This discussion hinges on the “level of openness” being a big deal and a common factor explanatory of many different things. I think the first step should be making a case for this concept being useful, rather than a salient but fuzzy label that one likes to attribute and then rationalize as having explanatory power.
You want me to explain, cite, and defend the entire Big Five model with its factorization and the predictive powers thereof?
I’m sorry, I’m not going to do that. The goal of this article was to discuss some research leading to what I considered interesting speculation and questions. I linked to Wikipedia for the massive background such a veiled question wants (or you could read the rest of the book these short extracts were from, whose fulltext I also linked for readers such as you).
You want me to explain, cite, and defend the entire Big Five model with its factorization and the predictive powers thereof?
Vladimir_Nesov does not, but I do. More specifically, I’d be interested in seeing a defence of this or any other model, such as Myers-Briggs or what have you. To me, all these models sound suspiciously similar to each other as well as to horoscopes, but I could be wrong.
Don’t get me wrong, I’m not asking you to do so OMG RIGHT NOW or anything, but I would be interested in reading a post on the topic.
I think you are laboring under a slight misapprehension about personality research. Myers-Briggs isn’t solid science. The eneagram isn’t solid science. Astrological personality models aren’t solid science. I think you have correctly noticed that “psychological traits” are a ripe area for epistemically unsound belief systems that appear to bear on something people hold near and dear (ie understanding other people) so you’re justifiably suspicious of a mention of personality, which is laudable.
But you’re asking for a defense of “all that crazy stuff”, and a good defense of “all that crazy stuff” can’t honestly be provided, because most of it really is bunk, or at least it has so much bunk mixed in that its only good for psychoceramic data or maybe to pan for gold that might be hiding in the crazy. The big five personality model is an attempt to do actual science in the same space in order to produce reasonably valid and reliable dimensions of human “personality” variation. The point of the big five is that there is solid research and a deep literature and so on, in contrast to all the crackpot stuff.
If someone uses the big five and you’re suspicious and ask for a defense of personality systems in general, that’s like someone using geometry and you being suspicious because you’re only aware of a lot of crackpots who keep trying to square the circle and so you ask them to defend the squaring the circle stuff, (which was proved to be impossible in 1882) before you’ll accept analysis of evidence that legitimately makes use of a “suspiciously geometric” concept like the triangle inequality.
Unfortunately, defending established science quickly is hard because the content of science generally involves real inferential distances. If you want to start reading in this area, two useful keywords are Psychometrics and Trait theory.
In practice, “Openness to new experience” is the weakest part of the big five personality model. It can be measured reliably and predicts various things you’d expect it to predict and relatively naturally falls out when you settle for using 5 dimensions rather than 3 dimensions or 18 dimensions. However, when researchers tried the same thing on other cultures to see if this was a human universal, it turned out that the other four (Extraversion, Conscientiousness, Agreeableness, and Neuroticism) were relatively universal but in some cultures (if I remember correctly it was things like agrarian peasant societies?) basically everyone is pretty low in Openness relative to measurement norms derived from Japan or the US or whatever.
I think you are laboring under a slight misapprehension about personality research. Myers-Briggs isn’t solid science. The eneagram isn’t solid science.
Your understanding is consistent with mine. Myers-Briggs is really frustrating, because some of its ideas are anecdotally compelling (Introversion vs. Extraversion, Thinking vs. Feeling), while others are esoteric (Judging vs. Perceiving and Sensing vs. Intuition). At least on the types, INTP probably refers to a real phenotype (which is common on LW), but I don’t know if any of the other type combinations are real.
Interestingly, the MBTI seems to almost reduce down to the Big Five according to this study.
Big five personality traits are kind of like that. From what I’ve read, they’re better understood as mostly-orthogonal surface regularities with causal explanations from many different levels and sources rather than as fundamental causally coherent essences. Lots of people seem to expect human traits to coherently cause human behaviors, so it is worth emphasizing how liable such thinking is to produce error.
