The analysis of names from the 2011 Census of Population

Previous research conducted at UCL has demonstrated that a name very often provides an open and accessible statement of the cultural, ethnic and linguistic characteristics of its bearer (e.g. Mateos et al 2011). Additional light may be shed upon these characteristics by parental choice of fore-(given) name, while changing fashions often render forenames a valid indicator of age and other geographic and social characteristics. This information has been used to develop working classifications of names, and they have been successfully used to augment incomplete data records for audit purposes - for example in gauging the success of NHS preventive care initiatives across different ethnic groups. However, these classifications have been developed using incomplete address registers (such as the public version of the Electoral Roll) and telephone directories.

There are a number of shortcomings to the data sources hitherto used in this kind of research that limit the usefulness of the resulting classifications when applied to new datasets:
1. The data sources underlying the classifications provide incomplete and probably biased representations of the population-at-large. For example, public electoral registers do not include (young and immigrant) non-voters or (privacy sensitive) 'opt out' individuals, and public telephone directories provide less than universal coverage and few given names.
2. Commercial classifications of the age profiles of given names are typically restricted to the 16+ age cohorts, and supplementation with ONS birth name data (e.g. www.ons.gov.uk/ons/rel/vsob1/baby-names--england-and-wales/2012/stb-baby-names-2012.html) is error prone because young children may move abroad and immigrants may bring young children with them. Thus these sources do not allow an inclusive snapshot of the population resident in the UK at any specific moment in time.
3. Whilst 'crowd sourced' validation is possible (e.g. www.onomap.org), there is no comprehensive means of comparing predicted and objective (e.g. age) or self-assigned (e.g. ethnicity) characteristics.
4. Little focus has been developed upon refining attempts to classify 'hard to reach' groups, such as Caribbeans, whose ethnicity can likely only be ascertained through subtle associations between forename-surname pairings.
5. The clustering procedure has been largely aspatial, in significant part because of unevenness of geographical coverage and the absence of highly granular location information.

This research will address these shortcomings through use of the best available secondary dataset for developing an enriched classification and conducting sensitivity analysis to refine and improve its universal application across the UK. Individual level Census data will be used in order to develop a classification of given and family names into cultural, ethnic and linguistic groups, by extending the methodology of Mateos et al (2011). Crucially, and for the first time, the results of this classification will be compared to individual and household data on Ethnicity, National Identity, Country of Birth, First and Second Language Spoken and Nationality. This will make it possible to investigate the causes of apparent errors in the classification, and to identify the small geographic areas in which they are concentrated. Through an iterative procedure, secure online facilities will be used to improve the classification in the light of these results.

Comparison of household and individual classification results with self assignments in terms of Census measures of identity will make it possible to make the classification tool sensitive to indicators of cultural assimilation, whether through inter-marriage or duration of residence, at scales from the local to the national.

The 'surname regions' will also be used to add regional context to the 2011 ONS Output Area Classification.

The UK is becoming an increasingly diverse society in ethnic terms, and the evidence of the 2011 Census is that ethnic minorities are becoming increasingly dispersed throughout the UK. For the time being, the 2011 Census of Population provides a very valuable snapshot of the size and detailed geography of ethnic minority populations, but it is one that will not be refreshed at this level of granularity for nearly 10 years, and only then if there is a 2021 Census. If the next 10 years see similarly rapid population changes to those of 2001-11 (e.g. of Romanian immigrants), ONS will find the classifier of considerable use in coding up administrative data sources, such as NHS records, in order to provide detailed analysis of the local and regional effects of demographic change. This is a primary motivation for the ONS contribution to this research.

A great strength of the classification will be that it can be run 'on site' at any location where a dataset containing personal names is stored. The software will be written so as to provide aggregate as well as individual estimates of ethnicity, and by suppressing the latter option it is perfectly possible to allow only aggregate results that do not create disclosure risks. As such, it is will prove useful in a range of settings. For example, crime statistics and police interventions may be coded by ethnicity; higher education institutions will be able to monitor the effectiveness of their constituent divisions in implementing widening participation initiatives; and NHS trusts will be able to profile users of services such as breast screening, A&E or GP referrals. Indeed, any organisation that has legitimate access to individual names records will be able to undertake audits of the extent to which their activities draw upon local labour markets or serve local (or broader) communities.

Although calibrated on UK data, the names classification is likely to be of wider geographical use, for example by researchers and government bodies with access to administrative datasets elsewhere in the EU.

The geodemographics industry will benefit from the research. Value added reseller CACI already uses the UCL names classification in its ACORN system, and so the classification produced in this research is likely to improve the discriminatory power of future versions of the classification. The software tool will be of value, therefore, in updating geodemographic classifications over time. In a similar way, the retail industry will benefit by becoming better able to differentiate between its customers and to better tailor their offerings to them.

In the much longer term, the research will have positive impacts upon amateur genealogy. 2011 Census data will only be released in 2111, and will be error prone with regard to names because of typographic errors for online Census submissions and OCR errors on forms completed by hand. The results of the data cleaning that will be a necessary part of this research will be retained alongside Census records by the ONS, to the tangible benefit of future generations.