Ethnic Residential Segregation in Northern Ireland

by Michael A. Poole and Paul Doherty
Published by the University of Ulster, Coleraine 1996
ISBN 1 85923 048 2
Paperback 299pp £10.00

Copies are available in bookshops or, by post, from:

T: (01265) 324666 or 324165
F: (01265) 324917
E: mp.shortt@ulst.ac.uk

This material is copyright of the Centre for the Study of Conflict and the author(s) and is included on the CAIN web site with the permission of the publisher. Reproduction or redistribution for commercial purposes is not permitted.

The Centre for the Study of Conflict is a research centre based in the University of Ulster. Its main work is the promotion and encouragement of research on the community conflict and to this end it concentrates on practical issues to do with institutional and community structures and change. It publishes papers and books arising out of this work including: a series of research papers particularly designed to make available research data and reports; a series of Majority-Minority reports; and a series of occasional papers by distinguished academics in the field of conflict.

This new publication by Paul Doherty and Michael Poole is the first of a two volume study of ethnic residential segregation in Northern Ireland. The first volume deals specifically with the internal geography of Belfast. The second deals with the urban space of the province as a whole and thus also includes Belfast, but this time with only selected summary information and set in the context of all other towns. The authors have collected data on residential segregation for many years and most of the material presented here is new or reanalysed.

The Centre is also publishing a number of other new reports at this time on topics such as Sport and Community Relations, Peer Mediation in Primary Schools, Parades, Education in a Divided Society, the Role of the Police and the Quaker Peace Education Project. A full list is printed at the back of this volume.

Seamus Dunn
March 1996.

Return to publication contents

1.1 Context and justification

'There is widespread belief among Americans, regardless of their own attitudes toward racial integration, that stable racially integrated neighbourhoods are a rare phenomenon. Even staunch supporters of integrated living who currently live in integrated neighbourhoods believe that theirs is one of the few such neighbourhoods in the country. Our study has uncovered information that contradicts these beliefs. We find that integrated neighbourhoods are much more common than most Americans think' (Bradburn et al., 1971).

There is much in these rather startling observations on the United States racial situation which is applicable to the division between Catholic and Protestant in Northern Ireland. There appears to be the same general belief here in the virtual ubiquitousness of intense religious residential segregation, so, when individuals know that they themselves live in a neighbourhood or a town without such segregation, there is a tendency to assume that it is exceptional. Only after talking to people from many different places does the suspicion begin to emerge that such environments are not exceptional after all, and that quantitative evidence might show that religious integration in the province's housing is, in fact, remarkably common. It is this suspicion, and the anomaly behind it, which provided the original stimulus over twenty years ago for what became the long-running research programme reported on in this volume.

The report has been written in conjunction with our previous publication Ethnic Residential Segregation in Belfast (Doherty and Poole, 1995), and hence much of what was said in the introduction to that volume is equally germane here. It is therefore appropriate to commence by summarising some of the main points of the introduction to that work before dealing with issues which are more specific to this second publication. In the final part of this introduction, we indicate the principal themes to be developed in the present volume, itemising them as a series of eight objectives.

Both publications can be viewed against the background of a paradox: on the one hand, religiously identified territory plays a fundamental role in the life of Northern Ireland people, but on the other hand it is our belief that comparatively little is known about the intensity and extent of religious residential segregation in the province. Certainly, that which is known tends to have a strongly Belfast focus. We are therefore seeking to fill substantial gaps in knowledge, unravel complexities, and perhaps to expose a few myths.

Our work is situated in the social geographic tradition of Estyn Evans, Emrys Jones, and Fred Boal, who have all contributed so significantly to research into religious residential segregation in Northern Ireland, particularly in Belfast. We would also place our research in the wider academic context of sociological and geographical studies which concentrate on the measurement of the intensity of residential segregation. An especially strong effort is made in relation to this wider context in the present volume because the comparative urban analysis which is focused upon here has only one antecedent within Northern Ireland itself and, indeed, very few within Britain or Ireland. This explains the deliberate attempt to set, in particular, the methodological framework for the analysis in international context. Specifically, this means primarily United States research, for American writers have so dominated the world of segregation analysis, as the reader edited by Peach (1975) and the reviews by James and Taeuber (1985) and Massey (1985) have clearly demonstrated.

Early work in Northern Ireland's social geographic tradition adopted a twin-pronged approach to justification. This suggested that segregation is both an interesting topic and also one that is fundamentally important (Evans, 1944; Jones, 1956, 1960a). However, following the outbreak of the political 'troubles' in the late 1960s, the view that segregation is important has tended to dominate, particularly because of the relationship between segregation and conflict. It has been demonstrated that residential segregation responds to violence via, for example, refugee movement (Darby and Morris, 1974), and also that it has an impact on the geographical location of violence (Poole, 1983a, 1990). The significance of these issues has been succinctly stated by Darby (1976: 25):

Just as one cannot hope to understand the Northern Ireland conflict without an acquaintance with its history, it is impossible to appreciate its pervasiveness without some knowledge of the background, extent and effect of residential segregation between Catholics and Protestants. This is both the cause and consequence of the province's history of turbulence.'

This relationship between segregation and violent conflict provides a powerful justification for the empirical analysis of religious residential segregation in Northern Ireland. However, it is only part of the general policy relevance of the residential segregation issue, as illustrated in the American context by the contributors to the volume edited by Goering (1986). This United States literature has given top priority to 'racial' segregation, and, from the pioneering research of Taeuber and Taeuber (1965) to the prolific recent work of Douglas Massey and his colleagues (Massey, 1990; Massey, Condran and Denton, 1987; Massey and Denton, 1989, 1993; Massey, Gross and Eggers, 1991), this has emphasised the issues of inequality, discrimination and the denial of human rights.

This human rights debate has provided much less of the rationale for segregation analysis in Northern Ireland, largely because the only allegations of systematic religious discrimination in housing concerned a small number of the pre-1973 authorities operating specifically in the public housing sector (Whyte, 1983, 18-21; Birrell, 1994, 110). On the other hand, there are wider policy issues, and these are addressed both by McKittrick (1994: 40-46), whose analysis is distinctly regretful of the drift towards increased segregation, and by Boal (1976: 74-75). The latter makes clear his disapproval of simplistic universal integrationist policies, for, instead, he emphasises the policy relevance of identifying whether specific instances of segregation are voluntary or involuntary, and whether they are a by-product of class segregation. Above all, as was argued in our companion volume,

'What is more important from a policy perspective is whether there is a sufficient choice of a wide range of social environments to make a voluntary decision possible' (Doherty and Poole, 1995: 13).

This applies both to decisions to live in segregated environments and those to live in mixed ones. Our research in both volumes is therefore driven by the policy view that people have a right to meaningful choice, as well as by the salience of the established relationship between segregation and conflict.

Ethnic Residential Segregation in Belfast concentrated its analysis on Northern Ireland's primate city, defined in built-up area terms. This single-city focus placed it in the major tradition of residential segregation research, both inside the province and further afield, for this tradition takes one urban settlement and subjects its internal geography to segregation measurement. There is, however, a second spatial scale-level at which residential segregation analysis can be conducted, and this involves studying the geographical variability between urban settlements. Such a study is essentially a comparison of the between-town spatial differences in the within-town neighbourhood variation which constitutes residential segregation. It therefore has a rather more complex research design than does the single-city focus, and it is such a complex comparative urban approach which is adopted in this volume.

This two-level design is stimulated partly by the exploratory multi-settlement analysis of Poole (1982), which provided preliminary evidence that Northern Ireland towns vary radically in the intensity of their religious residential segregation. However, a further stimulus is provided by the American comparative urban research on the segregation of blacks from whites. For example, Massey and Denton (1987: 823) concluded that the most likely way for segregation to be relaxed would be via black residential mobility to small and medium-sized cities where few blacks currently live. This view has more recently been endorsed by Farley and Frey (1994: 41), but they add that the newer cities of the South and West seem to offer a better prospect for relaxing segregation than the older metropolitan areas of the North and the Old South.

The combination of the local and the international evidence suggests that there can be a substantial geographical variation between towns in the intensity and trend of segregation. This implies a need to avoid falling into the trap of assuming that all Northern Irelands towns are alike - of assuming, for example, that, because Belfast has neighbourhoods which could perhaps be described as ghettos, then all Northern Ireland towns must have ghettos. It also implies a requirement for fresh theoretical perspectives to embrace the empirical reality of inter-urban variation, for a comprehensive theory of segregation must account for the contrasts of multi-settlement space. This last requirement is, we propose, one of the appropriate responses to Boal's (1987: 120) plea for "the development of sound theoretical perspectives for the study of segregation, judiciously tempered by appropriate empirical material". A further justification for this comparative urban volume is therefore that it makes a contribution towards this call for theoretical development by highlighting some of the complications which improved theory must be capable of handling.

