Typing—One More Time

In an earlier post (“Typing Is So Twentieth Century”), I addressed the practice of using

Can Do–Four Personalities. Roy Blumenthal. 2008. CC BY-SA 2.0

combinations of traits to create personality types. Type theorists propose that everyone falls into one type out of a set of two or more—as in you’re either a morning person or a night person. The Myers Briggs Type Indicator famously has 16 such types. That model has been sufficiently discredited. In my earlier post, I pointed out that the 16 MBTI types leave out 80% of the population! What do the rest of us do? Wander the dessert without an identify? Of course not! Eschew such “typing.”

But, alas, researchers persist. In recent articles in Scientific American and The Washington Post, among others, journalists have reported on new research that yielded four water-tight personality types based on an impressive sample of 1,500,000+ subjects. These articles for the general public were based on the scholarly article published by a research team from Northwestern University: Gerlach, M., Farb, B. Revelle, W., & Amaral, L. A. N. (2018). A robust data-driven approach identifies four personality types across four large data sets. Nature Human Behavior. https://doi.org/10.1038/s41562-018-0419-z .

Their research proposes four types. I give the name of their types in the first column, then the levels of the Big Five associated with each type, and then a translation of the trait level symbols into adjectives:

Name of type

Associated Big Five traits

Reserved N-E-O-A+C+ resilient, solitary, traditional, adaptive, disciplined
Role Model N-E+O=/+A+C+ resilient, outgoing, moderate/progressive, adaptive, disciplined
Average N=E=O=A=C= situationally resilient/reactive, ambiverted, moderate, negotiator, situationally disciplined/casual
Self-centered N=E+O-A-C- situationally resilient/reactive, outgoing, traditional, aggressive, spontaneous

The researchers claim age and gender effects—that the Role Model is more common among females over 40, and that it increases with age; that males are underrepresented in the Average type; and that the Self-centered type is more common in teen boys, decreasing with age. However, because of the overrepresentation of females and 20-somethings in the sample (see below, Crtique #4), I am doubtful of these findings.

The study used four web-based data sets of 100,000-500,000 each. Each sample was factor analyzed with the hope of similar results:

  1. IPIP; 145,388 respondents to 300 items (a version of the NEO-PI-R)
  2. Johnson-120; n=410,376
  3. myPersonality-100; n=575,380
  4. BBC-44; n=386,375

All four studies converged on the same solution (except O is borderline mid to high for Role Model)—an impressive, even convincing, achievement. However, my critique that follows suggests cautions for interpreting and using their results.

My Critique:

  1. Applicability. While the clusters are real, they cover a small portion of the population. It is as though we used four regions—say, Scandinavia, Southeast Asia, the Middle East, and New England—to cover the entire globe. In addition, the number in each sample who fit the cut points are not included, so we do not know if the four types represent 90-100% of the total, or only a smaller percent. I went to my norm group of July 16, 2008, a group of 7,459 U.S. adults who had taken our WorkPlace Big Five Profile in English as a part of programs conducted by members of our Big Five consultants’ network. I counted the number of individuals who exhibit all five trait levels for the four types. For example, I counted the number of “Reserved” by looking for the number of individuals who scored N<45 AND E>45<55 AND O<45 AND A>45<55 AND C>45<55. The results:
    1. Reserved = 22
    2. Role Model = 48
    3. Average = 0
    4. Self-centered = 2

That’s 72 individuals out of 7,459, or .9%–just a hair under 1%–who exhibit the four types found by this research[1]. What about the rest of us 99%. I question the utility of such a model!

  1. Comparison to MBTI. In my earlier blog, the 16 MBTI types comprised closer to 20% of the sample. Why was it larger? Because a) the MBTI types used only four elements (E/I, S/N, T/F, and J/P) for each of the 16 formulas, and b) the MBTI used 16 types while this study used only four. If we were to adjust for these differences, the percentage of the sample accounted for by the types in each model would be close.
  2. Unclear cut points. They do not mention how they set cut points in determining whether to assign someone to a type or not. I suspect they use the midpoint, whereas we recommend using .5 SD above/below the mean.
  3. Biased sample. They report gender and age frequencies as fractions and age brackets, not real numbers. That said, it appears that 18-25 year olds are vastly overrepresented, with those over 40 disappointingly low, such that age inferences are unreliable. Females appear to outnumber males by about two-to-one. This could have been remedied by balancing their samples. I would not trust either their gender- or age-effect conclusions. I wonder how the factor analysis might have been different with a sample balanced by age and gender. Females and 20-somethings are determining the outcome, and we know that significant changes occur during that first decade of adulthood. At Paradigm, we balance our norm groups based on the current U.S. Census.
  4. No theory. While the Big Five are not based on a theory of personality, but rather simply describe the structure of personality, type theories are generally based on a theory. The MBTI is based on Jungian theory. The ARC-type model is based on Karen Horney’s theory or moving against/toward/away from others. This set of four types has no such theoretical underpinning. They put far more emphasis on their statistical procedures than on explaining their results. Perhaps this should be expected from a research team comprised of three engineering “types” and only one psychologist. They do compare their findings to those of the ARC-type model[2], which comprises three types: Resilient (N=E=O=A=C+), Overcontrolled (N=E-O-A=C=), Undercontrolled (N=E=O=A-C–). While there is an affinity of their four types to the three ARC-types (Role Model is similar to Resilient, Reserved is similar to Overcontrolled, and Self-centered is similar to Undercontrolled), they do not explain the departures, nor where future research should lead/explore.
  5. Linear bias. Personality researchers are finding that many traits have curvilinear relationships to other variables. For example, the A/Accommodation/Agreeableness trait has a curvilinear relationship to leadership effectiveness, which is a similar construct to this study’s “Role Model” construct. That means that while lower scores on A are associated with effective leadership, if they get too low on A, the effectiveness begins to decrease rather than increase linearly. However, the Northwestern model does not allow for such non-linear connections.

Role models are good, depending on what kind of role you wish to model. If you want to model leaders, then perhaps this study’s model is noteworthy. However, if you want to model research scientists or long distance runners, you might look to other role models. Also, knowing that “self-centered” occurs with such infrequency—only 2 out of 7,459—is gratifying! In short, we need many more “types” if we want to address the population at large. I say abandon such typing—the Big Five already covers everyone! Learn how an individual’s levels of the various traits contribute to their uniqueness and suitability for specific kinds of work, play, and relationships.

[1] For those curious about how this count compares to what we would expect to find, we would expect 1/3 of the sample to score below 45 on N, E, and O, and 1/3 to score above 55 on A and C , to get the expected frequency for Reserved. That would be 1/3 x 1/3 x 1/3 x 1/3 x 1/3, or about .4%–about half a percent. Using the same formula for the other three “types,” we would expect them to cover 4 x .4, or 1.6%, of the population. In the analysis of our norm group above, we found .9%. That’s a rather close result for actual versus expected.
[2] See Connellan, M. B., & Robins, R. W. (2010). Resilient, overcontrolled, and undercontrolled personality types: Issues and controversies. Social and Personality Psychology Compass. https://doi.org/10.1111/j.1751-9004.2010.00313.x .

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

Blog at WordPress.com.

Up ↑

%d bloggers like this: