# Thread: MBTI and Big 5 correspondence

1. Hahaha OK I will make a JavaScript version that does it tmrw

2. As McCrae and Costa long ago noted in that article the OP cited, the MBTI dimensions appear to be tapping into the same real underlying personality dimensions as four of the Big Five.

The best way to "convert" your Big Five type to your MBTI type is to simply assume that, e.g., if you're a Big Five introvert, you're an MBTI introvert. The small correlations with the other Big Five dimensions should be ignored.

The MBTI S/N and J/P dimensions have a mild correlation (probably from test messiness, IMHO), but that doesn't mean you should determine whether you're an MBTI S by way of a formula that takes both your S and J scores into account. And the same is true of the small, messy correlations between each MBTI dimension and the (mostly) non-corresponding Big Five dimensions.

3. Originally Posted by reckful
As McCrae and Costa long ago noted in that article the OP cited, the MBTI dimensions appear to be tapping into the same real underlying personality dimensions as four of the Big Five.

The best way to "convert" your Big Five type to your MBTI type is to simply assume that, e.g., if you're a Big Five introvert, you're an MBTI introvert. The small correlations with the other Big Five dimensions should be ignored.

The MBTI S/N and J/P dimensions have a mild correlation (probably from test messiness, IMHO), but that doesn't mean you should determine whether you're an MBTI S by way of a formula that takes both your S and J scores into account. And the same is true of the small, messy correlations between each MBTI dimension and the (mostly) non-corresponding Big Five dimensions.
It would be helpful if you could directly reply to each point of my post above. I have taken a plethora of advanced statistics / modeling / machine learning courses.

The Big 5 is a high dimensional rotation of the MBTI. The questions were not designed to cleanly capture single MBTI dimensions. The most powerful way to explain the MBTI variance with the Big 5 variance is to use the entire model. The power of those "small" "messy" correlations, summed together, can outweigh the "significant" ones.

4. Originally Posted by mingularity
It would be helpful if you could directly reply to each point of my post above. I have taken a plethora of advanced statistics / modeling / machine learning courses.

The Big 5 is a high dimensional rotation of the MBTI. The questions were not designed to cleanly capture single MBTI dimensions. The most powerful way to explain the MBTI variance with the Big 5 variance is to use the entire model. The power of those "small" "messy" correlations, summed together, can outweigh the "significant" ones.
As just one example: Let's say a man tests Extraverted and Agreeable on the Big Five test and you want to determine his MBTI F preference by "converting" his Big Five type. You've taken a .22 "correlation" statistic (not exactly the right term, right?) between MBTI F and Big Five E and you're saying it justifies using the man's Big Five E score to give a positive boost to the man's Big Five Agreeable score — as compared to the result you'd get if you just straightforwardly used his Big Five Agreeable score to determine his MBTI F score.

Buuuut... by your reasoning, shouldn't the same .22 positive boost also apply to a conversion in the other direction — i.e., converting a man's MBTI F score to his Big Five Agreeable score? If so, isn't it obvious there's something wrong with using the correlation that way? If you give him a positive boost converting from Big Five Agreeable to MBTI F, and then take the resulting MBTI F score and convert it back to Big Five Agreeable using the same method, it won't get you back where you started, right? The more conversions you do back and forth, the higher and higher the man's Big Five Agreeable and MBTI F scores will go.

Also: Where do those correlational figures you're using come from? Are they from a study that used that 44-item Big Five Inventory you're using in this thread? When a subject takes two different Big Five tests, the scores don't typically match up to a degree that would justify using correlation statistics from one test and applying them to another — even if the way you were applying the correlations made sense in theory.

Finally: Can you point me to any respectable online source that supports your method of "converting" from one personality typology to another?

5. Originally Posted by mingularity
^ Significance levels are arbitrary. The correlations still carry information. The key question is about power, not significance. Keeping them in increases power because it keeps all the information around, which means we are using all the data from all the subjects.

The Big 5 is a high dimensional rotation of the MBTI - all variables do in fact correlate significantly. The question is how big of a sample would you need to achieve standard significance levels.
Dude, you missed the point of my post. :/ I wasn't arguing that Big 5 factors don't correlate with MBTI's factors. I was saying that if you're using factors like "extraversion" from the Big 5 to calculate S or N, your algorithm is inherently invalid. But it's okay, carry on.

6. Originally Posted by reckful
[...]
Since you didn't reply to my points I'm going to restate one of them with respect to your last post.

The Big 5 questions were not designed to explain the MBTI variance. Language is high dimensional, thus, the Big 5 is a high dimensional rotation of the MBTI, and vice versa. The Big 5 is not the MBTI. It doesn't matter if the trait mappings don't appeal to your intuition. That's just how the model works!

The model I am using is called a linear combination aka regression model. I am implementing the javascript version right now, which should be done soon, after which I will do more work on the statistics (probably intermittently throughout the week).