There is a further call for research to which this volume is responding, and that concerns the size of towns included in the comparative analysis. Northern Ireland is, in the international context, a region of very modest proportions. It has a land-area about the same as the United States' third smallest state of Connecticut and a population only half as big (Hunter, 1994: 1381, 1399). Therefore, the range of settlement sizes in this Ulster analysis is quite different from that characteristic of American comparative urban analyses of segregation. For example, the important series of segregation papers by Massey on the 1980 census used a set of 60 metropolitan areas, of which the smallest was Corpus Christi, with a population of 326,000 (Massey and Denton, 1987; U. S. Bureau, 1981, 18-20). Northern Ireland only has one town above this threshold.

The only American studies remotely comparable with our Northern Ireland investigation in terms of urban size are those of Hwang and Murdock (1983) on the 1980 census in Texas and of Murdock et al. (1994) in a follow-up 1990 analysis. Of 348 cities included in these studies, 55 were outside metropolitan areas, and a population threshold of 10,000 is referred to. The authors emphasise the original contribution they are making by extending residential segregation measurement to nonmetropolitan settlements, and they call specifically in both papers for more nonmetropolitan segregation research. With an even higher proportion of nonmetropolitan settlements than in the Texas study, as well as a much lower population threshold, our Northern Ireland analysis in this volume is clearly filling a gap in the international research on segregation.

There is a final methodological respect in which we depart from the mainstream tradition in segregation research, and that is in relation to the choice of indices. Northern Ireland analysts have generally followed the conventions of international segregation research by using whatever indices were normal at the time. For example, the dissimilarity index alone was applied to Belfast by Poole and Boal (1973) and in the multi-settlement study by Poole (1982). Later, in response to changing norms, Doherty (1990a) and Keane (1990) employed the isolation index in conjunction with the dissimilarity index, and it is this combination which was used in our companion Belfast publication.

This second volume will also use this combination, but it will be supplemented by two other indices. One is the replacement index, which is simply an elaboration of the dissimilarity measure. It was applied by Farley and Taeuber in 1968, but has rarely been used since in empirical work. The other, labelled the linear dominance index, has been developed especially for this volume, though it expresses a concept described verbally as long ago as 1973 by Poole and Boal. The proliferation of indices arises from the recognition that segregation is one of those concepts which sounds uncontroversially unambiguous when you first think about it: you might consider that you should just be able to define it and measure it, but this is deceptive. A useful - albeit not very complex - analogy is the problem of asking how big a person is. It might appear exceedingly simple to measure 'bigness', but what does the questioner really mean? Is it weight that defines bigness, or is it height - or, indeed, width? You suddenly realise that bigness is itself a rather vague idea.

Segregation is no different, for it, too, is rather a vague concept. Like bigness, segregation must be carefully defined before it can be measured, and again, like bigness, there are alternative ways in which it can be defined. The result is the modern recognition that segregation is a multidimensional concept (Massey and Denton, 1988) requiring a multiple index approach to measurement. The actual identification of dimensions is still an area for considerable controversy (James and Taeuber, 1985; Stearns and Logan, 1986; Morrill, 1991), and, in fact, our linear dominance index represents both a measure and a dimension not incorporated into any of these rival schemes. The index is therefore presented here as an innovative contribution to the multi-dimensionality debate, allowing us to try to meet Boal's (1987: 120) requirement "to earth our work in the material and social worlds, by means, if at all possible, of methodologically inventive investigations."

1.2 Objectives

The preceding attempt to contextualise and justify the analysis in this volume has necessarily provided some indication of its scope and coverage, but it should help the reader further to furnish a list of eight basic objectives.

1. To return to first principles in examining alternative conceptualisations of segregation in order to discover whether the battery of alternative indices currently available fully matches the set of concepts underlying current thinking on the subject.

2. To fill any gaps identified in meeting the first objective by making an original contribution to segregation index development, providing a theoretical analysis in the context of the interpretation of more conventional indices.

3. To review the pre-existing literature on ethnic residential segregation in Northern Ireland in order to identify gaps in the research completed to date and in order to set the empirical work in this volume in local context.

4. To explore methodological alternatives for the analysis of residential segregation, not only to justify the decisions taken when undertaking the analysis presented in this volume, but to help target topics for further empirical research both in Northern Ireland and elsewhere.

5. To apply a series of segregation and exposure indices to the description of contemporary ethnic residential segregation in the towns of Northern Ireland, emphasising the implications of the frequency distributions.

6. To map each index of segregation and exposure in Northern Ireland in order to search for evidence of regionalisation in the geographical patterns identified.

7. To examine the contemporary correlates of segregation and exposure in Northern Ireland in an attempt to provide part of the explanation for the variation between towns identified when tabulating and mapping the indices.

8. To provide information on the change and stability in segregation in Northern Ireland during the course of the twentieth century in order to provide a historical explanation to complement the discussion of correlates.

To a large extent - though not entirely - these objectives are each dealt with in a separate chapter, and they are presented here in the sequence in which they are addressed in the rest of the volume. They are also repeated in section 9.2 of the conclusion, where they provide the framework for a summary of the findings of the preceding chapters - a summary which, in turn, furnishes the point of departure for the final overview of the fundamental theme of the volume.

Return to publication contents

9.1 Preamble
At the risk of being repetitious, it will be convenient to begin this conclusion by summarising the main theoretical and empirical findings of the study. This recapitulation lays the foundations for the main part of this closing chapter, which addresses three tasks. Firstly, it develops a classification of Northern Ireland towns on the basis of the segregation and exposure dimensions explored in this volume. This, in turn, will help to shed light on the superficially simple issue of discovering how much ethnic residential segregation there is in the province, and it will also help to demonstrate how strongly regionalised the province's urban settlements are. Secondly, the conclusion explores the possible chain of causality which interlinks the five indices used in the empirical analysis. It does this by focusing on the theme of change and stability from Chapter Eight and trying to make inferences from these findings. Finally, an attempt is made to point to some directions for future research, for, despite the lengths to which this two-volume work expanded, it does only represent a superficial brush with the theme of segregation in Northern Ireland. Much work remains to be done, some of it requiring new data, but some simply needing further analysis of information already available.

9.2 Summary of Findings
This recapitulation of the main findings will be developed by repeating and addressing, in turn, the original objectives of the study.

(1) To return to first principles in examining alternative conceptualisations of segregation in order to discover whether the battery of alternative indices currently available fully matches the set of concepts underlying current thinking on the subject.

Segregation analysis in the social sciences deals primarily with the two distinct spheres of the geographical and the occupational. The concern in this volume is with just one type of geographical segregation that in the residential arena - and the focus is on the analysis of pattern rather than process. Even within this context of pattern analysis, there are alternative ways of conceptualising segregation, and the key-issue is defining zero segregation. There are two principal alternatives, each providing a separate benchmark from which to measure the intensity of segregation. One is the minority percentage-share of the population of the town as a whole, and, since this definition of zero segregation implies an even distribution of the minority percentage across all subareas in the city, it involves a conceptualisation of segregation as unevenness.

The second of the two main definitions of zero segregation involves different social groups being found in equal numbers. In a two-group situation, like that of Catholics and Protestants in Northern Ireland, this arises where the 'minority' percentage is 50% in a particular subarea. When this is used as a benchmark, it implies a totally different conceptualisation of segregation from that involved in unevenness. This alternative idea has been here labelled 'dominance' because it measures the extent to which one group is in the majority.

These two concepts of unevenness and dominance are distinct dimensions of segregation and can be distinguished in the segregation index literature. A number of indices, including the well-known dissimilarity index (D), use evenness as their measurement benchmark. Dominance is referred to much more rarely, though it arises in frequency distributions and in choropleth mapping, and there have so far existed only a few unsatisfactory indices for systematically measuring it. A second area of the literature which requires a careful definition of zero segregation is that concerned with desegregation policy, though much more has been written about schools than about housing. The case of school desegregation in the United States and Northern Ireland illustrates both a lack of consensus over the ultimate zero segregation target and the fact that the two conceptualisations of segregation used are again unevenness and dominance.

The multidimensionality of segregation, implied in this discussion, has received much emphasis in the recent index literature. However, while the use of D has kept unevenness at the forefront of attention, dominance has remained a highly neglected concept. Instead, it is exposure which has received much prominence through the use of P*. This is a concept which is regarded as a further dimension of segregation by some writers but as a separate issue by others. Even the latter view it as an important concept, however, and it forms a third conceptual strand to the measurement in this volume, which thus involves unevenness, dominance and exposure.

(2) To fill any gaps identified in meeting the first objective by making an original contribution to segregation index development, providing a theoretical analysis in the context of the interpretation of more conventional indices.

The choice of measures to express the dimensions of unevenness and exposure was straightforwardly conventional, involving the dissimilarity and isolation indices respectively. The replacement index is added because of its interpretative value, but it is modified from its conventional form to make it a standardised replacement index. For measuring exposure, there are two separate isolation indices - one for Catholics, and one for Protestants - and they need to be applied together. In considering whether P* is really a segregation measure, we develop three different ways of interpreting this index.

