Vulnerability and multiple deprivation perspectives on economic exclusion in Europe: A latent class analysis Open Access

In recent years general agreement has emerged that, despite the continuing vagueness of the term ‘social exclusion’, its main value lies in drawing attention to issues of dynamics and multidimensionality (Berghman 1995; Room 1999; Sen 2000). However, there is a tension in the social exclusion literature between an emphasis on a heterogeneity of trajectories and, on the other hand, an accumulation of disadvantages involving a ‘spiral of precariousness’ (Paugam 1996). This tendency is stressed in Room's (1999: 171) discussion of notions of continuity and catastrophe in the social exclusion literature. Thus as Whelan and Whelan (1995: 29) argue, while no one would wish to deny that social exclusion arises from of a variety of processes or that it is experienced as involving a good deal more than an income deficit, an uncritical insistence on multidimensionality could paradoxically have the effect of obscuring the processes involved in generating social exclusion.

In this paper we wish to address the issue of multidimensionality by following recent contributions by Breen and Moisio (2003) and Moisio (2004a, b) in applying latent class models in order to identify groups who are vulnerable to what we will label ‘economic exclusion’. In order to apply such methods it is necessary to provide a theoretical justification of the indicators employed. Our focus will be restricted to a small number of dimensions but ones whose interrelationships we consider to be crucial to understand. The notion of social exclusion is not an entirely new one. Thus Townsend (1979) in his seminal work considered poverty to involve exclusion through lack of resources. The European Union has conceived poverty in a similar manner defining poverty as exclusion from the minimally acceptable way of life of the Member state one is resident as a consequence of inadequate resources (Atkinson et al.2002). This provides a rationale for relative income approaches to measuring poverty on the basis that such thresholds are intended to identify those falling more than a certain ‘distance’; below the average and are as a consequence excluded from the minimally acceptable way of life.

The major problem with this approach is that low income turns out to be quite unreliable in identifying households experiencing distinctive levels of deprivation (Ringen 1987). However, to focus solely on deprivation would mean abandoning concern with the resources component of Townsend's definition and would seriously restrict our capacity to understand how deprivation is generated. In recent years it has been possible to further our understanding of the, apparently paradoxical, weakness of the relationship between income and deprivation. This involves trying to take into account unreported income, savings, and other assets, availability of support from family, friends and neighbours, non-cash income and differential needs (Perry 2002). Despite such efforts, the conclusion to be drawn from a substantial portion of the literature on multi-dimensional analysis of social exclusion is that, not only do different methods lead to different conclusions regarding levels of exclusion, but quite different groups are identified as excluded depending on the indicator on which one focuses.¹ As Nolan and Whelan (1996a: 3) argue, until we successfully grapple with the issue of the limited overlap between income and deprivation, efforts to develop a multi-dimensional approach seem unlikely to be fruitful.

Here we intend to focus on three key indicators – relative income poverty, a measure of life-style deprivation that has been found to be more strongly related to income poverty than alternative measures relating to spheres such as housing and social environment and finally a measure of subjective economic strain (Whelan et al.2001). Our objective is to identify groups who are vulnerable to economic exclusion in the sense of being distinctive in their risk of falling below a critical resource level, being exposed to life-style deprivation and experiencing subjective economic strain.

As Moisio (2004b) notes, implicit in the notion of multi-dimensional measurement of exclusion is the assumption that there is no one ‘true’ indicator of the underlying concept. Instead we have a sample of indicators that tap different aspects of a complex phenomenon. If we are to move beyond the accumulation of a mass of descriptive detail we need to develop a measurement model that enables us to understand the manner in which our indicators are related to the latent concept. In this paper we make use of latent class modelling to achieve this objective. The basic idea of latent class analysis is long established and very simple (Lazarsfeld 1950; Lazarsfeld and Henry 1980). The associations between a set of categorical variables, regarded as indicators of an unobserved typology, are accounted for by membership of a small number of latent classes. Latent class analysis assumes that each individual is a member of one and only one of N latent classes and that, conditional on latent class membership, the manifest variables are mutually independent of each others. Conditional independence is just a version of the familiar idea that the correlation between two variables may be a result of their common dependence on a third variable. The logic is identical but explanatory variable is unobserved and must be identified statistically.²

