Research Notes and Comments

315

3 Baxter, M. J. Paractical Problems of Calibrating a Model of Informal Recreational Day Trips. . Final Report to Social Science Research Council, Tourism and Recreation Research Unit, Edinburgh University, 1978. 4. Baxter, M. J., and G. 0. Ewing. Calibration of Production Constrained Trip Distribution Models and the Effect of Intervening Opportunities. Journal of Regional Science, in press. 5 Canter, D. The Psychology of Place. London: Architectural Press, 1977. . 6 Cesario, F. J. A Generalised Trip Distribution Model. I w d of Regional Science, 1 (1973), . 3 233-47. 7 W e l d , B. S. The Nature of Recreational Travel Space. In Recteatiaal Ecmutnics and . Analysis, edited by G.A.C. Searle. London: Longman, 1975. 8 Edwards, S. L ,and S. J. Dennis. Long Distance Day Tripping in Great Britain. Iournal of . . Tmnsport Ecaomics and Policy, 10 (1978), 237-56. 9 McAUister, D. M., and F. R. Klett. A Modified Gravity Model of Regional Recreation Activity . k 21-34. with an Application to S i Trips. Journal of Leirure Resea7ch, 8 (1976), 1 . Openshaw, S , and C. J. Connolly. Empirically Derived Deterence Functions for Maximum 0 . 17) Performance Spatial Interaction Models. Enoironment and P h i n g A, 9 ( 9 7 , 1087-79. 1 . Sack, R. D. A Concept of Physical Space in Geography. Geogwphical Analysis, 5 (1973) 1 18-34. 1 . Sayer, R. A. A Critique of Urban Modelling. PlOgrRFs in Planning, 6 (1976). 2 . 13. Stetzer, F. Parameter Estimation for the Constrained Gravity Model: A Comparison of Six Methods. Enoinmment and Planning A, 8 (1978), 873-83. 14. Stetzer, F., and A. G. Phipps. Spatial Choice and Spatial Indifference: A Comment. Ge~gtt~~hical Analysis, 9 ( 9 7 ,400-403. 17) 1 . Wilson, A. G. Entropy in Urban and Regional M&Uing. London: Pion, 1970. 5

A Consistent Conceptual Definition of the Index of Dissimilan 9 3 by John F. Jakubs

1. Introduction Many methods of measuring segregation have been formulated and proposed [ l , 2 . Some can be dismissed because their dependence upon particular 1 characteristics of the point pattern, such as population proportions by race, hs makes meaningless any intercity comparisons. Fundamentally, t i implies that obtaining the same index value for two cities would not necessarily mean that the same degree of segregation is present. Of the methods that remain, the most popularly used is the index of dissimilarity. Originally formulated in 1947, it has been employed notably by the Taeubers in their 1965 and subsequent work [3, 6, 71. The index has some attractive properties such as its range, from zero (no segregation) to 100 (totally segregated), and the fact that it is dimensionless. It is plagued by some problems, however, not the least of which is the confusion that has arisen over
John F. Jakubs is assistant profmsm of geography and public and envimrnental affairs, Indiana University.

OOlS7363/79/0779-0315$00.50/0

0 1979 Ohio State University Press G ~ R A P H I C ANALYSIS, 11, no. 3 (July 1979) AL vol.

316

GeographicalAdysk

its conceptual meaning [4, 9, 101. This paper addresses the question of conceptual definition and, by identdjmg the assumptions behind two alternative definitions, distinguishes one definitional form as preferable, since it is consistent for all cases. Two benchmarks are used in conceptual definitions of the index of dissimilarity. One is the condition of zero-leuelsegregation and the other is the situation of maximal segregation. The former occurs when all areal units comprising the region have identical racial proportions, equal to the aggregate or regional racial proportions. The latter exists when each areal unit has members of only one racial group within its borders. In this case, the effort required to change the pattern into one of zero-level segregation is maximized and the index has a value of 100. One definitional form employs the case of zero-level segregation and the effort required to achieve it. This definition claims that the index measures the number of black (or white) people who would be required to move to achieve zero-level segregation expressed as a proportion of the total black (or white population in the region [5, 7,111. Thus, the index of dissimilarity is

D = M ~ / B , or

D=M~/w, (1) where B and W are the regional black and white populations, respectively, and M B and M W are the numbers of black and white people who would have to move to achieve zero-level segregation. Call this definitional form 1. A second definition holds that the index measures the number of black (or white) people who would have to move to achieve zero-level segregation expressed as a proportion of the number of black (or white) people who would have to move in the maximally segregated pattern to achieve zero-level segregation [3, 4, 91. In this case, we have
D = M B / * M B , or