The main methodological innovation, however, concerns the final conceptual dimension - dominance. The relationship was explored between this concept and the idea of diversity, one element of which also involves measurement from a benchmark of numerical equality. This suggests that indices developed to express diversity could be adopted to measure dominance. The main social science measure of diversity was developed in ecology as the Simpson index, and this is a nonlinear measure which expresses the dominance of social interaction resulting from the dominance of social composition.

This issue of linearity or nonlinearity for measuring deviation from the zero segregation benchmark is a controversial one. However, the merits of the linear dissimilarity index for measuring unevenness were clear, and the desire to maintain consistency with D stimulated an attempt to develop a correspondingly linear measure of dominance, since there appeared no such index in the diversity literature. The nonlinear Simpson index could be justified as a measure of the social interaction consequences of dominance, but the nature of its causes, in terms of population movement, provided a strong theoretical justification for a linear dominance index. Such an index was developed specially for this volume by taking the dissimilarity index as a model and applying its procedures to the dominance concept.

This series of decisions on index choice led to the application of a total of five measures to the Northern Ireland segregation data. These were the dissimilarity and standardised replacement indices (D and 2R) to measure unevenness, the linear dominance index (LD) to express dominance, and the Catholic and Protestant isolation indices (cP*c and pP*p) to measure exposure.

(3) To review the pre-existing literature on ethnic residential segregation in Northern Ireland in order to identify gaps in the research completed to date and in order to set the empirical work in this volume in local context.

Belfast has continued to dominate the limited volume of work published on this type of segregation, and there is still very little information available on the smaller towns of the province. This has not prevented some authors from offering sweeping statements on the frequency of segregation, but the overall standard of writing has improved in the last ten years. Nonquantitative commentaries are more cautious and more specific, and a higher proportion of contributions build on quantitative analysis either of their own or of other researchers. Some authors still write as if Belfast were typical of Northern Ireland, but this is getting less common as the awareness increases that the province is subject to tremendous internal spatial diversity. In particular, the quantitative evidence suggests that segregation is much more widespread in the Belfast core than in the remoter periphery of Northern Ireland.

These comments are based on evidence of actual ethnic composition, but there has also accumulated a body of quantitative research on perceived ethnic environments since 1968. The findings suggest, with a high degree of consistency, that there has been very little change in the proportion of Northern Ireland people living in segregated and in mixed environments. Moreover, there is a major differential between Belfast and most of the rest of the province. This regional differential, along with the prevalence of both segregated and mixed ethnic environments, are two common threads running through both the quantitative evidence on actual segregation and the survey data on perception.

In interpreting both types of evidence, it should be borne in mind that the conceptualisation of segregation employed in the province-wide studies is invariably that of dominance. Since this is fundamentally different from the idea of unevenness, it is difficult to compare the findings with the research on individual towns, which only used D and P*. Therefore, the decision in this volume to apply unevenness, dominance and exposure measures will make the results from individual towns much more comparable to the province-wide findings reported.

(4) To explore methodological alternatives for the analysis of residential segregation, not only to justify the decisions taken when undertaking the analysis presented in this volume, but to help target topics for further empirical research both in Northern Ireland and elsewhere.

The focus in this volume is on individual towns, partly for its own sake and partly to construct regional patterns from basic building blocks. This focus makes the study conform to the main convention of segregation analysis worldwide, and it acknowledges the significance of the individual housing market in the segregation process. There is still a choice of three alternative conceptualisations of what constitutes a town - the political unit, the built-up area, and the functional city. American analysts have focused on the political and functional cities, while their British counterparts have tended to study local authority units. This volume will, however, continue the tradition of Northern Ireland segregation analysis of employing the built-up area because of both its perceptual clarity and the availability of data. The unfortunate implication is that placing our results in international context has to be done with considerable caution.

A second decision on the input data concerns the scale-level issue, involving the choice of the size of subareas whose ethnic composition is being measured within each town. The criterion selected for defining scale was the number of households, and, to ensure meaningful comparison, the same spatial mesh was used in each town. Specifically, the average size of subareas was made a constant of 200 households, and significant scale-mixing was avoided by allowing the size of individual subareas to vary only in the range from 120 to 300 households. Other scale-level issues were judged to be theoretically important, but were not addressed empirically in this volume to avoid too much methodological complexity.

The decision on the specific scale-level has implications for the minimum size of town that can realistically be subjected to segregation analysis. We estimate that this is a threshold of 1000 households, and it is the entire set of 39 Northern Ireland towns above this threshold that is the focus of attention in this volume.

The data for the analysis of 38 of these towns is taken from the 100 metre grid-square version of the Small Area Statistics from the 1981 census of population. That for the 39th, Belfast, has had to be estimated from the 1991 enumeration district data. This is an obvious weakness of the Belfast data, as is the fact that the boundaries are fixed lines that we must accept, even though they often fail to coincide with real social boundaries. In the 38 other towns, the 100 metre squares were used as building blocks to create customised subareas which, to a much greater degree, could be made to approximate to homogeneous neighbourhoods on the ground.

The census classifies people according to religious affiliation, and it is this that we have used to approximate to ethnic identity. More specifically, people stated to be Catholic have been allocated this ethnic identity, while all other categories based on stated religious denominations have been grouped together as a Protestant identity. That leaves people failing to state their religion, together with, in 1991, those who claim to have no religious affiliation and both these groups are omitted from the analysis.

(5) To apply a series of segregation and exposure indices to the description of contemporary ethnic residential segregation in the towns of Northern Ireland, emphasising the implications of the frequency distributions.

The frequency distribution of D is strongly positively skewed, which implies that, contrary to much opinion about the prevalence of this type of segregation in Northern Ireland, most towns are only very weakly segregated. For example, 31 out of the 39 towns have a D below the theoretical midpoint of 50. On the other hand, the five towns which differ markedly from the rest by having an index over 70 do include the two largest towns in the province, Belfast and Derry.

The second measure of unevenness, 2R, is a simple function of D and of the deviation of a town's overall Catholic population-share (P) from 50%. This means that 2R is only a little less than D in towns with a very mixed overall ethnic composition, but 2R is very much less than D in towns very dominated by a single ethnic group - usually, in practice, by Protestants, since very Protestant towns are much commoner than very Catholic ones. The result is that 2R always gives an impression of lower segregation than does D - often much less - but the two frequency distribution shapes are rather similar. For direct interpretation, it is R which is relevant, for it measures the minimum proportion of a town's total population that would have to move to achieve zero segregation. It follows that, in terms of sheer simplistic numbers, this could easily be achieved in the large number of very Protestant towns. Even the median R is only 8.7%.

The linear dominance index (LD) is strongly influenced by both D and P. Thus a value of P close to either zero or 100% - i.e. a very Protestant or a very Catholic town - will yield a high LD, and the same effect will result from a high degree of unevenness. In practice, the very high LD values in Northern Ireland result from Catholic population-shares close to zero rather than from either very high P or very high D indices. There are so many of these towns with a high LD that, in complete contrast to D and 2R, the frequency distribution is negatively skewed. Thus 29 of the 39 towns have a linear dominance index above the theoretical midpoint of 50.

The result is that, again in total contrast to the conclusion on unevenness, examination of dominance shows that the majority of Northern Ireland towns are strongly segregated. Moreover, the towns with the less intense levels of dominance are almost all small towns, so the urban population of the province is heavily concentrated into those towns that are strongly segregated. This all confirms the theoretical reasoning that LD is measuring a very different dimension of segregation from unevenness, and the final evidence of this is the almost zero correlation between LD and both unevenness indices.

All three of the true segregation indices presented thus had strong skewness in their frequency distribution, but the Catholic isolation index (cP*c) had a much more symmetrical shape. In fact, 24 of the 39 towns had indices below 50, indicating that the average Catholic lives in a majority Protestant environment in these 24. The Protestant isolation index (P*p), on the other hand, has strong negative skewness, and, in no less than 33 towns, the average Protestant lives in a majority Protestant neighbourhood. This primarily reflects the fact that Northern Ireland has far more towns with Protestant majorities than with Catholic ones and that unevenness is insufficiently widespread to offset this effect.

(6) To map each index of segregation and exposure in Northern Ireland in order to search for evidence of regionalisation in the geographical patterns identified.

Mapping the D values reveals little systematic relationship to housing and labour market areas, but there are some other elements of spatial clustering present. In the western and southern Border belt, segregation in all seven towns is either high or moderate. Moreover, in an area focused on Belfast, it is uniformly weak except in Belfast itself, which has the highest value in the whole province.

The quite high correlation between D and 2R means that the maps of these two indices are very similar. However, there is an even stronger cluster of very low 2R values in the Belfast Lough region than there is of very low D values. And high 2R values are found especially in a specifically southern Border belt. The first region has very low Catholic population-shares as well as low D values, while the second has very mixed populations overall as well as high or moderate dissimilarity.

