iiinova — where your digital self keeps evolving

Jason Radford, PhD @jsradford.bsky.social · 05/10/2026

This is on my list but I think it's going to go the way of Monster and Gintama - bits and bites over the better part of several years (or decade for Gintama at this point)

000

Jason Radford, PhD @jsradford.bsky.social · 05/10/2026

The big surveys should feel more like the NASDAQ where the constructs measured and mixes change as the underlying society and social theories change. (They do change but longitude for longitude sake isn't that valuable for the cost of these surveys).

010

Jason Radford, PhD @jsradford.bsky.social · 05/10/2026

Was having a conversation earlier about the ANES and GSS and how survey research has been handcuffed to weird questions that get weirder over time to maintain longitudinal integrity. (Looking at you authoritarianism.) There's value of course but we've tied ourselves to measures over constructs.

110

Jason Radford, PhD @jsradford.bsky.social · 23/09/2026

These new classes of LLMs seem to make the classification process more opaque, making it more difficult to troubleshoot and improve. At the National Internet Observatory, we're paying close attention to the field because our researchers need a path to fast classification at scale.

010

Reposted by Jason Radford, PhD

James Medlock @jdcmedlock.bsky.social · 17/09/2026

Republicans, in their infinite wisdom, left a big gambling loophole in the Medicaid work requirements. I'm no fan of the gambling industry, but if you are on Medicaid and struggling to deal with the work requirements, taking a $580 bet on both sides of an outcome is one way of getting around them.

401998531

Jason Radford, PhD @jsradford.bsky.social · 08/09/2026

Want to explore it yourself? We turned our documentation into a NotebookLM so you can work through study designs with it directly: notebook.google.com/notebook/8bb...

notebook.google.com

100

Jason Radford, PhD @jsradford.bsky.social · 08/09/2026

If you missed us at #APSA2026, here's where to catch up. Our data documentation walks through what data we collect and how: drive.google.com/drive/folder...

drive.google.com

Data Descriptions for Researchers (PUBLIC) - Google Drive

120

Reposted by Jason Radford, PhD

Brendan Nyhan @brendannyhan.bsky.social · 07/09/2026

Interesting new research on the effects and content of Instagram Reels and podcasts ceciliaysui.com/files/Influe... preprints.apsanet.org/engage/api-g... link.springer.com/epdf/10.1007...

14411

Jason Radford, PhD @jsradford.bsky.social · 06/09/2026

Finally got into the second half of John Levi Martin's book and it's been so illuminating in thinking about field theory. This essay does a great job of laying out the many ways AI, society, and social science intersect and hang together. Definitely a good time to drill into your essay yet again.

020

Jason Radford, PhD @jsradford.bsky.social · 06/09/2026

NIO just hit 30 onboarded projects this past week. Researchers are now using NIO data to study things like how AI shapes access to government services, the dynamics of change.org petitions, and cross-platform content curation. nationalinternetobservatory.org/approved-pro...

nationalinternetobservatory.org

Approved Projects

The National Internet Observatory aims to help researchers understand how people behave online and how platforms structure what people see. This will be accomplished through creating a large panel of ...

000

Jason Radford, PhD @jsradford.bsky.social · 02/09/2026

You can find their work and the qualitative coding schemas they used at their website - www.ai-observatory.org

000

Jason Radford, PhD @jsradford.bsky.social · 02/09/2026

The big takeways for data collection is 1) we need data collection that's focused on gathering data from different models embedded in different places and 2) data collection has to be continuous so we're able to track the very dyanmic changes in how people use AI and how AI responds.

100

Jason Radford, PhD @jsradford.bsky.social · 02/09/2026

At NIO, we've seen the explosion of exposure to AI content most clearly in the increase in AI summaries in search. But even in surveys, we see high penetration of intentional AI use.

100

Jason Radford, PhD @jsradford.bsky.social · 02/09/2026

Second, how society relates to AI is changing quickly. Social interactions with AI models have been rising since 2023. Conversations have been getting longer since 2023 with more turns and more tokens. Even different versions of the same company's model produce distinct patterns of use.

