SOCIAL MOBILITY IN EUROPE
Richard Breen1
Nuffield College, Oxford
This draft: Monday, 19 July 2004
Words: 8768
This paper is based on joint work undertaken with Ruud Luijkx as part of the `National Patterns
of Social Mobility: Convergence or Divergence?' project that was partially funded by the
European University Institute, Florence, and supported by Nuffield College, Oxford.
1
richard.breen@nuffield.oxford.ac.uk

2
Social Mobility in Europe
Introduction
Today I want to talk about the findings of a project, and forthcoming book, Social Mobility in
Europe (Richard Breen, ed., Oxford University Press, 2005), that looks at social mobility in 11
European countries over a period of almost 30 years. I want to summarize some of the main
results of the book and discuss how they relate to previous research in this area; and I also want
to make some comments about the consequences of this research and of these findings for the
study of social mobility. But I will begin by briefly explaining what the project was all about.
11 countries are represented in this study: they are Britain, France, Ireland, West Germany, the
Netherlands, Italy, Sweden, Norway, Poland, Hungary and Israel, and the period covered is from
the early or mid-1970s to the mid- or late 1990s. The book contains chapters dealing with each
country, written by an author or authors from that country (see Table 1). In each case the authors
base their analyses on data sets comprising as many as possible high quality, nationally
representative surveys carried our during this period. There are some common elements to each
of these chapters but, by and large, each authorial team was asked to analyze and discuss the
trends in social mobility in their own country and provide some explanation of them. This has
necessarily led to different approaches as circumstances dictated. But, in addition to the 11 single
country chapters, there is a further empirical chapter, the aim of which is explicitly comparative.
Here the data sets from all the countries have been put together to allow formal analyses of the
differences between countries and the changes through time. By using this design it was hoped to
marry the advantages of an edited collection of country chapters (namely the insight that can be
brought by authors who have extensive knowledge of their own language, culture and
institutions) to those of a proper comparative study (the ability to test, rather than simply
hypothesize, patterns of similarity and difference between societies and of variation over time).
[TABLE 1 HERE]
The motivation for the project was our lack of detailed knowledge about how patterns of social
mobility had evolved since the 1970s and whether societies were becoming more similar or more

3
diverse in this respect. We took the early 1970s as the baseline for our study because we know a
great deal about national variations in social mobility at that time largely thanks to the CASMIN
(Comparative Analysis of Social Mobility in Industrial Nations) project which culminated in The
Constant Flux (1992). Here Erikson and Goldthorpe used cross-sectional data (that is, one
mobility table per country) drawn, for the most part, from the late 1960s and early/mid 1970s, to
compare patterns of social mobility between, in total, 12 European countries plus the USA,
Australia and Japan. There is considerable overlap in the coverage of Erikson's and Goldthorpe's
study and the present one: nine of the countries represented in their work are included here (11 if
we count England and Wales and Scotland separately, though in our analyses we consider them
together) and, indeed, the first mobility table in the time series of tables for some of these
countries is the same as that used by Erikson and Goldthorpe.
Data
The data used in the project comprise 117 mobility surveys covering the period 1970 to 2000.
The sources of the data are shown in Table 2. The number of tables per country ranges from two
in Israel and Italy to 35 in the Netherlands.
[TABLE 2 HERE]
In general the age range of the respondents in our mobility tables is 25 to 64, and we coded
social class according to the seven-class `CASMIN' scheme (see Table 3). This identifies classes
I+II (the service class); III (routine non-manual class); IVab (petty bourgeoisie with and without
employees); IVc (farmers); V+VI (skilled manual workers, technicians and supervisors of
manual workers); VIIa (unskilled manual workers not in agriculture) and VIIb (farm workers).
One consequence of choosing this categorization is that it allows our results to be compared with
those of The Constant Flux where the same categories were used.
[TABLE 3 HERE]
Methodological issues
As Table 2 showed, the 11 countries contribute rather different numbers of mobility tables to our

4
cross-national analyses. Sweden, for example, has a table for every year from 1976 to 1999,
whereas Poland and Ireland have only three tables each, covering the years between the early
1970s and 1994. The amount of information we possess regarding change over time, and the
reliability of the conclusions based on this information, will vary between countries. If we have a
small number of observations, any one of them may be very influential in determining whether
or not the data display a trend (as we shall see) and this will inevitably lead to uncertainty in the
conclusions we draw. All else equal we must, as a consequence, attach more credence to results
about temporal trends drawn from countries with a larger number of observations (Sweden, the
Netherlands, Great Britain and Germany).
Furthermore, the data that we use are never free of error, and differences in data quality may
easily be mistaken for substantive differences. We have used the best quality data available from
each of our 11 countries, but we still need to be aware of the potential for differential reliability
and validity to induce spurious cross-national variation and temporal change. As far as the
differences between countries are concerned, the fieldwork for the surveys we use was in all
cases carried out according to internationally accepted procedures and the subsequent coding of
the variables ­ notably class origins and destinations ­ followed a common, and widely
implemented procedure. Nevertheless, while adherence to such norms is some reassurance that
the data attain high standards of quality, the surveys in the various countries were carried out
independently of each other, and so we should be cautious about what we infer from them
concerning cross-national differences. As far as change within countries is concerned, we can
have more faith in our findings when the various surveys have been administered in a consistent
fashion. In three cases the data always come from the same survey series: these are France (the
FQP ­ Formation- Qualification Professionnelle ­ surveys), Britain (the General Household
Survey) and Sweden (the ULF series). In a further five countries the data sets come from highly
comparable sources: these are Ireland (where the three surveys were all carried out by the same
fieldwork organization), Hungary (where the four surveys were all fielded by the Hungarian
Central Statistical Office), Italy (where a number of the same academics were involved in the
design and execution of the two surveys), Germany and Israel. But in the remaining three cases ­
Norway, Poland and the Netherlands ­ the data come from various sources within each country
and thus the possibility that variations in data quality might be mistaken for temporal change is

