Sunteți pe pagina 1din 14

Capitol publicat in David B. Grusky (ed.). 2001.

Social Stratification: Class, Race, and


Gender in Sociological Perspective. Second Edition. Boulder: Westview Press.

\ I

li

li I
PETER M. BLAU AND OTIS DUDLEY DUNCAN,
WITH THE COLLABOR ATION OF ANDREA TYREE -i
1i
1l
The Process of Stratification I
1l
1'
.l

Stratification systems may be characterized in


I
various ways. Surely one of the most impor-
l
I

tant has to do with of


e ine tha of 1
I

ome ascrlptrve leatures


one extreme we can rmag- system may be regarded as vestiges of an ear-
ine- that thI circumstances of a person's lier epoch, to be extirpated as rapidly as pos-
birth-including the person's sex and the per- sible. Public policy may emphasize measures
1!
fectly predictable sequence of age levels designed to enhance or to equalize opportu- 1l
through which he is destined to pass-suffice nity-hopefully, to overcome ascriptive obsta- 1l
to assign him unequivocally to a ranked sta- cles to the full exercise of the achievement nl
a\
tus in a hierarchical system. At the opposite principle. .J
extreme his prospective adult status would be The question of how far a society may real- at
wholly problematic and contingent at the istically aspire to go in this direction is hotly tl
.l
time of birth. Such status would become en- debated, not only in the ideological arena but tlr
tirely determinate only as adulthood was in the academic forum as well. Our contribu-
t
reached, and solely as a consequence of his tion, if any, to the debate will consist largely 1r
own actions taken freely-that is, in the ab- rn suDmtttrng measurements and estl 1l
sence of any constraint deriving from the cir-
t,
{l
cumstances of his birth or rearing. Such a pure .l
achievement system is, of course, hypothetical, aJ
in much the same way that motion without a{
friction is a purely hypothetical possibility in tem is ultimately a quantitative one. We have
pushed our ingenuity to its limit in seeking to
t1
1i
contrive relevant quantifications. 1J
The governing conceptual scheme in the -{
analysis is quite a commonplace one. ub
se- !i
1*
Originally published in 1967. Please see complete ly by a set of classificatory
source information beginning on page 891. or quantitative measurements taken at succes- {a
890
{
t
dt
{fl
The Process of Stratification 391

sive stages. Ideally we should like to have un- describe the maior outlines of status changes
der observation a cohort of births, following in the life cycle of a cohort. Thus a study of
the individuals who make up the cohort as the relationships among these variables leads
they pass through life. As a practical matter to a formulation of a basic model of the pro-
we resorted to retrospective questions put to a cess of stratification.
representative sample of several adjacent co-
horts so as to ascertain those facts about their
life histories that we assumed were both rele- A Basic Model
vant to our problem and accessible by this
means of observation. To begin with, we examine only fivey:rialrles.
Given this scheme, the questions we are For expository convenience, when it is neces-
continually raising in one form or another sary to resort to symbols, we shall designate
are: how and to w clrcum- them by arbitrary ietters but try to remind the
rth condition su uent status? reader from time to time of what the letters
stand for. These variables are:

V: Father's educational attainment


X,@
fficratic ones. Current policy discussion and IJ:
,W:
action come to a focus in a vaguely explicated
notion of the "inheritance of poverty." Thus a X Status o f res pondenti-occupatlo.n-i..1_
spokesman for the Social Security Administra- 1962
tron wntes:
Each of the three occupational statuses ts
It would be one thing if poverty hit at random and scaled by the [socioeconomic] index described
no one group were singled out. It is another thing [elsewhere],3 ralglng from g to 96. The two
to realize that some seem destined to poverty al- education variabl6 are scored on the follow-
most from birth-by their color or by the eco- ing arbitrary scale of values ("rungs" on the
nomic status or occupation of their parents.l "educational ladder") corresponding to specl-
fied numbers of years of formal schoolins
Another officially sanctioned concept is that completed:
of the "dropout," the person who fails to
graduate from high school. Here the emphasis No school
is not so much on circumstances operative at Elementary, one to four years
birthbutont@ tl. Elementary, five to seven years
Elementary, eight years
achievem
High school, one to three years
time of uncertain employment,"2 probable as- High school, four years
signment to jobs of inferior status, reduced College, one to three years
earning power, and vulnerability to various College, four years
forms of social pathology. College, five years or more (i.e., one or
In this study we do not have measurements more years of postgraduate study)
on all the factors implicit in a full-blown con-
ception of the "cycle of poverty" nor all those Actually, this scoring system hardly differs
variables conceivably responding unfavorably from a simple linear transformation, or "cod-
to the achievement of "dropout" status. . . . ing," of the exact number of years of school
This limitation, however, is not merely an an- completed. In retrospect, for reasons given
alytical convenience. \We think of the selected [elsewhere],a we feel that the score implies too
quantitative variables-il being sufficient to great a distance between intervals at the lower
392 lU / Genenating Ineruality

