Sample Design

There are significant differences in the approach to sampling for surveys of managed housing stocks against small-sample private sector surveys. Here we deal with the sample design for surveys of public sector housing stocks

Sample Size

The selection of appropriate sample size will, obviously, depend on the specific requirements of the survey - as finally defined, and the budget! However, careful appraisal is warranted: subject to the particular stock features, or policies, etc. of the stock owner concerned. 

In NBA's view, there are two alternatives :-

For most housing stocks, dependent on the uniformity of the stock and its geographical distribution, a sample of between 10% and 15% is adequate to ensure that this coverage is met on a property/zone basis. A sample of between 15% and 20% would normally be sufficient to extend coverage to each property type in each street.

Sample Selection

Extrapolation / 'Cloning' or Grossing

Normal statistical survey techniques employ a disproportionately stratified sample, selected randomly. Final survey data is held against only the surveyed properties - this data 'grossed' to the full stock through the application of weights (or multiply factors): thus no information is held directly for the unsurveyed percentage of the stock

This approach, while appropriate to private sector surveys will not, in our view, provide the level of detail that is required for the management and maintenance of a Council or Housing Associations own stock. In NBA's view, surveys designed to yield data for on-going use in planned maintenance must provide complete data records against each property on the housing register (thus allowing interrogation, and up-date, at an individual property level).

Consequently, for surveys of managed housing stocks a further step is normally required, whereby the survey data is 'extrapolated' (or 'cloned') to unsurveyed properties to ensure that data records are provided on a Per Property basis.

In order to permit this extrapolation, the sample stratification process requires more detailed appraisal and must be designed to ensure coverage of each dwelling variant within each defined location. Data for unsurveyed properties can, thus, be derived from a surveyed property of the same type within the same defined location. This is consistent with the ‘sub-stocks’ approach documented in DETR’s Collecting, Managing & Using Housing Stock Information – a Good Practice Guide, 2000. But goes well beyond this to ensure highly structured stratification consistent with accurate data ‘cloning’

With the known uniformity and the repetition of types prevalent within public sector housing the extrapolation of data by repetitive dwelling types is, obviously, the most accurate and reliable method of expanding the data to all 100% of the dwellings.

Defining Sample Criteria : Property types and Location

At the simplest level, 'Generic' Variant Codes can normally be established from the stock owners existing database information - providing codes against each dwelling that defines the property types and its location with reference to the following criteria:-

A hierarchical system of location referencing:- Area > Estate > Street > Block

Sample selection would be undertaken to ensure survey coverage of each defined variant code (above) within each defined location (e.g. Estate/ Street/ Block). Frequency Tables of Property Type & Location would be drawn up to determine the optimum sample size. A full evaluation of the sampling options and alternatives is then undertaken, with the relative merits of stratifying the sample, within the type criteria defined.

Sample Accuracy

The sample methodology that is outlined above and is required to ensure accurate extrapolation of data to unsurveyed properties differs considerably from normal probability sampling for which accuracy statements can be given. The highly structured sample that is required to allow for extrapolation divides the stock into very small unique sampling groups (i.e. each dwelling variant in each location).

However, in order to provide an approximate guide, sample error levels for an ‘average’ housing stock for reporting within each maintenance category would be expected to be around:-

Access & Reserves

For highly structured samples, that are designed to optimise the accuracy of fully extrapolated data, as discussed above, the emphasis should be on ensuring that all selected 'dwelling variants' are inspected. This means that if access fails to a given target property, it is not sufficient to select 'just another dwelling' but an alternative address of the same 'variant'. In order for this to be practicable there needs to be a reserve address for every dwelling variant to be targeted.