5
greatest here.2
We believe that more reliance can be placed on estimates of trends within
countries than measures of differences between them: thus our discussion, later in this chapter, of
which countries are more or less open in their mobility regime, should be interpreted with some
caution. Finally, while the data that we have are probably adequate for presenting a picture of
broad trends and differences, we would have less confidence in the extent to which they allow
the specifics of the pattern of social fluidity to be compared across either time or countries. This
consideration has then dictated our choice of models. Rather than seeking to develop detailed
models of the fluidity regime we prefer instead to fit rather general models and to assess their
adequacy using several measures (including the conventional chi-squared goodness of fit test and
the index of dissimilarity).
Absolute mobility and class structure
In contemporary studies of social mobility a key distinction is drawn between observed patterns
of social mobility, sometimes referred to as `absolute mobility', and social fluidity (or `relative
mobility'). Absolute mobility is concerned with patterns and rates of mobility, where mobility is
understood simply as movement between class origins (the social class in which someone was
brought up) and class destinations (the class they occupy at the time of the survey). Social
fluidity concerns the relationship between class origins and current class position: specifically it
is based on the comparison, between people of different class origins, of their chances of being
found in one destination class rather than another. If these chances were the same regardless of
origins then the mobility table would display perfect mobility: the class in which a respondent
was found would not depend on (would be independent of) the class in which she or he was
brought up. Social fluidity is often interpreted as an index of equality in the chances of access to
more or less advantageous social positions between people coming from different social origins
(in other words, as an index of societal openness),3
and contemporary research on social mobility
2
There are probably two major factors that will lead to change in the quality of mobility data within each country.
First, response rates tend to decline over time and so the representativeness of mobility tables derived from survey
data may worsen (though this is not an inevitable consequence of falling response rates and it also quite plausible
that when response rates are lower the quality of the data that are collected is higher). Secondly, it seems reasonable
to suppose that there may be variation over time in the quality of treatment of the actual data collected: that is to say,
in the collection, coding and processing of questionnaires.
3
The terms association and social fluidity can be used interchangeably: greater social fluidity implies lower origin ­
destination association. In the log-linear modeling context in which this paper is situated this association is captured
by odds ratios.

6
accordingly pays much more attention to social fluidity than to absolute mobility. I will follow
that precedent here and so will say only a few words about absolute mobility.
In fact, our results concerning absolute mobility can be summarised quite easily. There has been
a marked convergence in the class structures of European countries and in their patterns of
absolute mobility, and these things are true for both men and women.
(a) Convergence in class structures has been driven by some internationally consistent
trends, such as the growth in the service class, I+II, and the decline in manual work,
particularly of the unskilled kind. Among women, increased rates of labour force
participation have been associated with a reduction in international variation as more and
more of them enter occupations in the white collar classes, I+II and III. But the single
biggest cause of this convergence has been the declining significance of the farm classes,
IVc and VIIb, in those countries (such as Poland, Ireland, and Hungary) where a large
farm sector persisted until the last quarter of the 20th
century.
(b) This trend towards convergence in class structures has occurred together with decreasing
variation between countries in their rates of overall mobility, of vertical, of upward and of
downward mobility ­ and, again, this is evident among both sexes, as Tables 5 and 6
show. But, further, the distribution of people in the mobility tables of the different
countries has also grown more similar. If we calculate the s from comparisons, between
all pairs of countries, of their entire mobility tables, we find that the average  (the
average difference between countries) falls from 43 per cent in the 1970s, to 33 in the
1980s and 30 in the 1990s, among women, with the comparable figures for men being 39,
30 and 30 per cent. And the variance around these means has also declined: from 163.2 to
62.6 to 41.6 among women and from 137.5 to 62.9 to 56.1 among men.4
Although
4
The model of common social fluidity among countries within each decade misclassifies between three and four per
cent of cases. But if, instead of allowing each country to have its own distribution of origins and destinations, we
force these to be common in the same way that social fluidity is common (so we fit the model C OD) we find that
such a model misclassifies, among men, 24 per cent of cases in the 1970s, 19 per cent in the 1980s and 20 per cent
in the 1990s, and, among women, 29, 22 and 21 per cent. Because this model sets both fluidity and the origin and
destination distributions to be the same in all countries, and because its fit to the data (measured by ) improves
over decades, this is further confirmation that absolute mobility flows are becoming more similar. In addition, the
difference in  between this model and the common social fluidity model can be seen as an approximate index of
the importance, for absolute mobility, of differences between countries in their origin and destination distributions.
Evidently these differences are of declining importance; in particular they declined between the 1970s and 1980s.

7
European countries continue to show differences in their absolute mobility flows, these
have become smaller.
[TABLES 4, 5 and 6 HERE]
Absolute mobility concerns the observed rates and patterns of flows between origin and
destination classes and, in mobility analysis, is treated as the consequence of social fluidity (the
relative chances of people from each origin being found in each destination class) operating
within fixed origin and destination distributions. A model in which origins and destinations are
independent, given the observed distributions of these two in each country and at each point in
time, correctly classifies over 80 per cent of cases, while a model which also assumes a common
level and pattern of social fluidity correctly classifies around 95 per cent of cases. It is evident,
therefore that changes over time, and differences between countries, in absolute mobility are
driven by variation in the origin and destination distributions rather than in social fluidity.5
Can such variation be said to follow a pattern? We believe that the answer, in very broad terms,
is yes. We might imagine societies following a developmental path that incorporates two major
transitions: from an agricultural to an industrial society, and from an industrial to a post-
industrial society. The consequences, for the class structure, of the former transition are a decline
in the proportions in classes IVc and VIIb and a growth in the remaining classes, especially
(among men) the manual working classes V+VI and VIIa. The transition to a post-industrial
society sees the decline of V+VI and VIIa and the growth of I+II and III.6
Everywhere the
decline in agriculture is either more or less complete (Britain, Germany, Sweden, Israel, the
Netherlands) or well underway while, in eight of our 11 countries (Ireland, Poland and Hungary
being the exceptions), between the 1970s and 1990s, the class structure saw a steady fall in the
proportion of men in classes V+VI and VIIa and a consistent increase in the proportion in I+II
and III. Among women the pattern was exactly the same. These differences mean that some
countries display a post-industrial class structure with a heavy concentration of people in classes
5
This point is widely recognized. Compare, for example, Grusky and Hauser: `intersocietal differences in observed
rates must be attributed to variations in occupational distributions' (1984: 29) and Erikson and Goldthorpe: `if we
wish to understand cross-national variation in absolute rates, it is on differences in the structural contexts of mobility
that our attention must, almost exclusively, be focused' (1992: 213-4).
6
For the majority of countries the decline occurred in class VIIa and not in V+VI.

