It sounds more like different data set that's gathered by observation. If you can keep the observer's bias to a minimum this method may tell you more about how your feature is received and used. It is also the most time consuming and costly approach and doesn't really scale.