A single-choice relatively-few-words question can't be expected to indicate one's MBTI accurately, because putting the question to a group of known-for-sure ABCD types would throw up a distribution of responses and one is doing a 'reverse lookup' of that distribution in posing the question as a means of indicating an individual's Type. It would though be interesting to know for which Type(s) the single question had the highest accuracy.
Improving prediction-strength necessarily requires more input ie (i) more questions or (ii) having the respondent rank the answers to the one question so that negative weights can be attached to the Types associated with disliked choices or (iii) more words in the single multiple-choice question - this gravitates towards self-choosing between full descriptions of the 16 Types or (iv) a combination eg self-choosing from potted Type descriptions would normally throw a good half into the bin pretty well immediately and a good half of the remainder after a little thinking [or a little feeling
], leaving typically three choices from which the final choice is made by administering the appropriate customised test from a library of 1820 four-possibility tests, 560 three-possibility tests and 120 two-possibilty tests.
Improving prediction-strength necessarily requires more input ie (i) more questions or (ii) having the respondent rank the answers to the one question so that negative weights can be attached to the Types associated with disliked choices or (iii) more words in the single multiple-choice question - this gravitates towards self-choosing between full descriptions of the 16 Types or (iv) a combination eg self-choosing from potted Type descriptions would normally throw a good half into the bin pretty well immediately and a good half of the remainder after a little thinking [or a little feeling