I agree! People are very inconsistent. They change their beliefs, lie to people, are unsure about things, etc. A lot of sociologists would argue that this is a good reason to combine this kind of quantitative data with in-depth ethnographic observation of real behavior (e.g. https://shamuskhan.files.wordpress.com/2013/11/talk-is-cheap...). When we want to see how someone's opinion actually plays out in a real situation, that's usually our best tool.

Also, a lot of survey data captures demographic characteristics, not just individual beliefs-- think censuses. Surveys are generally considered a robust instrument for measuring things (like income, occupation, gender) that are relatively fixed over time.