8
I+II and III: this is particularly true of the male class structure in Britain and the Netherlands and
it is true of the female class structure in several countries. But the important thing, from the point
of view of the study of absolute mobility, is the recent rapidity of the transition out of
agriculture. Similarly, we saw in our comparative analysis, that the shift towards a concentration
of women in the white-collar classes has been more rapid in countries such as Hungary and
Poland where the class distribution in the 1970s differed most from this. The result has been the
growing similarity in destination distributions that we have already remarked upon. But because
countries embarked on this developmental path long before the first of our surveys was fielded,
there is also decreasing variation in class origins. The mean value of the  between class origins
for each pair of countries fell from 33 per cent in the 1970s to 23 in the 1980s and 24 in the
1990s.7
Absolute mobility flows converged because their main determinants did.
This convergence chiefly occurred between the 1970s and 1980s and whether the trend will
persist, or even strengthen, is, of course, difficult to say. Clearly, if the working classes continue
to decline in those countries where the decline has begun, and if this extends from VIIa to V+VI,
then further convergence will be inevitable as men, like women, come to be heavily concentrated
in classes I+II and III. Recent historical experience of the location of industrial production would
suggest that we can expect further convergence: in any event, it seems unlikely that any of these
countries will display a growth in classes V+VI and VIIa, while some at least will experience a
decline. As for the countries in which these classes have not yet begun to decline (Ireland,
Poland and Hungary), the outlook seems less certain. In Ireland the growth of classes I+II and III
has outstripped that of V+VI and VIIa over this period, but this is not true of the male class
structure in Poland and Hungary. On the other hand, among women in Poland and Hungary there
has been a steady growth in classes I+II and III and an increase, then a decline, in V+VI and
VIIa, suggesting that the second transition may be under way. Much here depends on the nature
of economic development. Foreign direct investment in manufacturing, as in the Irish case, is
one mechanism by which the size of the working class may be sustained and the rate of
convergence consequently slowed.
7
These figures are for men. For women the figures are 36, 24 and 24 per cent. The slight differences arise because
our samples of women include only those in the labour force and we have no data for women in Ireland.

9
Social Fluidity
In our comparative analysis we found that trends in social fluidity are very similar among men
and women, showing a widespread tendency towards greater fluidity. Britain is the sole clear
exception to this: here there has been little or not change. In other cases ­ notably Germany ­
there is no statistically significant change, though the trend, at least for men, is towards a weaker
association between origins and destinations. Elsewhere ­ in France, Ireland, Sweden, Poland,
Hungary and the Netherlands ­ there is a statistically significant increase in fluidity, though, the
small number of observations for Ireland (three), Poland (three), and Hungary (four), and the
lack of a consistent pattern of change in these countries, must leave some room for uncertainty.
But in contrast to absolute mobility we see no evidence of convergence among countries in their
social fluidity. Figures 2 (for men) and 3 (for women) show these within-country trends in the
form of the annual  coefficients from a LmSF model8
with common local association among all
the yearly tables of a given country.
[FIGURES 2 and 3 HERE]
Figure 4 (for men) and Figure 5 (for women) show the LmSF  parameters from a model,
applied to decade data from each country, which assumes common local association in each
table, varying only by . The value for Britain in the 1980s is set to unity. Among men, Figure 4
shows that, in the 1970s, levels of social fluidity were lowest in Germany, France, Italy, Ireland,
Hungary and the Netherlands and highest in Britain, Sweden, Norway, Poland and Israel.
Fluidity increased in France, Sweden and the Netherlands and possibly in Ireland, Hungary and
Poland too. The increases in the Netherlands and Hungary were particularly marked. These
different trends have left several countries ­ Sweden, Norway, Poland, Hungary and the
Netherlands ­ with, as far as we can tell, rather similar rates of fluidity, followed by Britain
(where the absence of change has led to a shift in its relative position), Ireland, France, Italy and
Germany, which remains the country with the strongest association between class origins and
class destinations. At the other extreme, Israel is consistently more open than any other country.
Overall, however, we can find no convincing evidence of convergence in fluidity regimes: for
8
LmSF means `log-multiplicative social fluidity' (by analogy with CnSF ­ `common social fluidity'). We use LmSF
to refer to the unidiff (Erikson and Goldthorpe 1992) or log-multiplicative layer effect model (Xie 1992) when the
local association is modeled in a completely unspecified way (as it is in CnSF).

