Do different means of recording sexual orientation affect its relationship with health and wellbeing?

Analyses of an individual's sexual orientation over time are desirable for policy evaluations and in estimating causal effects. We explore whether accounting for those who change sexual orientation over two time points, to create a measure of fluidity, produces substantially different results compared to sexual orientation measured at one time point and extrapolated to subsequent survey waves. We use seven waves of the UK Household Longitudinal Study which asked sexual orientation identity questions at two time points: waves three (2011-2013) and nine (2017-2019). Using the relationship with sexual orientation and various health outcomes as an empirical example, via a correlated random effects estimation approach, we find that the infrequent reporting of sexual orientation could over-estimate the negative impact for lesbian, gay and "other" individuals and under-estimate the negative impact for bisexuals. We further test the feasibility of the fluidity measure by examining attrition by sexual orientation identity and find small but statistically significant probabilities of attrition. Correction for attrition bias through inverse probability weighting makes little difference to the results. These results highlight the importance of accounting for changes in sexual orientation in empirical analysis and that doing so is feasible.