SPENCER: Hi, Colin. Welcome to the Clearer Thinking Podcast.

COLIN: Thanks. Great to be here.

SPENCER: Everyone knows intuitively what personality is, but what is it really? How do you define personality scientifically?

COLIN: Well, scientifically, I would say that personality is just any kind of persistent pattern of behavior and experience that people have. So if there's something about you that you could use to describe yourself, that other people would be able to pick you out later on, that's part of your personality.

SPENCER: Would that include things like beliefs — if you think the world is a good place or a bad place, or you think the Republicans are good or bad? Would you encompass that? Or would you say that's separate?

COLIN: No, that's absolutely part of personality as well. So it's really just any kind of persistent feature of your emotional patterns, behavior, cognition, motivation; it's all personality.

SPENCER: It seems it has to really be persistent. It's not just a trait that you had today, but that you don't have it tomorrow.

COLIN: Yeah. In fact, we wouldn't even really call that a trait, right? A trait itself is something that persists over time. So something you do once can't really be part of your personality. It's something you do on a regular basis.

SPENCER: How stable is it? If you were to make the most accurate measure possible of someone's personality at age 20 and then again at age 40, to what extent are they the same person?

COLIN: For this, we probably would need to talk about some different parts of personality because the definition I just gave is really broad. It's just anything that's persistent about you. But we usually distinguish between these kinds of broad personality trait tendencies, things like extroversion, or are you generally an anxious or not anxious person, or are you imaginative or not? Those broad patterns that we usually call personality traits are different from some of the more specific things, like the habits that we have, beliefs that we may have, and some of the other stuff that you mentioned that people may not be as likely to intuitively think of as part of personality. In my work and some other people in the field, we refer to those things as characteristic adaptations, which is kind of a mouthful, but it's basically just ways that you've specifically adapted to your own life circumstances. So let's distinguish those from these broad personality traits. The traits tend to be pretty stable. The other stuff obviously can change more often. Your habits when you're in your teens are going to be different from when you're in your 30s, etc. But the traits become more and more stable as you get older.

SPENCER: So for children, they'd be most unstable. And then, what age do they kind of start solidifying? Is it in their 20s?

COLIN: Well, there are two different patterns that are important to understand. One is what's called rank order stability, and that's basically, if I measure a bunch of people now and then I measure them again in a year, what's the correlation between their scores at those two different times? Is the person who was the most extroverted the first time still the most extroverted the second time?

SPENCER: Oh, comparing them to each other, as a group, you're saying they are the most extroverted, still the most extroverted and so on.

COLIN: Yeah, exactly. So, not thinking about their absolute levels, but just in terms of how they are relative to other people. That tends to be lowest in little kids and just kind of steadily gets more and more consistent over the life course. But the reason it's useful to think about another kind of stability is that there's also your basic mean level stability. If we took the average of everybody at different ages, would that change? We see changes there too. So there are normative trends in how people's personalities change. When kids get to be around 12 or 13, hitting puberty, they start to get lower in their conscientiousness and in their agreeableness. Right around 15, kids are pretty impulsive and disagreeable — any parent is probably familiar with this stage. Then they start going back up again. By the time they are around 20, they're sort of back at a level that's as high, or maybe even higher than they were before, in those kinds of traits. If we just think about, "When are the largest, most dramatic periods of personality change?" That's probably adolescence. But even in adulthood, personality can change, and it tends to change differently for different people. It's not likely that your traits are going to change dramatically, but you can still see personality change over the whole lifespan.

SPENCER: I don't know if this is known, but if you take a trait like neuroticism, which is the tendency towards negative emotions and mental health challenges, we know that there are methods that can help people improve their mental health. Whether it's cognitive therapy or whatever method works for that person. Would you tend to see that it actually improves, reduces their neuroticism? Or is it more that they still have the same level of neuroticism, but they're much more functional at that level?

COLIN: No, it actually does change their levels of neuroticism. There was a great meta-analysis that came out a couple of years ago, done by Brent Roberts, which shows across all different types of psychotherapy, many different experimental trials, they found that personality changes with psychotherapy, and the biggest change is reductions in neuroticism.

SPENCER: Now, to what extent is that sort of a circular thing, where the way we measure neuroticism is essentially asking the same questions that we ask to measure mental health? So you can't have one change without the other, versus we can actually get at this thing that's the personality of neuroticism, that's not merely a measure of whether they have mental health challenges.

COLIN: No, we can't really distinguish that entirely. One of the things that is becoming clearer and clearer in psychology is that there isn't actually a firm distinction between mental health problems and personality. Any persistent mental health problem is by definition part of your personality. The one time when you can see a mental health problem that is clearly different from this person's personality is when, let's say, somebody is generally a happy person, but they go through a period, maybe a couple of months, of really intense depression, so much so that they end up with a diagnosis of an episode of major depression. Then they recover and go back to their normal happy self. A year later, you would say, "Well, okay. That was something that was a relatively isolated incident, and that really seems to contrast with the person's personality." That's quite rare, though. It does happen occasionally because terrible things happen to people; they get themselves into situations that make them depressed for one reason or another. But mostly what you see is that if somebody is prone to depression, they're likely to have multiple episodes of depression over the course of their lives, and they are also likely to, even when they're not in an episode of clinically diagnosed depression, still be relatively high in their tendency toward being depressive. That's the part of their personality that's persistent. They're generally kind of neurotic and depressive, even if it's not so bad right now that they've got a diagnosis.

SPENCER: I like this metaphor of waves of different frequencies for thinking about people, the way that you can kind of build up a complex wave from waves of low frequencies and higher frequencies. We have personality, which is sort of this very low frequency wave. Yes, it can change over the lifespan, but it's very stable. Then we have these very high frequency waves, which is, "What happened one second ago? How was the environment for one second ago? Did someone just spit in your face?" That's going to make you feel a certain way. Did someone just give you a compliment? Then we have these other intermediate things, like how well you slept last night, which maybe affects you for five hours or 10 hours. How would you think about this stack of things that are at different frequencies impacting a person in their behavior at a given moment?

COLIN: Yeah, I really like that analogy. In fact, I'm not sure it even is an analogy. It may just be a good mathematical way to describe patterns of behavior. We have individual behaviors that are changing from moment to moment based on our situations and based on whatever goals are most operative for us at that moment. In order to understand personality, we basically need to be able to say, "How can there be patterns in that kind of fluctuating behavior that are stable over time?" One way to think about this, I really like the work of Will Fleeson, who has done a lot of experience sampling research. What he showed, and he was really the first person to do this systematically, was that if you ping people a few times a day — now we would use their phones — back when he started, they were giving people pagers or BlackBerrys or something to carry around with them so that they could do assessments multiple times a day. You can just ask them, "What are you doing right now?" You give them a sort of list of how much they are talking, how much they are anxious, those kinds of things. You can basically get an assessment of the same kinds of behaviors that we think about in terms of these broad personality traits, but actually in terms of the specific moment-to-moment behavior. What is usually done is you ask people, in the last half hour or the last hour, "What was your behavior like?" What he found was that when you do that, let's say you do that to somebody for a week, and then maybe you take a break for a couple of weeks, and then you have them do it again for a week, people have tons of variation in their behavior. Even the most extroverted people are sometimes quiet. If you do that and you compare the average across those two different times, those two different weeks, it's very stable. People's average tendency amounts to their personality traits. Those are very persistent across time, even though within the person there's tons of change, responding to the situation or the mood or the goals they have.

