How the other half of differential psychology lives, Part 1

Share on FacebookShare on Google+Email this to someoneTweet about this on Twitter

When we talk about individual differences in psychology here at GNXP, they are almost always differences in some form of intelligence — the real-world consequences of these differences, what inter-group variation there is in the trait, the lower-level biological correlates of such differences, and so on. This is all well and good, but we shouldn’t forget the other half of differential psychology: personality. Unfortunately, more of the history of personality theory has been plagued by cultish trends and airy-fairyness compared to the history of intelligence investigation — probably the first names that come to your mind are Freud, Jung, and so on; whereas the first names that come to mind for intelligence researchers would be bona fide scientists like Spearman, Jensen, and so on. I’ve put together a little whirlwind tour of personality theory because of its inherent importance to those interested in: 1) what dimensions we differ on from one person to another; 2) what the lower-level biological correlates are of such differences, including what alleles of which genes are involved; 3) what inter-group variation there is in such traits; and 4) how personality traits relate to the study of human evolution, especially by the force of selection. Now, each of these topics could easily fill a journal issue or two, so I emphasize that this will be a whirwind tour. This first Part looks at the first two questions; the second will look at the latter two. For those interested in greater exposition and references, the best textbook on the topic is Personality Traits by Matthews, Deary, & Whiteman, which covers the first two topics I’m addressing, though not the second two topics.

Let’s begin with the psychometric basics. Unlike intelligence, where there emerges a single general factor (g) among factor analyses of performance on intelligence tests, there is no such thing in the domain of personality. The two personality questionairres most researched, validated, and used by researchers are the Eysenck Personality Questionnaire (EPQ) and that developed by the Big Five researchers, the NEO-PI. Factor analysis is performed on the responses to reveal what higher-level dimensions of personality exist. Though the EPQ nominally measures three traits, while the NEO-PI measures five, they actually measure quite similar traits. The two factors Extraversion and Neuroticism are common to both questionairres, and arguably go back to Galen’s typology of the four temperaments. The third trait measured by the EPQ, Psychoticism, appears to be a higher-level combination of the two orthogonal NEO-PI factors Agreeableness and Conscientiousness — that is, someone who scores high on P would score low on both A and C. Eysenck’s P factor shows a skewed distribution in the population, whereas the others show more normal distributions, suggesting that it is the A x C interaction that is measured by P. Big Five researchers also measure Openness to Experience, but this is the most open to question of the five — it is non-trivially correlated with intelligence and may or may not appear in other cultures — so I’ll mostly look at the other four on the NEO-PI, which allows me to talk separately about A and C instead of just P. (Another popular questionairre, Cattell’s 16PF, shows significant intercorrelations among some of the factors, and grouping them into higher-level factors leads to something like the Big Five factor structure.)

Aside from the vague connotations the names carry, what do they actually measure? See here for the description of what the NEO-PI measures at the factor and facet levels. The one caveat in trying to keep straight what these measure is that Neuroticism measures one’s emotional stability vs instability — it does not necessarily reflect “Woody Allen” neuroticism. Just as P is the likely A x C interaction, so this Woody Allenesque neuroticism is the result of low-E x high-N interaction (“melancholics” in Galen’s typology). Also bear in mind that one’s experience of positive vs negative emotion are the result of the two orthogonal factors E and N — higher E goes along with greater frequency of positive emotions, while higher N goes along with greater frequency of negative emotions. So, it is possible to experience a heightened frequency of both positive and negative emotions if one is both highly extraverted and highly emotionally unstable (“cholerics” in Galen’s typology). Or put another way, scoring low on N doesn’t make one happy but simply calm, and likewise for scoring low on E. Such a greater tendency toward equanimity is what characterizes Galen’s “phlegmatics.” The common assumption that higher E protects one from negative emotions is confusing higher E with the high-E x low-N interaction (“sanguines” in Galen’s typology).

I’ve mentioned Galen’s typology of the four temperaments because it was the first model that sought to combat two typical objections to psychometric measurement: 1) that the tests measure what the tests measure, i.e., have little real-world relevance; and 2) that the traits measured are either fleeting non-traits or that they’re mere social constructs (though I doubt the latter was bandied about much in his time!). Now, Galen was completely wrong about what the underlying biological causes of personality differences were (who even knows what the “black bile” is which he thought melancholics had an excess of?), but then most medical investigation until roughly the 19th C was a mishmash of unfounded conjecture and quasi-religious superstition. He was correct, however, to look for enduring biological differences that could account for the stable trait differences in personality. Let’s look at the answers to these two objections in more detail.

