@ingorohlfing@mastodon.social

Ingo Rohlfing @ingorohlfing@mastodon.social

· 5mo ago

On the (Mis)Use of "On the Mis(Use) of..." in article titles (here 2x in same issue):
- On the (Mis)Use of Machine Learning With Panel Data
https://onlinelibrary.wiley.com/doi/full/10.1111/obes.70019
- On the (Mis) Use of the Fixed Effects Estimator
https://onlinelibrary.wiley.com/doi/10.1111/obes.70031

onlinelibrary.wiley.com

2

0

Ingo Rohlfing @ingorohlfing@mastodon.social

· 4mo ago

Replying to

@druedin@sciences.social I admit I belong to those reviewers who tend to wait until the deadline is close. In this view, tighter deadlines do improve turnaround times.

1

0

Ingo Rohlfing @ingorohlfing@mastodon.social

· 5mo ago

Kleiner Statistikfehler in der Überschrift der SZ. Die Rate der Kaiserschnitte hat sich von 0,15 in 1991 auf 0,33 in 2025 erhöht, was so im Artikel steht. In 1991 war die absolute Anzahl der Geburten allerdings höher als in 2025 (830019, https://de.statista.com/statistik/daten/studie/235/umfrage/anzahl-der-geburten-seit-1993/). Absolut gesehen ist die Zahl der Kaiserschnitte in 2025 daher um ca. 73% höher, nicht mehr als 100%.

de.statista.com

1

0

Ingo Rohlfing @ingorohlfing@mastodon.social

· 5mo ago

Replying to

universities and other teaching- and research-centered institutions without an elaborate application process. This has its own downsides, but would be the more radical consequence than infusing the funding system with more AI. 5/

1

0

Ingo Rohlfing @ingorohlfing@mastodon.social

· 5mo ago

RE: @avatter@mastodon.social

Es geht in dem Artikel ja nur um Open Data, nicht um Open Access. (Ironie off)

mastodon.social

1

0

Ingo Rohlfing @ingorohlfing@mastodon.social

· 11mo ago

Replying to

@matherion @auzdavenice I fully agree. The fact that many concepts and ideas originated in the quantitative domain make them appear "quantitative". This makes it harder to argue that they can be also used in qualitative research (maybe not all, but some). Seen from the other side, it makes it easier for those opposing these ideas to argue that they are not appropriate for qualitative studies. (based on my experience in political science)

1

0

Ingo Rohlfing @ingorohlfing@mastodon.social

· 11mo ago

Replying to

@auzdavenice Preregistration can always be done, yes. I meant a preanalysis plan as a standalone plan or part of a registered report with credible prior unaccessibility of the data. This probably can be done more often in observational research than many may think, but not always. As opposed to experiments.

1

4

0

Ingo Rohlfing @ingorohlfing@mastodon.social

· 45mo ago

Replying to

@uebernerd@nerdculture.de It would be good to open more file drawers (or hard drives, nowadays), but I don't think this is a disadvantage because barely anyone reads final project reports

2

0

Ingo Rohlfing @ingorohlfing@mastodon.social

· 12mo ago

Replying to

I guess the mechanism is not clear. Could be randomization levels the playing field and diversifies pool of recipients, or that more researchers feel encouraged to submit who would have also got funded without randomization. In either case, randomization seems to have a diversifying effect 2/

0

3

0

Ingo Rohlfing @ingorohlfing@mastodon.social

· 5mo ago

Replying to

I know that many work on AI-assisted or AI-based individualized learning, which may or may not work. I believed this would not be something a university pursues actively, at least not one that charges tuition fees because this is not something one can charge much for. Or what am I missing here? 2/

0

Ingo Rohlfing @ingorohlfing@mastodon.social

· 5mo ago

Replying to

@philipncohen@mastodon.social The demonstration that an LLM can produce a "publication ready" regression table that probably would be accepted as legit at least 7 times out of 10. (this is my guess) I am not sure such invention of regression tables is the biggest threat. 1) In this case, the data is publicly available, probably making it a matter of time until someone would fail to replicate the results and start asking questions. 2) Without a data-sharing requirement, the bigger risk may be that 2/