SPENCER: This might be hard to answer, but I've met people who are sort of personality skeptics. They're like, "No, it's all about the environment." "What determines your behavior? Environment." If someone holds a gun to your head, everyone's going to kind of behave the same way. Okay, maybe not everyone, but most people will kind of behave the same way. Then there are people who are like, "Oh no, your personality just determines your life outcome." Maybe there's random fluctuations, but ultimately your personality is going to determine everything. Then there's everything in between. One way to think about this, you could almost think of it as a linear equation with interaction terms. You've got the effects of the environment on you, you've got the effects of your personality on you, and then you have the interaction, the multiplication of that. Do we have any sense, or is there even any way to quantify how much each drives what happens in someone's life?

COLIN: Yeah, that's tricky. It's analogous, but not exactly the same question as the nature-nurture question. How much is genetic and how much is environmental, and predicting differences between people?

SPENCER: Yeah, famously hard to pull apart, there are different contributors.

COLIN: Yeah, but there are tools that can be used to do that. This is called the person-situation debate, instead of the nature-nurture debate. It's how much does the person bring to each situation versus how much does the situation do to that person? When you have these kinds of dichotomies, usually the answer is both. Somebody who says, "Well, it's all genetic" is wrong. Somebody who says, "It's all environmental" is wrong. It's a mixture of both. The same thing goes for this question as well. If you're talking about somebody's specific behavior, you always have to consider their general tendencies and the specific situation they are in. In terms of quantifying it, it's probably even harder to quantify than the nature-nurture question. One of the things that Will Fleeson has pointed out is that there's usually about as much variability, or maybe even a little more, in the behavior of people within the person as opposed to across people. If you do this kind of experience sampling that I was talking about, you get a density distribution of these behavioral states that you've measured. Everybody has some kind of distribution, and they will have an average that usually looks roughly normally distributed. There's your average tendency, which corresponds to your personality traits, but then there's all this variation around that tendency, and there's about as much variation around the individual person's mean as there is in terms of the variation of the means across people.

SPENCER: That suggests to me that actually there's a lot of variability that's not explained by personality, is that correct?

COLIN: If we're talking about variability in moment-to-moment behavior, then yes, there's a huge amount of variability that's not explained by personality, because we're constantly reacting to the environment and the situation that we're in. It doesn't make sense. You don't just do the same thing all the time.

SPENCER: Although you might think that personality can get some of that, because your personality might affect the environments you choose to be in.

COLIN: It does that, yes, selection effects and evocation effects. You can choose what environment you're in, and the kind of person you are sometimes produces different effects from your environment. If you're a disagreeable, argumentative person, you're going to end up in conflictual environments a lot, and it's going to seem like, "Oh, maybe it's an environmental effect," but in fact, you helped to create it.

SPENCER: Is your sense that that interaction term — not the effects the environment would have on everyone, and not the effects that your personality would have for anyone who has that personality — but the way that they interact? Do you have the sense that that actually drives a lot of human behavior, or do you think that's sort of a correction term that's more minor in the equation?

COLIN: No, I think that drives a lot of behavior. I think there's a way in which personality is inherently an interactive concept. So if I say that I'm really extroverted, what do I mean? Well, we know that extroversion seems to reflect people's underlying tendency to be sensitive to reward possibilities. And that's why extroverts are more outgoing and assertive, and whatever else they are, experience more positive emotion, because they're just experiencing more things as potentially rewarding in their environments. But there still has to be something that's a potential reward. If I am literally just by myself alone in a room, I cannot act very extroverted. I might be a little bit happier than an introvert in that situation. That's an interesting finding. Extroverts actually experience more positive emotion, even when they're alone, than introverts do. Extroversion is not just about being sociable. It's actually this more fundamental difference in terms of people's sensitivity to reward.

SPENCER: It calls into question a little bit this idea of popularizing this idea that, "Oh no, you can be an introvert," and you're just getting pleasure out of different things. You're getting pleasure out of sitting alone and reflecting or having a quiet conversation. But it suggests that there actually might be something that's not that introversion is bad, per se, but there might be costs to introversion.

COLIN: Yeah, I wouldn't ever say that an introvert can't get pleasure out of doing the things that you just listed. Of course they can. I guess the cost to being an extrovert or to being an introvert is that your reward system is a bit less sensitive. Now, that means you might experience a little bit less joy in life, but it also means that you're going to make fewer stupid, risky decisions.

SPENCER: So there's an extraversion cost as well. It's just a different kind of cost?

COLIN: Yeah, with any of these personality traits, there are usually trade-offs.

SPENCER: Do you think that all personality traits are that way, that essentially, there's a trade-off occurring where there are ways it's better to be high on it and ways it's better to be low on it, not necessarily to max it out or minimize it, but that we're really talking about a trade-off curve?

COLIN: Yeah, I think that's almost universally true. Neuroticism turns out to be the hardest one to identify. There are benefits to being quite neurotic.

SPENCER: What about safety? Because you can imagine, and you're right. Today, let's suppose you're living in a safer environment. If you're worried a lot and stressed a lot and depressed a lot, that doesn't seem good, but the tendency towards worry could keep you safer, especially in an environment that has a lot of danger.

COLIN: Yeah, and it may have to do with social environments, and also potentially social environments that are more like the ones that we evolved in. But there's this really unfortunate finding, which is that people who are high in neuroticism not only perceive situations as more stressful, they also end up in more stressful situations, as reported by others.

SPENCER: Later on, after they've been measured for their neuroticism, you can predict that in the future, they or other people perceive them as getting themselves into more stressful situations.

COLIN: Yeah, in as much as we can identify more objective measures of stressful, challenging, or traumatic situations, people who are high in neuroticism are more likely to have those kinds of bad things happen to them.

SPENCER: And that's not just because people who live in difficult situations tend to develop more neuroticism and continue to live in those bad situations. We actually think it's derived; somehow the personality trait is driving it.

COLIN: Well, I think that's a really fascinating question, and that would probably be what needs to be controlled. What we can rule out is that it's not just that they perceive their situations to be more stressful. We know that that's true also. You give people the same situation, and the person who's higher in neuroticism will perceive it as more stressful. But there's also some way in which they end up in more stressful situations. Now, what you're suggesting, of course, is that they may have been in a persistently stressful situation that also helped to make them more neurotic. I think that's a really important possibility.

SPENCER: What do we know about what causes people to have different personality traits? There's a genetic component. How strong is that genetic component? And do we know anything about what sort of environments create different sorts of people?

COLIN: Yeah, that's a great question. The basic answer is that, yes, it's partially genetic and partially environmental.

SPENCER: It's like everything is 30 to 70% genetic, as far as I can tell.

COLIN: Personality is no different. I would say we still don't know exactly where it is in that range between 30 and 70% because there are various drawbacks to different forms of measuring it. But let's call it 50/50, or something like that, just for the sake of discussion. The genetic piece, obviously, there's not much you can do about. Maybe, if you've got a genetic propensity toward a particular type of dysfunction, we can find some kind of pharmaceutical that can address the way that your brain happens to function. But so far, I wouldn't say that our track record with that is too great. The brain is too complicated, and the drugs — we talk about have side effects, but really, they're just effects, and some of them we want, and some of them we don't.

SPENCER: It seems that our standard procedure seems to be that we accidentally discover a drug has some effect, then we make up an explanation of how it works on the brain. I'm talking about psychiatric meds, and we give it to people for a long time and eventually realize it's a mixed bag. Maybe it helps a lot of people, but it often does things that we don't necessarily want as well, right?

COLIN: It's very hard to predict who it's going to help and who it's not, even if they have the same diagnosis. It's complicated, but the environmental side is what you were really getting at. What do we know about what happens to people that change their personality? In terms of adult personality change, or even just adolescence and young adulthood, there was a lot of optimism that we would find a bunch of things that affected personality in predictable ways, like, "Oh, maybe when you have kids, that's going to change your personality." It turns out that for most things, people's personalities do change in response to dramatic life events, but not necessarily in predictable ways. As far as I know, there are only two things that have been found to predict people's personality change in reliable ways, and those are getting into your first serious romantic relationship and getting your first job.