The way I’ve heard it explained goes something like this: “you don’t like art because you are high in Openness. You are high in Openness because you like art.”
Of course, since the Openness scale has reliability, you can make predictions about how someone would respond to one question from the scale if you know what they would respond to another item. Whether that’s because of one underlying trait, or because of a bunch of converging traits, is an empirical question.
At least on the types, INTP probably refers to a real phenotype (which is common on LW), but I don’t know if any of the other type combinations are real.
Just wondering, are you generally classified as INTP? I’ve noticed that people consistently put in one of the types are more likely to think that their type is real.
Like wedrifid, I test as an ENFP on online tests, but if I answer questions like I would have if I hadn’t learned social skills, I come out as an INTP. The INTP profile I mentioned is freakily accurate, and not just in a horoscope type of way.
No. It compresses into 4 bits plus a whole bunch of extraneous knowledge of humanity and the environment. Sure, you can say it compresses down to four bits so long as you consider the language itself to already know all the basics about humans and the difference between the this and the other 15 combinations.
Some parts more than others I expect. Each of the details is supposed to be considered separately and with an “are more likely to” attached rather than a strict conjugation.
The interesting question is how many compressed bits the 4 bits convey. I’m guessing about 3.
Just wondering, are you generally classified as INTP? I’ve noticed that people consistently put in one of the types are more likely to think that their type is real.
At a guess yes, Hugh strikes me as someone who is naturally ‘INTP’ like. But the thing with the way the Myers Briggs test questions is that personal ideology and learned skills have rather too much influence. ie. Last time I did one of those tests I came out as ENFP. Which I’m definitely not, and wouldn’t have got if I didn’t answer the questions strictly literally.
At least on the types, INTP probably refers to a real phenotype (which is common on LW)
Myers-Briggs ultimately derives from the psychodynamic theories of Carl Jung, who was himself an INTP. Thus, it makes sense that INTP roughly corresponds to an actual personality type; Jung simply described himself, and then turned to his existing theories to explain away why he was the way he was.
Of course, since the Openness scale has reliability, you can make predictions about how someone would respond to one question from the scale if you know what they would respond to another item.
And this is the central point of the whole thing. They aren’t meant to represent a deep meaningful biological reality. Just clear correlations that are useful.
@JenniferRM—“However, when researchers tried the same thing on other cultures to see if this was a human universal, it turned out that the other four (Extraversion, Conscientiousness, Agreeableness, and Neuroticism) were relatively universal but in some cultures (if I remember correctly it was things like agrarian peasant societies?) basically everyone is pretty low in Openness relative to measurement norms derived from Japan or the US or whatever.”
I would note that the very framework of personality traits can be questioned as WEIRD bias. But personally, I’m fond of explanations of personality like trait theory. There is an attractive elegance to such models and the research is immense. On the other hand, defenses still can be made for trait theory, even for openness. One would predict that agrarian peasant societies, with above average rates of pathogens and parasites, would measure as below average specifically on openness. That precisely fits the point made in the above piece.
it turned out that the other four (Extraversion, Conscientiousness, Agreeableness, and Neuroticism) were relatively universal but in some cultures (if I remember correctly it was things like agrarian peasant societies?) basically everyone is pretty low in Openness relative to measurement norms derived from Japan or the US or whatever.
Why is that a problem? ISTM it simply means that it’s not a fixed-at-birth sort of trait. (i.e., like the others, it’s learned to at least some degree.)
Lack of universality isn’t a problem if you understand that the concepts are tricky to use for inference and should regularly, step-by-step, be checked against reality to make sure you’re still tracking reality. For example, Openness may actually be the most heritable trait in populations where twin studies were done despite being the least likely to be a human universal (recent mean heritabilities: Openness=57%, Extraversion=54%, Conscientiousness=49%, Neuroticism=48%, Agreeableness=42%).
