Sample
All methods were performed in accordance with the relevant guidelines and regulations. Participants from 23 countries on four continents (see Table 1 for list of included countries) were recruited using a stratified sampling approach created by the survey platform Cint. We recruited 1000 participants per country such that participants from each country were representative of the nation in terms of gender, age groups. Among the 28,229 participants who provided any data, the research team excluded participants who did not correctly respond to validity checks (e.g., not selecting the option ‘apple’ after being instructed to do so), provided nonsense responses to open-ended questions, or did not complete the survey, and reopened the survey to more potential participants (n =7,263). We also removed participants who did not identify as a man or woman (n = 164) for this study, resulting in a sample of 20,802 participants.
Measures
Respondents rated the frequency of their consumption of various classes of food from "1 = Never" to "11 = Two or more times per day". We calculated land animal consumption by summing scores for the categories "Cow/beef", "Pig/Pork", "Chicken or other fowl", and "Other land animals". As the items measure frequency within a category rather than portion size, we assumed that they represent the amount of meat eaten without being greatly influenced by gender differences in general caloric intake.
We used the Human Development Index (HDI), a multi-scale measure of human development that is published by the United Nations [34], to rank countries according to their level of development on the three dimensions: health (as measured by life expectancy at birth), education (as measured by the years of schooling and the expected years of schooling), and standard of living (as measured by gross national income per capita). This index is viewed as superior to the Global Domestic Product (GDP) as a measure of development as it, unlike the latter, represents not only average growth in income but also considers health and education as the key components of human development [35]. The data was derived manually from the website of the United Nations Development Program in January of 2023, where the data from 2021 was freely available.
We used the Global Gender Gap Index (GGGI), a multi-scale measure of gender equality that is published by the World Economic Forum [36], to assess national differences in gender norms. The score is calculated based on four indicators: economic participation and opportunity, educational attainment, health and survival, and political empowerment. The data from 2021 was derived from the Global Gender Gap Report 2021, which was published by the World Economic Forum in 2021.
Analyses
We first examined levels of study variables across countries to provide an initial indication of overall levels of meat consumption and gender differences across countries. We then tested hypotheses using multi-level intercept-and-slopes-as-outcome models with level 1 individual- and level 2 country-level predictors. We constructed increasingly complex models, starting with an intercept-only model and ending with a cross-level interaction model. All our models were constructed to be nested within less complex ones, giving us the ability to statistically compare the variance explained by the more complex models. In the first step, an intercept-only model was constructed to investigate the ratio of the variance of meat consumption that the country levels alone explain and to test whether our multi-level approach was necessary and justified. This model was specified as:
Yij = γ00 + µ0j + rij [1.1]
with Yij representing the combined meat consumption score for each individual i in each nation j, which is composed of γ00, the average standardized meat consumption score across the population of all countries j (i.e., the overall mean), µ0j, the individual deviation of each country j form this mean, and each Individual i's deviation from these means, rij.
In the next step, we added the Level 1 predictors of binary gender, linear age, and quadratic age, to compose a random intercept model predicting meat consumption. We included age because of research indicating that meat consumption varies across age, peaking at age 20–49 with lower consumption at younger and older ages (Daniel et al., 2011). The random intercept model was specified as:
Yij = γ00 + γ10 (genderij) + γ20 (ageij) + γ30 (ageij2) + µ0j + rij [1.2]
With γ10 being added to the equation as the regression slopes for gender across countries, and γ20 and γ30 as the regression slopes for age and quadratic age across countries. In this model, intercepts can vary randomly, with slopes remaining constant across countries.
Next, we tested a random coefficients model, for which we added the random coefficients for the slopes of gender, age, and the quadratic term of age. As described below, the age slopes did not explain additional variance. The resulting random coefficients model was specified as:
Yij = γ00 + γ10 (genderij) + γ20 (ageij) + γ30 (ageij2) + µ0j + µ1j (genderij) + rij [1.3]
With µ1j added as the varying slope of gender in each country j, representing how the extent of gender effects varies across countries.
In a final step, we added the human development and gender equality indices as moderators of the gender effect into the equation to create an intercepts-and-slopes-as-outcome model with a cross-level-interaction. These variables correlated r.= 63 in our data. We added them as variables in separate models to avoid multicollinearity.
The final models were specified as:
Yij = γ00 + γ10 (genderij) + γ20 (ageij) + γ30 (ageij2) + γ01 (moderator j) +
γ11 (moderator j) x (genderij) + µ0j + µ1j (genderij) + rij [1.4]
With moderator representing respectively the HDI for the model containing human development, GGGI for the model containing gender equality, γ01 representing the effects of development/equality across countries, and γ11 for the effect of the cross-level-interaction of development/equality and gender across countries and individuals.