Constituency and th.' Analysic ed Ingh:]ati:e
Politics: A Study uf the United 01810: He>u::H of
Representatives in thu Eighiv--ol hih Cone 0:
JOHN LAWRENCE 1:CNEY
A DISSERTATION PI ESENTED TO THE GRADUATE COUNCIL OF
THE UNIVERSITY; of FLORIDA IN PARTIAL
FULFILLMENT OF THE REQUIREMENT FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY
UNIVERSITY OF FLORIDA
John LawLrence K~orey
ACKN OWLE DGME NTS
I would like to acknowledge my gratitude to Dr. Ruth McQuown,
Chairman of the Supervisory Committee, for her guidance and counsel
throughout my graduate training, particularly during the time of this
study' s preparation. Thanks are also due to the other members of the
committee: Drs. Richard L. Sutton, Nrorman C. Keig, Manning J. Dauei,
and Alfred B. Clubok.
Countless others also helped in various phases of th~e analysis.
Of these, two in particular deserve mention. Philip Bell' s generosity with
his time and programming skill made possible the cluster analysis pre-
sented in Chapter III, and Dr. Jesse Marquette provided crucial and timetly
assistance in the completion of the regression analyses set forth in
Chapters IV and V.
The data utilized in this study were made available in partially
proofed form by the Inter-University Consortium for Political Research.
Computer analysis was carried out at the University of Florida Computing
Center. Funds for this analysis were provided by the Department of
Political Science and the College of Arts and Sciences of the University
Obviously, none of the persons or institutions acknowledged
is to blame for any errors of commission or omission. They do not bear
any responsibility for either the analysis or interpretations presented
Finally, I would like to thank m~y wife, Mary, for her under-
standing and patience in the twvo long years during which the analysis
was undertaken and prepared. To her, this study is lovingly dedicated.
TABLE OF CONTENTS
ACKNOWLEDGMENTS . . . . . .
LIST OF TBLES .. .. . .. ..
ABSTRACT . . . . . . . . . . . . . .
CHAPTER (I. INTRODUCTION AND OVERVIEW ... ..
CHAPTER II. DIMENSIONS OF LEGISLATIVE CONSTITUENCIES
CHAPTER III. A NUMERICAL TAXON OM~Y OF LEGISLATIVE
CONSTITUENCIES . . . .
CHAnPTER IV. LEGISLATIVE INPUTS: CONSTITUENCY AND
THE SELECTION OF LEGISLATOR. .. ..
CHAPTER V. LEGISLATIVE OUTPUTS: CONSTITUENCY AN D
ROLL CALL BEHAVIOR ..............
CHAPTER VI. SUMMARY, CONCLUSIONS, AND SUGGESTIONS
FOR FUTURE RESEARCH .............
APPENDIX . .. . . .. . .
SELECTED BIBLIOGRAPHY . . . . .
BIOGRAPHICAL SKETCH .. . .. . . .
LIST OF TABLES
1. Correlations Among Constituency Attributes . ... 20
2. Rotated Factor Loading Matrix: Con sti tue ncy
Attributes ................... .... 21
3. Within Group Heterogoneity (as Per Cent of
Total Variance) for Final Fifteen Steps in
Grouping Procedure .................. 44
4. Constituency Attributes of Inductively Derived
District Types ... ........... .. 46
5. Cross -cla s sifi ca tion of Indu ctive Typology wi th
Threo "Traditional" Areal Classification Schemos,
by Column Percentages .. .. .. .. .. .. .. 51
6. Malapportionment Scores, by District Type .. .. 80
7. Representation Strength of District Types --
Actual Versus Proportionate .. .. .. .. .. .. 80
8. Correlations of Malapportionmnent Scores with
Constituency Attributes .. .... .. .. .. 81
9. Turnout Levels, by District Type .. .. .. . ... 87
10. Correlation of Turnout Scores with
Constituency Attributes .. .. .. .. .. ... 89
11. Regression of Turnout Scores on Constituency
Attributes ................... 92
12. Turnout Levels, by Section .. .. .. .. .. .. 97
13. Regression of Turnout Scores on Constituency
Attributes and Section ................. 99
14. Writhin-section Regression of Turnout Scores on
Constituency Attributes .. .. .. .. .. .. .. 104
15. Actual vs. Predicted Turnout, by District Type .. 108
16. Democratic Proportion of Two-party Vote, by
District Type ................... .. 114
17. Party Affiliation of District Types'
Representatives, by Column Percentages .. .. .. 114
18. Correlations cfc Democratic P~roportion of Two-party
Vote w~ith Constituency Attributes .. .. .. . 115
19. Regression of Democratic Proportion of Two-party
Vote on Constituency Attributes . .. .. ... .. 117
20. Democratic Proportion of Two-party Vote, by
Section .. ... .. .. . ... .. .. . 120
21. Party Affiliation of Representatives from the
South and th~e Non-South, by \-olumn Percentages .. 120
22. Regression of Democratic Proportion of Twc-party
Vote on Constituency Attributes and Section .. .. 121
23. V~ithin-section Regression of Democratic Proportion
of Two-party Vote on Constituenicy Attributes .. .. 124
24. Actual vs. Predicted Democratic Proportion of
Two-party Vote, by District Type . .. .. .. 127
25. Level of Inter-party Competition, by District
Type .. .. .. .. .. ... .. .. .. .. .. 129
26. Correlations of Inter-party Competition with
Constituency Attributes ... .. .. . .... 130
27. Regression of Inter-party Competition on
Constituency Attributes . ... .. .. . ... .. 131
28. Levels of Inter-party Competition, by Section . .. 132
29. Regression of Inter-party Competition on
Constituency Attributes and Section .. .. .. .. 135
30. Regression of Inter-party Competition on Constituency
Attributes, Section, and Party Preference ... .. 136
31. Regression of Inter-party Competition on
Constituency Attributes -- Non-South Only .. .. 138
32. Regression of Inter-party Competition on
Constituency Attributes and Party Preference --
Non-South Only .. . .... .. .. . ... 139
33. Actual vs. Predicted Inter-party Competition,
by District Type ................... 141
34. Rotated Factor L~oading Matrix: Roll Call
Variables... ................... .. 150
35. Correlation Coofficients Amone five Roll Call Indices 156
36. Roll Call Voting Indices, by District Type .. .. .. 157
37. Correlations of Roll Call Indices with
Constituency Attributes .. ... .. .. .. .. 160
38. Regression of Roll Call Indisco~ onl Constituency
Attributes .......... ............. 16
39. Roll Call Voting -Indices by~ Sec tion .. .. .. .. 168
40. Regression of R~ol) Call Indices on Constituency
Attributes and Section .. ... .. .. .. ... 170
41. Squared Partial Correlations of Roll Call Indices
with Constituency Attributes and Section .. .. .. 172
42. Roll Call Voting Indices by Party Affiliation
of Representatives ................... 175
43. Regression of Roll Call Indices on Constituency
Attributes, Section, and Party .. .. ... . ... 177
44. Squared Partial Correlations of Roll Call Indices
with Constituency Variables (Including Section)
and Party ................... .... 180
45. Writhin-Party Regression of Roll Call Indices on
Constituency Attributes and Section .. .. .. .. 183
46. Correlations Between Roll Call Indices and
Inter-party Competition . ... ... .. .. .. 188
47. Regression of Roll Call Indices on Constituency
Attributes, Section, Party, and Inter-party
Competition ................... .. 190
48. Wi~ithin-party Correlations Between Roll Call
Indices and Inter-party Competition .. .. .. .. 193
49. Wriithin-party Regression of Roll Call Indices
on Constituency Attributes, Section and
Inter-party Competition .. .. ... .. .. .. 194
50. Safe Seats vs. Competition Seats -- Regression of
Roll Call Indices on Constituency Attributes, Section,
;cnd Party ................,.......~ 198
51. Actual vs. Predicted Scores on Roll Call Indices,
by District Type ................... 202
Abstract of Dissertation Presented to the
Graduate Council of the University of Florida in Partial Fulfillment
of the Requirements for the Degree of Doctor of Philosophy
CONSTITUENCY AND THE ANALYSIS OF LEGISLATIVE
P'OLITICS: A STUDY OF THE UNITED STATES HOUSE OF
REPRESEN\TATIVES IN\ THE EIGHTY-EIGHTH CONGRESS
John Lawrence Korey
Chairman: Dr. O. Ruth McQuown
Maj or Departmnent: Political Science
This study deals with certain conceptual and methodological
problems involved in the specification of constituency environment as a
variable in the analysis of legislative politics. The primary goals of thec
study were to develop measures of constituency that would combine a
high degree of both comprehensiveness and parsimony, and to apply these
measures in an examination of selected substantive problems in legisla-
The research began wyith selection of twenty-five social,
economic, and demographic attributes for each of the 435 districts of the
United States House of Representatives in the Eighty-eighth Congress.
Six: of these attributes were discovered to be conceptually and empirically
redundant wvith other included characteristics and were eliminated from
further study. The remaining nineteen attributes were subjected to a
factor analysis which yielded six dimensions accounting for approximately
four-fifths of the total variance in the variables analyzed.
Employing these same nineteen attributes, an holistic typology
of district: environments was derived inductively, thus offering an alter-
native to traditional deductive typologies based either on urban-rural and
related distinctions, or on classifications based on geographic location.
Thle procedure employed was a modification of the error sum of squares
hierarchical clustering algorithm, which attempts to group objects in such
a way as to maximize group homogeneity. The procedure succeeded in
clustering all 435 districts into a relatively homogeneous five-fold typology.
District types, attribute dimensions, and individual attributes
were then all employed in an examination of some of the substantive
questions concerning constituency and legislative politics which have
long been foci of controversy among legislative analysts. In the first
phase of this portion of the research, selected legislative input variables --
mialapportionment, voter turnout, voter party preference, and inter-party
competition -- were studied. Generally modest associations were found
between constituency types and attributes and malapportionment. Much
closer relationships were discovered between constituency and both turn-
out and party preference. In all, about four-fifths of the variance among
districts in voter turnout in the 1962 congressional elections, and about
three-fifths of the variance in patterns of party preference were shown to
be explainable in terms of the constituency variables included in the study.
Moderately high relationships were also found between constituency and
levels of inter-party competition. Closer ex~amination,however, revealed
that much of this relationship could be explained as a function of the
association between constituency and party preference, on the one hand,
and,on the other, the correlation in the 1962 congressional elections be-
tween competition and Republican party preference. It wa s speculated
that much of the inconsistency in the findings of previous research on
this subject might be explained as a failure to properly distinguish between
party preference and party competition.
The study then shifted to an analysis of constituency influence
on congressional roll call behavior. Five roll call indices were selected
as dependent variables. Twfo were derived from a factor analysis of roll
calls taken during the first session of the Eighty-eighth Congress, and
were designaited: as Partisanship and Ideology. The three re ma inicq indices
were Domes'.ic Prosidential Support, Foreign Presidential Support, and
support for e Larger Federal Role. Only moderate associations \rtPc found
between constituency variables and roll call behavior. Party afm.Kion,
by contrast, was found to be a variable of extremely high explanet'y
power, even when constituency variables were held constant, :t~oral
marginality, on the other hand, was demonstrated to be of litti ta~rtance
in explaining congressional voting behavior, particularly whe*West and
constituency variables were controlled.
INTRODUCTION~ AND OVERVIEW
Analysts of American legislative bodies increasingly have
come to adopt what the writers of one leading textbook refer to as a more
" sociological" perspective.1 In other words, there has been a gr-owing
realization that institutional action cannot be adequately understood apar-t
from an institution's environment, i.e., the larger social system which
provides the context in which action takes place. One element of the
legislative environment which has been of special interest, on, both normna-
tive and scientific grounds, to students of democratic theory has of course
been the district of the individual representative. Numerous published
studies have sought to examine the relationship between what has been
called the "ecological" base of legislative districts -- the social, econom-
ic, and demographic variables that make up the constituency environment --
and, on the other hand, the various aspects of legislative organization and
In spite of this extensive inquiry, many very basic questions
concerning the relationship of constituency to legislative politics remain
u nre solved There are undoubtedly many reasons for this failing, but
perhaps the most basic has been the inadequate manner in which constitu-
ency has been measured. Political scientists and others have given a
great deal of attention to measurement problems in dealing wyith roll calls,
and some of the approaches they have developed will be employed in this
work. With respect to constituencies, however, measurement often remains
at an extremely primitive level. In large part, this is a reflection of
similar shortcomings which have in the past characterized all areas of
social science dealing with population aggregates. This is becoming in-
creasingly loss true, however, since in recent: years social scientists,
including political scientists, have given this problem increasing attention.
While much still remains to be done, significant progress has been made.
Unfortunately students of legislative behaviior have only begun to incorporate
these advances in analyzing representatives' constituencies. Development
of sound theoretical generalizations about the impact of constituency on
legislative politics must await more adequate specification of constituency
as a variable. A brief overview of the problem and of the strategy chosen
to encounter it will be presented in this chapter.3
Investigation of the constituency environment of legislative
politics involves one or both of two approaches. The first, which will be
called the attribute approach, seeks to determine associations between
given aspects of the legislative process and one or more social, economic,
or demographic characteristics (or attributes) of legislative districts such
as population density, racial heterogeneity, median income, and so on.
The second approach will be referred to here as the areal approach, since
those using it do not seek to describe directly specific individual or
collective traits of a district's inhabitants but rather to operate at a higher
level of abstraction by characterizing the area in which a district is located.
