VICTORIAN PANEL STUDY

Transcription

1 1 VICTORIAN PANEL STUDY A pilot project funded by the Economic and Social Research Council Professor Kevin Schürer, Dr Christine Jones, Dr Alasdair Crockett UK Data Archive paper for the BSPS conference, Canterbury, September 2005 Introduction We are now 14 months into an 18 month pilot study designed to explore the feasibility of a Full Victorian Panel Study in the future. This paper will not attempt to summarise in general terms the Pilot VPS, since such a summary has already been published in the BSPS newsletter 85 (June 2005), in Local Population Studies 73 (Autumn 2004) 73-7, and in the newsletter of the Family and Community History Research Society (March 2005). Instead it focuses on two specific aspects of VPS missing data and sample design. User Requirements The main requirements of the VPS data set by the research community are Large sample size in particular the number of individuals who are traceable from 1851 to 1901 and who can be linked to their descendants, a concept analogous to the Constant Panel size of contemporary longitudinal household surveys. The successful linking of individuals over time will be of great importance to the richness of a Full VPS, since this provides the longitudinal aspect. Many important research questions can only be answered if the data track substantial numbers of individuals (and their descendants) over time, as opposed to having discrete samples at every ten year census

2 2 point. For example a Full VPS database would allow definitive investigation of intergenerational occupational and social mobility, rates and patterns of mortality, fertility, nuptiality and migration, or occupational studies. Household information the usefulness of the data will be greatly enhanced if full household information (as opposed to isolated individual information) exists. Thus gathering information on whole households needs to be part of the VPS sample design. Examples of the research questions that can be addressed are marriage markets, and the patterns and change over time of household characteristics, including domestic service. Four types of potential usage of the Full VPS data by the research community have been identified. The population of Great Britain it is important to create a sample that is representative of the British population both cross-sectionally at each census year from 1851 to 1901 and in longitudinal terms over the 1851 to 1901 period. However, in the longitudinal sense, no sample can be representative of the changing British population between 1851 and 1901 because of immigration. In technical terms, the VPS sample will aim to be representative of how the British population, as it was constituted at 1851, changed over the period to National Populations some researchers will have interests that focus on only one or two of the countries of Scotland, England and/or Wales. The Full VPS, based on the British population can be used to analyse change within a constituent country via the application of weights.

3 3 Regional and Local Populations for local and regional historians, there is an interest in tracking changes in a given area over time. Even if the 1851 VPS sample is geographically clustered, this clustering will be diluted over time due to migration. This will make the main VPS sample less useful for studying local communities over the 1851 to 1901 period, particularly in areas of in-migration. To serve this research community, it would be desirable to include additional samples that are explicitly geographically delineated. In these additional samples the rule of observation and refreshment would simply be residence in the area at a given census year, with origin and destination noted for those who enter and exit observation due to migration. Minority Populations: immigrant, ethnic or occupational sub-groups since all immigrants will need to be traced for methodological reasons [of which more later] the creation of immigrant samples that follow a randomly selected booster sample of immigrants will have fewer additional cost and time implications for the Full VPS than might be expected. Data Requirements The Pilot VPS has done considerable work assembling, facilitating and checking the quality of available and potentially available machine readable data sources. The requirements can be summarised as follows Full databases of individual census returns for the total population of Britain from 1851 through to 1901 Full databases of Civil Registration data [births, deaths and marriages] for the 1851 to 1901 period

4 4 National and local [Registration District] level summary statistics for various population characteristics. These will be used to add community level data to the individual and household level data in the main VPS sample and for the calculation of post-stratification weights. Clearly not all these data sources are currently available in machine readable format and some of them are controlled by commercial organisations, resulting in financial restrictions to their availability to the research community. Missing Data A principal decision for the Full VPS project will be whether to use the existing two per cent sample from the 1851 census created by Michael Anderson [UK Data Archive study number 1316] or, if a full machine-readable database of the 1851 census were available, to make a fresh household based simple random sample. Initially it was thought that missing data could be a substantial methodological issue for VPS, but work during the pilot study has revealed that in the 1851 Anderson sample the problem is not as widespread as feared. There are two aspects of missing data to be considered. Unit nonresponse, may be thought of as a household missing from the enumeration. An example from 1851 would be the enumeration of vessels, which has not survived. Item non-response is where census returns are incomplete or illegible. This is analogous to the situation in contemporary social surveys in which someone agrees to take part in the survey but then refuses or forgets to answer specific questions. Alasdair Crockett has run tests on completeness of name information and whether partial or complete absence of name information is random with respect to other socio-economic

