Sean Taylor

Who they are

Taylor holds a PhD in Information Systems from New York University and a BS in Economics from the University of Pennsylvania. His profile at Stanford Data Science lists him as a data scientist at OpenAI focused on model evaluation, and, earlier, as the founder of Motif Analytics, a company that built data visualisation and analysis software, a research scientist at Lyft, and a member of Facebook's Core Data Science team.

Aral, the paper's senior author, was his doctoral advisor, a relationship Taylor states himself in his own account of the study. The experimenters did not simulate a news site. They worked with a live one, whose administrators applied the treatments.

Key papers

Muchnik, Aral and Taylor (2013), Social Influence Bias: A Randomized Experiment, Science 341(6146), 647-651, doi:10.1126/science.1240466.* On a social news aggregator, the first comment on each of more than 100,000 posts was randomly assigned to receive an up-vote, a down-vote or no intervention. A fabricated up-vote raised the likelihood of a later positive rating by 32 per cent and final ratings by 25 per cent on average. A fabricated down-vote was corrected away by later readers.

In a post summarising the paper, Taylor wrote that the result is a causal effect of prior ratings on rating behaviour, which creates path dependence, so a high or low rating does not by itself show quality. He also wrote that the up-treatment and down-treatment effects roughly cancel in aggregate long-run ratings, and that the treatments appeared to change opinions and not only who voted, which he called at least one significant component of the effects.

Key ideas

The design assigned the first vote on each comment by chance, so that the difference between conditions is the effect of the vote itself.

The paper reports an asymmetry: approval compounded and disapproval did not, because readers pushed back on an unearned down-vote and let an unearned up-vote stand.

---