This second approach canl in turn be subdivided into two categories. The
first is based on a distinction between urban and. rural areas, or on some
modification of this distinction. T'he second depends upon one or another
form of subdivision into geographical sections of the population of legisla-
tive districts under study.
The areal and attribute approaches are, of course, not radically
different, s:'nce it is usually assumed that areal distinctions in fact tap
di strict environments which differ in their specific chara cteristics Still,
it is useful to begin discussion by considering the two approaches separately.
In the first place, the relationship between the two is more often than not
very unclear and it will be in part the task of the present study to try to
dispel the resulting confusion. Secondly, the two approaches are best
used to fulfill complementary functions. Because it retains greater detail
of information, the attribute approach offers a more powerful strategy for
detailed exploration of relationships among variables. The areal approach,
however, can be advantageous in that it makes it possible to view con-
stituency environment holistically. It thus potentially has great utility
for descriptive purposes, for it permits summarization of a great deal of
information in succinct fa shion In fact, many of the most substantively
interesting questions asked about the legislative process are those taking
an holistic view of the legislative district through use of areal categori-
zations. Thus one finds researchers investigating such questions as:
Do metropolitan and non-metropolitan districts tend to form opposing voting
blocs within legislative bodies?4 Whly do Southern legislators receive
more than a proportional share of leadership positions within Congress? 5
Were suburban districts more severely underrepresented than urban districts
prior to state actions made necessary in the past decade by S'upreme Court
decisions on malapportionment?6
The two approaches further complement each other because con-
clusions from one will suggest inferences concerning the other. For example,
finding that given district attributes correlate with "conservative" roll call
behavior may lead an analyst to generalizations about liberal and conserva-
tive types of areas.7 Conversely, a finding that legislators from districts
located in metropolitan areas vote more "I'berally" than their rural col-
leagues may- lead a researcher to inquire as to why such differences occur
and to suggest explanations in terms of the differential attributes of the
two kinds of areas.8 While the strategy cilosen will of course depend upon
the wYay in which the investigator formulates his questions, the above com-
ments suggest that the most useful results will be achieved when the two
approaches are employed in conjunction.
Unfortunately serious shortcomings exist in the manner in which
both the attribute and the areal approaches have been utilized. These
shortcomings are less methodological than conceptual. During the past
two decades, great improvements have been made in availability of data
and in techniques of data manipulation. These advances in information
and in methodology have far outrun the ability of legislative analysts to
develop a framework capable of accommodating them successfully. If this
limitation can be overcome, considerable progress can be achieved, using
already acquired data and already existing techniques, in meeting the
purely technical problems involved in studying constituency. It will,
therefore, not be the purpose of this study to introduce any methodological
innovations, but instead to examine alternative ways of specifying con-
stituency and to explore the extent to which already familiar procedures
can aid in this process.
Most previous research relating, district attributes to legislative
politics has included only a handful of constituency variables, usually with
little or no justification for their selection. In recent years some studies
have employed comprehensive data sets, t-ut these pioneering efforts have
barely begun to scratch the surface of neniced analysis. Reorientation of
research in this direction, moreover, pose:; new problems. As greater
numbers of attributes are included in a study, there is a tendency for
analysis to become unwieldy and results difficult to interpret, pointing
to a need for some systematic effort at data reduction in this area. Un-
doubtedly there is no ideal solution to this dilemma, but the possibility
of attaining a more nearly optimal tradeoff between parsimony and com-
prehensiveness requires exploration. In Chapter II this problem will be
taken up in greater detail, and an attempt will be made, through use of
factor analysis of a set of nineteen social, economic, and demographic
attributes, to describe the basic dimensions of the constituency environ-
ment of one important legislative body, the United States House of
Studies employing the areal approach have suffered from the
fact that areal distinctions, in legislative research and elsewhere, have
usually been derived in a rather ad hoc fashion, and their social and
political meanings are frequently quite vague. They often appear to have
been chosen largely on grounds of convenience and ready availability and
have advanced little beyond the common sense level. If this level is to be
transcended, more rigorous means must be set forth for developing and
evaluating district classifications than have been employed to date. The
search for such means will constitute the :ocus of Ch~apter III. There it
wrill be shown that most previous aircal typologies are linked to attribute
studies by the assumption that eachi type rejpresents a shorthand means of
tapping a cluster of districts with similar attributeses, but that this assump-
tion is rarelyl tested or even stated in test; ble form. Proceeding induc-
tively where previous research has been la -gely deductive, and employing
specific evaluative criteria, a new typolog J of legislative districts will
be developed from the same data base used in Chapter II.
The remainder of the analysis will involve application of the
measures set forth in Chapters II and III toward examination of some of
the substantive problems which have long confronted past researchers.
Chapter IV will examine the relationship between constituency and an
aspect of the input side of the legislative process, the electoral arena
through which representatives are recruited to Congress. Four specific
dependent variables -- malapportionment, voter turnout, partisan prefer-
ences of the electorate, and inter-party competition -- will be studied.
Each of these variables will be analyzed at three levels: with respect to
district types, with respect to constituency dimensions, and with respect
to individual constituency characteristics. Chapter V will present the out-
put side of the legislative system, examining constituency influence on
the roll call behavior of representatives in Congress. Analysis will proceed
along the same general lines as in Chapter IV. Finally, in Chapter VI, the
findings of the study will be reviewed, remaining problems will be dis-
cussed, and suggestions for future research set forth.
Constituency and electoral dati( for the study will be taken from
statistics available for the districts of the Eighity-eighth Congress, which
was elected in 1962. Roll call data will be taken from the first session
(1963) of the Congress. The Eighty-eigihth Congress was chosen on
pragmatic grounds. For Congresses prior 'o the Eighty-eighth, extensive
census data at the congressional district Javel are much less readily
available. Selection of a later Congress, on the other hand, would have
unduly lengthened the gap between the time when the constituency data
were gathered (1960) and the occurrence of the political behavior studied,
thus contributing to measurement error.11
The present inquiry must clearly be classified as a case study,
with all the limitations that this implies. Only very speculative general-
izations to other legislative systems from the findings set forth here are
possible. The Eighty-eighth Congress, in fact, is probably quite atypical
not only of American legislative bodies in general, but even of post-Wrorld
Wrar II congresses. Poised between the Eisenhower and Vietnam eras in
American politics, the Eighty-eighth Congress is not easily classified in
either category. The districts of this Congress will, therefore, be treated
as a population rather than as a sample, and statistical techniques appro-
priate to samples, such as significance tests and confidence intervals, will
not be employed.
Probably the most extensive study of constituency influence in
the Eighty-eighth Congress to appear to date is a doctoral dissertation
written several years ago by ~Jck: Roland Vanderslik.12 The present en-
deavor especiallyy, but not exclusively, .he portion presented in Chapter V)
traverses from a somewhat different perspective much of the same substan-
tive ground covered by Vanderslik's earlier wYork. This writer's intellectual
debt to himi is great and will become increasingly apparent as the analysis
NOTES FOR CHAPTER I
Malcolm E. Jewell and Samuel C. Patterson, The Legislative
Process in the United States (NewT York: Random House, 1966), p. 3.
Heinz Eulau, "The Ecological Basis of Party Systems: The
Case of Ohio, Midwest Journal of Political Science, I (August, 1957),
In the present chapter, only a brief outline of the main topics
to be covered will be given. More detailed reviews and citations of rele-
vant literature will be set forth at appropriate points in subsequent chapters.
David R. Derge, "Metropolitan and Outstate Alignments in
Il:linois alid M;issouri Legislative Delegations, Ameia Political Scie.ee
Review, Li1 (December, 1958), 1051-1065; David R. Derge, Urba n-
Rural Conflict: The Case in Illinois, in John C.Wahlke and Heinz Eu:u,
eds., Leyiisative Behavior: Ai Reader in Theory and Research (Glencoe,
Ill. : Free ]Press of Glencoe, 1959), pp. 218-227.
Raymond E. Wolfinger and Joan Heifetz, Safe Seats,
Seniority, and Power in Congress, American Political Science Review,
LIX (June, 1965), 337-349.
Andrew Ha cker, Congres sional Di stricting: The Issue of
Equal Representation (Washington, D. C.: The Brookings Institution,
1963), pp. 80-87; Congressional Quarterly Weekly Report, XX
(February 2, 1962), 153-154, and XXII (August 21, 1964), 1786.
Jack Roland Vanderslik, "Constituencies and Roll Call
Voting: An Analysis of the House of Representatives for the 88th Congress"
(Ph.D. dissertation, Michigan State University, 1967), pp. 136, 188.
Bruce L. R. Smith, "Isolationist Voting in the United
States House of Representatives, in Carl J. Friedrich and Seymour E.
Harris, eds. Public Policyv, Vol. XII (Camnbridge: Harvard University
Press, 1963), pp. 351-353; Murray Clark Havens, "Metropolitan Areas
and Congress: foreign Policy and National Security, Journal o~f
Politics, XXVI (N~ovember, 1964), 769-774.
These datai are published in U. S. Bureau of the Census,
Conoressionall District Data Book: Districts o~f the ~ 8th Congress A
Statistical Abstr Supplement (W2ashington, D. C.: U. S. Government
Printing Office, J963).
The roll call data used in this study may be found in the
Congressional QAroly lmna 1963 (W~ashington, D. C.: Congressional
QuarterlySorvice,19G3), pp. 594-655.
The prGSent study was undertaken too early to employ the
results of the 1970 Census.
Va ndGTSlik Con~situencie s an:d Roll Call V/oti ng ." Part
of this study has been published in Jack Ro~land Vanderslik, "C~onstituncy
Characteristics and Roll Call Voting on Negro Rights in the 88th Congress, "
Social Science Quarterly, XLIX (December, 1968), 720-731.
DIMENSIONS OF LEGISIATIVE- CONSTITUENCIES
In his review of developments in analysis of congressional
politics since World War II, Robert L. Peabody writes:
Over the last two decades and especially since 1960 a
proliferation of empirical studies has yielded much newi~
data, richer insights, and a better understanding of the
internal workings of Congress, executive-legislative
relations, and the representative process. Despite these
significant advances, political scientists have not yet
produced a conceptually clear and comprehensive theory
of congressional behavior .1
Nowhere is this more apparent than in the literature on the effects of coi.-
stituency environment on the legislative system, yet: no area is more
crucial to both normative and empirical theory concerning rule-making in
a representative democracy. In this chapter, an attempt will be made to
outline some of the obstacles to the development of theory encountered in
efforts relating constituency attributes to legislative politics, and to
suggest means for overcoming these obstacles.
In a highly developed science, k~ey independent variables can
often be deduced from well-developed axiomatic theories, and extraneous
variance can be controlled at least reasonably well. Without the exist-
ence of highly developed theory to guide research efforts, legislative
analysts are largely left to rely upon their own creativity and common
sense in deriving categories and selecting indicators by which to measure
con stitue ncy chara cteri sti cs Conversely, the lack of adequate means
of measuring constituency characteristics is one very fundamental reason
for the slow growYth of theory noted by Peabody.
In practice, this problem has-usually been resolved by ad hoc
selection of, at most, a handful of variables. Typically little or no
justification is presented for the selection of the particular variables
chosen. The situation is similar to that described by Haddlen and Borga to
in their study of American cities:
Traditionally human ecologists have examined the re-
lationship of a few independent variables to a single
dependent variable and having exhausted the inter-
relationships of those few variables have proceeded
in subhsequent analysis to add an additional variable
or two. Sometimes this situation has reached pro-
portions where a variable 1~ associated with an in-
vestigator as his variable.
At one time this situation might have been blamed on inadequate sources
of data, or means for processing such data, particularly for representative
bodies as large as the United States House of Representatives. Today this
is no longer true due to the availability of greater amounts of data,4 to the
development and dissemination of increa singly sophisticated statistical
techniques and data processing equipment, and to experience gained in
research areas with problems structurally and conceptually similar to those
faced by analysts of constituencies.
In recent years there has been a growing realization of both the
need for and the feasibility of a more comprehensive approach to the study
of constituency attributes, and several works have appeared which have
employed a large number of measures of characteristics in analysis of
legislative politics. One purpose of the present analysis will be to con-
tinue in the direction in which these studies point, for much more work is
needed in order to acquire the knowledge necessary for the development of
legislative theory. In addition, an effort will be made to explore a new
problem created by the inclusion in research designs of large numbers of
Thie difficulty is that if there does not exist an adequate theoreti-
cal basis for selecting measures of constituency before analysis, the same
lack of highly developed theory makes interpretation of results of analysis
very difficult. By itself, addition of more variables can serve only to
create greater confusion when the researd~.er is forced to order findings in-
volving many variables all treated individually. There is thus a danger
that even the most rigorous and methodolocgically sophisticated research
can become impressionistic at the point at which the investigator comes to
draw conclusions from his study.
Concluding an analysis of the Eighty-eighth Congress in which
he relates each of twenty-one constituency variables to roll call voting,
There is another approach to the data I have which
probably should be used. That is factor analysis.
Because I wanted to be able to detect and distinguish
relationships between particular independent variables
and particular roll call dimensions, I chose the
methods which have been employed. The patterns in
both constituency data and the roll call voting dimensions
indicate strong intercorrelation. Factor analysis
is probably a better technique_ for assembling the
related elements of the data.