The difference between the two dimensions of unevenness and dominance is emphasised by mapping LD, for its spatial distribution shows a clearer pattern. This is a dichotomisation of Northern Ireland space between the core Travel to Work Areas of Belfast and Craigavon, with high dominance values, and the remainder of the province, which has substantially lower values. This suggests a labour catchment area hypothesis for the explanation of LD.

Mapping the Catholic isolation index shows that the very highest values are primarily found in the same western and southern Border strip revealed earlier as the location of most of the high and moderate dissimilarity indices. The slightly less high Catholic indices are located in such a way that the two groups of towns are highly regionalised into the west and south of the province. This means that the lower Catholic isolation indices are equally regionalised into the north-east - both the Belfast region and the County Antrim pocket to the north. The strong negative correlation between the two isolation indices means that the Protestant map is almost the mirror-image of the Catholic one. The main exceptions are caused by the four towns of Belfast, Lurgan, Portadown and Armagh, which have the unique distinction of scoring highly on both indices.

(7) To examine the contemporary correlates of segregation and exposure in Northern Ireland in an attempt to provide part of the explanation for the variation between towns identified when tabulating and mapping the indices.

Only a narrow range of environmental variables have been experimented with as possible correlates of the segregation and exposure indices, and these all involve demographic measures of minority-size and urban size. When trying to explain inter-urban variation in D, the most successful variable was the absolute size of the Catholic population. No attempt was made to correlate this type of independent variable against the standardised replacement index because the latter is simply a function of D and P. It is therefore interpreted as a consequence of D rather than of the environment.

By far the most successful correlate of LD was the local minority percentage, but, since LD can itself be conceived to be a joint function of D and the minority percentage, the correlation is tautological rather than a valid environmental explanation. A repetition of the same exercise on the two isolation indices (P*) found that both the Catholic and the Protestant index are far better correlated with the Catholic population-share (P) than with any other environmental variable. However, this, too, is tautological, for P itself is the minimum possible value for the Catholic index and (100- P) is the corresponding minimum for the Protestant index. Thus isolation can be interpreted as the level to which unevenness in the social distribution lifts the index above this minimum.

Because of the element of tautology here, a further multiple regression analysis was carried out, calculating the partial coefficients for the relationship between P* and the environmental variables, holding P constant. The result is a set of moderately high partial coefficients between Catholic isolation and the environmental variables, and these are much higher than the corresponding coefficients for Protestant isolation. In particular, Catholic isolation is closely correlated with the absolute size of the local minority. Since there is a mathematically exact relationship between the two isolation indices, this raises the possibility that environmental variables influence only the Catholic isolation index, while the Protestant index simply follows automatically from the Catholic one.

This issue of causality, and the way in which one index affects another, needs careful reconsideration. For example, it can be argued that P*, rather than being partly a result of D, is its cause. Not only is this case argued in the American literature in the context of racial segregation, but it gains in plausibility from the average composition concept of P* being so much simpler for residents to perceive than the unevenness concept of D. This implies a cause and effect chain in which D is simply a consequence of P*. The most relevant index to correlate with environmental variables is therefore P*.

(8) To provide information on the change and stability in segregation in Northern Ireland during the course of the twentieth century in order to provide a historical explanation to complement the discussion of correlates.

Manuscript census data was used to measure ethnic residential segregation for 20 towns in 1911, expressed at the same scale-level as our 1981 analysis. This material has been fed into a study of change between 1911 and 1981. The frequent political violence during this period has contributed to a general rise in D: the median rose from 31 to 40. The high correlation (r² = 0.72) between the two sets of indices, however, indicated that the pattern of inter-urban variation stayed very stable, though Portadown and four far western towns showed increases well above average. The other unevenness index, 2R, has a similar pattern, with a g eneral rise over the seventy years, but an even higher r² of 0.79. The significance of this measure is its demonstration that not only has it got generally more difficult to achieve integration, but there is considerable uniformity in this result and the difficulty is not expressed by D alone. Still some towns go against the trend, especially Newry as a consequence of its sharply diminished Protestant population-share.

The dominance dimension has also been subjected to a general upward trend as segregation has intensified - the median LD rose from 62 to 71 - but there is a weaker correlation between the towns at the two dates: r² is only 0.59. This is mostly because of some exceptionally large increases, especially in Strabane, where a large rise in unevenness and a sharp fall in the Protestant population-share had a powerful combined effect. Strabane is part of a general far western region of soaring LD indices involving the same four towns as had exceptional rises in D.

Exceptions to the general trend are less frequent and less substantial for the isolation indices, for the correlation between the two dates is exceptionally close: r² is 0.92 for the Catholic index and 0.88 for the Protestant one. However, whereas there was a very marked upward trend in cP*c, there was very little change in pP*p. The increases in Catholic isolation were caused by the combined overall upward trend in both D and P. The steepest rises were in Portadown and the western Border belt again, with an area of rather lower increase in the southern Border strip. The upward trend in P implies a widely reduced Protestant population-share, but this was just sufficiently offset by the increasing unevenness to cause pP*p to stay stable overall, though notable exceptions were the steep drops caused by the considerable Catholicisation of Newry and Strabane.

The Protestant isolation index is the only one of our five measures not to show a moderate increase in its overall level between 1911 and 1981. In terms of the amount of variability between towns, pP*p is exceptional, too, for its stability contrasts with a tendency with the other four indices for the gap to widen a little between the more segregated towns and the more integrated ones. With respect to the pattern of inter-urban variability, the two isolation indices are both remarkably stable, but even the three segregation indices show rather high stability. Moreover, the weakest correlation - for LD - would have been substantially higher but for the case of Strabane.

9.3 Classification of Towns
In seeking an overview of the performance of Northern Ireland towns on the five indices of segregation and exposure analysed in this volume, the values for each index will be classified into three categories. Those above 66.7 will be viewed as high, those below 33.3 as low, and the intermediate values will be regarded as moderate. After classifying towns in this way, it is easier to cross-tabulate one index against another to generate classification categories. Rather than attempting to cross-tabulate all five simultaneously, however, it is convenient to begin with the two separate dimensions of segregation. We shall use the dissimilarity index as the measure of unevenness, preferring it to the replacement index because it is a purer expression of unevenness alone. The dominance dimension will, of course, be measured by LD.

The cross-tabulation of three categories of D against three of LD yields a theoretical total of nine cells, but, as Table 9.1 shows, only six of these are, in fact, occupied. The reason that some cells are empty can most easily be explained with reference to the top right-hand cell. Where D is high, the implied unevenness of the two ethnic groups makes segregated environments so common that low LD scores are avoided. In fact, high D values even eliminate moderate LD scores, for the five towns with a dissimilarity index over 66.7 all have a linear dominance index over 66.7, too. Correspondingly, low values of LD imply that the majority of neighbourhoods have a very mixed ethnic composition, so this makes unevenness impossible to achieve. Indeed, even moderate values of D fail to materialise when LD is low.

That leaves six groups of towns, the smallest of which only has three within it, though the largest has eleven. In order to gain a better understanding of the overall profile of these six groups, Table 9.2 has been compiled to show the performance of the other three indices. In each case, these other variables, too, have been trichotomised according to the position of the values with respect to the thresholds of 33.3 and 66.7. Thus all measures are classified as high, moderate or low. It is now proposed to examine the full suite of indices in order to provide a description of the towns in each of the six types.

We begin this description with the three categories which have low dissimilarity indices. This lack of unevenness implies that, in all these towns, most neighbourhoods have an ethnic composition similar to that of their respective town as a whole. For example, since Comber is only 3% Catholic, the lack of unevenness indicates that most neighbourhoods are approximately 3% Catholic. Conversely, in Magherafelt, which is 43% Catholic, most subareas will be roughly 43% Catholic. It therefore follows that LD, which measures the deviation of neighbourhood composition from a situation of 50% Catholic, will vary closely with P.

In fact, all towns in the first category, which are characterised by a low LD value as well as a low dissimilarity index, possess an overall Catholic proportion which is between 36% and 64%. Thus these six towns all have an ethnic composition which is very mixed overall. In consequence, they also have isolation indices which are at a moderate level for both ethnic groups. This indicates that both the average Catholic and the average Protestant in these towns live in environments which are between one-third and two-thirds Catholic. It should be added that the high degree of ethnic mixing that the indices reveal in these towns does not express the achievement of some utopian ideal, for mixing does not imply, for example, harmony. In fact, there have been some well-known cases of increased segregation and other manifestations of conflict in these settlements, but such events have simply not gone to the extreme of precipitating major segregation.

The second group of towns, which is the largest, combines moderate dominance with low unevenness. These towns all have an ethnic minority which is between 17% and 37% of the total population, and it is this moderate minority-size which, in the absence of unevenness, gives them a moderate level of dominance. In only one of the eleven towns is the ethnic minority Protestant: this is Warrenpoint, which is 80% Catholic overall. This lifts the Catholic isolation index in Warrenpoint to a high level. Conversely, the substantial Protestant majority in the other ten towns means that, despite the low D values, Protestant isolation indices are high in all ten.

