The Question

Digital traces, posts, likes, messages, usage patterns, and browsing behaviour, contain information about who people are. Researchers have explored both whether human observers can infer personality from digital data, and whether machine learning algorithms can do so more accurately. Hinds and Joinson (2024) conducted a two-part meta-analysis synthesising both literatures simultaneously.

What They Found

In the first part, five meta-analyses of human perception studies across 24,124 individuals rated in 30 independent samples showed moderate convergent validity between digital-data-based personality inferences and self-reported personality across all Big Five traits, ranging from ρ = .38 for Neuroticism to ρ = .57 for Openness (Hinds & Joinson, 2024). Humans can read personality from digital data with meaningful but imperfect accuracy.

In the second part, a multilevel meta-analysis of 42 computer prediction studies reporting 534 effect sizes found moderate convergent validity for machine learning predictions at ρ = .30, somewhat lower than human perception on average (Hinds & Joinson, 2024). This is perhaps a surprising finding given the enthusiasm for AI-based personality inference; human observers drawing on digital data appear to perform at least as well as algorithms on average.

Moderator analyses found that certain platforms and data types, including X, Facebook, Sina Weibo, videos, and smartphones, had a negative impact on prediction accuracy. Openness was the most identifiable trait across both human and algorithmic approaches; Neuroticism was the least (Hinds & Joinson, 2024).

Why It Matters

As digital personality assessment moves into applied contexts, including hiring and organisational development, the evidence base for its accuracy matters enormously. This meta-analysis provides the most comprehensive synthesis available, and its findings suggest that both human and algorithmic inferences from digital data carry real signal while remaining substantially imperfect (Hinds & Joinson, 2024). The gap between commercial enthusiasm for these tools and the accuracy levels the evidence supports remains a concern.

Reference

Hinds, J., & Joinson, A. N. (2024). Digital data and personality: A systematic review and meta-analysis of human perception and computer prediction. Psychological Bulletin. Advance online publication. https://doi.org/10.1037/bul0000430