Read on Twitter
1/ Our new paper is out today in @NatureComms! “Experimental Evidence of Scale-induced Category Convergence across Populations.” In a time of so much political division, I hope this study can help remind us of the fundamental ways we are all connected. https://www.nature.com/articles/s41467-020-20037-y?utm_source=feedburner&utm_medium=feed&utm_campaign=Feed%3A+ncomms%2Frss%2Fcurrent+%28Nature+Communications+-+current%29
2/ We find that cultural processes of communication can lead separate large populations to independently arrive at similar category systems for new phenomena; however, in small populations, communication more often leads to divergent category systems.
3/ Our study designed a novel experimental platform called the “Grouping Game” which allowed us to observe how social groups collaboratively formed category systems in real time for a novel continuum of ambiguous images.
4/ To date, the emergence of similar categories across societies has been explained in psych/anthro in terms of “what people bring with them wherever they go – i.e. psychobiology”; but people also bring communication networks, and these can also induce cultural similarities
5/ This can be explained by a simple mathematical model. Ppl varied widely in their categories, yet a few labels were more likely to appear. I extend the birthday paradox to show that popular categories are more likely to reach critical mass and spread in larger populations.
6/ This is also reproduced by a naive agent-based model; across methods, we find that large populations amplify weak biases in categorization; this is surprisingly consistent with the Fisher-Wright model of drift, as a nice link between bio and cultural evolution
7/ In small groups, cognitive biases were not strong enough to drive outcomes; small groups regularly converged on idiosyncratic categories, even when the more popular categories were attempted in these small groups!
8/ This suggests that category formation is not driven solely by cognitive properties of the labels themselves (e.g., their 'natural' descriptive fit), but also by their frequency in the population and the likelihood that they reach critical mass and spread.
9/ To test this, we ran additional trials where a minority of confederates worked to spread rare categories in large populations; across the continuum, subjects adopted rare labels as a result of concerted social influence.
10/ Adopting these labels during the experiment altered how subjects privately described the images on their own in a post-test survey; this suggests that the confederates may have reframed their categorizations (and potentially their perceptions).
11/ I am now developing a method of crowdsourcing that harnesses communication networks to reduce bias and increase replicability in the categories used by human coders, eg in identifying what counts as hate speech in content moderation over social media. https://www.asc.upenn.edu/news-events/news/centola-and-guilbeault-receive-facebook-grant
12/ This study is the first step of many (I hope!) in my goal to empirically illuminate the fundamental relationship between the structure of society and the structure of the category systems that it produces, and which produce it in turn
13/ To any students thinking of pursuing a PhD in computational sociology (w. interest in categorization), or to anyone interested in hearing more about this research, please reach out over twitter or at [email protected]. Thank you all who read through!
