A growing body of empirical work shows that social recognition of individuals' behavior can meaningfully influence individuals’ choices. This paper studies whether social recognition is a socially efficient lever for influencing individuals’ choices. Because social recognition generates utility from esteem to some but disutility from shame to others, it can be either positive-sum, zero-sum, or negative-sum. This depends on whether the social recognition utility function is convex, linear, or concave, respectively. We develop a new revealed preferences methodology to investigate this question, which we deploy in a field experiment on promoting attendance to the YMCA of the Triangle Area. We find that social recognition increases YMCA attendance by 17-23% over a one-month period in our experiment, and our estimated structural models predict that it would increase attendance by 19-23% if it were applied to the whole YMCA of the Triangle Area population. However, we find that the social recognition utility function is significantly concave and thus generates deadweight loss. If our social recognition intervention were applied to the whole YMCA of the Triangle Area population, we estimate that it would generate deadweight loss of $1.23-$2.15 per dollar of behaviorally-equivalent financial incentives.