←
Will Schulz176 posts
Will Schulz avatar

Will Schulz

@small_schulz

Postdoc studying algorithms and online speech @CSMAP_NYU

◷ Joined June 2012▧ 19 Photos and videos
Jay Van Bavel, PhD avatar
⟳ Will Schulz retweeted
Jay Van Bavel, PhD@jayvanbavel·Aug 24

Social media discourse is driven by false #polarization Analyzing 1.7 million social media posts, a series of controlled experiments, and a large global survey covering 40 countries (total N = 49,222), we find that (1) strong ideologues post their opinions significantly more often than moderate individuals (2) extreme attitudes are posted more often than moderate attitudes. (3) This general pattern replicates robustly across 40 countries (4) the opinions that users post are more hostile and toxic than the beliefs they keep to themselves. Across studies we find extreme opinions are overrepresented on social media by at least 6% and up to 22% compared to their true prevalence in our samples. osf.io/preprints/psyarxiv/7z… This project was led by @small_schulz and @ClaireRobertson and includes @H_Sjastad @AshuAshok @LisaFOswaldo @junghwanyang @simonsaysnothin and @andyguess

Post attachment
Will Schulz avatar
Will Schulz@small_schulz·Aug 24

Check out the preprint here! osf.io/preprints/psyarxiv/7z… This was a massive collaborative effort, so a huge thank you to our WONDERFUL coauthors, @H_Sjastad (who spearheaded global data collection), @AshuAshok, @LisaFOswaldo, @junghwanyang, @jayvanbavel, and @andyguess!!

Will Schulz avatar
Will Schulz@small_schulz·Aug 24

Therefore, much of the intense polarization we observe online today may not be caused directly by platform algorithms pushing people apart, but rather by the selective posting behavior of extreme voices over-representing their most extreme views.

Will Schulz avatar
Will Schulz@small_schulz·Aug 24

And, troublingly, the opinions users choose to post online are measurably more hostile and off-putting than the beliefs they keep to themselves.

Will Schulz avatar
Will Schulz@small_schulz·Aug 24

This general pattern replicated robustly across 40 out of 40 countries examined, spanning diverse political cultures, and held true across diverse political issues regardless of general political interest or left-right ideology.

Will Schulz avatar
Will Schulz@small_schulz·Aug 24

Using mixed methods (including sequential behavioral experiments, analysis of over 1.7 million social media posts, and a global survey of nearly 48,000 people across 40 countries) we quantify the true scope of false polarization online.

Will Schulz avatar
Will Schulz@small_schulz·Aug 24

Is social media truly making us more politically polarized, or does it simply over-represent our most extreme opinions? In our new preprint, Claire Robertson and I present *punishing evidence* that social media systematically overrepresents users’ most extreme political opinions.

Post attachment
Danny R. Daneri avatar
⟳ Will Schulz retweeted
Danny R. Daneri@DannyDaneri·9 Apr 2025

My article, Labor vs. Big Business, is out! it's about efforts of companies like Uber to reform gig worker laws and voter cue-taking from unions and business. TLDR; progressive voters trust union cues to substitute for policy information link.springer.com/article/10…

Post attachment
Political Analysis avatar
⟳ Will Schulz retweeted
Political Analysis@polanalysis·3 Feb 2025

Currently in FirstView: In “What Would You Say? Estimating Causal Effects of Social Context on Political Expression,” @small_schulz develops a method for estimating social influence over political expression. He uses an experiment to estimate respondents’ lexical ideology.

Post attachment
Will Schulz avatar
Will Schulz@small_schulz·27 Dec 2024

6/ reach out if you’d like to try using the WWYS method yourself! I can set you up with the survey instrument and model code. I’m also looking for collaborators to apply in different languages/country settings!

Will Schulz avatar
Will Schulz@small_schulz·27 Dec 2024

5/ and I also derive causal estimates of self-censorship by analyzing treatment effects on a second “outspokenness” dimension generated by my Wordsticks model (details in paper)

Will Schulz avatar
Will Schulz@small_schulz·27 Dec 2024

4/ by manipulating the social context in the WWYS question in a between-subjects experiment, I derive causal estimates of ideological “code-switching” caused by liberal and conservative interlocutors (using close-friend speech as a reference/control)

Post attachment
Will Schulz avatar
Will Schulz@small_schulz·27 Dec 2024

3/ I scale the data using a model similar to Wordfish (but since it’s for ordinal data, I call my model WordSTICKS): the ideological content of each phrase is defined by the latent ideal points of the people who use/avoid the phrase

Post attachment
Will Schulz avatar
Will Schulz@small_schulz·27 Dec 2024

2/ I collect data using the “What Would You Say?” (WWYS) question, which asks respondents whether they would use various politically-charged words and phrases in a given social context (like a conversation with a close friend)

Post attachment