10
example, the within decade variance of the s shown in Figure 4 is largest for the 1980s and the
 for CnSF across countries in each decade is larger for the 1990s than for the 1980s.
[FIGURES 4 and 5 HERE]
The picture among women (Figure 5) is very similar. Once again, the points for the 1970s are
above those for the 1980s which are above those for the 1990s, indicating a general tendency for
fluidity to increase, with Britain being an exception. The average  falls from 1.28 in the 1970s
to 1.14 in the 1980s and 1.05 in the 1990s. France and Germany are the least fluid societies,
Britain, Sweden Poland, and, by the 1990s, the Netherlands are the most fluid. Hungary presents
a different picture for women than men, the former showing much lower fluidity, compared with
other countries, than the latter. In Israel the values are 0.84 in the 1970s and 0.71 in the 1990s.
Taken together with the results for men this is evidence of the exceptionally fluid nature of
Israeli society.
Overall, the results from our 11 countries point to a clear conclusion: there is a widespread
tendency for social fluidity to increase, even though this might not be a statistically significant
trend in ever case. Among men, fluidity is greater at the end of the period than at the start in
every country except Britain and Israel (where the values remain the same). Furthermore, of the
20 decade to decade changes in fluidity reported in the book, we find that in 16 of them fluidity
increased and it declined in three ­ in Ireland and Britain between the 1970s and 1980s and in
Norway between the 1980s and 1990s. There is just one further notable case in which fluidity
fell (but which is obscured by the use of aggregated decade data) and that is Hungary where
fluidity declined significantly between the 1992 and 2000 observations. Although there are some
cases (such as Sweden) where we cannot be unequivocal about an increase in fluidity, we can
say with confidence that nowhere (with the possible exception of post-Communist Hungary) is
there any evidence of a trend in the opposite direction.9
For women the picture is very similar. Of
18 decade to decade changes, two of them show a decline in fluidity (Germany between the
1980s and 1990s, Britain between the 1970s and 1980s) while 14 show an increase.
9
Our belief that this change in Hungary might indeed reflect an underlying increase in the rigidity of the mobility
regime is given support from a recent finding by Gerber and Hout (forthcoming) of a decline in Russian fluidity in
the 1990s.

11
Previous research
These results differ quite substantially from those of the previous major comparative study of
social mobility, namely The Constant Flux, in which Erikson and Goldthorpe argue strongly in
favour of the so-called FJH hypothesis of a basic similarity in social fluidity in all industrial
societies `with a market economy and a nuclear family system' (Featherman, Jones and Hauser
1975: 340) and they also claim (1992: 367) that `relative rates possess a high degree of temporal
stability'. Indeed, our results are somewhat closer to those of the slightly earlier comparative
study by Ganzeboom, Luijkx and Treiman (1989). They used 149 mobility tables for men drawn
from 35 countries spanning the period 1947-86. They claim that their show that `although ...
there is a basic similarity in mobility patterns ... at the same time there are substantial cross-
national and cross-temporal differences in the extent of mobility.' Furthermore, `a smaller but
significant part of the variance in mobility regimes can be explained by the trend towards
increasing openness over time' (Ganzeboom, Luijkx and Treiman 1989: 47). However, although
our results and GLT's results are rather similar, it should be noted that the latter have been
heavily criticised, notably by Wong (1994: 138) who reanalyses their data and finds that `the
model of temporal invariance cannot be rejected for a majority of countries ... Hungary and
Sweden are the only countries giving irrefutable evidence of temporal variation.'.
Why do we find evidence for change and cross-national difference when Erikson and Goldthorpe
did not? There are three main reasons, some methodological, others substantive. As far as
temporal change is concerned, because Erikson and Goldthorpe have only one mobility table per
country, their claim that there is little systematic temporal variation in patterns of social fluidity
within countries is based on analyses in which age groups are taken to represent different birth
cohorts. But as Breen and Jonsson (2003) have pointed out, this approach confounds lifecycle
and cohort effects and makes no allowance for either selective mortality or recall errors. In an
assessment of the reliability of measures of class origin and class destination in mobility tables,
Breen and Jonsson (1997) found that reliability was lower for origin information from older age
groups, implying that the common practice of using age groups to draw conclusions about cohort
change over time in mobility regimes may be unsound. More simply, one cannot properly make
inferences about change from cross-sectional data.

12
An important reason why we observe a trend towards increasing social fluidity and Erikson and
Goldthorpe do not is that our data, by and large, refer to later-born cohorts. Although our
analyses, like those of the overwhelming majority of mobility studies, are based on period data,
there is, I believe, good reason to suppose that social fluidity is driven by cohort, rather than
period, effects.
Erikson and Goldthorpe's data represents cohorts born between, approximately, 1900 and 1945,
whereas our data extends this to cohorts born around 1970. This means that our data contain a
larger share of people who benefited from what has been called the Golden Age of Welfare
Capitalism ­ in other words, the long post-war economic boom and the generally more
egalitarian educational and social welfare policies that followed in its wake. Work that I have
done with Janne Jonsson on Sweden clearly shows that here the major reduction in class
inequalities in educational attainment took place among cohorts of people born between 1910
and 1950 and that since then there has been no further equalization. But this means that, as we
compare period data for the 1970s, 1980s and 1990s, we gradually lose the older, less fluid
cohorts, who are replaced by younger, more fluid cohorts. In the case of Sweden such a process
will continue until the second decade of the 21st
century, after which we expect to see no further
increase in period social fluidity. But it seems likely that the specific cohorts which benefited
from the Golden Age will have differed between countries, so leading to variation between them
in the fluidity they display in particular periods.
[FIGURE 6 HERE]
Whereas we argue that countries display considerable variation in social fluidity, Erikson and
Goldthorpe argue the contrary. In this case, Erikson and Goldthorpe are simply mistaken in the
interpretation of their own results. They report (Erikson and Goldthorpe 1992: 381) that the
logarithm of each odds ratio in their mobility table from Czechoslovakia (which is the country in
which they find the greatest fluidity) is two-thirds as strong as in Scotland (the country with the
least fluidity). So, for example, given an odds ratio of 3 in Czechoslovakia, the comparable odds
ratio would be 5.3 in Scotland. On this basis it is difficult to argue for commonality in social