The low D indices do ensure that the average member of the minority community in each of these towns normally lives in an environment which is less than one-third minority, for these isolation indices are low. Two exceptions occur, for, in both Antrim and Randalstown, the Catholic isolation index is moderate - by a small margin. Apart from these two exceptions, however, the pattern in these eleven towns is to have a combination of a high majority isolation index with a low minority index. This is radically different from the uniform pattern in the first category of towns, all of which had moderate isolation indices for both ethnic groups.

The third group of towns is superficially paradoxical in that it combines high segregation in the dominance sense with low segregation on the unevenness dimension. However, the high dominance simply follows from the fact that, in the absence of unevenness, all neighbourhoods faithfully reflect the very low Catholic population-share in these towns. All P values are, in fact, below 15%, 50 not only is there a single dominant ethnic group in each town as a whole, but this ethnic group is uniformly Protestant. This has very simple implications for the isolation indices, for, in all nine of these towns, the Protestant index is high and the Catholic one is low. In this respect, of course, these nine towns do not differ from most of those in the second category identified - though closer examination reveals that the indices are more extreme amongst the nine in the third category.

No mention has been made so far of the standardised replacement index in discussing these three types of town. This is because 2R has to be either equal to or less than D, so, since all D values so far considered have been below 33.3, it follows that all 2R values must have been below this threshold, too. Therefore, in all three categories, every single town has had a low replacement index.

When attention is turned to the two groups of towns with moderate dissimilarity indices, this uniformity with respect to 2R is lost, for some towns have low standardised replacement indices, while others have moderate ones. The pattern of variation is readily explicable, however, if it is recalled that 2R is a function of P as well as D. In the case of the eight towns in the two groups we are now considering, low 2R values occur in towns whose minority is less than 25% of the entire population, while moderate 2R levels arise where the minority is 40% or more.

These P values also affect the dominance categorisation of the towns, so the set of towns with a minority of less than 25% are largely in the category which combines moderate unevenness with high dominance. On the other hand, those towns with P values between 40% and 60% have only a moderate level of dominance. It should be noted that the towns which combined such a high degree of mixing overall with low D values had low levels of dominance - Cookstown and its five companions. However, when D rises above 33.3, there is sufficient unevenness for moderate dominance to develop.

There are, in fact, five towns which combine moderate unevenness with moderate dominance, and they all have moderate 2R values, too, except for Ballymena. The low 2R value in the latter town is caused by its anomalously low Catholic population-share, for all the other towns in this group have a very mixed ethnic composition overall. Again with the exception of Ballymena, the isolation indices in this group are a mixture of high and moderate values. The situation looks complex, for each town has a different pattern, but the majority community in each town normally has a high isolation index, while the minority community most often has a moderate index.

This pattern of the majority community having a high isolation index is carried over into the fifth group of towns, especially since this community is in at least an 83% majority in each of the three towns involved. It is the combination of these large majorities in the overall population with moderate unevenness which generates high dominance also, but it is the effect of these large majorities which counteracts moderate D values to cause low replacement indices. The curious feature of this group is that, in all three towns, it is the Catholic community which is in the majority, so all P values are very high, and it is the Catholic isolation index which is high. Protestant isolation indices, on the other hand, are low or moderate.

That leaves only one group of towns to consider. These are the five settlements which are extreme cases by virtue of combining high unevenness with high dominance. They reveal a wide range of P values, for Belfast and Portadown are only 26% and 23% Catholic respectively, while Armagh and Lurgan are much more mixed overall: they are 56% and 52% Catholic respectively. Derry has a similar majority proportion as Belfast and Portadown, 72%, but this time it is the Catholic community which forms the majority. This is the only group in which P plays very little role in affecting dominance, for unevenness is so great that there exist many highly segregated neighbourhoods. Therefore, regardless of the overall ethnic composition, dominance is high.

Similarly, the proliferation of these segregated neighbourhoods means that both Catholic and Protestant isolation indices are uniformly high, regardless of P. The only index affected by P, therefore, is the standardised replacement index, for it is almost as high as D in the two towns with almost equal numbers of Catholics and Protestants. On the other hand, in the three towns with a clear majority of one group over the other, the departure of P from 50% is enough to bring 2R down to a moderate level.

In view of the success of the map analysis in the earlier chapters, it is important to consider the spatial distribution of these six groups of towns, and Map 0 presents the relevant information. Some of these groups are much more strongly regionalised than others. Especially clustered is type 3, combining low D with high LD, for these towns are all located within 25 kilometres of Belfast. True, they are not the only types of town in this region, though, if those of type 2 - with equally low D values but a moderate LD - are added, virtually all towns here are of one of these two types. The only major exception is Belfast itself, which is in one of a set of three types - 4, 5 and 6 - which are found primarily in the western and southern border belts, with a north Armagh extension as far as Lurgan. The common feature of these three types is that all have dissimilarity indices which are either moderate or high, combined with LD values which are also either moderate or high.

We therefore have the emergence of two distinct regions which have been noted before. There is a Belfast region of low D but higher LD, though it excludes Belfast itself, and there is a border belt of towns which are moderate or high on both dimensions simultaneously. These two regions do not occupy the whole of the province, but they do cover most of it. Moreover, most of the remainder is in a north coast belt which combines low dissimilarity with low or moderate dominance.

The summary of segregation captured in Map 0 and Table 9.1 allows final conclusions to be drawn about the prevalence of segregation in Northern Ireland towns. In particular, it is possible to re-examine and refine Poole's 1982 conclusion that the majority of Northern Ireland towns were integrated in their housing but that the majority of its urban population lived in segregated towns. The relevant data, classified into the six types just identified, is presented in Table 9.3, and it shows that perhaps the most interesting group is the sixth. This combines high dissimilarity with high dominance, and it is a rare type in that it only contains five of the total of 39 towns. However, these have no fewer than 58% of the province's urban population. This is not just because this group contains the province's primate city of Belfast, for, in addition, it includes two more of the five largest towns in Northern Ireland. This finding reinforces the paradox identified before.

N.B. Percentages in parentheses show population in each group as a proportion of total urban population

A separate, but related, issue is concerned with each of the two dimensions of segregation separately. The five towns in group 6 on the map are the only ones with a high level of dissimilarity, so, on this unevenness dimension, only a small minority of Northern Ireland towns are highly segregated, even though they contain 58% of its total urban population. However, when attention is switched to the dominance dimension, a completely different picture emerges. There are 17 towns classified as highly segregated in terms of dominance, and they contain 78% of the province's urban population. Therefore, the conclusion on the prevalence of high levels of segregation in Northern Ireland depends entirely on which dimension of segregation has been examined. The contrast is even greater with respect to towns with low levels of segregation, for 26 towns, with 30% of the urban population, have low levels on the unevenness dimension, whereas only six towns, with a mere 4% of the urban population, have low levels in terms of dominance. There could be no better indication of the crucial importance of recognising the dominance dimension of segregation as well as that of unevenness.

9.4 Cause and Effect
This discussion of town classification on the basis of five indices has demonstrated empirically that there is interdependence amongst the indices. It was also made clear at the end of the analysis of contemporary correlates of segregation in section 7.10 that there is not only a mathematical relationship between two of the indices but that there is a causal interlinkage amongst them all. It is the purpose of this section of the conclusion to extend this discussion of cause and effect by drawing from the debate in section 8.7 on change and stability during the twentieth century. That debate focused on the five indices themselves, ignoring any possible environmental variables, yet the discussion of so many of the indices in the preceding sections of Chapter Eight had necessarily taken account of one of these variables - P, the Catholic population-share. It is with this variable that the observations in this section must begin.

Table 8.11 showed that 15 of the 20 towns, whose change between 1911 and 1981 was analysed, experienced increases in the Catholic share of their total population. Moreover, the falls, where they did occur, tended to be much smaller than the rises. Consequently, there was an overall increase in P across the entire spectrum of values. The median and the mean both rose by between three and four percentage points - the median from 23.4 to 26.9, the mean from 34.1 to 37.9 - and the lower and upper quartiles both rose also. It is the consequences of this widespread rise in the Catholic population-share that must now be explored.

If such an increase in P, with the implied associated decline of the Protestant population-share, were distributed uniformly across the neighbourhoods of a town, then the effect would be very simple. The Catholic population-share in each neighbourhood would rise, and the Protestant proportion would decline. It could therefore be expected that the average Catholic would be living in a more Catholic environment, and the average Protestant in a less Protestant environment. This, in turn, implies an increased Catholic isolation index and a reduced Protestant isolation index.

