The European Journal of Personality (EJP) invites articles that estimate how strongly individual differences in personality traits can be predicted from combinations of multiple brain variables. Let’s call this “predict-personality-from-the-brain” initiative.
How well can we predict personality traits from the brain?
The associations of brain variables, representing either structural or functional brain variations, and personality traits have often been investigated by linking a) individual brain variables with traits b) within single samples. Documenting such associations can be, and has been, highly valuable for understanding the neural basis of individual differences.
A complementary research strategy would be to explore the overall degrees to which personality traits can be predicted by combining many (tens, hundreds, and maybe ultimately even thousands of ) brain variables into a single prediction model, training and validating it in independent samples or sample partitions to avoid over-fitting.
In short, this means combining personality neuroscience, a research topic with decades-long history, with increasingly popular predictive modelling.
For a parallel, training prediction models (“polygenic scores”) almost atheoretically based on thousands of genetic variants and estimating their predictive accuracy in independent samples has become a routine practice in quantitative genetics, including behaviour genetics. This polygenic modelling has resulted in key insights into how genetic variants are (not) linked with phenotypic variables, including personality traits. The most widely accepted conclusions are that behavioural traits are highly polygenic (i.e., influenced by very numerous genetic variants with each playing only a tiny role) and pleiotropic (i.e., same variants play roles in many traits). Explanations of genetic variance in traits now must obey the constraints that these conclusions impose. Also, polygenic scores are routinely used for testing causal hypotheses, including those that involve links among multiple phenotypes. Some have argued that the awakening to polygenicity in genetics also provides a model for psychological research (Götz, Gozling, & Rentfrow, 2021).
Similarly, it is possible that researchers will be able to create "polybrainvariable scores", which aggregate signals from many-many brain variables for a) the best possible prediction of personality traits in independent participants, b) without necessarily worrying about the theoretical meaning or statistical significance of individual brain variable-personality trait associations. For example, such research is already happening for cognitive abilities (for a brief review see Deary, Cox, and Hill, 2021).
Again, such a research approach based on aggregating many predictors to explore the limits of predictive accuracy would not negate the more common practice of linking individual variables with the traits of interest; instead, the approaches would be complimentary.
How predictable are personality traits from brain signals? It seems possible that brain signals allow for stronger predictions of trait variance than the genomic variations used in polygenic modelling. How large should sample sizes be for reliable prediction models? What kinds of variables (e.g., structural or functional, from MRI or EEG) will allow for more accurate predictions? Which theoretical constraints can such findings impose? Which new research questions can be explored? For example, a prediction-focused approach could be used to explore how different traits overlap in their brain-correlates, evidenced by brain-models that were trained for one trait also being able to predict other traits.
Theoretically, it is possible that the overall degree to which personality traits can be reliably (out-of-sample) predicted from a combination of many-many brain variables hints at the upper limits of explanatory models based on such variables. It could be very useful to know what the limits are, whatever they end up being. We can hope that they will be reasonably high; maybe up to 20-30% of trait variance can be predicted from certain combinations of brain variables? Maybe even more? If the percentages end up being lower than expected, however, this can also reduce the pressure on personality neuroscience studies to produce unrealistically large effects for individual brain variable-trait associations. But different kinds of brain variables (e.g., functional vs. structural) may provide very different upper limits, potentially suggesting research avenues that may be particularly worthwhile of being pursued, at least when larger effects are desired.
Methodologically, it is plausible that striving for the most accurate possible predictions (whatever the upper limit ends up being) will also result in important modelling advances. This surely has happened in genetics.
How to proceed? An invitation for proposals.
This invitation is broad and EJP is open for very different kinds of proposals, as long as they aim at estimating the predictability of personality traits from the brain in a methodologically rigorous way.
· There is no fixed timeline.
· We assume that in the proposed research, prediction models are trained and validated in independent samples or sample partitions, for otherwise there is an increased risk of over-fitting (Yarkoni & Westfall, 2017). This means that reasonably large samples must be used.
· To achieve the aim of estimating the degree of variation in personality traits that can be reliably predicted from brain variables, we encourage collaborations among research groups so that the work can benefit from the largest possible data sets. We know from genetics that prediction accuracy tends to increase with sample size and the same may apply to predicting traits from brain variables. Different data sets can be pooled for model training and/or be used to separate training and validation of the models.
· Proposals can be based on both new and existing data sets, including widely available ones such as the UK Biobank or the Human Connectome Project, or on any combination of them.
· We encourage pre-registering the proposed research, although we do not assume this; Registered Reports are a viable form for these proposals.
· Null findings (little or no predictive accuracy) are as welcome as any other findings.
Of course, regular papers that rigorously estimate individual brain variable-trait associations to test specific theoretical hypotheses are also welcome, as they have always been, as are proposals for special issues and target articles on personality neuroscience. However, these papers and proposals do not fall within the scope of the “predict-personality-from-the-brain” initiative described in this post and should be submitted via our regular submission channels.
Please do not hesitate to get in touch with any ideas, questions or proposals (rene.mottus@ed.ac.uk).
René Mõttus
References
Deary, I. J., Cox, S. R., & Hill, W. D. (2021). Genetic variation, brain, and intelligence differences. Molecular Psychiatry, 1–19. https://doi.org/10.1038/s41380-021-01027-y
Götz, F. M., Gosling, S. D., & Rentfrow, J. (2021, January 10). Small effects: The indispensable foundation for a cumulative psychological science. https://doi.org/10.31234/osf.io/hzrxf
Yarkoni, T., & Westfall, J. (2017). Choosing Prediction Over Explanation in Psychology: Lessons From Machine Learning. Perspectives on Psychological Science, 12(6), 1100–1122. https://doi.org/10.1177/1745691617693393