Cookies on the Gambling Commission website

The Gambling Commission website uses cookies to make the site work better for you. Some of these cookies are essential to how the site functions and others are optional. Optional cookies help us remember your settings, measure your use of the site and personalise how we communicate with you. Any data collected is anonymised and we do not set optional cookies unless you consent.

Set cookie preferences

You've accepted all cookies. You can change your cookie settings at any time.

Skip to main content

Report

Gambling Survey for Great Britain - Year 1 (2023), wave 1 report: Official statistics

Gambling Survey for Great Britain - Year 1 (2023), wave 1 report

  1. Contents
  2. Weighting strategy

Weighting strategy

The data was weighted to take account of non-response, bias, and improve representativeness. As there was no disproportionate sampling, selection weights were not required. The weighting method consisted of two stages4:

  1. a logistic regression model for number of responses within a household (run for households with more than one eligible adult), and

  2. a calibration to population estimates.

For the first stage, forward and backward stepwise logistic regression models were used to test which variables were associated with propensity to provide more than one response within a household. This model was run only for households with more than one eligible adult. Area-level variables (from the 2021 census for England and Wales and the 2011 census for Scotland) and household-level variables were tested. The final regression model included all variables that were significant in stepwise regressions: household tenure, household income, household type, index of multiple deprivation (IMD) percentiles (tertiles for England and bitiles5 for Wales and Scotland), 2021 output area classifications (for England and Wales only, as not yet available for Scotland), percentage of residents in National Statistics socio-economic classification (NS-SeC) categories 1 and 2 (interacted with a flag for Scotland, as the census 2021 and 2011 measures of NS-SeC had a different base), percentage of households that are owner occupiers, percentage of households with access to one or more cars, percentage of adult residents aged 55 years and over, and percentage of adult residents aged 65 years and over. Region of residence was also included in the model.

The predicted probabilities from this model were used to create response weights for households with more than one eligible adult. Weights were checked for outliers and left untrimmed. Weights for responding households with only one eligible adult were set to 1.

The response weights were then calibrated to estimates of the eligible population, residents of Great Britain aged 18 and over. Calibration weighting adjusts the weights so that characteristics of the weighted achieved sample match population estimates, reducing bias. The following variables were included in the calibration: age categories by sex, region, IMD percentiles (quintiles for England and bitiles for Wales and Scotland), tenure, and ethnicity.

Estimates of the GB population by age, sex, and region of residence were taken from Office for National Statistics (ONS) 2021 mid-year population estimates, which were for 2022 in England and Wales, and for 2021 in Scotland Population estimates for the UK, England, Wales, Scotland and Northern Ireland - Office for National Statistics (opens in new tab). Population estimates for IMD percentiles within each country were taken from ONS England and Wales (opens in new tab) and National Records of Scotland (opens in new tab). Population estimates for tenure and ethnicity were taken from the most recent Labour Force Survey data available, which was gathered between April and June 2023 Labour Force Survey performance and quality monitoring report: April to June 2023 (opens in new tab).

After calibration, the weights were checked for outliers and left untrimmed. The final weight for the 4,801 productive individuals has a design effect of 1.25, an effective sample size of 3,854, and efficiency of 80 percent.

References

4This same method was also used to weight Experimental Phase data, with the notable difference that highest level of education has not been included in the calibration variables for official statistics data collection. This is because the qualification questions in the GSGB are too different to those included in the Labour Force Survey (LFS) to be confident that they are measuring the same thing. Both the Experimental Phase and GSGB year 1, wave 1 response datasets show significant divergence in education profiles compared to LFS estimates. Therefore, calibration to LFS estimates of education would not be reliable and has the potential to increase bias rather than reduce it. Alternative high quality estimates of education levels are not available.

5 Bitiles tertiles and quintiles refer to the number of percentiles the data is divide into. The data is divided into; two percentiles for Bitiles, three percentiles for tertiles and five percentiles for quintiles.

Previous section
GSGB Year 1 (2023), wave 1 report - Questionnaire completion times
Next section
GSGB Appendix A - Online questionnaire
Is this page useful?
Back to top