But if we’re talking about something that is “not a fixed-at-birth sort of trait” then lack of heritability is one of the things I would have naively expected of the trait, you know? Its hard to shoot from the hip in this area—the terminology is stable and meaningful, and there is a relatively deep literature, but each new claim will require confirmation by experiment to extend your conclusions with any kind of rigor.
Water and milkshakes are both “arrangements of drinkable matter” and I can buy both sorts of matter in a cup at a fast food restaurant in the US. I can also expect to find water on Europa… but if I tried to use the fact that water and mllkshake were both “arrangements of drinkable matter” to predict finding milkshake on Europa as well, I’d have gone wrong in my thinking. It turns out that “arrangements of drinkable matter” isn’t a very useful category for naively deploying in sweeping extrapolative predictions.
Big five personality traits are kind of like that. From what I’ve read, they’re better understood as mostly-orthogonal surface regularities with causal explanations from many different levels and sources rather than as fundamental causally coherent essences. Lots of people seem to expect human traits to coherently cause human behaviors, so it is worth emphasizing how liable such thinking is to produce error. That doesn’t mean trait theory and the big five are bunk, it just means that you have to use the concepts with a measure of care, and some traits require more care than others. Openness is one of the tricky ones.
Lots of people seem to expect human traits to coherently cause human behaviors,
LOL. (ok, more like I went “bwahahahaha.. seriously?”) Wow. Yeah, I guess I can see why people might think that way, and I guess I must have thought that way at some point in the past. I can’t think that fuzzily about people any more, I’ve spent too much time inspecting behavior at lower levels.
In other words, when I saw the big five, I simply assumed they were summaries of behavioral patterns, and that it’d be daft to treat them as really predicting anything.
That is, saying someone has one of the five traits doesn’t explain anything, it just says, “this person is likely to do these things, because they’ve done these other things that seem to go together”.
(Or at least, if the big five are claiming to do anything more than that, I’d certainly be skeptical.)
[Edited to add: I suspect the downvoters have confused my LOLling at an instance of Fundamental Attribution Error with LOLling at JenniferRM’s comment. That is, they have probably failed to notice I am actually laughing with her, not at her.]
In other words, when I saw the big five, I simply assumed they were summaries of behavioral patterns, and that it’d be daft to treat them as really predicting anything.
Call me “daft”, then, because I still don’t get it. Does the “big five” model have any predictive power at all ? Does knowing the approximate position of a person in this five-dimensional space help us discover anything else about the person—specifically, something that we can verify empirically ? I’m not singling out the “big five” model for criticism; I’d ask the same question of Myers-Briggs, or enneagrams, or horoscopes. If your model has no predictive power, then it’s not very useful, regardless of how elegant it is.
Does the “big five” model have any predictive power at all ?
Sure, but only in the same sense that terminology like “rubes” and “bleggs” does. I more specifically meant that it has no explanatory predictive power—i.e., it doesn’t really tell you that, say, “Openness” is a valid physical construct in people’s brains, and not an accidental byproduct of some combination of other factors that are harder for us to notice.
...it doesn’t really tell you that, say, “Openness” is a valid physical construct in people’s brains...
I think this all depends on what your model predicts and with what accuracy. To use an analogy, a thing like “temperature” doesn’t really exist—it’s just an illusion caused by the motion of particles—but it’s still a very useful concept. In most cases, we can close our eyes and pretend that “temperature” is a measurement that really does refer to some physical quantity out in the real world.
So, is “Openness” (or any other of the Big Five axes) like “Temperature” ? Can we even measure a person’s “Openness” value reliably ? If so, what does it tell us ? What verifiable predictions can we make based on it ?
In the case of Myers-Briggs, AFAIK, the answers are “no”, “nothing much”, and “none”. In the case of temperature, we can definitely answer “yes” to the first question, and list a whole bunch of things like melting points and specific heat values etc. in response to the other two.
