Work on MoSeS has involved aggregating 2001 UK Population Census (01UKPC) Output Area (OA) data to Ward level for Leeds Local Authority District. Due to the application of a Small Cell Adjustment Method (SCAM), the populations and household counts for OAs do not add up to those at Ward level for the same 01UKPC variables (ONS, 2005a). In general the application of the SCAM means that counts aggregated from smaller spatial resolutions are likely to be smaller although the mismatch varies depending on the 01UKPC Table, the region being examined (variable counts), and the application of other Disclosure Control/Protection Measures (ONS, 2005a).
For many reasons it would be useful to have a single count for the number of people and the number of households at OA level that is exact. Alas there is no 01UKPC Table that can provide this, and it is likely that the best that can be done is an estimate. This page is to organise collaboration on producing estimates. How best to proceed?
Oliver Duke-Williams and Andy Turner first met on 2006-02-09 to discuss this issue.
John Parsons expressed an interest on 2006-02-10 and commented that a UK dataset would be of wide general use. He would be a user and potentially interested in collaborating on this work.
Phil Rees , Paul Norman , Ludi Simpson and John Stillwell may also be interested in collaborating? Any others?
Oliver is looking into aggregations from the Census Interaction Datasets (CIDs). Due to the nature of CID data SCAM has rendered the OA level data almost useless (ref)... Although aggregated CID are not likely to be the best sources of 01UKPC OA household or population counts, there are both population and household tables in CIDs. Examining the variation in these counts is of interest.
Look into aggregations from the 01UKPC OA CAS for the UK.
For total population the Univariate Sex Table UV003 which has two cells looks promising. Table UV003 has the same definition for each country. In theory, if all OAs have values of more than 3 in each internal cell, it should be safe to use this one table?
The internal cell values for Table UV003 were examined. The minimum value was found to be 6...
Summary Statistics for comparing the totals from Table UV003 and Table CAS001:
nRecords 223060
meanDifferenceBigDecimal 0.0001613915538420156011835380615081144086792791177261723303147135299919304223078992199408230969245943
sumOfSquaredDifference 13001299
maxDifference 36
totalAbsoluteDifference 1671616
totalDifference -2673
standardDeviationOfDifferenceBigDecimal 0.0232907143889545807213238562696696935332188528414381065351060847917037618500851520355055349821439354
For a household count things are more difficult (Rees et al., 2005). As yet nobody has tackled this for the UK. There are numerous 01UKPC OA CAS tables containing household counts. These have a range of numbers of cells. Some analysis of the variation in totals would be necessary along the lines of already published work, (Rees et al, 2005). Because there are slight differences in UKPC OA CAS Tables in different countries, and differences in how SCAM is applied in difference countries, there is much to do to work out how best to do this work...
Examine the internal cell values for Table UV056...
Look up defintions of:
All household spaces
With residentsand
With no residents.
Investigate the use of Super Output Area (SOA) aggregations for presenting results, (ONS, 2005c). There is no other common aggregation to OAs for all the UK.
Population and address counts are variables in the All Fields Postcode Directory and in something called the postcode headcount file (Rees et al., 2005). These data offers a means of comparison. What else is there?
A count of communal establishments broken down by type and of vacant households might also be useful.
Supposing a set of population and household counts can be produced at OA level, what else can be done to reduce the errors and uncertainites that have arisen from the application of SCAM?
The details of the SCAM were not formally disclosed, but it is believed that the same method was used by both ONS and NISRA.
For the (OA level) STS Level 3, which relates to Scotland, SCAM _was_ used. This may well be the only case of it's use in Scotland.
SWS Level 3 were not produced for Northern Ireland due to disclosure fears.