Reference back to Table 8.14 provides a reminder that the Catholic isolation index did indeed rise between 1911 and 1981: the median rose by eleven percentage points, for example, and the mean by six. However, the opposite did not happen in the case of the corresponding Protestant index, for its mean remained the same and its median declined by just a single point: indeed, the two quartiles actually rose. Thus Catholic isolation rose by substantially more than would have been predicted simply from the rise in P, while the Protestant index was a long way from falling by the amount predicted.

It could be argued that these figures indicate Catholic pressure to live in neighbourhoods which are getting more Catholic more quickly than is happening in the town as a whole. However it seems more plausible to suggest that the impetus for the change in the isolation indices is coming from the Protestant community because they are the population to whom something unpalatable is happening. Thus the implication of the rising Catholic population-share overall is that Protestants would live in ever more Catholic environments, and it is suggested here that such a scenario is seen as problematic by at least some Protestants in each town. In consequence, Protestants adjust to the threatened implications of the Catholicisation of their town by moving house just enough to ensure that they continue to live in the same kind of local environment as before. Clearly, some individual Protestants will be occupying a more Catholic neighbourhood than before, and some a more Protestant neighbourhood, but the average Protestant is trying to maintain the same degree of isolation as he or she enjoyed previously.

This argument is rather reminiscent of the reasoning of Massey and Gross (1991) referred to in section 7.10. They proposed that the fundamental control on the amount of desegregation which could take place in American housing was the attitude of the white community to having a black minority in its neighbourhood. Thus tiny black minorities are tolerated, but not those above a threshold of perhaps 5%. Analogous Northern Ireland thresholds are substantially higher in most places, but a similar line of reasoning suggests that the Protestant response to an increased Catholic population-share overall in a town is to move just enough to maintain former isolation levels. That, we propose, is why the Protestant isolation indices stayed so stable between 1911 and 1981, while the Catholic indices rose. Indeed, this is why the Catholic isolation indices rose so much more than P, for this is something which happened to the Catholic community as a result of the Protestant response to rising P.

The form of residential adjustment which Protestants are hypothesised here to be using to avoid their neighbourhoods getting more Catholic as a result of the general rise in P has to be segregation. Only by increasing the level of segregation can the local implications of rising P be countered. The problem, however, for the analyst of the situation is to identify which dimension of segregation it is which is directly adjusted by the Protestant community. Is it unevenness or is it dominance?

It is not possible to solve this problem from the quantitative evidence, but it is worth trying to reason our way to an answer. In the first place, the whole thrust of the argument with respect to the Protestant community is that its members are, on average, trying to maintain the stability of their isolation index. If this statement about the average further suggests that individual Protestants are attempting to maintain the stability of their neighbourhood composition, then that implies that the level of pre-existing dominance is being continued, for the benchmark for dominance is a fixed value of 50%. Thus these people are trying to maintain not only the same isolation but the same level of a majority or minority situation.

On the other hand, the benchmark for the unevenness dimension of segregation varies - and, expressed as the Protestant population-share, Q, it has generally fallen over time in these Northern Ireland towns. It follows that maintaining stability in the Protestant isolation index would lead to a changed deviation from this benchmark. Specifically, if Protestants move house in such a way as to try to maintain their local population-share, then the deviation between the local and citywide ethnic proportions must increase wherever the local Protestant proportion began by being greater than the corresponding citywide value. True, this deviation will narrow where this local share is less than the citywide one, but, by definition, most Protestants will be living where their local share exceeds their proportion in the town as a whole. Therefore, maintaining the Protestant isolation index constant in the face of increasing P implies a widened average deviation between p and P. This, in turn, implies increased unevenness at the same time as the stable deviation of pP*p from 50 implies a stable level of dominance.

The conclusion to this argument must be that the chain of cause and effect begins with a change in the Catholic population-share, P. In the absence of any other action, this has automatic implications for both isolation indices, but the Protestant community reacts against the implication for its index by ensuring that it remains stable. This then leads to a substantial increase in the Catholic isolation index. The stability in pP*p maintains a constant level of dominance in Protestant neighbourhoods, but the sharp rise in its Catholic equivalent leads to a substantial rise in Catholic dominance elsewhere. The overall result is that the average dominance dimension of segregation intensifies, and LD rises. Only then does there arise an impact upon the unevenness dimension, so only then, in a cause and effect sense, is there a consequence for D.

The second line of reasoning stems from the argument in section 7.10 that people are much more aware of the dominance dimension of segregation than they are of unevenness. This suggests that, where the Protestant community tries to maintain stability in their level of isolation, they are consciously attempting to maintain a pre-existing level of dominance because they wish to see either their current majority status continued or, less often, their present minority level not subject to decline. They are not consciously trying to have their neighbourhood remain, in particular, as disproportionately Protestant as before because they do not know how disproportionately Protestant it was in the first place. This, in turn, is because their interest in P is much less than their concern with a 50% composition. This makes dominance much more likely to play an early part in the cause and effect chain and for unevenness to be simply an unlooked for consequence.

Both lines of reasoning therefore support the view that the isolation indices come first in the cause and effect chain, influenced fundamentally by the environmental variable, P. In particular, the Protestant isolation index has consequences for its Catholic equivalent. Both then, in turn, have implications for the dominance dimension of segregation, and only then is there an impact upon unevenness. These tentative findings are illustrated in Figure XIII, and they are similar, but not identical, to the conclusion reached from the analysis of contemporary correlates of segregation in Chapter Seven and presented in Figure VII. There is agreement that exposure comes before segregation itself in the sequence of cause and effect, and there is agreement, too, that unevenness comes at the end, for the dimension represented by D is simply a consequence of interaction preferences.

Figure XIII: Proposed cause and effect chain linking the separate indices of segregation and exposure and based on analysis of the change in Northern Ireland between 1911 and 1981

However, there is dispute over whether it is Catholic isolation which affects its Protestant equivalent or whether the opposite applies. Clearly, the analysis in this volume does not represent the final word on these issues. Indeed, this is not surprising since we are breaking new ground not only by introducing the linear dominance index but also by seeking to identify a cause and effect chain in our framework of multiple segregation dimensions. Further research, both in Northern Ireland and elsewhere, is certainly required to illuminate the problem, and it is to this general issue of the requirements for further research that attention is now turned as our final theme in the conclusion.

9.5 Agenda for Further Research
The most glaring deficiency of the research reported in this particular volume, obvious to any reader, is its dependence on the 1981 population census. This creates two problems. Firstly, it is already well out of date - a snapshot in time 15 years before the publication date of this analysis. Secondly, the administrative difficulties relating to that specific census created a severe impediment to the reliability of its data. Nevertheless, despite experiments with alternative sources of religious information in Northern Ireland, especially from the clergy, the census remains the most satisfactory source for spatially extensive analysis. Therefore, one exceedingly obvious avenue for future research on ethnic segregation in Northern Ireland is to employ the 1991 census.

Even this latest census has problems, however. Unless researchers are prepared to pay large sums of money for grid-square data and devote immense amounts of time to the creation of customised subareas, this will not have the same unique flexibility which was exploited in the 1981 analysis in this volume. In spite of this, the enumeration district data, which is available for 1991, has the potential to be a moderately successful substitute, at least in large and medium-sized towns. Moreover, it is always useful to obtain more up-to-date data, though the significance of this is easily exaggerated, for our analysis of the period from 1911 to 1981 has emphasised the extraordinary stability of ethnic segregation in the province: ten more years, even of political violence, are unlikely to have made much difference. More important is the higher response-rate associated with the 1991 census, though, even here, the increased secularisation of society, as stressed in our companion volume on Belfast (Doherty and Poole, 1995: 95-96), has made a question which focuses purely on religious affiliation a progressively less satisfactory way of capturing ethnic identity.

Because of these qualifications, it is actually less important to update this analysis than to extend it backwards in time. The explanation for a contemporary spatial pattern has to be historical, especially for a phenomenon subject to the high degree of stability over seventy years that has been demonstrated. It will be useful to plug some of the gaps between 1911 and 1981 when, and if, manuscript data eventually becomes available for the censuses of 1926, 1937 and 1951. However, far more fundamental is to explore back in time before the twentieth century. This is very difficult, of course, because only occasional fragments survive both of actual population census material and of census substitutes, but the flourishing family history literature is providing a valuable service in documenting what is available (Grenham, 1992). An example of the kind of analysis which is possible is the work done many years ago on Armagh in 1770 by Wilson (1978), and similar work is currently in progress on Derry, Coleraine and Limavady in 1831 (Poole, forthcoming). Such research will enable us to learn whether the stability of the twentieth century is simply a continuation of a much longer period without much change or whether Ulster's residential environments were transformed and restructured in, for example, the nineteenth century.

The parallel to the issue of extending the analysis over time is to extend it over space. In one sense, this issue does not even arise because the analysis reported in this volume is a 100% coverage of all settlements over a certain threshold-size - specifically, over 1000 households. On the other hand, there is the possibility of researching the settlements below this threshold. These settlements are of two basic types - those in tiny towns and in villages with fewer than 1000 households, and the residue consisting of open countryside.

