Going beyond simplicity: Using machine learning to predict belief in conspiracy theories

Abstract

Public and scientific interest in why people believe in conspiracy theories (CT) surged in the past years. To come up with a theoretical explanation, researchers investigated relationships of CT belief with psychological factors such as political attitudes, emotions, or personality. However, recent studies have put the robustness of these relationships into question. In the present study, a prediction-based analysis approach and machine learning models are deployed to detect and remedy poor replicability of CT belief associations. The analysis of a representative dataset with 2025 UK citizens supports the assumption that the current simplicity of the field's analysis routine, exhibiting high sample-specificity and neglecting complex associations of psychological factors with CT belief, may obscure important relationships. The results further point towards key components of conspiratorial mindsets like general distrust and low sociopolitical control. Important implications for building a coherent theory of CT belief are derived.

Identifying important individual‐ and country‐level predictors of conspiracy theorizing: A machine learning analysis

“…The first four of these predictors were more powerful than any country‐level predictors. For the most part, these individual‐level predictors are in line with prior theory and research, including the results of a similar machine‐learning analysis by Brandenstein (2022). Research has found that conspiracy theorizing is be associated with a negative view of elites, disillusionment with society, paranoia and frustrated psychological needs (Douglas et al., 2017, 2019).…”

“…The first four of these predictors were more powerful than any country-level predictors. For the most part, these individual-level predictors are in line with prior theory and research, including the results of a similar machine-learning analysis by Brandenstein (2022).…”

“…Recently, it has been argued that machine learning analyses can complement existing theory by facilitating the rigorous exploration of large datasets, casting a broader net and identifying potentially overlooked relevant predictors (Van Lissa, 2022a, 2022b). In a recent analysis of this type, Brandenstein (2022) analysed several predictors of conspiracy theorizing based on Douglas et al.’s (2017) framework of epistemic, existential, and social needs. Brandenstein conducted a machine‐learning analysis on a representative dataset of over 2000 UK citizens.…”

“…This analysis revealed that the relationships between epistemic, existential, and social factors and conspiracy belief—which have been observed in previous research—were largely supported. Brandenstein (2022) therefore sets a precedent for using a machine‐learning approach to study the predictors of conspiracy belief. In the current study, we go further by (a) examining a wider range of psychological predictors including many that have previously been untested, (b) examining a range of country‐level predictors, and (c) including a wide range of national contexts.…”