end of the scalel but the resultant distortion is prepared for another study5 using the OCG
minor in view of the very small proportions data, however, suggest that approximately
scored 0 or 1. one-eighth of the respondents report a com-
A basic assumption in our interpretation of bination of age at first job and education that
regression statistics-though not in their cal- would be very improbable unless (a) they vio-
culation as such-has to do with the causal or Iated instructions by reporting a part-rime or
telrporabrdrurrsef these variabl.i I" t..r"s school-vacation job as the first job, or
of the father's career we should naturally as- (b) they did, in fact, interrupt their schooling
sume precedence of V (education) with re- to enter regular employment. (These "incon-
spect to X (occupation when his son was 16 sistent" responses include men giving 19 as
years old). \7e are not concerned with the fa- their age at first job and college graduation or
ther's career, however, but only with his sta- more as their education; 17 or 18 with some
tuses that comprised a configuration of back- college or more; 74, 15, or 16 with high-
ground circumstances or origin conditions for school graduation or more; and under 14
the cohorts of sons who were respondents in with some high school or more.) I7hen the
the Occupational Changes in a Generation two variables are studied in combination
(OCG) study. Hence we generally make no as- with occupation of first job, a vety clear ef-
sumption as to the priority of V with respect fect is evident. Men with a given amount of
to X,. in effect, we assume the measurements education beginning their first jobs early held
on these variables to be contemporaneous lower occupational statuses than those begin-
from the son's viewpoint. The respondent's ning at a normal or advanced age for the
education, U, is supposed to follow in time- specified amount of education.
and thus to be susceptible to causal influence Despite the strong probability that the U-W
from-the two measures of father's status. Be- sequence is reversedfor an appreciable minor-
cause we ascertained X as of respondent's age ity of respondents, we have hardly any alter-
16, it is true that some respondents may have native to the assumption made here. If the
completed school before the age to which X bulk of the men who interrupted schooling to
pertains. Such cases were doubtlessly a small take their first jobs were among those ulti-
minority and in only a minor proporrion of mately securing relatively advanced educa-
them could the father (or other family head) tion, then our variable W'is downwardly bi-
have changed status radically in the two or ased, no doubt, as a measure of their
three years before the respondent reached 16. occupational status immediately after they fi-
The next step in the sequence is more prob- nally left school for good. In this sense, the
) correlations between U and \X/ and between
'W
and Y are probably attenuated. Thus, if we
had really measured "job after completing ed-
which stipulated "the first full-time job yoo ucation" instead of "first job," the former
had after you left school." In the years since would in all likelihood have loomed some-
the OCG study was designed we have been what larger as a variable intervening between
made aware of a fact that should have been education and 1962 occupational status. 'We
considered more carefully in the design. Many do not wish to argue that our respondents
students leave school more or less definitively, erred in their reports on first job. We are in-
only to return, perhaps to a different school, clined to conclude that their reDorts were re-
some years later, whereupon they often finish alistic enough, and that ir *", ou, assumprion
a degree program.s The OCG questionnaire about the meaning of the responses thar
contained information relevant to this prob- proved to be fallible.
lem, namely the item on age at first job. The fundamental difficulty here is concep-
Through an oversight no tabulations of this tual. If we insist on ctny uniform sequence of
item were made for the present study. Tables the events involved in accomplishing the tran-
The Process of Stratificatrcn 393

sition to independent adult status, we do vio- In proposing this sequence we do not over-
lence to reality. Completion of schooling, de- look the possibility of what Carlsson calls
parture from the parental home, entry into "delayed effects,"s meaning that an early
the labor market, and contracting of a first variable may affect a later one not only via in-
marriage are crucial steps in this transition, tervening variables but also directly (or per-
which all normally occur within a few short haps through variables not measured in the
years. Yet they occur at no fixed ages nor in study).
any fixed order. As soon as we aggregate indi- In translating this conceptual framework
vidual data for analytical purposes we are into quantitative estimates the first task is to
forced into the use of simplifying assump-
tions. Our assumption here is, in effect, that
"first job" has a uniform significance for all
men in terms of its temporal relationship to 1 supplies the correlation matrix on which
educational preparation and subsequent work much of the subsequent analysis is based. In
experience. If this assumption is not strictly discussing causal interpretations of these cor-
correct, we doubt that it could be improved relations, we shall have to be clear about the
by substituting any other single measure of distinction between two points of view. On
initial occupational status. (In designing the the one hand, the simple correlation-given
OCG questionnaire, the alternative of "job at our assumption as to direction of causation-
the time of first marriage" was entertained measures the gross magnitude of the effect of
briefly but dropped for the reason, among the antecedent upon the consequent variable.
others, that unmarried men would be ex-
cluded thereby.)
Thus., if = ..541,, *,.
,ro*
say that
:"" il ig.+
One other problem with the U-W'transi-
tion should be mentioned. Among the
younger men in the study, 20 to 24 years old,
are many who have yet to finish their school- tur From another point of view we are more
ing or to take up their first jobs or both-not concerned with net effects. If both first iob
to mention the men in this age group missed
by the survey on account of their military ser-
vice.7 Unfortunately, an early decision on tab-
ulation plans resulted in the inclusion of the Y consists in a transmission of the prior influ-
20 to 24 group with the older men in aggre- ence of X. Or, thinking of X as the initial
gate tables for men 20 to 64 years old. We cause, we may focus on the extent to which
have ascertained that this results in only mi- its influence on Y is transmitted by way of its
nor distortions by comparing a variety of prior influence on Wi
'We
data for men 20 to 64 and for those 25 to 64 may then, devote a few remarks to the
years of age. Once over the U-Vl hurdle, we pattern of gross effects before presenting the
see no serious objection to our assumption apparatus that yields estimates of net direct
that both U and lV'precede I except in re-
gard to some fraction of the very young men
just mentioned. TABLE I
Simple Correlations for Five Status Variables
In summarS then, we take the somewhat
idealized assumption of temporal order to
Variable
represent an order of priority in a causal or
Yr 1962 occ, status 54r 596 405 322
processual sequence, which may be stated dia- W: First-job status 538 332
grammatically as follows: U: Education .438 453
X: Fatherrs occ. status Dlb