Although the analysis reported in this paper is based on cross-sectional data, a temporal dimension is implicit in our use of latent class analysis. Not all respondents identified as vulnerable to economic exclusions are excluded at a particular point in time. However, by identifying individuals as vulnerable we clearly wish to convey that their prior or subsequent risk of such exposure is significantly greater than that of those allocated to the non-vulnerable class. This approach is entirely consistent with recognition that poverty is not a static phenomenon. Research based on panel data (Bane and Ellwood 1986; Jenkins and Rigg 2001; Fouarge and Layte 2005) has shown that movements into and out of poverty are a great deal more frequent than had been supposed and that a far greater proportion of the population experience poverty than revealed by cross-sectional data. Using these findings and their own research based on German data on social assistance claimant spells, Leisering and Leibfried (1999) have gone on to argue that most poverty spells are actually of a very short duration, tend decreasingly to be associated with structured disadvantage, and are actively overcome by most people experiencing them. The broader context of this perspective is Beck's (1992) argument that individuals are increasingly forced to act on their own initiative to ‘construct’ their own life course. Interpreted in a strong fashion the ‘individualisation’ thesis suggests that poverty and social exclusion are structured far more by life-course transitions than by factors such as social class and employment status. A weaker version would see conventional stratification variables as mediated by particular life events. However, we share with Moisio (2004a) a concern that an interpretation of the evidence relating to poverty dynamics which views entry and exit events as the ‘causes’ of poverty, may obscure the structural context within which such events unfold.

In the analysis that follows we will seek to establish the role which traditional stratification factors play in structuring vulnerability to social exclusion within a framework that acknowledges the dynamic nature of poverty and the potential impact of biography and life-events. We shall also seek to address issues relating to levels of economic exclusion conceived as multiple deprivation. The process by which people come to be exposed to multiple deprivation has been a central concern of the social exclusion literature. Berghman (1995) views social exclusions as involving a social process in which the creation and reinforcement of inequalities leads to a state of deprivation and hardship from which it is difficult to escape. However, as Whelan et al. (2002) note, despite the influence th is perspective has had on both academic and policy discussions, conceptual analysis has remained imprecise and empirical evidence modest. By incorporating both vulnerability and multiple deprivation issues in the same analysis we hope to overcome some of these limitations.

In this paper we make use of the European Community Household Panel (ECHP) survey as released by Eurostat in December 2002 under the User Data Base (UDB) format. The ECHP is a harmonised cross-national longitudinal survey focusing on household income and living conditions. In the first wave (1994) a sample of some 60,500 households i.e., approximately 130,000 adults aged 16 years and over were interviewed across 12 member states. In wave 2 (1995) Austria, entered and wave 3 (1996) Finland joined the ECHP. For our present analysis we use data for eleven countries from the first wave and for Austria and Finland we draw on the third wave. Our choice of waves is motivated by need to have all three indicators available and to take advantage of large sample sizes to avoid the problems associated with spare cell counts. Our unit of analysis is the individual.

Although we will not use welfare regime type as a variable in our analysis we shall address the extent to which the patterning of our results is consistent with this typology (Esping-Andersen 1990; Goodwin et al.1999). In pursuing this issue we allocate Denmark and Finland to the social democratic regime with its emphasis on a substantial redistributive role, seeking to guarantee adequate economic resources independently of market or familial reliance. As Muffels and Fouarge, (2004) note, The Netherlands is something of a hybrid case having moved from being primarily a corporatist ‘breadwinner’ state to one characterised by active employment policies and more restrictive welfare policies but in a context of a safeguarding of principles of equality, uniformity and universality. For our present purposes we allocate it together with Germany, Austria, Belgium, and France to the corporatist regime with its emphasis on welfare as primarily a mediator of group-based mutual aid and risk pooling, with rights to benefits depending on being already inserted in the labour market. The UK and Ireland are located in the liberal regime, which acknowledges the primacy of the market and confines the state to a residual welfare role, social benefits typically being subject to a means test and targeted on those failing in the market. The Southern Mediterranean countries, comprising Italy, Spain, Greece and Portugal, we will take as constituting a distinctive welfare regime with family support systems playing a distinctive role and the benefit system being uneven and minimalist in nature (Ferrera 1996; Arts and Gleisen 2002).