13
fluidity, and, indeed, these results are quite similar to ours: in our data the logarithm of the odds
ratios in Israel are around half those in Germany. So, an odds ratio of 3 in Israel would be an
odds ratio of 7 in Germany. Given this, one might ask why Erikson and Goldthorpe conclude that
there is little variation in fluidity. The reason is the following. All comparative mobility studies
find that the difference in the goodness of fit of between a model in which social fluidity is
common across all countries (the model of Common Social Fluidity or CnSF) and a model in
which fluidity can differ between them, is small, relative to the difference between the CnSF
model and one in which origins and destinations are independent. It is certainly true that the
model of variation in fluidity is a significantly better fit to the data than CnSF, but this
incremental improvement to any goodness of fit measure is small. For example, when we
analyzed our data according to decade, a very small index of dissimilarity was returned by a
model that allowed for no temporal or cross-national variation in social fluidity (3.95 per cent for
men and 3.81 for women) and allowing for such variation only improved  by, at most, two
percentage points. This compares with a  of around 15 per cent in models in which origins and
destinations are independent. Much the same picture emerged when we used annual data, and
arguments like this usually lead to the conclusion that most social fluidity is common and
invariant over time. Sometimes the same point is made using the deviance, rather than , as the
yardstick, and here the result is even more extreme. For example, 90 per cent of the deviance
returned by the model of independence disappears when we add common (among countries and
over decades) social fluidity, and a model allowing fluidity to change over time and differ
between countries improves it only by a further seven per cent. Taken together, the
decompositions of the deviance and of  would seem to indicate that more than 85 per cent of
social fluidity is common over nations and time.
This then seems to conflict with our finding of large variation between countries in their odds
ratios. The problem with using arguments based on  or the deviance to support the argument
that social fluidity is largely invariant is that variation in fluidity is assessed using measures of
the consequences of fluidity for the whole mobility regime (i.e. for absolute mobility), and, as we
have seen, these consequences are unanimously agreed to be quite minor. An analogy may help
to make the point. In a linear regression, Y =a + bX; X (analogous to social fluidity) may display
a lot of variation, but it will have little impact on Y (analogous to overall mobility) if the

14
coefficient, b, is close to zero. But simply because b is small we could not then claim that there
was little variation in X. Measures such as the change in  or in G2
might be said to capture the
strength of effect of fluidity on overall mobility, but they do not measure the variation in fluidity
itself, and it is therefore mistaken to conclude, on this basis, that social fluidity is common and
invariant. It is empirically the case that, between countries or over time, large variations in social
fluidity can be found which nevertheless have little impact on the overall mobility regime. To
illustrate this: if we take the fluidity pattern from the 1997 Italian men's table in our data and
insert it into the 1991 Israeli men's table, while preserving the Israeli marginal distributions, the
 between the real and the constructed Israeli tables is six per cent.10
When we consider that the
Israeli and Italian mobility regimes are close to the extremes of the range of fluidity found in our
data this suggests that six per cent represents the maximum impact of differences in fluidity on
the distribution of individuals in the mobility table. The conclusion to be drawn from these
apparently contradictory measures of the variation in fluidity is not that it is common or
invariant, but, rather, that even quite substantial differences in fluidity have little impact on the
distribution of cases over the mobility table ­ i.e. on observed, absolute mobility flows.
Explanation
Thus far I have said nothing about how we might explain patterns of social mobility or social
fluidity, aside from some rather vague remarks about the impact of policies pursued during the
Golden Age of Welfare Capitalism. In broad terms, explanations of variations in social fluidity
are usually pitched at a macro-sociological level: that is, fluidity is assumed to be related to
other, societal level measures. The very influential `liberal theory of industrialism' for example,
argues that fluidity will increase with economic development. In our case, the stage of economic
development of our countries varies rather little, but, even so, we could find no evident link
between their ranking in fluidity terms and their GDP per capita. Nor did we find any support for
Erikson and Goldthorpe's argument that fluidity is related to income inequality: there is no
significant association between a country's fluidity and its Gini coefficient. Overall we could
discern no tendencies towards either convergence or divergence in fluidity, and thus the
hypothesis that, as nations have come to follow different policy trajectories ­ particularly in
10
We use the observed Italian fluidity pattern, and thus the magnitude of the difference that we report does not
depend on the adequacy of any particular model of fluidity.

15
economic policy ­ so we might see growing differences between them in fluidity, receives no
support. There is some indication, however, that fluidity is greater in state socialist (Poland and
Hungary) and social democratic (Norway and Sweden) countries, and the argument for such a
political explanation receives additional support from the finding of declining fluidity in
Hungary during the 1990s. But, on the other hand, we observe very high fluidity in Israel and
data from the General Social Survey (made available to us by Mike Hout), shows that fluidity is
high in the United States. This leads to the conclusion that direct political intervention of the
kinds associated with state-socialist and social democratic societies may be one means by which
a society can reach relatively high rates of fluidity, but it is not the only one.
One major difficulty in devising theories to explain variation in mobility or fluidity is that
mobility tables, especially period-based tables, reflect a large number of underlying processes ­
artefactual, contingent and substantive. For one thing, this aggregation of processes renders it
difficult to explain variations in fluidity; for another, it may also be the case that some of the
commonality that has often been observed in comparisons of social fluidity derives from the
mixing together in the mobility tables of processes that, when investigated separately, might
show greater and more systematic societal and temporal differences.
While it is reasonable to suppose that fluidity in a nation is shaped by government policy, the
education system, the workings of the labour market, and suchlike, it is also the case that what
we observe in a mobility table may also reflect purely artefactual sources of variation arising
from differences in the way that the data themselves represent the underlying phenomenon of
interest. Furthermore, what we might call contingent factors, which are usually omitted from any
theoretical discussion of social fluidity, may play an important role in shaping what we observe.
The chapter on Germany in the current project provides a good example. Walter Müller and
Reinhard Pollak attribute the high fluidity they find among people born in the 1920s to the
massive migration from the eastern part of Germany that occurred following the Second World
War. The measured class origins of this cohort are thus their pre-migration origins, which had
very little relevance in shaping their subsequent mobility patterns: the physical detaching of a
large share of the cohort from their true origins led to higher measured social fluidity. The same

