AJS Volume 112 Number 6 (May 2007): 1775­1810 1775
2007 by The University of Chicago. All rights reserved.
0002-9602/2007/11206-0004$10.00
Explaining Change in Social Fluidity:
Educational Equalization and Educational
Expansion in Twentieth-Century Sweden1
Richard Breen
Yale University
Jan O. Jonsson
Stockholm University
The authors analyze social fluidity among Swedish men and women
using a series of 24 annual surveys, 1976­99 (Np63,280). A theoretical
model suggests that changes in fluidity are normally driven
by cohort rather than period effects. The results support this argument:
changes in fluidity between the mid-1970s and late 1990s
were due to the successive replacement of older and less fluid, by
younger and more fluid, cohorts. Cohorts differed in their fluidity
because the effect of class origins on educational attainment declined
(an equalization effect) and because greater shares of each cohort
had higher levels of educational attainment, which placed them in
labor markets that operate more meritocratically (a compositional
effect). The article discusses the relevance of these results for other
countries and for policy.
INTRODUCTION
In studies of social mobility, Sweden has long been recognized as holding
a distinctive place: class origins (the social class in which a person is
1
Versions of this article were presented at the meeting of International Sociological
Association Research Committee 28, Mannheim, April 2001, and at seminars at Yale
University in October 2005 and the University of Cambridge in January 2006. We are
grateful to the participants on these occasions for their comments and suggestions. We
extend particular thanks to Robert Erikson, David Firth, Ruud Luijkx, Louis-Andre´
Vallet, and the AJS reviewers for helpful comments on earlier drafts. Jonsson acknowledges
financial support from the Swedish Council for Working Life and Social
Research (FAS D2001-2881 and D2002-2893). Direct correspondence to Richard Breen,
Yale University, Department of Sociology, New Haven, Connecticut 06529-8265. Email:
richard.breen@Yale.edu, janne.jonsson@sofi.su.se
.
American Journal of Sociology
1776
brought up) appear to have a smaller influence on class destinations (the
class which the person comes to occupy as an adult) than in most other
countries (Erikson and Goldthorpe 1992; Breen 2004), and there has been
a fairly lengthy period during which the impact of origins on destinations
has steadily weakened. In this article we reaffirm this trend, but we go
beyond most research in social mobility by developing a theoretical model
of how class origins are linked to class destinations and in providing an
explanation, which we test empirically, of the trends that we find. It is
no secret to say that contemporary research in social mobility is, for the
most part, technically sophisticated but theoretically weak. In this article,
we work to strike a better balance by providing a firm theoretical footing
for the empirical study of social mobility.
The term "social fluidity" is often used to refer to the degree to which
class destinations depend on class origins: the weaker the statistical association
between origins and destinations, the greater social fluidity is
said to be.2
Accordingly, 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, and studies
of whether societies are moving toward greater "openness" use social fluidity
as a key indicator. Whether countries differ substantively in their
degree of social fluidity and whether fluidity changes over time are questions
that have been much debated in the stratification literature.Research
in the 1980s and early 1990s (e.g., Grusky and Hauser 1984; Erikson and
Goldthorpe 1992) found support for the "FJH hypothesis," or variants of
it, which claims that social fluidity is similar in all industrial societies
"with a market economy and a nuclear family system" (Featherman, Jones,
and Hauser 1975, p. 340). Some of these studies also argued for "a high
degree of temporal stability" (Erikson and Goldthorpe 1992, p. 367; also
Goldthorpe 2000, chap. 11) in social fluidity. But particularly in more
recent research, greater emphasis has been placed on change and variation.
Ganzeboom, Luijkx, and Treiman (1989, p. 47) claimed that "there
are substantial cross-national and cross-temporal differences in the extent
of mobility," and Breen (2004; also Breen and Luijkx 2004a) documents
significant variation in social fluidity among the countries of Europe and
a fairly widespread temporal trend toward greater fluidity in the closing
decades of the 20th century (see also the review in our earlier work [Breen
and Jonsson 2005]).3
Our findings in this article support the latter position insofar as we
2
In the log-linear modeling context in which this article is situated, the association
between origins and destinations is measured using odds ratios.
3
And some former proponents of the FJH hypothesis now reject it (see Hauser 1995,
pp. 176­77).
Educational Equalization
1777
report change in Swedish social fluidity over the course of the 20th century,
but we also provide a theoretical framework for understanding how social
fluidity may change, and we use this framework to account for the change
that we observe. Hitherto, the task of explaining temporal change within
a country has often been an ad hoc affair and, although some studies
(esp. Hout 1988) have provided valuable insights into the mechanisms
that might underlie change, it has not been possible wholly to account
for temporal trends by the addition of variables which represent plausible
processes by which fluidity might change. But this is what we do in the
present study: we show that variables capturing two simple processes can
indeed account for trends in social fluidity in 20th-century Sweden.
Previous research shows that the association between class origins and
destinations in Sweden was relatively stable during the first decades of
the 20th century (Carlsson 1958) but weakened during the postwar period
and up to the beginning of the 1980s (Erikson 1983, 1987).4
This trend
toward increased openness continued to 1991, particularly for women
(Jonsson and Mills 1993; Jonsson and Erikson 1997). In this article we
first examine changes in social fluidity among Swedish men and women
by comparing successive birth cohorts. This allows us to cover almost the
entire 20th century, because the oldest cohorts in our data were born
around 1912 and the youngest in the early 1970s. On the other hand, and
in contrast to this cohort perspective, we can also compare social fluidity
across the different surveys in which our data were collected. These were
carried out between 1976 and 1999, and so this period perspective allows
us to focus on changes that took place in the final quarter of the 20th
century. But the period and cohort perspectives are different ways of
examining the same data, and so the second goal of the article is to relate
the two. As we explain below, we believe that there are good reasons for
supposing that changes in fluidity are normally and mainly--though not
exclusively--driven by cohort-related, rather than period-related, factors.
If this is true we should expect that most, if not all, of any period change
that we detect will prove to be a consequence of changes occurring between
cohorts.
The third, and major, goal of the article is to account for the pattern
of change that we observe. We look at the role of education, given the
widespread recognition that it is one of the major channels through which
intergenerational class reproduction occurs (Ishida, Mu¨ller, and Ridge
1995), and we identify two processes through which education might cause
4
Ganzeboom et al. (1989), using five Swedish data sets collected between 1950 and
1983, claim that Sweden shared in the general worldwide trend to increasing fluidity
that they identify. Wong (1994), in his reanalysis of their data, found that this trend
was evident only for Hungary and Sweden.
American Journal of Sociology
1778
social fluidity to change. The first, which we call "equalization," is a decline
in the association between class origins and educational attainment: in
other words, class origins come to exercise a weaker effect on educational
attainment. In several European countries, including Sweden (as we document
below), reforms of the educational system have been undertaken
with exactly this goal. Equalization affects social fluidity because, for a
given association between education and class destinations, a lesser impact
of class origins on educational attainment will weaken the overall association
between origins and destinations. The second process we call "compositional":
if there is an association between origins, education, and destinations
such that the origin-destination association is weaker at higher
levels of education, and if educational expansion places increasing shares
of each cohort in those educational levels where the association is weakest,
then this compositional change can be expected to lead to an overall
reduction in the gross association between origins and destinations.5
A
three-way interaction between class origins, educational qualifications,
and class destinations may be present when, for example, higher qualifications
are a powerful signal for employers that leaves little leeway for
social network effects, or when the job markets in which degreeholders
operate are particularly meritocratic. A weaker origin-destination association
at higher levels of education is in fact reported from the United
States (Hout 1988), France (Vallet 2004), and Sweden (Erikson and Jonsson
1998), making it possible that an expansion of higher education across
cohorts in these countries led to increasing fluidity. In our analyses we
document the existence of both equalization and compositional effects,
and we also seek to quantify the importance of each in accounting for
cohort changes in social fluidity.
In the next section of the article we present a theoretical model of the
processes that underlie change and stability in social fluidity, and we
explain why these processes are more likely to manifest themselves as
cohort than as period effects. Successive sections present our data and
5
A stronger result can be shown in the case of linear systems. Let X, Z and Y be
continuous measures of, respectively, class origins, educational attainment, and class
destinations. We then have a two-equation stochastic system: (i) ;Z p a a X u0 1 1
and (ii) . By substituting (i) into (ii) and arrangingY p b b X b Z b XZ u0 1 2 3 2
terms we see that the derivative, (i.e., the unconditional effect of X on Y), isY/ X
given by , and the derivative of this with respect to a changeb b a b a 2b a X1 2 1 3 0 3 1
in the overall level of educational attainment (i.e., ) is given by . So if there is noa b0 3
interaction between origins and education ( ), changes in the mean level of ed-b p 03
ucation will not change the gross association between origins and destinations. But if,
as in the case under discussion, , such a change will reduce this association (theb ! 03
compositional effect). But the association will always change for a change in the origineducation
relationship (captured by , the equalization effect).a1
Educational Equalization
1779
our analyses, and a final section summarizes our results and draws
conclusions.
