Per Engzell

About five years ago, I remember a colleague passionately arguing we need pre-registered seeds. Starting to think they were right...

Part of the beauty of academic writing is that it's modular: you can get the gist from the abstract, and choose to deep dive into whichever part is relevant for you.

On the one hand, yes. On the other, it feels like the time when we would read full papers from start to finish was gone long before LLMs, if such a time ever existed.

Yeah that's the question. On the one hand, you'd imagine that LLMs would free up more time to explore data, go down alternative paths etc. On the other, it doesn't feel to me like that is happening? I can't say why though

That's another aspect of this: the equivalent of Claude has always existed, for people at well-resourced research unis with an endless supply of research assistants...

We often treat education as the primary engine of social mobility, but conventional models may be overestimating its impact.

Not correcting for selection into schooling will artificially inflate any indirect effect through schooling...

1/3

#socialmobility #sociology

There’s something obviously alienating about that in a simple Marxian sense, and it's not hard to spot the parallels with earlier waves of automation. /end

Last thought: feels like we haven't yet grasped the psychological and social consequences of researchers going from being workers to being managers of research production. 11/

If you've ever spent a day researching a paragraph that later ends up a footnote, that later ends up cut... can you honestly say you've never regretted it? 10/

The fact that writing sprawls out in all directions at first *feels* like an important part of the thought process. That seems right, but is it always so? 9/

Another difference: an LLM writes exactly what is required and nothing more. Normally writing starts with an excess of text and results, before you start pruning it down. 8/

The idea in this case is good. But it's still the kind of project where you might once have thought: interesting result, but not quite worth a whole paper. 7/

Part of what makes the paper 'feel LLM' is we're not used to seeing full papers on ideas that might be interesting but not necessarily enough so to justify the writing effort. 6/

Before LLMs, you might spend a couple of hours exploring a throwaway result. But it would rarely feel worth the effort to write the whole paper. I wonder if that might change now. 5/

It’s obvious that newer models got better at coding. But they’ve also become better at writing, especially adapting tone to a specific field, journal, or even author. 4/

Eerily this *felt* distinctly like other papers written by the same person, as something they might plausibly have produced - which is also exactly what makes it feel LLM. 3/

This was a short paper proving a basic point - incrementally important but perhaps not enough so to justify the high opportunity cost of writing a full paper pre-LLM. 2/

Another colleague shared a complete, and by all accounts sound, empirical paper written by LLM, with some prompting. Some thoughts (and some of you will be sick of hearing this by now)… 1/n

Published is another matter...

One potential upside as the cost of writing a paper goes to zero: null results might finally get written up

After 5 years I can finally share a full WP of our project conducting cognitive interviews of life satisfaction reporting.

Main findings:
1. LS scales are psychometrically valid, but...
2. Standard statistical assumptions made when analysing LS data are not credible.

osf.io/gv5e3/files/...

When you collect data online, are the results from humans or AI? In a project led by Booth PhD student Grace Zhang, we estimate the prevalence of AI agents on commonly used survey platforms:
osf.io/preprints/ps...
🧵

Good shout out! (Btw half of the authors overlap and everyone is in the same department so I think George knows...)

Excited to share new work with Yaming Cao @zew.de and @kayaez.bsky.social @cu-cardiffbusiness.bsky.social published in Labour Economics! Students with gritty peers in high school are more likely to succeed in the labor market. Gritty peers are pivotal for #FirstGen students. doi.org/10.1016/j.la...

My department really needs to stop hiring philosophy professors, we can't even stop the ones we have from posting on BlueSky

It goes like this the merge, the fix
A set of dummies does the trick

Dragon Ball Z meme: It's over 9000!!!!

When people ask me what my article's word count is

I love it when people explain basic LLM/AI stuff to me, it just gives me the standpoint to be an even better ally to women

A person with a microphone saying “ because I think that something awaits me in the divine for resisting“

Them: why do you keep calling yourself a “statistician” instead of a “data scientist”

Me:

Nothing accelerates theoretical clarity like a dataset that refuses to cooperate