Evaluating small area population estimates and projection for sub-council areas in Scotland

Abstract

This thesis is a collaborative research project with the National Records of Scotland and seeks to provide a fresh assessment of small area population estimates, and the first evaluation of small area population projections in Scotland. Population estimates and projections are valuable tools for planners and policy makers, with small area population statistics becoming increasingly important as demand grows for more detailed data. From planning school place provision to adequate water services, there are many aspects of planning and policy making which depend on having knowledge of the population size and structure at a neighbourhood level. This research uses a mixed methods approach, evaluating historical estimates and projections using statistical techniques; as well as using qualitative analysis to examine how local users of these statistics engage with, and accommodate for, the potential for error inherent in these estimates and projections. This research focuses in particular on estimates produced by the Cohort Component method currently used in Scotland, comparing this approach to alternative methods such as those employed by other statistical agencies, and less data intensive, simple methods. A significant finding from this comparison is that methods favoured by other UK statistical agencies outperformed the Cohort Component method. Results show that both the Ratio Change method used in England and Wales, and the Average method, used in Northern Ireland, both produced the most accurate estimates. When exploring how these methods varied in accuracy across areas, results also found evidence of bias. The most striking finding from this evaluation was the relationship between estimation bias and deprivation, with population estimates in the most deprived areas, tending to be under-estimated and the most affluent areas over estimated. This was a finding which was present across all of the complex methods of population estimates that were included in this study, however the effect was strongest for the Cohort Component method. While these findings may suggest that Cohort Component method may not be the most appropriate for producing population estimates, it was the best performing method when evaluating population projections. Here, it was found that this method outperformed all the simple approaches included in this study. However, there is some evidence to support the use of these simpler methods in some circumstances, over short projection periods. While the simpler methods were less accurate than the Cohort Component method, all approaches included in this study met the threshold of 80% of projections within 10% of the true population, which the shelf life literature defines as a reliable projection. This suggests that, over short projection periods, the simple methods can be considered reliable and useful, despite marginally higher levels of error. These simple methods could therefore be recommended to local users who wish to produce their own projections. This desire from users for locally produced statistics was evident in this research. Participants felt that local knowledge, particularly regarding special populations, could improve the assumptions used in producing population statistics. Participants also felt that more could be done to provide greater context to the population change presented in the projections in order for them to be taken seriously by non-expert audiences. Taking these views into account, closer links between national and local bodies, is a key recommendation of this research when considering improving user experience.