A THEORETICAL MODEL OF SOCIAL FLUIDITY
The association between origins and destinations (parents' class and respondent's
class) depends on the degree to which factors associated with
class position in the parental generation can influence which classes their
children come to occupy.6
Figure 1 provides a simplified depiction of this
process. In the parental and filial generations occupation of a particular
class position depends on assets (such as educational qualifications) and
gives rise to consequences (such as income). In the parental generation
the impact of a given asset on class position is labeled , and in the filiala
generation, , while the "return" to class in terms of consequences is giveng
by and by in the two generations.7
The extent to which a memberl J
of the filial generation possesses a particular asset depends on the direct
transmission of that asset between generations (this effect is labeled )t1
and on the degree to which the consequences of class position in the
parental generation are associated with the securing of more of this asset
in the filial generation (this effect is labeled ). The parameterst2
all play a role in social fluidity because they shape thea, l, g, t , and t1 2
two fundamental mechanisms that underlie it: (a) the association between
filial assets and parental class, which we call transmissibility (given by
),8
and (b) the association between filial assets and filial classat lt1 2
destinations, or class returns ( ). Together these determine the associationg
between class origins and destinations as mediated by a particular asset
(given by ). Variations in fluidity, over time or between coun-g[at lt ]1 2
tries, will be largely dependent on institutional arrangements and demographic
circumstances that determine the transmissibility of, and the
returns to, assets.
The direct transmission of assets (through ) includes the inheritancet1
of property, aspirations, and aptitudes and, perhaps most important, genetic
factors. The indirect transmission, via , captures the idea thatt2
occupying a particular class position entails consequences, which themselves
influence the accumulation of assets in the next generation. These
6
This section draws particularly on ideas in Breen (1997) and Bowles and Gintis (2002).
7
In some cases one might want to assume that one or both of these effects was constant
over generations: .a p g, and/or l p J
8
Transmissibility, as we define it, is rather than simply , because weat lt t t1 2 1 2
are interested in the intergenerational transmission of assets only insofar as it contributes
to the association between parental class and filial assets. From this perspective
transmissibility must include the link between parental class and parental assets.
American Journal of Sociology
1780
Fig. 1.--Influence of parental position on class location of offspring
consequences may be psychological, as in the case of preferences shaped
by class position and transmitted to children, or material, as in the case
of the rewards that accrue to particular classes. How far such consequences
can influence the assets for mobility possessed by the next generation
depends on the degree to which classes generate differential consequences
(which is captured in fig. 1 by ) and the extent to which thesel
can be translated into assets for the attainment of class positions by their
children ( ). Income is an obvious example. The greater the variation int2
income between classes and the more families can use their income to
purchase advantages for their children, the stronger will the origin-destination
association be. Such would be the case where class differences
in income are great and public provision of education is poor: here high
income can be used to purchase education that then functions as an asset
for achieving a privileged class position. But if progressive taxation reduces
the inequality between classes in their market income ( is small)l
and/or the public educational system is as good as the private system
( is small), then the association between origins and destinations will bet2
correspondingly weaker. A similar argument can be advanced in respect
of other institutions of the welfare state that reduce the extent to which
inequalities in rewards in the parental generation translate into a differential
distribution of mobility assets among their children.
Change in fluidity can come about through changes in either or both
of the two fundamental mechanisms mentioned above: the class returns
to assets and their transmissibility. Educational qualifications, which we
focus on in this article, may become less dependent on origins due to
school reforms, for example, which would suppress transmissibility. Or
education may become less important in the labor market, in which case
the class returns will decrease. In both of these cases, social fluidity will
increase. But in real life several assets are likely to be of importance, and
so changes could be hard to predict. In the case of two assets, such as
Educational Equalization
1781
social networks and educational attainment, we may find that even if the
transmissibility of education was decreasing, this could be counteracted
by increasing class returns to networks. And if the transmissibility of
networks was stronger than that of education, but neither changed over
time, social fluidity would in fact decline if the class returns to networks
increased relative to the returns to education.
A more complex situation arises if the return to one asset depends on
the returns to another asset. In this case, a shift in the distribution of one
of these assets can lead to a change in social fluidity. If social networks
are consequential for class position among those with low education but
not at high levels of education, and if the distribution of education changes
so that a larger share of the filial generation has higher education (even
though the correlation between education and parental social class remains
the same), then the average effect of social networks on class position
will nevertheless decline, and, therefore, social fluidity will increase.
While changes in social fluidity, according to our model, follow changes
in transmissibility or class returns, they could take effect in two major
ways, namely as cohort effects or period effects. The former works through
cohort replacement, the importance of which for social change has been
stressed by several sociologists (Mannheim 1952; Ryder 1965). For example,
increased social fluidity in the working-age population would come
about through the entry of more fluid younger cohorts and the exit of less
fluid older cohorts. Period changes, however, are differences in fluidity in
the population at different historical time points that arise from some
processes that affect all, or a large share, of the population, more or less
simultaneously. The crucial issue, then, seems to be whether changes in
transmissibility or class returns are spread across the entire working-age
population or whether their influence is restricted to those cohorts entering
the labor market.
We first consider changes in transmissibility: one can imagine circumstances
in which these yield period effects, such as when the inheritance
of property--an asset that may be transferred at any age--is suddenly
discontinued (an example is the postwar collectivization of the Hungarian
agricultural sector; cf. Erikson and Goldthorpe 1992, pp. 152­54). Nonetheless,
transmissibility is overwhelmingly likely to lead to cohort effects
for the simple reason that most transmission of assets occurs during childhood
and early adulthood, manifested particularly in educational
qualifications.
Changes in the class returns to assets will give rise to period effects on
fluidity provided that these changes influence a sufficiently large share of
the working population. Period effects are therefore most likely to be
visible given the large-scale restructuring of labor markets, which involves
the separation of workers from their current job and their reintegration
American Journal of Sociology
1782
into employment in a different class because the assets they possess no
longer secure the same class positions.9
We may observe this when, for
instance, certain areas of the economy, such as manufacturing or mining
in some industrialised countries, decline and their workers are relocated
from jobs in the skilled manual class to unskilled work, or when new
economic opportunities appear (as in the growth of the IT industry) which
make certain assets more valuable than hitherto. In recent times the restructuring
of the formerly state socialist economies of central and Eastern
Europe has provided striking examples. Gerber and Hout (2004) show a
period decline in social fluidity in Russia, and a similar finding is also
reported for Hungary by Ro´bert and Bukodi (2004). Again, however,
period change appears to follow rather uncommon events.
Changes in returns can lead to cohort, rather than period, effects when
changes in the returns to an asset take effect gradually, as when the
possession of an asset is associated with the occupation of a particular
class or classes among older workers but not among younger ones. An
example is found in the declining viability of small farms: typically, older
men continue to work as farmers despite having relatively small amounts
of land, whereas younger men who inherit such holdings are unlikely to
become farmers. Indeed, we suspect that changes in the returns to assets,
just like transmissibility, are more likely to appear as a cohort rather than
a period effect. Recall that by returns to an asset we mean the returns in
terms of the chances of acquiring a given class position. It is widely agreed
that changes in an individual's social class position are relatively rare
after the age of about 35 (Goldthorpe 1980, pp. 51­52, 69­71; Erikson
and Goldthorpe 1992, p. 72; Breen 1994). In Sweden, between circa 1960
and 1990, around 80% of gainfully employed men and women were class
stable between ages 35 and 40, and 80%­90% between ages 45 and 50
(Jonsson 2001, fig. 8.1). So any change in the class returns to different
assets is likely to have its strongest effects on that share of the workingage
population under about 35 years of age, with relatively little effect
on the majority, except in unusual circumstances.
Our conclusion is that period change in fluidity is likely to be the
exception rather than the rule and that most changes in fluidity that we
observe in stable democratic societies will arise from processes of cohort
replacement. Yet, studies of change in social fluidity using multiple observations
from a country have almost always adopted a period, rather
than a cohort, perspective (e.g., Luijkx and Ganzeboom 1989; DiPrete
and Grusky 1990; Jonsson and Mills 1993; Vallet 1999; and the contri-
9
Large-scale unemployment itself would not be sufficient to bring about an observed
period effect because the convention in mobility research is to assign the unemployed
and others without a job to the social class of their last occupation.
Educational Equalization
1783
butions to Breen 2004; exceptions include Hout [1988] and Hauser and
Huang [1997]), and often there has been little explicit consideration of
whether change is in fact driven by period or cohort effects or both.10
For
the Swedish case, however, we can now test the assertion that change in
social fluidity is predominantly driven by cohort replacement. Because
we have a series of repeated cross-sectional mobility surveys we can examine
period change and cohort change simultaneously. If, indeed, fluidity
is chiefly responsive to cohort-related factors, rather than to those associated
with periods, then period change should disappear when we control
for the former.
Given our theoretical model, and our conjecture that change in social
fluidity is likely to be cohort driven, what changes do we expect to see
in Sweden? First, there is reason to believe that a decline in , the returnsl
to class position, has led to increasing social fluidity. Erikson (1996)
showed that, over the course of the 20th century, more equal conditions
among social classes in Sweden led to more equal educational attainments
between children of different origins: this may have led to increasing
fluidity during the 1980s and 1990s because of the more egalitarian conditions
that prevailed during the childhood of the cohorts that entered
the labor market then (who were born in the 1960s and the 1970s) as
compared with the cohorts that they replaced (born approximately in the
1910s and 1920s). Political means toward this end include progressive
taxation and redistribution mainly through the welfare state, for example
by the introduction of the child allowance and various benefits in cash
and in kind (Korpi and Palme 2004).
Political strategies to equalize opportunity in Sweden were not exclusively
directed toward reducing inequality of condition. Policies also
sought to reduce the consequences of such inequality (captured by int2
fig. 1). These included the abolishing of fees for secondary and tertiary
education, the introduction of free school meals and health care in schools,
and free books and teaching aids in primary and secondary school, all of
which have made educational opportunities less dependent on economic
resources in the family of origin. Educational reforms, such as the comprehensive
school reform in the 1950s, have aimed at increasing opportunities
and educational attainment, particularly among those from less
well-off families (e.g., Erikson and Jonsson 1996).
