Further Considerations on the Methodological Analysis of Segregation Indices

Author(s): Charles F. Cortese, R. Frank Falk and Jack K. Cohen

Source: American Sociological Review, Vol. 41, No. 4 (Aug., 1976), pp. 630-637
Published by: American Sociological Association
Stable URL: http://www.jstor.org/stable/2094840
Accessed: 04-03-2015 20:30 UTC

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
http://www.jstor.org/page/info/about/policies/terms.jsp

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content
in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship.
For more information about JSTOR, please contact support@jstor.org.

American Sociological Association is collaborating with JSTOR to digitize, preserve and extend access to American Sociological
Review.

http://www.jstor.org

This content downloaded from 129.108.202.79 on Wed, 04 Mar 2015 20:30:35 UTC
All use subject to JSTOR Terms and Conditions
 FURTHER CONSIDERATIONS ON THE METHODOLOGICAL
ANALYSIS OF SEGREGATION INDICES *

CHARLES F. CORTESE, R. FRANK FALK and JACK K. COHEN

University of Denver

American Sociological Review 1976, Vol. 41 (August): 630-637

The process of developing an adequate measure of segregation occupied the literaturefor

over a decade and culminated in the widespread use of the Index of Dissimilarity. The
inadequacies of this index, although identified by the Duncans (1955), remain with us
and largely have come to be ignored. This research further explores the difficulties per-
taining to limitations in the use and interpretationof the Index of Dissimilarity, demon-
strates some of the systematic biases resulting from these inadequacies and provides a
mathematical refinement which overcomes some of the major problems inherent in the
use of this index.

The concept of ecological segregation article was followed by criticism as well

has never been dealt with adequately in as the developmentof differentmeasures
definitionalterms (cf. Duncanand Duncan, (Hornseth, 1947; Jahn et al., 1948; Jahn,
1955:217). Instead of defining segregation, 1950; Williams, 1948; Cowgill and Cow-
most workhas consideredhow the opposite gill, 1951) whicheventuallyled to Duncan
of segregation-often called assimilation- and Duncan (1955) demonstratingthe
should be defined. A clear and proper mathematical relationships between the
definitionof assimilationis especiallyneces- segregationindices previouslypresented.
sary since,in fact, most attempts(including The Duncans suggested that the Index
ours) to measure relative segregation are of Dissimilarity(D) was the most useful
basedon a priori "ideal"distributions.That of these indices (1955:214-5) and it
is, the complicateddetails of geographical has, in fact, achieved preeminencein the
distributionand clusteringare ignored in measurementof segregation.Taeuber and
an effort to get a simple overall index of Taeuber (1965) have presented what is
segregation.Heretofore,the "ideal"distri- probably the most detailed discussion on
bution has usually been taken to be the the measurementof segregation to date.
even distribution.Jahn et al. (1947) were Figure 1, reproducedfrom Duncan and
concernedwith the questions inherent in Duncan (1955), shows a geometricaldef-
such operationalizationof the concept of inition of D as the maximumvertical dis-
"ecologicalsegregation."In that early arti- tance between the "segregationcurve"and
cle, they formulatedtheir index of dissim- the curve Y = X. The segregationcurve
ilarityon the expectedvalue of the central is the plot of the cumulativeproportionof
tendency of a random distribution.They whites versus the cumulativeproportionof
stated: "This means that if there is no nonwhites, where the respective cumula-
segregation,then members of a minority tions are obtained by taking the census
racial group will be distributedrandomly tracts in order of increasingnonwhiteper-
throughoutthe various census tracts of a centages.
city. For example, if ten percent of the The Duncans provideda set of inade-
population of a city is Negro, then each quacies and precautionarycommentsabout
census tract would be expected to have the interpretationof segregation indices.
a Negro populationof approximatelyten Our articlebuilds upon the Duncans'com-
percent" (Jahn et al., 1947:293). Their ments and further explores the "mathe-
maticalproperties[of the indices]of which
* This research was funded in part by a grant
their proponentswere unaware,and which
from N.S.F. through the UndergraduateResearch
Participation Program, Grant Number GY-9959. lead to difficultiesof interpretation" (Dun-
The position of authors' names does not imply can and Duncan, 1955:210). The objec-
ordinal properties. tions to D have resultedin earlierattempts
630