I don’t want to investigate this issue. But for the reason I gave, I can’t take anything away just from reading this post, which excludes me from its target audience. This doesn’t make the post bad, but is probably a consideration worth being aware of. (Making the burden of argument overwhelming, as you implied, wasn’t my intention in any case.)
You’re right, of course, but the case has already been made in just about every freshman psychology textbook.
It is reasonable to assume that an educated member of the general public will be familiar with the Big Five personality traits, and that a typical LW reader will have noted and remembered the Big Five as one of a handful of psychological ‘findings’ that actually do have demonstrable explanatory power. There is moderately compelling evidence that they are, e.g., heritable, that they can be affected by drugs, that they correlate strongly across cultures, years, and survey versions, and that they can be used to prospectively predict various mental disorders and success rates. The Wikipedia article http://en.wikipedia.org/wiki/Big_Five_personality_traits does a reasonably good job of documenting this. The Big Five are also the lead feature in Chapter 2 of “Psychology Applied to Modern Life,” (10th ed.), which I believe has been strongly recommended on LW before.
No doubt you meant well, but by suggesting that gwern used a “fuzzy label that one likes to attribute and then rationalize as having explanatory power,” you also suggest that gwern is indulging in sloppy mental habits and that gwern’s article lacks an appropriate foundation. Neither suggestion is really fair; the Big Five personality theory is so well-established that there is nothing improper, even among us skeptics, about relying on it and building on it to make a series of interesting points.
by suggesting that gwern used a “fuzzy label that one likes to attribute and then rationalize as having explanatory power,” you also suggest that gwern is indulging in sloppy mental habits
This shouldn’t be a problem, I believe. To the extent it becomes objectionable to doubt anyone’s sanity without a solid case, we are losing ability to guard against error. It should be a routine matter, like washing your hands.
Given these, plus Miller’s sloppiness in using “costly signaling” verbiage, I must agree with Nesov’s objections. The post wants to lead us to some conclusions, but these are only vaguely implied not stated, and the heavy use of quotations from the book, wall-o-text fashion, distracts from noticing that the post doesn’t really get to any particular point.
Which ones specifically seem weak? The problems with “garbage in, garbage out” and the lack of explanatory power of factor analysis seem serious enough to me. Basically, the WP page says that there is no “Big Five personality theory”, contrary to what the comment by Mass_Driver suggests.
What there is is a bunch of questionnaire items, the answers to which seem to somewhat reliably predict the answers to other questionnaire items when subjected to CFA; the label “Openness” for instance is arbitrarily given to one of those five “factors”, and if you look at the “Sample openness items” from the WP page it’s not clear that any of these questions in fact have to do with a person’s being open to new experiences.
Critics argue that there are limitations to the scope of Big Five as an explanatory or predictive theory. It is argued that the Big Five does not explain all of human personality.
If people claim that it is an all encompassing model, then that would be a serious criticism. I don’t actually know if researchers claim that, but it seems unlikely to me.
The methodology used to identify the dimensional structure of personality traits, factor analysis, is often challenged for not having a universally-recognized basis for choosing among solutions with different numbers of factors.
This is a weaker criticism than it seems. It stems from using classical stats. Latent factor (such as FA) models are less confusing from a Bayesian point of view. The particular indeterminacy they are talking about disappears when you specify a model and prior more clearly and try to update in a principled manner. A better criticism would be that the results of personality models might be sensitive to model assumptions. However, that sensitivity is an empirical matter and isn’t established by referencing the indeterminacy of FA. Another good criticism of Big Five research might be that factor analysis is not a good tool because it could be hard to figure out how sensitive your results are to the particular analysis.
Another frequent criticism is that the Big Five is not theory-driven. It is merely a data-driven investigation of certain descriptors that tend to cluster together under factor analysis.
What I’d like to know, primarily, is whether the Big Five traits correlate with anything other than answers to questionnaire items: in particular, actual behavior.
Pointer to studies appreciated; Google hasn’t turned up much. Some studies of job performance have C having the kind of influence you’d expect, then there’s things like this.