However, in Sweden as in many other European countries, for many
years educational expansion per se was not an important strategy, and
this contrasts with the United States, where expansion played a central
role (Hout and Dohan 1996). Expansion could be seen as an attempt to
10
But see Gerber and Hout (2004), whose empirical approach to distinguishing cohort
from period effects is derived from our earlier work (Breen and Jonsson 2001).
American Journal of Sociology
1784
weaken the importance of both and because parental assets andt t1 2
resources alike would be less decisive for admission to higher education.
In our terms, expansion would promote equalization. But this supposes
that children of less advantaged class backgrounds are the primary beneficiaries
of the increased opportunities--something which is not necessarily
the case. Jonsson and Erikson (in press) show that in Sweden expansion
has been relatively unsuccessful in reducing educational
inequality, probably because middle-class children with lower grade levels
but high aspirations are often the first to take advantage of increasing
opportunities. Nevertheless, expansion may lead to a reduction in the
association between class origins and class destinations even though it
does not weaken the inequality between class origins and educational
attainment. This can come about if the conditions hold for the compositional
effect we discussed earlier, so the returns to assets other than
education (e.g., social networks) are higher at lower levels of education.
Then educational expansion may lead to increased social fluidity even
though it does not reduce the transmissibility of assets, but, rather, because
it alters the average class return to other assets.
In sum, we expect that social fluidity has increased in Sweden and that
this change has been largely driven by cohort replacement; and we believe
that the process behind this reflects a combination of educational equalization
and a compositional effect following educational expansion.
DATA AND VARIABLES
The data set is a compilation of the annual surveys of living conditions
(ULF) 1976­99, conducted by Statistics Sweden (Vogel et al. 1998). Each
survey is representative of the adult Swedish population, ages 15­75. The
sample fraction is around one one-thousandth and the yearly sample sizes
are around 6,000 respondents.11
Nonresponse rates vary between 15% in
the early surveys to around 22% in the later ones. Generally, the data
quality is high. Some basic requirements for inferring change from repeated
surveys are fulfilled: the same fieldwork and data preparation
organization has been used throughout, classifications have followed national
standards, and, since there has been a continuity of purpose, the
11
The 1976­79 surveys have a household design, and from 1980 there is a panel element
(where 50% of the sample is included in the sample eight years later). This may lead
to within-person correlations across observations and so we use weights to adjust
sample sizes accordingly. More specifically, respondents in households where both
spouses are interviewed are down weighted by 0.5, as are those in a panel wave who
have responded twice; those who have participated in three different waves are assigned
a weight of one-third for each wave.
Educational Equalization
1785
survey methods (face-to-face interviews) as well as the questionnaire design
have been largely the same during the whole period.
The social class schema that we use (Andersson, Erikson, and Wa¨rneryd
1981) is similar to the internationally used EGP class coding (Erikson and
Goldthorpe 1992; Erikson, Goldthorpe, and Portocarero 1979), and therefore
we use that notation.12
We identify six classes. As can be seen in table
1, the distribution of classes among fathers is significantly different from
that of sons and daughters.13
The table reflects the upgrading of the social
class structure with an increase in service-class jobs (classes I and II), as
well as a sharp decline of the farming class.14
Respondents' educational
attainment--coded into six categories, using the CASMIN (Comparative
Analysis of Social Mobility in Industrial Nations) educational classification
(Mu¨ller and Shavit 1998, table 1.2b)--is shown in table 2. To exemplify
the change in the educational structure we report the distribution
of educational credentials for the older and the younger parts of our
sample (born 1912­43 and 1944­74, respectively). The most noteworthy
change is the rapidly decreasing numbers who leave school after the
compulsory years and the expansion at tertiary levels--although, as can
be seen, Sweden still has relatively few people with university degrees.
12
Foremen, supervisors of manual workers, and lower grade technicians do not form
a separate class (class V in the EGP schema). Instead some more qualified supervisors
(such as verkma¨stare) go into class II, foremen are in general classified in class III,
and blue-collar technicians (a group that is relatively uncommon in Sweden) go mostly
into class VI. Occupations normally organized in the manual workers' trade union in
Sweden (LO) are classified into classes VI and VII, and these include some occupations
that in the EGP coding schema are found in the unqualified strata of the nonmanual
classes (IIIb). Among these are lower-grade salespersons and shop assistants as well
as lower-grade service workers (employed, inter alia, in hotels, restaurants, and in
offices) and nurse's aides. It should be mentioned that the dividing line between class
I and class II in the class schema was changed in 1982 leading to a small reduction
(around two percentage points) of the percentages in the higher class (changes concerned
both origin and destination classes). This is unlikely to have any major effects
on the results, and our inspection of the results shows no break in the OD associations
between the period up to 1981 and following 1982.
13
The basis for the origin classification is a question on father's main occupation and
employment status during the respondent's childhood (0­16 years of age). Unfortunately,
before 1984 information about mother's occupation was collected only for respondents
who did not live with their father during childhood, so only in those cases
is information on the mother used to indicate origin class.
14
The gradual decline of unskilled positions (both classes IIIb and VII) is evident from
more detailed analyses. Here, we have merged the skilled workers with the unskilled,
partly because we want to save degrees of freedom for our models, partly because
these classes show fairly similar intergenerational mobility propensities (as well as high
internal mobility). To avoid empty cells, we could also have merged farmers with the
other self-employed, but previous research shows important mobility barriers to exist
between jobs in the agricultural and nonagricultural sectors (Erikson and Goldthorpe
1992).
American Journal of Sociology
1786
TABLE 1
Marginal Distributions For Origin (Father's Class) and Destination Class
for Sons and Daughters
Class Description
Origin Destination
All Men Women
I ................ Professionals, higher executives
or administrators, employers
with 20 employees or more
6.0 14.7 8.1
II ............... Semiprofessionals, medium level
administrators and officials
9.1 17.2 17.5
IIIa ............ Lower routine white-collar
workers (jobs demanding
some qualifications)
6.6 7.1 12.3
IVab ........... Self-employed, employers with
1­19 employees
14.6 10.3 4.3
IVcd ........... Farmers 16.6 2.8 2.1
IIIb,VI, VII ... Unskilled workers in service,
manual workers (skilled and
unskilled)
47.0 47.8 55.8
Total ..... 99.9 100.0 100.1
Note.--EGP social class schema; figures relate to repondents who were between 25 and 64 years of
age in 1976­99.
It is notable that women, as in many other countries, have lower educational
qualifications than men in older cohorts, but higher educational
attainment than men in younger cohorts.
The analysis is confined to those ages 25­64 and undertaken separately
for men ( ) and women ( ). We group the data inton p 33,281 n p 29,999
four-year tables, giving us six periods (1976­79, 1980­83, . . ., 1996­99).15
Within each we identify 10 four-year age groups (25­28, 29­32, . . . , 61­
64) which allow us to define 15 overlapping age cohorts (see table 3). The
oldest cohort was born in 1912­18, the youngest in 1968­74. By definition
not all cohorts can be observed in every period: the oldest and youngest
are each observed only once, while the cohorts born in the 1930s and
1940s are observed in all six periods. The result is that the cross-tabulation
of origin (O) by destination (D) by cohort (C) by period (P) is not rectangular:
of the possible 3,240 ( ) cells of the table, 1,080p 6 # 6 # 15 # 6
are structural zeroes.
15
We used data for all men with a class code and all women who were not registered
as home workers because among home workers there are some cases that might have
their class position assigned to them on the basis of their spouse's occupation.
Educational Equalization
1787
TABLE 2
Educational Distributions for Men and Women, Born 1912­43 and 1944­74
Qualification Description
Men Women
1912­43 1944­74 1912­43 1944­74
1ab ............. Compulsory 40.0 15.9 38.6 12.4
1c ............... Lower vocational 23.7 35.6 32.9 40.3
2ab ............. Lower secondary 4.0 4.2 7.4 5.1
2c ............... Upper secondary 14.6 15.8 4.4 9.8
3a ............... Lower-level tertiary 8.3 15.4 8.7 18.8
3b .............. University degree 9.3 13.2 8.0 13.6
Total ......... 100.0 100.0 100.0 100.0
Note.--CASMIN educational schema.
RESULTS
Our analytical strategy is straightforward, but requires several steps. It
is also rather technical, and so we here summarize the logic of the analyses
and our main findings. We begin by testing for, and finding, change in
social fluidity across birth cohorts, and then we carry out three analyses
of changes between periods. In the first of these we ask whether, for each
birth cohort, there is evidence that its fluidity changes as it ages: we find
that there is not. Second, we use the panel element in the data to test
whether, among those individuals who were interviewed on two occasions,
patterns of fluidity differ at the later as compared with the earlier interview.
Again, we find that this is not the case. Third, we make simple
comparisons between each of the surveys in our data and ask whether
fluidity changes between them. In this case we find that it does. Having
found change between cohorts and between surveys we then ask whether
the latter is in fact driven by processes of cohort replacement, and so we
analyze cohort and period change in the same model. We find that when
we control for cohort changes, period differences vanish. The final set of
analyses then seeks to explain why fluidity changed over cohorts, and we
do this by modeling the equalization and compositional effects of education
and showing that, taken together, they are sufficient to account for
all the observed cohort change in social fluidity.