D =M ~ / * M ~ , (2) where an asterisk indicates reference to the maximally segregated pattern. This second definitional form may be stated more generally. For any pattern, consider the effort that would be required to alter the distribution of population across areal units such that zero-level segregation would prevail. Similarly, consider the effort that would be necessary to change the corresponding maximally segregated pattern into one of zero-level segregation. This latter effort cannot be exceeded by that required for any other pattern. The index of dissimilarity is defined as the ratio of these two efforts, and utilizes as an indicator of effort the number of household relocations that would be required in necessary pattern alteration. Although this population movement is hypothetical, devised only to measure the level of segregation present in a pattern, and is not necessarily linked to any public policy, an academic debate has emerged as to the nature of such relocation [4, 101. In this paper, it is shown that under definitional form 2, the type of relocation does not affect the index. The interpretation of the index of

Research Notes and Comments

317

dissimilarity is unaffected, whether movement is considered to consist of: (1) locational exchanges between members of two groups, which necessitates no changes in the housing stock or, (2) relocation of one race or class only, either immigration or emigration, with concomitant housing construction or abandonment. It is a ratio of effort levels when effort is defined in terms of household movement requirements, or it is, as defined originally but less generally, -the sum of the differences between observed numbers of Negroes residing in the respective census tracts of a city and the numbers that would occur if there were no segregation, divided by the sum of the differences that would occur if there were complete segregation [3, p. 2941.
2. Framework

Consider a region subdivided into a set of mutually exclusive and exhaustive areal units. Let pi, B,, and W, indicate the total, black, and white populations, respectively, in areal unit i , and let P, B, and W denote, for the region, the total, black, and white populations, respectively. Divide the set of areal units into two subsets: those characterized by an overrepresentation of black people, or where

B,/P, > B / P ,

(3)

and those which are not. Indicate elements of the former by a subscript G and the latter by a subscript N. The index of dissimilarity, in ratio form, has been shown [ 4 ] to be

D = (PB, - B P , ) / B W .

(4)

Here the index has been divided by its scalar factor of 100. Its range is between zero and one. Movement to change a segregated pattern into one of zero segregation can be interregional, involving migration into and out of the area in question or intraregional, consisting only of population redistribution within the area under study. Below it is shown that, whereas definitional form 1 changes under these assumptions, definitional form 2 is stable. First consider the assumption of interregional population flows.

3. Interregional Migration
To desegregate a pattern in this manner, black movement into the region is necessary only to N-type zones and white in-migration is required only for G-type units. Likewise, out-migration aids in the process of desegregation only for blacks leaving G-type zones and whites leaving N-type units. For Gtype units, then, the condition of zero segregation requires that
B / P= ( B, -

w)/( pi - qB Mi ). +

(5)

Solution of the above for M f , and summation over the set of G-type zones

318

Geographical Analysis M ~ =, ( PB, - BP, - B M ~ , ) /W,

result in

(6)

where M B , and M W , represent the numbers of black out-migrants and white in-migrants, respectively, pertaining to the set of G-type zones. For N-type zones zero-level segregation implies that

B / P = ( B~ M ~ ~ pi /-(q w + + ) M:).
Again solving for obtain

(7)

MiB,

but now summing over the set of N-type zones, we

(8)

M B N = ( P B G - B P G - B M W N ) / W,

where M B N and M w N represent black in-migrants and white out-migrants, respectively, pertaining to the set of N-type zones. The total movement requirements by race are

M W = M W , + M w N and M B = MB,+ M B N .
Thus, from (6) and (B),

(9)

M B = [ 2(PB, - BP,) - BM
For white movement we have that

I/

W.

(10)

M W = [ 2(PBG - BP,) - WM] /B.

(11)

Expressions (10) and (11) are indeterminant, since various combinations of white and black levels of movement can result in zero-level segregation. Consider the relationship between black and white movement requirements, however. Let

M ~ = N M ~ ,

(12)

where N is nonnegative. This expression is general, because specification of the races is arbitrary and movements of members of either one race only or of both races are possible since N can be zero. Substitute for M B in (10) and solve for M W to obtain

MW=(2PB,-2BP,)/(WN+B).

(13)

For definitional form 1 to hold, it is apparent from ( 1 ) and (4)that the following relationship must be true:

MW=(PB,-BP,)/B.

(14)

Research Notes and Comments

319

Conditions (13) and (14) together are satisfied only when N = B / W , or when
M~ = ( B / w)M".

(15)

This definition is not generally true for the case of interregional migration. Consider definitional form 2 and the maximally segregated pattern. Recall that the index of dissimilarity, by this definition, is the ratio of movement requirements for an actual pattern and for a corresponding maximally segregated pattern. Thus, restrictions on the type of movement necessary to achieve a pattern of zero-level segregation, as in (12) above, must be applied similarly to both patterns, and (12) can be restated as MB=NMW, and

* M~ = N( * M ").

(16)

For interregional migration, recall (13) and apply it to the maximally segregated pattern. Further recall that in such a pattern, B,, Pc, and B are identical. Thus (13) becomes
* M W = 2 B W / ( WN+ B ) .

(17)

Divide (13) by (17) to obtain M "/( * M ") = (PB, - BPc)/ B W , (18)

and it is clear from (4) that definitional form 2 holds in general, when movement requirements are conceptualized as interregional in nature.
4. Intraregional Migration

Two types of intraregional population movement can change a segregated pattern into one of zero-level segregation: black movement from G- to N-type zones and white movement in the opposite direction. Thus, each move has two effects, and for this type of population redistribution
C MB=MB = M B N , and

M"=M",=M",.