I think Bryan Caplan talks about the Big Five some and is usually pretty clear thinking. He has a paper on the topic which gives some references for correlations between the Big Five and other things. For example:
A particularly noteworthy aspect of the Conscientiousness—job performance link is that Conscientiousness is highly correlated (.5 to .6) with various measures of educational achievement but uncorrelated with measured intelligence. (Barrick and Mount 1991, p.5) Conscientious people are more successful in both school and work. In consequence, rate of return to education estimates that fail to control for Conscientiousness are likely to be biased upwards.
And
Criminals are on average markedly lower in both Conscientiousness and Agreeableness than non-criminals, even holding other variables fixed.1 (Costa and Widiger 1994; Wilson and Herrnstein 1985)
I’d be interested in hearing if you find interesting results.
I am also interested in that. My impression from hearing people talk about such things is that yes, they do, but I don’t actually know much about this area. That result is interesting.
Not just weak, but confused, and not even clearly criticisms, as they look more like snarky suggestions for improvement (might be Wikipedia page’s framing).
This discussion hinges on the “level of openness” being a big deal and a common factor explanatory of many different things. I think the first step should be making a case for this concept being useful, rather than a salient but fuzzy label that one likes to attribute and then rationalize as having explanatory power.
You want me to explain, cite, and defend the entire Big Five model with its factorization and the predictive powers thereof?
I’m sorry, I’m not going to do that. The goal of this article was to discuss some research leading to what I considered interesting speculation and questions. I linked to Wikipedia for the massive background such a veiled question wants (or you could read the rest of the book these short extracts were from, whose fulltext I also linked for readers such as you).
Vladimir_Nesov does not, but I do. More specifically, I’d be interested in seeing a defence of this or any other model, such as Myers-Briggs or what have you. To me, all these models sound suspiciously similar to each other as well as to horoscopes, but I could be wrong.
Don’t get me wrong, I’m not asking you to do so OMG RIGHT NOW or anything, but I would be interested in reading a post on the topic.
I think you are laboring under a slight misapprehension about personality research. Myers-Briggs isn’t solid science. The eneagram isn’t solid science. Astrological personality models aren’t solid science. I think you have correctly noticed that “psychological traits” are a ripe area for epistemically unsound belief systems that appear to bear on something people hold near and dear (ie understanding other people) so you’re justifiably suspicious of a mention of personality, which is laudable.
But you’re asking for a defense of “all that crazy stuff”, and a good defense of “all that crazy stuff” can’t honestly be provided, because most of it really is bunk, or at least it has so much bunk mixed in that its only good for psychoceramic data or maybe to pan for gold that might be hiding in the crazy. The big five personality model is an attempt to do actual science in the same space in order to produce reasonably valid and reliable dimensions of human “personality” variation. The point of the big five is that there is solid research and a deep literature and so on, in contrast to all the crackpot stuff.
If someone uses the big five and you’re suspicious and ask for a defense of personality systems in general, that’s like someone using geometry and you being suspicious because you’re only aware of a lot of crackpots who keep trying to square the circle and so you ask them to defend the squaring the circle stuff, (which was proved to be impossible in 1882) before you’ll accept analysis of evidence that legitimately makes use of a “suspiciously geometric” concept like the triangle inequality.
Unfortunately, defending established science quickly is hard because the content of science generally involves real inferential distances. If you want to start reading in this area, two useful keywords are Psychometrics and Trait theory.
In practice, “Openness to new experience” is the weakest part of the big five personality model. It can be measured reliably and predicts various things you’d expect it to predict and relatively naturally falls out when you settle for using 5 dimensions rather than 3 dimensions or 18 dimensions. However, when researchers tried the same thing on other cultures to see if this was a human universal, it turned out that the other four (Extraversion, Conscientiousness, Agreeableness, and Neuroticism) were relatively universal but in some cultures (if I remember correctly it was things like agrarian peasant societies?) basically everyone is pretty low in Openness relative to measurement norms derived from Japan or the US or whatever.
