Understanding Society: design overview

Key takeaways

content: "@buckUnderstandingSocietyDesign2011" -file:@buckUnderstandingSocietyDesign2011

Reading notes

Imported on 2024-06-23 13:33

⭐ Important

• & Understanding Society builds on the success of the British Household Panel Survey (BHPS). (p. 5)
• & Understanding Society is a major addition to this portfolio. As a longitudinal study it follows individuals over time, regularly collecting data about each participant, together with similar data about all other members of his or her household. The design allows it to provide information on the persistence of such states as unemployment, child poverty or disability; on factors which influence key life transitions, such as marriage and divorce, labour force entries and exits and retirement; (p. 6)
• & The household panel design of Understanding Society contrasts with that of the cohort design of many other longitudinal studies. In these cohort studies, a sample of individuals from a particular age group is selected and followed. There are birth cohorts, where the sample is selected around birth and followed thereafter, cohorts of young people, often selected from a particular school year, or ageing cohorts, where people over a threshold age are followed. In the household panel design, a sample of the whole population is selected in their household context. It is important to stress here that the longitudinal elements, just as in the cohort studies, are the individual people. It is not a longitudinal study of households, since arguably households have no coherent existence over time, and focusing analysis only on households whose composition does not change between waves leads to severe biases (see Duncan and Hill 1985). Rather, it is a study of individuals in their changing household contexts and this context is very important for analysis of many life domains (Giles 2001). (p. 7)
• & There are three key distinctive features of the household panel design compared with the cohort design. Firstly, while a birth cohort study is representative of the population in that particular cohort, the household panel is a representative survey of the whole population of all ages. (p. 7)
• & Research from studies with a household panel design supports direct inferences ab out the whole population. Since the study has a full range of age cohorts, and because births to sample members join the sample, there is a representative sequence of new cohorts constantly replenishing the study. Consequently, studies with a household panel design are an important complement to cohort studies by supporting generalisations beyond specific cohorts. (p. 7)
• & Secondly, it is a survey based on households. Multiple social environments shape behaviours and life circumstances. For example Skew and Wolke, in this Special Section, examine bullying in relation to school and home. However, households can be closely observed in Understanding Society. Economic welfare, income and material conditions are normally assessed at the household level, because of the degree of sharing of resources. Households also provide a context for understanding the social and cultural resources available to individuals, both children and adults. The collection of comparable data from each individual in the household at each wave provides a natural way of collecting rich household level data. It also provides a very important resource for the study of how households are changing and the demographic processes which lead to household composition change. (p. 7)
• & Thirdly and related, the household focus also provides a way of understanding the interrelationships between individuals within households and families. Many of the key decisions which individuals make are influenced by other household members. A focus on households also provides an opportunity to investigate the inter-relationship with families. Families clearly extend beyond household boundaries and not all household members will consider themselves family members. The study supports research on the interactions over time with family members outside the household, and the evolution of relationships within the household. For example, household panel surveys have contributed to the study of resident and non-resident parents and their contributions to the developmental outcomes for children (Ermisch 2008). (p. 7)
• & Observing multiple generations and all siblings allows examination of long-term transmission processes and isolates the effects of commonly shared family background characteristics. Articles in this issue make use of household relationships in their analyses. (p. 8)
• & The survey’s large sample size is a key attribute. The target of 40,000 households across the study’s samples will permit exploration of questions for which other longitudinal surveys are too small to support effective research. Many relatively rare events or sub-populations can be studied with the survey. It permits analysis of small sub -groups, people who moved to the UK as children or disabled people, or regional and sub-regional levels. It allows examination of the effects of geographical variation in policy, for example differences between the countries of the UK. The large sample size also allows high-resolution analysis of events in time, for example focussing on single-year age cohorts. As an example, with a total of approximately 1,000 births to sampled women per year, it will be possible to study births to teenage mothers. (p. 8)
• & Understanding Society will provide important new information about ethnic minorities through over-sampling of ethnic minorities and the collection of additional measurements relevant to their life experiences. These additional measures are asked of members of the boost sample (oversample), of members of ethnic minority groups not sampled as part of the boost sample, and by a comparison group of around 1,000 adults from the general population sample (Berthoud et al 2009) . (p. 8)
• & Understanding Society has four sample components: a) the general population sample, b) the ethnic minority boost sample, c) the Innovation Panel, and d) the sample of participants from the BHPS. The sample designs are similar in having multi-stage sample designs mostly with stratification and clustering. However, the sample design of each sample component has some unique features, (Lynn 2009) which are discussed below. (p. 9)
• & The general population sample is a stratified, clustered, equal probability sample of residential addresses drawn to a uniform design throughout the whole of the UK (including north of the Caledonian Canal). The Northern Ireland sample is not clustered. Within Great Britain, the Primary Sampling Units (PSUs) are postal sectors stratified by nine regions of England plus Scotland and Wales), population density and minority ethnic density. 2,640 postal sectors were selected systematically, with probability proportional to size (number of addresses). Within each sampled sector, 18 addresses were selected systematically, resulting in an equal-probability sample of a total of 47,520 addresses in Great Britain. In Northern Ireland, 2,400 addresses were selected systematically from the Land and Property Services Agency list of domestic properties, thus making a total of 49,920 selected addresses in the UK. Since constraints of survey capacity meant that fieldwork needed to be spread over a two year period, the overall sample was divided into 24 monthly sub-samples, each independently representative of the UK population. (p. 9)