(v x1- (u)- (w)- (Y). vi Father's educBtion


lV / Genenating lnequality

and indirect effects. Since we do not require a FIGURE T

causal ordering of father's education with re- Path coefficients in basic model of the
process of stratihcation
spect to his occupation, we may be content
simply to note that r*u = .51'6 is somewhat Father's ^-^\\
.dJ9 Resp
lower than the corresponding correlation, ryu educatron ?ln \ edr
rr--tt
- .596, observed for the respondents them-
selves. The difference suggests a heightening \.zss
of the effect of education on occupational sta- :Y +\\3
us D etwee lTTllffatileHFarlurne 5 on s gen c ra - Occ. in
tions. Before stressing this interpretation, 1962

ftlvever, we must remember that the meas-


urements of V and X do not pertain to some
actual cohort of men, here designated "fa-
First
thers." Each "father" is represented in the job-
data in proportion to the number of his sons f.r1
who were 20 to 64 years old inMarch 1.962.
The first recorded status of the son himself
is education (U). \7e note that rru is just But if the diagram is logical for the respon-
slightly greater than rr,r. Apparently both dent's generation, we should have to assume
measures on the father represent factors that that for the fathers, likewise, education and
may influence the son's education. occupation are correlated not only because
In terms of gross effects there is a clear or- one affects the other but also because com-
dering of influences on first job. Thus r*u > mon causes lie behind both, which we have
twx > ruy. Eclucation is most st not measured. The bidirectional arrow merely
lated with serves to sum up all sources of correlation be-
iiil. and then bv fa tween V and X and to indicate that the expla-
1 status in 1.962 (Y) a nation thereof is not part of the problem at
hand.
b; but our earlier discussion of the The straight lines running from one meas-
rst-iob measure suggests we should not ured variable to another represent direct (or
overemphasize the difference between r", net) influences. The symbol for the path co-
and r"u. Each, however, is substantially efficient, such as Py1y, canies a double sub-
greater than r"r, which in turn is rather more script. The first subscript is the variable at
impressive than rrr. the head of the path, or the effect; the second
Figure 1 is a graphic representation of the is the causal variable. (This resembles the
system of relationships among the five vari- convention for regression coefficients, where
ables that we propose as our basic model. The the first subscript refers to the "dependent"
numbers entered on the diagram, with the ex- variable, the second to the "independent"
ception of r*y, are path coefficients, the esti- variable. )

mation of which will be explained shortly. Finally, we see lines with no source indi-
First we must become familiar with the con- cated carrying arrows to each of the effect
ventions followed in constructing this kind of variables. These represent the residual paths,
diagram. The link between V and X is shown standing for all other influences on the vari-
as a curved line with an arrowhead at both able in question, including causes not recog-
ends. This is to distinguish it from the other nized or measured, errors of measurement,
lines, which are taken to be paths of influence. and departures of the true relationships from
In the case of V and X we may suspect an ln- additivity and linearity, properties that are as-
fluence running from the former to the latter. sumed throughout the analysis.
The Process of Stratification 395