16
argument may explain the high level of fluidity in Israel, a country in which a very large share of
the population is comprised of immigrants.
Origins, Education and Destinations
Perhaps the simplest model of the mobility process that sociologists and others have used is the
so-called `OED triangle' illustrated in Figure 6. This is an attempt to capture the main paths that
link class origins with class destinations. It is widely accepted that educational attainment is the
major factor in mediating social fluidity (Ishida, Müller and Ridge, 1995; Marshall, Swift and
Roberts, 1997), and the OED triangle allows for this by positing an effect of class origins on
educational attainment (arrow A) and an effect of education on class destinations (B). Aside
from this, there is then a residual direct effect from origins to destinations (C) which captures all
that part of the origin ­ destination association that is not mediated through education. Of course,
the model could be expanded to allow separate paths for other factors that have been identified as
mediating the origin ­ destination association and in this way make it similar to the more
complex path-analytic models associated with work in the status attainment tradition, a tradition
initiated by Blau and Duncan (1967).
[FIGURE 7 HERE]
In the absence of well developed and testable behavioural theories of the social fluidity regime, a
first step in furthering our understanding would be to determine the degree to which, in each
country, changes in fluidity are driven by changes in each of these paths. A second step would
then be to seek to account for them, whether this is in terms of changes in the impact of `factors
of production' or through some other means. In the log-linear and log-multiplicative modelling
framework in which we, and the authors of the country chapters, have been working, although it
is possible to estimate models for all paths of the OED triangle, it is not possible to carry out
what is known as `path analytic' decomposition. In this instance, a path analytic decomposition
would measure the direct impact of class origins on destinations (path C) and its impact via
education (paths A and B). As a result, although we can discuss trends in each of these paths, we
cannot (though see the appendix to this chapter) make definitive assessments of their relative
importance for social fluidity.

17
Six of our country chapters analyze the role of education in social fluidity, though in the German
case, a cohort, rather than a period perspective is taken. For the other five, the country chapters,
together with other published research and some additional analyses that we have carried out
(and which are available on request from the authors), allow us to draw the following
conclusions about the three paths shown in Figure 6:
(a) Origins to education (path A in Figure 6): class inequality in educational attainment has
declined in this period in France, Sweden, and the Netherlands but not in Ireland or
Britain.
(b) The effect of education on class destination, controlling for class origins (path B), has
grown weaker over the period in France, Sweden, Ireland (see Whelan and Layte 2002),
Britain and the Netherlands.
(c) The partial effect of origins on destination, controlling for education (path C), remains
constant in Ireland and Britain but declines in the Netherlands.
(d) In the French case, Vallet reports a compositional effect deriving from an interaction
between origins, destinations and education. The association between origins and
destinations is weaker among people in higher educational categories, and, as more
people reach those categories, so there is an overall reduction in the strength of the
association between origins and destinations. Hout (1988: 1388) earlier attributed some
of the increase in social fluidity he observed in the USA to this compositional change.
Our own analyses show that this effect is also present in Sweden.11
(e) It is well known ­ and several of the country chapters confirm it ­ that education mainly
mediates the hierarchical component of mobility and has little or no effect on other
elements, particularly the tendency for self-recruitment among farmers and the petty-
bourgeoisie.12
Our own analyses (described in the appendix to this chapter) suggest that
the overall extent to which education mediates the impact of origins on destinations
increased over the last decades of the 20th
century but continues to vary considerably
11
If, in the French and Swedish cases, we did not take account of this effect by including a three-way interaction
between origins, destinations and education in our model, then it would appear as a declining partial effect of origins
on destinations (that is, the same change as we observe in the Dutch data).
12
The Irish case may be thought typical in this respect: `Education served to mediate about half of the effects
associated with position in the class hierarchy. However, it played almost no role in accounting for the inheritance or
property effects that also serve to determine class outcomes' (Whelan and Layte, this volume: 000).

18
between countries. Its role is greatest in Sweden (which might therefore be described as
the most meritocratic of our countries) and weakest in Britain.
In summary, we find several different mechanisms through which the increase in social fluidity
in France and the Netherlands and possibly Sweden, and its constancy in Britain and possibly
Ireland, might be explained. In all five countries, we observe a weakening of the link between
education and class destination, but in France, Sweden and the Netherlands we see two further
effects neither of which is found in Britain or Ireland. First, the link between class origins and
educational attainment has weakened; and, secondly, the direct partial effect of origins on
destinations, controlling for education, has also declined. In France and Sweden (though not in
the Netherlands) this seems to be due, at least in part, to the growth in the proportion of people
with higher levels of educational attainment.
Applying such arguments to the OED triangle we should expect a weakening of paths A and C
and a strengthening of path B. What we in fact see is that all the paths either show a tendency to
remain unchanged or to weaken. This certainly implies declining ascription, and, indeed, we
have seen a general tendency for social fluidity to increase. But it does not imply a growth in the
importance of achievement, at least as this is captured in our measure of educational
qualifications. Furthermore, although, as we noted earlier, education is considered to be the
major factor mediating social fluidity, our results show that it nevertheless plays a minor role
when compared with the direct partial effect from origins to destinations. And it is this path, of
course, which captures the workings of all the heterogenous factors that Bowles and Gintis's
(2002) arguments would point towards as important determinants of the association between
origins and destinations.
The complexity of social fluidity, especially in a period perspective such as we have adopted
here13
, makes it resistant to simple explanation. We have seen that fluidity can and does change
for several reasons, and the end result is a consequence of several diverse processes. This means
13
A period perspective means focusing on change over historical time as opposed to, say, a cohort perspective,
according to which we would compare mobility among groups born at different times. There are strong arguments
for focusing on cohorts as well as periods in mobility analysis. The German chapter in this volume provides a good
example in which changes in fluidity ­ first a decline then an increase ­ can be attributed to specific historical events
that affected particular birth cohorts but which, because period data aggregate the experiences of different cohorts,
cannot be seen there.