Th~is suggestion has much to recommend it, for it offers a way
of reducing analysis to manageable proportions, not by arbitrary a priori
selection of variables or by attempts to interpret quantities of output after
the analysis is completed, but rather by introducing a powYerful and systemn-
atic data reduction technique into the analysis itself As Hadden and
Borgatta note in proposing factor analysis as a technique for analyzing
aggregate data in urban studies,
Modern resources, including high speed electronic
computers, make possible a less personalistic and
more systematic analysis. It is now possible to
begin investigations by ordering the relationships
among a large number of variables. .. .
Casentially, factor analysis is iln attempt to represent the vari-
ance contained in a relatively large set of neasurements in terms of a
smaller number of dimensions or "factors, while minimizing loss of detail
of information. Thus it offers a promising approach to attainment of an
optimal tradeoff between comprehensiveness and parsimony. It is a tech-
nique already quite familiar to most political scientists, and especially
to legislative investigators, who have used it extensively as a tool in the
management of roll call data.8 Factor analysis has been applied, by
political scientists and others, in numerous studies of almost every con-
ceivable kind of population aggregate, yet, except for some work that has
been done in analyzing behavior in the United Nations General Assembly,
little seems to have been done to apply factor analyses specifically to
attributes of legislative constituencies.10
At the same time it must be noted that factor analysis possesses
several limitations which suggest that it be applied cautiously in studying
legislative constituencies. In the first place, factor analysis does involve
some loss of information concerning the variance in the measures input to
the analysis, and it is conceivable that this "unique" variance may be of
great importance in explaining the behavior of the legislative system.
Secondly, dealing with a small number of relatively general factors can
cause the researcher to overlook important nuances that might be uncovered
through analysis of specific characteristics. Finally, and most importantly,
there is a danger that studying constituency attributes in terms of "under-
lying dimensions" might cause analysis to become too abstract, to separate
both the researcher and his readers too fai from the data, and to lead to
highly misleading conclusions.
The strategy that will be followed, therefore, in subsequent
chapters will be to use, as measures of constituency environment, both
specific attributes and factors derived from them.11 There are two reasons
for doing this. The first is that the advantages of one approach are the dis-
advantages of the other, hence employing both should provide better results
than would either employed separately. In this way there is a better chance
of staying close to the data while at the same time having a tool with which
the data can be ordered. Secondly, retention of specific attributes in the
analysis will provide a means for testing the efficacy of factor analysis in
studying legislative constituency. If it can be shown that a small number
of factors can provide the investigator with as much or nearly as much
information as a large number of individual variables, then clearly the
former would be preferable on grounds of parsimony. The remainder of the
present chapter wvill then be devoted to selection and factor analysis of
the constituency attributes to be used in the remainder of the study.
From each of the 435 House districts of the Eighty-eighth
Congress, nineteen social, economic, and demographic attributes were
chosen. Originally, twenty-five variables had been selected, but six
were deleted on grounds of redundancy since they woro closely related to
other includedj variables both empirically (as indicated by Pearson's
correlation c,,efficient of +.7 or more) and conceptually. Data were
based on the, results of the 1960 Census. The variables selected, along
with abbreviated names that will be employed subsequently in tables, are
1. Population growYth, 1950 to 1960 (Pop. Gro.)
2. Population per square mile (Density)
3. Bla ck population/total population (Bla ck)
4. Foreign stock: persons foreign born or of foreign or mixed
parentage/total population (For. Stk.)
5. Persons age sixty-five and over/total population (65+)
6. Persons under eighteen years of age/total population (-18)
7. Geographic mobility: persons five years old and older with
r-esidence in 1960 different from that in 1955/total population (Mobil.)
8. Elementary school students in private school/all elementary
school students (Priv. Ed.)
9. Technical illiteracy: population twenty-five years of age
and older with less than five years of schooling/total population twenty-
five years of age and older (Illit.)
10. Persons twenty-five years of age and older with four or more
years of college/total population twenty-five years of age and older
11. Families with annual income less than ,$2,000/all families
12. Families with annual income $15,000 or more/all families
13. Gini index of economic inequality (Ineq.)
14i. Unemployed civilian labor force/total civilian labor force
15. Persons employed in agriculture/total employed civilian
labor force (Agric.)
16. Persons employed in manufacturing/total employed! civilian
labor force (Manuf.)
17. White collar workers/total employed civilian labor force
18. Owner occupied housing units/all housing units (Own.
19. Overcrowding: housing units with more than one person pc r'
room/all occupied housing units (Crowd.)
In selecting variables this writer has attempted, within the
limits of availability of data, to be fairly catholic without being indis-
criminate. No claim is made that the list is in any way definitive. As
Hofferbert comments in presenting a list of variables for a factor analysis
of American states,
One could complain that this list...contains no readily
apparent theoretical rationale for the particular variables
employed. I suppose that I could construct some post
hoc rationale that might be moderately adequate as a
beginning theory. And, of course, the list could have
been extended by using a variety of subdivisions of some
of the items, .. .. But the test of this particular list either
in terms of its cohesiveness or its inclusiveness, mu st
be the same as for any alternative. Namely, howY much
promise does it hold for relating in a theoretically inter-
esting manner to the dependent variables we seek to
A few specific points about the list should be mentioned. Some
analysts have included geographic section as a variable in factor analysis
of population aggregates.14 In this study this was not done because, as
will be explained in m~ore detail in the next chapter, one goal of the
present work is to separate and compare the effects of sectional culture,
on the one hand, and social, economic, and demographic variables on the
Inclusion of the variable "per cent of elementary school students
onrolled inl private sch;ools" requires some explanation. Unfortunately, for-
present purposes, census data include no information about religious
affiliation. Since most non-public elementary schools in this country are
Roman Catholic, this variable provides a measure, admittedly crude, of
Though two recent articles by Dyel6 are notable exceptions,
economic inequality has been largely neglected as an independent variable
in political research. The index of inequality employed in the present
study is a trapezoidal approximation to the Gini coefficient of concentra-
tion and is based on the seven categories of family income reported in
the 19 60 Census which are available at the congressional district level.
In estimating aggregate income, and in computing the index, the average
within-category income for each of the categories, except the open-ended
one, was assumed to be the midpoint of the category. For the open-ended
category ($15,000 and over), the average was estimated for each district
by fitting Pareto curves to the data.
Table 1 presents a matrix of Pearson's correlation coefficients
among all nineteen variables. From this matrix, six factors were extracted.18
Communalities were derived iteratively with squared multiple correlation
coefficients used as initial estimates. The extracted factors were rotated
to an orthogonal varimax solution. The rotated factor loading matrix is
presented in Table 2. The factor 10adings show the degree to which a given
variable is associated with a given factor, and when these figures are
squared they represent the proportion of variance in a variable that is "ex-
plained" by the factor. The communalities in the right hand column of tl e
table represent the amount of variance in each variable explained by all
six factors together. Eigenvalues, shown at the bottom of each column o~f
factor loadings, constitute the amount of variance overall which each factor
explains, and can be computed as the sum of squared factor loadings for
each factor. Table 2 also shows the cumulative per cent of common and of
total variance explained by each successive factor.
Factor one has an eigenvalue of 3.47, and accounts for 18 per
cent of the total variance in the unreduced correlation matrix, and 23 per
cent of the variance common to the six extracted factors. The factor is
most closely associated with variables related to poverty: severe inequal-
ity, and high levels of technical illiteracy and low income families. On
LD 10 N
m LD LD N LD
O LD L O H L
m Oc c -
' US rJod
V0 co a
I I I I I I I~r~U
cO NLJm~ o CH
I I I ImlVm0
NN wD er't
~On <0 o <0O
I I Ir~~nu
II II I illi~
H LDN 0 3
H CO M YN-U33P~'
m c W LD 0 (N LD
pW N (0 10-
- I I-t~
mL LD t~rcu
II I ~~rO~-
I I I
ROTATED FACTOR LOADING MATRIX:
I II III IV V VI h2
Depri- Afflu- Home Indust- SFC Popu-
vation ence Ow~ner- rializa- lation
ship tion Dyna-
Ineq. .84 -.01 .07 -.48 .12 -.02 .96
LowY Inc. .74 -.40 .17 -.43 -.01 -.11 .94
Illit. 73 -.31 -.17 -.17 -.32 -.12 .81
Black .67 -.20 -.26 11 -.15 .09 .60
Crowd. .56 -.35 -.11 -.35 -.55 .12 .90
Coll. Ed. -.09 .92 -.05 .01 .02 .22 .90
BHigh Inc. -.15 .88 -.17 .14 .02 .02 .84
W~ht. Col. -.35 .83 -.21 .11 .06 .19 .90
Unempl. .13 -.43 -.20 .04 .00 .05 .25
Own. Occ. -.25 .07 .91 .01 -.08 -.08 .92
Density .02 .02 -.78 .10 .11 -.07 GB
For. Stk. -.53 .30 -.56 .15 .07 -.21 .7i
Priv. Ed. -.43 .27 -.44 .29 .09 -.34 .60
Manuf. -.21 -.09 .02 .76 -.03 -.27 .71
Agric. .21 -.30 .34 -.73 .01 -.16 .80
65+ -.10 -.12 -.09 -.19 .85 -.21 .83
-18 .15 -.20 .56 -.18 -.75 .03 .97
Mobil. .06 18 .01 -.13 -.16 .93 .94
Pop. Giro. -.29 .37 .24 .09 -.32 .45 .59
value 3.47 3.40 2.71 1.93 1.90 1.52
per cent of
iance 23 46 64 77 90 100
per cent of
iance 18 36 50 61 71 79
the other hand, variables related to high socioeconomic status, such as
college education, white collar employment, and high family income, have
only slight loadings on this factor. Black population has a substantial
loading on factor one, and to a lesser extent the same is true of overcrowd-
ed housing conditions. The factor is negatively related to ethnicity and
to Catholicism (as measured by the proxy variable of private elementary
edu ca tion) Factor one will be designated as "Deprivation. "
Actor twvo, with an eigenvalue of 3.40 also accounts for 18 and
23 per cent of total and common variance respectively. This factor is
clearly defined by the high 10adings of proportion of adults with college
educations, proportion of families with annual incomes of $1.5,000 and
over, and Iproportion of white collar workers. The relatively low 10adings
of such varied bles as unemployment, technical illiteracy, and family in-
comes below $2, 000 per annum show that the factor is much more clearly
associated wiith cultural enrichment than with deprivation. Factor two will
be interpreted as "Affluence."
Factor three has an eigenvalue of 2.71, accounting for 14 per
cent of total variance and 18 per cent of common variance. Owner occupancy
of housing is easily the variable most closely associated with this factor,
which will therefore be designated as "Home Ownership. Not surprisingly,
population density has a high negative loading on this factor. To a lesser
degree, factor three is negatively associated with ethnicity and with Cathol-
icism, and positively with proportion of population under eighteen years of
Factor four has an eigenvalue of 1.94, and accounts for 10 per
cent of total, and 13 per cent of common variance. It is defined primarily
by a positive loading of manufacturing, and a negative loading of agricul-
ture. Economic inequality and low family income have moderate negative
loadings on the factor which will be designated as Indu strialization "
Factor five, with an eigenvalue of 1.90, and accounting for 10
and 13 per cent of total and common variance respectively, is associated
positively with proportion of population sixty-five years of age and over
and negatively with proportion of population under age eighteen. To a
lesser degree, it is negatively related to overcrowding. Factor five will
be identified as "Stage in Family Cycle (SFC). "
Factor six has an eigenvalue of 1.52, accounting for 8 per cent
of total variance and 10 per cent of common variance. It is primarily de -
fined by geographic mobility, and, to a much lesser extent, is associate d
with population growth. Factor six will be interpreted as "Population
As Table 2 indicates, the six factors together explain 79 per
cent of the variance contained in the nineteen measures analyzed. Thus
the goal of achieving a high degree both of comprehensiveness and parsi-
mony seems to have been met fairly well.
The final step in the analysis involves computation of factor
scores for each of the 435 districts. Factor scores enable the researcher
to locate each case (in this instance, district) on each dimension.19
These scores, not presented here, will be made use of in subsequent
NOTES FOR CHAPTER ~II
Robert L. Peabody, "Research on Congress: A Coming of
Age, in Ralph K. Huitt and Robert L. Peabody, eds., Congre ss: Two
Decades o~f Analss (New York: Harper &; Row, 1969), p. 3.
Almost any number of such studies could be cited. See, for
example, Charles F. Andrain, "A Scale Analysis of Senators' Attitudes
Toward Civil Rights, We~stern Politi cal Q~ularterly XVII (September,
1964), 488-503; Robert W. Becker et al., "Correlates of Legislative
Voting: Michigan House of Representatives, 1 954 -1 961 ," Midwe st
Journal o~f Political Science, VI: (November, 1962), 384-396; Thomas R.
Dye, "A Comparison of Constituency Influences in the Upper and Lower
Chambers of a State Legislature, We s';ern Politi cal Quarterly, XIV
(June, 1961), 473-480; Thomas A. Flinn, "Party Responsibility in the
States: Soma7 Causal Factors, American Political Science Review, LVIII
(March, 1961:, 60-71; Thomias A. Finii and Harold L. Wolman, Con-
stituency and Roll-Call Voting: The Case of Southern Democratic Congress-
men," Midlwestt Journal o~f Political Science, X (May, 1966), 192-199;
Lewis A. Froman, Jr., Con~gressmen and Trheir Constituencies (Chicago:
Rand McNally & Company, 1963); Charles; H. Gray, "The Social Base of
the Coalition of Southern Democrats and N\orthern Republicans, Political
Science, XrVII (March, 1965), 31-33; Charles H. Gray and Glenn W.
