Sebastian Karcher

@adam42smith

Director, Qualitative Data Repository (personal account). Data, Zotero, Social Science Methods

847 Following1,436 Followers2,515 Posts

Bluesky

Is this your profile?Claim it and adam42smith.dirty.social is yours.Claim profile

Sebastian Karcher@adam42smith· 17d

The 5yo's German is pretty limited, but he did go through the botanical garden singing "Farn, farn, farn, auf der Autobahn," so I'm quite pleased

Sebastian Karcher@adam42smith· 20d

You don't actually have to have had gay sex to qualify as bi according to the studies that get up to 30%. This graph is pretty great (it doesn't literally show the above, but reflects it) (The data source -NSFG - is reputable, even though the graph is from an IFS blogpost)

Graph showing that the percentage of people who have had gay sex is only slightly increasing, while the percentage of folks identifying as LGB has doubled between 2012 and 2020.

Sebastian Karcher@adam42smith· 29d

Contact buttons on Emily Wilson's homepage...

Contact forms for:
Interviews/Speaking Requests
General Inquiries
Misogynistic Trolling
Kind Words and Praise

Sebastian Karcher@adam42smith· 29d

This is straight out of Recoding America.

Each PIT Crew will include product managers, designers, engineers, user researchers and data experts who will work alongside City agencies and New Yorkers to design, build and launch digital solutions on accelerated timelines. Instead of waiting years for new technology, agencies will be able to move from idea to implementation in a matter of months.

Sebastian Karcher@adam42smith· Jul 11

This seems like a pretty major glitch in the AI summary:

Phone screenshot showing the ongoing Wimbledon women's final with Muchova trailing in the 3rd set, followed by the AI summary claiming she won after a 3rd set comeback

Sebastian Karcher@adam42smith· Jul 8

The +37 number is apples to oranges: a) folks can disapprove of most pols but still pref one party and b) net positives for individuals exclude "don't know enough" (high esp. for Shapiro). Notably, Mamdani's lower net + is driven all by higher disapproval Source: apnorc.org/wp-content/u...

Table from AP-NORC poll showing Shapiro with 21 unfavorable, 41 favorable, Mamdanig with 39 unfavorable, 41 favorable

Sebastian Karcher@adam42smith· Jul 7

TIL this wasn't even the worst case of interference by a US politician in soccer world cup history...

Minutes before the game, Argentinian President Jorge Rafael Videla visited the Peruvian locker room accompanied by former American Secretary of State and organiser of Operation Condor, Henry Kissinger[14] and read a message to the players from the Peruvian dictator Francisco Morales Bermúdez on the Argentine-Peruvian brotherhood.[15] Argentina subsequently won the match 6–0, rising above Brazil in the Group B table on goal difference, and advancing to the final against the Netherlands.

Sebastian Karcher@adam42smith· Jul 1

Zotero has been busy! (And finally batch edits! )

Screenshot from Zotero forum:
Available for beta testing: Undo support
Available for beta testing: Batch editing
Available for beta testing: Multi-collection selection
Available for beta testing: More-advanced advanced search

Sebastian Karcher@adam42smith· Jun 15

From a collaborator's out of office message. I appreciate this a lot...

If your message is truly urgent, please resend it with “URGENT” in the subject line. Please note that I am using “urgent” in its older, sterner sense: immediate peril, official catastrophe, or the sort of problem that causes reasonable people to say, “Oh dear,” but with genuine feeling.

Sebastian Karcher@adam42smith· Jun 15

I giggled.

Post: This is a naive take. You clearly have no idea what's going on online with kids

Sebastian Karcher@adam42smith· Jun 14

Explaining to the kids why it's important to cheer for Japan www.youtube.com/watch?v=VX5c...

Aber eine Sache gibt's
Da bin ich meganational:
Es kam über die Jahre
Und jetzt sitzt es ziemlich fest
Solang's um Fußball geht
Hass' ich Holland wie die Pest

Sebastian Karcher@adam42smith· Jun 6

I'll nominate Germany as a dark horse. German cafe/restaurant breakfast is mostly pretty forgettable, but German bakeries are generally top and nothing compares to the variety of freshly baked Brötchen you get on a weekend breakfast in a random German home. Option of cheese, sausage, nutella or jam

A basket with a large variety of rolls, some whole wheat, some lye, some whit with poppyseed and sesame

Sebastian Karcher@adam42smith· Jun 4

Oh to be young an innocent again...

