We have developed new, efficient, methods to fit interpretable and generative network models to survey data. We present recent work analysing the friendship questions from both Understanding Society and the BHPS with comparison to the US via the General Social Survey and American Life Panel. These allow us to return posteriors on parameters of a kernel function that gives the probability of individuals having a friendship conditional on their relative social co-ordinates: this kernel quantifies the homophilic preferences of the respondents. We find that for the UK the parameters of the social kernel, and thereby respondent’s homophilic tendencies, are broadly unchanging over the duration of the surveys and, further, we find the parameters are similar between the US and UK. Equipped with this kernel, and given the joint distribution of respondent characteristics, we define a social access statistic which returns the proportion of all individuals in the survey a randomly chosen individual from the survey can befriend. We find that this statistic is declining significantly over time in the UK, likely driven by changes in the distribution of ages and strong age-related homophily. We also find that the major factor modulating the degree of social access is spatial separation followed by age-separation, sex differences and occupation. These results are recapitulated for the US data. We will conclude by discussing how new survey designs, mobile phone data and large-scale behavioural data can all provide additional avenues to probing the important dimensions of homophily while pointing to their health relevance.