The income measure we employ is total annual household disposable income of the year prior to that in which data collection took place, including transfers and after deduction of income tax and social security contributions. In order to take account of differences across households in terms of size and composition of the household we adjust the household income by using the modified OECD³ equivalence scale. The poverty threshold is then calculated as the 70 per cent median equivalised income line.⁴

The ECHP supplies information about the living condition of the households and we identified thirteen household items, which could serve as indicators of a concept of life-style deprivation. These items are considered to cover a range of what we term Current Life-Style Deprivation (CLSD). A further eleven items relating to housing and the environment, which in principle meet our definition of deprivation, have been excluded because they have been shown to form quite distinct clusters to the CLSD measure and to have significantly weaker correlations with income (Whelan et al.2001). The format of the items varied, but in each case we seek to use measures that can be taken to represent enforced absence of widely desired items. Respondents were asked about some items in the format employed by Mack and Lansley (1985): for each household it was established if the item was possessed/availed of, and if not a follow-up question asked if this was due to inability to afford the item. The following six items took this form:

• A car or van.

• A colour TV.

• A video recorder.

• A microwave.

• A dishwasher.

• A telephone.

In these cases we consider a household to be deprived only if absence is stated to be due to lack of resources.

For some items the absence and affordability elements were incorporated in one question, as follows: ‘There are some things many people cannot afford even if they would like them. Can I just check whether your household can afford these if you want them?’ The following six items were administered in this fashion:

• Keeping your home adequately warm.

• Paying for a week's annual holiday away from home.

• Replacing any worn-out furniture.

• Buying new, rather than second hand clothes.

• Eating meat chicken or fish every second day, if you wanted to.

• Having friends or family for a drink or meal at least once a month.

The final item relates to arrears; we consider a household as experiencing deprivation in terms of this item if it was unable to pay scheduled mortgage payments, utility bills or hire purchase instalments during the past twelve months. An index based on a simple addition of these thirteen items gives a reliability coefficient of 0.80.⁵ For our present purposes we use a weighted version of this measure in which each individual item is weighted by the proportion of households possessing that item in each country. As a consequence deprivation of an item such as a video recorder will be counted as a more substantial deprivation in Denmark as compared to Greece.

The weighted CLSD measure makes it possible to identify for each country and for the 70 per cent median income poverty line a corresponding deprivation threshold. This deprivation threshold is simply the level at which a similar percentage of individuals who are defined as income poor are also deprived. In other words if in Denmark we have identified 18 per cent of individuals as income poor the deprivation threshold is the score value where 18 per cent of individuals have the highest score of deprivation.⁶

The subjective measure of economic strain we employ is based on the following question asked to all household reference persons in the ECHP:

Respondents were offered six response categories ranging from ‘with great difficulty’ to ‘very easily’. The economic strain variable is constructed as being those reporting either ‘great difficulty’ or ‘difficulty’.

As well as these three key variables we use a number of socio-economic characteristics of the reference person that include social class position and employment status. For social class position we employ an aggregated version of the CASMIN class schema (Erikson and Goldthorpe 1992) and distinguish between: manual, self-employed and non-manual. Regarding the employment status we use the principal economic status of the reference person and we distinguish also three categories: being at work (at least 15 hours), being unemployed and being inactive.

In applying latent class analysis, each of our indicators is taken as an imperfect measure of economic exclusion. In order to provide us with sufficient degrees of freedom our income poverty variable has four categories distinguishing between those below 50 per cent median income, between 50–60 and 60–70 and above 70 per cent. Our results will be reported in terms of the conditional probabilities of being below each of the three median income lines.⁸ Our deprivation outcome reports the conditional probability of being above the threshold that corresponds to that relating to 70 per cent of median income. The economic strain variable distinguishes those households that have ‘great difficulty’ or ‘difficulty’ in making ends meet from all others. Thus both the income and deprivation variables are explicitly defined relative to national standards. The economic strain variable allows for the impact of within and between country comparisons but the available evidence indicates that the vast bulk of the variation is within country (Whelan et al.2001). Our objective is not to explain between country differences in these indicators but to compare the within country patterns of differentiation between the economically excluded and others.

