### Normalized Test Results

Posted:

**Sat Apr 03, 2010 4:18 am**I have temporarily included--within the sociotest data on the user's sociotype page--the normalized results. Normalization in this context has a specific meaning. Start with the assumption that the probability distribution for the 16 sociotypes is equal in any randomized population sample (e.g., there are as many IEIs as LIEs, etc.). Normalizing the test data effectuates this assumption by removing any bias in the questions. As long as each question is minimally probative for the trait it is trying to test for (and this is an important assumption), then even if the question's wording is very biased towards one trait over the other, the normalization should nevertheless correct for this bias.

'Put in more simple terms, if 16 people take the test, regardless of the answers they provide (assuming though that there is at least minimal variation), normalizing the data would result in each person being a unique type from every other person. If 128 people take the test, there will be exactly 8 persons of each sociotype. I actually haven't checked to see if this is what is actually occurring, but this is what should be happening.

Normalizing the data becomes the optimal solution when (1) the sample size moves towards infinity and (2) the sample size becomes a more accurate representation of the general population (i.e., a randomized unbiased selection process). Unfortunately, neither of these conditions is present for my test; thus the normalized results will often be very innacurate (e.g., 10 LIIs could take the test yet only one of them may get a result of LII). Thus i don't put much weight on the normalized results and they are not used in the sociotype algorithm.

However I do find them helpful for certain aspects (the j/p dichotomy in particular), and more misleading in others (mistyping intuitives as sensors). The more biased the sample of test takers in a certain trait, the more inaccurate it will be.

If you have taken the sociotype test, you can see your normalized results by going here and clicking on "sociotype profile"; or if you haven't taken the test yet, go here to take it. If nothing else, think of your normalized result this way: you fit that sociotype better than at least 93% of the other users on the site, even if you aren't actually that type.

'Put in more simple terms, if 16 people take the test, regardless of the answers they provide (assuming though that there is at least minimal variation), normalizing the data would result in each person being a unique type from every other person. If 128 people take the test, there will be exactly 8 persons of each sociotype. I actually haven't checked to see if this is what is actually occurring, but this is what should be happening.

Normalizing the data becomes the optimal solution when (1) the sample size moves towards infinity and (2) the sample size becomes a more accurate representation of the general population (i.e., a randomized unbiased selection process). Unfortunately, neither of these conditions is present for my test; thus the normalized results will often be very innacurate (e.g., 10 LIIs could take the test yet only one of them may get a result of LII). Thus i don't put much weight on the normalized results and they are not used in the sociotype algorithm.

However I do find them helpful for certain aspects (the j/p dichotomy in particular), and more misleading in others (mistyping intuitives as sensors). The more biased the sample of test takers in a certain trait, the more inaccurate it will be.

If you have taken the sociotype test, you can see your normalized results by going here and clicking on "sociotype profile"; or if you haven't taken the test yet, go here to take it. If nothing else, think of your normalized result this way: you fit that sociotype better than at least 93% of the other users on the site, even if you aren't actually that type.