Cohort Change
Table 4 shows the goodness of fit of six log-linear models fitted to the
three-way ODC table.16
Model 1 is the model of common social fluidity,
16
Throughout we model social fluidity in a not very parsimonious way. As Hout (1988,
pp. 1374­76) has noted, more powerful tests would use fewer degrees of freedom. But
in this case, given our large sample size (63,280, almost 10 times bigger than Hout's
American Journal of Sociology
1788
TABLE 3
The Relationship between Cohorts, Age Groups, and Periods
Age
Groups
Periods
1976­79 1980­83 1984­87 1988­91 1992­95 1996­99
25­28 ... 10 11 12 13 14 15
29­32 ... 9 10 11 12 13 14
33­36 ... 8 9 10 11 12 13
37­40 ... 7 8 9 10 11 12
41­44 ... 6 7 8 9 10 11
45­48 ... 5 6 7 8 9 10
49­52 ... 4 5 6 7 8 9
53­56 ... 3 4 5 6 7 8
57­60 ... 2 3 4 5 6 7
61­64 ... 1 2 3 4 5 6
Note.--Nos. 1­14 in the table cells represent cohorts. Birth years are, for cohort 1, 1912­18; cohort
2, 1916­22; cohort 3, 1920­26; cohort 4, 1924­30; cohort 5, 1928­34; cohort 6, 1932­38; cohort 7, 1936­
42; cohort 8, 1940­46; cohort 9, 1944­50; cohort 10, 1948­54; cohort 11, 1952­58; cohort 12, 1956­62;
cohort 13, 1960­66; cohort 14, 1964­70; cohort 15, 1968­74.
stating that the OD association is the same in all cohorts: this can be
rejected using the standard chi-squared goodness of fit test.17
In model 2
a common pattern of local OD association is assumed to vary log multiplicatively
over cohorts in what has come to be known as a model of
uniform difference, or unidiff (Erikson and Goldthorpe 1992; Xie 1992).
Specifically, letting and index origins andi p 1, . . . , 6 j p 1, . . . , 6
destinations, respectively, and cohorts, we writek p 1, . . . , 15
lnv p b lnv . (1)ijFk k ij
Here denotes the odds ratiovijFk
F /Fijk ij k
, (2)
F /Fi jk i j k
where is the fitted value in the ijkth cell of the table and is a cohort-F bijk k
specific multiplier, and we set .18
The model says that the differenceb p 11
in the log-odds ratios between any two cohorts is proportional to the
sample), even the global tests that we use will have reasonable power to detect change.
In table 4 and all other tables we report both the deviance (likelihood-ratio chi-square)
and the BIC statistic (Raftery 1986). We base our model selection on the former; we
report the latter because it is a widely used criterion of fit, though, for reasons advanced
by Weakliem (1999), we do not make use of it.
17
Henceforth when we write that something is or is not significant we are referring
to the criterion.P ! .05
18
In table 3 and subsequently we attach a superscript to b to indicate the variable
over which fluidity is changing.
Educational Equalization
1789
TABLE 4
Goodness of Fit of Models for the Origin # Destination # Cohort Table
for Men and Women
No. Model df
Men
(n p 33,281)
Women
(n p 29,999)
Deviance P BIC Deviance P BIC
1 ... OD 350 402.86 .027 3,242 423.47 .004 3,185
2 ... ODbC
336 394.81 .014 3,104 366.26 .123 3,098
3 ... ODbC
k diag 330 315.85 .703 3,120 350.76 .207 3,051
4 ... ODkbC
diag 343 331.12 .667 3,240 358.79 .268 3,177
5 ... ODkbC
349 398.65 .033 3,235 379.01 .130 3,219
Note.--All models include the terms OC DC. O p origin; D p destination; C p cohort; diagp 6
parameters fitted to cells on main diagonal of the O-D table; .k p 1, . . . , 15
difference in their parameters, and thus declining values of this param-b
eter over cohorts correspond to increasing social fluidity. This model uses
14 degrees of freedom (i.e., number of cohorts minus one) more than the
independence model, but reduces the deviance by only eight for men. For
women, however, there is evidently more change as captured by the unidiff
model--the reduction in deviance is around 57, which is clearly a significant
improvement in fit. The third model introduces parameters for
each of the cells on the main diagonal of the mobility table. This imposes
the constraint that the sum, over all cohorts, of the frequencies in each
cell of the main diagonal of the origin by destination table is fitted exactly.
The log-odds ratios under this model are given by
lnv p b lnv d d d d ,ijFk k ij ij ij i j i j
d p 0, if i ( j. (3)ij
Here denotes the diagonal parameters. Because these parameters aredij
constant over cohorts, the difference between cohorts in their log-odds
ratios is still proportional to the size of their coefficients. As equationb
(3) makes clear, the diagonal effects operate over and above the evolution
of the pattern of local odds ratios and so, even though the diagonal parameters
do not vary over cohorts, this does not mean that propensities
for individuals to be found in their class of origin remain constant over
cohorts: rather, this propensity changes according to the common logmultiplicative
evolution of the whole pattern of local association. What
does remain constant is the tendency for class inheritance that exists over
and above that which is implied by this pattern.19
Adding these diagonal
19
We tested models in which the diagonal parameters were allowed to vary over
cohorts, but this never yielded a significant improvement in fit.
American Journal of Sociology
1790
parameters makes a substantial difference to the fit of the men's tables,
though rather less difference for women. This is not surprising, since it
is well known that men are more likely than women to be found in the
same class as their father (e.g., see Jonsson and Mills 1993 for the Swedish
case). In model 4 we impose a linear constraint on the evolution of the
parameters: so now we writeb
lnv p kblnv d d d d ,ijFk ij ij ij i j i j
d p 0, if i ( j, k p 1, . . . , 15. (4)ij
For both men and women model 4 is preferred to model 3 and to model
1. Last, model 5 removes the diagonal effects but retains the linear constraint.
This considerably worsens the goodness of fit of the model for
men, but has less impact on women, once again illustrating the lesser
importance of inheritance effects among women.
Figure 2 shows the parameter values for the diagonal cells generated
by model 3. As is evident, there is strong inheritance in the upper service
class--for women, in particular--and in the petty bourgeoisie; and among
men inheritance effects are very strong in the farming class.20
Of more
interest to us, however, are the coefficient estimates from model 3 andb
also the linear estimated from model 4, both of which are shown inb
figure 3. They display a trend toward increased social fluidity over cohorts,
with an estimated slope for the association between origins and destinations
of .036 for men and .043 for women, though the cohort-specific
s suggest that, from the 1948­54 cohort onwards, the trend toward moreb
fluidity disappears.21
It should be borne in mind that for the oldest and
youngest cohorts we have only one observation, two for the second oldest
and second youngest, and so on, so the end points on this and the other
figures depicting cohort change should be regarded as particularly subject
to uncertainty (and they also exercise less influence on the estimate of the
slope). If we compare the coefficients for the 1964­70 cohorts, we seebk
that they are two-thirds or less of the value for those for the 1916­22
cohorts, indicating a substantial increase in fluidity over the 20th century.
Age Effects
Having established that fluidity changed over cohorts in 20th-century
Sweden, we now turn to period change. The first question we address is
20
There is also some "disinheritance" among men in class II and women in class IIIa
(see also figs. 7 and 8, below), suggesting that these are origin classes which are disproportionately
likely to be vacated by those born into them.
21
We included a quadratic term in addition to the linear term for the evolution of the
log-odds ratios, but this was not a significant improvement for either sex.
Educational Equalization
1791
Fig. 2.--Parameter values for the diagonal cells generated by model 3
whether there is any indication of substantive changes in fluidity within
cohorts as they progress in their career: in other words, an age effect.22
We cannot address this issue for our very oldest and youngest cohorts,
for each of which we have only one table. Table 5 shows the result of
fitting, to each of the remaining 13 cohorts, a model that assumes constancy
over periods in the OD association (PO PD OD) and a model of
uniform change in fluidity across periods (PO PD OD P
). In every caseb
the model of common fluidity fits the data and the model of uniform
change never yields a significant improvement in fit. The result is overwhelmingly
comforting for the cohort view: there seems to be no substantive
change in fluidity across periods for people in a given cohort.23
This conclusion can be checked by using the panel element in our data,
though with a smaller sample.24
For some respondents we have information
on their class position at two points, eight years apart, which we
call D1 and D2. Information on D1 was collected at the initial ULF
interview between 1979 and 1991 and on D2 between 1986 and 1999.
22
An age effect cannot explain change over periods, since, by definition, an age effect
is specific to an age group but constant over periods and cohorts. Thus, to explain
change, age effects must change--as they would, for example, in an age by period
interaction.
23
Note that this does not mean that members of a cohort do not change occupations
or classes across their careers (they most certainly do, at least up to the age of 30­40),
only that these changes are unrelated to their social origins.
24
We thank an AJS reviewer for suggesting this analysis.
American Journal of Sociology
1792
Fig. 3.-- coefficient estimates from model 3; linear estimated from model 4b b
Respondents born before 1921 were too old, and those born after 1966
too young, to have a value on D2: furthermore, respondents born between
these dates may also not have a value for D2 depending on how old they
were when they were first interviewed. Finally, only half of each year's
sample was reinterviewed eight years later. As a result we have panel
data on 9,906 respondents (compared with the 63,280 used elsewhere in
our analyses).