In Table 1 we display results for model fit, class size and conditional probabilities. Given the large sample sizes, ranging from 7,660 to 22,812, any parsimonious model is unlikely to fit the data according to strictly conventional criteria of fit. Our model gives such a fit only in Denmark, Finland, Germany and France but comes very close to so doing in The Netherlands and Austria. Nevertheless it does remarkably well across all thirteen countries in accounting for the patterns of association between the three indicators. Full details of model fits for the independence and latent class models are provided in Appendix Table A1 The size of the G² and degree of mismatch in the independence model provides an index of the strength of the association between the indicators that requires explanation. For the independence model, which assumes no association between the variables, the G² goodness of fit statistic ranges from 1167 in Denmark to 7557 in Italy with 10 degrees of freedom. In ten of the thirteen countries the latent class model, which uses an additional six degrees of freedom, reduces the independence G² by over 99 per cent and in no case is the reduction less than 98.3 per cent. Focusing on the index of dissimilarity (▵) or percentage of cases misclassified, which is unaffected by sample size, we find that, while the level of misclassification for the independence model ranges from 12.9 per cent in Denmark to 25.6 per cent in the UK, in contrast for the latent class model the highest number of cases misclassified is 1.4 per cent in Spain and in ten of the thirteen cases the figure is 1 per cent or less. To accept the model only where the criterion of conventional statistical fit is fulfilled would mean that we would reject it in nine cases where the reduction in the deviance associated with the independence model lies between 98.3 and 99.5 per cent. Thus overall we conclude that the evidence strongly supports the hypothesis that the relationships between the three indicators arise because of the division of the population into two latent classes.

The conditional probabilities express the risk of exposure to each type of disadvantage depending upon whether or not one is or is not a member of the vulnerable class (NV. v V). The patterning of such probabilities provides strong evidence that they discriminate between those vulnerable to economic exclusion and those who are buffered from this experience. The size of the vulnerable class, as shown in Table 1, ranges from a low of 15 per cent in Finland to a high of 38 per cent in Greece. Denmark, Finland, The Netherlands, Belgium and Germany are found in the range running from 15 to 21 per cent. For Austria and France and Italy it rises to 24–25 per cent before increasing further to 28 per cent in the UK. The highest values are found in Ireland and the remaining Southern European countries where the size of the exclude group goes from 30 to 38 per cent. Thus, while there is clearly some overlap, the average level of economic exclusion increase as one moves from the social democratic welfare regime countries at one end of the spectrum to the residualist regimes at the other. The major line of differentiation, however, is between the social democratic and corporatist countries on the one hand and the liberal and residualist regime countries on the other.

The key variable differentiating the classes is that referring to life-style deprivation. For those respondents vulnerable to economic exclusion the risk of being found above the deprivation threshold varies from a low of 0.67 in Austria to a high of 0.94 in The Netherlands and Germany. In nine of the thirteen countries the proportion is above 0.80. In contrast in the non-vulnerable class the highest proportion above the threshold is 0.10. In eight of the thirteen countries the relevant proportion does not rise above 0.06. Thus the non-vulnerable class are largely insulated from the risk of being above the deprivation threshold. On the other hand, for those found in the vulnerable class, in every case a substantial majority are found above the threshold.

For income poverty we find a clear, but less sharp differentiation, than in the case of deprivation. The social democratic regime countries constitute exceptions with the risk of being found below the 70 per cent of median threshold if one is located in the vulnerable class being as low as one in three. For the remaining countries the relevant proportion lies in the extremely narrow range running from 0.52 in Austria to 0.66 in the UK, with eight countries being found in the range running from 0.52 to 0.59. The number below the 60 per cent line does not rise above one in five for the social democratic countries with the remaining countries lying between 0.36 in The Netherlands and 0.50 in the UK. For the 50 per cent threshold the figure does not rise above one in eight in Denmark, Finland and Ireland and elsewhere ranges between 0.19 in Austria and 0.38 in Italy. For the non-vulnerable class the proportion below the 70 per cent threshold lies in the narrow range between 0.11 and 0.18. Similarly narrow ranges are observed for the other income lines.

Focusing on economic strain we again find a distinctive pattern of differentiation between vulnerable and non-vulnerable class. The proportion in the vulnerable class runs from 0.41 in Germany to 0.94 in Greece and is above 0.60 for ten of the thirteen countries. For the non-vulnerable class the relevant figure runs from a low of 0.02 to a high of 0.31 in Greece. However, outside the Southern European countries it does not rise above 0.13. Thus in most countries membership of the vulnerable class carries an extremely high risk of experiencing economic strain. The contrast between the vulnerable and the non-vulnerable classes, however, is on average less sharp than in the case of deprivation.

