In contrast to classical social norm experiments, we conduct experiments that semicontinuously randomise the share of individuals who are taking a particular action in a given environment. Using our experimental results, we are able to estimate the distributions of individual tipping points across our settings. We find that tipping points are very heterogenous, and that a substantial share choose to do the action (or not) regardless of what others are doing. We also show that, once embedded in dynamic models, our estimates predict that individuals will end up doing very different things despite engaging in copying-like behaviour.