Our aim was to investigate how socio-demographic factors influence trends and age-related trajectories of fish consumption. We examined consumption of total, fried and recommended fish (white and oily fish, and shellfish) in the Whitehall II study over 11 years in participants aged 39–59 years at phase 3. The cohort included 8358 British civil servants who completed a FFQ at phase 3 (1991–3), phase 5 (1997–9, n 5430) and phase 7 (2002–4, n 5692). Occupational grade, ethnicity, marital and retirement status were collected at each phase. To analyse changes in age-related trends of fish intake over time according to socio-demographic characteristics, we applied a random mixed-effect model. Over the follow-up a significant increase in consumption of ‘recommended’ (mean: 1·85 to 2·22 portions/week) and total fish (mean: 2·32 to 2·65 portions/week) and a decreasing trend in fried-fish intake (mean: 0·47 to 0·43 portions/week) was observed. Recommended, fried and total fish consumption differed by occupational status, ethnicity, marital status and sex. The trend of age-related fish intake diverged significantly by ethnicity. In South Asian participants (n 432), slope of recommended-fish consumption was significantly higher compared with white participants (0·077 v. 0·025 portions/week per year). For black participants (n 275) slope of fried-fish intake was significantly higher compared with white participants (0·0052 v. − 0·0025 portions/week per year). In terms of public health, our descriptive and analytical work allows detailed understanding of the impact of socio-demographic factors on fish intake and its age-related trends. Such information is valuable for food policies that seek to promote health equity.

This paper shows how credibility theory can be encompassed within the theory of Hierarchical Generalized Linear Models. It is shown that credibility estimates are obtained by including random effects in the model. The framework of Hierarchical Generalized Linear Models allows a more extensive range of models to be used than straightforward credibility theory. The model fitting and testing procedures can be carried out using a standard statistical package. Thus, the paper contributes a further range of models which may be useful in a wide range of actuarial applications, including premium rating and claims reserving.