6 Ellie Oppenheim et al. Understanding of statelessness was measured using the following binary question ‘Do you know what it means to be a stateless person?’. Political orientation was measured on a scale from zero to one hundred where zero was leftwing and one hundred was right-wing. Scores were equal to the number the individual chose on the scale. Personal migration experience was measured by participants providing a binary response as to whether they had ever lived in a country other than their home country. Contact was measured by participants providing a binary response to the question ‘have you ever had any friends, relatives, or acquaintances who were or are asylum seekers, refugees or stateless?’. Social policy attitudes towards stateless people, asylum seekers, and refugees were separately assessed using measures based on Hartley and Pedersen (2015). Participants rated their level of support for policies aimed at each group on a seven-point Likert scale. Three policy statements were given for each group separately, and were as follows: ‘[stateless people/asylum seekers/ refugees] should have immediate access to all social services such as education, housing and healthcare’; ‘[stateless people/asylum seekers/refugees] should have the right to work as soon as they enter the UK’; ‘there is too much effort put into the care and support of [stateless people/ asylum seekers/refugees] in the community’ (reverse scored). A mean score of the three policy items was calculated with higher scores reflecting more restrictive social policy attitudes. Reliability analysis yielded McDonald’s ω of .86, .89 and .87 for social policy attitudes towards stateless people, asylum seekers, and refugees, respectively. Data analysis Preliminary analyses included calculating means, standard deviations, Spearman’s correlations, and difference tests between groups. Mixed graphical network analysis Mixed Graphical Modelling (MGM), i.e. network modelling, was used to explore the relationship between the study variables and social policy attitudes and prejudice towards all three groups. MGMs offer several advantages over hierarchical regression models. MGMs excel in handling diverse data types simultaneously, capturing complex and bidirectional relationships between variables, and providing a comprehensive network structure (Haslbeck et al., 2021). They perform automatic variable selection and handle missing data more effectively. MGMs are particularly useful for exploratory analysis, revealing unexpected connections and reducing dimensionality in high-dimensional datasets. Whilst hierarchical regression models focus on linear relationships and unidirectional influences, MGMs offer a more flexible and holistic approach to modelling complex systems, making them better suited for understanding intricate relationships and interdependencies among variables. While a nested structure is defined within a hierarchical regression, within network models, the structure emerges from the data. This provides useful insights into how the variables used in this study may work independently and together to inform social policy attitudes and prejudice. We estimated MGMs, in which measures were added as either continuous or categorical. In estimating the network, an elastic net regularisation reduces the inclusion of spurious edges, resulting in a sparse network with higher specificity (Epskamp et al., 2018). The regularisation parameter was selected with 10-fold cross-validation and specified that estimates across neighbourhood regressions should be combined (AND rule). As the regression on social policy https://doi.org/10.1017/S147474642510119X Published online by Cambridge University Press