This study is based on secondary analysis from a prior study where the aim was to understand the views of the public about lifestyle behaviours [12]. Favourable ethical opinion for data collection was granted by the Loughborough University Ethical Committee for Human Participants (reference: 17722). All methods were performed in accordance with the relevant research ethics guidelines and regulations. Written informed consent was obtained from participants when they agreed to take part in the primary study. This study uses data from a nationally representative sample of adults from across the UK recruited via the UK Ipsos KnowledgePanel. Panellists are recruited using a random probability unclustered address-based sampling method where every household in the UK has a known chance of being selected to join the panel [13]. Letters are sent to selected addresses in the UK inviting occupants to become members of the panel. Up to two household occupants are permitted to join the panel. If an occupant does not have access to the internet, they can register to the KnowledgePanel by post or telephone, and are given a free tablet, an email address, and access to the internet so that they are able to complete surveys online. These inclusive methodological approaches not only improve population coverage, but also provide a more effective means for recruiting hard-to-reach members of the public.

At the time of joining the KnowledgePanel, panellists completed a general online administered informed consent process. As the KnowledgePanel is a random probability survey, panel invited samples are stratified when conducting waves to account for any profile skews within the panel. At the time of this research the KnowledgePanel contained 14,016 available panellists. Of these, 4000 were selected at random and invited to take part in the primary study on which this secondary study is based. The sample was stratified by region and education. The sample drawn was then reviewed against population statistics on ethnicity, index of multiple deprivation (IMD) [14], urbanity of home postcode (rural vs urban), age and self-reported gender. IMD contains seven domains of deprivation: income, employment, education, health, crime, barriers to housing and services; and living environment. The sample size was fixed hence no formal sample size calculations were made, however the sample size is large enough to satisfy the general rules of thumb of two participants per variable and 10 events per variable required for adequate estimation of linear and logistic regression coefficients respectively [15].

Data was collected between 24–30 June 2021. On joining the KnowledgePanel, panellists were asked to complete a core survey profile, which included questions about age, self-reported gender, ethnicity, income, country of residence, postcode, highest level of education and the political party voted for at the 2019 UK general election. Data for self-reported height and weight were collected as part of a separate parallel survey. Postcode was used to calculate IMD quintile and to classify the political party that won the constituency where participants lived at the 2019 general election. All data received by the research team were de-identified.

No patients or members of the public were involved in the planning or design of this study.

To ensure the findings were representative of the UK population, the below weighting specification was applied to the data in line with the target sample profile. As only two members per UK household are allowed to register with the KnowledgePanel, a design weight to correct for unequal probabilities of selection of household members was employed. Calibration weights were applied using the latest population statistics relevant to the surveyed UK population. England and Wales, Scotland and Northern Ireland were each weighted separately while an additional weight was created for the UK to account for any over or under sampling within each of these countries. Two sets of calibration weights were applied. Calibration weighting was applied using region and an interlocked variable of self-reported gender by age (both use Office of National Statistics (ONS) 2020 mid-year population estimates as the weighting target [16]. Demographic weights were then applied to correct for imbalances in the achieved sample and the data weighted on education, ethnicity, IMD (quintiles), and number of adults in the household. Estimates from the ONS 2020 mid-year population estimates and the Annual Population Survey were used as the weighting target [16]. Statistics are reported as unweighted frequencies and weighted percentages to maximize transparency, unless indicated otherwise as weighted results.

The political party categories of interest at the 2019 UK general election were Conservative, Labour, and the Liberal Democrats as these typically share the largest proportion of electoral support in the UK [17]. A small number of participants voted for other political parties (i.e., Scottish National Party, Brexit Party, Plaid Cymru, Green Party, UK Independent Party, British National Party, Democratic Unionist, Ulster Unionist, Sinn Féin, People Before Profit & Alliance). These were included in the statistical modelling as a combined group, and whilst it was important for these data to remain in the dataset for analysis purposes, their comparison with other political parties were not of interest and not reported. Analyses for the political party of participants’ constituency Member of Parliament (MPs) and living with obesity outcomes compared participants who voted for the Conservative, Labour and the Scottish National (SNP) parties as there were more voters living in constituencies won by the SNP than for the Liberal Democrats Party, the Liberal Democrats Party being included in the combined group for this analysis.

Linear regression models were used to explore the association between political party affiliation/voting and BMI score in unadjusted and adjusted analyses. Unadjusted models included political party voting as the only independent variable; adjusted models also included age group, self-reported gender, IMD quintile and country of residence (England, other). Age was treated as a categorical variable to allow for its non-linear relationship with BMI.

Logistic regression (unadjusted and adjusted models) was used to explore the differences between political party voting and BMI category in two different ways: people with obesity versus people without obesity, as well as using the BMI categories (underweight, healthy weight, people with overweight, people with obesity) (ordinal logistic) [18]. Similar models were used to investigate whether participants who live in constituencies won by Conservative Party MPs at the 2019 general election were more likely to have a higher BMI score and to be people with obesity than participants who live in constituencies won by MPs from other political parties. Model performance included R2 (linear regression models); pseudo R2 (logistic models); and variance inflation factor (VIF), where VIF > 5 was considered severe multicollinearity. Proportional odds assumptions were tested using the Brant test. Two-sided p values < 0.05 were considered statistically significant. Primary analysis was performed on complete cases with multiple imputation by chained equations (100 imputations) included as a sensitivity analysis using Stata version 17. Combined calibration and demographic weighting was performed using the SVYSET command in Stata.