Профиль

I'm back to a condition where I can't consider methodology (I'm on a bus heading to a glacier) but my reply was to your comment about how laying out the data this way prevents all in cohort comparison. It is still trivial, you just pivot on the label. If that's not what you went, sorry.

I think a lot about how I wanted to go on to a PhD but couldn't afford to do it (and my family, being angry, tore out every one of the limited supports I had in place. And then I see folks who have PhDs, my age, where I think "I could have been doing a job like that. This isn't fair." Alas.

Yeah, that's how I feel too (although I think I'm less afraid of regressions because of the methodologies I prefer to use).

To the first part yes: each id should appear twice. To the second part no, its not any more difficult.

So long as we set allegations of p-hacking aside, I agree this paper isn't the best I've ever read. It's just an effect strong enough to be worth investigating.

That's not "collective ownership" in the public sense, it's "collective ownership" in the syndicate sense.

I'm not sure that Joel wound necessarily find it a burn. He's a nebula youtuber and they're all convinced that the only problem with the copyright system is that they don't collectively own the publisher, hence Nebula.

I think the long-form-fixed-shape format is extremely common now because of the benefits of a uniform shape when using it with dataframe libraries and many R and Julia libraries expect this sort of shape.

But there are pre and post scores for each lecture for AI and not AI. Using visidata, I can see distinct IDs repeated with the "lect_AI" labeling field. So, is your objection that you'd prefer a wide data format rather than a long data format?

The analytic dataset is public. I can just look at it. Please substantiate your very serious claim with the actual rows in the data.

Well, the analysis reports 316 session observations. - 142 AI - 174 classroom One session for each would be 388. They have 316. There is data missing, but it's not half. What am I missing here?

That's a very serious accusation. What action did you think they took *after* seeing the results? The typical definition is "exploiting a researcher'a degree of freedom to misrepresent the data". Which one do you believe the data shows they exploited?

Well if you don't think it's a problem that's fine by me. But what do you mean by "half the data?"

Are you suggesting that half the data is median and the other half is mean?

I and others who've looker don't see "half the data" missing. I don't know where you get "half the data". But this format isn't very good for discussing the finer points of data analysis. Maybe you wanna write up something?

I don't see why this affects the outcomes? I think your criticism about the unexplained exclusions is good, I share it. It's why we need replication. But we can at least verify that the regression was done correctly and the effect size is big enough that even quartered it'd be worth pursuing.

1. This New York Post cover: “Thousands of 9/11 families demand Mamdani be banned from 25th-Anniversary” is a brazen lie. How do I know? I interviewed the organizer of the petition.

And it's a preprint, ymmv: I find the conclusions here in line with the hypothesis I've been circling in on: AI assistance is a multiplier on student learning, students who are good and motivated get results. Students who are just trying to pass the test don't see durable learning either way.

There are some papers I've been trying to get access to without blowing my research budget like this one.

Things are a bit less chaotic this morning than last time we talked so I have my notes, not just "hey Google and ChatGPT help me find what I recall I read on the Kestin paper" mode. There has been a check on the data from a source I like a lot. It gives reasonable caveats and just checks the data.