How does social influence shape collective outcomes? When does it lead to lock-in on inferior options?
In our 🚨 new preprint 📝 osf.io/preprints/so... we make three contributions
w/ @alexgelas.bsky.social Alex Jochim @leostnbrk.bsky.social Peter Steiglechner & @pantelispa.bsky.social
🧵1/4
7/ Our results provide a solution to a recent puzzle about whether and when lock-in on an inferior option is possible and connect a behavioral phenomenon to macro-level popularity dynamics. They also highlight the importance of small winning margins.
6/ This theoretical prediction is corroborated in a reanalysis of data from multiple-world experiments in three domains (political preferences, matters of fact, matters of taste). In the picture, when the MME exceeds quality difference (d), lock-in is observed (blue), but rarely otherwise (red).
5/ Importantly, lock-in becomes possible when the size of the marginal majority effect exceeds the quality difference of the two alternatives (i.e., difference in choice probability without social influence).
4/ Main finding: Small majorities matter!
People are much more likely to choose A if it is even slightly more popular than B.
We call this the *marginal majority effect*
3/ We develop a mathematical theory that describes popularity dynamics between two competing alternatives when people's choices are influenced by the previous choices of others.
We then validate it using data from recent behavioral experiments.
2/ We all know social influence matters.
People are more likely to choose something if others chose it first: music, platforms, political opinions, etc.
But here’s the puzzle:
Sometimes bad options can lock in an early popularity advantage, while in other cases better options take over.
Why?
1/ Why do inferior but popular things remain popular?
Excited to share our new paper with @pantelispa.bsky.social, @glemens.bsky.social & @arnoutvanderijt.bsky.social
"The marginal majority effect: When social influence produces lock-in"
www.science.org/doi/epdf/10....
Four travel grants available for PhD students and postdocs to present their posters at the 13th Toulouse Economics and Biology Workshop! Apply by Feb 27th.
www.iast.fr/conferences/...
I am hiring a postdoc for a DFF-funded project on social influence, and the decision processes that fuel rich-get-richer dynamics in the online/offline world. The position is for up to a year, competitive Danish salary, remote work possible. Interested or know somebody? DM me or share!
It is the time of the year when I tell you about my favorite post-doc ever 👇 Unless you are allergic to the French, this one sits up there with the Nuffield postdoc (life style, productivity, interdisciplinary stimulation). Share widely! Apply! #poliscky
www.iast.fr/research-fel...
There are two possible explanations: 1. distribution for some traits is highly non-gaussian, 2. being average in a trait means less likely to be average in another.
No. 2 would be more interesting. To exclude 1, I would look at the percentage of people within +/-0.25sd for each trait (must be ~20%)
Go home evolution, you're drunk!
📣 We are excited (or should we say disgusted?) to announce the conference 'Disgust across borders' at IAST! @iast.fr
🗓️ Join us on Dec 4 & 5 for two days full of disgust research across species and disciplines.
Registration and abstract submission are open (until Aug 31): forms.gle/QBmkUB3aLDjY...
As some of you guessed, the two models are statistically indistinguishable. The answer is Model 1, but there was no way to tell from the data. I was very surprised when I first saw this. Here we show that this confounding of heterogeneity and reinforcement is general: www.pnas.org/doi/10.1073/...
Norrköping campus at sunset
We are hiring postdocs in Computational Social Science
📍SweCSS, Norrköping, Sweden
⏰Deadline June 3
🔗https://liu.se/en/work-at-liu/vacancies/26854
Please apply // help us spread the word
My point is that the only way time enters the picture in Model 2 is through the error terms, which are i.i.d, so this pattern shouldn’t be surprising.
(Disclaimer: I know the answer to the original question.)
Interesting. Wouldn't you expect model 2 to (also?) have this property?
Did Model 1 or Model 2 generate these sequences? Why?
Model 1: Pólya urn, starting with one black and one red ball. 1 (0) stands for black (red).
Model 2: Empty probit model, random intercepts with mean 0, ICC = 1/2.
All 100,000 sequences here: tinyurl.com/35pkyfxa
📢 We're now on Bluesky! Follow us for insights on: 🧠 Interdisciplinary research 🎤 Talks, workshops & conferences 🎧 Podcasts 🧐 Science outreach #Science #Research #Interdisciplinary #IAST #Toulouse #AcademicLife
📢 We're now on Bluesky!
Follow us for insights on:
🧠 Interdisciplinary research
🎤 Talks, workshops & conferences
🎧 Podcasts
🧐 Science outreach
#Science #Research #Interdisciplinary #IAST #Toulouse #AcademicLife
Science of Science in Copenhagen, June 2025 #scisci