Because our goal is to check whether the origin-destination association
is the same over the life course we form the table of origins by D1 by D2
for each sex, and distinguish two broadly defined cohorts. In the older
cohort both D1 and D2 occurred after the age of 34: it is made up of
those born 1921­44 together with those born 1945­48 who were 35 years
old or older at D1. In the younger cohort D1 occurred before 35. These
respondents were born 1949­66 or were born 1945­48 and were 34 years
old or younger at D1. Forming two cohorts in this way allows us to
distinguish those whose life-course mobility may have occurred before
the age of occupational maturity (which we take to be 35) and those
whose mobility, if any, occurred after this point. The small number of
cases forced us to move to a four-class categorization for origins, first
destination (D1), and second destination (D2). We distinguish class I, clas-
Educational Equalization
1793
TABLE 5
Social Fluidity Models Fitted to Each Cohort
Cohort Table
Deviances
Men Women
PO PD OD PO PD ODbP
PO PD OD PO PD ODbP
2 ....... 1916­22 2 15.42 15.32 25.19 19.02
3 ....... 1920­26 3 44.82 41.17 31.75 30.73
4 ....... 1924­30 4 64.57 64.24 66.41 65.89
5 ....... 1928­34 5 72.81 70.65 102.92 100.36
6 ....... 1932­38 6 119.65 112.28 131.04 125.34
7 ....... 1936­42 6 105.41 103.97 104.79 99.73
8 ....... 1940­46 6 109.11 108.18 137.88 134.43
9 ....... 1944­50 6 103.28 98.53 85.02 79.10
10 ...... 1948­54 6 119.55 113.17 136.71 129.14
11 ...... 1952­58 5 112.66 107.43 108.62 98.80
12 ...... 1956­62 4 61.24 61.07 70.57 68.34
13 ...... 1960­66 3 50.62 49.69 46.08 45.97
14 ...... 1964­70 2 22.27 22.17 22.78 22.20
Note.--Degrees of freedom for PO PD OD model p (no. of tables 1) # 25, and degrees of freedom
for PO PD ODbP
model p (no. of tables 1) # 24; df for the comparison of the two.
ses II and IIIa, classes IVa, IVb, and IVc, and classes IIIb, VI, and VIII.25
The test is simple: we fit the model OD1 OD2 D1D2 and we compare its
goodness of fit with the same model in which the OD1 and OD2 associations
are fixed to be the same (which forces the association between
origins and destinations to be the same for both destinations). The test
has nine degrees of freedom and in no case is the constrained model a
significantly poorer fit: the deviances are 13.25, 13.86, 10.37, and 14.87
for the older and younger cohorts of men and the older and younger
cohorts of women, respectively.26
Period Change
The absence of change within cohorts as they age does not mean, of course,
that social fluidity does not change across periods, and so table 6 uses the
same set of five models as table 4, this time applied to test for period
change, now ignoring cohorts. Evidence for period change is much weaker
than for cohort change, but adding parameters for the main diagonal once
25
The sample sizes in each of the origin by D1 by D2 tables are 3,493, 1,503, 3,375,
and 1,535 for older and younger men and older and younger women, respectively.
26
We carried out the same test using the model OD1 OD2 (i.e., omitting the D1D2
term capturing association in life course mobility) with the same results (details available
on request from the authors).
American Journal of Sociology
1794
TABLE 6
Goodness of Fit of Models for the Origin # Destination # Period Table
for Men and Women
No. Model df
Men
(n p 33,281)
Women
(n p 29,999)
Deviance P BIC Deviance P BIC
1 ... OD 125 180.3 .001 1,121 184.6 .001 1,104
2 ... ODbP
120 167.5 .003 1,082 167.4 .003 1,070
3 ... ODbP
l diag 114 133.3 .105 1,054 160.0 .003 1,015
4 ... ODlbP
diag 118 142.8 .060 1,086 167.8 .002 1,049
5 ... ODlbP
124 174.4 .002 1,117 175.1 .002 1,103
Note.--All models include the terms OP DP: O p origin; D p destination; P p period; diag p 6
parameters fitted to cells on main diagonal of the O­D table; .l p 1, . . . , 6
again leads to a large improvement in fit for men (compare models 2 and
3). On the other hand, models 3 and 4 again fit the data and are a
significant improvement over both models 1 and 5.27
Comparing models
3 and 4, the latter is more parsimonious. Among women, although none
of the models fits the data according to the deviance, model 5 would be
preferred according to this criterion, once again indicating the lesser importance
of inheritance. If, however, we take model 4 as the preferred
model for both sexes, we find that the slope of is .060 for men andb
.050 for women.28
Period and Cohort Change
Having established a trend toward increasing social fluidity over both
cohorts and periods, we turn, in table 7, to models that allow for both
types of change. All the models reported in table 7 fit the OPC and DPC
margins: thus we allow the origin distribution and the destination distribution
to vary over both cohorts and periods. Our interest is in the OD
association, and we test whether, given change over periods, cohort change
persists, and vice versa. We take as the point of departure two models of
change in social fluidity incorporating both period and cohort change, as
before using the unidiff model in its unconstrained (models 2­4) and linear
(models 5­7) forms to model change (including time constant parameters
27
There is no evidence that the diagonal parameter values vary over periods.
28
If period fluidity were simply a weighted sum of the fluidity in each cohort represented
in that period's table, and if all cohorts were the same size, then the period
and cohort slopes in our data would be identical (within sampling error). This is
because, given a cohort slope b, the association in an entering cohort in any period
is equal to ( b # 10 # the association in the exiting cohort), and as there are 10
cohorts in every period survey, the difference between periods is then b.
Educational Equalization
1795
TABLE 7
Goodness of Fit of Models for the Origin # Destination # Cohort # Period
Table for Men and Women
No. Model df
Men
(n p 33,281)
Women
(n p 29,999)
Deviance BIC Deviance BIC
1 ... OD 1,475 1,404.28 13,955 1,493.23 13,712
2 ... ODbP
l diag 1,464 1,361.65 13,883 1,472.42 13,620
3 ... ODbC
k diag 1,455 1,317.37 13,883 1,420.52 13,579
4 ... ODbP
lbC
k diag 1,450 1,311.85 13,787 1,412.11 13,536
5 ... ODlbP
diag 1,469 1,370.84 13,925 1,481.12 13,663
6 ... ODkbC
diag 1,469 1,332.54 13,964 1,428.55 13,715
7 ... ODkbC
lbP
diag 1,468 1,332.18 13,954 1,427.77 13,706
Note.--All models include the terms OPC DPC. P-values are not reported: they are all greater than
0.4. O p origin; D p destination; P p period; C p cohort; diag p 6 parameters fitted to cells on main
diagonal of the O­D table; k p 1, . . . , 15; l p 1, . . . , 6.
for the main diagonal). These models allow a common pattern of fluidity
to vary over either or both the C and P margins simultaneously: in other
words, the pattern of social fluidity is multiplied by a cohort-specific and
a period-specific b parameter. Ignoring, for ease of notation, the effects
of the diagonal parameters, in the case of model 4 we have
C P
lnv p b b lnv , (5)ijFkl k l ij
where k indexes cohorts and l periods, while for model 7 we have
C P
lnv p kb lb lnv ,ijFkl ij
for k p 1, . . . , 15, and l p 1, . . . , 6. (6)
The results show that, given a model which includes cohort change,
the addition of change over periods (model 3 compared with model 4,
and model 6 compared with model 7) does not significantly improve the
fit of the model, but the reverse is not the case (model 2 compared with
4 and model 5 compared with 7). That is to say, when we control for
cohort effects, the period effects vanish. Reflecting this, in model 7, which
is the direct counterpart to model 4 in the earlier cohort and period analyses,
the partial period slope is now estimated as being not significantly
different from zero ( 0.008 for men and 0.015 for women), whereas the
partial cohort slope is almost unchanged from its unconditional value
( .034 for men, compared with .036 and .045 for women, compared
with .043). These results, together with the finding, reported in table 5,
that fluidity does not vary within a given cohort, and the results of our
analysis of the panel element in our data, lead us to conclude that period
American Journal of Sociology
1796
TABLE 8
Goodness of Fit of Models for the Origin # Education # Cohort Table for
Men and Women
No. Model df
Men
(n p 33,281)
Women
(n p 29,999)
Deviance P BIC Deviance P BIC
1 ... OE 350 442.66 .001 3,203 406.32 .020 3,202
2 ... OEbC
k 336 386.19 .031 3,112 365.61 .128 3,098
3 ... OEkbC
349 405.01 .021 3,229 385.43 .087 3,212
Note.--All models include the terms OC EC. O p origin; D p destination; C p cohort; E p education,
.k p 1, . . . , 15
change in fluidity in the last quarter of the 20th century was the consequence
of the replacement of older, less fluid cohorts, with younger, more
fluid ones.
Understanding Cohort Change
How are we now to understand the change in social fluidity across cohorts?
Perhaps the most plausible mechanism to explain cohort change
concerns the transmission of assets during childhood, pointing to the important
role of education for social mobility. Increasing social fluidity may
then come about in two different ways, which we earlier labeled equalization
and compositional effects. We will look at each of these in turn.
A common explanation of increasing social fluidity is that it is driven
by a weakening association between social origin and educational qualifications.
Such a weakening did occur in Sweden among cohorts born
approximately between the early 1920s and the 1950s (Jonsson and Erikson
2000), and this weakening is likely to have brought about changes
in the gross association between origins and destinations of the kind that
we have shown above. In our data we find the same result. Table 8 reports
three models fitted to the origin # education # cohort table, the first of
which is constant association between origins and education (OE), the
second of which allows for uniform change in the OE association over
cohorts, and the third of which constrains this change to be linear. As
with the OD association, the model of linear change provides a good
account of the data. In figure 4, the coefficients from the second of theseb
models show the decline in the association flattening out after the 1950s:
this is very similar to the trend shown in figure 3, which provides some
prima facie evidence that the change in social fluidity was driven, in some
part, by changes in educational inequality. Our estimates, which we presented
below, suggest that around half of the total association between
Educational Equalization
1797
Fig. 4.-- coefficients for uniform change in the OE association over cohortsb
origins and destinations is mediated via education, showing that a weakening
of this indirect path through a decline in the origin-education association
can indeed lead to increases in social fluidity.
