We consider how to credibly and reliably assess the opinions of individuals using their social media posts. To this end, this paper makes three contributions. First, we assemble a workflow and approach to applying modern natural language processing (NLP) methods to multi-target user stance detection in the wild. Second, we establish why the multi-target modeling of user stance is qualitatively more complicated than uni-target user-stance detection. Finally, we validate our method by showing how multi-dimensional measurement of user opinions not only reproduces known opinion polling results, but also enables the study of opinion dynamics at high levels of temporal and semantic resolution.