Your understanding is consistent with mine. Myers-Briggs is really frustrating, because some of its ideas are anecdotally compelling (Introversion vs. Extraversion, Thinking vs. Feeling), while others are esoteric (Judging vs. Perceiving and Sensing vs. Intuition). At least on the types, INTP probably refers to a real phenotype (which is common on LW), but I don’t know if any of the other type combinations are real.
Interestingly, the MBTI seems to almost reduce down to the Big Five according to this study.
The way I’ve heard it explained goes something like this: “you don’t like art because you are high in Openness. You are high in Openness because you like art.”
Of course, since the Openness scale has reliability, you can make predictions about how someone would respond to one question from the scale if you know what they would respond to another item. Whether that’s because of one underlying trait, or because of a bunch of converging traits, is an empirical question.
Just wondering, are you generally classified as INTP? I’ve noticed that people consistently put in one of the types are more likely to think that their type is real.
Like wedrifid, I test as an ENFP on online tests, but if I answer questions like I would have if I hadn’t learned social skills, I come out as an INTP. The INTP profile I mentioned is freakily accurate, and not just in a horoscope type of way.
Wow! All that compresses down to just four bits!
No. It compresses into 4 bits plus a whole bunch of extraneous knowledge of humanity and the environment. Sure, you can say it compresses down to four bits so long as you consider the language itself to already know all the basics about humans and the difference between the this and the other 15 combinations.
I can’t see 16 such highly detailed descriptions covering more than a small fraction of humanity.
Nu, with horoscopes each of 12 descriptions cover all of humanity!
Some parts more than others I expect. Each of the details is supposed to be considered separately and with an “are more likely to” attached rather than a strict conjugation.
The interesting question is how many compressed bits the 4 bits convey. I’m guessing about 3.
At a guess yes, Hugh strikes me as someone who is naturally ‘INTP’ like. But the thing with the way the Myers Briggs test questions is that personal ideology and learned skills have rather too much influence. ie. Last time I did one of those tests I came out as ENFP. Which I’m definitely not, and wouldn’t have got if I didn’t answer the questions strictly literally.
Myers-Briggs ultimately derives from the psychodynamic theories of Carl Jung, who was himself an INTP. Thus, it makes sense that INTP roughly corresponds to an actual personality type; Jung simply described himself, and then turned to his existing theories to explain away why he was the way he was.
And this is the central point of the whole thing. They aren’t meant to represent a deep meaningful biological reality. Just clear correlations that are useful.
@JenniferRM—“However, when researchers tried the same thing on other cultures to see if this was a human universal, it turned out that the other four (Extraversion, Conscientiousness, Agreeableness, and Neuroticism) were relatively universal but in some cultures (if I remember correctly it was things like agrarian peasant societies?) basically everyone is pretty low in Openness relative to measurement norms derived from Japan or the US or whatever.”
I would note that the very framework of personality traits can be questioned as WEIRD bias. But personally, I’m fond of explanations of personality like trait theory. There is an attractive elegance to such models and the research is immense. On the other hand, defenses still can be made for trait theory, even for openness. One would predict that agrarian peasant societies, with above average rates of pathogens and parasites, would measure as below average specifically on openness. That precisely fits the point made in the above piece.
https://benjamindavidsteele.wordpress.com/2022/02/06/weird-personality-traits-as-stable-egoic-structure/
Why is that a problem? ISTM it simply means that it’s not a fixed-at-birth sort of trait. (i.e., like the others, it’s learned to at least some degree.)
Lack of universality isn’t a problem if you understand that the concepts are tricky to use for inference and should regularly, step-by-step, be checked against reality to make sure you’re still tracking reality. For example, Openness may actually be the most heritable trait in populations where twin studies were done despite being the least likely to be a human universal (recent mean heritabilities: Openness=57%, Extraversion=54%, Conscientiousness=49%, Neuroticism=48%, Agreeableness=42%).