SPENCER: Oh, wow. How do they tend to change it? I guess I would predict conscientiousness might go up with the first job.

COLIN: Yeah, they tend to make people more conscientious and more agreeable. You get this sort of change toward a more mature personality profile.

SPENCER: For both of those?

COLIN: For both of those things. But for everything else, there was a period where people were starting to do a lot of large longitudinal studies and just not finding very much. Now what we've moved on to, I think, is a really cool development, basically looking at individual differences in the way that people react to the same events. For example, some people have kids for the first time, and it makes them get their lives together. They become more conscientious, more organized, they start giving up on frivolous things they were doing and really getting their stuff together. Other people, it's the opposite. They have kids, and it's just too much, and now suddenly everything's falling apart. They used to be relatively on top of things, but this is just one thing too many and a very big thing, and they become disorganized and less conscientious. You can see why you wouldn't see a prediction that was across the population. It's like, "Oh yeah, having kids makes you X because it makes some people X and makes other people not X."

SPENCER: Oh, wow. So the average effect doesn't do a good job of capturing the effect of these events. We actually have to look at the individual and say, "How does this affect different sorts of people?"

COLIN: We have not just changed, but individual differences in change.

SPENCER: If you ask me to guess what would have an effect on personality, I think I would guess that extreme trauma or difficult life circumstances would affect personality. Is that not the case?

COLIN: Oh no, that absolutely is the case. So, okay, and that, I think we can especially see that with studies of development earlier in life, that maltreatment of one kind or another, neglect, abuse, various kinds of traumatic events in people's development certainly puts them at higher risk for all kinds of mental health problems, makes them more neurotic. Basically, I was saying there's this connection between psychopathology and personality, mental health and personality. Each one of the personality trait dimensions corresponds to some particular type of mental health problem that people can have, and so those tend to be affected accordingly.

SPENCER: Oh, really, each trait corresponds to, or you're saying some personality traits correspond to mental health challenges.

COLIN: Well, if we think about the major personality trait models, the Big Five model, for example.

SPENCER: Maybe you could just walk us through the five, actually, it's a good time to do that.

COLIN: So, okay, I've been talking about extroversion versus introversion. And I guess the one thing to say about that is that people who have come from the background of being familiar with the MBTI and Jung's idea of extroversion and introversion often have the wrong idea about what introversion is from the modern scientific perspective, because Jung thought of it as sort of about being inwardly focused and more imaginative and creative in the life of the mind and that kind of stuff. But that turns out not to be the opposite of extroversion. The opposite of extroversion, and therefore, maybe unfortunately, what tends to get called introversion now in scientific psychology, is basically just being quieter, more submissive, less outgoing, less talkative, just the opposite of the things that we usually associate with extroversion. The stuff about being more imaginative, creative, interested in fantasy, interested in mental and intellectual life, that's actually seen as a distinct trait, and that's the openness to experience dimension. That's one of the Big Five. So there's extraversion, there's openness, there's neuroticism. We talked about that; that's basically sensitivity to negative emotions of all kinds. There is conscientiousness, and that's how organized and self-disciplined you are versus being disorganized, scattered, impulsive. And then the last one is agreeableness, which basically collects all of the traits that are related to being altruistic and cooperative versus being selfish, exploitative, and antagonistic. In the same way that there are these models of personality that are mapping out these major dimensions, people have been doing that same thing with symptoms of mental health problems as well. And it turns out that there are five major dimensions of mental health problems that basically correspond very closely to the Big Five.

SPENCER: Oh, wow, I was actually not familiar with that. Could you walk us through what are those five dimensions of mental health challenges and how they map on?

COLIN: Yeah. One place to start looking for this is research on the hierarchical taxonomy of psychopathology, which is abbreviated to HiTOP. That's a consortium of up to now probably over 200 people at this point. The whole goal of that was basically to try to do something similar for symptoms of psychopathology as was originally done for personality that led to the big five. In other words, rather than relying on traditional medical wisdom to say, "How are people's mental health problems organized?" To instead let the data speak for themselves. Basically, to look at patterns of co-occurrence of different types of symptoms and say, "Well, if somebody has this type of symptom, are they more likely to have this or that?" When you do that, there's a lot of work using factor analysis, the same kind of approaches that were used in personality. The HiTOP structure basically shows that there are these six major dimensions. Five of them correspond pretty closely to the big five. I'll come back to that in a minute. Then there's a sixth one that's a little bit more tentative, that is about somatoform problems, when people have problems that they experience as being physiological, but there's no detectable physical problem.

SPENCER: So with hypochondria and psychosomatic hypochondria?

COLIN: Hypochondria and psychosomatic hypochondria are one of the core features of that. It's mental health issues that are expressed in terms of physiological concerns, basically. I should also mention that one of the things that HiTOP doesn't do yet, which is a real lack and we're working on, I'm part of the consortium, is to cover symptoms of diseases that are mainly diagnosed in childhood. For example, it doesn't currently incorporate autism symptoms very well. There will most likely be added another spectrum that's going to be called the neurodevelopmental spectrum, where you'll see a lot of that stuff falling. But for now, there are these five core spectra. There's one that's called detachment that is basically problematically low extroversion. It's like, "I can't make connections to other people. I never experience any pleasure."

SPENCER: So an example that might map onto this would be schizoid personality disorder, where people have really low interest in social connections and are detached from society.

COLIN: Yes, so schizoid would sort of be the prototypical example of that. There's nothing necessarily else that's wrong, except for a crippling inability to connect to other people and experience positive emotions and be at all assertive for oneself. All those components of extroversion are basically missing, and there's nothing wrong with being introverted, with being less assertive, or less of a bubbly person, but it can get to be so extreme that it causes people real challenges in life.

SPENCER: But would that also include things like social anxiety, where you want to connect with people but struggle to do it, or is that actually a different factor?

COLIN: Social anxiety would usually be maybe a bit of a blend of some of that, but also just the more what's called internalizing, which is basically just the same thing in personality as neuroticism, a general tendency toward anxiety and negative emotions. That can manifest, obviously, socially, and it's more likely to manifest socially for someone who is farther out on that detachment spectrum or more introverted. If they're also anxious, that's likely to then manifest in social anxiety, but you can find people who are not particularly neurotic but who are still just extremely detached. They just have a hard time connecting to other people and acting in the social world, being agentic.

SPENCER: And what else would go in internalizing? So depression goes in there as well.

COLIN: Yeah, depression, anxiety, panic, types of problems. There's debate within the sort of HiTOP research about how much sexual problems and eating disorders are part of internalizing or should be modeled as their own separate spectra. One of the nice things about HiTOP is that it's designed to be an empirical system, so that it can evolve as the evidence accumulates and more scientific research is done. But those problems are pretty strongly linked to the more classic internalizing problems like anxiety and depression as well.

SPENCER: Would anger not be internalizing? Because I think of anger as part of neuroticism, but I think you're going to tell me it's not internalizing.

COLIN: That's a complicated one. Yes, within the normal personality space, the tendency toward experiencing irritability, hostility, and anger is linked to neuroticism. It's probably a bit in psychopathology. It's partly that. People who are anxious and depressed often have problems with getting irritated and upset easily as well, but it's also linked to these what are called more externalizing problems, which have to do with the spectra of disinhibition, which is like low conscientiousness. That's like having trouble controlling your behavior well, and then the antagonism spectrum, because one of the ways that anger manifests problematically is when it's explosive and directed at other people, and then that becomes more like the kind of antagonism, disagreeableness features.