Gregory, "Military Spending and Senate Voting: A Correlation Study,"
Journal o~f Peace Research, I: (19 68) 41-54; Havens, Metropolitan
Areas and Congress"; Leroy N. Rieselbach, The Roots o~f Isolationism
(Ilndianapoli s, Inc. : Bobb s-Merrill 1 9 66) pp 114-137; W. Wayne
Shannon, Party, Con stituency a nd Conrssoa Votin: A S~tudq of
Legislative Behavior in the United States House of Representatives (Baton
Rouge: Louisiana State University Press, 1968).
Jeffrey K. Hadden and Edgar F. Borgatta, American Cites
Their Social Characteristics (Chicago: Rand McNally & Company, 1965),
For data on districts of the United States House of Representa-
tives, a major breakthrough was achieved with the publication of U. S.
Bureau of the Census, Congressional District Data Book: Districts o~f
the 87th Congress A Staitia Abstract Supplement (Washington, D. C.:
U. S. Government Printing Office, 1961). The data base was greatly ex-
panded in U. S. Bureau of the Census, Congressional District Data Book:
Districts of the 88th Congress.
Va nd erslik "Constituencies and Roll Call Voting"; Hugh
LeBlanc, "Voting in State Senates: Party and Constituency Influences, "
Midwest Journal o~f Political Science, XIII (February, 1969), 33-57;
John G. Grumm, "A factor Analysis of Legislative Behavior, Midwest
Jouna o~f Political Science, VII (November, 1963), 3434-354; John G.
Grumm, "The Systematic Analysis of Blocs in the Study of Legislative
Behavior, Western Political Quarterly, XVIIZI (June, 1965), 359; John
E. Jackson, ";Statistical Models of Senate Roll Call Voting, Am eri ca n
Political Science Reviewv, LXV (ue19 1) 39-470; Jack E. Vincent,
"Predicting Voting Patterns in the General Assembly, Am eri ca n Poli ti cal
Science Review, LXV (June, 1971), 471-498; Jack E. Vincent, "National
Attributes ar, Predictors of Delegates Attitudes at the United Nations, "
American Political Science ReviewY, LXII (September, 1968), 916-931;
Jack E. Vincent, "An Analysis of Caucusing Group Activity at the United
Na tions Journal o~f Peace Research, II (1970), 133-150. One might
also add to ;ihis list a number of studies which have not focused primarily
on the legislative system, but which have included aspects of legislative
politics as y artt of a larger analysis. See, for example, Thomas R. Dye,
Politics, Economics, and! the Public: Policyl Outcomes in the~ American
States (U~hirago: Rand McNally & Company, 1966); Allan G. Pulsipher
and James L. Weatherby, Jr., "Malapportionment, Party Competition, and
the Function-il Distribution of Governmental Excpenditures, American
Political Science Review, LXII (December, 1968), 1207-1219; Ira
Sharkansky r ad Richard I. Hofferbert, "Dimensions of State Politics,
Economics, and Public Policy, American PolticalScience Review, LXZIII
(September, 1969), 867-879.
Vanderslik, Constituencies and Roll Call Voting, p. 228.
Hadden and Borgatta, American Citis, p. 5.
8Factor analysis will be applied to roll call data in Chapter V
of the present work, and previous roll call factor studies will be cited at
An early application may be found in Harold F. Gosnell,
Machine Politics: Chicago Model (Chicago: University of Chicago Press,
1937), Appendix B. For a comprehensive review and critique of the litera-
ture in this area, see Carl- Gunnar Janson, "Some Problems of Ecological
Factor Analysis, in Mattai Dogan and Stein Rokkan, eds., Quatiatv
Ecoogial nalysis in the Social Sciences (Cambrid ge: Ma ssa chu sett s
Institute of Technology Press, 1969).
Vincent, "National Attributes as Predictors of Delegate
Attitude s"; Vin cent "An Analysis of Caucusing Group Activity";
Vincent, "Predicting Voting Patterns. "
This strategy will to some extent parallel that used by Vincent,
who employs zero-order correlation, principal components analysis, and
canonical correlation in analyzing voting patterns in the United Nations
General Assembly. See Vincent, "Predicting Voting Patterns. "
These variables were taken from the U. S. Bureau of the
Census, C~ongre~ssiona District Data Book:~ Districts o~f the 88th Congress.
The variables deleted were median years of schooling, median family in-
come, proportion of housing sound units with all plumbing, median value
of owner-occupied housing units, median value of renter-occupied housing
units, and proportion of single-family housing unlits. For a similar screen-
ing procedure, see Sharkansky and Hofferbert, "Dimensions of State
Politics, p. 869.
Richard I. Hofferbert, "Socioeconomic Dimensions of the
American States: 1890-1960, Midwest Journal o~f Political Science,XI
(August, 1968), 407.
Hadden and Borgatta, American Cities; Charles M. Bonjean
et al., "Toward Comparative Community Research: A Factor Analysis of
U. S. Counties, Sociological Quarterly, X (Spring, 1969), 157-176.
For a similar application of this variable, see Robert L.
Lineberry and Edmund P. Fowler, "Reformism and Public Policies in
American Cities, American Political Science Review, LXI (September,
1967), 706. As a partial check on the validity of private elementary edu-
cation as a measure of Catholicism, the present writer correlated available
statewide data on the two variables and obtained a Pearson' s r coefficient
of .84. Data on Catholic populationwere obtained by aggregating diocesan
figures obtained from the National Catholic Almanac 1961 (Patterson, N.J.:
St. Anthony's Guild, 1961), pp. 440-443. Since a few dioceses cross state
lines, it was necessary to combine some state totals before computing the
correlation. Though used here as a measure of Catholicism, it should be
noted that to some extent, the variable "private elementary education" is
also in part a measure of socioeconomic status. For research showing
positive correlations between status and parochial education among Catho-
lics, see Andrew M. Greeley and Peter H. Rossi, "Corrolates of Catholic
School Attendence, School Review, LXXII (Spring, 1964), 52-73.
Thomas R. Dye, "Income Inequality and American State
Politics, American Political Science Review, LXIII (Mvarch, 1969),
157-162; Thomas R. Dye, "Inequality and Civil Rights Policy in the
States, Journal o~f Politics, XXX:I (Novemnber, 1969), 1080-1097.
For the formula used in applying the Pareto curve see
Herman P. Miller, Incomie Distribution in the United States (Wa shington ,
D. C.: U. S. Government Printing Office, 1966), pp. 215-216. In twYo
districts, :to seventeenth and twventy-sixth of New York, Pareto curves
could not appropriately be applied since th~e highest income categories
contained mo~-e families than did the second highest category. Following
Miller, the figure of $24,000 was selected as the average income in the
highest category for these cases. For the formula for the trapezoidal
approximati~n to the Gini coefficient, see, ibid., pp. 220-221. An equiva-
lent: method is presented in Oliver Benson, Political Science Laboratory
(Columbu s cOhio: C. E. MeLrrill Pub. Co. 1969), pp. 8-11. Economists
seem to fee~l that the trapezoidal approximation provides excellent results
with eight or more income categories. Since only seven categories were
available at .ho congressional district level, a validity check was made
by aggregati;,g categories on a statewide basis, and correlating the resulting
Gini indices with those produced at the statewYide level in two earlier studies
using more detailed data. The results of these studies can be found in
Thomas D. Hopkins, "Income Distribution in Grant-in-aid Equity Analysis, "
National Tax Jounal XVIII (June, 1965), 209-213; Ahmad Al-Samarria
and Herman P. Miller, "State Differentials in Income Concentration, "
American Economic Review, LVII (March, 1967), 59-72. The obtained
correlations were .968 with the Hopkcins figures and .998 with those pub-
lished by Al-Samarria and Miller.
Solutions were also derived in which different numbers of
factors were extracted and rotated. The seven-factor solution was almost
identical to the six-factor solution, with the seventh factor having no
loading higher than 24. Four-and five-factor solutions yielded results
which seemed to be a bit less clear than those produced with six factors.
The formula for the computation of factor scores is:
X = c .Xi
f i-1 1
where X, = the factor score on factor f,
Bi = the regression coefficient on factor f for the i th
p = the number of variables.
See Lee F. Anderson et al. Legislative Roll-Call Analysis (Evanston,
Ill.: Northwestern University Press, 1966), p. 1412.
A N\UMERICAL TAXONOMY OF
While legislative analysts have devoted a great deal of atten-
tion to the study of constituency attributes, an equally voluminous litera-
ture has grown up around studies of what Eulau and H-inckley call the
"Miliou of the legislative district.1 In the latter studies an attempt is
made to characterize the nature of the area in wYhich thle district is located
and to then compare different types of arer s with respect to legislative be-
havi or. In ttlis chapter areal classifications schemes employed in legisla-
tive research will be reviewed, criticisms leveled against them, and an
alternative strategy for the development of areal typologies proposed and
The most common basis of areal classification of districts has
included concepts such as "urbanism, "ruralism, "metropolitanism, "
and related notions. For want of a better term, such classification schemes
will be referred to here as "urban-related" typologies.
Most urban-related typologies are based on simple urban-rural
dichotomies. The operational basis of this dichotomization takes many
forms. Some studies employ the Bureau of the Census term "urban place"
to differentiate urban areas from rural.2 Other census terms, "urbanized
area" and "Standard Metropolitan Statistical Area (SMSA)" have also been
used.3 An occasional researcher has derived his own cut-off between
urban and rural areas.
A number of attempts have been made to devise typologies based
on more complex distinctions. Friedman, for example, classifies state
legislative districts in Tennessee as "urban" (those containing cities of
greater than 50,000 population), "mixed" (those with cities of at least
15,000 but not more than 50,000), and "rural" (those with no city as large
as 15,000). A Congressional Quarterly typology, which has been used!
in research by Smith6 and by V. O. Key,7 divides congressional seats into
"rural, "small town, "mid-urban, and "metropolitan" districts." In
recent years several investigators have sought to make distinctions with-
in metropolitan areas and, in particular, to Incorporate thet colicept "subuib"
into studies of legislative constituency.9
Another form of areal categorization which has appeared freque ntly
in legislative research is that based on groupings of districts by geographic
location.1 As with urban-related typologies, sectionally based district
typologies have taken many forms. At the national level a distinction is
almost always made between North and South, though the exact boundary
line between the two varies from study to study. Another widely used com-
parison is that between coastal and interior sections. Some sectional
classifications are quite simple, others involve a fairly large number of
categories.11 Whlile such distinctions are most common at the national
level, they are also sometimes employed within state boundaries.12
Sectional and urban-related typologies are by no means incompatible, a nd
often districts are simultaneously classified in both ways, for example,
" rural- Sou th ern "
The extensive use that has been made of both urban-related and
sectional typologies invites examination of the rationale behind their con-
struction. Unfortunately, students of the legislative process have on the
whole been Jacking in introspection in this regard, though given the
necessity for academic division of labor this is perhaps inevitable. In
the following discussion, then, it will be necessary to draw upon insights
not only of past legislative analysts but of researchers in other areas of
social science in which similar distinctions have been employed. In this
discussion criticisms will be set forth which cast grave doubt upon the
general usefainess of classifications based on urbani-rural and sectional
distinctions. At the same time, an understanding of these criticisms will
provide the basis for a more satisfactory approach to the development of
Among legislative analysts the most frequent, but perhaps least
damaging, criticism of past areal typologies of legislative districts has
been that significant differences do not, in fact, occur in the political
behavior of voters and representatives from different types of constituency.
While there may be merit in this argument, it is less than wholly convinc-
ing. For one thing, such criticisms have been directed primarily against
classifications based on urban-rural dichotomies, the crudest of the areal
typologies discussed above. Many of the more sophisticated classification
schemes which have been developed have yet to be subjected to extensive
testing. further, while many studies have failed to reveal inter-type
legislative differences, other studies can be cited which have reached the
opposite conclusion.14 Finally, even if a typology fails to reveal inter-
type variance, a finding that different types of constituency environment
produce similar kinds of political behavior might itself be of major import-
ance, and does not necessarily demonstrate that the typology in question
is not useful.
A more telling criticism deals with the units of analysis employed
in making arcal distinctions. Legislative district boundaries seldom cor-
respond to the boundaries used in areal classification. "Urban places, "
for example. usually cover a much smaller geographic area than do legis-
lative districts, and a single congressional district may contain several
fairly large t rban places, along with their suburbs, as well as extensive
rural area s. Major metropolitan areas, on the other hand, often encom-
pass a number of legislative districts, as do sectional divisions. In
either case, this lack of boundary congruence can create serious problems
for ca s s ifi ca tion If a legislative district contains, for example, about
equal proportions of urban, suburban, and rural zones, how ought it to be
classified ? Aside from this practical difficulty, an even more serious
theoretical problem lies in the fact that the entirety of a district must be
categorized on the basis of the nature of only a part. Conversely, when
a unit used in developing areal distinctions encompasses several legisla-
tive districts, the analyst is forced to place all such districts in the same
category, regardless of differences among them. In short, the unit of
analysis on the basis of wYhich areal typologies are derived ought to be
the same as the unit of analysis the researcher wishes to study. This is
not the case with either urban-related or sectional typologies.
