General surveys are self selective. Not everyone in a group will take a survey. There is too much bias and therefore are not really scientific. How can one control the bias's in results in these types of surveys? Drop out subjects are not accounted for.
But, now, with the advent of the internet, millions of users can be typed by personality and used for real observational and experimental psychology...
Cambridge was psychologically profiling users from interactions. Basically, a 'big five' personality test was used.
openness - Do you like new experiences?
conscientiousness - Are you a perfectionist?
extroversion - Do you enjoy parties?
agreeableness - How much empathy can you feel?
neuroticism - Do you get upset easily?
Scientists say the algorithm used gets better and better at predicting your behavior the more you interact with it. Supposedly even better than person to person. Not just Facebook likes and such, but all your posts, shares, photos, messages, emails, etc.
Cambridge was hired by Trump to gather data on over 85 million people. So, depending on your personality traits certain types of ads would've been sent your way to induce your thinking mode.
BTW, Trump isn't the only one to use these services...