Information on data quality

Household Projections: Data quality

The household projections are a demographic projection informed by the projected population, produced by the Office for National Statistics, and projected household representative rates informed by Census and Labour Force Survey data. Household projections are therefore driven by assumptions on future levels of fertility, mortality and net migration, and household formation behaviour, that is how this population groups into household units (as measured by household representative rates).

Analysis of the components of household growth, presented in Live Table 415 shows that the population level is the main determiner of future household numbers, accounting for 72 per cent of growth in the 2008-based projections. The changing age structure of the population accounted for a further 20 per cent of household growth. Household formation behaviour makes a relatively small contribution to household numbers, accounting for 16 per cent of growth. Variant projections have been produced which provide an indication of the sensitivity of the projections to key assumptions in the population projections (net migration, mortality and fertility) and are presented in the Live Table 416 (see above link).

As the population projections are the main driver of household growth, a short summary of the accuracy of historic population projections is presented here. The ONS published an analysis of the accuracy of historic population projections in 2007 in Population Trends 128 - Fifty years of United Kingdom national population projections: how accurate have they been?(external link, PDF, 16 pages, 869kb).

The general conclusions from this analysis are that the error tends to increase with the projection period, so nearer projections are more accurate than longer run projections. The ONS analysis compared the historic projections with the latest estimates of the UK population up to mid-2005. When considering projections 20 years ahead (available from the 1955-based to 1985-based projections) the analysis found that the mean absolute error of the projected total UK population was about 2.5 per cent overall, and lower than 2 per cent when just the most recent (1975-based to 1985-based) projections were considered. This would correspond to around 1.4 to 1.7 million people in the UK (2.0 to 2.5 per cent mean absolute error) calculated on the 2008-based principal projection for 2028.

The table below roughly translates this mean absolute error in the UK population projections into the corresponding impact onto projected number of households in England by assuming an average household size.

	2028 (20th Projection Year)
Population in England*	59.05m people
+/- 2 to 2.5 per cent mean absolute error	+/- 1.18 to 1.48m people
Average household size	2.19 (persons per hh)
Households in England	26.47m
Impact of population error	+/- 0.54 to 0.67m households
per cent household error due to pop	+/- 2.0 to 2.5 per cent

* The population error has been adjusted to an England level to enable comparison with the household projections which are for England.

Whilst this is a simplistic approach, it may also be an overestimate of the aggregate error in household projections arising from this observed error in the population projections. This is because the ONS analysis found that the largest differences between projected and actual populations were in the youngest and oldest ages, while projections of the working age population were found to be comparatively accurate. Error in the population projections for the younger age groups will have little effect on the household projections.

In general terms, the greater the degree of disaggregation in the projections data, whether by household type, spatial area etc, the greater the risk of error arising between the projected data and the future numbers.

Gaps between the projected population and the current population estimate can also arise from errors in the population base. This was observed for projections made between the 1991 and 2001 Censuses where there was a substantial, and consistently growing, base population error. The 2001 Census showed that the population estimates rolled forward from the 1991 Census (on which the projections made between the Censuses had been based) had increasingly overestimated the population of the UK. Therefore, the projections are more likely to contain greater error the further the base period is from the latest census point. This is a general point which can arise from any statistic produced in an inter-censual period.

The 2011 Census data will therefore provide an excellent opportunity to assess the accuracy of historic projections and to consider whether there is evidence to further refine the methodology used to produce the household projections.

Subject to Department for Communities and Local Governments' central Statistical Planning activities and resource availability, we intend to undertake further work when the required data are available from the 2011 Census. In broad terms, it is intended that this work would:

assess differences between projected 2011 household numbers and the 2011 census results both in aggregate and by key dimensions
explore potential reasons for these differences and the implications for accuracy of the household projections.
Inform how the methodology for projecting household formation from Census trends and Labour Force Survey data can be improved.

It is anticipated that this work would be carried out as part of, or in the preparation for, the post-Census update of the household projections.