This content downloaded from 129.108.202.79 on Wed, 04 Mar 2015 20:30:35 UTC
All use subject to JSTOR Terms and Conditions
 ANALYSIS OF SEGREGATIONINDICES 631
throughoutthe population. Our research
Figure I
has focused on an examination of the
random fluctuationsaround this expecta-
tion as significantinformation.It is our
argument that the opposite of segrega-
tion is more appropriatelya randomness
D of distributionratherthan an hypothetical
"evenness." It might be natural, there-
fore, to construct an index which takes
the value of zero when the distributionis
0.
random.However,due to the extensiveuse
of D, we preferto accountfor randomness
by computing the expected value of D
o X. Cum. Prop. of Nonwhites ? under the hypothesisthat the distribution
of nonwhites in any census tract is ran-
dom. That is, one could compute a value
to correct the measure (cf. Leasure and for D which would be expected,given that
Stem, 1968), but thus far it seems the the minority population were distributed
major problems are yet to be dealt with randomlythroughoutthe entirepopulation.
adequately. More explicitly,we assumethat for a cen-
sus unit with Ti residents, we randomly
Objectionsto D select without replacement, Ti individuals
from the total populationof the city. To
The major objections to the Index of computethe expectedvalueof D, one needs
Dissimilarityfollow: the probabilitythat the number of non-
1. The expectationof "evenness"as the whites (Ni) has some particular value
oppositeof segregationis not as use- (ni). The assumptionsof samplingwith-
ful in most cases as the concept of out replacementand a populationdivided
"randomness."9 into two categories (i.e., white and non-
2. D is affected by differencesin the white) specifies that this probability is
proportion of the minority in the given by the hypergeometricdistribution
population, thus preventing inter- (see equation (2) below). It is our con-
city comparisons. tention that the comparisonof the actual
3. D is affected by the size (number D for a city with its expected value con-
of households) of the areal unit of tains more informationthan D itself and
analysis. can alleviate some of the limitations of
4. The present interpretationof D as D. This comparisonis most sensibly car-
the proportion of nonwhites who ried out in units of the standarddeviation
would have to change their tract of of D (cf. Bardwell,1971) underthe same
residenceto make the distributionof assumptionsused for the expected value
the minority even throughout the of D. To compute the standarddeviation
city (Duncan and Duncan, 1955: of D, one needs the joint probabilitythat
211) is misleadingsince it does not in two census units with populationsTi
include the concept of replacement and Tj, the numbers of nonwhites (N,
of the relocatedminority. and Nj) in those units have some particular
Some of these objections are obviously values (ni and nj).
interrelated.We will deal with each briefly. In general, anytimethat more than one
The idealof evennesshas been used pre- census unit is involved in the placement
dominatelyas illustratedby the fact that of a given total number of individuals,
the zero point of most of the indices dis- interdependencyexists. This interdepen-
cussed by the Duncans (1955), including dency is accountedfor by the joint proba-
the Index of Dissimilarity itself, occurs bility distributionin the calculationof the
when the minority is distributed evenly standarddeviation.

