STUDY OBJECTIVE: To examine changes in regional variance in all cause mortality rates in Great Britain from 1931-91 using formal statistical modelling procedures, and to follow up the suggestion by Illsley and Le Grand that there has been a reduction over time in the regional variance in younger but not older age groups. DESIGN, SETTING, AND PARTICIPANTS: Data were the age and sex specific death rates around each census from 1931-91 for the British regions, reconstructed to make them comparable with the 1981 regional definitions. Regional variance was modelled using bootstrap simulation tests and by age-period and age-cohort models. MEASUREMENTS AND MAIN RESULTS: While there was some evidence of a decline and levelling off of the regional variance over time in older age groups (over 35), the decline in younger age groups was more marked. This broadly confirms previous findings. Parametrising the period effect into linear and quadratic components, with allowance for an increase in regional variance in the war years, gave broadly comparable fit to the data as a model with period as a factor. Models for the changes in regional variance which were based on period effects seemed to provide a better description of the observed variances than those based on birth cohort effects. In the younger (but not older) groups there was evidence of a rise in the regional variance between 1981 and 1991. CONCLUSIONS: The decline in regional variance is larger in younger than in older age groups when allowance is made for the increase in regional variance over the war years. Statistical modelling can provide insights into the data which are not always detected by descriptive analyses. Moreover, they provide a capacity for generalisation beyond the particular data; relationships found can form the basis for studies of replicability, for example, in other countries.
STUDY OBJECTIVE: To examine changes in regional variance in all cause mortality rates in Great Britain from 1931-91 using formal statistical modelling procedures, and to follow up the suggestion by Illsley and Le Grand that there has been a reduction over time in the regional variance in younger but not older age groups. DESIGN, SETTING, AND PARTICIPANTS: Data were the age and sex specific death rates around each census from 1931-91 for the British regions, reconstructed to make them comparable with the 1981 regional definitions. Regional variance was modelled using bootstrap simulation tests and by age-period and age-cohort models. MEASUREMENTS AND MAIN RESULTS: While there was some evidence of a decline and levelling off of the regional variance over time in older age groups (over 35), the decline in younger age groups was more marked. This broadly confirms previous findings. Parametrising the period effect into linear and quadratic components, with allowance for an increase in regional variance in the war years, gave broadly comparable fit to the data as a model with period as a factor. Models for the changes in regional variance which were based on period effects seemed to provide a better description of the observed variances than those based on birth cohort effects. In the younger (but not older) groups there was evidence of a rise in the regional variance between 1981 and 1991. CONCLUSIONS: The decline in regional variance is larger in younger than in older age groups when allowance is made for the increase in regional variance over the war years. Statistical modelling can provide insights into the data which are not always detected by descriptive analyses. Moreover, they provide a capacity for generalisation beyond the particular data; relationships found can form the basis for studies of replicability, for example, in other countries.