7. Originally Posted by Honor
Dude, you missed the point of my post. :/ I wasn't arguing that Big 5 factors don't correlate with MBTI's factors. I was saying that if you're using factors like "extraversion" from the Big 5 to calculate S or N, your algorithm is inherently invalid. But it's okay, carry on.
See my post above. It is not inherently invalid. Each Big 5 trait explains some amount of variance of each MBTI trait. With an infinite number of subjects, we would know exactly how much variance. We don't have that many, however, so we have to work with what we've got. Ignoring information because it's not "significant" is tantamount to throwing information away. We want to keep as much information around as we can because it increases our statistical power.

Originally Posted by Honor
I was saying that if you're using factors like "extraversion" from the Big 5 to calculate S or N, your algorithm is inherently invalid.
It doesn't actually work out that way. The MBTI throws out questions from the E/I scale that explain variance in the S/N scale. However, the Big 5 was not designed that way. Hence, an extraversion question from the Big 5 scale almost certainly does explain variance in the MBTI S/N scale. Also, the MBTI scales are in fact correlated themselves. If we had an infinite number of subjects, it would be easier to see. Actually, we can probably look up how correlated they are, and that would be interesting to know.

8. Originally Posted by mingularity
Since you didn't reply to my points I'm going to restate one of them with respect to your last post.

The Big 5 questions were not designed to explain the MBTI variance. Language is high dimensional, thus, the Big 5 is a high dimensional rotation of the MBTI, and vice versa. The Big 5 is not the MBTI. It doesn't matter if the trait mappings don't appeal to your intuition. That's just how the model works!

The model I am using is called a linear combination aka regression model. I am implementing the javascript version right now, which should be done soon, after which I will do more work on the statistics (probably intermittently throughout the week).
I'm not the one who's failing to reply to relevant points, amigo.

Since you aren't willing to say where your correlation figures come from (and whether they even relate to the test you're using here), and you aren't willing to explain why your technique wouldn't goofily result in the same incremental boost to the type preferences (rather than the opposite boost) when you're going in the opposite direction, and you aren't willing to cite any source for the applicability of the conversion technique you're using to dimension scores from personality typologies — and no, those Wikipedia math links don't qualify — I suspect it's time for me and the other readers of this thread to conclude that the emperor has no clothes.

9. Originally Posted by reckful
Since you aren't willing to say where your correlation figures come from
Originally Posted by mingularity
• McCrae, R. & Costa, P. (1989). Reinterpreting the Myers-Briggs Type Indicator From the Perspective of the Five-Factor Model of Personality. Journal of Personality.
Originally Posted by reckful
(and whether they even relate to the test you're using here)
The test I'm using here was designed to explain as much of the variance in the original Big 5 as possible, given the number of questions. It's also free for non-commercial use.

Originally Posted by reckful
and you aren't willing to explain [...]
Originally Posted by mingularity
Figuring out the best way to do the conversion is a work in progress.
Why don't you lend a hand and try to make it actually work? I know an entire legion of experts in statistics (read: my friends) and already had a brainstorming session with a couple of them last night. We agreed this should work in principle, and that it's a fun project. Details are certainly important, but trying to go from details to "you're wasting your time" isn't really helping me out much.

Originally Posted by reckful
the emperor has no clothes.
Are you picturing me naked?!

10. I haven't delved into the specific stats or sources, but I have bothered to use the equations.

Scores of:
N 2
E 70
O 95
A 93
C 92

from another test (please don't kill me)

Yield:
EI 15.49 I
SN 23.38 N
TF 24.66 F
JP -29.28 J

Works well enough in this case. edit: I'd be interested to hear more about turning correlations into coefficients, how (and if) statistical significance comes into play, and so on.
Originally Posted by Honor
Dude, you missed the point of my post. :/ I wasn't arguing that Big 5 factors don't correlate with MBTI's factors. I was saying that if you're using factors like "extraversion" from the Big 5 to calculate S or N, your algorithm is inherently invalid. But it's okay, carry on.
It's not a flaw in the algorithm; it's all up in the stats.

Suppose we have some typology systems--perhaps Grant's Temperament System (GTS) and a Tinkerbell Classification (TC). GTS has the the factors Extraversion, Friendship, and Orderliness. TC has Extraversion, Emotion, Caffeine Intake, and Chicken Consumption.

Suppose also that we've had a bunch of people who've taken both the GTS and TC.

Data could show that those who have scored as Extraverts on the GTS have also tended to score as Extraverts on the TC scale. There's no stopping the data from showing also that Extraverts on the GTS (perhaps slightly) tend to score high on the TC's "Caffeine Intake" factor.

(Hell, the data could also show that Extraverts on the GTS tend to score as hardcore "Chicken Consumers" on the GTS as well. This would mean that the GTS has factors that aren't quite independent from one another, which is not a good thing.)

#### Posting Permissions

• You may not post new threads
• You may not post replies
• You may not post attachments
• You may not edit your posts
Single Sign On provided by vBSSO