SPENCER: And where would things like alcoholism or drug addiction go? Would that be disinhibition?

COLIN: Yeah. So those are located primarily within disinhibition.

SPENCER: It's really cool. You think about the DSM, which started as a kind of committee of people meeting and saying, "Well, maybe we should add this, and maybe we should add that." There's some kind of meeting that happens, and nobody really knows how the sausage gets made. They end up putting really weird things in weird buckets, but it's improved over time. We've seen some improvements, for sure, but definitely not an entirely empirical project, right? There are a lot of opinions involved. I love the work you're doing to try to say, "What does the data really tell us? Let's actually try to carve reality at the joints and not make this based on just our subjective judgments."

COLIN: And that's sort of exactly the goal of HiTop, in fact. Ultimately, its greatest ambition, which is probably fairly distant at this point, would be to replace the DSM as a system for diagnosing people. Instead of being fitted, because one of the things that we know, and there's very clear evidence for, is that mental health problems are not these categorical entities where you either have it or you don't. We're getting used to the idea of things being on a spectrum. But it turns out that everything's on a spectrum, and everyone's on every spectrum, because you either have a high level of it or a low level of it, or somewhere in between. I think that moving to that kind of system would actually be very good in terms of reducing stigma around mental illness, because there's this sense in which we tend to find, "Oh, what is your problem? Oh, you have ADHD. Oh, that explains so much. You're that type of person. We put you in a box. Then we treat you as if you were different from the rest of the world." But in fact, that box doesn't exist. You just have more of certain traits than other people do, and sometimes those traits cause people problems.

SPENCER: It seems that putting things in boxes is such a problem that plays on so many systems. The human mind just really wants a box for it. You can see this in personality, where the most popular personality system in the world is the Myers-Briggs style, where you get one of 16 categories. People want to know what your type is. Are you an ENTJ or whatever, or Enneagram? What's your type? You also see it in the DSM, putting people into mental health categories. I feel like a similar thing almost happened in genetics, where we wanted to say, "Oh, you have this gene, and this gene means you have this thing or you don't." But it turns out it's very hard to find things that work that way. Very few human traits we care about are just one gene. It turns out you actually need a giant model that's taking in thousands of genes at the same time in order to predict anything of interest, except in these sort of super rare cases.

COLIN: Right. Yeah, and Fisher predicted this over 100 years ago. There are very few what are called Mendelizing traits, where one gene or a couple of genes are doing all the work. I remember because I was involved in molecular genetic research very early on for behavioral genetics. In my dissertation in the early 2000s, I had some studies of genetic variants, and at that time, we were looking for individual genes that might have a measurable effect. We didn't think there would be three or four genes that determine schizophrenia, but at least we thought maybe there would be 100 or 200, and we could find them. Well, no, it turns out that there are thousands of genes influencing any one of these complex traits, whether it's normal personality variation or whether it's risk for different kinds of mental health problems. So, in a way, that is kind of analogous. We always want things to be these tidy boxes.

SPENCER: If you were to imagine plotting people in n-dimensional space, each person's one point in space. Imagine in three-dimensional space, each person is a point. We could live in a world where there are nice clusters. "Oh, there's a bunch of people over here, and there's a gap. You don't really see anyone in between." But it seems almost nothing with humans works that way. There may be a few weird exceptions, but almost nothing. It's almost always this continuous, right?

COLIN: Yeah. There are no zones of rarity where you would get a clear distinction. This has actually just been studied empirically. You see a pretty smooth distribution. Now, there are places that are denser than others within that multi-dimensional space. Because there are correlations among different traits, but you can find people with any combination. Even if they're rarer, they're not that rare. So, there's no clean way to carve nature at its joints and make it into categories. You have to be using these dimensional assessments.

SPENCER: Yeah, one kind of mystery that some people might wonder about is where's the sort of really super scientifically accurate 16 categories of personality or whatever. Trying to do what Sir Myers-Briggs did, but in a super empirical way? And I think the answer is that it doesn't carve reality at the joints very well to try to do that.

COLIN: People have tried, and it doesn't work for empirically demonstrable reasons, which is exactly what you're saying: there just aren't these gaps in the multidimensional space. The interesting thing is there are these statistical programs that are designed to find categories. If you get a big sample of personality data and you feed it into one of those programs, it will give you categories because that's what it's designed to do. It will carve it. But the problem is that it's not carving it in ways that are replicable or robust. If you do the same thing in another sample with the same assessments, you'll get different groupings, and that's because the program is designed to force things into categories, but it's working with data where the underlying structure really isn't categorical, so you don't get robust or reliable categories. So there will never be a scientific version that says there are 16 types of people, or there are 20 types of people, or there are six types of people, or whatever.

SPENCER: So we've actually tried to do this with our own personality test. We ran a clustering algorithm. We said, "Okay, people like 16. We'll just look for 16 categories, and the algorithm will find them, and it will capture variability." You can do it. It will capture variability. You know, people within a type are more similar to each other than people not in the type. But as you say, there's an instability there. It's not a fundamental "We found the 16 types of people." It's like, "In this set of questions, we can group people in this way, and people in one type are more similar than across types," but you have not learned something about human nature so much as you've grouped people in a way that compresses the data. If we do clusters, we don't seem to get these repeatable patterns that we can find in every data set, but with factors, we do find something more repeatable. So can you tell us about that and how that led to the construction of the Big Five?

COLIN: So if we've got a big sample of descriptions of personality, individual items, a lot of this work was originally done just by going to the dictionary and extracting adjectives that can be used to describe people. The benefit of that was that it's a way to get a relatively unbiased sampling of a very broad range of descriptions of people's personality. Before that, before the work that led to the Big Five, there were tons of personality questionnaires being used in psychology, but everybody had sort of the one that they created that they liked, based on their ideas. There wasn't really a sense of, "How can we integrate all this empirically and synthesize?" And the dictionary served as a pretty good and relatively unbiased pool of items. If you get people to rate themselves on lots of these items, hundreds of different adjectives or other types of personality descriptions, then you can do factor analysis. What factor analysis is designed to do, basically, is just to pull out clusters of items that vary together, so things that tend to go together, and to identify these underlying dimensions of variability that are associated with many different individual items. You see things like being anxious, depressed, nervous, upset, and irritable all tend to appear together in the same people. Those are correlated with each other. That would be describing the neuroticism factor. Once people started doing this, they found that there was this five-factor solution that was quite replicable across different samples and even across different languages, that seemed to be a good parsimonious way to describe the major dimensions of patterns of covariation among different personality traits.

SPENCER: So whereas clustering is trying to find groups of people, factor analysis is trying to essentially find groups of questions you could ask about people and create these spectrums. Is that right?

COLIN: That's exactly right. Yeah. So it's identifying groups of items that seem to all be related to each other and tapping into some broader underlying dimension.

SPENCER: And if someone rates themselves as organized, they also tend to rate themselves as on time and perfectionistic. So we get these five groupings. Now, how robust are these five groupings? If we go to different countries, do we get the same effect? If we try asking the wording in a slightly different way, do we get the same result?

COLIN: Yeah, that's a good question, and maybe slightly more complicated than it seems on the surface. This lexical approach of going to the dictionary and extracting words that describe people has been done in languages all over the world, more than two dozen different languages. That's actually where one of the competitors to the Big Five has come from, because it was found that there was a six-factor solution that appears in more languages than the Big Five does. The Big Five doesn't appear in all languages, but even the six-factor solution also doesn't appear in all languages. In fact, if you want to get the personality framework that works across the most possible languages, what you end up with is a two-factor structure, one that's referred to as dynamism, which relates to being extroverted, intelligent, powerful, and competent, and then the other that's often referred to as social propriety or social self-regulation. It's basically about being a good, dependable, reliable person that others would want to have around and want to be responsible for things. Those two major dimensions can be found in basically all languages.