19
that, as far as policy prescriptions for raising the level of social fluidity are concerned, things are
equally complex. In our analysis of the OED triangle we found a consistent weakening of the
link between education and destination. As long as education is positively correlated with class
origins, a decline in the positive partial association between education and destination, holding
constant the partial origin ­ destination association, should result in an increase in fluidity. But,
not only is this effect not well understood, it does not lend itself to any policies that a
government might want to encourage and, indeed, by itself it may not always be sufficient to
increase fluidity significantly, as the British case shows. This leaves three mechanisms, any of
which is able to contribute to greater fluidity. In those cases where social fluidity is greater
among those with higher educational qualifications, a simple change in the distribution of
education towards a greater share of more highly educated people can cause a general rise in
fluidity. This seems to have been particularly important in France and, adopting a cohort, rather
than a period, perspective, Breen and Jonsson (2003) show that changes in fluidity between
successive Swedish birth cohorts can largely be attributed to changes in the distribution of
educational attainment. But a necessary condition for this is that the origin ­ destination
association should indeed differ by educational level, and there is no reason to suppose that this
will always be the case, as the Dutch example shows. Furthermore, a policy to increase
enrolments in higher education with a view to increasing social fluidity will not be effective if
this also changes the degree to which labour markets for the more highly educated operate on a
meritocratic basis. In fact, Vallet finds exactly this trend in France: `as education has expanded
and the highest educational categories have grown in size, the capability of advanced education
to weaken the `ascriptive effect' has declined.'
The second mechanism seems to have been partially responsible for the increase in Dutch
fluidity: this is the weakening impact of origins on destinations when the effect of education is
taken into account. Such a change is capable of exerting a large effect on social fluidity, though
this may be unsurprising given that this `residual' path captures all the non-educational
influences on social fluidity. These include avenues of inter-generational transmission based on
the inheritance of property, on unmeasured (in mobility studies) factors that may be contextual
(such as access to particular networks), individual (preferences and abilities whose effects are not
mediated via education), and processual (discrimination and the hiring practices of employing

20
organisations), as well as any contingencies that induce an association between origins and
destinations. Evidently what is required is some understanding of the exact nature and relative
importance of these which would then yield a basis on which to assess whether and how they
might be susceptible to deliberate change.
Lastly, a decline in the association between class origins and educational attainment will also
tend to lead to greater fluidity, but we should be cautious about the possible extent of this. For
one thing, as the effect of education on destination also diminishes, changes in the origin ­
education association will have a smaller payoff. In addition, the effect on social fluidity of
changes in the origin to education and education to destination paths will depend on how much
fluidity is accounted for in this way. In Sweden, a great deal of it is mediated in this way, and so
further reductions in class inequality in educational attainment will be more consequential for
social fluidity here than would the same reductions in, say, Britain.
Conclusions
The experience of this project should lead us to question the balance that mobility research has
struck between social fluidity and absolute mobility. The emphasis has lain heavily on the former
but, insofar as we are concerned with the mobility regime, this now seems inappropriate. This is
by no means to deny that social fluidity tells us important things about the prevailing degree of
inequality in the chances of attaining one class position rather than another,14
and may be
indicative of other characteristics of society. Nevertheless, although one would not want to say
that fluidity can never make a difference (since we can easily construct examples in which
extreme patterns of fluidity will be highly consequential for the distribution of cases in a
mobility table), within the advanced industrial and post-industrial societies, the range of fluidity
that we observe is relatively inconsequential in determining variation in mobility flows and in the
life chances of individuals and families as these are captured in measures of class position. Many
previous authors (such as Grusky and Hauser 1984; Goldthorpe 1985) have called for more
attention to be paid to structural change, but, as Erikson and Goldthorpe (1992: 104, 189)
suggest, it is not clear how such change should be explained nor, indeed, whether it might not
14
And, for this purpose, odds ratios are an appropriate object on which to focus since, as Marshall and Swift (1996:
376) put it, `the concept of equality is inherently comparative: it necessarily invites us to ... assess (the advantages
of different groups) relative to one another' (parentheses added).

21
better be approached as a matter of historical description rather than sociological explanation.
But while this might be a valid concern if we conceive of class structures as macro-sociological
phenomena, it may be less so, and may leave open the possibility of sociological explanation, if
we were to turn our attention to the detailed evolution of businesses and firms and of the jobs
that constitute classes.
Furthermore, it now seems to me that a period perspective on social mobility may not be the
most appropriate if, as I have suggested, social fluidity in driven by factors related to cohorts,
rather than periods. The cohort perspective on social mobility has not been entirely neglected,
but it has certainly played a very secondary role to the usual period view. Under the cohort
approach, change in the association between origins and destinations is believed to occur in
specific cohorts, while, from a period point of view, this association is considered to change
among all cohorts at similar historical points in time. The importance, for social change, of the
replacement of older by younger cohorts has been stressed by several sociologists (Mannheim
1952; Ryder 1965). In the specific field of mobility research, Erikson and Goldthorpe (1992), for
example, argue that equality of condition is essential for equality of opportunity, and that
changes in the latter will then predominantly take effect during a person's upbringing and
schooling. As a result, the overall level of inequality of opportunity in society will change mainly
through cohort replacement, with younger cohorts experiencing a different degree of inequality
of educational attainment than older. It has long been education is the major channel of social
mobility: if this is so, then social fluidity should respond to changes in the level of class
inequality in educational attainment, which, because educational inequality is itself a
characteristic of cohorts, implies a cohort explanation of change in fluidity.
Though rarely spelled out, there is a tension between the cohort interpretation of change in social
fluidity and the belief that period effects drive variation in the origin-destination association over
the lifecycle through labour market changes that simultaneously affect different birth cohorts.
We believe that there are strong arguments for adopting a cohort rather than, or in addition to, a
period perspective. It is widely agreed that changes in an individual's social class position are
relatively rare after the age of about 35 (Goldthorpe 1980: 51-2, 69-71, Erikson and Goldthorpe
1992: 72). Such stability implies that, except in unusual circumstances in which this stability is