Another difficulty with m1any areal typologies is the bewYildering
array of operational definitions which are often given to many of the terms
employed. Hadden notes that at least three different operational definitions
of the concept of "suburb" are curr-ently in usage. Similarly, the term
"rural" has been operationalized in radically different ways. Sometimes it
is defined as including any area outside the built-up ring around cities of
50,000 or more, at other times as only those places outside incorporatoJ
areas of 2,500 or more.
The problem is much more fundamental than simply one of dis-
agreement over operational definitions, which is a difficulty encounterec
in almost any research problem in the social sciences. The conceptual
basis for the terms used in areal classification is very amorphous. Behind
ad hoc operational definitions lies conceptual confusion.
Dewey, in a review of the sociological literature dealing with
the concept of urbanism, found that no less than forty different meanings
have been associated with the term by various authors. While there was
some overlap in definitions, there was nothing like consensus on the mean-
ing of the concept.17) Kurtz and Eicher demonstrate that a similar con-
fusion has existed in the literature with respect to the suburbanism con-
cept,8 and Shryock remarks that "suburb is used almost as loosely by
social scientists as by laymen. "9Smith writes that "the sectional hy-
pothesis has...always suffered from the rather amorphous and elusive
nature of a 'section' for the purpose of analysis."20 The following com-
ment by Hauser, made in reference to research employing urban-rural and
folk-urban distinctions in studies of non-Western societies, has general
applicability to areal classification concepts.
There is evidence, by no means conclusive as yet, that
both parts of these dichotomies represent confounded
variables and, in fact, complex systems of variables
which have yet to be unscrambled. The dichotomiza-
tion3 perhaps represent all too hasty efforts to syn-
thesize and integrate what little knowledge has been
acquired in empirical research...integration is often
achieved of that which is not yet known.2
El en if the individual concepts used in a given typology could
ber- rdei~fribed the problem of setting fror-th-r rationale for choosing these
concepts as the basis for areal classification would still remain. As thle
brief surve: earlier in the chapter should make evident, there are an al-
most infinite number of ways in which areal distinctions can be made.
Faced with this embarrassment of riches, how is the analyst to choose?
Is a metropolitan-outstate distinction better or more appropriate than a
differentiation between urban and rural area s? Within metropolitan areas
how important is it to distinguish between central city and suburb? Should
metropolitan areas be further differentiated according to their overall size?
Into how many and which sections ought districts be divided ? Do urban-
related and sectional bases of classification complement each other, or
are they alternative approaches to fulfillment of the same function? The
answers to these questions and, in fact, the criteria which one would use
in order to answer them are far from obvious.
It may be, as Eulau and Hinckcley charge in their review of
literature comparing urban and rural legislative districts, that researchers
employing areal typologies are guilty of reification. Two 'sides,' 'force s ,
or 'interests' are regarded as opposed to each other, as if they were real
collectivities, and the cleavage postulated to exist between them, often
derived from quite untenable historically grounded axioms, is accepted as
a basic explanatory factor of legislative behavior.,2
If such classifications really are reifications, of course, it
might be better to dismiss them entirely, as at least one analyst seems to,
suggest. It will be argued hero, however, that such distinctions re-
present, not reification, but, in a tentative and groping way, a perfect!}
valid insight about measurement of constituency environment. Legislati~ e
analysts, usually only implicitely, seem to be using urban-related and
sectional typologies as shorthand representatives of complexes of con-
stituency attributes. One can occasionally find in the literature (on legis-
lative analysis and elsewhere in the social sciences) explicit statements
of this assumption. Smith, in a study of the Eighty-sixth Congress, admits
that "a great many factors other than mere city or county residence are, of
course, concealed beneath the rubric of ruralism-urbanism, but argues
that "the rural-urban difference still makes convenient categories for the
purpose of analysis. "24 Similar sentiments are found in V. O. Key's
metropolitanism andi ruralism are broad and deceptive
terms that mask other differences that are probably
more influential politically than is the fact that some
people live in the city and others in the country.
Nevertheless, urban-rural differences makd a conven-
ient peg on which to hang a discussion.
Sociologists Gist and Fava point out,
Typically, rural-urban definitions achieve their utility
by a very simple method. The definition consists only
of somne easily measurable items such as size, density,
or legal community status. These simple items are
assumed to be associated, in ways that may or may not
be fully understood, with more complex items such as
socioeconomic behavior. Therefore, by employing the
definition based on the simple ~tpns one has an index
to the more complicated items.
Friedman, in an essay critical of urban-related classifications, states
the assumption made by socia) scientists in using
urban-rural distinctions of this kind is that the term,
while not precise, provides a useful tool for distin-
guishing socioeconomic group interests which tend to 2
congregate in the two different type s of environment.
Similar assumptions often lie behind use of sectionally based
typologies. Marwell states that
the importance of region in structuring roll call
voting invites again the question of why region
should act: as an important independent variable....
Perhaps the best we can do at this point is to specu-
late that region tends to act as a sort of collective
variable adding the effects of a variety of influences
all distributed geographically. 2
Similarly, Turner writes that
sectional differences in congressional voting behavior
may be in many cases manifestations of sectional dif-
ferences in economics, religion, culture, or other
In short, the areal and attribute approaches have the same
empirical basis and differ only in that the former seeks to deal with con-
stituency characteristics holistically, with the attendant advantages and
limitations suggested in the introduction. This may be belaboring the
obvious, but, unfortunately, few researchers demonstrate that the differ-
ent areal types they employ in fact tap different distributions of constitu-
ency chara cter isti cs Even when this is done, there is no assurance
that another of the wide range of alternative schemes available to the re-
searcher might not have succeeded better ini discriminating among types.
Despite these deficiencies, such studies suggest a criterion
for developing and evaluating areal typologies wYhich can serve as a start-
ing point for more adequate classification Presumably, when a researcher
classifies districts as, for example, urbar., suburban, and rural, an as-
sumption is being made that districts fallilig into the suburban category,
while not identical to each other in terms o~f the variable complex which
the classification scheme taps, are at least relatively similar to each
other and relatively dissimilar to districts falling in the urban and rural
categories. A typology o~f legislative constunce should, then, be
regarded as adequate t~o the extent that it groups districts in such a way
as t~o maximize within-group honogeneity and bIietwe-ru heterogeneity
with respect t~o a specified set o~f attributes. In past research it usually
has been simply assumed that this criterion, or something like it, is met
satisfactorily by whatever typology the investigator is using. What is
needed is an approach that will treat within-group honogeneity not as an
assumption but as a goal to be achieved in a rigorous and systematic
fashion. To do this it will be necessary to classify areas not on the basis
of a theoretically premature deductive method, but inductively, beginning
with specified constituency attributes according to which constituency
attributes can be grouped.
Inductive typology construction is certainly no novelty. As far
back as 1927, Rice attempted to group legislators into clusters whose
members had high frequencies of agreement in their voting behavior.31
His technique, subsequently subjected to some modification, has by nowN
become a standard tool in roll call analysis.
Inductive grouping procedures have perhaps been most exten-
sively exploited in biology. It was in this discipline that the term "nume-ri-
cal taxonomy" was coined to describe the practice of grouping objects irto
relatively homogeneous clusters on the basis of detailed characteristics. 33
An interesting synthesis of clustering traditions in biology and in roll ca~ll
analysis has been attempted by Grumm.3
Even in the study of population aggregates, inductive techni-
ques are increasingly employed in construction of typologies, and have
been used to classify primitive tribes,35 census enumeration districts,36
census tracts,3 7 citie s,3 8 counties,3 states,4 and nations.4 Indeed,
it can be said with little exaggeration that such techniques have been em-
ployed in grouping almost every kind~ of population unit except legislative
In the present study, the districts of the Eighty-eighth Congress
will be grouped on the basis of the same nineteen attributes discussed in
Chapter II. The procedure used involves three steps: (1) data reduction
through use of principal components analysis, (2) grouping of districts
using a hierarchical clustering algorithm proposed by Ward, (3) refinement
of clusters produced by Wrard's algorithm to improve group honogeneity.
Each of these steps will now be described.43
(1) For reasons of economy and of program limitations, it
was decided to reduce the data space before carrying out subsequent stcps
in the grouping process. Principal components analysis was chosen for
this purpose. This technique is similar to factor analysis, but begins with
unities in the diagonal of the correlation matrix and does not iterate on
these initial communality estimates. The result is that the axes extract.d
from the matrix are based on all the variance in the matrix rather than, as
in factor analysis, the common variance only. Principal components ane l-
ysis involves no assumptions about underlying dimensions in the data
matrix, and thus is inappropriate when analysis requires interpretation of
common factor stru cture.4 For present purposes, however, principal com-
ponents analysis was felt to be preferable to factor analysis, since the
only goal at present is a mathematical transformation of the data into as
few dimensions as possible. Since it is to be used solely as an inter-
mediate step in the analysis, interpretation of dimensions is not necessary.
All nineteen of the social, economic, and demographic variables
were subjected to principal components analysis. The first eight com-
ponents accounted for 91 per cent of the total variance in the original
data matrix and so it was decided to retain these eight for further anal-
ysis. Retention of a ninth component would have contributed only an
additional 2 per cent of variance. Since interpretation of components was
not required, no rotation of axes was performed. Component scores were
computed in the same way as that used in Chapter II for computing factor
One further modification was made of the data before the dis-
tricts were grouped. Since interest is in the original variables, rather
than in the principal components themselves, each component was weighted
by the amount of variance it explained in order to remain as close as
possible to the initial data. This was don. by multiplying all component
scores by the square roots of their respective eigenvalues.45
(2) Using weighted principal component scores, districts
were grouped using Ward's error sum of squares hierarchiical clustering
algorithm.46 In the procedure proposed by Wnard, each case is initially
considered to be a separate "group. At this point, since each group con-
tains only one case, each is perfectly homogeneous. Then those two
groups are joined whose union will result in the smallest increment in
within-group heterogeneity. Within-group heterogeneity (or variance) is
measured by computing the squared difference among all cases within
each group on each variable (in the present study, each weighted com-
ponent) and summing the results over all variables and groups. This total
within-group variance is referred to by Ward as the "error sum of squares
(ESS)." With the union of the two groups to form a single group, the num-
ber of groups is reduced by one. The amount of within-group heterogeneity
provides a measure of the tightness of the newy grouping. The process is
repeated until all cases have been joined into a single group, at each
step those twYo groups being joined whose union wvill result in the least
increment in heterogeneity.
In the past, solution of this algorithm for problems involving
as many cases as are being analyzed here would have been difficult if not
impossible even on a high speed digital cilmputer. Philip Bell and
Steven Gladin~, however, have recently dev-ised a method for solving Ward's
algorithm in a wYay which can e~conomicaliv accommodate problems far
larger than the one encountered at present.48
Ward's arror sum of squares measure is an especially useful
statistic. It provides an operational mean a for evaluating the adequacy
of a given classification scheme according to the criterion set forth
ea rlier-. The measure is precise, mathematically simple, and has a high
degree of face validity as a measure of the "tightness" of a classification
scheme. As will be demonstrated later in this chapter, it can be used to
evaluate typologies derived by means other than Ward's procedure.
The task remains of determining the number of groups to be re-
tained for further analysis. It is usual to select a grouping level occur-
ring just prior to a sharp loss in homogeneity, since this indicates a point
at which an additional small gain in parsimony must be achieved at the
cost of a relatively large loss in detail of information. Ward recommends
that, to facilitate this task, output from a clustering program should in-
clude, after each step in the clustering process, the error sum of squares,
the increase in this figure over the previous step, and the acceleration
(the increase in the increase) over the previous step. In Table 3 these
figures are given for the last fifteen steps of the clustering algorithm.
All figures are expressed as percentages of the total variance of all dis-
tricts on all weighted components.
The decision was made to retain five groups for analysis.
Eight groups would perhaps provide a more natural breaking point, but at
this level three groups consisted of only twrenty-five, twenty-two, and
seventeen districts respectively. It was felt that, in an exploratory study
such as this, there would be little of themaetical or substantive interest
to be gained by analysis of these relatively small groups and that the finer
distinctions which such analysis might yield canl better be left for future
(3) One major shortcoming of Wrard's method is that once
two groups have been joined at any stage in the clustering process, their
members can never be separated at a subsequent stage even when such
separation would result in c1 more homogeneous grouping.49 It is possible,
however, to in part meet this difficulty by moving districts to groups with
which they have a better fit than the groups to which they were assigned
by Ward's method. For the present problem, this was done according to
the following procedure: Once hierarchical clustering is completed, and
a given grouping level selected, the squared difference between the score
WrITHIN GROUP HETEROGENEITY'
(AS PER CENT OF TOTAL VARIANCE)
FOR FINAL FIFTEEN STEPS IN GROUPING PROCEDURE
of each case and the mean score of each group is computed for all com-
ponents and summed. Cases are then tentatively reassigned if they are
closer to the centroid of a group other than their own. The error sum of
squares resulting from this new arrangement is computed and if it is lower
than the initial error sum of squares, the reassignments are completed.
Since such reassignments alter group means, the process is repeated in
It had originally been intended to continue iteration until no
further reas.<.ignments were indicated or until such reassignments failed to
further reduce within-group variance. However, after the first iteration,
subsequent iterations produced very small, improvements but did not reach
convergence. After the sixth iteration produced an improvement in the
error sum of squares of less than one-twentieth of 1 per cent, the de-
cision was r ade to terminate the process at that point. The error sum of
squares after six iterations was 48. 65 per cent of total variance.