But if we’re talking about something that is “not a fixed-at-birth sort of trait” then lack of heritability is one of the things I would have naively expected of the trait, you know? Its hard to shoot from the hip in this area—the terminology is stable and meaningful, and there is a relatively deep literature, but each new claim will require confirmation by experiment to extend your conclusions with any kind of rigor.
Water and milkshakes are both “arrangements of drinkable matter” and I can buy both sorts of matter in a cup at a fast food restaurant in the US. I can also expect to find water on Europa… but if I tried to use the fact that water and mllkshake were both “arrangements of drinkable matter” to predict finding milkshake on Europa as well, I’d have gone wrong in my thinking. It turns out that “arrangements of drinkable matter” isn’t a very useful category for naively deploying in sweeping extrapolative predictions.
Big five personality traits are kind of like that. From what I’ve read, they’re better understood as mostly-orthogonal surface regularities with causal explanations from many different levels and sources rather than as fundamental causally coherent essences. Lots of people seem to expect human traits to coherently cause human behaviors, so it is worth emphasizing how liable such thinking is to produce error. That doesn’t mean trait theory and the big five are bunk, it just means that you have to use the concepts with a measure of care, and some traits require more care than others. Openness is one of the tricky ones.
LOL. (ok, more like I went “bwahahahaha.. seriously?”) Wow. Yeah, I guess I can see why people might think that way, and I guess I must have thought that way at some point in the past. I can’t think that fuzzily about people any more, I’ve spent too much time inspecting behavior at lower levels.
In other words, when I saw the big five, I simply assumed they were summaries of behavioral patterns, and that it’d be daft to treat them as really predicting anything.
That is, saying someone has one of the five traits doesn’t explain anything, it just says, “this person is likely to do these things, because they’ve done these other things that seem to go together”.
(Or at least, if the big five are claiming to do anything more than that, I’d certainly be skeptical.)
[Edited to add: I suspect the downvoters have confused my LOLling at an instance of Fundamental Attribution Error with LOLling at JenniferRM’s comment. That is, they have probably failed to notice I am actually laughing with her, not at her.]
Call me “daft”, then, because I still don’t get it. Does the “big five” model have any predictive power at all ? Does knowing the approximate position of a person in this five-dimensional space help us discover anything else about the person—specifically, something that we can verify empirically ? I’m not singling out the “big five” model for criticism; I’d ask the same question of Myers-Briggs, or enneagrams, or horoscopes. If your model has no predictive power, then it’s not very useful, regardless of how elegant it is.
Sure, but only in the same sense that terminology like “rubes” and “bleggs” does. I more specifically meant that it has no explanatory predictive power—i.e., it doesn’t really tell you that, say, “Openness” is a valid physical construct in people’s brains, and not an accidental byproduct of some combination of other factors that are harder for us to notice.
I think this all depends on what your model predicts and with what accuracy. To use an analogy, a thing like “temperature” doesn’t really exist—it’s just an illusion caused by the motion of particles—but it’s still a very useful concept. In most cases, we can close our eyes and pretend that “temperature” is a measurement that really does refer to some physical quantity out in the real world.
So, is “Openness” (or any other of the Big Five axes) like “Temperature” ? Can we even measure a person’s “Openness” value reliably ? If so, what does it tell us ? What verifiable predictions can we make based on it ?
In the case of Myers-Briggs, AFAIK, the answers are “no”, “nothing much”, and “none”. In the case of temperature, we can definitely answer “yes” to the first question, and list a whole bunch of things like melting points and specific heat values etc. in response to the other two.
Well yes, but the average kinetic energy does exist. So it just turns out that temperature didn’t quite mean what we thought it means.
I don’t want to investigate this issue. But for the reason I gave, I can’t take anything away just from reading this post, which excludes me from its target audience. This doesn’t make the post bad, but is probably a consideration worth being aware of. (Making the burden of argument overwhelming, as you implied, wasn’t my intention in any case.)
You’re right, of course, but the case has already been made in just about every freshman psychology textbook.
It is reasonable to assume that an educated member of the general public will be familiar with the Big Five personality traits, and that a typical LW reader will have noted and remembered the Big Five as one of a handful of psychological ‘findings’ that actually do have demonstrable explanatory power. There is moderately compelling evidence that they are, e.g., heritable, that they can be affected by drugs, that they correlate strongly across cultures, years, and survey versions, and that they can be used to prospectively predict various mental disorders and success rates. The Wikipedia article http://en.wikipedia.org/wiki/Big_Five_personality_traits does a reasonably good job of documenting this. The Big Five are also the lead feature in Chapter 2 of “Psychology Applied to Modern Life,” (10th ed.), which I believe has been strongly recommended on LW before.
No doubt you meant well, but by suggesting that gwern used a “fuzzy label that one likes to attribute and then rationalize as having explanatory power,” you also suggest that gwern is indulging in sloppy mental habits and that gwern’s article lacks an appropriate foundation. Neither suggestion is really fair; the Big Five personality theory is so well-established that there is nothing improper, even among us skeptics, about relying on it and building on it to make a series of interesting points.
This shouldn’t be a problem, I believe. To the extent it becomes objectionable to doubt anyone’s sanity without a solid case, we are losing ability to guard against error. It should be a routine matter, like washing your hands.
Objections have been voiced and not addressed. Not just here but on the WP page as well.
Given these, plus Miller’s sloppiness in using “costly signaling” verbiage, I must agree with Nesov’s objections. The post wants to lead us to some conclusions, but these are only vaguely implied not stated, and the heavy use of quotations from the book, wall-o-text fashion, distracts from noticing that the post doesn’t really get to any particular point.
Those WP criticisms seem surprisingly weak.
Which ones specifically seem weak? The problems with “garbage in, garbage out” and the lack of explanatory power of factor analysis seem serious enough to me. Basically, the WP page says that there is no “Big Five personality theory”, contrary to what the comment by Mass_Driver suggests.
What there is is a bunch of questionnaire items, the answers to which seem to somewhat reliably predict the answers to other questionnaire items when subjected to CFA; the label “Openness” for instance is arbitrarily given to one of those five “factors”, and if you look at the “Sample openness items” from the WP page it’s not clear that any of these questions in fact have to do with a person’s being open to new experiences.
If people claim that it is an all encompassing model, then that would be a serious criticism. I don’t actually know if researchers claim that, but it seems unlikely to me.
This is a weaker criticism than it seems. It stems from using classical stats. Latent factor (such as FA) models are less confusing from a Bayesian point of view. The particular indeterminacy they are talking about disappears when you specify a model and prior more clearly and try to update in a principled manner. A better criticism would be that the results of personality models might be sensitive to model assumptions. However, that sensitivity is an empirical matter and isn’t established by referencing the indeterminacy of FA. Another good criticism of Big Five research might be that factor analysis is not a good tool because it could be hard to figure out how sensitive your results are to the particular analysis.
This doesn’t seems like a criticism to me.
What I’d like to know, primarily, is whether the Big Five traits correlate with anything other than answers to questionnaire items: in particular, actual behavior.
Pointer to studies appreciated; Google hasn’t turned up much. Some studies of job performance have C having the kind of influence you’d expect, then there’s things like this.
I think Bryan Caplan talks about the Big Five some and is usually pretty clear thinking. He has a paper on the topic which gives some references for correlations between the Big Five and other things. For example:
And
I’d be interested in hearing if you find interesting results.
I am also interested in that. My impression from hearing people talk about such things is that yes, they do, but I don’t actually know much about this area. That result is interesting.
Not just weak, but confused, and not even clearly criticisms, as they look more like snarky suggestions for improvement (might be Wikipedia page’s framing).