SPENCER: How controversial is this? Is it generally accepted like, "Okay, these two-factor solutions are really the best if you want to go all the way around the world?" Or is that still up for debate?

COLIN: That's pretty well accepted at this point. The evidence is pretty extensive, and even proponents of the Big Five, like I prefer the Big Five system, would agree that it doesn't necessarily replicate in every language where you go to the dictionary and pull words out. But something important to note is that if you take a measure designed to measure the Big Five and translate it into another language, it usually still produces the same Big Five structure. The more you get into countries that are less economically developed and have lower levels of education, the more that's likely to break down. There has been some interesting research that shows that if you are careful to use items that are more appropriate for that culture, you can still successfully see the Big Five structure there. There are various challenges with transporting these systems into different cultures. One way to think about it is that the two-factor structure works, the five-factor structure works, the six-factor structure works. They're not that different, especially the five and the six, which are really just rearranging some of the same pieces in slightly different ways. Since a factor structure, if we think of a hierarchy, we sometimes think about it as having what's called simple structure, where "Here's an item that only measures one of the Big Five." In fact, usually for any given item, it's going to be correlated with multiple of the Big Five. It'll probably have its strongest correlation with one of them. We are often inclined to use it to measure that one.

SPENCER: Like perfectionism, you could imagine there's a certain perfectionism that involves anxiety around performance, and that might be partly neuroticism instead of just conscientiousness.

COLIN: Yeah, that's a perfect example. Perfectionism is going to have what's called a cross-loading on neuroticism, even if it has its primary factor loading on conscientiousness. In as much as that's true, personality doesn't really have what's called simple structure, where everything is only related to one factor. That means you can slice it up in slightly different ways. It's not like one of them is necessarily right and one of them is wrong; they're just slightly different ways of slicing up these patterns of covariation. What you're looking for are ones that are reasonably robust and replicable, and that have good validity in terms of reflecting real underlying patterns of behavior and sources in the brain and that kind of thing.

SPENCER: It seems to me that there are some different, slightly different claims. They seem similar, but are meaningfully different. One is, if you go to different countries and use their local language, and you pull all the adjectives in the dictionary, and you have people self-rate, and you factor analyze that, do you get the Big Five right? Part of not getting the Big Five in some countries could be due to the way their language works. Maybe you need multiple words strung together to get all the concepts right. It doesn't mean they don't have that concept, but maybe they don't have it as an adjective. So that's one thing. Another thing you could be claiming is, okay, let's say you translate an actual Big Five survey. It has all these questions for the Big Five into different languages. Do you get the Big Five? That's, in a way, a much easier challenge because you're already limited to just these five factors. The question is, do the items group the way you expect? Does it group in an organized way of being perfectionistic, or does it group in some other weird way? But that's a much weaker claim because an additional thing you could do, which is a little bit unspecified, is to imagine you give someone every personality question you could possibly imagine, whether it's in their dictionary or not, whether it's in the Big Five or not. Then do you get the Big Five, or is it inherent to the nature of personality?

COLIN: I think going across languages, I've always wanted to do something that was kind of like getting the best of both of those two approaches because there is this inherent limitation where, if you're starting from the natural language in a place, you are going to end up with a different set of variables than what you got somewhere else in their dictionary, and that could create differences in factor structure that are based on the inputs rather than on the underlying structure. What if you went to all of the different dictionaries that have been studied, created a pool of essentially all the words across all the languages, and then, as much as possible, translated all of them into each other? Let's say, in the end, you've got a thousand words, and maybe in some languages you need to use a short phrase instead of just a word for one of the items. But everybody then is getting basically the same thousand items across the world. They have come from everywhere across the world as well. You have the unbiased spread in the representative pool of descriptors, but you also have the same variables in each language. That would be a really cool study.

SPENCER: I love it. That seems like the ultimate test. What do you think we would find? Would we get the Big Five, or would we only get these two, dynamism and social propriety?

COLIN: I think we'd get the Big Five. I think we would get the two because the two is so robust, even across languages where you're not necessarily having exactly the same inputs.

SPENCER: But do you think you would also get the five on top of that?

COLIN: But I think we would probably also get the five. One of the reasons I think that is that there is currently a project that is basically attempting to map out every reasonably coherent, unidimensional facet-level trait of personality. A facet is a relatively narrow trait. Neuroticism is the broad trait, and then irritability would be a narrow trait, or anxiety would be a narrow trait. They are basically trying to empirically identify every distinct, narrow-level trait. This work is being done by David Hughes and Paul Irwing. It's a great project, an ambitious project, and they have just gotten to the point where they have a list. It's now somewhere in the 70s, I think, in terms of, like 71 or 75, something like that, different facets. When they factor that, they get the Big Five. That would still need to be done across many different languages. But what it suggests is that there's another way to get to the Big Five that doesn't just rely on the natural language approach, on the lexical approach, and that if you try to get an exhaustive list of reasonably distinguishable facets, many of the adjectives are not really distinguishable from each other, empirically. Is anxious different from nervous? It's probably not really tapping into something different.

SPENCER: And you kind of look at the correlation structure to make those decisions. You say, "Look at these two facets. Are the two correlated? We're just going to call it one facet." Is that how it's done?

COLIN: Right. Well, there are multiple ways you can do it. You can look at the degree to which they're correlated, the degree to which they predict the same or different things. You can see what's the network of associations with all the other variables. Is it the same or different? If two things are really distinct from each other, they should have what's called discriminant validity. They should predict different things.

SPENCER: And then you could use conceptual analysis, saying, just like other concepts, that they're measuring the same.

COLIN: Yeah, they're starting with conceptual analysis basically just to try to find the broadest range of possibilities. But then they're using empirical analysis to narrow it down and say, "Which of these are really distinct?" You can use confirmatory factor analysis to say, "Are all these items really just measuring one thing or not, and how tightly?"

SPENCER: Now, why is nobody arguing for the big seven or the big 14?

COLIN: It's just empirical. There was, for a while, a big seven, which was basically the big five plus factors labeled as positive valence and negative valence. This was also when people were going to the dictionary, but they weren't excluding things that seemed purely evaluative. We don't usually call someone awful, horrible, wonderful, or great — we don't usually think of those as personality descriptors because they have no specific content. They're just a pure evaluation. It's like, "Well, what's wonderful? How are you acting that actually makes people judge you that way?" But you can include them; they are descriptive of people. The short answer for why people are only focusing on two, five, and six is that that's just empirically what's in the data.

SPENCER: What does it mean when you say what's in the data? What happens if I make my Spencer's eight-factor model of personality? What will I see in the data that will show me, "Nope, it's the wrong number?"

COLIN: A great question, and there are a couple of ways to go about that. One way you can do this in factor analysis is through parallel analysis, where you create a random data set that has the same number of variables as the data set you're using with the same kind of distributional properties of the variables, but it's just random data. Then you run the factor analysis. Interestingly, just because of noise, there are patterns of covariation among the variables even in a random data set. You can identify the longest axis of covariation through that multidimensional cloud in the same way that you can in real data. That will be your first factor. Then you go to your second factor, which is the orthogonal axis that's the longest next to that. If you do that with a bunch of random data sets, you can identify, "Okay what would it look like if we have the pattern of eigenvalues, which are basically the length of those vectors, would look like in random data?" Then you can compare it to your real data. You can look to see where those curves cross because the random data usually looks like a line going down, and the real data looks like a curve. Where the curve of the real data drops below the curve of the 95th percentile of the random data, you say, "Well, everything above that are probably real factors because there's more than what you would just see if you were analyzing noise."

