This project set out to utilise a rare quantitative source, the 1994 National Survey of Ethnic Minorities (NSEM), to inform on specific aspects of the case for the proposed religion question in the 2001 Census, focussing on the South Asian population. There were three specific aims. First, to explore the potential of an inferential method of quantifying religion (using some measure of country of origin as a proxy) as an alternative strategy to meeting the needs for data on religion. Second, to explore the ways in which religious sub-groups of the South Asian population differed from each other in substantive ways, and the extent to which this reveals new aspects of heterogeneity over and above that provided by ethnic group. Third, to use the findings of NSEM analysis to address questions concerning the way in which a religion variable might be incorporated into census outputs.

Based solely on ethnic group:

• For Pakistanis and Bangladeshis (just over a third of the South Asian population in Britain), ethnic group provides a statistically reliable basis for inferring religion at individual level and consequently at any required spatial aggregation.
• For those of Indian origin, religious affiliation can only be estimated with real confidence at the national level, due to evidence of substantial areal variation (e.g. Indian Muslims, just 7% of Indians nationally, make up more than a quarter of the Indian population in the East Midlands. Disaggregation of the 1994 NSEM sample allows some sensitivity to this at regional level, but small numbers mean confidence intervals for these estimates are wide, especially for regions outside the main areas of Indian concentration.
• Given the highly uneven spatial distribution of South Asians by ethnic group, the adequacy of the method varies considerably for different South Asian concentrations, being much less applicable in the South where Indians dominate the Asian population, than in some northern cities like Bradford where Pakistanis are the majority Asian group.
• While the Muslim status of Pakistanis and Bangladeshis has remained consistent over time the religious profile of Indians has varied. Indian religious populations are thus a ‘moving target’ and the accuracy of estimates will be undermined where religious weights are based on a different snap shot of the population to the one to which they are applied (as is the case when the NSEM is used to weight Census counts).

Incorporating other predictors:

Analysis of the NSEM found a number of other census variables that were associated with religious affiliation. Attempts were made to incorporate these as predictors to improve the sensitivity of inferential estimation to spatial variation in the association between religion and ethnicity.

• Weighting of district level estimates using country of birth and social class distributions gave some measure of relative difference in the religious composition of Indian populations.
• At individual level, the statistical associations were not strong enough to enable development of a multivariate model for predicting religion effectively at the individual level.

The suspicion is that without considerably more detailed information on the geographical origins of Indian communities the inferential method will remain insensitive to local level variation in religious composition.

2. DIFFERENCES BETWEEN RELIGIOUS SUB-GROUPS

Analysis of the South Asians sample in the 1994 NSEM revealed substantial differences between religious groups which are only partially captured in the census ethnic group classification. Crucially, the patterns appear to challenge any simplistic interpretation of South Asian ethnic differences in terms of a Muslim:non-Muslim axis. Working with a combined ethno-religious classification, two key observations emerged from the 1994 NSEM:

• Religious sub-groups of the Indian population show considerable differences across a range of demographic, social and economic characteristics, though the ‘ordering’ of groups is rarely consistent across topic areas (e.g. Sikhs appear to fare relatively badly in areas like employment, but emerge as having the best housing conditions).
• In many topics, Indian Muslims show considerable dissimilarity with other South Asian Muslims (i.e. they report characteristics more like other Indians than other Muslims). In some areas, however, (for example in female economic activity), they show obvious similarities with other Muslims.

Both points are particularly evident in the field of economic activity, which formed the basis of a more targeted analysis of the 1994 NSEM data. Key differences in patterns of economic activity remained even after controlling for a range of other individual and area level factors. For example, over and above the well established differences between Indian and Pakistani and Bangladeshi ethnic groups it was observed that:

In general it is argued that most applications of a proposed religion variable are likely to be served best by groupings based on an interaction of ethnic and religious identity. Current proposals for Census Area output include just two tables involving religion, making it impossible to do more than a headcount of the ethno-religious groupings that the Census White Paper and the NSEM analysis have identified as appropriate basis for Asian representation. Since the highly concentrated distribution of the South Asian population in Britain means that tables employing a detailed ethno-religious classification will not have national relevance, a proposed compromise is to generate a set of extra area tables produced as standard output only for selected areas with large Indian populations. On the basis of NSEM research, useful variables tabulate with this classification would include age, economic activity, tenure, and income (if asked) – all of which showed substantial variation across ethno-religious groups.

--- (2000) "Quantifying the Muslim Population in Europe: Conceptual and Data Issues," International Journal of Social Research Methodology, Theory and Practice 3, No.2, 87-101.