• & The goal for the ethnic minority boost sample was to provide samples of at least 1,000 adults in each of the five largest ethnic minority groups: Indian, Pakistani, Bangladeshi, Caribbean and African. Such a sample would support groupspecific analyses of these ethnic groups (Berthoud et al 2009). While the sampling targets are defined in terms of numbers of adults, the sample is of households. (p. 10)
• & The sampling approach first identifies geographic areas with at least 5% density of ethnic minority groups. Because the 2001 Census was becoming outdated, the density estimates were adjusted using more recent survey estimates. These high density sectors were 36 per cent of the total sectors and accounted for 85% of all members of minorities. Further sub-sampling of the high density areas was done to increase the efficiency of the yield. Thus, a higher sampling fraction was used for areas anticipated to yield three or more households while successively smaller fractions were used for areas expected to yield two, one or zero ethnic minority households. (p. 10)
• & Understanding Society incorporates the BHPS sample members into the overall sample design beginning in Wave 2. The extensive longitudinal data of the BHPS has great scientific value, including the opportunity for early longitudinal analyses of Understanding Society. The BHPS was a random sample of Great Britain, excluding the Scottish Highlands and Islands. In its first wave in 1991, it achieved a sample of 5,500 households. Boost samples of Scotland and Wales were added in 1999 and of Northern Ireland in 2001. These modifications were motivated by interest in analyses in these countries, related to political changes associated with devolution in the UK. (p. 10)
• & After consultation, it was decided that it was most important to ensure the integration of BHPS into the new study (Laurie 2010). So instead of having its fieldwork concentrated between September and December, as was the practice up to 2008, fieldwork is distributed evenly over the 12 months of the first year of data collection beginning in January 2010, as part of wave 2 of Understanding Society. (p. 10)
• & rom wave 2 onwards the BHPS sample has the same questionnaire as the Understanding Society general population sample. Jenkins and Taylor in this issue present rates of employment from 1991 to 2009 from BHPS and Understanding Society data. (p. 10)
• & The overall achieved initial sample was targeted at 40,000 households: approximately 26,000 from the general population sample, 4,000 from the ethnic minority boost, 1,500 from the Innovation Panel, and 8,400 from the BHPS participants at wave 18 of that study. The total achieved numbers across these four components were 39,802 households containing 101,086 individuals, including children. (p. 11)
• & Understanding Society has a complex sample design and will be used in various ways by data analysts. Consequently the weighting strategy is also complex. Understanding Society must provide weights for the household and individual levels, for units that respond or do not respond to different instruments, e.g. the self-completion instrument, for responding to different combinations of study waves, and for the diverse sample components. Sampling information, including primary sampling unit and strata identifiers, will be available on the data set. (p. 11)
• & In general, weights are the product of a design weight to convey the probability of selection, adjustment for non-response, and sometimes poststratification, to make the distribution a closer match to the population distribution. (p. 11)
• & The development of weights also takes the time pattern of response into consideration. For example, weights for complete longitudinal responses will be produced. These would take into account differential probabilities of attrition after wave 1. They would include those for Waves 1 and 2 or Waves 1, 2, and 3. Weights for other combinations of waves will be developed to support important analyses based on data from those waves. Cross-sectional weights and weights for single year samples waves will also be produced. This brief summary of the weighting strategy can be supplemented by Lynn and Kaminska (2010). (p. 11)
• & Figure 1: Timetable for data collection waves 1 to 4 by quarter (Q) 2009-2012 (p. 12)
• & Most of the data collection uses computer assisted personal interviewing (CAPI). There are several instruments for members in selected households. The structure is similar to the BHPS. One household member completes the household enumeration grid and the household interview, which takes about 15 minutes. Each person aged 16 or older has the individual adult interview (32 minutes) and self-completed questionnaire (8 minutes). Youth aged 10 to 15 are asked to respond to a self-completion questionnaire, which is a paper and pencil instrument. (p. 12)
• & The initial four waves of data collection are face to face, a mode of administration that is typically more costly but more likely to reduce attrition when we are establishing the study. An experiment in the second wave of the Innovation Panel compared groups issued to face-to-face interviewing, vs those initially issued to telephone administration with varied procedures for interviewing outstanding household members faceto-face (Lynn, Uhrig and Burton 2010) . (p. 12)
• & The overall response rate at wave 1 for the general population sample at the household level response rate was 57.2%, which is somewhat below the target rate of 60%. The response rate for the ethnic minority boost sample after screening for eligibility was 56.9%, somewhat above the 55% target rate. These rate s are typical for multi purpose surveys of this sort in the UK. Surveys with a more specific focus of particular relevance for the sampled individuals, e.g. interviewing mothers about their children, tend to get rather higher response rates. The target for household response rate in Wave 2 is greater than 80% and for Wave 3 is greater than 90%. Burton, Laurie and Lynn (2011) provide more information about wave 1 response rates. (p. 13)
• & Table 1. Outline of questionnaire content (p. 14)
• & Table 2: Outline of content of youth self-completion (p. 15)
• & The ability to link Understanding Society survey data with other data sources is a central goal for the study. The added data will greatly enhance its scientific research capacity. (p. 15)

⛔ Weaknesses and caveats

• ! Understanding Society is also noteworthy for its ability to contribute to the understanding of ethnic minorities, which are relatively poorly covered by other longitudinal studies. The UK population can be characterized as having a relatively large number of minority groups, each with small population shares. Study of ethnic variation in general population surveys requires over-representation, using boost samples of minorities in order to achieve sufficient sample size. An ethnic minority boost was also incorporated in the Millennium Cohort Study (Dex and Joshi 2005). (p. 8)