Polling and statistical generalisations hide a lot of detail - they can generally identify the what, but whether they get the why is much more hit and miss.

Most obviously without anecdotal reports on what people believe, pollsters would not know what questions to ask.