Issue weights

If your analysis includes multiple time points or waves of Innovation Panel interviews you should use a longitudinal weight.

With each refreshment sample we create a new issue weight at an enumeration level. The issue weight consists of two parts: a cross-sectional weight for the new boost sample, for which participation in the panel is for the first time in that wave, and a longitudinal weight for the continuing part of the sample. 

You can see more information on refreshment samples on the Sample Design page.

Cross-sectional part

Refreshment (boost) samples are equal probability samples for Great Britain. Thus no correction for unequal probabilities is required. In order to correct for nonresponse at a household level we obtain predictors from our sampling frame, from interviewer observations in case the boost was administered fully by face-to-face interviewers, and from outside information. The outside information is obtained from official statistics resources (e.g. Census, Office for National Statistics (ONS)) and is linked to our dataset based on LSOA level. The predictors are checked for missingness and imputation is used where required. Response at a household level is then predicted using logistic regression. The inverse probabilities of response are then applied to everyone who was enumerated in responding households. This weight is then scaled to the mean of 1.

Longitudinal part

The continuing part of the panel requires correction for attrition between the previous boost wave and the current boost wave. Predictors from the household level questionnaire, household grid and enumeration information are used to predict this attrition. The inverse probabilities are then computed. To create a longitudinal weight these inverse probabilities are multiplied by the issue weight from the previous boost. Children born between the two boosts are given their mother’s longitudinal weight. The longitudinal weight is then scaled to the mean of 1. For further information take a look at our online training material Creating tailored weights for UKHLS. 

Combining longitudinal and cross-sectional parts

A new weight is created by giving a longitudinal weight to the continuing panel members and cross-sectional weight to the new panel members selected through the boost. The cross-tabulation of age, gender and region is then checked against ONS mid-year equivalent estimates and post-stratification is applied to match IP data to the population statistics. The final weight is then scaled to the mean of 1.

Cross-sectional weights

All cross-sectional weights are created in the same way using the most recent issue weight as a base weight. The enumeration nonresponse between issue weight and the relevant instrument is predicted using logistic regression. Predictors come from the household grid, household interview and enumeration information in the recent refreshment wave. The inverse response probabilities are multiplied by the issue weight. The resulting weight is then shared to TSMs through a weight share procedure. After this those who still have no valid weight value are given it through a nonzero weight-share method. The final weight is then scaled to the mean of 1.