5 5 demographic and geographic variables. Completeness of name information was chosen as it is obviously the biggest factor when it comes to record linkage success or failure, although age and birthplace information is also important. Before doing any record linage it is possible to anticipate attrition biases emerging in the VPS longitudinal sample if the completeness of name information correlates with age, sex, household type, occupational group or place of enumeration. The results are encouraging. It will be virtually impossible to link people where name information is completely missing, such as when only initials are given in the enumeration of some institutions. Those for whom name information is partial, mainly because the transcriber was doubtful about the correct reading of the name, can often be linked using other supporting information from the age and birthplace fields, or from other household members. Considering both personal and surnames together 93.5% of the 1851 sample have complete information, a further 5.7% have partial information and only 0.8% have no information. Because of the very small numbers involved, the effect of being unable to link those with no name information will be slight, but the effect of failing to link those with only partial name information could bias the VPS sample. However, the calculations show that the profile of the partials is very similar to that of the completes in terms of age, sex occupation, household type, position in household, and country of birth. There are some differences of absents, and these are largely intuitive: slight biases towards being younger and male (a pattern seen in contemporary surveys), in unusual or unknown forms of household, unknown status within the household, and being patients or lunatics. In the course of this analysis it was thought at one stage that there was a considerable bias by occupation and by county of enumeration, the absents being very high among people working

6 6 in textile fabrics [almost 19%] and people enumerated in the counties of Cheshire and Lancashire [over 5% each]. However, on closer inspection it was found that this was due to just 3 enumeration districts Little Leigh [Cheshire, HO107/2165] Spotland and Wardleworth [both Lancashire, HO107/2246-7] with a total of 2,366 records for which name and birthplace information was entirely absent, but curiously occupation data were present, and they were overwhelmingly textile workers. Checking the documentation relating to SN1316 it was found that Michael Anderson s team of transcribers in the 1970s had been unable to read the photocopies of these CEBs. A visit to the FRC shed further light on the problem. The films of piece numbers HO107/1466 to 2531 had been found to be of poor quality and were replaced in The new films were checked and were found to be legible, though still not easy to read. Whether or not VPS uses this sample, we will be recommending that these 3 EDs are retranscribed to enhance the value of SN1316. Once this work is done there will no longer be a bias of name information by occupational type or by county of enumeration. Once similar work is done to investigate any possible biases by birthplace information the problem of missing data will at least be measurable and compensating methodologies can be devised. What has yet to be examined is whether there are biases in consistency of name information (between two census years), since this will also impact upon the success of definitive or probable record linkage. Selection of the 1851 sample The size of the 1851 sample is also an issue of fundamental importance. The required 1851 sample size rests on the proportion of individuals that via linkage can be maintained continually in observation from 1851 to 1901 (or until death). In turn this rests upon the rate of

7 7 attrition (the cumulative magnitude of failed record linkage) between 1851 and However, one advantage of creating an historical survey is that the 1851 sample size does not need to be predetermined. If the estimates of attrition rates by the Pilot VPS are too low, the Full VPS can easily augment the 1851 sample (following the same sampling rules) in order to obtain a constant sample of adequate size. Such flexibility is in contrast to contemporary longitudinal surveys when the sample size than can be observed for the duration of the survey is inevitably fixed by the size of the initial survey. If the VPS is thought of as a long-term historical resource that might have several waves of funding, the main VPS sample could be augmented at a future date, provided the additional sampling methodology was compatible to 1901 samples The 1861 to 1901 samples are a product of the 1851 sample and the rules of observation and refreshment that determine who is included in the main VPS sample from 1861 onwards. Rules of observation relate to which persons included in the previous census should be included/observed at the following census. Rules of refreshment relate to persons who were not observed in the previous census but who are entered into the VPS sample at the next census. Both observation and refreshment rules are required to keep the main VPS sample broadly representative of the British population over the 1851 to 1901 period. The VPS defines three types of sample member, original, additional and temporary, as follows: OSM Original Sample Member everyone in the households selected in the main 1851 VPS sample COSM Child of Original Sample Member all children of OSMs born after 1851 census [those co-residing children born before the census will be OSMs by definition]

