Replenishment
Rationale
In order to run the model beyond the length of the Understanding Society data (11 waves, currently from 2009 - 2020), we needed to generate a replenishing population that would maintain the age structure, as well as being representative of the British population into the future in terms of sex and ethnicity.
Data Sources
Counts by age, sex, and year were generated for 2008 - 2070 from 2 sources: 1. National Midyear estimates for 2008 - 2020 1. These were obtained using the nomisweb data query tool 2. Principal Population Projections (PPP) for 2021 - 2070 1. Principal Population Projections
Counts by ethnic group were obtained from a previous project involving Nik Lomax called Ethpop. These were counts by ethnic group by age and LAD, which were summed to generate national level counts of ethnicity by single year of age. See this link for more information about ethpop.
Method
What we did to ensure that the simulation maintains the age structure of the data moving into the future, was to begin with a population derived from a single year of understanding society (we picked 2018). The replenishing population is also derived from US 2018, but instead takes just the 16 year olds. Each year in the simulation, every simulant is transformed and ages by 1 year, which means the 16 year olds at time t will be 17 at time t+1, meaning there are no 16 year olds in the simulation. We then take the 16 year olds from the replenishing population, which have been reweighted to be representative for that year by sex and age (see below) and add them into the simulation to replace the 16 year olds that have aged.
Reweighting
As part of the data generation pipeline, we take the final_US datafiles (after a number of pre-processing and imputation steps) and use them to generate both the starting and replenishing populations. The model begins with a population derived from Understanding Society in 2018 (wave 9), from which the analysis weights provided by the survey are then modified based on the counts from both the age-sex projections and the counts by ethnicity. Replenishing populations are also derived from US wave 9, where we take all 16 year olds from wave 9, duplicate them for every year from 2009 - 2070, and reweight these populations by age, sex, and ethnicity. The actual reweighting calculation is very simple:
new_weight = (old_weight * count_by_group) / total_weight_by_group
Complication
A slight complication to this method of replenishing is that education states now have to be transformed over time in the model. This is handled by the Education module, which is described on a different page.