Just as the results of a factor analysis must be given a sub-
stantive interpretation, so too must the types just derived be supplied with
descriptive labels. Three tactics were employed in order to do this.
First, the mean scores of each type were examined, both on the
six constituency factors derived in Chapter II, and on the nineteen indi-
vidual characteristics from which the factors were derived. Table 4 lists
the means (1J) and standard deviations (e) for each type of each constit-
uency factor and individual characteristic. The table also lists, for each
variable, the between-group variance expressed as a per cent of total
CONSTITUENCY ATTRIBUTES OF INDUCTIVELY
DERIVED DISTRICT TYPES
TABLE 4 -- (Continued
II III IV
fa ctur- perou s
Chara cteri stics
TABLE 4 -- (Continued
I II III IV V
Ma nu- Pr-os- Popu- Unde- Unde- TotalIs
factur- perous- Jation veloped voloped-
ing growth Center deprived
y .22 .14 .28 .09 .04 .15
0 .11 .07 .12 .07 .05 .12
y .06 .06 .10 .07 .18 .09
0 .02 .03 .04 .04 .05 .06
9 .07 .11 .07 06 .06 .07
0 .02 .03 .05 .02 .02 .03
o .08 .08 .10 .17 .27 .14
0 .03 .03 .04 .06 .08 .09
p .05 .07 .05 .03 .02 .04
0 .02 .04 .05 .01 .01 .03
9 .32 .34 .35 .37 .41 .36
0 .02 .04 .04 .03 .03 .04
y .05 .04 .07 .05 .06 .05
0 .02 .02 .02 .02 .02 .02
TABLE 4 -- Continued
variance. This statistic is known as nl2, or the squared correlation ratio.
Note that the higher the value of the squared correlation ratio the greater
the homogeneity of the types.
Second, in order to interpret the inductively derived typologies
and to relate them to more traditional modes of district classification, the
types were cross-classified with three "traditional" groupings. The results
are given in Table 5. The first traditional typology is a simple urban-rural
dichotomy. Districts are classified as urban if more than 50 per cent of
their inhabitants resided in urban places in 1960, and as rural otherwise.
The second typology is a more complex classification scheme devised by
the Congressional Quarterly Service, which categorizes districts as urban,
suburban, rural, or mixed. Finally, a regional typology is considered
in which districts are classified as Northeastern, Midwestern, Border,
Southern, and Western.5
A third more impressionistic tac ic used in interpreting induc-
tive types was to examine the locations of districts in each type. In this
the maps found in the Congressional District Data Book proved helpful, as
did the capsule descriptions of district locations provided by Congressional
Type I: Ma nu facturi ng Di stri cts Type One, with 136 districts,
is the largest of the five clusters. Districts in this type will be designated
as "Manufacturing" districts, since the average district in the type has 36
per cent of its labor force employed in manufacturing, a figure which is
easily the highest of any of the types. Type One districts, predominantly
CROSS- CLASSIFICATION OF INDU CTIVE TYPOLOGY
WITH THREE "TRADITIONAL" AREAL
CLASSIFICATION SCHEMES, BY COLUMN PERCENTAGES
(N's in Parentheses)
I II III IV V
Manu- Pros- Popu- Unde- De- Totals
fa ctur- perous- lation veloped veloped
ing growth Center deprived
Urban 86(117) 9 9(79) 100o(4 2) 4 5(4 2) 30(25) 7 0(3 05)
Rural 14( 19) 1( 1) 0( 01 55(52) 70(58) 30(130)
Totals 1.0 0(13 6) 10 0(8 0) 1 0 0(4 2' 10 0(9 4) 10 0(8 3) 10 0(43 5)
Urban 24( 33) 29 (23) 9 5(4 0) 2( 2) 6( 5) 24(103)
Suburban 16( 22) 33(26) 5( 2) 1( 1) 0( 0) 12( 51)
Rural 32( 44) 9( 7) 0( 0) 88(83) 83(69) 47(203)
Mixed 27( 37) 30(24) 0( 0) 9( 8) 11( 9) 18( 78)
Totals 100(136) 100(80) 100(42) 100(94) 100(83) 100(435)
Northeast 52( 71) 11( 9) 60(25) 3( 3) 0( 0) 25(108)
Border 4( 5) 10( 8) 5( 2) 17(16) 7( 6) 9( 37)
South 0( 0) 20(16) 0( 0) 16(15) 90(75) 24(106)
Midwest 37( 50) 18(14) 24(10) 44(41) 9( 0) 26(115)
We st 7( 10) 41(33) 12( 5) 20(19) 2( 2) 16( 69)
100(136) 10 0(8 0) 100(4 2) 10 0(9 4) 10 0(83) 100 (43 5)
located in the Northeast and the Midwest, are for the most part not found
in this country's great metropolitan areas, but rather contain one or more
cities of moderately large size, such as Trenton, New Jersey (1960 popu-
Jation: 34,913); Newbhurgh, New York (30,979); and New Haven, Con-
ne cti cut (1 52 ,04 8) Though on the whole of only medium population
density, the bulk of districts in Type One are classified as "urban" accord-
ing to the Census Bureau's "urban place" criterion. Congressional Quarterly's
typology classifies many districts in this type as suburban or rural, a fact
which points up somec of the weaknesses of this classification scheme. At
least some of the suburban districts included in this category (for examp.'e,
the districts encompassing Camdon, New Jersey and East St. Louis, Illi-
nois) do not closely resemble the usual image connoted by the term subv~;b. "
Among the "rural" districts included here are the Sixth District of Wisco Isin
(embracing Sheboygan and Oshkosh and having 40 per cent of its labor
force employed in manufacturing) and the Eighth District in the same state
(which includes Green Bay and has 33 per cent of its labor force employed
Aside from its industrial character, an outstanding feature of
Type One districts is their relative lack of economic inequality. The mean
Gini index of Type One districts is only .32, the lowest of any type. This
seems to result from a combination of about average levels of extreme af-
fluence (the type differs by less than 2 per cent from the average for all
districts in white collar employment, college educated persons, and fami-
lies with income of $15,000 or more per annum) together with lower than
average extr-eme poverty. (This is the lowest of all types in overcrowding,
and is tied with Type Two for smallest proportion of technical illiterates
and families with less than $2,000 per annum income.)
Demographically, Type One districts have, on the average,
high concentrations of persons of foreign stock and Catholics. The exodus
of Blacks fr-om the South during the 1950s, however, seems to have largely
bypassed Type One, which averages less than half as many Blacks as does
the average district in the nation as a whole. More generally, Type One
districts seem to have experienced below -average in-migration during the
years prior to 1960.
Type II: Prosperou s-growth Districts. Type Two districts are
well dispersed regionally, though a disproportionate number are found in
the We st Northwestern and Midwestern districts in Type Two are com-
posed for the most part of high status area, on the fringes of the nation's
large metropolitan areas (for example, Na asau and W'estchester counties
in New York). Southern Type Two districts are located pr-imarily in the
peripheral South. Six such districts are in Florida and five in Texas. The
district which includes Atlanta, Georgia is also found in Type Two. Of
the thrity-thr~ee Western districts in Type Two, no less than twenty are
found in California .
Two major qualities distinguish most of the districts in this
type: rapid population growth and a high level of affluence. Type Two
districts will, therefore, be cla ssified as "Prosperous -growth di stricts .
The geographic distribution of districts noted in the last paragraph reflects
the shift of population in the United States during the 1950s to the West
and to certain parts of the South and, within geographic regions, away
from both "rural" and inner-city areas toward "suburban" locations. The
mean growth rate of Type ?Two districts (72 per cent) was more than three
times that experienced by the nation as a whole. Type Two had, on the
whole, fewer elderly persons than did any other type.
More than any other, Type Two enjoyed a high degree of pros-
perity at the time of the 1960 Census, leading all types in per cent college
educated, per cent in white collar occupa'ions, and proportion of families
with annual incomes of $15,000 or more, and suffered somewhat loss un-
employment than any other type.
Type III: Population Ceniter Districts. Type Three, with forty-
two districts, comes remarkably close to f'tting Louis Wirth's classic
definition of urbanism as consisting of siz 2, density, and heterogenoity.56
Without exception, all districts in this type arc located in or near very
large metropolitan areas. Just one-half are in either the Greater Nrew York
or Chicago areas. Other areas represented here include parts of Boston,
Baltimore, Detroit, Cleveland, Los Angeles, San Francisco, and all of
Perhaps the most outstanding feature of Type Three is its con-
centration of population. The average district in the type contains over
28,000 persons per square mile, more than seven times that of the average
for all districts. Only 28 per cent of housing in this composite of Type
Three districts are owner occupied, a figure which is more than doubled
by each of the other types. Because of their extremely high mean density,
Type Three districts will be designated as "Population Center" districts.
As to heterogeneity, Type Three has a higher proportion of per-
sons of foreign stock and more Catholics than any other type, and in pro-
portion of Blacks is surpassed only by Type Five.
A few other facts about Type Three deserve comment. Perhaps
somewhat surprisingly, employment in manufacturing is only slightly above
the average for all districts, undoubtedly reflecting the fact that America's
largest cities have reached a "post industrial" phase of development, with
high levels of employment in tertiary industries (no indicators of which
were included in this study) .
Many of the districts in Type Three score high on deprivationi
or include very deprived areas within their boundaries, and many of the
nation's largest ghettos fall in such districts. It would be a mistake,
however, to categorize Type Three districts in general as areas of pover-ty.
In New York, both the Seventeenth (the "silk stocking") District and the
Eighteenth District in Harlem are in Type Three.
On balance, Type Three districts stand out as far as age com-
position is concerned. The average of this type for proportion of popula-
tion under eighteen years of age is lower than that of any other type.
Types Three and Four are tied for highest average proportion of elderly
Finally, it might be noted that Type Three was the only type to
register an absolute loss of population during the 1950s. At the same time,
population mobility was only a little below average. These indicators of
a combination of population turnover together with little net population
change coincide with the popular image of the inner core of large metro-
politan areas during this period.
Type IV: Undoveloped Districts. Type Four districts are
heavily concentrated in the Midwest, and also show strength in Border
states, in the peripheral South, and in the Rocky Mountain section of the
West. By almost any definition of the term (except by the "urban places"
criterion), it is the most "rural" of all types, with very low population
density, very little manufacturing, and much agriculture. All but a few
districts in this typo are located outside of Standard Metropolitan Statisti-
cal Areas and, even by the "urban places" criterion, most are classified
as rural. Type Four districts will, therefore, be designated as "Undeval-
oped districts .
Population in Type Four districts, which experienced relatively l
little growth in the 1950s, is very homogenous, consisting of proportion-
ately few Blacks, Catholics, or persons of foreign stock. Type Four dis-
tricts are not outstandingly prosperous, but neither do they tend, in gen-
eral, to be areas of extreme poverty.58 Elderly persons are somewhat
overrepresented in this type.
Type V: Undeveloped- deprived Districts. Type Five districts
are located primarily in the deep South, including all of Mississippi,
Louisiana, Alabama, and South Carolina, plus all of North Carolina but
the Greensboro- Durham- High Point area, all of Georgia but the Atlanta
area, and all of Tennessee but the Nashville area.
In some respects Type Five closely resembles Type Four, but
by most measures is somewhat less "rural. The strongest factor differen-
tiating the two types is the extreme poverty of districts in Type Five.
Whrerea s Type Four districts average only 7 per cent technical illiteracy,
compared to 9 per cent for all districts, Type Five districts average 18
per cent, easily the highest of any type and twice the nationwide average.
Type Five districts also score lowest in per cent of labor force in white
collar occupations, highest in proportion of families with annual income
under $2,000, and highest in economic ineqcuality. Type Five districts
will be de si g nated a s Undevel oped -deprived "
It is interesting to note that, despite their sparse density,
Type Five districts evidence considerable overcrowding in housing con-
ditions, leading all other types by a wide margin in this category. Also,
despite the nationwide high negative corre'.ation between density and
owner occupancy of housing, Type Five districts are below average on the
Even more than Type Four districts, Type Five districts have few
Catholics and few persons of foreign stock. The latter type, however,
leads all others in its proportion of Black residents. Despite their low
mean growth rate, Type Five populations tend to be younger than average,
having the highest proportion of persons under eighteen years of age of
any type, and a proportion of elderly persons exceeded by all groups but
All of the above descriptions and labels must be viewed with
considerable caution. They constitute attempts to make generalizations
about each group as a whole and it must be emphasized that these general-
izations may be more or less inaccurate when applied to a given group
member. Conversely, a member of one group may have many of the charac-
teristics of a group other than that to which it ha s been a signed .
Overall, the district topology presented here succeeds moder-
ately well in meeting the stated objective of classifying districts into
homogeneous groups. The error sum of squares presented earlier shows
that the five types are differentiatted by 51 per cent of the total variance
in the eight weighted principal components. Somewhat more meaningful
figures can be derived by averaging the nk~oofficients in Table 4. MH~en
this is done it can be seen that the types account for an average of 40 pecr
cent of the total variance in the six constituency factors and, most im-
portant, for 47 per cent of the variance in the nineteen individual charac -
teristics. These figures indicate a considerable loss of information from
that contained in the original data but, after all, reduction of 435 categor-
ies to five represents no little increase in parsimony, and it is to be ex-
pected that a price must be paid for this gain. It is to be hoped that future
research will be able to improve on these results, but it is likely that in
large measure the failure to produce more homogeneous groups is due to
the fact that congressional districts simply do not fall into neat clusters,
and that the universe with which the student of congressional constituencies
deals is a complex and inelegant one.