The indicators are independent of each other within latent class. As a consequence the conditional probabilities indicate the proportion that are currently exposed to disadvantage on one of the other indicators who are also disadvantaged on the indicator to which the coefficient relates. For example, for the vulnerable class in the UK 66 per cent of those above the deprivation threshold are also below 70 per cent of median income. Conversely, 91 per cent of those below the income threshold are above the deprivation threshold. In the non-vulnerable class hardly any of those above the income threshold in any of the countries are also found above the deprivation threshold.

In Table 2 we focus on the extent to which those found to fulfil the respective condition are located in the vulnerable class. For income poverty there is considerable variation across countries with the relevant percentage ranging from less than 40 per cent in Denmark and Finland to close to over 70 per cent in Greece. There is a broad contrast that involves income being a more powerful predictor in the liberal and residualist welfare states. The vast majority of those above the deprivation are also members of the vulnerable class. The lowest membership rate is found in Belgium and Germany where just under two-thirds of this group are vulnerable. In ten of the thirteen countries membership levels reach three-quarters or more. Those reporting economic strain are generally more likely to be in the vulnerable class than in the case of the income indicator although there is relatively little difference in the cases of Spain, Portugal and Greece.

Advocates of the individualisation thesis argue that poverty and social exclusions cannot be traced back to common and easily identifiable causes. The poor are increasingly seen as constituting a heterogeneous group affected by variety of causal processes. We are not in a position to offer a comprehensive evaluation of the individualisation thesis since its propositions relate to changes over time. However, a comparison at a point of time of the relative importance of different factors can clearly give us a sense of how far such a process has advanced and to what extent we are required to significantly alter our understanding of the current determinants of economic exclusion.

In Table 3 we set out the relationship between social class and risk of vulnerability to economic exclusion. The simplest model is one where social class influences latent class membership and has no further effect on the indicators. However, in order to achieve a satisfactory fit across the range of countries included in our analysis it is also necessary to allow social class to have an effect on income poverty and economic strain within latent classes.⁹ The former is by far the most important effect, in particular, as it captures the impact of self-employment within latent class. Deprivation is affected by social class only through its impact on latent class membership.¹⁰

Information regarding the fit of this model and all others referred to in this section is provided in Appendix Table A2. The percentage of cases misclassified ranges from 0.6 per cent of cases in Denmark to 2.7 per cent of cases in Belgium. In all countries vulnerability levels are relatively low among the non-manual class. The risk level ranges from a low of 0.07 in Spain and Portugal to 0.18 in Denmark. With the exception of Austria, the risk levels for the self-employed in Northern Europe are much closer to those for the non-manual class that the manual. In the Southern European countries on the other hand vulnerability rates for the self-employed are substantially higher and are much closer to those for the manual class. The risk rate ranges from 0.07 in Germany and Belgium to 0.38 in Greece. In social democratic regime countries risk rates for manual workers do not rise above 30 per cent. The majority of the corporatist countries are found in the range between one in four and one in three. With the exception of Italy, for the liberal and residualist regimes the figure lies between just over four out of ten and just over one-half.

In the final column of Table 3 we report the odds ratios for being vulnerable rather than non-vulnerable for manual as opposed to non-manual classes. These ratios are not affected by variations in levels and summarise the degree of inequality in risk level between the two groups. Three broad groups emerge. For the social democratic regimes together with the odds ratio does not exceed two to one to one. For the remaining corporatist regime countries, together with the UK and Italy, the value ranges between two-and-a-half to five-and-a-half to one. Finally, the highest values are found in Ireland and the remaining residualist welfare regimes, other than Italy, where the relevant value lies between six to one in Greece and twelve to one in Portugal.

In Table 4 we look at the impact of employment status. The residualist welfare countries have relatively high proportions in work vulnerable to socio-economic exclusion with the relevant figure varying between a low of one in five in Italy and a high of one on three in Greece. For the inactive we observe distinctively high values for the residualist welfare states other than Italy but also for the liberal regime countries. The risk level varies from four out of ten in Ireland to over one in two in Greece. Finally, for the unemployed in only one case, that of Finland, do we find less than one-in-two in the vulnerable class. The countries with relatively low values are the social democratic countries together with The Netherlands, Germany and Portugal where the value ranges between 44 and 59 per cent. For the remaining countries between two-thirds and three-quarters of the unemployed are in the class vulnerable to economic exclusion, with the remaining corporatist countries being at the lower end of that continuum.