This content downloaded from 129.108.202.79 on Wed, 04 Mar 2015 20:30:35 UTC
All use subject to JSTOR Terms and Conditions
632 AMERICAN SOCIOLOGICAL REVIEW
The joint distributionwas derived by
conceptuallyselecting Ti plus Tj individ-
uals for the two census units and then
selecting Ti of these for the ith unit. This
joint probabilityis given by the product
of two hypergeometric distributions as
shownin equation(3) of the appendix.
We proposethat the "standardscore"of
D,
ZYD
(=1
is more meaningfulthan D itself.
An additionalmotivationfor building
uponD is that even thoughevennesswould
not occur naturally, evenness is often a
social goal. Examples of this are recent
policies and attemptsto distributeminority
teachersevenly througha school system.
The other three objectionsor limitations
raised above have been previouslypointed
out but not clearly demonstrated(Dun-
can and Duncan, 1955:216). Jahn et al.
(1947:294) stated that "a satisfactory
measure of ecological segregationshould
... not be distorted by . . . the proportion
of Negroes."D is affectedby the propor-
tion of nonwhitesin a particularcity and
is furtheraffectedby the numberof house-
holds per unit of analysis (Taeuber and
Taeuber, 1965:231-5). We demonstrate
these effects in Table 1 which provides
the expected value of the Index of Dis-
similarityfor computer-generatedrandom
placementwith varyingproportionsof non-
whites and under the ideal conditions of
many areal units with an equal numberof
householdsin each unit.
The expected values in Table 1 shows
that the random fluctuationscan make a
significantdifferencein the expectedvalue
of D. The pattern is apparent-as q in-
creases to .5, the expected value of D de-
creases. This relationship,of course, re-
verses as q exceeds .5. For example,a city
with only 5% of the populationnonwhite
and a low density, such as 10 households
per block, has an expected value of D of
.63 even when the population has been
) residentiallydistributedwithout regard to
race or any other demographiccharacter-
istic. An observed D value of .63 easily
can be misinterpretedas evidence of a
highly segregatedcity when, in fact, it cor-
respondsto no segregation.In otherwords,
this score is a result of particularvalues
of q and N and not of any reality of seg-
regation.It is thereforemeaninglessto use
D to compare the relative segregationof
this city to a city, which is 40% black
and has one hundredhouseholdsper block,
which has an expected score of .08 when
there is an equal lack of segregation.If
we were simply to subtractout the effects
of proportionand density (accomplished
by observed D minus expected D), we
would have a more realisticstartingpoint
than the conventional zero assumption.
We have suggestedthat the magnitudeof
this differenceis most sensibly carriedout
in units of the standarddeviation.Thus the
scores in Table 1 shouldbe consideredthe
zero points under the random assignment
of the populationto the census units. By
employingthe standardscore correspond-
ing to D for a particularcity with its par-
ticular proportionof nonwhites and with
its particularblock structure,we obtain an

Table 1. Expected Value of the Index of Dissimilarity I with Random Placement of Nonwhites
throughout a City
(N) Number of Households in Each Areal Unit (e.g., Block)
10 25 50 100 1000
.01 .914 .786 .611 .370 .127
(q) Proportionof .02 .833 .615 .372 .273 .090
Minority in the .05 .630 .369 .264 .180 .043
City's Population .10 .387 .272 .185 .132 .042
.20 .301 .196 .140 .099 .032
.30 .266 .176 .122 .087 .028
.40 .250 .161 .114 .081 .026
.50 .246 .161 .112 .080 .025
a The range of Index of Dissimilarity may vary from 0.000 to 1.000. The values in this table

were computed by use of the binomial approximation;see Appendix, equation (21).