An important feature of this kind of causal presumed to operate in some other way it
scheme is that variables recognized as effects must be represented in the diagram in accor-
of certain antecedent factors may, in turn, dance with its causal role, even though it is
serve as causes for subsequent variables. For not measured. Sometimes it is possible to de-
example, U is caused by V and X but it in duce interesting implications from the inclu-
turn influences W'and Y The algebraic repre- sion of such a variable and to secure useful es-
sentation of the scheme is a system of equa- timates of certain paths in the absence of
tions, rather than the single equation more of- measurements on it, but this is not always so.
ten employed in multiple regression analysis. A partial exception to the rule that all causes
This feature permits a flexible conceptualiza- must be explicitly represented in the diagram
tion of the modus operandi of the causal net- is the unmeasured variable that can be as-
work. Note that Y is shown here as being in- sumed to operate strictly as an intervening
fluenced directly by W U, and X but not by variable. Its inclusion would enrich our un-
V (an assumption that will be justified derstanding of a causal system without invali-
shortly). But this does not imply that V has no dating the causal scheme that omits it. Sociol-
influence on Y V affects U, which does affect ogists have only recently begun to appreciate
Y both directly and indirectly (via W). More- how stringent are the logical requirements
over, V is correlated with X and thus shares that must be met if discussion of causal pro-
in the gross effect of X on X which is partly cesses is to go beyond mere impressionism
direct and partly indirect. Hence the gross ef- and vague verbal formulations.e 'We are a
fect of V on Y previously described in terms long way from being able to make causal in-
of the correlation r"u, is here interpreted as ferences with confidence. and schemes of the
being entirely indirect, in consequence of \/s kind presented here had best be regarded as
effect on intervening variables and its correla- crude first approximations to adequate causal
tion with another cause of Y. models.
On the empirical side, a minimum test of
the adequacy of a causal diagram is whether it
Path Goelficient$ satisfactorily accounts for the observed corre-
lations among the measured variables. ln
'S7hether
a path diagram, or the causal making such a test we employ the fundamen-
scheme it represents, is adequate depends on tal theorem in path analysis, which shows
both theoretical and empirical considerations. how to obtain the correlation between any
At a minimum, before constructing the dia- two variables in the system, given the path co-
gram we must knoq or be willing to assume, efficients and correlations entered on the dia-
a causal ordering of the observed variables gram.1o $Tithout stating this theorem in gen-
(hence the lengthy discussion of this matter eral form we may illustrate its application
earlier in this chapter). This information is ex- here. For example,
ternal or a priori with respect to the data, ryx= Pyx+ Pyurux+ Pvtdwxl
which merely describe associations or correla-
tions. Moreover, the causal scheme must be and
complete, in the sense that all causes are ac- Twx=Pwx+Puturux.
counted for. Here, as in most problems in-
volving analysis of observational data, we We make use of each path leading to a given
achieve a formal completeness of the scheme variable (such as Y in the first example) and
by representing unmeasured causes as a resid- the correlations of each of its causes with all
ual factor, presumed to be uncorrelated with other variables in the system. The latter corre-
the remaining factors lying behind the vari- lations, in turn, may be analyzed; for exam-
able in question. If any factor is known or ple, r*y, which appeared as such in the first
3S6 lU / Genenating Inequality

TABLE 2
equation, is broken down into two parts in
Partial Regression Coefficients in Standard Form
the second. A complete expansion along these
(Beta Coefficients) and Coefficients of Determination,
lines is required to trace out all the indirect for Specified Combinations of Variables
connections between variables; thus,
Indeoendent Varlables a C@fflcient of
rvx= Pyx* PvuSux+ PvuPuf vx+ Pvv?wx Dependeot Determlmtion
Variablee (R2)
* PvwQvtPux + Pvw?wuPuvrvx'
Ub .279 .3r0 .26
Now, if the path coefficients are properly .433 .214 -026 .33
.224 .33
estimated, and if there is no inconsistency in I
440
. L20 -.014 .43
the diagram, the correlations calculated by a rn . 394 . r15 .43
Y .428 .42
formula like the foregoing must equal the ob-
av: Fatrer's education.
served correlations. Let us compare the values
X: Fat}lerrE @c. status.
computed from such a formula with the cor- U: ReBpondentrs educatlon.
W: First-job status.
responding observed correlations: Y: 1962 @c status.
bBeLa c@fficients ln tiese set8 talen as estlmates of
T*u=P1v,vtvv+PwuTuv path c@fflclents for Figure l.

= (.224)(.s16) + (.440)(.4s3)

= .116 + .199 = .315 Table 2 records the results of the regression


which compares with the observed value of calculations. It can be seen that some alterna-
.332; and tive combinations of independent variables
were studied. It turned out that the net regres-
Tvv = P vur uv + P vxrxv + P vu[wv sions of both'$7 and Y on V were so small as
to be negligible. Hence V could be disre-
= (.3e4)(.4s3) + (.115)(.516) + (.281)(.31s) garded as a direct influence on these variables
without loss of information. The net regres-
= .326 sion of Y on X was likewise small but, as it
(using here the calculated rather than the ob- appears, not entirely negligible. Curiously,
served value of r*u), which resembles the ac- this net regression is of the same order of
tual value, .322. Other such comparisons- magnitude as the proportion of occupational
for roy, for example-reveal, at most, trivial inheritance in this population-about 10 per
discrepancies (no larger than .001). cent, as discussed [elsewhere].12 'We might
\7e arrive, by this roundabout journeS at speculate that the direct effect of father's oc-
the problem of getting numerical values for cupation on the occupational status of a ma-
the path coefficients in the first place. This in- ture man consists of this modest amount of
volves using equations of the foregoing type strict occupational inheritance. The remain-
inversely. We have illustrated how to obtain der of the effect of X on Y is indirect, inas-
correlations if the path coefficients are much as X has previously influenced U and
known, but in the typical empirical problem \7, the son's education and the occupational
we know the correlations (or at least some of level at which he got his start. For reasons
them) and have to estimate the paths. For a noted [elsewhere]13 we do not assume that the
diagram of the type of Figure 1 the solution full impact of the tendency to take up the fa-
involves equations of the same form as those ther's occupation is registered in the choice of
of linear multiple regression, except that we first job.
work with a recursive system of regression With the formal properties of the model in
equationsll rather than a single regression mind we may turn to some general problems
eQuatton. confronting this kind of interpretation of our
The Process of Stratification 397