At the same time, it must be made clear what the typology pre-
sented in this chapter does not accomplish. The procedures followed are
designed to maximize inter-type differences in constituency attributes.
From this it does not follow that the resulting grouping will maximize dif-
ferences on any other variable, such as constituency electoral behavior,
or legislator roll call behavior. This fact does not necessarily detract
from the typology's value, for political similarities among constituency
types may be as important as differences. For example, conventional
wisdom holds that both "suburban" and "raral" areas produce conservative
political behavior, though the two are usually thought of as being quite
different in social, economic, and demographic attributes. If two very
different kinds of constituency produce simiilar patterns of politics, this
fact is as wroithy of investigation and explanation as any political dif-
ferences that a comparison of constituency might uncover. Thus the typol-
ogy developed here differs in its purpose from the cluster-bloc analysis
of legislative behavior cited earlier, since the two approaches, while
methodologically similar, differ in the criterion used to define homogeneity.
Before concluding this chapter, a few further comments are
necessary concerning the relationship between the typology developed
here and traditional areal classification schemes. In the first place, as
indicated in Table 5, the types derived inductively in the present analysis
do bear some relationship to the types found in traditional deductive classi-
fication schemes. Manufacturing districts are generally located in the
Northeast and urban Midwest. Most Prosperous-growth districts are
Western and many can be classified as suburban. Population Center dis-
tricts are clearly metropolitan. Undeveloped districts are mostly rural
and Midwestern, while Undeveloped-deprived districts are found primarily
in the Deep South.
The correspondence of the two modes of classification is, how-
ever, only very rough aind approximate. Further, each of the three tradi-
tional classification schemes presented in Table 5 compar-es unfavorably
to the inductive procedure! in terms of homogeneity. When the error sum
of squares is computed for each of the deductive typologies, the follown-
ing results are obtained: for the two-fold urban-rural typology -- 85
per cent of total variance on eight weighted principal components; for thie
four-fold C~ongressional Quarerl typology -- 75 per cent;9 and for the
five-fold sectional typology -- 71 per cent. It should be noted that in
each case, this figure is clearly higher than that shown in Table 3 for
comparable grouping levels of hierarchical clusters, even without refine-
ment of the latter according to the procedure described above. The se
results bear out the argument advanced earlier in the chapter. On the
one hand, they indicate that there is indeed some degree of validity in
the use of urban-related and sectional typologies as "shorthand" variables.
On the other hand, they confirm the need for systematic rather than intui-
tive methods of district classification. It is true, of course, that, had
different deductive typologies been used for comparison, a closer corres-
pondence to the inductive groupings might have been obtained, but this
fact hardly detracts from the necessity for rigorous means of developing
and evaluating classification schemes.
There is another way in which the results of the analysis car-
ried out in this chapter may be relevant to the evaluation of traditional
typologies. Most such groupings have contained only two or three classi-
fica ti ons The figures in Table 3 show that, at the three-group level,
hierarchical clustering produces within-group heterogeneity amounting to
about two-thirds of total variance in the weighted principal components.
At the two-group level, over three-fourths of total variance is within-
group. It: is likely that, whether out of necessity or convenience, many
future researchers will continue to rely on traditional modes of district
cla s sifica ti on The figures just cited strongly indicate that in such situa-
tions even a minimum of prudence would suggest careful examination of
whatever groups are used to insure that they in fact are differentiated by
constituency characteristics and would argue for employment, if at all
possible, of a classification scheme containing several categories.
Finally, it is necessary to consider a possible alternative to
the notion advanced in the above pages that traditional areal classifica-
tion schemes are best seen as ways of tapping complexes of more specific
constituency attribute s. While this might be true of much of the use of
these classifications in past research, it is not clear that such is univer-
sally the case. Social scientists and others sometimes speak of distinc-
tive "ways of life" associated with concepts such as urbanism, suburbism,
or sectional location. It could be argued that these areal concepts incor-
porate historical and cultural factors that cannot be fully measured by
characteristics contained in contemporary census reports, and that typolo-
gies derived from these characteristics, therefore, cannot adequately ful-
fill the function of traditional modes of classification. While plausible,
this line of reasoning is subject to two objections.
In the first place, assuming that there are important cultural
differences which are not measured by the various social, economic, and
demographic attributes available from the conses results, there is no very
persuasive evidence available to indicate the nature of' these differences,
and still less to show that such differences bear any very close corres-
pondence to urban-related and sectional areal typologies. At the present
stage in the development of social scientific theory, it would, on the
whole, seem far wiser to concentrate research efforts on typologies based
on specific district attributes before attempting to deal with the rather
vague notions suggested by a "way of life" approach.60
In the second place, the two alternatives are not as different
as might at f rst appear to be the case. It is possible that certain district
attributes interact in such a way that their combined political effect is
different from the additive effect of the attributes considered by themselves.
For example, the effect of income on liberalism might depend in part upon
level of education. Hence, particular combinations of characteristics
might produce political cultures quite different from those predicted by an
additive model. Again, however, it would seem advisable to begin with
an attribute approach to areal typology construction. If one can arrive at
groupings which are homogeneous in terms-of their attributes, one will
have then isolated combinations of characteristics which occur frequently
and which are thus most likely to be the source of important interaction
effects. Major discrepancies between actual political behavior and that
predicted by an additive model might then provide clues as to the nature
of such interactions.
In general, then, the position taken here will be that the "way
of life" argument is at best somewhat premature. This position, however,
must admit of at least one exception. There is considerable evidence to
support the proposition that the Southeastern portion of the United States
possesses unique cultural traditions growing out of and associated with a
past history of slavery, devastation and defeat in the Civil War, and re-
construction after that war. This historical factor would be very difficult
to measure throughh contemporary Census data and in fact may grossly dis-
tort the significance of much of these data. Thus, due to sectional atti-
tudes, legal and illegal restrictions on political participation, and the one-
party system, all deeply rooted in the South's history, "per cent Black"
has in the South a political meaning which is in many ways diametrically
opposite to its meaning outside the South. Similarly, organized labor is
even weaker in the South than its numbers would indicate and, in general,
Southern politics is felt by many expert observers to be more conservative
than would be predicted solely on the basis of social, economic, and
demographic attributes. 61
On the other hand, the politics of the South is, of course,
affected by its contemporary characteristics as well as by its historical
traditions, and it is important to distinguish to the greatest extent possible
between the two influences. In order to take both into account, the anal-
ysis in the following two chapters will include, along with the study of
district: attributes and district types derived from these attributes, an
attempt to measure the impact of Southern geographic location on the be-
havior of Congressmen and their constituents.
One other sectional variable will also be included in part of
this study, not because it is as firmly grounded theoretically or empirically
as is the Southern hypothesis, but because it has for so long been a matter
of controversy. In the folklore of the discipline considerable attention has
been given to the existence of "Midwvestern isolationism. Therefore,
when patterns of roll call voting on foreign policy are analyzed in Chapter
V, an effort w~ill be made to assess the effect of Midwestern location.
NOTES FOR CHAPTER III
Heinz Eulau and Katherine Hinckley, "Legislative Institu-
tions and Proces ses ," in James A. Robinson, ed., Political Science
Annual, I (Indianapolis, Ind.: Bobbs-Merrill, 1966).
See, for example, Dye, "A Comparison of Constituency
Infl ue nce s"; Leroy N. Rieselbach, "The Basis of Isolationist Behavior, "
Public Opin on Quarterly, XXIV (W~inter, 1960), 645-657; Murray Clark
Havens, Ciier Versus Farm (University, Ala.: Bureau of Public Adminis-
tration, University of Alabama, 1957); Andrain, "Ai Scale Analysis";
Wilder Crane, Jr., "A Caveat on Roll-Call Voting Studies of Party Voting, "
Midwest Jou!rnal of Political Science, IV (August, 1960), 237-249;
Sha nnon Party, Constiuenc and Congressional Votng Flinn, Party
Responsibil:Ly in the States"; Vanderslik, "Constituency and Roll Call
For example, Havens, "Metropolitan Areas and Congress";
Le Bla nc "C oting in State Senates "; Gray and Gregory, Military Spend-
ing and Senrate Voting"; Wrolfinger and Heifetz, Power in Congre s s";
Ira Shark-ansky, "Voting Behavior of Metropolitan Congressmen: Prospects
for Changes with Reapportionment, Journal o~f Politics, XXVIII (November,
1966), 774-793; Thomas A. Flinn, "The Outline of Ohio Politics, "
We stern Poltia Quarterly, XIII (September, 1960) 702-721. See also
Julius Turner, Pa ty and Constituency (Baltimore: Johns Hopkins Press,
1951), p. 74n' George L. Grassmuck, Sectional Biases in Congress on
Foreign Policy (Baltimore: Johns Hopkins Press, 1951), p. 109n.
Gray, "Coalition of Southern Democrats and Northern
Republicans, p. 33; William C. Havard and Loren P. Beth, The Politics
of Misrepresentation: Rural-Urban C~onlflit in thLe Floid Legislature
(Baton Rouge: Louisiana State University Press, 1962), p. 15.
Robert S. Friedman, "The Urban-Rural Conflict Revisited, "
Western Political Quarely, XIV (June, 1961), 486. For similar classifi-
cation schemes, see Frank M. Bryan, "The Metamorphosis of a Rural
Legi sla ture Polity, I (Winter, 1968), 193n; Claris McDonald Davis,
Legislative Malapportionment and _Roll-Call Voting in Texas: 1961-1963
(Austin: Institute of Public Affairs, University of Texas, 1965), pp. 18-19.
Smith Isola tio ni st Voti ng p. 340.
V. O. Key, Jr., Public Opinion and American Democracy
(New Yorkc: Alfred A. Knopf, 1961), pp. 283-286.
Congressional Quarterly Almanac 1956 (Washington, D. C.:
Congressional Quarterly Service, 1956), p. 788.
Congressional Ouarterly Weekly Rogort XX' (February 2, 1962),
pp. 153-169, and XX~II (August 21, 1964), pp. 1784-1798; Hacker, Con-
_gressional Districtins, pp. 80-83; Leo M. Snowiss, "'Congressional1
Recruitment a~nd Repre senta tion American Political Science Review, LX
(September. 1966), 628-634; Sheldon Goldman, Roll-all Behavior in the
Massachuse'ts House of Representatives: A Test o~f Selected Hypotheses
(Amherst: Eu~reau of Governmental Riesearch, Universi ty of Ma ssa chu setts ,
1.968), pn. 30-31, 38n.
For example, Gra ssmuck Sectional Biases in Congress;
Rieselbach, The Roots of Isolationism, pp. 106-114; Ralph H. Smuckler,
"The Region of Isolationism, American Political Science Review, XLVII
(1954), 386-401; Gerald Marwell, "Party, Region, and the Dimensions
of Conflict in the House of Representatives, 1949-1954, American
Political Science Review, LXI (June, 1967), 380-399; George Robert
Boynton, Southern Conservatism: Constituency Opinion and Congress-
ional Voting, Public OpinionQuarterly, XXIX (Summer, 1965), 259-
269; Leroy N. Rieselbach, The Ba si s of Isolationi st Behavior, Public
Opinion Qurtrl, XXIV (Winter, 1960), 645-657.
One early study, for example, employed nine sectional
categories. See, Hannah Grace Roach, "Sectionalism in Congress
(180-190) "American Political Science Review, I Ags,12)
See, for example, Flinn, "The Outline of Ohio Politics";
Friedman, "The Urban-Rural Conflict Revisited"; Havard and Beth,
The Politics o~f Misrepresentation; Havens, CityVerssFa~lrm.
See, for example, Derge, Metropolitan and Outstate
Alignments"; Derge, Urba n- Rural1 Conflict "; Friedman, "The Urban-
Rural Conflict Revisited"; Havens, CityVersus Farm; Davis, Legislative
Malapportionment and Roll Call Votig
Even here, moreover, it is possible to counter these negative
findings with other studies that have revealed urban-rural differences. See,
for example, Murray C. Havens, "Metropolitan Areas and Congress";
Sharkansky, "Voting Behavior of Metropolitan Congressmen"; Julius Turner,
Party and Constituency, chap. iv.
For attempts to deal with this problem, see C~ongesioa
Quarterly Weekly Report XX (February 2, 1962), 153-169, and XXII
(August 21, 1964), 1784-1798; Hacker, Congressional Districting,
pp. 80 ff.
Jeffrey K. Hadden, "Use of Ad Hoc Definitions, in
Edgar F. Borgatta, ed., Sociological Methodology 1969 (San Francisec:
Jossey~-Bass, 1969), pp. 277-278.
Richard Dewey The Rural- UJrban Continiuum: Real But-
Relatively ULnimportant, Ame~~orican ora of Sociology, LXVI
(July, 1960), pp. 60-61.
Richard A. Kurtz and Joanne B. Eicher, "Fringe and Suburb:
A Confusion of Concepts, Social Force s, XXXVII (October, 1958), pp. 32-37.