Table 9 uses the origin # destination # education # cohort table to
address the compositional question. Here there are two sets of models:
2­4 fit completely unrestricted log-linear models, and 5­7 fit unrestricted
log-multiplicative uniform difference models. In models 5 and 7 we denote
the log-multiplicative evolution of the OD association over educational
levels as , where m indexes educational levels from 1 to 6. In bothE
ODbm
sets of models we reach the same conclusion: once we allow for variation
in the OD association between levels of education, there is no significant
change in fluidity over cohorts. So, model 2 is not a poorer fit to the data
than model 4, but model 3 is, indicating that the omission of the ODC
term is not statistically significant once the ODE term is included in the
model, whereas the ODE term cannot be omitted even when ODC is
included. Likewise, model 5 is not a poorer fit than model 7, but model
6 is, indicating that the term is required but that the termE C
ODb ODbm k
is not.
This result does not mean that it is the compositional effect that is
wholly responsible for partialling out cohort change, because the model
of necessity includes the OEC and EDC terms: thus the trend in fluidity
over cohorts may depend on both the equalization effect--which is already
included in the models reported in table 9--and on the compositional
effect. The basis of this compositional effect is shown in figure 5, which
reports the coefficients for each educational level, taken from model 5b
American Journal of Sociology
1798
TABLE 9
Goodness of Fit of Models for the Origin # Destination # Education #
Cohort Table for Men and Women
No. Model df
Men (n p 33,281) Women (n p 29,999)
Deviance BIC Deviance BIC
1 ... OD 2,225 1,928.98 21,239 1,918.85 21,018
2 ... ODE 2,100 1,677.76 20,189 1,689.20 19,960
3 ... ODC 1,875 1,571.07 17,953 1,545.22 17,784
4 ... ODE ODC 1,750 1,335.32 16,887 1,326.36 16,714
5 ... ODbE
m diag 2,214 1,841.16 21,213 1,875.45 20,948
6 ... ODbC
k diag 2,205 1,877.20 21,083 1,874.66 20,857
7 ... ODbE
mbC
k diag 2,200 1,834.34 21,074 1,859.45 20,820
Model Comparisons Term Tested df Deviance P Deviance P
2 versus 4 .......... ODC 350 342.44 .603 362.84 .307
3 versus 4 .......... ODE 125 235.75 .001 218.86 .001
5 versus 7 .......... ODbC
k 14 6.82 .941 16.00 .313
6 versus 7 .......... ODbE
m 5 42.86 .001 15.21 .010
Note.--All models include the terms OCE DCE. O p origin; D p destination; P p period; C p
cohort; E p education; diag p 6 parameters fitted to cells on main diagonal of the O-D table; k p 1,
. . ., 15; m p 1, . . . , 6.
of table 9. This shows a clear pattern, though not a linear one, of weaker
association between origins and destinations at higher educational levels.
Thus, as successive cohorts have come to have higher levels of education
(see table 2) so the gross association between origins and destinations has
weakened.
This effect is illustrated in figure 6, which shows the weighted sum of
the estimated b values for each educational level in every cohort (where
the weights are the relative sizes of the educational categories). In other
words, this would be the social fluidity in each cohort if that were simply
the weighted sum of the fluidities in each educational category. The picture
shown in figure 6 is similar to that in figures 3 and 4: a decline, parallel
for both sexes, until the cohorts born in the 1950s, then stability. The
temporal coincidence of the equalization and compositional effects derives
from the fact that the equalization that affected cohorts born in the first
half of the century had the consequence of expanding the middle and
upper levels of the educational system; this expansion translated into a
compositional effect because labor markets were more meritocratic the
higher the level of educational qualifications.
Although the weighted averages of the educational level bs in each
cohort point to similar trends to those shown by the cohort bs themselves,
the relationship between the unconditional cohort trends and the corresponding
trends across educational levels is more complicated than this.
Educational Equalization
1799
Fig. 5.-- coefficients for each educational levelb
Drawing on results by Goodman (1972, pp. 1070­75), we know that we
can find an unconditional cohort effect, (that is, ODC) even when the
partial ODC terms are zero, provided that the partial ODE and EC associations
are nonzero. From this it follows that the three-way ODC term
when ODE is not in the model depends not only on the educational effect
but also on the distribution of educational categories across periods. It is
therefore not a simple matter to infer what pattern of unconditional cohort
effects is implied by a given set of education effects, and figure 6 should
be taken only as illustrative.
Education and Social Fluidity
In the final part of our analysis we try to make a more formal assessment
of the effects of educational equalization and compositional change on
the trend over cohorts in social fluidity. Essentially this involves determining
how much of the association between origins and destinations is
mediated via educational attainment and, following from this, how much
of the change in fluidity over cohorts comes about through, on the one
hand, changes in the effects of origins on educational attainment and of
educational attainment on class destinations, and, on the other, the shift
of the population into those educational categories in which origins have
a weaker effect. Given continuous measures of social position we could
do this using path analysis, but with categorical variables this is not
possible. We therefore use an approximation, following Breen and Luijkx
(2004a).
American Journal of Sociology
1800
Fig. 6.--Weighted sum of estimated b values for each educational level, every cohort
We must begin with a measure of the evolution of the gross OD association
over cohorts, and so we could fit the log-multiplicative ODbC
model to the three-way origin by destination by cohort table (as reported
in the models of table 4). The obvious next step might then be to use the
four-way table of origins by destinations by cohorts by education to fit a
model which includes the partial effects of education on destination controlling
for origins, and the partial effects of origin on destination controlling
for education. This would, in fact, be one of the models reported
in table 9 and, if the partial OD association were fitted using a logmultiplicative
specification, it might seem that we could compare the bs
from this model (the partial bs) and compare them with the bs from one
of the models of table 4 (the gross bs). But such a comparison would be
invalid because the pattern of local OD association will be different in
the two cases: that is, the pattern of OD association that evolves log
multiplicatively over cohorts depends on whether we control for the effect
of education on destinations or not.29
We attempt to overcome this difficulty
by constraining the pattern (though not the strength) of the OD
association in the partial model to be the same as the estimated gross OD
association. This allows us to make a comparison of the b parameters
from the two models and so measure the relative strength of the association
with and without controlling for the effects of education.
29
This is because the local OD association and the b parameters are estimated together
in the log-multiplicative model. In the unconditional model they depend on the gross
association between origins and destinations, whereas in the conditional model they
depend on the association between origins and destinations holding education constant.
Educational Equalization
1801
But things are not so simple because we can reasonably suppose that
the pattern (and not just the strength) of the OD association will differ
significantly depending on whether education is in the model or not. Educational
attainment is a more important asset for mobility to some class
destinations rather than others: in particular, entry into self-employment
or farming among children born into these classes is a question of inheritance,
rather than of educational attainment (Ishida, Mu¨ller, and Ridge
1995). With this in mind we use, as our baseline model, model 4 of table
4, which includes the set of parameters applied to the main diagonal of
the table (but whose effects are held constant over cohorts and which we
henceforth referred to as "diag").
Our model for the unconditional OD association can be written OC DC
ODkbC
diag: this allows the OD association to evolve linearly over
cohorts. The first of our models for the conditional or partial OD association
is OEC EDC XOD
kbC
diag. XOD
here represents the OD association
which is fixed to be equal to that estimated from the gross model,
and this is, once again, constrained to evolve linearly over cohorts. We
allow the diagonal parameters to differ between the partial and gross
models and we fit the OEC and EDC margins exactly in order to focus
on the difference between the bs from the gross and partial models. Comparing
the bC
estimates tells us the extent to which the OD association,
and its trend over cohorts, weakens once we take into account the association
between origins and education and that between education and
destination.
However, we can go further than this and fit a second partial model
in which we allow the association to vary over both cohorts and educational
levels as follows: . Now the bC
pa-OD E C
OEC EDC X b kb diagm
rameter tells us the slope of the OD association over cohorts when we
also allow that association to vary over educational levels. Here the constrained
OD association, XOD
, evolves linearly over cohorts, as before, but
varies freely over educational levels.
Table 10 contains the results of these analyses. Model 1 repeats model
4 of table 4. Model 2 is a poorer fit to the data than the counterpart
model, which preserves the linear trend but estimates the OD association
freely (the difference in deviance is 60.4 for men and 88.3 for women;
25df), but nevertheless still provides an adequate fit to the data,30
as does
model 3. Once we control for education (model 2), the strength of the OD
association--as measured by the value of the log-odds ratios in model 2
30
The large increase in the deviance for model 2 compared with model 6 of table 8
suggests that the pattern of the OD association in the off-diagonal cells also changes
when education is taken into account--but this is something which our model cannot
capture.
