Local Authority FRS Reweighting Demo : Darlington

This is a simple demonstator which shows how you can reweight the Family Resources Survey dataset to so that when aggregated the data matches the characteristics of an English or Welsh Local Authority (as recorded in the 2011 Census).

The Council data is extracted from Nomis.

The program generates a set of weights which are the smallest distance (in some sense) from uniform weights. See Creedy [2003] and Deville and Sarndal (1992) for a discussion.

To use, select a Council for the menu at the top, check the boxes below for each criterion you want the data to match, and press 'generate'. If the reweighting succeeds, a .csv datafile will be created with a household reference numbers and weights; this can be merged with FRS data using the reference as the key. If the reweighting fails, try adjusting the parameters at the bottom of this page

This is very much a prototype. To do (at least):

  • Add extra targets, esp. benefit receipts, council tax bands;
  • Generate uprating factors for wages, unearned income, housing costs (needs research!);
  • Fix multiple user-interface problems (error-checking, menus too long ..), show/hide tables;
  • Better error handling and progress indication.

The source code and support files are available.

Possible Targets

Age in ranges: 0-4, 5-10, 11-15, 16-24, 25-34, 35-44, 45-54, 55-59, 60-64, 65-74, 75+
LA TotalsUniform-Weighted FRS
Genders (M/F) All residents
LA TotalsUniform-Weighted FRS
Ethnic Group: Aggregated (white, mixed, asian, black, other)
LA TotalsUniform-Weighted FRS
Tenure Type: Aggregated (owned outright, mortgaged inc. shared ownwership, social rented, private rented, rent-free)
LA TotalsUniform-Weighted FRS
Owned outright13,81016,257
Mortgaged inc. Shared Ownwership16,65714,772
Social Rented7,2288,974
Private Rented8,9756,667
Number of rooms (aggregated: 1,2-3,4-5,6+)
LA TotalsUniform-Weighted FRS
Number of rooms (disaggregated: 1..9 or more)
LA TotalsUniform-Weighted FRS
Occupation (aggregated: white_collar,blue_collar)
LA TotalsUniform-Weighted FRS
White Collar23,71324,787
Blue Collar25,30119,727
Occupation (Standard Occupational Classification 2000 disaggregated)
LA TotalsUniform-Weighted FRS
Managers, Directors and Senior Officials4,6396,801
Associate Professional and Technical5,7416,940
Administrative and Secretarial5,8865,011
Skilled Trades5,8084,697
Caring Leisure and Other Service4,8324,103
Sales and Customer Service5,4612,995
Process Plant and Machine Operatives3,5623,122
Elementary Occupations 5,6384,810
Economic Activity (ISO) Aggregated
LA TotalsUniform-Weighted FRS
Economic Activity (ISO) Aggregated
LA TotalsUniform-Weighted FRS
Full Time Employee30,33729,720
Part-Time Employee11,5559,580
Looking After Home or Family3,0813,301
Long Terms Sick or Disabled / Other Inactive/ Students9,85111,637
Housing Benefit
LA TotalsUniform-Weighted FRS
Housing Benefit9,8407,864
Council Tax Benefit (total recipients)
LA TotalsUniform-Weighted FRS
Council Tax Benefit (total recipients)11,9609,547
Housing Benefit and Council Tax Benefit
LA TotalsUniform-Weighted FRS
Housing Benefit and Council Tax Benefit (total recipients)21,80017,411
Income Support
LA TotalsUniform-Weighted FRS
Income Support (total recipients)2,3202,803
Job Seekers' Allowance
LA TotalsUniform-Weighted FRS
Job Seekers' Allowance (total recipients)3,5501,977
Income Support and Job Seekers' Allowance
LA TotalsUniform-Weighted FRS
Income Support and Job Seekers' Allowance (total recipients)5,8704,780
Working Tax Credits and child tax credits
LA TotalsUniform-Weighted FRS
Working Tax Credits and Child Tax Credits(total recipients)9,2009,528
Pension credits
LA TotalsUniform-Weighted FRS
Pension Credits (total recipients)5,1603,629

Available FRS Years

Reweighting Settings

The program minimises some measure of the dispersion of the weights, compared to uniform values. Five such 'distance functions' are availaible.

Chi-square (unconstrained) just minimises the sum of squares; this is not recommended as it can produce negative weights. Chi-Squared (constrained) is similar but puts upper and lower bounds on the weights. See the Deville and Sarndal paper for a description of the other measures; note that these frequently fail on this data and sometimes produce negative weights (indicative of a possible programming error). It's probably best to leave the distance function type as "constrained chi".

The upper and lower bounds represent how far from uniform the weights are allowed to go when using a constrained distance function; a lower-bound 0.1 means the weights cannot be less that 1/10th of the uniform weight, and an upper bound of 10 means that no household can have a weight greater than 10x uniform.

The weights generation can fail if the council population characterstics are very different from the FRS population. If this happens try widening the lower and upper bounds.

Weights can be generated so that the whole of the FRS sample grosses up to the Council's totals, or so that a subset of the FRS does - just England, or just the Government region containing the council. Using a subset might help ensure that the dataset better resembles the council in other ways, such as wage or rent levels.

Version Information

                        $Revision: 16246 $
                        $Date: 2013-07-05 11:58:31 +0100 (Fri, 05 Jul 2013) $