This content downloaded from 129.108.202.79 on Wed, 04 Mar 2015 20:30:35 UTC
All use subject to JSTOR Terms and Conditions
 ANALYSIS OF SEGREGATIONINDICES
improvedmeasure of segregationfor that
city. Of course, the issue of the proper
unit of analysis remains moot and our
proposed analysis can be carried out for
any choice for which the appropriatecen-
sus data are available.
Despite the fact that the value of D
is affectedby the proportionof nonwhites
and the block structure,and thus partially
vitiates intercity comparisons,such com-
parisons have often been made in the
literature (cf. Taeuber and Taeuber,
1965; Bahr and Gibbs, 1967). The Taeu-
bers (1965:215) have stated, "Our pur-
pose in computingsegregationindexes is
to permit intercitycomparisons,and there
must be some sense in which equal values
mean equal degrees of segregation."The
proposedstandardscore measureincreases
the validityof such comparisons.
We have suggested another objection
which pertainsto the interpretationof the
index. D is interpretedas the proportion
of nonwhites who would have to change
their tract of residenceto make the distri-
butionof the minorityeven throughoutthe
city (Duncan and Duncan, 1955:211).
Such an interpretationis misleadingin that
D gives the proportionof the minoritythat
would have to be moved withoutreplacing
them with whites. In particular,if the cen-
sus tract were wholly populated by a
minority, the entire tract would have to
be evacuatedin this interpretation.Actu-
ally, what is often desiredis the proportion
of minoritypopulationwhich would have
to be exchangedwhile keepingthe number
of householdsper unit constant.It will be
shown in equation (9) below that this
"exchange proportion"is, in fact, given
by (l-q)D where q is the proportionof
the minorityin the population.'
The Standard Score Index
Our proposalis to computethe expected
value of D (,AD) and the variance of D
(0f2D) under the hypothesis of random-
ness and to use the standardscore
1 Professor Taeuber has reminded us, in cor-
respondence, that Duncan had pointed this out
(see Taeuber and Taeuber, 1965: 30f.) and that
Farley and Taeuber (1968:955) have used this
formula, identifying it as the "replacement in-
dex."
This content downloaded from 129.108.202.79 on Wed, 04 Mar 2015 20:30:35 UTC
All use subject to JSTOR Terms and Conditions
633
Z DEEAD
D
as an improved measure of segregation.
Here D is the measure of deviationfrom
evenness employed by the Duncans. The
quantity,1D may be interpretedas the value
of D which would be achievedif race had
no effect on the residential distribution.
The value of Z gives the degree to which
the actualdistributiondiffersfrom random-
ness as measuredin standardscores.
Since the exact distributionof Z is yet
to be determined,no probabilisticinterpre-
tation of Z is yet possible. However,what
we have providedis a necessaryfirst step
which does improveD in allowing,appar-
ently for the first time, the comparisonof
cities with differentnonwhite percentages.
In addition, studies of one city over
time have always been suspect due to the
strong possibilitythat q has changed over
the time period. Our improvementnow
allows us to deal with this.
If our basic idea for improvingD is ac-
ceptable,the next step is the determination
of the distributionof Z exactly or approxi-
mately or empirically.By empiricallywe
mean a categorizationof cities by their
Z-scores. For example, the knowledge of
which cities have Z-scores between 0 and
1, between 1 and 2, etc. would, by itself,
be valuableinformation.In fact, we believe
that such informationis probably more
important to sociologists than the exact
statisticalprobabilitiesof occurrence.These
calculationscould be accomplishedusing
the formulaein our mathematicalappendix.
While the details of computing Z are
left to the appendix,we now make precise
what we mean by the "hypothesisof ran-
domness."Suppose that a given city has
k census units containingrespectivelyT1,
T2, . . ., Tk households and suppose that
the city-wide proportion of nonwhites is
q. We conceive of each unit of size T1 as
being a randomsamplefrom the total pop-
ulation, T. We assume the samples are to
be selected without replacement,that is,
we exclude the possibility that the same
person occupies two or more residences.
Therefore,the distributionof N1, the num-
ber of nonwhites in unit, is hypergeo-
metric: that is,
634 AMERICAN SOCIOLOGICAL REVIEW

qT
{ 8 (l-q)T For our second example,we again con-
T
ni sider a city composedof two census tracts,
Prop (N1 = n )Ti-n 2)
O but now N1 = 0, Wi 150, and N2= 30,
-

(Ti)
W2 = 0 (i.e., a completelysegregatedcity).
Here, to bring about evenness without
We conclude this section by making a replacement,we must close down tract 2
general observationabout indices. If one completelyand move all its Ns to tract 1.
has a useful index and also a theorywhich 30
explainssome of the obvious and less rel- This is a proportion of -3 1 of the
30
evant factors (e.g., random effects), one minority.Fromthe formulafor D, we have:
can alwayspartialout these factorsby con-
structinga Z-score analogousto that pre-
sented above. D= ?/2(l 0 150 30 0

The Exchange Proportion Problem again in agreementwith our interpretation.

