Can topic modelling of local newspaper texts enhance understanding of neighbourhood effects on health?

Abstract Social attributes of neighbourhoods, like heritage, and low-level social disorder, are not reflected in official metrics such as deprivation indices. However, research suggests these attributes are important for understanding spatial variations in health and social outcomes. This exploratory study investigated whether recurring themes in local newspaper articles capture meaningful social characteristics that help explain neighbourhood health resilience, defined as a dearth of illness after adjusting for deprivation. Topic modelling of geo-referenced texts identified and quantified 55 themes of commonly occurring words in Edinburgh, which capture salient neighbourhood attributes. Correlations between the themes and domains of the Scottish Index of Multiple Deprivation (SIMD) were weak, suggesting that newspaper themes captured characteristics beyond those in the SIMD. Reassuringly, expected correlations were observed between crime metrics from newspapers and the SIMD domains. Stepwise regression modelling revealed theoretically plausible themes associated with neighbourhood health resilience/vulnerability. Themes on heritage and community sports identity were positively associated with health resilience, whereas low-level social disorder (e.g. littering, antisocial behaviour) and 'local politics' were negatively associated. This study underscores the potential of using area-based topic modelling of newspaper texts to capture neighbourhood aspects neglected in official statistics but could further explain spatial variations in neighbourhood health outcomes.

Keywords: neighbourhood metrics, topic modelling, population health, resilience/vulnerability, index of multiple deprivation.