|
Executive summary
In this report we describe the ways we have projected a set of tables of the future population distribution across seven variables, relevant in the analysis of child poverty, across 13 regions. The seven variables are:
-
age
-
number of dependent children in families
-
type of family
-
size of households
-
ethnic group
-
number of earners
-
housing tenure.
We have used three different approaches to projection:
-
existing official projections (for age, some household categories, employment forecasts)
-
extrapolation of inter-census trends or FRS statistics (number of dependent children, type of family, size of households, number of earners, housing tenure)
-
our own cohort-component projections (ethnic group).
For each region we projected seven sets of marginal tables, each containing number of people and each summing to the most accurate total, the total regional population estimated by combining GAD national (2004-based) and ONS sub-national (2003-based) projections. This ensured consistency of the key variable tables.
We produced a seven variable by 13 region array by aggregating the individual records in the 2001 census Individual Sample of Anonymised Records. We built a seven dimensional iterative proportional fitting algorithm and associated implementation software to adjust this 2001 census-based array to marginal tables for 2010 and 2020. If this was successful, then it would be possible to choose combinations of the seven variables as tables for re-weighting the microsimulation results produced by IFS, using the FRS. In the event, it proved difficult to achieve convergence of the adjusted array and agreement between the summed marginals and the external constraints. Nevertheless we were able to improve on initial results by re-ordering the solution sequence and ironing out inconsistencies and errors in the constraint tables.
This was a pioneering attempt to quickly generate consistent projections of the UK population across more socio-economic dimensions than has been attempted before. The results are reasonably robust and can be used to re-weigh microsimulation results for 2010 and 2020 to yield an improved understanding of whether policy will achieve the goal of child poverty eradication. A more sophisticated approach will require moving to a dynamic microsimulation model, which is an altogether bigger undertaking.
|
|
Related documents
- 754Kb ~ 4 min (114 pages)