(19)

Arguments made in section 3 demonstrate that the condition of zero-level segregation requires
M B c = (PB, - BP, - BM",)/
W,

320

Geographical Analysis

or for intraregional migration,

M~ = (PB, - BP, - B M ~ )W. /

(21)

This is indeterminant since either exclusively black, white, or any of various combinations of black and white movements can completely desegregate a pattern. As above, however, consider the relationship between black and white movement requirements. Again, let

M ~ = N M ~ .
Substitution for M B in (21) and rearrangement of terms leaves

(22)

= ( PB, - BP,)

/( WN+ B ).

(23)

Regardless of the type of migration considered, definitional form 1 requires that (14) hold. Conditions in (14) and (23) together require that N= B/P.

(24)

Once again the truth of definitional form 1 requires a restricted conception of the movement requirements to achieve zero-level segregation. Here, for intraregional migration, it is necessary that

M~ = (B / P )M w.

(25)

Any other relationship between white and black migration requirements would make this definition invalid. This is not the case, however, for the second definition. Again, as for interregional migration, the restriction in (22) requires, for a maximally segregated pattern, that

* M B= N( * M ").

(26)

For a maximally segregated pattern, where B G , P, and B are equivalent, (23) , becomes

* M ~ w= B / ( W N + B ) . /
Divide (23) by (27) and

(27)

M W / ( * M W )= (PBG - BP,)/BW.

(28)

The second definitional form places no restrictions on movement conceptualization when intraregional migration is considered. This analysis has examined two conceptual definitions of the index of dissimilarity. It has been shown that considering the index as a ratio of effort

Research Notes and Comments

/ 321

requirements for an actual settlement pattern and a corresponding maximally segregated arrangement requires no restriction upon the type of population movement that would be necessary to achieve desegregation. An alternative definition was examined and it was demonstrated that this definition lacks the generality of the ratio definition. Debate in the literature pertaining to the implications of specific values of the index of dissimilarity in terms of amount and form of migration necessary for desegregation is meaningless when one considers the ratio definition. Since there are no restrictions on movement, residential exchanges can be considered, as can interregional or intraregional movements, in varying proportions by race.
5. c n 1 soz o c u i ts

The measurement of segregation is not likely soon to diminish in importance. Commitment to economic and racial desegregation in housing and educational systems requires defensible standards against which to measure progress. Measurement procedures have received attention from the judicial system recently and this is likely to increase [ 8 ] .It behooves the academic community to offer straightforward and noncontroversial definitions of measures used in developing and enforcing public policies. Until now the index of dissimilarity, the most popular measure by use, has had a precise computational procedure [7: and a precise geometric interpretation [2], but two distinct conceptual interpretations. This paper clarifies the movement assumptions behind these interpretations and argues for the definition that is consistent for all such assumptions. This definition minimizes confusion in interpretation, confusion that might inhibit acceptance of use of the index in public policy analysis.
LITERATURE CITED

1. Cortese, C. F., R. F. Falk, and J. K. Cohen. Further Considerations on the Methodological 1 1 7 ) 630-37. Analysis of Segregation Indices. American Sodobgicul Review, 4 ( 9 6 , 2 Duncan, 0 D , B. Duncan. A MethadologicalAnalysis of SegregationIndexes. Ameticrm . . . and Sodologfcol Review, 20 (1955), 210-17. 3 Jahn,J.. C. F. Schmid and C. Schrag. The Measurement of Ecological Segregation.Amerfwn . Wbgicul Review, 1 (1947), 2 293-303. . 4 Jakubs,J. F Residential Segregation:The Taeuber Index Reconsidered.Journal of Regtonal . Science, 1 ( 9 7 ,281-83. 7 17) 5 schnare. A. B.Restdenrial Segregation by Race in US.MdropoNtan Area9: An Analysis A m . Citfap and Over Time. Washington: The Urban Institute, 1 7 . 97 6 Soremen, A., K. E. Taeuber, and C. J. Hollingsworth, Jr. Indexes of Racial Residential . !3egregation for 1 9 Cities i the United States, 1940-1970.Sodobgicd Focup ( 9 5 . 0 n 17) 7 Taeuber, K. E ,and A. F Taeuber. Negme.s in Citfap. Chicago: Aldine, 1 6 . . . . 95 8. White Suburban Schools Rap Segregation Claim. Dollap Tima-HemZd. November 18,1977. - _ 9. Winship, C. A Revaluation of Indexes of Residential Segregation. sodnl Fcme.~, 55 ( 9 7 , 17) 1058-86. 10. Zelder, R E. On the Measurement of Residential Segregation: Reply. lOuma2 of Regtonal Science, 1 ( 9 7 ,299-303. 7 17) 11. Flacial Segregation i Urban Housing M r e s Journal o Regional Science, 1 n akt. f 0 (90, 1 7 ) 93-108.