Stata version 14.2 se

Descriptive statistics (e.g., proportions and means) were calculated using sampling probability weights to reflect the unequal probabilities in selecting study participants due to the study’s sampling design and controlling for clustering, i.e., similarity of participants at the settlement level. Wald tests were used to examine differences in the response proportions and average knowledge scores between FAARM intervention and control households. Linear regression models with both fixed and random effects (mixed-effects linear regression), applying sampling probability weights, and controlling for settlement-level clustering, were used to examine associations between household socio-demographic characteristics and mold mycotoxin knowledge score. Subsequently, a multivariable mixed-effect model was built with those socio-demographic variables associated in the crude analysis (p < 0.15), adjusting a priori for household wealth and the household head’s educational level. For ease of interpretation, household size and size of the homestead and agricultural land were categorized as illustrated in Table 2 and Supplementary Table S2, informed by categories used in previous research in Bangladesh [34 (link)]. Data management and analysis were performed with Stata version 14.2 SE (StataCorp LLC, College Station, TX, USA).