SIC Mapping in the UK Labour Force Survey
The UK Labour Force Survey
The LFS is a household survey spanning 1975 to the present day and is the source of most official UK labour market statistics produced by the Office for National Statistics (ONS). Details of the LFS can be found at the Economic and Social Data Service (ESDS) by clicking here. The LFS was biennial between 1975 and 1983, annual from 1983 to 1991, and quarterly data based on a rotating panel are available from 1992 Q2 onwards.
Over the period for which LFS data are available, the UK official Standard Industrial Classification has changed. In some cases, this change was prompted by a consensus that the existing industry classification had become partly obselete due to economic developments. In other cases, SIC changes were intended to improve harmonisation of industry classifications across Europe.
SIC Classifications in the LFS
The LFS has featured the following SICs as the basis for industry classification:
|SIC 1968||1975 to 1979|
|SIC 1980||1981 to 1993 Autumn seasonal quarter|
|SIC 1992||1993-94 Winter seasonal quarter to 2008 Q4|
|SIC 2007||2009 Q1 to present|
From 1994 onwards, the LFS has included at least one measure of SIC based on the previous classification. During the SIC 1992 period, classification by SIC 1980 1-digit Divisions is available, and during the SIC 2007 period the LFS has and will include variables coding observations into 60 SIC 1992 2-digit Divisions and 17 "1-digit" Sections as well as a further aggregation into 9 1-digit Sectors.
Do you need a one-to-one mapping or a proportional mapping?
A proportional mapping captures as accurately as possible the 'true' correspondences between two classifications. This accuracy is an obvious merit. A proportional mapping is useful when the focus is on aggregate or mean measures - for example, employment by industry or mean transition flow rates out of unemployment by industry.
However, a proportional mapping should not be used when it is of interest to observe industry for a given individual - for example, when measuring inter-industry transitions at individual person level. A proportionate mapping would not be suitable because such a mapping can imply that an individual changes industry (on the new classification) when in truth they did not (according to the initial classification). Consider an individual classified by the original SIC into industry A in time periods t and t+1, where industry A is mapped to a number of industries (X, Y and Z for example) according to the new classificaton. Assume the individual is classified to industry X at t. A proportional mapping would result in a probability less than 1 of being allocated to the same industry X at time t+1.
When consistency across time is required in the mapping, a direct, or one-to-one, mapping should be used. One-to-one mappings will entail less misclassification error if they involve mapping from a disaggregated to a more aggregated level. For example, publicly released quarterly LFS data from 2009 onwards includes industry information based on SIC 1992 generated by a one-to-one mapping of each 4-digit SIC 2007 Class to a single 2-digit SIC 1992 Division, to a single "1-digit" SIC 1992 Section and to a single 1-digit SIC 1992 Sector.
My Resources pages highlight sources of information about both proportional and direct one-to-one mappings for SICs.