SPENCER: So you're essentially saying some factors we see when we do factor analysis are essentially going to be noise, because even pure noise, you're going to find some factors there. Let's make sure we're only looking at things above and beyond the noise. And you're saying, I think that there aren't that many, maybe six, maybe five above the noise.

COLIN: Well, here's the interesting thing. So you can use that in some context. You can rely on that kind of what's more than noise. But it also turns out that if you just take one of these big data sets with 500 adjectives that people have rated about themselves, and you do that kind of parallel analysis, you're going to find more than five factors. It will say that there are more than five factors. And that's because when we're interested in a model like the big five, we know that we're grouping lower level traits together. We know that there's a hierarchy, and so anxious and nervous and tense might form one factor within neuroticism, and irritable and angry and peevish might form another factor. But we're just ignoring that because we're grouping them all into neuroticism. But in terms of patterns of co-variation in the data, those actually do cluster together. And so you're going to see that there's more than five factors. And so one of the other ways that people attempt to identify what's the right number of factors is to look at what's replicable. So can I do the same thing in another data set and find the same factors in terms of their content? Different patterns of larger numbers of factors than five or six don't tend to be replicable across lots of different data sets.

SPENCER: If I run on one data set, I get a different seven factor than if I run another data set. What is the seven-factor? I can't even name it because it's not stable. It's a completely different thing every time.

COLIN: Exactly. Yeah. Now there's interesting work. Gerard Saucier is a really great researcher in this area, especially in the lexical research he's done in languages across the world. He got interested in, "Can we do higher dimensionality analyses that might be reasonably replicable?" And he was able to find a solution that had, I don't know, 22 or 23 factors in some of these lexical data sets. He was only looking at English, but he was looking across multiple samples and large collections of adjectives. And he could kind of try to make it work for these larger dimensional structures, but they're just nowhere near as stable. You can try to make it work. But if you really want something that is this kind of balance between parsimony and stability, you just end up with these five or six factor solutions, basically.

SPENCER: Another claim I've heard is if you try to push to higher factors, you just start splitting them apart. If you go from three to four, you really feel like you're adding something for the fourth, but if you go from nine to 10, you're just like, "Oh, I just, I could see what happened here. I just took one factor before, and I just split it down the middle. Some half the items go here and half the items go there." Does that seem right to you?

COLIN: That does often happen. You can get things it's like, well, these are clearly just different flavors of neuroticism. Yeah, and this would be controversial, but I would say that even when you go from five factors to six factors, you're basically just splitting things slightly differently. Now the six factor people are going to vehemently disagree with this, I'm sure, but even they will acknowledge that two of the factors in the six factor solution are actually just a different rotation of two of the factors in the Big Five. It's just basically taking neuroticism and agreeableness and rotating them 45 degrees. So you get two different factors, but they're really just rotated versions of the Big Five factors, and then the sixth factor in the six factor solution is usually referred to as honesty and humility, the H factor. But the fact is that within the Big Five, you can group those as part of agreeableness, and the whole structure still works. So I would argue that that's basically just splitting agreeableness. The only reason it doesn't look like just a split in the six factor solution relative to the five is that these other two are turned relative to each other, and one of the ones that split is actually kind of split and turned.

SPENCER: I think it's interesting, though, because this question is, "How many factors of personality are there? It seems you could live in a world where there is no answer to that, but it seems like we live in a world where there's something to be said about that. And five versus six, okay, maybe that's not such a big difference, but it does raise certain questions, so this H factor, honesty, humility, the six factor model adds, it seems like it's something like how good or evil you are. Is that right?

COLIN: Well, if you allow, if you add words like good and evil into the analysis, right, then yes, they do tend to cluster around that factor. So it certainly has to do with people's tendency to be moral or immoral. But that's a crucial part of agreeableness. Because agreeableness is all about, are you altruistic and generous and helpful to others, or are you manipulative and selfish and hurting other people? So you can see why. You can also make the argument that that can just be described as part of agreeableness. Now, what I think is really crucial to understand about this question is that the right way to think about personality is as a hierarchy. So it's not like we're saying there's only these five traits, there's five traits, but then you can break them down into more traits, and then more traits below that, and that's the way you deal with this.

SPENCER: If you were to split it apart, if I were to take my five traits and say, "No, I'm going to treat consciousness as three things instead of one thing. I'm going to treat neuroticism as three things instead of one thing." And I predict things about the world, I will keep improving my accuracy as I break it down further and further, right?

COLIN: Yes, that's what the evidence suggests. For most things, you get more accuracy as you go to more fine-grained personality descriptions.

SPENCER: So basically, we can think of the Big Five as giving you really good bang for the buck. All you need is five numbers about a person, and you can predict all this stuff. You have a lot of accuracy for only five numbers. But hey, go down as deep as you want. You can keep subdividing. You can go all the way down to the individual questions and use those to predict, and that would be the most accurate.

COLIN: Yes, usually what's been found that has been done empirically is that if you go all the way down to the individual questions, there are certain statistical considerations that come in when you're working with individual items because they're not as reliable measures. But yeah, you can get more accuracy the more variables that you have, basically. But I like to think about these kinds of trade-offs and where the most informative levels are. So I often work with a system that actually splits each one of the Big Five into two major sub-dimensions. There are empirical reasons to do that, and it seems to produce some of the most useful distinctions for discriminant validity within each one of the Big Five. So conscientiousness, for example, splits into an industriousness factor and an orderliness factor, and industriousness predicts pretty different things than orderliness does. Industriousness tends to predict job performance. Orderliness doesn't. If you are already predicting using industriousness as a predictor, sometimes orderliness even predicts negatively. I was just going to give another example, which is that industriousness is generally negatively associated with risk for mental health problems, whereas orderliness is actually positively associated with risk for things like compulsivity and rigid perfectionism, these certain kinds of problems.

SPENCER: I think this points out that while the Big Five gives a lot of bang for the buck, you get a lot of compression. You only need five numbers. You can miss really important signals if you use it. One example is, if you're looking at differences between men and women in personality, if you limit yourself to the Big Five and don't go to deeper sub-facets, you might miss some of the variation.

COLIN: Yeah, you would absolutely. We've published on this for some reason. I have theories about why this might be, but it has become one of my highest cited papers, and I think that's partly because it was an early open access publication, so it's easy to find. We showed that at the level of the Big Five, the gender differences that you see are basically really only for neuroticism and agreeableness. Women score higher than men in both agreeableness and neuroticism.

SPENCER: And are we talking just a little bit higher, or are these big differences on average?

COLIN: Yeah, no, it's not a very strong effect. The strongest effect size there corresponds to, I'm trying to remember, something like a third of a standard deviation.

SPENCER: So we could, if we imagine these as bell curves, right? The individual variability is high. The distributions are just shifted a little bit away from each other, but mostly overlapping, right?

COLIN: That's correct. In fact, I can quantify that because we did. It's about 70% overlap for the biggest difference that we found, which was for agreeableness, so that you still have 70% of the distribution overlapping. That means there are lots of women who are less agreeable than the average man and vice versa. But once you go down to the 10-factor level, below the Big Five, you actually see differences in every one of the five domains. The reason is that if you go to extroversion, it turns out that men score higher in assertiveness, but women score higher in enthusiasm, which is basically a combination of sociability and positive emotions. If you go to conscientiousness, women score higher in orderliness. If you control for orderliness, men score higher in industriousness. At the zero order, you basically see very little difference in industriousness, and you mainly see women scoring higher in orderliness. Within openness, men score higher in the intellectual component, intellectual engagement and intellectual confidence, but women score higher in the openness component, more interest in aesthetics and imagination and artistic creativity kinds of traits. So you really see, once you break it down, much more pervasive differences.

