2d

I've been reading Good Statistics by I. J. Good. He's entertaining as a writer. He's also pretty good at letting you know exactly what he thinks in the most unimpeachably polite British way possible.

#statsky

Reposted by Will Lowe

Andy Seaton @aseaton.bsky.social · 2d

I've been reading Good Statistics by I. J. Good. He's entertaining as a writer. He's also pretty good at letting you know exactly what he thinks in the most unimpeachably polite British way possible.

#statsky

Will Lowe @conjugateprior.org · 2d

No idea, but it makes sense that conditioning should be easier than de/un-conditioning.

Will Lowe @conjugateprior.org · 3d

I am genuinely in awe of – and slightly terrified by – some academic's urge to let no research, research-adjacent or research-if-you-were-to-look-at-it-in-a-dark-light activity go to waste 🏃💨

Will Lowe @conjugateprior.org · 3d

I do get the instinct to think about decomposing 'good' papers into features and operationalizing those into citation metrics and whatever this ends up doing. But I also suspect we should not directly maximize 'good papers' even if we could. The medium (papers) just isn't the message (science).

Will Lowe @conjugateprior.org · 3d

At least one person is treating it like this! 😁

pure.qub.ac.uk/en/activitie...

Will Lowe @conjugateprior.org · 3d

Anyway, I hope that was slightly informative about your mysterious email requests. All additional welcome very welcome.

Will Lowe @conjugateprior.org · 3d

Personally, I'd love to know what Elsevier's role is, in particular because Elsevier, like Thompson-Reuters et al. are explicit that they are increasingly more 'analytics company' than publisher. Which makes since when print is less in demand but everyone pays to give you free data, because reasons.

Will Lowe @conjugateprior.org · 3d

The other folk seem to be, on the government side UKRI's Ben Steyn, on the Elsevier side David Plume, and Susan Guthrie at Rand. You can get a decent idea of what they're up to on the link above.

I thought I'd learn something about evaluation there, but my searches so far draw a complete blank.

Will Lowe @conjugateprior.org · 3d

And speaking of details, who are the people in all this? Based on a presentation at a pre-conference workshop of Metascience 2025 cassyni.com/events/AqjXf... the sender of the emails is Sarah Otner of the University of Sussex: "it's from me". I'm not sure why that wasn't in the email.

Cassyni | Science starts with a seminar

Seamlessly organise, run and publish academic research talks. Get started in minutes.

Will Lowe @conjugateprior.org · 3d

Ok so we all rate lots of papers. How are the indicator developers to be evaluated with our scores?

There is some discussion in the entrant hanbook (sic) noveltyindicators.challenges.org/support-to-e... tl;dr it's going to be "accuracy" and they'll figure out the details later & let everyone know 🤔

The winner of the Metascience Novelty Indicators Challenge will be awarded to the indicator that can achieve the highest level of accuracy towards the dataset of expert novelty scores.

The set of novelty ratings for the list of papers submitted by challengers for the final submission will be tested against the expert novelty score dataset. We are currently developing the best approach to delivering this test in a way that matches the characteristics of the expert rater dataset.
Challengers will be provided with a detailed description of this method alongside the metadata, which will be shared in December 2025.

Will Lowe @conjugateprior.org · 3d

A pessimistic view is that Elsevier is going to build the winning entry into its workflow & has, amazingly, gotten the government to indirectly fund measure development and a charity to pay off the developers afterwards.

Since this involves scientific publishing it's, eh, probably worse that that 😟

Will Lowe @conjugateprior.org · 3d

I can see why someone might think it would be... unpersuasive to ask academics to do (more) work for Elsevier, so left them out. Bit skeezy though, no?

An optimistic view is that Elsevier is there because Scopus chose the articles we are matched to. Access to their data is also part of the prize.

Will Lowe @conjugateprior.org · 3d

So who's running this thing? That's a bit harder to say. If your email is like mine, it mentions these folk below (left).

But if you follow that link too, three more orgs appear: Coefficient Giving (they're the money), Challenge Works (no idea who that is), and Elsevier, who need no introduction.

For further information on this project please visit our website.

[logos for]

UK Research and Innovation (a government entity apparently)

University of Sussex Business School's Science Policy Research Unit

Rand Europe

WHO IS INVOLVED?

The Metascience Novelty Indicators Challenge is a collaborative effort launched by the UK Metascience Unit (UKRI) with the support of Elsevier. Its design and implementation were supported by SPRU and RAND Europe, and it is being delivered by Challenge Works, a Nesta enterprise. Coefficient Giving will be providing the £300,000 prize to the selected winner through Challenge Works.

[logos]

Will Lowe @conjugateprior.org · 3d

It seems that we survey takers are in fact evaluators, albeit not very careful ones since the survey designers estimate 6 minutes to evaluate an article's novelty with a screen full of sliders. We're being asked to construct the test set for a competition whose winner gets £300,000 (~350k Euro).

HOW TO WIN

The Metascience Novelty Indicators Challenge is a global initiative to develop and validate scalable indicators that can identify novelty in research. We are inviting academics, data scientists, researchers, and innovators worldwide to create and submit their novelty indicators to be validated against a dataset of human expert assessments
- the best-performing indicator will win a prize of £300,000 alongside an opportunity to have their tool implemented in a real-world context.

Reposted by Alexander Wuttke

Will Lowe @conjugateprior.org · 3d

Perhaps you received a mysterious noreply email asking you to evaluate some publications 'for novelty'. Looked kinda dubious? Yup, that's the one.

So what's up with this 'metascience novelty indicators challenge'? 🧵

Will Lowe @conjugateprior.org · 3d

5318008

a cartoon dog is laughing and biting its nails .

Alt: Muttley laughs after seeing the calculator screen

media.tenor.com

Will Lowe @conjugateprior.org · 3d

Association-cancelling collider bias is my favourite collider bias.

Will Lowe @conjugateprior.org · 3d

Bad news: www.nytimes.com/2025/03/01/us/…

Will Lowe @conjugateprior.org · 4d

Anscheinend hat nicht jeder Zweig der Familie seinen Namen geändert, nachdem sie Gallien verlassen hatten.

Rohrfix

Will Lowe @conjugateprior.org · 4d

resigning from governing the prime minister, perhaps?

Will Lowe @conjugateprior.org · 4d

isn't it? While it goes against the grain (one can immediately see all the ways these individuals are not like-motivated) I do think it's often helpful to start with an extension and see if it's possible to induce an intension. CWL, per the article, is much more coherent than the sum of those guys.

Will Lowe @conjugateprior.org · 5d

One of the hazards of institutionalization and identification with particular countries, I suppose. Apropos, the US's "capacity for benevolent global leadership" was and remains the point where I step off the train. That's not a phrase that belongs near any form of liberalism.