The reason for these smaller settlements not being included in the current volume was that, at the scale-level chosen for analysis - with an average of 200 households per subarea - these places could only have been divided into four or even fewer subareas. This was regarded as too few for meaningful segregation analysis: even for the 1911 analysis presented here, fewer than four subareas was regarded as unacceptable. The extension of a province-wide comparison to smaller settlements could therefore only be done by going down to a more micro scale-level, and this would necessitate buying even more data. For example, the use of subareas averaging 100 households each would double the price of the necessary purchase-programme. Therefore, a more realistic option would probably be to undertake a separate analysis of the smaller settlements, possibly overlapping with those of between, say, 1000 and 2000 households included in the current study.

The open countryside is a more difficult problem, for quite different reasons. Any segregation analysis involving the unevenness dimension necessitates the identification of a benchmark value for P, from which to measure segregation. In a nucleated settlement like a town or even a village, this value of P is given by the Catholic population-share in the settlement as a whole. However, what is the equivalent for a district of open countryside? It has to be a meaningful entity, but it cannot be based on the extent of the built-up area, as in a town. Perhaps the boundaries of a functional entity, like a parish, could be selected, but these differ for separate religious denominations and are not even relevant for many. What certainly cannot be justified is simply taking an arbitrary segment of space.

This difficulty does not arise with the dominance dimension of segregation because the threshold is automatically and ubiquitously 50%. This partly explains why province-wide analyses which are comprehensive in the sense of including both urban and rural areas, like those of Smith and Chambers alluded to in Chapter Four, invariably employ the dominance dimension. There is certainly a great need to extend this work to a much more spatially detailed scale-level than has ever been attempted before in order to cover two aspects of analysis about which there is a great deal of understandable ignorance. One is a comparison of rural and urban micro-areas to explore the prevalence of integration and segregation in small-scale communities. The other is the investigation of the dominance geography of rural space to map the micro-level complexity spread across Northern Ireland.

These references to geographical scale-level draw attention to the discussion in Chapter Five, where some of the complexities of alternative scale options were reviewed. As pointed out there, this volume has taken a very simple route through the scale-level minefield, for only a single constant level has been employed. However, there is a long-recognised need for multi-level analysis (Poole and Boal, 1973), and amongst the applications of recent computer software in this field has been an interesting study of school segregation in Scotland by WilIms and Paterson (1995). Perhaps even more innovatively, there is a need to experiment with Boal's (1987: 94) ideas on varying the scale-level according to the size of settlement studied, and Moore (1993) has demonstrated the urgent need for this to avoid spurious results in the related field of deprivation analysis. The idea is to make scale more closely match the size of spatially homogeneous areas in each settlement. Such a flexible approach to scale may, for example, be the answer to the inclusion of tiny towns and villages in a general comparative investigation.

A further scale-related idea of Boal does not yet appear to have been followed up at all. This is his suggestion that research is required on people's perception of neighbourhood, so that a scale-level can be chosen for analysis which is meaningful to residents rather than simply convenient for the investigator. As explained in Chapter Five, this is an issue of which Smith and Chambers showed a welcome awareness, though they were not able to build the necessary data-collection into their investigation. This is not surprising in view of the difficulty of such research, as Boal emphasised, and it is even hard to see how its results could be implemented in segregation analysis if the findings were as complex as might be predicted. For example, it might be found that different residents in the same street or even the same household had sharply contrasting perceptions of what scale was relevant to them. Nevertheless, the problem remains. and so does the challenge for imaginative research.

The final point to be made in the context of scale is to underline that the two volumes in this set on ethnic residential segregation have each used a different conceptualisation of scale for their main analysis. The Belfast volume, for its grid-square analyses, held scale constant in terms of area by using kilometre squares, whereas this second volume has defined scale in terms of the number of households in each subarea. No comparative analysis of these two methods has been presented, but there does need to be a comparative exploration of these fundamentally different methodologies. This must involve an examination of the implications for segregation measurement.

The choice of subarea was shown in Chapter Five to be just one of a series of decisions for spatial analysis which could be conceived in a set theory framework. The principal other one was the choice of total set, or study-area, and this is perhaps one of the most important themes to pursue in undertaking further research on segregation in Northern Ireland. It was stressed that there is perhaps a better rationale for measuring segregation within functional cities, defined by labour and housing market areas, than within the physically defined settlements, bounded by built-up area limits, employed in this volume. In addition, comparison of the Northern Ireland results reported here with the vast volume of American segregation research is impeded when hardly any United States research defines cities in a physical sense: the majority of modern analysis there uses functional cities extending far into rural space. Such investigations in Northern Ireland would require a great deal of preparatory work defining functional cities in a more detailed fashion than is done with the current Travel-to-Work Areas, but it would produce a result which is at least as theoretically satisfactory as the material in this volume.

Regardless of which conceptualisation of a town is used, the understanding of the multidimensionality of segregation is enhanced if towns are classified into clusters with similar characteristics on all dimensions. This was the strategy adopted in section 9.3 of this conclusion, but there are three ways in which it should be further extended. Firstly, whereas the procedure earlier in this chapter involved choosing cell-boundaries which were meaningful in relation to the theoretical extremes of the indices, it would be useful to supplement this by applying a statistical grouping technique to the empirical data to search for natural clusters. Secondly, towns were grouped in section 9.3 purely on the basis of their 1981 segregation characteristics, but such an approach could be repeated for other dates to discover how consistent the groupings were over time. Thirdly, the cross-sectional approach still inherent in both these alternatives could be augmented by classifying towns on the basis of their change over time.

Throughout all these research agenda proposals, there has run a common assumption that the measurement of segregation involves the generation of a single value for a town as a whole. The significance of the subareas within each town is therefore simply that they are the entities whose social composition is being compared and contrasted for summarising into the single figure. However, it is now forty years since Jones (1956: 184) drew the distinction between this type of approach - which he suggested was typical of sociology at the time -and an alternative method which he characterised as more geographical. This alternative generated a separate segregation index for each subarea within a town and thus lent itself to mapping the way in which segregation varies geographically in intra-urban space.

Such intra-urban analysis was essentially the theme of the first of the contributions in this two-volume set on ethnic residential segregation in Northern Ireland (Doherty and Poole, 1995), whereas this second volume has dealt solely with the investigation of inter-urban differences. The first volume was, however, restricted to Belfast, so there is clearly a need to apply its methods to the smaller towns of Northern Ireland. Work of this type has already been published for Derry, for example, by Robinson (1970), but it needs to be done on a much more comprehensive scale by covering the totality of towns.

Such a switch in geographical focus also opens up a different set of problems for explanation from those considered in this volume. Thus, by contrasting the specific areas within a town which are most segregated with those that are most integrated, it leads to the generation and testing of hypotheses of the type considered by Jones (1956:185-187) in relation to Belfast. He produced a scatter-diagram plotting segregation intensity against occupational rank, demonstrating not only that segregation was strongest in working class areas on average, but that the relationship varied considerably from one sector of the city to another. The work of Jones predated the great surge of public sector housebuilding of the century's third quarter, so a modern extension of this analysis needs to test house tenure as well as occupational class. Above all, however, it needs to be extended beyond Belfast to the smaller towns of the province. Some of the work of this type already done has been referred to in Chapter Four, but a systematic and comprehensive analysis is required.

A further aspect of the difference between our two volumes is that the first deliberately experimented with alternative methods of using the answers to the census question on religion to determine ethnic identity. In this volume, on the other hand, we have adopted the much simpler solution of limiting analysis to those people stating themselves to be Catholic or Protestant. Since the first volume was limited to Belfast, it follows that the methodological experimentation on ethnic identity has been restricted to Belfast. Therefore, there is a need to extend this type of experimental work to the rest of the province. This cannot be done by simply examining data for Northern Ireland as a whole because, as the Belfast investigation showed, there can be considerable geographical variation in the ethnic identity both of those who fail to answer the religious question and of those who state themselves to have no religion. In consequence, analysis has to be undertaken for at least individual settlements, measuring segregation between alternative categories of answer to the religious question.

All the proposals made so far for further research involve the proliferation of segregation measurements - in time, in space, and to alternative set and subset conceptualisations. However, in addition, the research agenda must incorporate an expansion of the attempt at explanation. This will come partly from further historical measurements, but it will also come by extending the correlation analysis presented in this volume. The range of independent variables included here has been much less ambitious than has become normal in equivalent United States research. For example, Farley and Frey (1994) included the following as independent variables in a correlation and regression analysis of the dissimilarity index in 232 American metropolitan areas. There was a regional variable together with each city's functional specialisation, its age, its rate of housing expansion, and the income difference between its ethnic groups, as well as the size and ethnic composition of its population.