‪Bluesky exchange:

The Popehat Fits Fine‬
‪@kenwhite.bsky.social‬
”If you question or criticize Platner you are rooting for Collins” is the new “objectively pro-Saddam.”

Jacob Alexander Tice
‪@daemonspudguy.jacobatice.com‬
What's the context on the Saddam thing?

Sebastian Karcher@adam42smith· Jun 4

Also -- SSRN is not and has never been "open source". Words have meaning.

Screenshot of title and subtitle from Bainbridge's substack with the words "major open source" highlighted by me:
The Social Science Research Network Has Jumped the Shark.
Significant and unwelcome changes to what had been legal academia's major open source scholarly repository and notice system
Stephen Bainbridge

Sebastian Karcher@adam42smith· Jun 1

Me reading the abstract: Well, 'coding error' is a bit strong, isn't it, they just disagreed on some variable coding. Me getting to p. 3: Oh, OOOHHH, OH NO!!!!

B&B grouped the 71 teams into 13 disciplinary categories without providing a theoretical justification for such fine-grained coding. This choice has significant consequences: four of these categories contain only a single team. Consequently, the dummies for these categories identify individual teams, functioning equivalently to team fixed effects. Since ideology is measured at the team level and does not vary within teams, these four teams—and their 80 reported AMEs—are effectively excluded and do not contribute to the estimation of the ideology effect.

Sebastian Karcher@adam42smith· May 30

When the hitpieces end up sounding like profiles in wholesomeness.

Talarico and Menard frequently attend music concerts — including visiting Austin City Limits every year — and like to hike at Big Bend National Park annually, as well as take other trips, photos obtained by The Post show. They’re also avid Texas Longhorns college football fans and San Antonio Spurs devotees.

Brianna also has Texas roots with one ancestor, Michael B. Menard, having both a West Central state county named after him. He also signed the Texas Declaration of Independence.

Sebastian Karcher@adam42smith· May 23

At a kids theater show and this is among the music directors credits... Simultaneously least likely book to become a musical & most likely one to be a fringe show...

Kafkas Metamorphosis: The Musical, performed at various fringe festivals

Sebastian Karcher@adam42smith· May 19

Just saying....

Line chart showing smartphone penetration (left axis, blue) and estimated annual count of "bad" difference-in-differences papers (right axis, red) from 2000 to 2022. Both series rise sharply after 2007, the year of the iPhone launch. Bad DiD papers increase from near zero to approximately 4,550 by 2022. A confidence interval is shaded around the red line. Annotations note the iPhone launch, researchers' discovery of "staggered rollout," and the pre-publication status of Callaway & Sant'Anna at a key inflection point.

Sebastian Karcher@adam42smith· May 4

I'd guess attendance effects. Schools are still struggling with attendance -- it's worse with older kids, but definitely starts at the elementary level. Here are chronic absenteeism percentages for our (very high poverty, urban) school district

Chronic absenteeism percentage in Syracuse city schools. At elementary/middle school level, pre-pandemic number was 35%, went up to 53% in 2021/22, still at 41% in 24/25

Sebastian Karcher@adam42smith· May 1

One thing we'd really like to see is wider sharing of consent forms by researchers (maybe even a database?) _and_ IRBs willingness to pre-review more than just a narrow set of local template to allow researchers to use appropriate consent w/o massive effects on IRB review duration

One promising way forward is to develop a modular, open-access repository of consent tools that can be adapted across studies. In the biomedical sciences, templates for eConsent, key info sheets, and comprehension assessments are increasingly available. A similar effort in the social sciences could produce adaptable scripts, data sharing agreement templates, and visual aids, along with pedagogical instructions on how and when to use them. Participants at the workshop emphasized the need for practical, ready-to-use materials that lower the barrier to ethical innovation. Currently, in many instances, informed consent scripts are borrowed from prior studies, templates provided by IRBs, and other researchers, sometimes without the necessary rethinking of new research realities. This common practice can be leveraged to a better end. Rather than appealing to individual researchers to reinvent appropriate consent documents from scratch, institutions, journals, and funders could support centralized repositories of best practices addressing many different research scenarios. Such resources would be particularly valuable for early-career researchers, those operating in resource-constrained settings, or anyone navigating complex cross-cultural or interdisciplinary research. IRBs can facilitate the adoption of appropriate consent forms by pre-reviewing a larger range of forms so that researchers are not constrained to a single institutional template. Importantly, these tools should be iterative and flexible, informed by user feedback, including from research participants themselves.