22
disrupted, period effects will be a less important source of change in fluidity when compared
with cohort replacement and within-cohort change that occurs during the early years in the
labour force.
Let me end with two examples to suggest the usefulness of a cohort approach. Earlier I said that
we observe no period change in the German data ­ i.e. between the 1970s and late 1990s.
However, in work that they have carried out more recently, Müller and Pollak now claim to find
period change when they include data collected since 2000. The reason for this (go to slide 15) is
that the 1930-39 cohort, which shows a very low level of social fluidity, has finally exited the
labour force and has been replaced by a cohort with fluidity approximately the same as that of
the 1960-69 cohort.
I also pointed to the Netherlands as one of the countries in which we observe a marked increase
in period fluidity ­ and in this respect it contrasts sharply with Germany. Furthermore, fluidity
increased both between the 1970s and 1980s and between the 1980s and 1990s. But again, cohort
replacement is a very plausible explanation, as we can see in slide 17. In the 1970s, the first five
cohorts (born between 1906-15 and 1946-55) are represented in the period data. In the 1980s we
lose the 1906-15 cohort which is replaced by the very much fluid 1956-65 birth cohort, and in
the 1990s the 1916-25 cohort is replaced by the 1966-74 cohort ­ again a much more fluid
cohort is replacing a less fluid one.
Both these accounts are consistent with education acting as the driving force leading to change in
fluidity. It perhaps will not come as a surprise to hear that I am now engaged on comparative
mobility research from a cohort perspective and, although this approach is far from
unproblematic, it does have one distinct advantage over the usual period approach inasmuch as it
allows us to observe processes operating over a much longer period: in the Dutch case the period
view lets us look at 1970 to 2000, but we have cohorts born between 1906 and 1974. So if we
believe that social change usually operates rather slowly, then we are more likely to find
evidence for it when viewed on this larger time scale.
REFERENCES

23
Breen, Richard (ed.). 2004. Social Mobility in Europe. Oxford: Oxford University Press.
Breen, Richard and Jonsson, Jan O. 2003. `Period and Cohort Change in Social Fluidity: Sweden
1976-1999'. Unpublished paper.
Bowles, Samuel and Gintis, Herbert 2002. `The Inheritance of Inequality'. Journal of Economic
Perspectives 16: 3-30.
Erikson, R. and Goldthorpe, J. H. 1992. The Constant Flux: A Study of Class Mobility in
Industrial Societies. Oxford: Oxford University Press.
Featherman, D. L., Jones F.L. and Hauser R.M. 1975. `Assumptions of mobility research in the
United States: the case of occupational status'. Social Science Research 4: 329-60.
Gerber, T. and Hout, M. 2003. `Tightening Up: Social Mobility in Russia, 1988-2000'.
Unpublished paper.
Goldthorpe, J. H. 1985. `On Economic Development and Social Mobility'. British Journal of
Sociology 36: 549-573.
Goldthorpe, J. H. and Mills, C. 2004. `Trends in Intergenerational Class Mobility in Britain in
the Late Twentieth Century.' in R. Breen (ed.) Social Mobility in Europe.
Goldthorpe, J.H., Yaish, M. and Kraus, V. 1997. `Class Mobility in Israeli Society: A
Comparative Perspective'. Research in Social Stratification and Mobility 15: 3-27.
Grusky, D.B. and Hauser, R.M. 1984. `Comparative social mobility revisited: Models of
convergence and divergence in 16 countries'. American Sociological Review 49: 19-38.
Hout, M. 1988. `More universalism, less structural mobility: the American occupational structure
in the 1980s'. American Journal of Sociology 93: 1358-1400.
Ishida, H., Müller, W. and Ridge, J. 1995. `Class Origin, Class Destination, and Education: A
Cross-national Study of Industrial Nations'. American Journal of Sociology 101: 145-193.
Jonsson, J.O. 2004. `Equality at a halt? Social mobility in Sweden, 1976-1999.' in R. Breen (ed.)
Social Mobility in Europe.
Layte, R. and Whelan, C.T. 2004. `Class Transformation and Trends in Social Fluidity in the
Republic of Ireland 1973 to 1994'. in R. Breen (ed.) Social Mobility in Europe.
Lipset, S. M. and Zetterbeg, H.L. 1959. `Social mobility in Industrial Societies'. Pages 11-75 in
S. M. Lipset and R. Bendix eds. Social Mobility in Industrial Society. Berkeley: University of
California Press.

24
Mach, B. 2004. `Intergenerational Mobility in Poland: 1972 - 1988 ­ 1994.' in R. Breen (ed.)
Social Mobility in Europe.
Marshall, G. and Swift, A. 1996. `Merit and Mobility: A Reply to Peter Saunders'. Sociology 30:
375-86.
Marshall, G., Swift, A. and Roberts, S. 1997. Against the Odds? Social Class and Social Justice
in Industrial Societies. Oxford: Clarendon Press.
Müller, W. and Pollak, R. 2004. `Social Mobility in West Germany: The long arms of history
discovered?' in R. Breen (ed.) Social Mobility in Europe.
Vallet, Louis-Andre, 2004. ; Change in Intergenerational Class Mobility in France from the
1970s to the 1990s and its Explanation: An Analysis Following the Casmin Approach' in R.
Breen (ed.) Social Mobility in Europe.
Whelan, C. T. and Layte, R. 2002. `Late Industrialisation and the Increased Merit Selection
Hypothesis: Ireland as a Test Case'. European Sociological Review 18: 35-50.
Xie, Y. 1992. `The Log-Multiplicative Layer Effect Model for Comparing Mobility Tables'.
American Sociological Review 57: 380-395.
Yaish, M. 2002. 'The Consequences of Immigration for Social Mobility: The Experience of
Israel'. European Sociological Review18: 449-72.