100

Jason Radford, PhD @jsradford.bsky.social · 02/09/2026

I doubt this is a model-level phenomenon - grok isn't better at politics. Gemini isn't "the social model." I think it's a delivery phenomenon - grok data is laregly created by X users for X. Anthropic is an enterprise product sold mostly to big dev teams. The point remians, uses vary WIDELY.

100

Jason Radford, PhD @jsradford.bsky.social · 02/09/2026

First takeaway: people use different AI very differently. Grok skews toward politics and current events, Anthropic's models toward coding, Gemini toward more social use, OpenAI toward longform writing and homework help. No single company's report tells the whole story.

110

Jason Radford, PhD @jsradford.bsky.social · 02/09/2026

Researchers from 14 institutions looked at data from a variety of different sources of human-AI data, including the National Internet Observatory. They wanted to investigate what could be learned from comparing the datasets. Here's a couple of key takeaways:

100

Jason Radford, PhD @jsradford.bsky.social · 02/09/2026

In the API age, scholars could study people's public behavior at scale. In the AI age, there is little public data. There can't be any APIs. So how can we study AI? There's a new paper out using National Internet Observatory data (among others) to tackle this challenge.

100

Jason Radford, PhD @jsradford.bsky.social · 31/08/2026

Journal of Quantitative Description? Munger has done an excellent job getting that set up over there.

110

Jason Radford, PhD @jsradford.bsky.social · 31/08/2026

AI continues to fundamentally reshape politics. NIO is capturing what Americans see and do online, giving you the chance to study these changes in real time. Come to our #APSA2026 shortcourse to learn how to study politics and AI just in time for the midterms. connect.apsanet.org/apsa2026/pre...

connect.apsanet.org

Pre-Conference Short Courses

000

Jason Radford, PhD @jsradford.bsky.social · 27/08/2026

For all the political scientists coming to #APSA2026 in Boston, @pranavgoel.bsky.social will be teaching a short course on Wednesday Sept 2nd: "Collecting online activity data with the National Internet Observatory." Come learn about how to use NIO to study what people are actually doing online.

130

Jason Radford, PhD @jsradford.bsky.social · 24/08/2026

A macroeconomist can use it as a raw measure of commercial activity; communications scholars can study the concentration of all digital flows, and political scientists and sociologists can look at the shift in data flows around different events.

100

Jason Radford, PhD @jsradford.bsky.social · 24/08/2026

Network data gives the rawest measure of human digital activity. It's literally tracking whether any data is moving between a person and some external entitiy. This means you can use it to study a person's aggregate digital activity, regardles of mobile app (mobile apps send data lots of places)

100

Jason Radford, PhD @jsradford.bsky.social · 24/08/2026

NIO collects client-level data so that independent researchers can see how people really move through the internet. The network data is tied to surveys and cross-device activity, giving researchers real traffic tied to real people.

100

Jason Radford, PhD @jsradford.bsky.social · 24/08/2026

Researchers need to know what's changing, what's risky, what's working. But the data sits locked inside ISPs like AT&T and Comcast. When they share traffic data, it's stripped down past the point of tying anything to a real user or app, and it's just one company's slice of the internet.

100

Jason Radford, PhD @jsradford.bsky.social · 24/08/2026

Almost no one outside a handful of companies like Akamai knows how the internet actually works for people at any given time. NIO put together the data needed to do this. Last week, Scott Cambo and Dave Choffnes showed researchers at ACM SIGCOMM how to use NIO to study how data is actually flowing.

100

Jason Radford, PhD @jsradford.bsky.social · 23/08/2026

Empirical relativism has proven a lot more dangerous than so-called moral relativism.

020

Jason Radford, PhD @jsradford.bsky.social · 17/08/2026

I can hear this post very clearly. Also, if you heat the milk and add lemon juice, it'll curdle and you get paneer!

010

Jason Radford, PhD @jsradford.bsky.social · 17/08/2026

This past month, Northeastern moved NIO to a new office at the Christian Science Plaza(designed by I.M. Pei no less!). It's one of the architectural jewels at the heart of the city of Boston. It's a great place to come to work every day.

