We investigate the stability of individuals' affiliation to ethnic group categories using the Longitudinal Study of England and Wales linked between the Censuses of 1991 and 2001. While membership of the White category is stable, between seven and nine per cent of those recorded in an Asian group in 1991 have changed to a different group by 2001, while 23% of the Caribbean and African groups have
... [Show full abstract] changed. We quantify the separate influences of question unreliability, changes in categories, and conscious change of affiliation, finding that the latter contributes little instability over the period 1991-2001. The unreliability of the question is significant, due partly to the ambiguity of the categories for some people, and partly to imprecise imputation of missing values. We find the best correspondence between the different classifications used in 1991 and 2001 using empirical measures of fit. We investigate compatible classifications using larger and smaller numbers of categories.