To obtain the result,we introduceT=
We now turn to a discussion of the Ni + WI, the total numberof residentsin
interpretationof D as the proportionof tract i; and T = N+ W, the total number
minoritywhich would have to change cen-
sus tracts to bring about evenness. We of residentsin the city; qi N the pro-
asserted above that D actually gives this portion of nonwhitesin tract i; and q
proportion on the assumption that the N
minority residents are not replaced by the city-wideproportionof nonwhites.
majorityresidents.We illustratethis point
with two simpleexamplesbefore passingto Considera census tract i for which we
the general case. If in the ith census tract have too high a proportionof the minority,
there are Ni nonwhitesand WI whites and i.e., qi>q. We must remove enough non-
if in the city there are N E Ni non- whites from this tract to bring this propor-
i tion down to q. Say we remove RI: since
whites and W . WI whites, then (see we do not replace these people we require
-

i -
Duncan and Duncan, 1955) q Ni Ri (4)
k N1 W,
D =1/2 (3) Solving the equationfor RI, we have
| N W
R= Ni -Tiq (5)
(5 T,(qj-q)
1- q 1- -q
Here k denotes the total numberof census
tractsin the city. Thus, from all the census tracts with too
For our firstexample,we considera city large a proportion of nonwhites, we re-
with only two tracts for which N1 = 10, move
W1 150, and N2 30, W2 =150.
Clearly, to bring about evenness without EvRj 1q vTj(q -q).
replacement, we need merely to move
10 Ns from tract 2 to tract 1. This is a qi>q qj>q
10 That is, a proportion
proportion ?1/4of the minority.From
the above formulafor D we have: qj > q
D =1/2 /10-3150 +30 _150 1/4
(6)
40 300 40 300~ (1 q) qT q >q
in agreementwith our interpretation.Note is removed (recall q = N/T). We now
that if tract 1 had only 160 (N1 + W1) claim that the quantityon the right hand
residentialunits, this would not be a prac- side of equation (6) is equal to D. To es-
tical way of achievingevenness. tablish this, we note that

This content downloaded from 129.108.202.79 on Wed, 04 Mar 2015 20:30:35 UTC
All use subject to JSTOR Terms and Conditions
 ANALYSIS OF SEGREGATION INDICES 635