The relevant question about the residual is


from Fi not really its size at all, but whether the unob-
cients in the served factors it stands for are properly repre-
tors. that is. varia sented as being uncorrelated with the mea-
sured antecedent variables. 'We shall entertain
tation of the extent to which measured causes [elsewhere]1a some conjectures about unmeas-
ured variables that clearly are not uncorre-
lated with the causes depicted in Figure 1. It
tained from the coefficient of determination; turns out that these require us to acknowl-
if R?MU"\ is the squared multiple correla- edge certain possible modifications of the dia-
tion of Y on the three independent variables, gram, whereas other features of it remain
then the residual for Y is " t-=-nfrrrr.1 more or less intact. A delicate question in this
regard is that of the burden of proof. It is all
too easy to make a formidable list of unmeas-
ured variables that someone has alleged to be
crucial to the process under study. But the
f?ean to live in a so- mere existence of such variables is already ac-
ciety where nearly perfect explanation of knowledged by the very presence of the resid-
the dependent variable could be secured by ual. It would seem to be part of the task of the
studying causal variables like father's occu- critic to sbow, if only hypothetically, but
pation or respondent's education. In such a specifically, how the modification of the
society it would indeed be true that some causal scheme to include a new variable
are "destined to poverty almost from birth would disrupt or alter the relationships in the
. . . by the economic status or occupation of original diagram. His argument to this effect
their parents" (in the words of the reference could then be examined for plausibility and
cited in endnote 1). Others, of course, his evidence, if any, studied in terms of the
would be "destined" to affluence or to empirical possibilities it suggests.
modest circumstances. By no effort of their Our supposition is that the scheme in Fig-
own could they materially alter the course ure 1 is most easily subject to modification by
of destinS nor could any stroke of fortune, introducing additional measures of the same
good or ill, lead to an outcome not already kind as those used here. If indexes relating to
in the cards. socioeconomic background other than V and
X are inserted we will almost certainly esti-
mate differently the direct effects of these par-
ticular variables. If occupational statuses of
the respondent intervening between 'W and Y
that an explanation is correct or nearly so, were known we should have to modify more
whereas a low percentage of determination or less radically the right-hand portion of the
means that a causal interpretation is almost diagram. Yet we should argue that such modi-
certainly wrong. The fact is that the size of fications may amount to an enrichment or ex-
the residual (or, if one prefers, the propor- tension of the basic model rather than an in-
tion of variation "explained") is zo guide validation of it. The same may be said of
whatever to the validity of a causal interpre- other variables that function as intervening
tation. The best-known cases of "spurious causes. In theory, it should be possible to
correlation"-a correlation leading to an specify these in some detail, and a major part
egregiously wrong interpretation-are those of the research worker's task is properly de-
in which the coefficient of determination is fined as an attempt at such specification. ln
quite high. the course of such work. to be sure. there is
3S8 lV / Eenenating Inequality

always the possibility of a discovery that intersected more than once in one compound
would require a fundamental reformulation, path. Having traced all such possible com-
making the present model obsolete. Discard- pound paths, we obtain the entirety of indi-
ing the model would be a cost gladly paid for rect effects as their sum.
the prize of such a discovery. Let us consider the example of effects of ed-
Postponing the confrontation with an al- ucation on first job, U on W The gross or to-
tered model, the one at hand is not lacking in tal effect is r*u =.538. The direct path is Pwu
= .440. There are two indirect connections or
compound paths: from W'back to X then for-
ward to U; and from W back to X, then back
tion coefficient and the path coefficient have to V and then forward to U. Hence we have:
the same dimensionality. The correlation r"* rwu :
I Pr*Pu* I PwxrxvPw
Pwu
= .405 (Table 1) means that a unit change
(one standard deviation) in X produces a (gross) (direct) (indirect)
change of 0.4 unit in I in gross terms. The
path coefficient,pyx = .115 (Figure 1), tells us or, numerically,
that about one-fourth of this gross effect is a 8 = .440 + (.224) (.27
.5 3 9 \ + 1.224) (. 5 16) (.3 1,0)
result of the direct influence of X on Y (We
speculated above on the role of occupational
= .440 + .062 + .036
inheritance in this connection.) The remain-
der (.405 - .115 = .29\ is indirect, via U and
= .440 + .098.
VZ The sum of all indirect effects. therefore, is
given by the difference between the simple In this case all the indirect effect of U on \X/
correlation and the path coefficient con- derives from the fact that both U and W'have
necting two variables. \7e note that the in- X (plus V) as a common cause. In other in-
direct effects on Y are generally substantial, stances, when more than one common cause
relative to the direct. Even the variable tem- is involved and these causes are themselves in-
porally closest (we assume) to Y has "indi- terrelated, the complexity is too great to per-
rect effects"-actually, common antecedent mit a succinct verbal summary.
causes-nearly as large as the direct. Thus A final stipulation about the scheme had
rytv = .541 and p"* - .281', so that the aggre- best be stated, though it is implicit in all the
gate of "indirect effects" is .26, which in this previous discussion. The form of the model it-
case are common determinants of Y and W self, but most particularly the numerical esti-
that spuriously inflate the correlation between mates accompanying it, are submitted as valid
tnem. only for the population under study. No claim
To ascertain the indirect effects along a is made that an equally cogent account of the
given chain of causation we must multiply the process of stratification in another society
path coefficients along the chain. The proce- could be rendered in terms of this scheme. For
dure is to locate on the diagram the depen- other populations, or even for subpopulations
dent variable of interest, and then trace back within the United States, the magnitudes
along the paths linking it to its immediate and would almost certainly be different, although
remote causes. In such a tracing we may re- we have some basis for supposing them to
verse direction once but only once, following have been fairly constant over the last few
the rule "first back, then forward." Any bidi- decades in this country. The technique of path
rectional correlation may be traced in either analysis is not a method for discovering
direction. If the diagram contains more than causal iaws but a procedure for giving a quan-
one such correlation, however, only one may titative interpretation to the manifestations of
be used in a given compound path. In tracing a known or assumed causal system as it oper-
the indirect connections no variable may be ates in a particular population. When the
The Process of Stratification 3S9