Sebastian Karcher@adam42smith· May 1

Principle 2: Designing Consent Sheets Consent is too often a wall of text and many studies show it's often ignored, esp. in the social sciences. Keeping is short is important, but also visual presentation, as shown by Mary Politi's group's work: bmjopen.bmj.com/content/15/3...

Visual informed consent template with 8 information boxes, each with a heading and an icon and 2-3 bullet points

Sebastian Karcher@adam42smith· Apr 29

(I'll stop after this, promise). "Reasonable request" is used a lot in US Federal Regs, here a 1976 reporting requirement on interstate commerce. I suspect it made its way into data sharing language from there & I think timing is indeed early 2000s with NIH 2003 policy & the NA report I link.

Screenshot: "Availability of underlying data. All underlying data used in preparation of the material outlined above shall be made available for inspection upon reasonable request"

Sebastian Karcher@adam42smith· Apr 29

I thought it might be the APA ethics rules, which imply this, but they don't include the word 'reasonable'. www.apa.org/ethics/code

8.14 Sharing Research Data for Verification
(a) After research results are published, psychologists do not withhold the data on which their conclusions are based from other competent professionals who seek to verify the substantive claims through reanalysis and who intend to use such data only for that purpose, provided that the confidentiality of the participants can be protected and unless legal rights concerning proprietary data preclude their release. This does not preclude psychologists from requiring that such individuals or groups be responsible for costs associated with the provision of such information.

Sebastian Karcher@adam42smith· Apr 26

I see a full book button at the top for Cambridge OA books, e.g. www.cambridge.org/core/books/w...

Landing page of Working with Concepts book, showing a blue "Full Book PDF' button above the list of individual chapters

Sebastian Karcher@adam42smith· Apr 9

technically they are not saying "causal" 😝

🚨New preprint and our results are rather concerning..

We find the "boiling frog" equivalent of AI use. Using large-scale RCTs, we provide *casual* evidence that AI assistance reduces persistence and hurts independent performance.

And these effects emerge after just 10–15 minutes of AI use!

Sebastian Karcher@adam42smith· Apr 8

This is super cool work. It's also interesting that the uni-dimensionality of LLM fiction corresponds to LLMs flattening respondents views to demographic vars & single ideological dimension in 'synthetic' data as per @cbarrie.bsky.social and Cerina's recent paper: osf.io/preprints/so...

Ridge plot showing massively higher values for all LLM models than for GSS human participants.

Caption reads:" Figure 3: Global dependence by education. Bootstrap distributions of effective dependence, De = 1−|R|1/p, computed on the full polychoric correlation matrix (attitudes + persona variables) for the GSS and each LLM, shown overall and stratified by education. Right-side lollipops show the education gradient ∆ = De(high edu) − De(low edu): the point is the median and whiskers denote the 95% bootstrap interval. Lollipops are dark when the 95% interval excludes zero (light otherwise). Higher De indicates stronger global linear dependence in the belief system."

Sebastian Karcher@adam42smith· Apr 6

Yeah, I was looking just now and it's like --- rainbows are pretty, but not what your status history should look like:

Claude.ai status for the last 90 days with a lot of yellow and red especially in the last month

Sebastian Karcher@adam42smith· Apr 3

and that's why we don't switch. (This, btw., is a 'practice goat', according to her owner, an 11 year old)

A little goat with giant ears looking at you

Sebastian Karcher@adam42smith· Apr 2

3/3 I really liked this part of the study. Whereas 'old-school' bots gave you lousy answer, LLM agents will give you answers that may be _too_ good to be true: they're too consistent, too 'engaged', 'read' too closely, etc.

Sample qualities along eight dimensions: Atention, Comprehension, Honesty, Engagement, COnsistency, Open-ended quality, Overall Quality.
LLM survey takers outperform human average along all 8 and outperform best human samples (direct panles) on several dimensions.

Sebastian Karcher@adam42smith· Mar 31

Yeah, so reading reviews from other sources would obv have worked here -- see attached the most egregiously similar para (Guardian left, NYT right). But this e.g. was a translated book published in multiple languages, so would NYT be responsible for checking Le Monde, Liberation, FAZ, El Pais...?