Severity matters: Using network analysis to explain low and high levels of persecutory beliefs

Delusions, including persecutory beliefs, are theorised to arise from multifaceted interacting mechanisms. In our previous machine-learning study, 55 theory-derived predictors appeared to better account for lower levels of persecutory beliefs, whereas higher severity scores were underestimated. This raises the question of whether aetiological predictors differ in their relevance across the severity continuum. In this secondary analysis of data from a large, stratified online sample (N = 336), we applied network analyses using the ten strongest predictors from the prior model to compare their associations with low versus high levels of persecutory beliefs. Consistent with earlier findings, predictors accounted for more variance in subclinical than in clinically high persecutory beliefs. Association patterns also differed across levels of severity: In the low-severity network, persecutory beliefs were directly linked to aberrant salience and ostracism, whereas in the high-severity network, persecutory beliefs were most strongly linked to negative beliefs about mistrust and threat anticipation. These observations indicate that some findings from subclinical samples may not generalise to clinical populations. The emergence of distinct symptom dynamics across the continuum underscores the need for more targeted investigation, particularly at the more severe end of the delusion continuum. Future longitudinal, temporally fine-grained network analyses will be crucial for clarifying how mechanisms shift as persecutory beliefs intensify.