Human activities increasingly take place in online environments, providing novel opportunities for relating individual behaviours to population-level outcomes. In my talk I will consider a simple generative model for the collective behaviour of millions of social networking site users who make choices between different software applications that they can install. The proposed model incorporates two distinct social mechanisms: (1) imitative behaviour reflecting the influence of recent installation activities of other users; (2) rich-get-richer popularity dynamics where users are influenced by the cumulative popularity of each application. Interestingly, although various combinations of the two mechanisms yield long-time behaviour that is consistent with data, the only models that reproduce the observed temporal dynamics are those that strongly emphasize the recent installation activities of other users over their cumulative popularity. More generally this demonstrates that even when using purely observational data, as opposed to experimental research designs, temporal data-driven modelling can in fact effectively distinguish between competing microscopic mechanisms, providing novel insights into collective online behaviour.