Social learners choosing left in discordant and concordant treatments from experiment 1.

Results are from logistic regressions. Independent variables include an index for the 20 learning blocks in each session and the distribution of choices among demonstrators in the fifth trial of the block. This distribution is centered at zero. Thus, if p5 is the proportion of demonstrators choosing left in the fifth trial, we transform this as x5 = p5 − 0.5. Robust standard errors clustered on social learner account for any correlations due to multiple observations per social learner [48, 49], and we show these robust clustered standard errors in parentheses.