SPENCER: That's really interesting. Is there a way to quantify the multi-dimensional difference? Because you could say on each of these traits, the differences are always small between men and women. Is there a way to talk about how overall, when you take it all into account, how big a difference do you see between men and women?

COLIN: Yeah, there is. You can do a sort of multi-dimensional distance. After we published that paper, another researcher, Marco del Giudice, published a follow-up that basically reanalyzed our data to do precisely that, because we had just published all the univariate differences. They said, "Well, okay, how much of an overlap or not is there?" When you look at this in multivariate space, it drops from about 70% for the largest difference for the individual ones to only about, I think my memory is only about 30% overlap in the distributions for men and women at the multivariate level. I think that really helps to explain this kind of paradox, where people have this intuitive sense that men and women have quite different personalities, but then when we measure all of the individual traits, there's not much difference. There are average differences, but they're not large. If you think about it as a configuration of this whole collection of traits, rather than just these individual traits one at a time, you start to see why people get the impression that there's a bigger difference.

SPENCER: Interesting, interesting. So we still get meaningful overlap. But if we count up or measure all the different traits someone has, and you're putting each one on sort of the more feminine side or more masculine side, then you can start saying, "Oh, yeah, the score is a differential between men and women to a larger degree."

COLIN: Yeah, in multivariate space, there is less overlap, definitely.

SPENCER: Now, of course, as soon as we talk about the subject, people want to know, "Well, what's the cause of it? Is it inherent? Is it genetic? Is it changeable?" I think I just want to flag that the differences we measure don't mean that the differences are inherent or can't be changed, or desirable, or anything like that. Those are all kinds of things people add on top of it.

COLIN: Yeah, those are important caveats.

SPENCER: So we've talked about what's in the Big Five, but what's not in the Big Five? Are there parts of personality that you can only barely predict with the Big Five, or even not at all predict with the Big Five?

COLIN: Yeah. So there are certainly things that don't get categorized neatly into one of the Big Five. Some of those you can describe as a blend of multiple of the Big Five, like two or three of them.

SPENCER: I think that's sort of still in the realm of the Big Five, even though it doesn't line up right. But you could say, "Oh, you combine conscientiousness and agreeableness and extroversion, you still get it right." But what's really not in there?

COLIN: Yeah. Those are the interesting things that are harder to find aspects of religiosity. If you're thinking about participating in organized religion, that is correlated with agreeableness and conscientiousness, but probably not strong enough that you'd want to say, "Well, that's just a blend of those two things." Spirituality is associated with higher openness. But again, these things are a little bit outside of what you can easily capture, various things relating to sexuality, sexual motivation. There's some correlation.

SPENCER: We found that in our research. We were looking at the biggest differences between men and women, and the way they reported their personality. The single biggest difference we found was how they report their sexuality. When they meet an attractive person, do they think about sex? They think about sex all the time. We were like, "Oh, wait, is this in the Big Five? Not really." That's pretty interesting.

COLIN: Yeah, you're in a perfect position to come up with this list, probably better than I am with your data. What are the things that you can't easily map onto the Big Five? There are definitely things out there that don't map on. This is what I always try to explain to people. The Big Five are not supposed to be an exhaustive description of personality. What they are, are the major dimensions of co-variation among all the more specific personality traits. They're capturing something important, just because there are these major sorts of clusterings among personality variables. But there's plenty more to be interested in.

SPENCER: If anyone's curious about what Colin's referencing, we created a site called personalitymap.io where you can explore over a million human correlations, including searching for the Big Five and seeing what they're correlated with. There's lots of fun stuff to explore. I will put a link in the show notes for anyone interested in that.

COLIN: Spencer, that reminded me I wanted to ask you something about Personality Map. You said you've got over a million correlations, and a lot of those are individual items with other items. Are you familiar with the Eugene-Springfield community sample?

SPENCER: I've heard of it. It's on my list of data sets to look into. Do you think we should add it to our system?

COLIN: You could quickly add another 25 million or so correlations.

SPENCER: What? 25 million?

COLIN: Yeah, because this group started out with a thousand people in Eugene Springfield, Oregon, and they agreed to answer questionnaires for money by mail, and they did it for over 10 years. Many of the people in that sample have answered practically every personality questionnaire known to man, plus a bunch of other stuff. It's a really invaluable resource. If you ever have a question of whether something is correlated with something else, you can look in the Eugene Springfield community sample. Of course, it's just one sample of largely middle-class, middle-aged white people, but it's been really valuable for personality psychology, and all the data is publicly available from the Harvard Dataverse.

SPENCER: Oh, it's super cool. Definitely going to dig into that. Love it. So one of the things we do with Personality Map is we glue data sets together, and we could pull that in and then have it be accessible along with all the other data sets. So yeah, before we wrap up, I want to go back to mental health a little bit. I think it's so interesting how we are in this system where everything's sort of thought of in categories. You get a diagnosis, there's a medical code, insurance, billing, etc. As we talked about earlier, the data doesn't really point to this categorical model; the categories are something we're imposing on it. But you might also say, "Hey, look, we live in a practical world. At some point, an insurance company has to make a decision: do you have a disorder or not." Whereas with medicine, maybe we can detect the virus in your bloodstream; we're not there yet with the psychological stuff. We don't have the magical device that detects if you have a disorder. So, yeah, what do you think about that? Do you think that's sort of what we need to do living in the world we are, or is that just a fundamentally broken system?

COLIN: Well, you set me up perfectly for answering that question with this example of whether you have a virus or not, because that's basically the disease model of medicine, and that really is categorical; either the virus is there or it isn't. But there's a whole other area of medicine that isn't like that, and that's internal medicine. Where it's like your blood pressure. Nobody thinks that your high blood pressure is something that you have or you don't. We know that blood pressure is on a spectrum, and we know that once it starts to get past a certain level, it starts to become dangerous and have negative health consequences. The way we deal with that is we don't pretend that there's some fictitious category called hypertension; the measure is just a manifestation, an imperfect measure of it. We say, "No, blood pressure is fundamentally a dimensional phenomenon," but we're going to identify a point along that spectrum where we think that risk starts to increase enough that it's worth intervening.

SPENCER: We call it high blood pressure or something; we give it that name.

COLIN: Well, there's pre-hypertension, and then there's hypertension. There are these two points along the spectrum where medicine has decided if you get above this number, then we're going to probably offer you some kind of treatment, and then we'll give you a different treatment and more severe treatment if it gets above the second number. But no doctor thinks that there's something magic about the difference between 139 and 140 in your blood pressure. They know that that's just a semi-arbitrary cutoff based on various quantifications of risk that allows us to make that kind of treatment decision. And that's what we should be doing with psychopathology. The whole high-top idea is that instead of trying to put people into these imaginary boxes, we just say, "Okay, now you're elevated to this particular type of symptom. We might want to deal with that; maybe that's causing you certain kinds of problems. Maybe that's a risk for certain kinds of problems." Insurance can deal with that just fine; the same way that there's a code for hypertension, there could be a code for elevated internalizing problems.

SPENCER: So for a given patient, we break it down by the different subscales that they might be high in. Then we might organize our treatments that way too; instead of the treatment being targeted at depression, it would be like we've got treatments for internalizing symptoms. In this person, maybe they have both internalizing and detachment, and so then we would maybe give them multiple treatments for thinking of those as different variables. Is that the idea?

COLIN: That's exactly right. Yes, and internalizing is a good example here because there already exists a protocol for internalizing. It's called the Unified Protocol, and it's basically targeting neuroticism. It's a therapeutic treatment that targets neuroticism, which is the same as internalizing. The reason for that is that it's been discovered that there's relatively little specificity in treatments. The same treatments tend to work well for a bunch of different problems. It turns out that there are treatments that work well for most internalizing problems. This Unified Protocol was created to basically formalize that.