TABLE10
GoodnessofFitofModelstoDecomposetheODAssociation
ModelsDescriptionDeviancedf
RelativeLog-Odds
Ratio*
Proportional
Slope
Absolute
Slope
Men:
1.......OCDCODkbC
diag331.123431.036.036
2.......OECEDCXOD
kbC
diag1,956.592,230.53.057.030
3.......OECEDCXOD
bm
E
kbC
diag1,892.292,225.37.01.034.013.017
.43.52.018.015
.51.68.000.013
Women:
1.......OCDCODkbC
diag368.793431.044.044
2.......OECEDCXOD
kbC
diag1,993.712,230.58.066.038
3.......OECEDCXOD
bm
E
kbC
diag1,947.422,225.10.20.049.033.034
.51.43.021.025
.70.68.010.005
Note.--Formen,;forwomen,.np33,281np29,999
*Settolformodel1.
Educational Equalization
1803
Fig. 7.--Diagonal parameter estimates from models 1­3: men
compared with model 1--is reduced by just less than half. That is, once
we control for the path that links origins to destination via education, the
direct effect of origins on destination is about halved. The slope of the
log-odds ratios over cohorts, expressed as a proportionate decline in the
association for that model, strengthens in model 2 compared with model
1; however, if we express the slope as the change in the absolute value
of the log-odds ratios then it is only slightly weaker in model 2 than in
model 1.31
So, although controlling for class inequality in educational
attainment accounts for a good deal of the association between origins
and destinations, it does not explain much of the trend of change in this
association over cohorts. In model 3 we also allow the OD association to
vary over educational levels, and we report the strength of the origindestination
association at each educational level and also the absolute
slope within each educational level. In both cases the figures run from
the lowest to the highest educational level, and they should be read by
row. Not only does the association vary in strength quite noticeably over
educational levels (as we already saw in fig. 5), but, among men, the
absolute slopes are quite close to zero. Among women they are close to
zero at the higher educational levels but somewhat further from zero at
the lower levels. These results suggest that the declining trend in the
association between origins and destinations is mainly due to the com-
31
The proportional slope reports the proportional decline per birth cohort: so, in model
2, for men this is just less than 6%. But this is relative to the initial association, which,
in model 2, is weaker than that in model 1. If we then ask, What would this decline
be as a share of the original association? (i.e., that in model 1), we find that this 6%
proportional reduction equates to a 3% absolute decline.
American Journal of Sociology
1804
Fig. 8.--Diagonal parameter estimates from models 1­3: women
positional effect of educational expansion rather than to the process of
educational equalization. Finally, the diagonal parameter estimates from
the three models reported in table 10 are shown in figures 7 and 8. The
first bar of the histogram for each class repeats that shown in figure 2.
Overall the parameters show remarkably little change, indicating that the
tendency for classes to be self-recruiting (where this tendency exists) operates
largely independently of the educational mechanism.
CONCLUSION
Previous research on social fluidity has not been successful in accounting
for temporal change; nor, indeed, have researchers agreed on the extent
to which modern societies are characterized by change at all. Our analysis
documents trends toward greater openness in Sweden, and we extend
previous research by addressing the question of how this change came
about. On the basis of a theoretical model of the intergenerational transmission
of class position we argue that change in social fluidity will, under
normal circumstances, be driven by cohort replacement rather than by
period effects. Following from this, we pay special attention to the way
in which changes in children's educational attainment can account for
changing fluidity.
The Swedish data are particularly suitable for addressing the question
of temporal change: we have access to 24 annual surveys covering the
period 1976­99, with cohorts born between 1912 and 1974, and which
have comparable classifications of social class origins, educational qual-
Educational Equalization
1805
ifications, and class destinations (derived from current job). Our first conclusion
is that social fluidity increased in Sweden, particularly among
those cohorts of men and women born during the first half of the 20th
century. We also observed a trend toward greater fluidity in the workingage
population during the last quarter of the 20th century, but this period
change disappears when we control for differences between birth cohorts.
Moreover, we show that social fluidity does not vary across periods within
a cohort, and, using the panel element in the data, we find that fluidity
does not change across the life course. These results led us to conclude
that period change is, in fact, driven by a process of cohort replacement.
To the extent that change in social fluidity generally comes about through
cohort replacement, this may explain why many sociologists have failed
to discern any period change when analyzing two or more surveys that
are not so many years apart and that therefore mainly comprise samples
from the same cohorts. It may well be the case that true period changes
are likely to occur only in specific, and perhaps rather dramatic, circumstances,
but historical changes that equalize the opportunities for successive
birth cohorts may have a substantial impact that is only visible in
a longer time perspective.
Our second conclusion is that the evolution of fluidity over cohorts in
Sweden has been driven by educational equalization and by a compositional
effect based in the changing educational distributions of successive
birth cohorts. Hout and Dohan (1996) have portrayed the U.S. and Swedish
"strategy of educational equality" as very different: the first seeks to
expand educational opportunities while the second focuses on equality of
condition as a means of improving the relative chances of children from
disadvantaged origins. While there is little evidence that educational expansion
in the 20th century led to a decreasing association between class
origins and educational attainment in Sweden (Jonsson and Erikson, in
press), our results show that expanding the educational system nevertheless
helped to reduce the association between class origins and class destinations.
It allowed more children to reach educational levels that led
them to labor market segments where meritocratic selection was more
prevalent and origin characteristics counted for less. The equalizing of
educational chances in Sweden also led to increasing social fluidity. Moreover,
although one can have compositional effects without equalization,
equalization almost certainly implies expansion of the middle and/or
higher levels of education, given that equalization is unlikely to occur
through a reduction in educational participation by the middle classes.
This means that promoting such equalization is likely to be a very effective
strategy because it also advances compositional change. Our results provide
perhaps the clearest example, within the social mobility literature,
of how educational equalization can drive social fluidity: equalization
American Journal of Sociology
1806
implies expansion of the middle and/or higher levels of the educational
system, and, if social origin has less impact in the labor market for those
with such qualifications, compositional change accelerates the trend toward
increasing social fluidity. It is of course possible that expansion also
might lead to countervailing tendencies, with social origin reasserting itself
in the labor market for graduates as their number increases (something
that Vallet [2004, p. 142] reports for France), though we did not find any
such development in our data.
Our results indicate that the trend toward educational equalization in
Sweden--and thus also the trends in the compositional effect and in
overall social fluidity--ground to a halt in those cohorts born around
midcentury. If the situation had persisted among cohorts born after 1974
(and assuming the continued absence of any independent period influences
on fluidity) the trend of equalization in period fluidity in Sweden would
have been expected to come to a halt by 2020, as the Swedish workforce
came to consist only of cohorts born after the middle of the 20th century.
However, recent studies show that the impact of social origin on the
transition to upper secondary education diminished further during the
1990s (Gustafsson, Andersson, and Hansen 2001),32
which may, along with
the rapid expansion in the provision of higher education in the same
decade, allow the trend toward equalization in Sweden to continue.
Finally, we turn to the relevance of our results for other countries and
for questions of policy. Sweden is well known in the mobility literature
for its high level of social fluidity (see Breen and Luijkx 2004b, pp. 59,
72) and for its long period of gradual equalization. Nevertheless, the
educational reforms introduced in Sweden during the 20th century, to
which we referred earlier, are rather typical of many developed countries,
especially in Europe. This raises two questions: Has the same trend of
increasing social fluidity occurred in other countries? and Why is fluidity
higher in Sweden than elsewhere? One difficulty in answering the first
question is the dearth of analyses that adopt a cohort, rather than a period,
perspective: even so, there is now a large body of evidence from periodbased
studies to suggest a widespread trend toward increasing social fluidity
(Breen and Luijkx 2004a). Whether this can be attributed to the
same causes as the Swedish case is not known, but recent research has
demonstrated an equally widespread trend toward a weaker association
between class origins and educational attainment (for a comparative analysis
see Breen et al. [2005]; various single country studies are referred to
in Breen and Jonsson [2005, p. 226]).
Sweden's high level of fluidity might be attributed to two main factors.
32
We cannot observe this in our data because the individuals affected by it are too
young to have been included in our samples.
Educational Equalization
1807
On the one hand, equality of condition (especially with respect to income
and the risk of prolonged unemployment) has been attained to a much
greater level there than elsewhere, and this has played an important role
in weakening the transmissibility of mobility assets between generations.
On the other hand, Swedish employers appear to consider formal merits
rather than characteristics related to the family of origin when employing
those with higher education, particularly graduates (see fig. 5). This, in
turn, may be because the tertiary educational system is relatively homogeneous,
being free of fees and not displaying any marked differences
in prestige between institutions (at least not for the cohorts we analyze).
One implication, noted by Breen and Luijkx (2004a, p. 400), is that much
more of the link between origins and destinations is mediated via education
in Sweden than in other countries, and so reforms of the educational
system, as well as the mere expansion of it, are likely to have a greater
impact on social fluidity there than elsewhere. Thus, in Sweden, the organization
of the educational system has been allied with efforts toward
greater equality of condition. We might expect each of these to contribute
to increasing openness, but, if the Swedish case is any guide, their combination
seems to be particularly effective.
REFERENCES
Andersson, Lars-Gunnar, Robert Erikson, and Bo Wa¨rneryd. 1981. "Att beskriva den
sociala strukturen: Utva¨rdering av 1974 a°rs fo¨rslag till socio-ekonomisk indelning."
Statistisk Tidskrift 2:113­36.
Bowles, Samuel, and Herbert Gintis. 2002. "The Inheritance of Inequality." Journal
of Economic Perspectives 16:3­30.
Breen, Richard 1994. "Individual Level Models for Mobility Tables and Other CrossClassifications."
Sociological Methods and Research 23:147­73.
------. 1997. "Inequality, Economic Growth and Social Mobility." British Journal of
Sociology 48:429­49.
------, ed. 2004. Social Mobility in Europe. Oxford: Oxford University Press.