k k Therefore,once we have computedD, the

l Ti (qi -q) E(Ni Tiq) proportion of the nonwhite, white, and
the total population that must exchange
N-Tq=O; places to achieveevenness are respectively
hence, the proportion removed can be writ- given by
ten as (1 - q)D, qD, 2q(1 - q). (9)
1 k
2(1 -q)qT i-1 Conclusion
We feel that the objectionsto D which
1 k have been voiced previouslyby others or
2(1 -q)qT i
>1Ni-TIqJ (7) have been raised by us are not minor in
since theireffecton the use of the index. Table 1
clearly demonstratesthat the effects of q
Ni -Tiq =Ni -(Ni + Wi) q and N are not "loose"but do in fact pro-
(1-q) Ni-qWi, duce a systematicdeviation.
the proportion removed is The corrections suggested here should
go far toward allowing for meaningful
I k Ni Wi intercitycomparisonsas well as providinga
2 i-1 qT (q)T more practicalinterpretationof the index.
The former is made possible throughthe
1 i~~
k JN~~~wJ
Ni _Wi|D (8)
(8 use of the standardscores while the latter
is aidedby the use of the exchangepropor-
tion formulas(9).
as we see by equation (3). Therefore, our
It mustbe pointedout that the qualifica-
interpretation of D as the proportion of the
tions and modifications we have intro-
minority removed without replacement is
duced have equal, if not greater,applica-
established.
bility in the use of D outside the confines
As stated earlier, what is usually desired
of residential segregation. Multivariate
is the proportion of the minority which
analyses have used the index for several
would have to be exchanged with the ma- measures of differentiationor have in-
jority to achieve evenness. Fortunately this cluded the proportionof nonwhitesas an
quantity has a simple expression in terms
independent or intervening variable (cf.
of D as we shall now show. Since we now
Jiobu and Marshall, 1971). Our findings
replace those moved, equation (4) be- stronglysuggestthat there is a high degree
comes of confoundingeffect when q and D ap-
q Ni-Ri pear along with intercity comparisonsin
T1 multivariatemodels.
and in place of (5) we have
MATHEMATICAL APPENDIX
Rt= Ni - qTi Ti (qi - q).
Thus the proportion removed throughout In computingAD and C2D,we assume
the city is: that the Ti residentsof the ith census unit
are a randomsample selected without re-
I
N >iq = Y. Ti (qi -q) placementfrom the total populationof size
qi >q q i> q T. Thus the probabilitydistributionof N1,
(1 - q)D. the numberof nonwhitesin the ith census
unit, is given by the hypergeometricdis-
Here the last equality follows from (7, 8). tribution,
Similarly, the proportion of the majority
that must change tracts to bring about Pi Prob(NI ni)
evenness is qD and the proportion of the JT8(I q)Tj
total population that must move is: ni TVT -n,
(1 - q) D * T + qD yW 2q (1 -q) D. n () T-n * (1)

This content downloaded from 129.108.202.79 on Wed, 04 Mar 2015 20:30:35 UTC
All use subject to JSTOR Terms and Conditions
636 AMERICAN SOCIOLOGICALREVIEW
Furthermore,if we define Turning to the computation of a2D, we
have, from (7)
sij = ni + n; (2)
the joint distributionof the nonwhitepop- 1-f k
a2D= { Var(jNi Aij)+
ulationsNi and Nj on the ith and jth census i 1
units is given by
E Cov (INi -|, INjI- /tjl)}. (12)
Pii Prob(Ni =ni, Nj =nj)- (3) i7j
T, 8 Tj t qTy(l q)T Now
ni sJ
s- ni J sij ATi + Tj -sij Varf jNj ,Aij) =E((Ni /Ai)2)
( T ) -E2(INi-,I) =ai2- el2 (13)
iJ Ti +
Tj)(
and
Since the mean and varianceof a hyper-
geometric distribution are well known Cov(jNi -JIiI, Nj -j I)
=E(INi-iLj NjN-/Lj,)
(Hays and Winkler, 1970), we have at -E(INi /j-11) E(INj-MkjI) (14)
once =eij - ejej.
,ui~_E[NJ = qTi (4)
Thus,
and
1 k
T-T.i ,2D (2
Gr +
ai2 E[(Ni ,Ai)'] Tiq( - q) j-{ _et2)
i=1
(5) V (ejj -eej)) (15)
From equations (7, 8) of the text we
have 1 k k )
-2 { >S2 e( i2 + eij}.
k i 1 i-1 i7&j
D-2q(1 11 i Ni-qT1l. (6)
2q1- q)T -1 Finally,using (11),
Thus from (4), 1 k
a2D= 2 a{ i2D)2.
1 k
D - Y. Ni -All, (7) (16)
Bi 1
Equations (11) and (16) of this ap-
where
pendix constitute theoretical expressions
B 2q(1 -q)T. (8) for ADand a2D. In practice,however,vari-
This is the form of D that we shall use for ous simplifications and approximations
computingtheoretical expressions for /AD would be employed to simplify the calcu-
and a2D. lations. We now briefly indicate some of
We definethe followingexpectations: the most obvious of these.
In general,since E[(Ni -M)] 0, equa-
Ti tion (9) can be replacedby
ei =E[Ni y- il] Y Ini -/ilPi (9)
ni =? INT (i)
eij = E[lNi K-/il INj - /%jl]
ei=2 V (Ai -ni) Pi, (17)
ni=0
Ti+Tj si
- -
Ini AiI(sij ni) -,jI Pij where INT(,Ai)is the greatestinteger less
sij=O nI=O than or equal to Mi.
(10)
We now consider some simplifications
Then, we have at once, arising from approximatingthe hypergeo-
k metric distributioneither by the binomial
I or by the normal distribution.We employ
ID E(D) Y, ei. (11)
i 1 the notation,

