The problem is the distribution along the dichotomies. It's gaussian. Most of the people are borderline. Take eg. E-I. Most people are in the middle. Some people are "rather introvert" or "rather extravert"; only a few are "extremely introvert" or "extremely extravert".
Basically, if you have more than one polarizing question, the same person can answer an "introvert" answer to one question and an "extravert" answer to the other; putting the person in the middle again.
If you have only one, you will be inaccurate. You can't call me 100% extravert because I like to organize get-togethers. You can't call me 100% introvert because I like reading.
I know all this. This adds naturally to the difficulty of making any sorting mechanism. The "middle of the road" syndrome is something ANY instrument that aims to sort will have. This is an artifact of averaging and the tendency for people to pick the middle option.
Actually, your questions aren't polarizing, as they allow for "grey" answers. "Do you hate yourself more often than not" with only "yes" and "no" as answers, is polarizing.
Allowing "grey" answers was kind of the point. If you force people to chose between two options, it is the answer choice that polarize, not the question itself.
I am attempting to use strong language so that the question does polarize even if there is a continuum of answers. In other words, the distribution should have two high points in it, or the distribution should be skewed heavily to one side.
The majority of questions, there are also many reasons WHY someone might answer yes or no. It seems like a sorter device would have to look at multiple answers to triangulate intent/motivation, rather than just basing the sort on the individual answers to each.
I don't was yes or no answers because that
forces the polarity. I want the polarity on the questions to be real.
Yes, creating a sorter is complicated. If you just do a simple addition, you will have the central limit theorem pushing the distribution back to being Gaussian.
You may need to get a feel for how often the person picks middle of the road answers to weight a polarized response more for that person.
Also, I am actually thinking about making a sorting mechanism using some sort of learning algorithm to pick out questions that sort well and are correlated with each other...the number of dimensions will depend on how the polarizing questions correlate.
For example:
You can just put (for instance): "I hate day dreaming"
It sounds like you're fishing for S/N here, but on its own this question does not determine that.
An INTP daydreams of new worlds and theories.
An ENFJ daydreams about community development.
An ISFJ daydreams about getting married and her wedding.
An ESTP daydreams of all the cool things he will be doing after school today or the sale he will make.
If you ask each one of them if they hate day dreaming, you have no idea whether you'll get a yes or no, it all depends on how they read that question.
Yes. But in strong language, the person who actually answers that they hate day dreaming in a high percentage would be different from the norm, I think.