Some of these variables are impossible to obtain data for in Northern Ireland, while others are less appropriate in the Ulster context. Moreover, even after including all these variables, only 68% of the American variation in D was explained, and our 52% level of explanation in Northern Ireland, based on the single independent variable of Catholic community-size, does not compare too unfavourably with this. Nevertheless, all these qualifications do not fully counter the view that more independent variables need to be incorporated in future Northern Ireland correlation analysis of segregation.

Inspiration for ideas on what variables could be included in such a more comprehensive attempt at explanation in Northern Ireland could come not only from analogous research overseas, but also from an analysis of the residuals from our own correlation exercise. No attempt has been made in this volume to present the regression analysis corresponding to the correlation investigations, but clearly each relationship could be summarised in a regression equation and, more importantly, the resulting residuals could be analysed to identify which towns have their segregation and exposure indices least well predicted by the environmental variables so far incorporated. This can help generate ideas for further variables to be included in the attempted explanation as well as identify specific aberrant towns where perhaps a detailed historical investigation would be particularly worthwhile.

Before leaving the American attempts at explanation, it should be added that Farley and Frey not only attempted to explain the inter-urban variation in segregation in 1990 with respect to these variables. They also did a similar analysis in an attempt to explain the variation in the change in segregation between 1980 and 1990. This is an avenue which should really be explored in Northern Ireland, too - both over such short intercensal periods and also over a greater time-length like the 1911-81 period studied in Chapter Eight. Space precluded it in this volume, but it is an obvious way to extend the present analysis.

The attempt at explanation requires not only the correlation of segregation with independent environmental variables and the historical analysis of change and stability. It also must involve the causal interlinkage between the separate dimensions of segregation and exposure. This has been attempted in this volume in the development of possible cause and effect chains, but it is considered appropriate to mention it here as part of the research agenda for the future because the work undertaken so far is seen as exploratory in the extreme. No firm conclusions have been offered - only tentative suggestions.

If it is appropriate to follow the trails blazed by some of the American investigations referred to and to tailor at least some Northern Ireland research to make it comparable with segregation studies in the United States and elsewhere, it is also apposite to cite ideas for research overseas which would make it comparable with our Ulster investigations. Two suggestions may be made. Firstly, there is a need to supplement research in big cities with studies of small settlements. The calls made for such research by the Texas investigators have already been noted in Chapter One (Hwang and Murdock, 1983; Murdock et al., 1994), but our Northern Ireland work includes many settlements much smaller than the minimum size-threshold applied to the Texas work. As a concept, segregation is perfectly applicable in all its dimensions to quite small settlements. Therefore, the current metropolitan focus of so much segregation research worldwide needs to be supplemented with an additional small town orientation which both answers the call of Murdock and his colleagues and extends the threshold downwards

The second suggestion is that there is a need to supplement the work of Goering and Coulibably on what we have called the dominance dimension of segregation by extending its investigation and employing a less cumbersome measure like the linear dominance index which we have developed in this volume by analogy with the dissimilarity index. Segregation research needs to adopt a twin-track approach by focusing simultaneously on both unevenness and dominance, as well as exploring their interrelationship. Both facets of segregation can then be related to exposure, as measured by the isolation indices, so that a truly multidimensional approach can be developed and applied internationally to the analysis of residential segregation.

Continuing this methodological theme, there is still considerable opportunity for experimentation with alternative indices. This volume has deliberately employed a particularly simple set, but, while complexity should not be embraced where simplicity is sufficient, there is still a need to experiment with, for example, measures which involve squaring the deviation between local social composition and whatever benchmark is being used. This implies the use of eta squared and of a form of the Simpson diversity index. For example, we need to understand why the correlations between the linear and the squared indices were so extraordinarily high when applied to this Northern Ireland data. There are other indices, too, which the theoreticians of segregation and a handful of empirical workers have been enthusiastic over, such as those based on information theory and on the Atkinson equation (Zoloth, 1976; James and Taeuber, 1985), but which have never received much more than exploratory application in empirical analysis.

It might be convenient for the reader to finish this section by recapitulating on the proposals being made here for a research agenda on residential segregation. A total of 21 proposals is being made.

1. To extend the analysis comprehensively to the most recent census date of 1991.

2. To fill in some of the missing time between 1911 and 1981 by eventually investigating the intervening census years.

3. To extend the work backwards in time to 1901 and, more importantly, into the nineteenth century and earlier in the sporadic cases where data survives.

4. To widen the scope of segregation analysis to tiny towns and villages, though this will require work at a very fine level of spatial detail.

5. To include the open countryside in segregation studies, though this can probably only be done in relation to the dominance dimension.

6. To apply a hierarchy of spatial scale-levels to the measurement of segregation, while holding scale constant at each level.

7. To experiment with a flexible scale concept by systematically varying the spatial mesh at a single level of analysis, especially in relation to settlement-size.

8. To undertake primary research on residents' perception of neighbourhood to discover what scale-level is most meaningful to different types of people.

9. To explore the respective implications of the two alternative scale conceptualisations - one defining it in terms of area, the other in terms of people or households.

10. To apply the measurement of segregation to the functional city defined in terms of housing and labour markets.

11. To extend the classification of towns by applying statistical grouping techniques, by repeating the exercise for different dates, and by categorising towns on the basis of change over time.

12. To extend the study of the intra-urban variation in segregation, which was applied to Belfast in our first volume, to the smaller towns of Northern Ireland.

13. To test hypotheses - involving, for example, house tenure and social class - advanced to explain the differences in the intensity of segregation from one neighbourhood to another within the same town.

14. To repeat, in the small towns of Northern Ireland, our Belfast experiments with alternative methods of using the answers to the census religious question to define ethnic identity.

15. To incorporate more independent variables into a correlation and regression analysis of segregation at single cross-sections through time.

16. To analyse the residuals from the regression of segregation on environmental variables to identify aberrant towns as a source of ideas for both further correlation research and intensive historical investigation.

17. To extend the correlation-based approach to explanation to the study of change in segregation over time.

18. To investigate in much greater depth the tentative findings made linking the separate indices of segregation and exposure together into a cause and effect chain.

19. To remedy the neglect of small settlements which is such a characteristic of segregation research in the world outside Northern Ireland.

20. To apply the linear dominance index in segregation research elsewhere to balance the current overemphasis on unevenness and exposure.

21. To experiment empirically with alternative indices of segregation, especially those of a nonlinear form.

9.6 Closing Comments
Despite fifty years of serious quantitative research on residential segregation, therefore, the field is still at a youthful and experimental stage. The total volume of work may look quite impressive, especially in the United States, but it has largely sacrificed experimentation to the advantages for comparative analysis of index replication. What is now needed on an international front is a combination of both approaches - methodological experimentation in conjunction with large-scale comparative analysis of a wide range of measures, both traditional and innovatory. Only this can provide the evidence for methodological evaluation, as well as empirical information on the full complexity of segregation.

This volume is presented as a local contribution to what segregation research requires on this international front. In terms of methodology, it has stressed the multidimensionality of the segregation and exposure family of ideas, and it has contributed a new, albeit simple, index to assist in the expression of this multidimensionality. In particular, the argument has been made that unevenness and dominance are very distinct conceptualisations of segregation, so a potentially imbalanced and misleading conclusion may be reached by analysing one and ignoring the other. In addition, the possibility of linking the resulting complex of indices together into a cause and effect chain has been subjected to initial exploration. Finally, the volume has emphasised the need for a long-term time perspective as an approach to explanation, providing indispensable support to the correlation view of explanation.

On the empirical side, the multidimensionality of the approach has highlighted the oversimplification of earlier work which concluded very simply that most Northern Ireland towns were weakly segregated. It is now apparent that, while this is perfectly true of the unevenness dimension of segregation, the dominance dimension produces a more complex verdict. Thus only a handful of towns are weakly segregated in terms of dominance, while the rest are equally split between moderate and high levels of segregation. This difference is not a confusing ambiguity. It is empirical evidence of the need for conceptual clarification, for no question on the frequency or intensity of segregation can be answered without stating precisely what is understood by the term 'segregation'.

The identification of a cause and effect chain linking the multiple indices of segregation and exposure is at too early a stage of development to reach definitive empirical conclusions, but the historical perspective, whose importance has been stressed, has been invaluable for revealing the extraordinary stability of segregation intensities and patterns. Northern Ireland's towns have thus experienced only marginal changes in segregation in the course of a highly turbulent century, and this reinforces a social reproduction interpretation of Ulster society.

Moreover, the analysis has highlighted what must never be forgotten when studying Northern Ireland. The province is an irregular mosaic of radically different places - not just Catholic and Protestant, but with every conceivable level of mixing in between. All the pieces in the mosaic are tossed together into this single political unit, but the result is that nowhere is typical of Northern Ireland as a whole - not Belfast, not Ballymena, not Bushmills. These major differences between places, coupled with the extreme stability revealed over nearly three generations of time, emphasise that the social reproduction processes involved are intensely localised. It is primarily as a contribution to the investigation of this complex geographical localisation that this volume is offered.

Return to publication contents