First, though we’re most familiar with the real-world consequences of intelligence differences — most on display in outcomes of academic and job performance — personality differences also predict real-world outcomes. Extraversion predicts how socially engaging one is in social settings (for example, how long one talks to others). Neuroticism predicts susceptibility to anxiety or depression, as well as marital satisfaction. Agreeableness also predicts marital satisfaction (not surprising, as this measures how trusting, caring, and cooperative one is). Conscientiousness predicts job performance, especially the non-intellectual aspects of it — punctuality, for example. As mentioned before, Openness seems to measure aspects of intelligence, so it’s a predictor of intelligence differences.

As I assume most reading this are interested in what the typical personality profile looks like for top scientists, either to alter their personality in the desired direction (not easy, but not impossible, especially by taking mind-altering substances like caffeine), or to evaluate how promising an applicant for a science position is. The major study on point was Cattell’s 1965 survey of living eminent scientists, using his 16PF questionairre. They showed clear deviation from average toward the following poles: unsociable, intelligent, emotionally stable, dominant, brooding (or serious, introspective), undependable (not rule-following), bold (or venturesome), sensitive (as in Openness to Feelings or Aesthetics — not necessarily as in tender & caring), trusting, guilt rejecting (or self-assured), radical (not conventional: like high-O), self-sufficient (not group-oriented), and self-disciplined. They scored average on the traits imaginative vs practical, forthrightness vs diplomatic, and tense vs relaxed. These were “eminent researchers,” so there is undoubtedly variation — presumably at the lower level of “lab monkey,” dependability would matter much more, while at the more pioneering level it may be detrimental (the absent-minded professor too engrossed in his work to care about protocol and decorum). I looked at one real-world consequence of differences in such traits in my post on sex differences in scientific eminence.

Turning now to the search for lower-level biological causes of these traits, let’s start with a few simpler observations. The first is that these traits are that — stable traits. For some odd reason, a camp of social psychologists wanted to wish personality traits away by noting that scores on such traits did not predict all that well an individual’s behavior in a particular situation. The situation was therefore thought to mold behavior. What these psychologists failed to pick up in their intro statistics class is that one data point will give you a garbage estimate of how well the independent variable predicts the dependent variable. It turns out that when you examine a person’s behavior over a wide range of situations numerous times, their personality traits do predict what you’d expect — e.g., that the introverts tend to be less socially engaging, although they might be more sociable among their close relatives over Christmas dinner. Clearly traits are situation-sensitive — a disagreeable person is not constantly on the warpath, and neurotics are not forever freaking out — but they still represent biases in how the individual will behave. For example, two individuals get an F on a test — to a low-N person, this might be a cause for re-evaluating what objective measures can be taken to get a better grade on the next test; while a high-N person might overreact and turn their frustration inward, focusing on their negative emotions. Same situation, different “coping styles.”

Not only are these traits stable across situations, they are pretty stable across the lifespan, at least after about age 30, both on a population aggregate level (50 y.o.s are not pronouncedly different in aggregate personality measures than 30 y.o.s) and throughout an individual’s course. Even the decade from age 20-29 seems not to involve much change. Real personality changes occur mostly during adolescence, with maybe / maybe not some additional tweaks during the third decade of life, and little change after 30. Such persistence suggests a genetic component, and indeed twin and adoption studies give heritability estimates about equal to what you’d find for intelligence — roughly, narrow-sense ~ 0.35 and broad-sense ~ 0.5, with the environmental variance representing non-shared as opposed to shared factors. Interestingly, a recent study found a positive correlation between prevalence of the pathogen Toxoplasma Gondii and aggregate level of Neuroticism at a national level (more on national differences in Part 2). Whether or not one becomes infected, or whether or not an infected person sees an increase in Neuroticism, is largely a stochastic affair — presumably this sort of developmental noise accounts for most of the non-shared environmental variance.

As for the supra-gene / sub-psychometric level, the literature is pretty ambiguous. This may not be as bad as it may seem — whereas with intelligence, we’re talking about raw horsepower or ability, with personality traits we’re talking about context-sensitive behavioral strategies. We know that genes and non-genetic biological causes (like germs) serve to build something which results in stable biases across situations (but sensitive to them), but for all we know, this may be instantiated in an If/Then program that’s impossible to pin down on a given number of gross properties of the brain, like brain size, glucose metabolization rate, and so on, for intelligence. So, unlike intelligence research which focuses on combinations of these lower-level gross properties, the personality research seems to be moving more toward the cognitive level, while still acknowledging that this cognitive architecture must have a neural substrate, but not necessarily an easily observable property like volume of white matter in some brain region.