020

Jason Radford, PhD @jsradford.bsky.social · 13/08/2026

If you missed us at #ASA2026 on Tuesday, fear not! You can play around with our AI study designer. It's just our public data documentation put into a notebook so you can plot out potential study designs using NIO data: notebook.google.com/notebook/8bb...

notebook.google.com

220

Jason Radford, PhD @jsradford.bsky.social · 12/08/2026

Also begs the question if we should just pull a wheel of fortune and send all reviewers what an LLM would say. Save us all the time of coming up with the same RSTL&E.

000

Jason Radford, PhD @jsradford.bsky.social · 11/08/2026

Doesn't fix the problem with the APIs but if you're looking for stable access to platform data with known sampling frames, this is why we built the National Internet Observatory

130

Jason Radford, PhD @jsradford.bsky.social · 11/08/2026

Research shows the wealthy are one of the least mobile groups of people www.sup.org/books/sociol....

110

Reposted by Jason Radford, PhD

Deen Freelon @dfreelon.bsky.social · 11/08/2026

When we pulled data using the same sampling criteria and timeframe, we found zero overlap between the official API and the other two methods in most cases. No videos in common. 😶

25513

Jason Radford, PhD @jsradford.bsky.social · 10/08/2026

Come by tomorrow if you're at ASA. 8am workshop on using National Internet Observatory data for your own research: how participants are recruited, what data exists, how to write a project application that clears ethics review, and a look at the secure enclave you'd work in. #ASA2026 #ASA #sociology

042

Jason Radford, PhD @jsradford.bsky.social · 09/08/2026

Sociology will build rocket ships. The Copernican Revolution paved the way for a new notion of the universe; one in which it becomes conceivable that we could leave the Earth. By improving our understanding, sociology is opening the doors to a new kind of society akin to walking on the moon #ASA2026

000

Jason Radford, PhD @jsradford.bsky.social · 09/08/2026

Our science does cumulate. Sometimes I think it is difficult to see that when we read our latest papers or even look at the last 20 years of research. There are still so many open questions and debates. But when we look at what a generation has accomplished, it's clear to see how far we have come.

110

Jason Radford, PhD @jsradford.bsky.social · 09/08/2026

And just as the elites at the top of early renaissance institutions convulsed at the idea the Earth is not the center of the university; so too do our elites resist the notion that male and female are not innate, immutable categories; that the definitions of man and woman are not intrinsic.

100

Jason Radford, PhD @jsradford.bsky.social · 09/08/2026

The revolution in the science of sex and gender from feminism to queer theory to transgender work has slowly built an understanding of the two as social constructs rather than mere biological traits and social essences. With medicine, psychology, and research participants; we've built the knowledge.

100

Jason Radford, PhD @jsradford.bsky.social · 09/08/2026

I see the joint backlash to transgender rights and social science (and the broader anti-DEI push), as part of the Copernican revolution the social sciences have brought to society over the past 75 years. We've long been seen and seen ourselves as less cumulative than the "hard" sciences. But no.

230

Jason Radford, PhD @jsradford.bsky.social · 09/08/2026

If anyone at #ASA2026 wants to talk MCPs for data sharing and data analysis, hit me up. (They're APIs for AIs) We've been working on them for chip50.org and nationalinternetobservatory.org and they're really going to change the way research gets done in the next 5 years.

chip50.org

The Civic Health and Institutions Project | CHIP50

The Civic Health and Institutions Project

000

Jason Radford, PhD @jsradford.bsky.social · 09/08/2026

#asa2026 just's want to make sure you don't get lost.

Map showing how to get from one venue to the other by crossing the street. To be fair, the point of the map seems to be to show entrances and that there's a crosswalk which isn't obvious once you hit the street.

120

Jason Radford, PhD @jsradford.bsky.social · 09/08/2026

Solidarity at the solidarity session! Bodes well for my 8am on Tuesday 😅

010

Jason Radford, PhD @jsradford.bsky.social · 09/08/2026

Plus I got another sticker for my computer as a reminder of what great sociology looks like.

Computer sticker that says "Other countries have social safety nets. The U.S. has women" - Jessica Calarco.

161

Jason Radford, PhD @jsradford.bsky.social · 09/08/2026

I think there's also a lesson here for folks studying the return of nationalism to western politics. With the wholesale expulsion of immigrants, we expect to see women leaving the labor force to do the care work immigrants did as the price of childcare, elder care, and medical care rise.

150