Reinvestigating a study suggesting the presence of at least four personality types, scientists now claim they have an alternative interpretation of the results originally published in Nature Human Behaviour in 2018.
Applying conventional machine-learning techniques to a rather large data set, the study in question had found four meaningful clusters in personality dimensions. However, latest analysis reveals that a skewed distribution could have inaccurately led to the interpretation of suggestively meaningful clusters.
Two Major Views Concerning Distribution of Personality Types
Issues regarding the classification of personality into a single or various discrete types have been of significant concern. Such information could potentially lead to an enhanced understanding of human individuality and mental health issues.
It has been established that human personality is categorized by five dimensions – also referred to as traits or factors – namely extraversion, neuroticism, agreeableness, openness, and conscientiousness. However, an understanding of how these personalities are distributed in the five dimensional space (5D) remains unexplained. The two views that currently exist are the ‘dimensional’ view and the ‘categorical’ view.
The dimensional view suggests that the distribution of personalities is ‘unimodal’, and personalities are continuously distributed in 5D space. In contrast, the categorical view supposes that the distribution is ‘mulitmodal’, and each individual can be classified into the many different types/clusters/dense regions existing in 5D space.
Whereas the commonly used statistical tools of personality are based on the dimensional view, some scientists have suggested the categorical view, claiming the presence of personality types.
Findings of the Original Study
Gerlach et al developed an alternative approach to identifying personality types, and applied their method to four large data sets consisting of above 1.5 million participants. Refining previously suggested personality typologies, they claimed to have found substantial evidence for the existence of at least four distinctive personality types. Moreover, they demonstrated that the personality types existed as a small subset of a much larger set of unauthentic solutions in conventional clustering approaches. They also highlighted the major limitations in applying unsupervised machine learning techniques for analyzing large data sets.
New Finding Suggest New Possibilities
Revaluating the main part of the study by Gerlach et al, the analysis is in accordance to Gaussian mixture model, or GMM which is a probabilistic model for representing normally distributed subpopulations within an overall population). GMM denotes a given distribution by weighted sum of a finite number of normal (Gaussian) distributions and suggests five factor scores that provide positions of individuals in five dimensional space.
If cluster structures really do exist and each cluster is represented by a single normal (Gaussian) distribution, then each normal component may be linked to a single cluster. To evaluate whether each Gaussian component was a potentially meaningful cluster, Gerlach et al performed an analytical test based on the ‘null model’, a model generated with random samples of a specific distribution, assuming that the five factors are distributed independently. They found four meaningful clusters, concluding that four Gaussian components existed, hence four personality types exist in 5-D.
However, the occurrence of meaningful clusters can also be obtained from a target distribution that has skewness (distortion or asymmetry in a normal distribution). To confirm this, researchers applied the same procedure to artificially generated two dimensional (2D) data from a skewed, unimodal distribution. The best fitted GMM obtained consisted of seven components, three of which were categorized as ‘meaningful clusters’.
Based on these confounding findings, scientists have suggested that the result of the study by Gerlach et al does not necessarily reveal cluster structures and could instead be indicative of skewness of the distribution. The latter could possibly influence personality types and despite the pivotal work done by Gerlach and his team, the understating of whether personalities are distributed as categorical, dimensional or possibly intermediate in the five dimensional space still remains unconfirmed.