Lightnews — Scholar-powered news

Reposted by Thiago

Mark A. Hanson

@hansonmark.bsky.social

The solutions of the past 3 decades have failed to change the incentives of #PublishOrPerish. As a result, researcher funding, time, control, and trust has been lost.

The ONE CONSTANT in the wake of the serial crisis, #PlanS and #OpenAccess reform has been publish profit margins.

2/n

A table showing profit margins of major publishers. A snippet of text related to this table is below.

1. The four-fold drain
1.1 Money
Currently, academic publishing is dominated by profit-oriented, multinational companies for
whom scientific knowledge is a commodity to be sold back to the academic community who
created it. The dominant four are Elsevier, Springer Nature, Wiley and Taylor & Francis,
which collectively generated over US$7.1 billion in revenue from journal publishing in 2024
alone, and over US$12 billion in profits between 2019 and 2024 (Table 1A). Their profit
margins have always been over 30% in the last five years, and for the largest publisher
(Elsevier) always over 37%.
Against many comparators, across many sectors, scientific publishing is one of the most
consistently profitable industries (Table S1). These financial arrangements make a substantial
difference to science budgets. In 2024, 46% of Elsevier revenues and 53% of Taylor &
Francis revenues were generated in North America, meaning that North American
researchers were charged over US$2.27 billion by just two for-profit publishers. The
Canadian research councils and the US National Science Foundation were allocated US$9.3
billion in that year.

November 11, 2025 at 11:52 AM

Reposted by Thiago

Andrew Heiss

@andrew.heiss.phd

Wrote up a little intervention post/explanation for my class about why using LLMs for trying to learn programming (as first time learners!) is bad and detrimental datavizf25.classes.andrewheiss.com/news/2025-11...

In the first week of class, I had you read an article about how ChatGPT is philosophical “bullshit”, as well as this other post about AI and LLMs in this class.

I’ve seen a surprising increase in the amount of ChatGPT-based weekly check-ins and graph interpretations in the exercises. (In the past, I’ve even had students tell me they PDFs of the slides into ChatGPT to generate interesting and confusing things.) Don’t do this! LLMs cannot find three things that you personally found interesting from the readings. LLMs are not you.

It’s a huge waste of time for all of us to for me to spend time evaluating and grading output from ChatGPT. I want to read what you think and do, not what a computer invents.

NoteMy personal stance on AI and LLMs
Many of you have asked about how I think about and use LLMs in my own work (and in the class materials). Please check this short little page where I explain what I do with text, images, code, research, and learning.

In short, I do not use LLMs to create any of the materials for this class, and I do not use them to grade or evaluate any of your work.

I’ve even had several students submit completely made up stuff for mini projects and final projects. In your final project, your job is to take a dataset, make three different plots with it, and combine the plots into one graphic that tells some sort of story. I’ve (really truly) received projects like this:

(a ChatGPT-generated plot of completely made up data from gapminder)

At a very quick glance, that looks like a plausible image. But holy crap it falls apart fast, both comically and tragically. Comically because, like, look at the y-axis in the scatterplot; or the x-axis ticks in the population grown chart that are sometimes spaced every 20 years, sometimes every 10 years, and has 1980 there twice; or the weird blobs in the map.

Tragically—and most importantly—literally nothing in this image is real and none of the annotations make sense and none of the colors make sense and everything here is completely meaningless and devoid of any information. It’s pure philosophical bullshit—an image “produced without concern for its truth.”

That’s what I mean when I say that it’s a forgetful and unreliable resource. Lots of that code is helpful as an illustration of what’s going on. Lots of the explanation is good and fine. But it subtly changed its mind halfway through and didn’t realize it (because it can’t realize it! it’s just a chain of statistically likely text!)

Hopefully you can see where this is dangerous for learning. It can often “teach” you wrong, and you can only recognize that if you know stuff already.

It’s extra dangerous if you then take that example output, paste it into your exercise, and submit it.

