Surveys can sometimes be somewhat informative if done correctly. I once ran a morality survey (on TolunaQuick) where I quantified categories of morality across socioeconomic statuses using a three-tier questionnaire. The approach was much like that employed by common personality tests, where each measurement is made from multiple questions essentially targeting the same topic. By having related questions on different tiers (levels of magnitude or intensity), the subjective information was somewhat quantifiable across individuals. Like with many systems, it would be recommended to begin with a simple scenario (one type of event, one variable being measured) and to build from there.
Without having a method of quantification, the formulae for human interactions would remain as nothing more than abstract hypotheses. Hence, quantification is necessary for scientific analysis.
On a related note, I believe that one of the most detrimental flaws in everyday human communication is that magnitudes are expressed using highly subjective, discrete categories. For example, someone might ask, "How much fun did you have?". The answer would generally be limited to a scale having just several steps such as "a little bit", "a fair amount", "quite a bit", or "a lot". It is unfortunate that humans are unable to give a more fluid answer that is somehow standardised, such as by saying "3.5 F worth of fun". Another deplorable example is when journalists take scientific findings and say something like "X was found to be significantly more likely to give you cancer than Y" while not giving any details as to the absolute or relative quantity of difference. For all we know, X increases the risk of cancer by 0.01% compared with Y; and the general public seems ignorant of the uselessness of the statistical term "significant" when not accompanied by the numbers whence it sprang. Common tongue is inherently deceptive, and many people have a bad habit of taking advantage of this ambiguity for propagandist purposes.
How do you overcome the issue where, say, if we're asking our survey participants, "How often do you feel insulted by a family member in a typical day?", participants who experience frequent emotional abuse from their family may not even realise what they're experiencing are insults? "It's not an insult because it's true" kind of thing.
It seems that the nervous system pretty reliably adjusts its sensitivities to the recent norms, so maybe the answer to this predicament involves not only (a) asking participants such questions, but also (b) establishing for each person the current systemic bias. Similar to pain-threshold testing, perhaps the person could be exposed to social scenarios, such as in videos, while being monitored with an EEG or related device. The pre-screening done on the experiment participants in the 2001 film "The Experiment" could be seen as similar since it tests emotional reactions. Even persons from different worldly cultures and religions would presumably have different biases, so finding this bias for each individual would probably be very important if accurate analysis were sought. To give an analogy, asking people questions on feelings is like asking a multi-meter for a reading. The problem is that each person has a large set of calibration knobs, so the real-world measurements can not be known without first checking the knobs.