To return to the example of differing level of Neuroticism influencing response to a failure on a test, we might imagine a simple toy model of Neuroticism where environmental inputs are analyzed and coded, and these analyses then pass into a mechanism for determining reponse (remain calm, freak out, etc.). Individual differences in Neuroticism could be re-stated as differences in how high or low of a threshold the box has for outputting the “freak-out” response — let’s say stimuli are coded from 1 to 9 in increasing order of threatening potential. Stimuli coded as 9 would manage to freak out anyone — say, a rabid animal charging toward you — while stimuli coded as 1 would only manage to freak out Woody Allen — say, you stub your toe first thing in the morning. Thus, a high-N individual would have a low threshold (anything 1-9) for responding in a freak-out way, while a low-N individual would only lose their cool if they’d processed a level 9 stimulus. This is a cognitive model . Other cognitive models have the coding mechanism do most of the work in explaining differences in N — high-N people are biased to code stimuli as threatening and lock attention onto them, which low-N people would regard as non-threatening. Eysenck’s sub-cognitive view was that higher-N people had more easily excitable sympathetic nervous systems — the involuntary fear/fight/flight/sex system.

Having said this, there is extensive (though again, muddy) data on the biological correlates of E in particular. The original idea (well, after the four humors, that is) is due to Eysenck, who thought that extraverts had lower basal levels of cortical arousal, while introverts had higher levels. Appealing to the Yerkes-Dodson Law, he hypothesized that there was an optimal level of cortical arousal for performance in life’s tasks. Thus, those low in arousal (extraverts) would seek to artificially boost their arousal in order to function more comfortably — namely, by engaging in social stimulation, living a fast-paced life, and so on. Introverts, by contrast, would tend to avoid such stimulation, since their arousal levels are thought to be high enough already, at or above the optimal level. One prediction is that extraverts should tend to respond in a less extreme manner than introverts to increased levels of stimulation. Extraverts are better at performing tasks when there is background noise, they are less startled by abrupt loud noises, and they salivate less profusely when lemon juice is squeezed into their mouths. However, some of these results — especially measures of cortical arousal such as EEG measures — are not always replicated, or, worse, show the opposite of the predicted pattern. As with research into the biological causes of intelligence differences, some of these meager results are likely due in part to the non-representative samples used in studies — namely, squeeze 100 college students from Psych101 into the lab over the weekend, and get your results instantly.

In Part 2, I’ll look more at genes, since that will provide a nice transition to discussion of inter-group differences, as well as the significance of personality research to human evolutionary biology.

6 Comments

  1. Just one note: I realize the post isn’t copiously referenced — most of them come from the refs in the Matthews, Deary, & Whiteman book, or simple Pub-Medding. I’d go to the relevant section of Personality Traits, which is searchable at Amazon (and cheap in paperback), or Pubmed “cortical arousal extraversion” for example. I wanted more links, but this has been a back-burner post for awhile, and I wanted to get it out sooner rather than never.

  2. Reading Judith Rich Harris’ “No Two Alike” got me to wondering how reproducible are personality measurements. I recall IQ test scores correlate (test-retest) at about 80%. Using google scholar I dug up some articles on correlation (test-retest) for personality tests. I was surprised to see the dominant eigenvector (introversion) reliable at the 80% level, with other eigenvectors diminishing towards about 50% for the fifth eigenvector. All much better than I (an engineer) would have guessed.

  3. Agnostic, good post. I look forward to part 2.

  4. Jon — not only do short-term test-retest reliability coefficients range from 0.7 to 0.9 (excellent for psychometric measurement), but if you test someone now and 10 years from now, they won’t have changed much. The longer-term correlation will be somewhat lower, but when you take the disattenuated correlation, they go upwards of 0.9, indicating that the weakened correlation between now and 10 years hence is mostly an error of measurement, not a real change in one’s personality traits. (Data & refs in Personality in Adulthood by McCrae & Costa; search index for “retest reliability.”) 
     
    Such little change over time is true for crystallized and verbal IQ tests, but not for the more fluid and visuospatial ones like Ravens, the scores on which tend to deteriorate over time, esp. after ~50 (see Looking Down on Human Intelligence by Deary, Ch. 8). So in a sense, personality traits are more enduring aspects of a person’s being than reasoning abilities. Not too surprising, as intelligence is more of a skill like athleticism while personality is like a set of strategies. Natural selection may not have cared much if 60-y.o.s could perform pattern recognition like they could in their 20s, but anyone will still need to navigate social and personal life.

  5. More praise, and encouragement for part2. Nice work distilling the unifying skeleton of such disparates.

  6. What’s your opinion on the Myers-Briggs system of personality typing? I know the Big-5 test is more fashionable among psychologists, but I think the Myers-Briggs test works pretty well for a four-factor model. Also, the Myers-Briggs system seems to emphasize that there are inherent trade-offs in having one personality type over another, whereas the Big-5 system makes more of a value judgment (obviously openness, conscientiousness, and agreeableness are better than their alternatives).

a