I’m not even concerned about the plagiarism aspect of that—code is code and there are only a few ways to write {dplyr} code that will do this, and it’s easy to adapt from real human-based examples online, so I’m less worried about stealing other people’s work here.

I’m concerned that it’s not real data.

It made up values of 1, 2, 3, and 4 for some countries to illustrate grouping and summarizing. That’s good and normal.

None of that is real. Don’t treat it as real.

November 2, 2025 at 10:17 PM

Thiago

@csthiago.bsky.social

www.nature.com/articles/d41...
This ultra-competitive scenario makes me very anxious.
1) almost a lottery
2) It is possible that the lottery would be more fair, as at least it wouldn't have the conscious/unconscious bias (mainly the "weights" the institution of origin carries)

Is academic research becoming too competitive? Nature examines the data

Applications for European research grants increased in 2025. Scientists say they’re feeling the competition.

www.nature.com

October 20, 2025 at 12:39 PM

Thiago

@csthiago.bsky.social

@grahamkendall.bsky.social you could check the pattern of self citations of this group : journals.plos.org/plosone/arti...
Every paper of them in this topic cite almost all of the previous ones. Rarely citing other groups

Evaluating the cost-effectiveness of COVID-19 mRNA primary-series vaccination in Qatar: An integrated epidemiological and economic analysis

Qatar implemented a mass primary-series vaccination campaign to mitigate the impact of the coronavirus disease 2019 (COVID-19) pandemic. This study aimed to retrospectively evaluate the cost-effective...

journals.plos.org

October 19, 2025 at 4:51 PM

Reposted by Thiago

Nature Communications

@natcomms.nature.com

A study of ~7 million Brazilian births shows that infection with #chikungunya, #dengue or #Zika viruses during pregnancy is associated with increased risk of adverse perinatal outcomes

@csthiago.bsky.social

#viruses #epidemiology #pregnancy

Perinatal outcomes of symptomatic chikungunya, dengue and Zika infection during pregnancy in Brazil: a registry-based cohort study - Nature Communications

In this study of ~7 million Brazilian births, the authors show that chikungunya, dengue, or Zika infection during pregnancy is associated with an increased risk of adverse perinatal outcomes, with the magnitude of these effects varying by trimester of exposure.

bit.ly

October 15, 2025 at 6:15 PM

Thiago

@csthiago.bsky.social

Help!
Where can I find resources about causal mediation of ordinal outcomes? (using SEM)

Thanks 😄
#rstats @dingdingpeng.the100.ci

September 29, 2025 at 2:27 PM

Reposted by Thiago

Fjorida Llaha

@fllaha.bsky.social

I’ve had the same impression while looking at academic epidemiology postdoc offers. So much focus on AI, machine learning, LLM experience. Are these the core skills we need now? What about subject matter knowledge, study design, bias, statistics, population health knowledge? Outdated?

Lorenzo Fabbri @epilorenzofabbri.com · Sep 19

Looking at job offers (both academia and industry) and it’s basically all “AI” & Co. 😔
And WTF is “experience with LLM” for an academic post in epidemiology??

September 20, 2025 at 5:01 PM

Thiago

@csthiago.bsky.social

Sleuths,
Can someone explain to me how these CIs are possible?
www.nejm.org/doi/full/10....
It is not clear how they got RR from the Cox PH, as it would require choosing a time point (looks like an interpretation of HR as RR)

@lonnibesancon.bsky.social @sophieehill.bsky.social @retractionwatch.com

September 18, 2025 at 10:06 AM

Thiago

@csthiago.bsky.social

1/2
Does anyone know how to proceed for multiple imputations if one of the variables (fully observed) in the model has more than 5000 levels (in the outcome model, it is a mixed effect)?

#rstats #episky

September 5, 2025 at 7:10 AM

Thiago

@csthiago.bsky.social

Check out our last paper about the adverse consequences of arbovirus infection during pregnancy:
www.nature.com/articles/s41...

We have use the exact same approach for dengue, zika and chikunugnya. (including a negative control exposure!)
#dengue, #zika, #chikungunya, #epidemiology