Weighting LFS survey data to get population based estimates
The LFS collects data on a sample of the population. To convert this information to give estimates for the population, the sample data is weighted. Each case is given a weight which can be thought of as the number of people that case represents. This weighting factor takes account of differential non-response among different sub-groups in the population. This weighting procedure involves grossing data to sub-regional population estimates and then adjusting for the estimated age and sex composition, by region.
These weighting factors are provided to HSE by the Office for National Statistics (ONS) along with the LFS dataset itself.
In response to the coronavirus pandemic, the Office for National Statistics (ONS) made some changes to the LFS data collection and various refinements to the weighting1 methodology. From March 2020 (with the onset of the coronavirus pandemic), face-to-face interviews for respondents newly joining the survey2 were moved to telephone interviews. With a corresponding fall in response rates, the wave 1 sample was doubled in July 2020 to increase the achieved sample size. In April 2021, the introduction of the field strategy referred to as ‘Knock to Nudge’3 had a positive impact on the response rates and the sample was reduced. ONS found that since these changes were implemented certain characteristics were not as well represented as in earlier surveys, introducing an increased non-response bias to the survey. As a result, various improvements have been made to the ONS weighting methodology. For further details see Impact of reweighting on Labour Force Survey key indicators - Office for National Statistics (ons.gov.uk)
Whilst the ONS provided weighting factors take account of non-response to the LFS itself, they do not account for non-response to individual questions within it, in particular the Workplace Injury and SWI survey modules' screening questions. In 2001/02, just over 5% of eligible respondents did not respond to either or both of the screening questions within the workplace injury and SWI survey modules (most of whom were unavailable for interview); in recent years this has increased to over 15%. Since this non-response is not accounted for in the standard LFS weighting procedure, an adjustment is made to the weights to take account of this non-response and provide more reliable estimates.
For the non-response adjustments a procedure is implemented to identify demographic and employment-related variables and codes that best describe the pattern of response to the ill health and workplace injury screening questions. The procedure progressively splits the sample into groups between which the response rates differ to a statistically significant extent. The adjustment for non-response then involves multiplying the standard LFS weights by the reciprocal of the response rate within each of the groups identified.
In line with the ONS’ new weighting methodology, HSE’s further adjustment has also been revised to better account for the change in non-response bias. Housing tenure has been included in the model when identifying variables that best describe the pattern of response and completing the adjustments on a weekly basis has also been introduced.
ONS revise the LFS weights (estimates for the population) following a census. A new policy was introduced in 2016, to revise weights annually rather than periodically between censuses, providing users with the most up to date estimates. The frequency was changed to every other year from 2018. Revisions are generally made to the previous three years. HSE assess the impact on the work-related illness and workplace injury published estimates, and a judgement about the need to revise is made. Any subsequent revisions made are noted in the revision log.
- The LFS collects data on a sample of the population. To convert this information to give estimates for the population, the sample data is weighted. Back to reference of footnote 1
- For the LFS, people are interviewed in five consecutive quarters, with the first interview (wave 1) generally being face-to-face. Back to reference of footnote 2
- Interviewers visit sampled addresses where no telephone. number can be obtained and encourage respondents to provide their phone number and arrange an appointment by knocking on the door. Back to reference of footnote 3