When we look at the odds ratios for the inactive versus those at work we observe relatively modest variation. The highest values of between five to four to one are found in The Netherlands and the liberal regimes. The remaining values are clustered in the range running from one and a half to three to one. For unemployment the sharpest contrast is between the social democratic countries where the odds ratio is between five and seven-and-a-half to one and the liberal welfare regimes where it ranges between seventeen and twenty-three to one. Considerable variation is observed within the corporatist and residualist regimes.

The scale of the observed effects for both social class and employment status is difficult to reconcile with the notion that economic exclusion has come to transcend traditional stratification boundaries and has become an experience or stage in the life-course (Leisering and Liebfried 1999: 23). However, the continued importance of conventional stratification factors is not all inconsistent with the notion that social exclusion is associated with particular life-course events and it is to these that we now turn our attention.

In this section we address the impact of key socio-demographic factors and associated discontinuities in the life-course such as separation/divorce.¹¹ We could find no evidence of a pattern of systematic variation by age group across countries and it is clear that such variation cannot account for the significant socio-economic structuring documented in the previous section. From Table 5 we can see that households headed by women are consistently more likely to be vulnerable to social exclusion. The strongest effects are found in liberal welfare regime countries and Belgium where the odds ratios range between three to one and four to one. Gender effects are weak in both the social democratic and residualist welfare regimes. For both separation/divorce and lone parenthood the weakest effects are found in the Southern countries which seems to reflect differences in patterns of household formation and dissolution rather than welfare regimes as such (Berthoud and Iacovou 2003). In no case does a relevant odds ratio exceed three to one. The largest effects are observed in the liberal welfare regimes where welfare rules seem to play a more significant role. Those separated/divorced appear particularly disadvantaged in Ireland and lone parents are worst placed in the UK, in the former case the odds ratio is close to eight to one and in the latter it exceeds ten to one. For the remaining countries there is little in the way of systematic variation. In the majority of cases odds ratios vary between two and five to one. Neither the scale, nor patterning of the biographic or life event variables suggests that they can come near to accounting for the socio-economic structuring of vulnerability to social exclusion documented earlier.

Not only are manual workers and those not at work exposed to higher levels of vulnerability than other at risk groups they also constitute larger segments of the population. In Table 6 the composition of the vulnerable group is broken down by social class. In every case other than Greece manual workers form a majority of the vulnerable class. The Greek situation is accounted for by the fact that equally large numbers are drawn from the self-employed and manual workers.

Apart from Greece and Italy, the lowest manual composition levels are found for the social democratic welfare regimes together with Belgium where the relevant figure ranges between 53 and 56 per cent. Even here manual workers are significantly over represented. For the remaining countries the relevant figure ranges between three-fifths and three-quarters. The lowest figures of between 5 and 12 per cent for non-manual workers are found in Ireland, Spain, Portugal and Greece. In contrast the figure rises to close to four out of ten for Denmark, Finland and Belgium. In no case do we observe a pattern consistent with the notion of transcendence of class effects and even where the distribution is most heterogeneous in class terms a vulnerable respondent is substantially more likely to come from the manual rather the non-manual class despite the fact that the former are numerically superior.

The process by which people come to be exposed to multiple deprivation has been a central concern of the social exclusion literature. In the analysis that follow we shall document the extent to which respondents are currently exposed to social exclusion in the sense that they are simultaneously below 70 per cent of median income, above the corresponding deprivation threshold and experiencing subjective economic strain. This definition of social exclusion and the associated form of multiple deprivation is somewhat more circumscribed than many that have figured in the literature, which frequently make reference to factors such as social isolation. However, since the evidence for the significance of social isolation is weak,¹² it seems sensible to focus first on dimensions where evidence of at least moderate correlation exists and whose relationships have been subjected to considerable scrutiny.

In order to calculate levels of economic exclusion involving income poverty, life-style deprivation and economic strain we take advantage of condition of local independence whereby these indicators are independent of each other within categories of the latent class. Because of extremely low conditional probabilities for deprivation in the non-vulnerable class, calculation of social exclusion levels reduces to calculating them within the vulnerable latent class. It is also the case that within the vulnerable class, apart from Denmark, economic strain levels are so high that multiple deprivation levels involving all three dimensions are only marginally lower than those involving income poverty and deprivation alone. In Table 7 we show economic exclusion levels for the vulnerable class and the population as a whole. The lowest levels are observed for the social democratic regime with one in eight in Denmark and one in five in Finland being exposed to such exclusion. The Austrian and German levels are also comparatively low with one in five fulfilling the necessary conditions. Elsewhere variation is rather modest with levels varying between one in three and two in four. All of the remaining countries have values lying, approximately, between 30 and 40 per cent with Belgium, France and Ireland at the lower end of the continuum and the UK, Italy and Greece at the upper end.

