Regarding predictions in sociology (a point made in the answer given here by Olga Zeveleva), I would say that polls actually work pretty efficiently in many countries in a context of elections predictions from 1936 onwards, when Gallup introduced random sampling and predicted Roosevelt's victory. To my knowledge, forecasts in the USA failed in a couple of cases: in 1948 when Truman's victory came unexpectedly, and in the case of Trump (though it is difficult to predict a small percentage difference in an indirect election system, the reasons for the unexpected elections here are indeed multifaceted and complex). Also, the impact of social desirability (giving the answers the respondent thinks the interviewer will like to hear) differs across cultures.
As for qualitative sociology, it is not always about predictions, but also about associations of different phenomena, classifications of certain groups of people. For example, those who feel incapable of impacting the political situation in a country are more likely also to have negative attitudes towards migrants or other out-groups. In stating that, I am not trying to determine what comes first (I may even suppose that these things come simultaneously).