0

1

0

Ingo Rohlfing @ingorohlfing@mastodon.social

· 5mo ago

Replying to

These developments make the choice of applications for funding harder, so what to do? Article seems to settle on an "AI-native funding systems": AI can check how likely it is that the applicant can deliver on the promises; prediction models can evaluate most-likely breakthrough applications etc. 2/

0

2

0

Ingo Rohlfing @ingorohlfing@mastodon.social

· 5mo ago

Replying to

@philipncohen@mastodon.social someone pretends to have collected original data through a survey or some other data analysis method. This would make reproduction impossible without a data-sharing mandate. 2) We may see vibe-coded quantitative analysis sooner or later (maybe someone knows an example already). This means a dataset is feed into an LLM and the results are based on prompts only. The LLM would be able to produce code that supposedly would produce the results if used in R, Python whatever. 3/

0

1

0

Ingo Rohlfing @ingorohlfing@mastodon.social

· 5mo ago

Replying to

@philipncohen@mastodon.social Unless the researcher validates this, which is imo more unlikely than likely if one does a prompt-based analysis in the first place, the code is worth nothing. Plus the findings may be non-reproducible by the original researcher when the LLM is altered, which happens regularly, I think. With mandatory code sharing, this would be relatively easy to detect once a third party tries to verify the findings. Regardless, would be interesting to know why ASA has no policies adopted. 4/

0

1

0

Ingo Rohlfing @ingorohlfing@mastodon.social

· 5mo ago

Replying to

The article is attentive to the problems of using an "AI native" funding model, such as perpetuating biases into the process. In my reading, these risks are downplayed by saying that can be addressed. So, in my view this does not read like the funding system will be toppled, but perpetuated. 3/

0

2

0

Ingo Rohlfing @ingorohlfing@mastodon.social

· 5mo ago

Replying to

As agentic LLMs become more sophisticated, the application workload may decline even when the number of applications goes up, potentially supplemented by submission caps etc. However, the overall system would remain intact. imo, a toppled funding system would mean to redirect the money to 4/

0

2

0

Ingo Rohlfing @ingorohlfing@mastodon.social

· 2mo ago

Replying to

Das Faculty Network ist nach Themen sortiert, die sich alle auf Inhalte (Policies) beziehen: https://asg.uni-koeln.de/faculty-network Das ist zweifellos wichtig und die beteiligten Personen sind ausgewiesen. Governance braucht dennoch auch Umsetzung von Inhalten (Politics). In einigen Bereichen gibt es m.E. kein Erkenntnisproblem, sondern eines in der Umsetzung. 2/

asg.uni-koeln.de

0

1

0

Ingo Rohlfing @ingorohlfing@mastodon.social

· 3mo ago

Replying to

Substantively, I am skeptical about Q3, implications of the findings. I understand students are encouraged to think about what their findings mean beyond rejecting H0 or so. However, It seems to overstate what one can and should follow from one study. At least, the specification of substantive implications should be complemented by a discussion of the limited value of single studies, evidence synthesis etc., adding more constraints on time. 2/

0

Ingo Rohlfing @ingorohlfing@mastodon.social

· 2mo ago

Replying to

Das Fehlen von Politics zeigt sich darin, dass nur ein ausgewiesener PoWi (Thomas König) im Gründungsteam ist. Soziologie ist etwas stärker vertreten, aber der Fokus liegt auf Wirtschaftswiss. und Recht. Der Webseite zufolge wird das Netzwerk nach und nach erweitert, dann hoffentlich um eine Perspektive auf Politics. 3/

0

Ingo Rohlfing @ingorohlfing@mastodon.social

· 11mo ago

Replying to

@nicebread So, this means that non-preregistered studies should be evaluated more harshly because there is less transparency? I can imagine some social scientists (political science, sociology) would disagree if they cannot preregister their observational research. This maybe different for experiment-heavy psychology. In the end, there will always be a concern that some standard advantages one kind of research and disadvantages another.

