# Commentary: A Robust Data-Driven Approach Identifies Four Personality Types Across Four Large Data Sets

**Katahira et al (2020) · Skewness can mimic personality types**

*Katahira et al · Methodological · Measurement & Method, Critique & Boundary · Frontiers in Big Data · 2020 · Open access*

Katahira, Kunisato, Yamashita and Suzuki offer an alternative reading of Gerlach, Farb, Revelle and Amaral (2018), who fitted Gaussian mixture models to five personality factor scores from more than 1.5 million respondents and found four 'meaningful clusters', suggesting at least four personality types. A skewed distribution with no cluster structure, they argue, 'can erroneously lead to the seemingly meaningful clusters'. At stake is whether personality is dimensional, continuously distributed, or categorical, with dense regions into which individuals can be classified.

A mixture model can fit a non-Gaussian distribution by combining several Gaussian components, and Gerlach and colleagues tested each against a null model in which the factors are independent. Applied to simulated two-dimensional data from a unimodal, skewed distribution, the procedure gave seven components, three deemed meaningful clusters, into which only 45.5% of the samples were classified. Skew alone is not enough: dependence among the factors is also necessary, since the null model has the same marginals as the original distribution.

Gerlach and colleagues' factor scores were found to be skewed, Katahira and colleagues report, and there appears to some extent to be dependence between factors, so they suggest the results 'do not necessarily reflect the cluster structures and instead could reflect skewness of the distribution'. The mechanism, they argue, offers a natural explanation for Freudenstein and colleagues' (2019) finding that less than half (42%) of respondents in one data set were classified into the four clusters.

They demonstrated, they write, the possibility that skew can influence the reported types, without formally evaluating how far the skew affected Gerlach and colleagues' results, which may require novel statistical methods. We note that the simulation used two dimensions, not five. Whether personality is categorical, dimensional or in between remains, they conclude, an open question.

Psychological safety is not discussed, and the bridge to it is thin. We read the commentary beside Tom's article on personality profiling, which finds little evidential basis for Myers-Briggs, DISC and similar tools; the commentary addresses neither the tools nor teams, only whether personality comes in types at all.

- **Citation:** Katahira, K., Kunisato, Y., Yamashita, Y. and Suzuki, S. (2020) 'Commentary: a robust data-driven approach identifies four personality types across four large data sets', Frontiers in Big Data, 3, Article 8.
- **This page:** https://explore.psychsafety.com/n/katahira-et-al-2020/
- **View the source paper:** https://doi.org/10.3389/fdata.2020.00008
- **Interactive map:** https://explore.psychsafety.com/?mode=papers&node=katahira-et-al-2020

## Neighbours (2)

- [Critique of Personality Profiling](https://explore.psychsafety.com/n/the-fallacy-of-applying-complicated-models-to-complex-problems-aka-why-personality-profiling-is-bs.md)
- [Two Lines: A Valid Alternative to the Invalid Testing of U-Shaped Relationships With Quadratic Regressions](https://explore.psychsafety.com/n/simonsohn-2018.md) (paper)