SPENCER: So if you look at evidence-based treatments, you'll find that cognitive behavioral therapy is going to be used for all internalizing disorders, pretty much.

COLIN: You get various others, too.

SPENCER: And other disorders as well, absolutely. But if you get really in the weeds, you do start to see differences. For example, if you look at the behavioral part of CBT and you look at depression versus anxiety, for depression, it's going to be more about getting people to do pleasurable activities, whereas for anxiety, it's going to be more about getting people to put themselves in exposure situations where they're exposed to the things they fear. So you do actually start seeing differences, and with that in your model, or in the HiTOP model, would you just have to look at narrower sub-facets of internalizing to sort of think about th.

COLIN: That's why this concept of hierarchy is so useful, right? And that's why it's the hierarchical taxonomy of psychopathology for HiTOP because there are going to be certain treatments that are likely to work for the full range of internalizing problems. But then as you go down lower in the hierarchy and you characterize people's problems with more fine grain, there are going to be other treatments that are really specific to one or another. And what we want is to get basically the optimal mix of treatments for people.

SPENCER: Do you see any drawback to switching to this continuous system from a categorical system or is it just all wins?

COLIN: People are always trying to come up with why this couldn't work and why it's a bad idea, but to me, it just seems like all wins. Because people often point to the insurance problem and that's not a problem. We know how to deal with that already. People say, "Well, clinicians won't be used to it, they'll find it cumbersome." Well, there's already been a lot of research that showed that if you ask clinicians to use the DSM system alongside one of these dimensional systems, they actually like the dimensional system better, and they say it corresponds more closely to what they actually see in their patients.

SPENCER: It does allow interesting variation. You could have a patient that has some symptoms of this disorder, but not all. And they have other symptoms of this other disorder, and there's no problem with that. In Hi-TOP, it's just, "Oh, okay."

COLIN: And this happens all the time. And this is one of the problems with the DSM disorders, as they're very heterogeneous. Just empirically, you can get two people with the same diagnosis who have a very different set of symptoms because, let's say there's a list of nine symptoms. So let's take borderline personality disorder. There's nine symptoms, and you have to have five of them in order to get the diagnosis. That means you could have two people who have the same diagnosis and only one symptom overlapping.

SPENCER: That's interesting. One of the things that blew my mind when we were looking through correlations is that if you take a standard scale for measuring depression, PHQ-9, and a standard scale for measuring anxiety, GAD-7, both widely used, the correlation between the scores is over 0.7, which in psychology, that's almost unheard of. Nothing correlates that strongly. It's almost like they're measuring the same thing, which is insane.

COLIN: Well, they're measuring neuroticism. There are many measures that are supposed to be measuring depression or anxiety that are really just more broadly measuring neuroticism, like the Spielberger Trait Anxiety Inventory, which people use all the time in research. You look at the items, it's not specific to anxiety. Even conceptually, we can just, as you were saying a minute ago, it's like, well, anxiety, you might do exposure; depression, you might do more activation kind of stuff. There are conceptual differences, but the way they measure it often just gets smeared out so that they're basically all just measuring general neuroticism.

SPENCER: And if you look at the scales, you will see there are no symptoms in common between the GAD-7 and the PHQ-9, I'm pretty sure. So you will look at them conceptually and see they are distinct, but there are some items on that, such as irritability, that could go either way; it's almost arbitrary whether you put that in depression or anxiety, because both disorders could lead to irritability. Okay, maybe one leads to slightly more irritability. But is that really justification to stick it in one of the two scales?

COLIN: And in a personality context, we would say, "No, irritability is just another distinct facet." Anxiety is one facet, irritability is one, anhedonia is one, because that's sort of the core of depression that's different from anxiety. It's that low positive affect. And so, yeah, it's much more useful to think about this kind of hierarchically and break things down into the specifics, but when they're really just tapping into the same broader thing, then you've just got neuroticism.

SPENCER: I'll tell you. The one thing that I do like about categories is because it seems to map on better to the way the human mind works. That seems useful sometimes. And what I've observed, and this is totally anecdotal, I don't have the data to back this up, is that people who are a bit depressed don't have very much in common. People who are a bit narcissistic don't have very much in common. But the most depressed people, the people who are in the 99.99 percentile, seem to have a lot in common. And the most narcissistic people, you know, 99.99% narcissism, seem to have a lot in common, to the point where it feels like it starts making more sense to use categories at the extremes. And again, this is just anecdotal, but if I try to justify this with arguments, I would say something like, "Look, if you're that depressed, you can barely function, you can barely get out of bed; everything you're doing is imbued with the fact that you're depressed." "If you're that narcissistic, then everything you're doing is coming down to its effect on your ego, and that's your primary driving force in everything you do," and suddenly people start looking more similar to each other.

COLIN: Yeah, no, I think that's a really astute observation, and it does have to do with motivation. These people end up being driven in this very particular way because they have this one particularly strong motivational complex. But that's just true in general. When we talk about extroversion versus introversion as a dimension, we're not stopping people from identifying themselves or other people as extroverts and introverts. We're just kind of recognizing that that means people who are fairly far out on those spectra. And what we're doing is helping the people in the middle who are like, "I don't know what I am." It's like, "Well, that's just because you're near the average, and it's not a category." So, there's nothing wrong in some sense with the MBTI kind of ideas that say, well, you're either high or low in this, but it would be a lot better to categorize it into maybe three or four categories instead.

SPENCER: So most people are just normal on it, or in the middle or average.

COLIN: Right, most people are near the average. And so I always ask people, "Well, how many of you have done the MBTI? Do you know your type? And how many of you have done it more than once? And if you did it more than once, did you end up with a different result?" People very often end up with different results from one time to another, and that's because they're pretty near the average on something, and they happen to score a little above the average once and a little below the second time. And so it categorized them differently. And that creates this illusion of change or inaccuracy, when, in fact, it's actually just trying to turn something into a binary that's not really well captured by a binary. Even just having three or four categories that you split the spectrum up into would make it a lot easier. So I've been working on some feedback. I have a personality questionnaire that has these 10 factors that break each of the Big Five down into two, and I have a web version of it. I have been working on developing some new feedback for it to give to people. I decided four categories is probably the right amount where you can tell people you're high in this, you're below average or above average, or you are high. You know, you're low, below average, above average, high. The problem with that, from a categorical perspective, is that instead of then two times five, which is still manageable, you've got four times five, and that's too many categories. You have to get people to be willing to think about the spectra at the same time that they're willing to say, "Well, you know, I'm a low, medium, high kind of thing."

SPENCER: Sometimes people ask me what I think the most basic rational skills are. If you could teach it to everyone you're trying to improve people's critical thinking, what would you do? One of the things I come back to is this, going from categories to spectra. So many human errors that we make, whether it's making mistakes thinking about mental health or racism, are category problems. We're bucketing things that shouldn't be bucketed, and then we attribute all these weird things to the bucket that aren't really justified, and then, things go terribly, basically.

COLIN: Absolutely. The other thing that made me think of was, what is a correlation? In some ways, it's just another example of this categorical thinking. When you say these two things are correlated, people hear, "Well, everybody who's X is also Y." It's like, "No, no, it's just people who are higher on X in general tend to be more likely to be Y." When you say this is correlated with this, X is correlated with Y, they hear all X are Y. I deal with this with students all the time.

SPENCER: Sort of them applying categories, taking your continuous measure and trying to turn it into categories against your will. Colin, thank you so much for coming on. It's been a truly fascinating discussion. Thanks for coming on Clearer Thinking.

COLIN: Yeah, thanks so much for having me. It's been a lot of fun.