same interpretive structure is appropriate for what is specifically meant by "vicious circle,"
two or more populations there is something what are the operational criteria for this con-
to be learned by comparing their respective cept, and what are the limits of its usefulness.
path coefficients and correlation patterns. 'We To begin with, there is the question of
have not yet reached the stage at which such fact-or, rather, of how the quantitative facts
comparative study of stratification systems is
feasible. . . .

tus" (presumably reference is to the status o


the family of orientation)? '!7e have found
Ihe Concept ol a Uicious Gincle that the father-son correlation for occupa-
tional status is of the order of .4. (Assuming
Although the concept of a " cycle of attenuation by errors of measurement, this
poverty" has a quasi-official sanction in U. should perhaps be revised slightly upward.)
S. public policy discussion, it is difficult to Approaching the measurement problem in an
locate a systematic explication of the con- entirely different way, we find that the
cept. As clear a formulation as any that may amount of intergenerational mobility between
be found in academic writing is perhaps the census major occupation groups is no less
following:15 than seven-eighths as much as would occur if
there were no statistical association between
Occupational the two statuses whatsoever. or five-sixths as
They are associated with much as the difference between the "mini-
many. factors which make it difficult for individuals mum" mobility involved in the intergenera-
to modify their status. Position in the social struc- tional shift in occupation distributions and
the amount required for "perfect" mobility.tz

vicious circle in which each factor acts on the other


in such a way as to preserve the social structure in hibited by Lipset and Bendix to indicate the
its present form, as well as the individual family's contrary.) If the existing amount of modifica-
position in that structure. . . . The cumulation of tion of status is insufficient in terms of some
disadvantages (or of advantages) affects the indi- functional or normative criterion implicitly
vidual's entry into the labor market as well as his employed, the precise criterion should be
later opportunities for social mobility. made explicit: How mwch mobility mwst oc'
cur t
The suspicion arises that the authors in
preparing this summary statement were partly -G*t, take the postulate that occupational
captured by their own rhetoric. Only a few status (of origin) is "associated with many
pages earlier they had observed that the factors" and that "each factor acts on the
"widesoread variation of educational attain- other" so as "to preserve . . . the individual
family's position." Here the exposition virtu-
ally cries out for an explicit quantitatiue
rather than a determining role."15 But is an causal model; if not one of the type set forth
--enaDllng and motlvatrng role" Iogrcally aoe- in the first section of this chapter, then some
quate to the function of maintaining a "vi- other model that also takes into account the
cious circle"? In focusing closely on the pre- way in which several variables combine therr
cise wording of the earlier quotation we are effects. Taking our own earlier model, for
not interested in splitting hairs or in generat- want of a better alternative, as representative
ing a polemic. It merely serves as a convenient of the situation, what do we learn about the
point of departure for raising the questions of "associated factors"? Family "position" is,

-
400 M Eencnating Inequallty

indeed, "associated with . . . education," and pertinent results: r", = .400; R"r*rt = .425;
education in turn makes a sizable difference R",urr, = .651,. Note that adding the "associ-
in early and subsequent occupational achieve- ated factor" of father's education to father's
ment. Yet of the total or effect of educa- occupation increases very slightly our esti-
tion (U) on I occupational status mate of the influence of "family position" on
@ part consists in a occupational achievement. Including respon-
nsmtssron o dent's education, however, makes quite a
6-osition." at Ieast as thts rs tn meas- striking difference. Squaring these coefficients
+-i-, to yield an accounting of the total variation in
ured variables V (father's education) and X
(father's occupation). . . . A relevant calcula- respondent's 1,962 occupational status (Y), we
tion concerns the compound paths through V obtain these percentages:
and X linking Y to U. Using data for men 20 (i ) Gross (or total) effect of father's
to 64 years old with nonfarm background, we
education and occupation 18.06
find:

PvxPux= '025 (li) Education of respondent,


independent of (i) 24.32
PvxrxvPuv = '01'4
(iii) All other factors, independent of
PvxPwxPux= '014 (i) and (ii) 57.62