8 8 ASM Additional Sample Member sample members selected by refreshment rules CASM Child of Additional Sample Member child of sample member selected by refreshment rules TSM Temporary Sample Member anyone who is not an OSM or COSM but lives in the same household in a subsequent census as one or more of the OSMs or COSMs [spouse or other relative, servant, lodger, visitor] The rules applied to VPS, which are closely analogous to those used by the British Household Panel Survey [BHPS], are the most effective means of keeping the VPS broadly representative of the British population between 1851 and 1901, with statistical weighting used to correct for any biases that might arise. Rule 1 attempted record linkage for all OSMs and COSMs to be carried out from 1851 to 1901 to both Census and Civil Registration data. The implications of this are that all linked OSMs and COSMs remain in observation in the main VPS sample. Ceasing of observation is either due to known death [record linkage to civil registration death data], or an indistinguishable mixture of emigration and attrition caused by failed linkage to either census data [in which case the OSM or COSM is still alive] or to civil registration death data [in which case the OSM or COSM is dead]. Rule 2 attempted linkage of all ASMs and CASMs is to be carried out from their first census of selection onwards to 1901 to both Census and Civil Registration data. The implications are the same as for rule 1. Rule 3 attempted linkage of all TSMs is to be carried out from their first year of inclusion to both Census and Civil Registration data. However, TSMs are only

9 9 maintained in the main VPS sample if they continue to reside in the same household as one or more OSMs, COSMs, ASMs or CASMs. Other reasons for ceasing observation are death, emigration or failed record linkage. Refreshment of the sample via ASMs is essential to keep the VPS representative of the growing and changing immigrant population over the 1851 to 1901 period. ASMs could also be used to correct for any emerging mismatch between the VPS sample and the British population. Rule 4 ASMs to compensate for immigration the estimated new immigrant population [those arrived since the previous census] will be sampled at each census [1861 to 1901] to match the VPS sampling fraction of the rest of the population for each census year. Estimation of the population that has immigrated since the previous census is practical but resource intensive, requiring 100 per cent forward and backward linkage to Census and Civil Registration data of all overseas born persons for each census year. In this way, one can arrive at estimates of both absolute numbers and individual identity of persons who have immigrated since the previous census, and hence a sub-population from which to sample ASMs to compensate for immigration. This is the only viable method for keeping the VPS sample representative of the immigrant population over the 1851 to 1901 period. The method will be to start in 1901 and working back to 1861 [or 1851 if a 100% database is available] attempt to back link all persons recorded as being born overseas to obtain the earliest census at which they were observed. Allowing for some contamination due to failed record linkage, when the immigrant was in fact in Britain at an earlier census year but couldn t be traced back to it, this method provides both total numbers and individual identities of

10 10 persons arriving in Britain between each census. This information can be used to randomly sample ASMs from new immigrants at the appropriate sampling fraction and thereby keep the main VPS sample representative of the changing immigrant population over the 1851 to 1901 period. Although this method allows the greatest possible accuracy, the time/expense of attempting to back-link all overseas born people will be considerable Rule 5 Exit/ceasing to be observed for all types of VPS sample members exit will occur due to death [confirmed by linkage to Civil Registration death data], emigration or attrition [failed linkage to census data/person alive; failed linkage to Civil Registration death data/person dead]. For TSMs exit will also occur when the TSM is no longer resident in the same household as one or more OSMs, COSMs, ASMs or CASMs. Conclusion Since this is very much research in progress there are no conclusions, as such. However, we are coming to the conclusion that there is not going to be a single VPS, rather that we will need to create different versions for different user groups; a series of community profiles for which the geography is held constant. We see this as the only way to reconcile the disparate needs of research community. Otherwise, if we try to please everyone we will end up pleasing no one.