Study Population
We employed data from Wave 3 of the “Food Usage During Covid-19” omnibus panel survey, which was jointly commissioned by the Food Foundation and researchers at King’s College London. It was administered online through YouGov over 6-8 July 2020, shortly after the end of the first UK lockdown. YouGov holds a large panel of registered users, and it uses an active sampling design whereby a representative sub-sample of the panel is selected in terms of age, gender, social class, and education. In analyses, this sub-sample is statistically weighted to match the national profile of British adults aged 18 and over, including those who cannot access the Internet. These weights were derived from the Census, ONS population estimates and large-scale random probability surveys such as the Labour Force Survey (for further details about the data and the sampling strategy used, see https://yougov.co.uk/about/panel-methodology). Our sample consisted of 4,350 UK respondents aged 18+.
Mental Health Outcomes
Anxiety was measured using the General Anxiety Disorder-7 questionnaire (GAD-7) (29), and depressive symptoms using the Patient Health Questionnaire (PHQ-9) (30). Participants indicated how often, in the past two weeks, they experienced each of the experiences outlined in 7 and 9 statements, respectively. For example, “feeling nervous, anxious, or on edge” (GAD statement) or “feeling down, depressed, or hopeless” (PHQ statement). Response categories were 1) always, 2) often, 3) sometimes, 4) rarely, 5) never. These differed slightly from the original modules’ responses; the full modules and how responses were coded to reflect the original scales are shown in the Appendix Table 1. GAD-7 scores ranged from 0 to 21 and PHQ-9 from 0 to 27. Both were dichotomised, using the clinical threshold of ≥10 for anxiety (29) and ≥15 for depressive symptoms (30). Higher scores indicated elevated anxiety and depressive symptoms, indicating clinically significant levels of severe anxiety and depression.
Food insecurity
We used two food insecurity measures. Severe/moderate food insecurity was measured using a modified version of three questions from the United States Department of Agriculture’s (USDA) food insecurity module (31). In this module, experiences of food insecurity arising from financial reasons are captured, however, the wording of questions was adjusted to also account for food insecurity arising from an inability to acquire food for any reason, thus, capturing lockdown-related conditions. Participants reported if they or anyone in their household in the past month (June 2020), experienced: having smaller meals than usual or skipping meals/ ever being hungry but not eating / not eating for a whole day because they could not afford or get access to food. A binary food insecurity variable was created, where agreement with any of the questions noted above, were categorised as food insecurity. Our second measure was worry about food, also known as marginal food insecurity (32). We employed this measure as we expected that while some affluent people may still be protected from moderate/severe food insecurity, more may have experienced milder food insecurity. Participants were asked: “How worried, if at all, are you currently about getting the food you need?” (not worried at all/not very worried=0, fairly/very worried=1).
Other measures
A wide range of potential confounders related to demographic and socioeconomic characteristics were considered in all multivariate analyses. As mental health symptoms decrease with age, we controlled for respondent’s age using a quadratic term to capture the non-linear relationship. We also adjusted for gender (0=men, 1=female), household composition (0=partnered without dependent children (aged 16 or under), 1=partnered with dependent children, 2=unpartnered without children, 3= lone parents) and life-limiting disability (0=limited a lot, 1=a little, 2= not at all limited), as women, single parent households and the disabled are at higher risk for both food insecurity and poor mental health (9).
We created a composite indicator of socioeconomic status (SES) (high, moderate, low) as previous studies have shown that education, household income, and employment status are important risk factors for food insecurity (8). Respondents reported their work status, their highest educational qualification, and their monthly household income after tax in June 2020. We wanted to capture a group that typically, would be at low food insecurity risk, from a group which would be at higher risk and more likely to be experiencing wider financial insecurity. We were conservative in our categorisation of people to SES groups. Respondents had to meet all criteria to be included in the high or low SES group. If they did not meet at least one of the income, work status or education criteria then they were placed in the moderate group. Due to this strict categorisation, the low and high SES group are expected to be smaller compared to the moderate group. More details on the construction of this variable can be found in Appendix Table 2. We also adjusted for broader financial hardship indicators, i.e., respondents were asked since the coronavirus outbreak in the UK (March 2020), how often any of the following statements were true: 1) they found themselves behind or unable with paying household bills/ rent/ mortgage/ Council-tax payments, 2) they had credit-card debt or bank overdraft that carried over from month to month (0=never or sometimes true, 1=often true). Another variable inquired how secure or insecure they felt about their personal financial situation over the next 12 months (0=financially secure, 1=financially insecure).
No more than 7% of the observations were missing for all variables, except for household income, with 27% missing observations. Thus, we included the missing income values in our analyses as a separate category within the income variable, although we otherwise applied a complete case analysis.
Statistical analysis
Descriptive statistics (t-tests and chi-square tests as appropriate) were used to present characteristics of the full sample, and by food insecurity status. Predicted probabilities derived from logistic regression models were used to examine the relationship between food insecurity/worry about food with anxiety and depression, adjusted for demographic, socioeconomic and financial hardship characteristics (1, 8, 9, 33-35). We also stratified models to assess whether these relationships varied across socioeconomic groups. Stata 16 was used to perform all analyses.