TOTAL 100.00
PvwPwxrxvPuv = '008
An analogous calculation, derived from
$urn = .051
multiple-classification rather than linear-re-
gression statistics, was offered [elsewhere].18
The results are rather similar. Here we have
imputed to the measures of "family position,"
X and V their total influence, including such
sive of prior influence of father's education part of this as works through education; the
and occupation on respondent's first job) is 24 per cent contribution of respondent's edu-
PvwPrvu = .128, for a total of .535. Far from cation refers only to the part of the effect of
serving in the main as a factor perpetuating education that is net of the background fac-
initial status, education operates primarily to tors. Still. education has a
induce variation in occupational status that is ind
independent of initial status. The simple rea-
son is that the large residual factor for U is an vershadowing both these compo-
indirect cause of Y But by definition it is quite nents, of course, is the unexplg14gg! v4ria.lgign
uncorrelated with X and V This is not to of nearly 58 per cent, which can have nothinp
gainsay the equally cogent point that the de- to do with "perpetuating status."
'SThatever
gree of "perpetuation" (as measured by r"y) the merit of these observations.
that does occur is mediated in large part by they should at least make clear that statistical
education. results do not speak for themselves. Rather,
This conclusion is so important that we the findings of a statistical analysis must be
should not allow it to rest on a single calcula- controlled by an interpretation-one that
tion. The reader accustomed to a calculus of specifies the form the analysis will take-and
"explained variation" may prefer the follow- be supplemented by further interpretations
ing. For men 35 to 44 years of age with non- that (ideally) make explicit the assumptions
farm background (a convenient and not un- on which the analyst is proceeding. The form
representative illustration), we have these in which our results are presented is dictated
Tbe Process of Stratffication 40t

by a conception of status achievement as a other, then their combined effect will consist
temporal process in which later statuses de- largely in redundancy, not in "cumulation."
pend, in part, on earlier statuses, intervening This circumstance does not relieve us from the
achievements, and other contingent factors. ln necessity of trying to understand better bow
such a framework it may not be a meaningful the effects come about (a point also illustrated
task to evaluate the relative importance of dif- in a less fortunate way in Pearson's work). It
ferent causal factors. Instead. attention is fo- does imply that a refined estimate of how
cused on how the causes combine to produce much effect results from a combination of
the end result. From this ooint of view we can "associated factors" will not differ greatly
from a fairly crude estimate based on the two
or three most important ones. Sociologists
have too long followed the mirage of "in-
and to what extent thrs creasing the explained variance.". . .
'We
effect is transmitted via measured intervening do not wish to imply that the idea of
variables and, finally, to what extent such in- cumulation of influences, or even the particu-
tervening variables contribute to the outcome, lar form of cumulation describable as a "vi-
independently of their role in transmission of cious circle," is without merit. Our aim is to
prior statuses. In a balanced interpretation all call attention to the necessity of specifying the
these questions should be dealt with explicitly. actual mechanism that is only vaguely sug-
Our treatment seems to indicate the advis- gested by such terms. One legitimate meaning
ability of keeping in perspective the magnr- of cumulation is illustrated by the model of a
tude of the gross relationship of background synthetic cohort presented [elsewhere].20 In
factors and status of origin to subsequent this case what is cumulative is the experience
achievement. The relationship is not trivial, of an individual or a cohort of individuals
nor is it, on the other hand, great enough in over the life cycle, so that in the latter part of
itself to justify the conception of a system that the life cycle achieved status depends heavily
insures the "inheritance of poverty" or other- on prior achievements, whatever the factors
wise renders wholly ineffectual the operation determining those achievements may have
of institutions supposedly based on universal- been. The cumulation here consists in large
istic principles. measure of the effects of contingent factors
Our model also indicates where the "vi- not related to social origins or measured
cious circle" interpretation is vulnerable. In background factors.
the passage on the vicious circle quoted there The situation of the Negro American,
seems to be an assumption that because of the which is analyzed [elsewhere],21 exemplifies
substantial inter-correlations between a num- mechanisms inviting the label of a vicious cir-
'What
ber of background factors, each of which has cle. is crucial in this case is not merely
a significant relationship to subsequent that Negroes begin life at a disadvantage and
achievement, the total effect of origin on that this initial disadvantage, transmitted by
achievement is materially enhanced. Here, rn intervening conditions, has adverse effects on
other words, the concept of "cumulation" ap- Iater careers. Rather, what happens is that, tn
pears to refer to the intercorrelations of a col- addition to the initial handicap, the Negro ex-
lection of independent variables. But the ef- periences further handicaps at each stage of
fect of such intercorrelations is quite opposite the life cycle. lfhen Negroes and whites are
to what the writers appear to suppose. They equated with respect to socioeconomic cir-
are not alone in arguing from a fallacious as- cumstances of origin and rearing, Negroes se-'
sumption that was caustically analyzed by cure inferior education. But if we allow for
Karl Pearson half a century ago.1e The crucial this educational disadvantage as well as the
point is that if the several determinants are in- disadvantage of low social origins, Negroes
deed substantially intercorrelated with each find their way into first jobs of lower status
402 lV / Eencnating Incquality