Breen, Richard, and Jan O. Jonsson. 2001. "Period and Cohort Change in Social
Mobility in Sweden." Paper presented at the meeting of the ISA Research Committee
on Social Stratification and Mobility (RC28). Mannheim, Germany. April.
------. 2005. "Inequality of Opportunity in Comparative Perspective: Recent Research
on Educational Attainment and Social Mobility." Annual Review of Sociology 31:
223­44.
Breen, Richard, and Ruud Luijkx. 2004a. "Conclusion." Pp. 383­410 in Social Mobility
in Europe, edited by Richard Breen. Oxford: Oxford University Press.
------. 2004b. "Social Mobility in Europe between 1970 and 2000." Pp. 37­75 in
Social Mobility in Europe, edited by Richard Breen. Oxford: Oxford University
Press.
Breen, Richard, Ruud Luijkx, Walter Mu¨ller, and Reinhard Pollak. 2005. "Nonpersistent
Inequality in Educational Attainment: Evidence for Eight European
Countries." Paper presented at the meeting of the ISA Research Committee on Social
Stratification and Mobility (RC28). Los Angeles. August.
Carlsson, Go¨sta. 1958. Social Mobility and Class Structure. Lund: CWK Gleerup.
American Journal of Sociology
1808
DiPrete, Tom A., and David B. Grusky. 1990. "Structure and Trend in the Process of
Stratification for American Men and Women." American Journal of Sociology 96:
107­43.
Erikson, Robert. 1983. "Changes in Social Mobility in Industrial Nations: The Case
of Sweden." Research in Social Stratification and Mobility 2:165­95.
------. 1987. "The Class Structure and Its Trends." Pp. 19­42 in Welfare in Transition:
A Survey of Living Conditions in Sweden, 1968­1981, edited by Robert Erikson
and Rune A° berg. Oxford: Clarendon.
------. 1996. "Explaining Change in Educational Inequality: Economic Security and
School Reforms." Pp. 95­112 in Can Education Be Equalized? edited by Robert
Erikson and Jan O. Jonsson. Boulder, Colo.: Westview.
Erikson, Robert, and John H. Goldthorpe. 1992. The Constant Flux: A Study of Class
Mobility in Industrial Societies. Oxford: Clarendon.
Erikson, Robert, John H. Goldthorpe, and Lucienne Portocarero. 1979.
"Intergenerational Class Mobility in Three Western European Societies: England,
France, and Sweden." British Journal of Sociology 30:415­41.
Erikson, Robert, and Jan O. Jonsson. 1996. "The Swedish Context: Educational Reform
and Long-term Change in Educational Inequality." Pp. 65­93 in Can Education Be
Equalized? edited by Robert Erikson and Jan O. Jonsson. Boulder, Colo.: Westview.
------. 1998. "Qualifications and the Allocation Process of Young Men and Women
in the Swedish Labour Market." Pp. 369­406 in From School to Work: A Comparative
Study of Educational Qualifications and Occupational Destinations, edited by Y.
Shavit and W. Mu¨ller. Oxford: Oxford University Press.
Featherman, David L., Frank L. Jones, and Robert M. Hauser. 1975. "Assumptions
of Social Mobility Research in the U.S.: The Case of Occupational Status." Social
Science Research 4:329­60.
Ganzeboom, Harry B. G., Ruud Luijkx, and Donald J. Treiman. 1989.
"Intergenerational Class Mobility in Comparative Perspective." Research in Social
Stratification and Mobility 8:3­84.
Gerber, Theodore P., and Michael Hout. 2004. "Tightening Up: Declining Class
Mobility during Russia's Market Transition." American Sociological Review 69:
677­703.
Goldthorpe, John H. 1980. Social Mobility and Class Structure in Modern Britain.
Oxford: Clarendon.
------. 2000. On Sociology. Oxford: Oxford University Press.
Goodman, Leo A. 1972. "A General Model for the Analysis of Surveys." American
Journal of Sociology 77:1035­86.
Grusky, David B., and Robert M. Hauser. 1984. "Comparative Social Mobility
Revisited: Models of Convergence and Divergence in 16 Countries." American
Sociological Review 49:19­38.
Gustafsson, Jan-Erik, Anette Andersson, and Michael Hansen. 2001. "Prestationer och
prestationsskillnader i 1990-talets skola." Pp. 135­211 in Va¨lfa¨rd och skola. SOU
(Statens Offentliga Utredning) 2000:39. Stockholm: Fritzes.
Hauser, Robert M. 1995. "Better Rules for Better Decisions." Sociological Methodology
25:175­83.
Hauser, Robert M., and Min-Hsiung Huang. 1997. "Verbal Ability and Socioeconomic
Success: A Trend Analysis." Social Science Research 26:331­76.
Hout, Michael. 1988. "More Universalism, Less Structural Mobility: The American
Occupational Structure in the 1980s." American Journal of Sociology 93:1358­1400.
Hout, Michael, and Daniel P. Dohan. 1996. "Two Paths to Educational Opportunity:
Class and Educational Selection in Sweden and the United States." Pp. 207­32 in
Can Education Be Equalized? edited by Robert Erikson and Jan O. Jonsson. Boulder,
Colo.: Westview.
Ishida, Hiroshi, Walter Mu¨ller, and John M. Ridge. 1995. "Class Origin, Class
Educational Equalization
1809
Destination, and Education: A Cross-national Study of Industrial Nations."
American Journal of Sociology 101:145­93.
Jonsson, Jan O. 2001. "Towards a Post-Fordist Life-Course Regime? Generational
Changes in Transitions and Volatility." Pp. 1­28 in Cradle to Grave: Life-Course
Changes in Modern Sweden, edited by Jan O. Jonsson and Colin Mills. Durham:
Sociologypress.
Jonsson, Jan O., and Robert Erikson. 1997. "Klasstruktur och social ro¨rlighet under
1900-talet." Pp. 491­511 in Va¨lfa¨rd och oja¨mlikhet i 20-a°rsperspektiv 1975­1995,
edited by J. Vogel and L. Ha¨ll. Stockholm: Statistics Sweden.
------. 2000. "Understanding Educational Inequality: The Swedish Experience."
L'Anne´e sociologique 50:345­82.
------. In press. "Why Educational Expansion Is Not Such a Great Strategy for
Equality: Theory and Evidence for Sweden." In Expansion, Differentiation and
Inequality of Access in Higher Education: A Comparative Study, edited by Yossi
Shavit, Richard Arum, and Adam Gamoran. Palo Alto, Calif.: Stanford University
Press.
Jonsson, Jan O., and Colin Mills. 1993. "Social Mobility in the 1970s and 1980s: A
Study of Men and Women in England and Sweden." European Sociological Review
9:229­48.
Korpi, Walter, and Joakim Palme. 2004. "Robin Hood, St. Matthew, or Simple
Egalitarianism? Strategies of Equality in Welfare States." Pp. 153­79 in A Handbook
of Comparative Social Policy, edited by Patricia Kennet. Northhampton, Mass.:
Edward Elgar.
Luijkx, Ruud, and Harry B. G. Ganzeboom. 1989. "Intergenerational Class Mobility
in the Netherlands between 1970 and 1986." Pp. 5­30 in Similar or Different? edited
by Wim Jansen, Jaap Dronkers, and Kitty Verrips. Amsterdam: SISWO.
Mannheim, Karl. 1952. "The Problem of Generations." In Essays on the Sociology of
Knowledge. London: Routledge & Kegan Paul.
Mu¨ller, Walter, and Yossi Shavit. 1998. "The Institutional Embeddedness of the
Stratification Process." Pp. 1­48 in From School to Work: A Comparative Study of
Educational Qualifications and Occupational Destinations, edited by Yossi Shavit
and Walter Mu¨ller. Oxford: Clarendon.
Raftery, Adrian E. 1986. "Choosing Models for Cross-classifications." American
Sociological Review 51:145­46.
Ro´bert, Pe´ter, and Erzse´bet Bukodi. 2004. "Changes in Intergenerational Class Mobility
in Hungary, 1973­2000." Pp. 287­314 in Social Mobility in Europe, edited by
Richard Breen. Oxford: Oxford University Press.
Ryder, Norman B. 1965. "The Cohort as a Concept in the Study of Social Change."
American Sociological Review 30:843­61.
Vallet, Louis-Andre´. 1999. "Quarante anne´es de mobilite´ sociale en France: L'e´volution
de la fluidite´ sociale a` la lumie`re de mode`les re´cents." Revue franc¸aise de sociologie
40:5­64
------. 2004. "Change in Intergenerational Class Mobility in France from the 1970s
to the 1990s and Its Explanation: An Analysis Following the CASMIN Approach."
Pp. 115­47 in Social Mobility in Europe, edited by Richard Breen. Oxford: Oxford
University Press.
Vogel, Joachim, Lars-Gunnar Andersson, Uno Davidsson, and Lars Ha¨ll. 1988.
Inequality in Sweden: Trends and Current Situation. Report no. 58. Living
Conditions Series. Stockholm: Statistics Sweden.
Weakliem, David L. 1999. "A Critique of the Bayesian Information Criterion for Model
Selection." Sociological Methods and Research 27:359­97.
American Journal of Sociology
1810
Wong, Raymond Sin-Kwok. 1994. "Postwar Mobility Trends in Advanced Industrial
Societies." Research in Social Stratification and Mobility 13:121­44.
Xie, Yu. 1992. "The Log-Multiplicative Layer Effect Model for Comparing Mobility
Tables." American Sociological Review 57:380­95.