Two main types of weights are released with UKHLS data: longitudinal weights and cross-sectional weights.
Longitudinal weights
Longitudinal weights should be used for any analysis that includes information from more than one wave. Longitudinal weights are created for monotone longitudinal response, meaning that households or respondents (depending on the weight) participate in each previous wave. This also means that for nonmonotone longitudinal analysis, where some panel members have missed one or more waves previously but have participated in other waves that are used in a particular analysis, the weight provided will give representative estimates but will not use all the available information (and therefore may have potentially lower statistical power than if a tailored weight is created for such analysis). In most such situations the sample size is still sufficient for analysis with the longitudinal weights provided, but where there may be concern about statistical power, a tailored longitudinal weight may be created by the user to reflect their specific combination of waves. For this you can follow our online course on Creating Tailored Weights. Note, that analysis with a tailored weight by expectation should give the same point estimates as analysis with the provided longitudinal weights but should result in slightly narrower confidence interval around these estimates.
Cross-sectional weights
Unlike in a cross-sectional study where participants are selected at the time of interest, in a longitudinal study participants are selected at one point of time, though their information can be used cross-sectionally years later. Cross-sectional weights in a longitudinal study allow a user to represent a population cross-sectionally with some exceptions (note that a longitudinal study does not include people who spend a few months, or even a few years in the country, specifically recent immigrants since the last boost and their children are excluded from the Study, and if they remain in the country this is true until the following boost). Cross-sectional weights should only be used if all the information in an analysis comes from one wave.
Cross-sectional weights are created based on longitudinal enumeration weights. Enumeration represents response at a household level and gives a weight to all household members even if only one person in the household responds in a particular wave. Longitudinal enumeration weights are only available for OSMs (original sample members, those that were selected into the Study or their children born into the Study). But given that UKHLS is a household study and interviews all household members, it therefore also gathers information on TSMs (temporary sample members, those that were not in a household at the beginning, but joined a household sometime during the course of the Study). Information from TSMs is incorporated into the information from OSMs through a weight-share method. Any other cross-sectional weight is then modelled from the enumeration weight. Information on the sample design for Understanding Society can be found in the Study Design section.
Sharing information from OSMs to TSMs results in the situation where everyone who has a valid answer in a particular wave and is part of a continuously enumerated household has a positive non-zero cross-sectional weight. Yet there will still be some households with some missing waves, even though in a current wave of interest some or all household members have completed interviews. With each year a chance of a household missing one or more waves in the past but still participating is increasing, which results in an increase in potential zero cross-sectional weights. This situation is resolved through another stage of weight-share, although importantly there is a set of panel members whose weights always remain zero. This group is called TSMs from Wave 1, and they represent people who were not eligible for selection into Ethnic Minority Boost at Wave 1 of UKHLS or into Immigrant and Ethnic Minority Boost at Wave 6 of UKHLS, but who live in the same household with people eligible for selection into these boosts. TSMs from Wave 1 are different to those people who join household later in the Study (TSMs) as their selection probabilities are set to 0 from the start, and this defines their later weights which always have to be zero. Interviewing these people is still important as their information provides context to understanding immigrants and ethnic minority household life. Yet their interview information cannot be used in a similar way to that of other panel participants, which results in their zero cross-sectional weights. Note, that the number of TSMs from Wave 1 is very small and is decreasing with time as when such people leave the household they are not followed.