0

10

0

Ingo Rohlfing @ingorohlfing@mastodon.social

· 11mo ago

Replying to

@tomstafford @nicebread This would be a meaningful comparison. With regard to trusting reviewers, one has to trust them, but I am not confident how much this trust is earned, on average, given my review experiences

0

Ingo Rohlfing @ingorohlfing@mastodon.social

· 4mo ago

Are journals reducing the maximum turnaround time for peer review because they expect one to be late? Meaning that one is having a normal turnaround time of about 25-30 days (afaik, this is normal) instead of 40-50 days? I am asking because I just got invited to review a regular paper within 10 days. It is doable, but unusually short.

0

2

0

Ingo Rohlfing @ingorohlfing@mastodon.social

· 2mo ago

Ich habe in meiner Statistikveranstaltung aus aktuellen Anlässen diese Grafik vom RKI gezeigt, um zu illustrieren, wie sie Unsicherheit ausweisen über das Prädiktionsintervall. (aus https://www.rki.de/DE/Aktuelles/Publikationen/Epidemiologisches-Bulletin/2025/19_25.pdf?__blob=publicationFile&v=4). Kritisch und gleichzeitig leicht amüsant finde ich die Kategorie "grenzsignifikant". Die hätte es m.E. nicht gebraucht und hat mich hieran erinnert. https://mchankins.wordpress.com/2013/04/21/still-not-significant-2/

rki.de

0

Ingo Rohlfing @ingorohlfing@mastodon.social

· 3mo ago

Arguing with economists: the case for preregistration
https://kdoroc.substack.com/p/arguing-with-economists-the-case
Assuming this represents a broader picture in econ: I would have assumed that econ was more open to preregistration than other disciplines. This post suggests differently.
Attitudes toward preregistration are probably more a generational issue and less a disciplinary one. I think new generations of researchers are likely to be more open to prereg across disciplines.

kdoroc.substack.com

Arguing with economists: the case for preregistration

At a recent economics conference there was a long discussion on preregistration where we heard questions and comments from ~20 different people at varying levels of seniority.

0

Ingo Rohlfing @ingorohlfing@mastodon.social

· 3mo ago

A Comparison of Packages to Generate Codebooks in R | Crystal Lewis
https://cghlewis.com/talks/rladies_nyc/ This is a short and very useful overview over the alternatives for codebook generation. #rstats

cghlewis.com

0

Ingo Rohlfing @ingorohlfing@mastodon.social

· 2mo ago

Causal inference and effect estimation using observational data
https://jech.bmj.com/content/76/11/960 It's in the rubric "Glossary", which it isn't, but it's close. This is a very concise summary of causal inference, effect types and related terminology one needs to know. This could serve as a stepping stone in a causal-inference class for an in-depth discussion of the different dimensions of causal inference.

jech.bmj.com

0

Ingo Rohlfing @ingorohlfing@mastodon.social

· 5mo ago

Replying to

@petersuber@fediscience.org I assume this service does not come for free. Any information about how much this costs?

0

2

0

Ingo Rohlfing @ingorohlfing@mastodon.social

· 12mo ago

Replying to

@tomstafford@mastodon.online True, the more diverse applicant pool most likely derives from encouragement. The recipients of awards are also more diverse, which could be because randomization interferes with positive feedback and networks in peer review, or because the applicant's pool is more diverse, or both.

0

1

0

Ingo Rohlfing @ingorohlfing@mastodon.social

· 5mo ago

Replying to

@druedin@sciences.social @philipncohen@mastodon.social Interesting point. It is probably correct because, as the post notes, the model likely has been trained on similar research. Would then be interesting to let an LLM produce results for a more or less completely new phenomenon that is not part of the training data. The challenge then may be that no one has a good benchmark for evaluating the LLMs results, except for plausibility, which maybe good enough as a starting point.

0

1

0

Ingo Rohlfing @ingorohlfing@mastodon.social

· 4mo ago

Replying to

@druedin@sciences.social An indicator for what? For prestige because what counts as highly prestigious journals has low intended turnaround times?

0

1

0