The level of exclusion in the population depends not only on the level within the vulnerable class but also on the size of that class. In fact, cross-national variation in levels of economic exclusion depends largely on variation in the size of the vulnerable class. The lowest level of 3 per cent is found in the social democratic countries. This figure ranges between 5 and 7 per cent in the corporatist countries. In most of the liberal and residualist welfare states countries the figure lies between 9 and 11 per cent with Greece being distinctive in displaying a level of 15 per cent.

In Table 8 we look at the impact of social class on social exclusion levels. For the non-manual class levels are low in all countries. The highest level of 4–5 per cent is found in Italy and the UK. For the remaining eleven countries the figure ranges between 2 and 3 per cent. For the self-employed the figure is a good deal more variable. It is at its lowest level of 2–4 per cent for Germany, Ireland and the social democratic regimes. It rises to 7–8 per cent for the remaining corporatist regime countries, the UK and Italy. Consistent with our expectations, levels are comparatively high in residualist welfare regimes, reaching levels of one in nine in Spain and one in seven in Portugal and one in five in Greece. While the self-employed are significantly less likely to be found in the vulnerable class than manual workers, within this class they have particularly high probabilities of being below the income threshold. This is a significant factor in producing the observed levels of exclusion. In the social democratic regimes exclusion rates remain low even for manual workers and do not exceed 4 per cent. Amongst the corporatist states there is a range of variation. Belgium, Germany and Austria are at the lower end of the continuum with rates of approximately 5 per cent. The Netherlands and France are at the upper end of the continuum with rates of between 9 and 11 per cent. The highest level of economic exclusion for manual workers is found in the liberal and residualist welfare regimes where the figure rises to approximately one in eight. For the UK and the Southern European countries their overall high levels of economic exclusion are attributable to the relatively high levels experienced by both the self-employed and manual workers. For Ireland, on the other hand, its overall situation is almost entirely a consequence of the fact that manual workers are particularly disadvantaged.

In this paper we have argued the need for a more explicit treatment of theoretical and measurement issues relating to the conceptualisation of social exclusion as multidimensional. In addressing these issue, rather than seeking to deal with a wide range of dimension, we have focused on a smaller number that we consider to be important on theoretical grounds and to constitute crucial building blocks in efforts to construct reliable and valid indices of social exclusion. We have also concentrated on dimensions where sufficient previous work exists to provide us with adequate confidence in the individual indicators and a body of knowledge concerning the observed relationships between them. Our starting point involved treating these measures as imperfect indicators of vulnerability to economic exclusion. From such a perspective one can distinguish between vulnerability to economic exclusion of a broader social group and risk of exposure to multiple deprivation at a point in time applying to some subset of this group.

Applying latent class analysis to the thirteen countries in our analysis we found that in every case the latent class model accounted for the vast bulk of the associations between our indicators. The risk of vulnerability to economic exclusion varied across countries in a manner broadly consistent with our expectations based on their allocation to welfare regimes. The social democratic regimes are distinctive not only in having a relatively small class vulnerable to economic exclusion but also in exhibiting a weaker pattern of differentiation between the vulnerable and the non-vulnerable. For the remaining countries there is a striking similarity in the manner in which the latent classes are differentiated. In every case it is the deprivation indicator that has the greatest discriminatory capacity, followed by economic strain and then income poverty.

Our analysis thus confirms the value of a latent class approach to multidimensionality. Implicit in this approach is a dynamic perspective on exclusion in which vulnerability is translated into the actual experience of economic exclusion conceived as multiple deprivation. Thus, while at a particular point in time deprivation is the primary factor differentiating the vulnerable and non-vulnerable classes, our findings are entirely consistent with a perspective that requires that social exclusion must be understood as the outcome of a process in which the accumulation and erosion of resources over time interacts with variability in the demands with which households must cope (Nolan and Whelan 1996a). The clarity of the picture that emerges relating to the underlying structuring of economic exclusion can be contrasted with the impression of something close to apparent randomness that emerges when income is correlated with individual deprivation items (Mack and Lansley 1985).