than whites. Again, allowing for the handicap cious circle resulting from discrimination and
of inferior career beginnings, the handicap of poverty is a challenge a democratic society
lower education, and the residual effect of must face, in our opinion. To advocate this
low socioeconomic origins-even with all policy, however, is not the same as claiming
these allowances-Negroes do not enjoy com- that all ascriptive constraints on opportuni-
parable occupational success in adulthood. In- ties and achievements could or should be
deed, even though we have not carried our eliminated. To eliminate all disadvantages
own analysis this far, there is good evidence that flow from membership in a family of ori-
that Negroes and whites do not have equal in- entation-with its particular structure of in-
comes even after making allowance for the oc- terpersonal relationships, socioeconomic
cupational status difference and the educa- level, community and regional location, and
tional handicap of Negroes.z2 Thus there so on-would by the same token entail elimi-
surely are disadvantaged minorities in the nating any aduantages the family can confer
United States who suffer from a "vicious cir- or provide. If parents, having achieved a de-
cle" that is produced by discrimination. But sirable status, can ipso facto do nothing to
not all background factors that create occupa- make comparable achievement easier for
tional handicaps are necessarily indicative of their offspring, we may have "equal opportu-
such a vicious circle of cumwlatiue disadvan- nity. " But we will no longer have a family
tages; the handicaps of the Southern whites, system-at least not in the present under-
for example, are not cumulative in the same standing of the term. (This point has not been
sense.23 A vicious circle of cumulative impedi- misunderstood in radical, particularly Marx-
ments is a distinctive phenomenon that should ist, ideologies.)
not be confused with any and all forms of dif- We do not contemplate an effortless equi-
ferential occupational achievement. librium at some optimum condition where the
claims of egalitarian values and the forces of
family attachment are neatly balanced to the
satisfaction of all. A continuing tension be-
tween these ultimately incompatible tenden-
cies ma5 indeed, be a requisite for social
we must at least aftempt to avoid having our progress. !7e do contend that both equity and
position misunderstood. We have not vouch- effectiveness in the policy realm call for a
safed a "functional interpretation" that as- deeper understanding of the process of strati-
serts that somehow American society has just fication than social science and politics yet
the right amount of stratification and just the can claim.
appropriate degree of intergenerational status
transmission . We haue indicated that it is easy
to exaggerate the latter and, in particular, that
it is possible seriously to misconstrue the na-
ture of the causal relationships in the process Notes
that characterizes status transmission between 1. Mollie OrshanskS "Children of the Poor,"
generatrons. Social Security Bulletin, 26(Jdy 19 63).
2. Forrest A. Bogan, "Employment of High
School Graduates and Dropouts tn 1964," Special
Labor Force Report, No. 54 (U. S. Bureau of La-
bor Statistics, June 1965), p.643.
3. Peter M. Blau and Otis Dudley Duncan, The
American Occupational Structure, New York: The
al. To Free Press. 1957. ch.4.
4. rbid.
Inequality 403

'Who 1.4.lbid., ch. 5.


5. Bruce K. Eckland, "College Dropouts
Came Back," Haruard Educational Reuiew, 34 15. Seymour M. Lipset and Reinhard Bendix,
(1964\,402420. Social Mobility in Industrial Society, Berkeley:
5. Beverly Duncan, Family Factors and School Univ. of California Press, 1959, pp. 198-1'99.
Dropout: 1920-1960, U. S. Office of Education, 1,6.Ibid., p. 190.
Cooperative Research Project No. 2258, Ann Ar- 17. U. S. Bureau of the Census, "Lifetime Occu-
bor: Univ. of Michigan, 1965. pational Mobility of Adult Males: March L962,"
7.BIau and Duncan, op. cit., Appendix C. Current Population Reports, Series P-23, No. 11
8. Gosta Carlsson, Social Mobility and Class (May 12, 1964), Table B.
Strwctwre, Lund: C\VK Gleerup, 1958,p.724. 18. Blau and Duncan, op. cit., ch.4.
9. H. M. Blalock, Jr., Causal Inferences in Non- 19. Karl Pearson, "On Certain Errors with Re-
experimental Research, Chapel Hill: Univ. of gard to Multiple Correlation Occasionally Made
North Carolina Press, 1964. by Those'Who Have Not Adequately Studied This
10. Sewall Sfright, "Path Coefficients and Path Subject," Biometrika, 1,0(19141, 1.81-187.
Regressions," Biometrics, 16(1.960l,, L89-202; 20. Blau and Duncan, op. cit., ch. 5.
Otis Dudley Duncan, "Path Analysis," American 21,.lbid., ch. 6.
J ournal of Sociology, 72(19 66), 1-16. 22. See Herman P. Miller. Rich Man, Poor Man,
11. Blalock, op. cit., pp. 54ff. New York: Crowell, 1964, pp.90-96.
12.Blal; and Duncan, op. cit., ch.4. 23.BIau and Duncan, op. cit., ch.6.
|.1- |ota.- cn- .1-

S-ar putea să vă placă și