This content downloaded from 129.108.202.79 on Wed, 04 Mar 2015 20:30:35 UTC
All use subject to JSTOR Terms and Conditions
 ANALYSIS OF SEGREGATIONINDICES 637
Cowgill, Donald 0. and Mary S. Cowgill
B(n, q; x) (n )qx (1 q)n-x, (18) 1951 "An index of segregation based on
block statistics." American Sociological
for the binomial probabilities. In most Review 16:825-31.
demographicstudies,the census unit popu- Cramer, Harald
1946 Mathematical Methods of Statistics.
lations Ti are all much smaller than the Princeton: Princeton University Press.
total populationT. In this case, it is per- Duncan, 0. D. and Beverly Duncan
missibleto replace (1) by 1955 "A methodological analysis of segrega-
tion indices." American Sociological
Pi Prob(Ni Bni) B(Ti, q; ni). ( 19) Review 20:210-7.
Farley, Reynolds and Karl E. Taeuber
By a result derived by Cramer (1946), 1968 "Population trends and residential seg-
Bardwell (1971) and others, we have in regation since 1960." Science 159:953-
this approximation 6.
Hays, William L. and Robert L. Winkler
ei- 2 Si(1 -q) B(Ti, q; Si), Si 1+ 1970 Statitics, Vol. I. Holt-Rinehart-Win-
INTOAO), (20) ston.
Hornseth, Richard A.
and thus 1947 "A note on 'The measurement of eco-
logical segregation' by Julius Jahn, Cal-
1 k vin F. Schmid and Clarence Schrag."
,a- Y.ESiB (Ti, q; Si), Ti < < T. American Sociological Review 12:603-
qTi 4.
(21) Jahn, Julius A.
1950 "The measurement of segregation: der-
If furtherTi q( 1 - q) is large for all the ivation on an index based on the cri-
terion of reproducibility." American
T,, then the normal approximationto the Sociological Review 15:100-4.
binomialmay be employedin (21) and we Jahn, Julius A., Calvin F. Schmidt and Clarence
obtain Schrag
1947 "The measurement of ecological segre-
1 k gation." American Sociological Review
T\/27rq(1-q) i-1 12:293-303.
1948 "Rejoinder to Dr. Hornseth's note on
Ti<<T, 1<<Tiq(1-q). (22) 'The measurement of ecological segre-
gation.'" American Sociological Re-
Similarly,a topic of further investigation Review 13:216-7.
is the considerationof the analogoussim- Jiobu, Robert M. and Harvey H. Marshall, Jr.
plificationsof Pij and their consequences 1971 "Urban structure and the differentiation
for o2D. between blacks and whites." American
Sociological Review 36:638-49.
REFERENCES Leasure, J. William and David H. Stern
Bahr, Howard M. and Jack P. Gibbs 1968 "A note on housing segregation in-
1967 "Racial differentiation in American dexes." Social Forces 46:406-7.
metropolitan areas." Social Forces 45: Taeuber, Karl E. and Alma F. Taeuber
521-32. 1965 Negroes in Cities. Chicago: Aldine.
Bardwell, George Williams, Josephine J.
1971 Segregation-A Social Account. Den- 1948 "Another commentary on so-called seg-
ver: Colorado Civil Rights Commis- regation indices." American Sociologi-
sion. cal Review 13:298-303.

This content downloaded from 129.108.202.79 on Wed, 04 Mar 2015 20:30:35 UTC
All use subject to JSTOR Terms and Conditions