This pattern of predictable social structuring continues when we examine the impact of socio-economic characteristics of the household reference person on vulnerability to economic exclusion. While we are not in a position to examine trends over time, the consistency and scale of both social class and employment status effects seem entirely inconsistent with the notion that poverty and economic exclusion has come to transcend social boundaries. However, while there is clear evidence of socio-economic structuring across the range of countries, there is also significant variation. Social class effects are particularly strong in liberal and residualist welfare regime countries, with self-employment effects being particularly strong in the latter. Employment status effects are even stronger but are relatively weak in the social democratic regimes and particularly powerful in the liberal regime countries. Membership of the vulnerable class clearly does extend to the non-manual class but in every country, except Greece where the self-employed play a crucial role, manual workers constitute a majority of the class and, in all but four countries, the latter taken together with the self-employed comprise three-quarters of the socially excluded.

Such socio-economic structuring is entirely consistent with differentiation by individual biography and life–situation. However, age effects were found to be a good deal weaker and more variable. Gender effects were more uniform but again modest in comparison with socio-economic effects. Separation/divorce and lone parenthood were clearly associates with increased vulnerability but, apart from the exceptional impact of the former in Ireland and the latter in the UK, could not be interpreted in welfare regime terms. Indeed such effects were uniformly weak in the Southern European countries.

As with findings in relation to poverty dynamics our analysis has implications for the debate on universal versus selective policies. As Moisio (2004b) notes, an emphasis on the transitory nature of poverty, and arguments for individualisation, focus attention on the events that trigger poverty and seem to carry, at least implicitly, a recommendation for active targeted policies. The findings we have presented here regarding the structuring of social exclusion are consistent with recent analysis of poverty dynamics by Breen and Moisio (2004) who argue that poverty mobility has been overestimated and that there is a clear distinction between those who are almost entirely protected from poverty and those who move in and out of that condition. Although operating on the basis of cross-sectional rather than longitudinal relationships, and focusing on rather different outcomes, we are led to a similar conclusion to Layte and Whelan (2002: 231) that it is necessary to direct attention away from highly targeted policies aimed at multiply deprived groups and encourage a focus on more generalised responses directed at groups who may not be currently socially excluded but whose vulnerability means that a range of factors may precipitate this situation. As Kleinman (1998) notes one of the consequences of employing the term ‘social exclusion’ to denote multiply deprived groups is that it defines the key social cleavage as between a comfortable majority and a small excluded socially isolated minority. Despite the accumulating evidence which challenges this assumption¹³ there is likely to be considerable resistance to it since, as Whelan et al. (2002: 103) note, it directs attention to macro and expensive policies relating to the factors contributing to the vulnerability of broad class and status groups and thus refocuses attention on issues such as access to education, employment and operation of the tax and welfare system.

We would like to thank the anonymous referees, Richard Breen and Pasi Moisio and participants at an ESRI seminar for comments on an earlier version of this paper. The paper is based on analysis of the European Community Household Panel survey. The data are used with the permission of Eurostat. The authors are entirely responsible for the analysis and interpretations contained in the paper.

Christopher T. Whelan is a Research Professor at the Economic and Social Research Institute, Dublin. He is currently Chairperson of the Standing Committee of the Social Sciences of the European Science Foundation and Coordinator of the ESRI team participating in the Economic Change, Quality of Life and Social Cohesion (EQUALSOC) Network of Excellence under the EU Sixth Framework His current research interests include the causes and consequences of poverty and inequality, measurement and monitoring of poverty, social exclusion, and quality of life and social mobility. In recent years he has contributed on these topics to a range of academic journals. He is involved in a range of European projects involving analysis of sources such as the European Community Household Panel Study, the Irish component of the European Survey of Income and Living Conditions and the European Survey of Quality of Life.

Bertrand Maître is a Research Analyst at the Economic and Social Research Institute, Dublin. He has been at the Economic and Social Research Institute since 1997. He is a Graduate of Economics in the University of La Sorbonne Paris I in 1990. He is currently involved in a number of projects using a range of European and Irish data sets. His research interests include the distribution and packaging of household income and multidimensional approaches to poverty and social exclusion. He has published recently on these issues in the European Sociological Review and the Journal of European Social Policy.