*2.5. Statistical Analyses*

Descriptive statistics were used to summarise demographic, socioeconomic and household food spend variables, both overall and across away from home food outlet exposure levels. Study weights for Wave 1 cross-sectional analyses, prepared and provided by UKHLS, were used to account for participant non-response and clustered study design. Therefore, weighted mean percentages (with 95% CIs) are presented rather than raw frequencies.

Binary logistic regression was used to model high away from home food spend households by quintile of away from home food outlet density (Model 0). Crude models were adjusted for demographic variables (age and sex), proportion of restaurants, proportion of fast food outlets and proportion of cafés, total number of other food sources and total equivalised household food spend (Model 1). Lastly, Model 1 was additionally adjusted for socioeconomic variables (education and income) (Model 2). All statistical analyses were carried out in Stata version 14 (StataCorp LP., College Station, TX, USA) [39].
