The canvas() function from the ggview R package is very useful for previewing/tweaking a ggplot into publication-ready format: it renders a plot "as it would appear if saved to a file with the specified dimensions".
cran.r-project.org/web/packages...
#RStats #ggplot #ggview
Indeed, LLMs can generate papers, and are rapidly improving at it. But generating papers is not the purpose of science. Generating papers is not even the purpose of writing lit reviews. The purpose is, yes, "to move the frontier of [human] knowledge".
1/3
higher submission numbers because of submission fees. For publishers, a natural move seems to be to free journal space and collect much more revenue through APCs. The new bottleneck would then be researcher funding. Who can afford to spend 5x to 10x more on APCs than at present? 4/
This is attributed to the mindless use of LLMs for writing and paper mills. If serious submissions significantly increase on top of that and nothing changes on the peer review side, many journals should have rates of >75%.
2) It is an interesting note that journals/publishers benefit from 3/
across social science methods. ingorohlfing.wordpress.com/2026/02/22/t... Two notes related to the econ post:
1) If it ever meant anything, the desk rejection rate is a dead as a metric. Multiple editors report it has gone up over the past years because of obviously bad and badly fitting articles 2/
Research and Publishing Are Now Two Different Things
causalinf.substack.com/p/claude-cod... The post breaks down for econ how the expected massive increase of submission numbers will impact publishing. On the quant side, it resonates with my post on the uneven impact of LLMs 1/
Despite increasing use of observational data in public health, researchers are still hesitant to use strong causal language (for fear of overpromising).
New paper and comment in @bmj.com sets out new guidance on appropriate use/misuse of causal language
www.bmj.com/content/392/...
Der FAZ-Kommentar, in dem Malmendier mit solchen Sätzen abgetan wird: "Gemessen daran muss das Experiment Malmendier, die fern in Kalifornien lebt und arbeitet, als gescheitert gelten. Richtig Fuß gefasst hatte sie in der deutschen Debatte nie," hat mich an bekannte Hausberufungsskandale erinnert
Communicating uncertainty about our conclusions is an important scientific task.
Is reporting a point estimate and standard error good enough?
We propose a practical recipe for checking whether it is, and for improving our report when it is not. buff.ly/1RmUW1F
Sci publishers don't use DOIs in their content alert emails, do they? The URL one sees when hovering over a link in an email is not obviously a DOI. I guess they use something else to track who used a linked when? 2nd indication that DOIs are not used is that links in older emails often are dead.
📚 Special issue in Kölner Zeitschrift für Soziologie on Explanation and Causality in Sociology. Essential reading on where causal inference in sociology is heading
link.springer.com/collections/...
New blog post by me: The “LLM revolution” is likely to have uneven consequences for different social science methods
ingorohlfing.wordpress.com/2026/02/22/t...
I think adoption of LLMs will be less common among qual researchers with consequences for publishing and everything career-related to it
*maybe* this journal mistakenly thought the author submitted a registered report (or similar).
But if the position is "papers must not be pre-registered" I'd love to hear the justification.
There are packages that output intermediate results of pipes, which enhances transparency, but implies there is more output displayed or popping up in a tab. Personally, I became a bit less pipy again because I want to use more base R again.
With the advent of Claude Code and other tools, it may be possible to produce a curated journal policy database for disciplines with a limited input of resources. Ideally, publishers would maintain a database of journal policies, but this is highly unlikely to happen. 5/
do also matter in legislative decision-making. Knowing formal guidelines, but not the informal practices of a journal is better than not knowing either. Most reaons for sunsetting TOP seem to target misinterpretations of the TOP factor, not the TOP factor itself. 4/
some of the reasons that seem to account for the decision. Yes, there is a difference between stated guidelines and journal practice, but this is a common distinction that does not stop research in other fields. There is research on formal legislative procedures, not denying that informal norms 3/
hard, probably because journal guidelines are not always well organized and imprecisely formulated, as is noted in the post. To me, this is exactly the reason why a curated database is highly needed. I understand COS has limited resources and needs to set priorities. Still, I am not convinced by 2/
Sunsetting TOPFactor.org: What’s Changing and Why
www.cos.io/blog/sunsett...
@cos.io This comes unexpected. I know that the TOP factor drew criticism (what doesn't?) which is mentioned in the post.
I don't think this is a good development. Finding curated information about journal policies is 1/
Since 2020, TOP Factor has helped researchers understand how journals support open practices. On March 16, 2026, COS will be sunsetting the tool.
Read more about what prompted the decision, what we've learned, and the future of open and transparent policymaking: www.cos.io/blog/sunsett...
Meine Rede... Siehe mehrere Heisenberg Stipendiaten in meinen Bekanntenkreis, die irgendwie KEINE Uni überzeugen konnten bisher, sie mit diesen 3+2 JAHRE aufzunehmen. Ich meine... die bringen wortwörtlich über 1 Mio. mit. Aber nein. Und bevor es jemand einwirft: es liegt nicht an den Kandidaten!!
The volume of AI-enabled decent scientific work will overwhelm scientific journals and reviewers. Good post about this except it ends on a strange note, that Wiley and Elsevier AI tools might save us, which is both unlikely and would give even more power to them.
open.substack.com/pub/thebsdet...
I guess Stata is working on an AI integration bc it knows it will be toast otherwise in the medium run. At a reasonable pricing level, this may help in securing current subscribers. I am uncertain this will help in the long run bc it may not be attractive enough as an entry point to data analysis.
I'm sceptical of most "AI will kill [insert job/field/method]" takes, but I'm increasingly of the view that AI will kill Stata. What's the point in paying for it when a Claude Code Max subscription costs about the same and does the same things and more in R/Python, exponentially faster and better?
README Checklist by the Data Editors of Review of Financial Studies
review-of-financial-studies.github.io/readme.html An extensive checklist that is very useful. README files are likely something that is left for LLMs in the future, but it could be based on this template then. #OpenScience
The way I think about this is that the distribution of effects is approximately sparse (most effects are very weak but nonzero). Journals/researchers are interested in effects that are surprising given that prior distribution.
Yes but a 2c from this political scientist:
1) in polisci, which is more open to qual methods/soft data than econ, the claude code shock may mean that qual data & original measurements will increase value
2) that "hopefully" is doing *a lot* of work but again women tend to select in qual meth so 🤞
This, probably in combination with some R+R demands having marginal benefit for improving a manuscript.
Every academic in their 40s goes through this career/life stage where they decide that the journals are broken and stupid. It's mostly because they are bored of working on R+Rs given that the marginal benefit for their own career of another publication is close to zero.
New short blogpost!
There are probably many benefits of sharing clinical trial data, including:
Verifying results, better meta-analysis, understanding inconsistent results, further exploration, better clinical decision-making, learning how to run trials better, and reducing redundancy.
