Humor me if you will, but I find "N" is the hardest thing to test for**, given how it relates to nonlinear connect the dots function it is. Therefore, I'm willing to bet that about 50% of the people who test "N" actually turn out to be "S". So far my hypothesis has proved true nearly 100% of the times, so I tend to believe it to be true until proven otherwise. :)
** Based, for instance, on this test, where there are may be three questions that test for "N" vs "S":
http://www.humanmetrics.com/cgi-win/jtypes2.asp
Okay, people, the online tests try to come close, but they are not rigorously vetted / tested for test - retest and internal consistency like the official Myers Briggs Forms. It is very difficult hard to write questions that test / retest across a wide population.
If you want to critically discount MBTI, great. We can all discuss that, but realize that anyone can put anything on the internet.
Regarding the article that someone posted, I was fine with the opinion (disagree with its viewpoint though, as 16 personality group types offers a huge amount of variation). But the claim that 50% test / retest changes is misleading....
According to the MBTI Manual , third ed., 1990 Isabel Meyers et al. (I have an older copy), the Test - Retest across 4 weeks is close to 90% and up to 97% for each individual letter (Form M) and n=424. Test - Retest comparing Form G for under and over 9 months, and a population of n=1698 has each letter of retest at 59% (T-F) to 84% (E-I). Form M has higher reliability, but both are much higher than chance. Other controlled tests shows that mood did not significantly affect test-rest.
Over a 50 year time, starting in high school, test respondent (n=39) had 54% remain within one letter of the original. (Form G and restested on Form M). Given that MBTI theory allows for people maturing and the early test forms were not intended for people under the age of 18, this is very good for psychometrics results. (It is a fuzzy soft science, after all).
Organisms are not hard science and test statically like physics. Heck, even my hand measurement will change over 6 months, so why should we expect psychometrics to not have variation? As long as the correlations are strikingly high, across a large sample, then I am good with the statistics that support it.
So, the MBTI Test- retest reliability is quite high, not perfect, but much much higher than chance. T-F is the most likely scale to change. The recent question form "M" is better than the original form "G".
Oh, and college students show much higher test - retest correlations than high school students, or than lower achieving students, and ethnicity did not have a factor. So if you took the official Form G or Form M, and are a college grad, your retest consistency is likely very high.
The reason I like MBTI is that it allows for a lot of complexity across populations, has a huge sample base, and has a deep depth of research (statistical) to support it. All psychometrics at best self selected correlations, not cause and effect, but as a tool for thinking about life and what you want out of it, and why you want / do / say different things than others, it is great!