Henry S. Shryock, Jr., Population Redis tribution Within
Metropolitan Areas: Evaluation of Research," Social Forces, XXXV
(December, 1956), p. 155 (emphasis in original). These sentiments
are echoed in a more recent article by Hadden, who comments that "the
concept suburb remains largely undefined and at best ambiguously con-
ceptualized and concludes that "in short, ecologists have contrived a
concept that is operationally convenient, but its meaning is conceptually
questionable." See his "Use of Ad Hoc Definitions, pp. 277, 280
(emphasis in original) .
Smith, "Isolationist Voting, pp. 367-368.
Philip M. Hauser, "Observations on the Urban-Folk
Urban-Rural Dichotomies as Forms of Western Ethnocentrism, in
M. Hauser and Leo F. Schnore, eds., The Stuyo~f Urbanization
York: Wiley, 1965), p. 514.
Eulau and Hinckley, "Legislative Institutions and Processes. "
Triedman, "The Urban-Rural Conflict Revisited, pp. 485;,
Smith, "Isolationist Voting, p. 340.
V. O. Key, American State Politics (New York: Knopf 1956),
Noel P. Gist and Sylvia rleis Fava, Urban Societ, 5th ed.
Thomnas Y. Crowell Com~pany, 1964), p. 40.
"The Urban-Ruel~ Conflict Revisited, p. 481.
Gerald Marwell, "Party, Recion, and the Dimensions of
Con fli ct ,
Turner, Party and Constiuency p. 128.
A few users of traditional areal typologies do take pains to
demonstrate that their types in fact differ in their characteristics. See
Eulau, "The Ecological Basis of Party Systems, pp. 128-129; Snowi ss ,
" Congressional Recruitment and Repre sentation p. 629.
Stuart A. Rice, "The Identification of Blocs in Small
Political Bodies, American Political Science Review XXI (August,
For more recent applications, see: David Bicknell Truman,
The Congressional Party: A CaseStudy (New York: Wiley, 1959); Leroy
N. Rieselba~ch, "Quantitative Techniques for Studying Voting Behavior in
the United Nations General Assembly, International Organization, XIV
(Spring, 1960), 291-306; Arend Lijphart, "The Analysis of Bloc Voting
in the General Assembly, American Political Science Review LVII
(December, 1963), 902-917; Bryan, "The Metamorphosis of a Rural
Legislature. See also Duncan MacRae, Jr., Issues and Paris in
Legislative Voting: Methods o~f Statistical Analysis (New York:
Harper &( Row, 1970).
For a non-technical introduction, see Robert R. Sokal,
"Numerical Taxonomy Scientific American, CCXV (December 196 6) ,
Grumm, "The Systematic Analysis of Blocs."
Forrest E. Clements, "Use of Cluster Analysis with
Anthropological Data, American Anthrpoloist LVI (April, 1954),
Joel Smith, "A Method for Classification of Areas on the
Basis of Demographically Homogeneous Populations, American Sociolo-
gical Review (April, 1954), 201-207. See also Peter N~orman, "'Third]
Survey of London Life and Labor: A New Typology of London Districts, "
in Mattai Dogan and Stein Rokkan, eds., Quantitav Ecclogical Anaysis
in the Social Sciences (Cambridge: Massachusetts Institute of Technclogy
Pre s s, 1969), pp 371-396.
Robert C. Tryon, Identification o~f Social Areas byr Cluster
Analysis: A General Method with an Application t~o the San Fracic
Bay Area, University of California Publications in Psychology, VII
(1955), 1-100. Tryon's work is closely related to the better known Shevky-
Bell approach. See Eshref Shevky and Wendell Bell, Social Area Analysis:
Ther, Illustrative Appiato and Comoutational Procedures (Stanford:
Stanford University Press, 195 5) .
C. A. Moser and Wolf Scott, British Towns (Edinburgh:
Oliver & Boyd, Ltd., 1961); Richard L. Forstall, "A New Social and
Economic Grouping of Cities, The Municipal Yearbook 1970, pp. 102-
170; Lawrence R. Alschuler, "Political Participation and Urbanization in
Mexico" (Ph.D. dissertation, Northwestern University, 1967), chap.
ix; Kenneth J. Jones and Wyatt C. Jones, "Toward a Typology of American
Cities, Journal o~f Regional Science, X (1970), 217-224.
D. M. Ray and Brian J. L. Berry, "Multivariate Socio-
Economic Regionalization: A Pilot Study in Central Canada, in T. Rymes
and S. Ostry, eds., Regional Statistical Studies (Toronto: University
of Toronto Press, 1965), pp. 1-48.
Margaret J. Hagood, "Statistical Methods for Delineation
of Regions Applied to Data on Agriculture and Population, Social forces,
XI(March, 1943), 287-297; Richard Lee Sutton, "Level of Develop-
ment of State Environment and the Structure and Activities of State Plan-
ning and Development Organizations (Ph.D. dissertation, University
of North Carolina, 1970), chap. ii.
Bruce M. Russett, International Regions and the International
System: A Studyi P~olitical Ecology (Chicago: Rand McNally & Company,
1967), chaps. i-iii; A~rthur S.Banks and Thillip M. Gregg, Grouping
Political Systems: Q-factor Analysis of a Cross-Polity Survey, Am eri ca n
Behavioral Scientist, IX (November, 1960), 3-6.
The only exceptions that th-s writer was able to find involved
operationally defining concepts such as uIrbanism in terms of a single attri-
bute, or through simple cros s- cla ssificrrtion of at most two or three attri -
bu te s. The best example is probably Edgar Litt, The Political Cultures
of Massachusetts (Cambridge: Massachusetts Institute of Technology
Press, 1965). Litt's typology has also beren used by Goldman, who pro-
vides from personal correspondence with 11tt the operational definitions
used by the latter. See Goldman, Roll C~all Behavior in Massachusetts,
pp. 31-33, 70-74.
In general form, though not in all specifics, the outline
followed here adheres closely to that set forth in Ray and Berry, "Multi-
variate Socio-Economic Regionralization. "
Cf. Vincent, "Predicting Voting Patterns, pp. 476-479.
This is done so that the variance of the components will be
proportionate to their respective eigenvalues, that is, to the proportion
of variance in the original data matrix which they explain. Square roots
are used since multiplying a set of numbers by a constant produces a
new set of numbers with a variance equal to the variance of the original
set times the square of the constant.
For more detailed descriptions of Ward's Method see Joe
H. Ward, Jr., "Hierarchical Grouping to Optimize an Objective Function,"
Journal o~f the American Statistical Association, LVIII (March, 1963),
236-244; David Wishart, "An Algorithm for Hierarchical Classifications, "
Biometrics, XX'V (March, 1969) 165-170. For application of the method
see Joe H. Ward, Jr. and Marion Hook, "Application of an Hierarchical
Grouping Procedure to a Problem of Grouping Profiles, Educational and
P sy chol ogi cal M ea surement XXIII (Spring, 1963), 68-81; Ru s sett ,
International Regions and the International System pp. 49-58; Robert A.
Young, "A Classification of Nations According to Foreign Policy Outputs, "
prepared for delivery at the sixty-sixth annual Meeting of the American
Political Science Association, Biltmore Hotel, Los Angeles, California,
September 8-12, 1970. A very similar algorithm is employed in Ray and
Berry, "Multivariate Socio-Economic Regionalization. "
Of the studies cited in the previous footnote, Young's
study, which involves eighty-three nations, is the largest in terms of
the number of objects grouped.
The version of the program employed in the present stud; is
dimensioned for 1,000 cases and nine objects, and requires about 90,000
bytes of core storage. The program took 1.02 minutes of Central Process-
ing Unit time of the University of rlorida'; IBM 360-65 to cluster thie data
described in the text.
This criticism is advanced by Russett, International Regions
and the International System, p. 49.
Similar procedures for refining the results of hierarchical
clustering have been suggested by Ray and Berry, "Multivariate Socio-
Economic Regionalization" and by Brian J. L. Berry, "A Synthesis of
Formal and Functional Regions Using a General Field Theory of Special
Behavior,"in Brian J. L. Berry and Duane F. Marble, eds., Spacial
Analysis (Englewood Cliffs, N. J.: Prentice-Hall, 1967), pp. 419-428.
See also Norman, "Third Survey of London Life and Labor" and Sutton,
"Level of Development of State Environment. "
Linton C. Freeman, Elementary Applied Statistics (New
York: John Wriley and Sons, 1965), chap. xi.
In this typology, "urban" districts are those located pre-
dominantly in the center cities of Urbanized Areas plus thirteen satellite
cities with a population in excess of 100,000. Suburban" districts are
those located predominantly in the urban fringes of such center cities.
Rural districts are those predominantly located outside of Urbanized Areas.
Districts not predominantly located in any one type of area are classified
as "Mixed. "
The sectional breakdown made in this section is as follows:
(1) Northeast: Connecticut, Maine, Massachusetts, New Hampshire,
New Jersey, New York, Pennsylvania, Rhode Island, Vermont.
(4) Midwe st: Illinois, Indiana, Iowa, Kansas, Mrichigan, Minnesota,
Nebraska, North Dakota, Ohio, South Dakota, Wisconsin.
(2) Border: Delaware, Kentucky, Maryland, Missouri, Oklahoma,
We st Virgi nia .
(5) West: Alaska, Arizona, California, Colorado, Hawaii, Idaho,
Montana, Nevada, New Mexico, Oregon, Utah, washington, Wyoming.
(3) Sou th : Alabama, Arkansas, Florida, Georgia, Louisiana, Mississippi,
North Carolina, South Carolina, Tennessee, Texas, Virginia.
U. S. Bureau of the Census, Congressional Disrit Datca
Book: Districts of the 88th Congress; Congressional Quately Weekil
Report (August 21, 1964), pp. 1784-179Y.
Type I bears a loose correspondence to "mid-urban" cate-
gories in typologies developed in Hacker, Congressional Districting,
pp. 80-83 and in Congressional Quarterly Almanac 1956 p. 788.
Louis Wirth, Urbanism as a Way of Life, American
Journal of Sociology, XLIV (July, 1938),124
The only two districts in Type III not classified urban by the
Congressional Quarterly Service are in northern New Jersey: the eleventh
(Newark and vicinity) and the fourteenth (Jersey City).
Of the variables measuring socioeconomic status, the type
differs most from the national average in proportion of low income families.
This is, moreover, probably not a highly reliable statistic in this case
since it does not take into account the value of producer-consumed farm
This comparison is perhaps a bit unfair in the case of the
Congressional Quarterly typology, since the "mixed" category used here
actually combines several subcategories in the original typology.
Cf. Herbert J. Gans, "Urbanism and Suburbanism as Ways
of Life: A Reevaluation of Definitions, in Arnold Rose, ed., Human
Behavior and Social Processes (Boston: Houghton Mifflin Company,
1962), pp. 625-648.
V. O. Key, Southern Politics in State and Nation (New
York: Vintage Books, 1949).
LEGISLATIVE INPUTS: CONSTITUENCY AND
THE SELECTION OF LEGISLATORS
In this chapter, an attempt will be made to study constituency
environment with respect to selected aspects of the input side of the
legislative process. The first portion of the chapter will examine the
question of malapportionment in the Eighty-oighth Congress. The re-
mainder of the chapter will be devoted to analysis, at the aggregate
district level, of voter behavior in the 1962 congressional elections.
Attention will focus on the questions of voter turnout, voter party pre-
ference, and inter-party competition.
On February 17, 1962, Mr. Justice Hugo Black, delivering the
United States Supreme Court's decision in the case of Wesberry v. Sanders,
declared it to be a requirement of the Federal Constitution that, "as nearly
as is practicable one man's vote in a congressional election is to be
worth as much as another's."1 On this date, the Eighty-eighth Congress
was in its second session. The district data for this Congress, therefore,
provide an opportunity for examining patterns of malapportionment at the
time that the Court's historic decision was rendered.
If politics is definable as "the authoritative allocation of values
for society, then in studying the politics of malapportionment, it is
important to determine as methodically as possible just which segments of
society receive other than a proportionate share in the "value" of represen-
tation. In the literature on malapportionment, discussion has usually been
cast in terms of areal typologies. Researchers have described the over-
representation of rural areas at the expense of the cities which existed at
the time of the Court's reapportionment decisions, and have debated the
possible impact or lack of impact of reapportionment on urban-rural con-
flict.3 A number of observers have also described suburban under-represen-
tation, noting that the suburbs, rather than the cities, were the most
severely under-represented in Congress and in state legislatures.$
Since these classifications are, as argued in the preceding
chapter, at best rather imprecise, the above statements about them conv-y
only rather impressionistic information about the kinds of districts which
were under- or over-represented, or about the characteristics of their in-
habi ta nts Employing the district typology developed in the third chapter,
it should be possible to come closer to this goal than has been done through
use of traditional modes of classification. Thus an attempt will be made to
determine how well each of the five constituency types was represented in
the Eighty-eighth Congress in proportion to its population.
While an examination of inter-type differences will reveal more
precisely which kinds of districts received a disproportionate share of
representation, it will uncover only in a general way the constituency
attributes associated with malapportionment. The relationship of constit-
uency attributes to malapportionment is a matter that with some exceptions
seems to have received relatively little attention in previous research. In
the present work, malapportionment will be correlated with the six constit-
uency dimensions isolated in Chapter II and with the nineteen social, eco-
nomic, and demographic characteristics from which they were derived.
The goal of this section is rather modest in that it will be con-
fined to description rather than causal explanation. Correlations between
malapportionment and the constituency attributes included in this study are
quite likely to be "spurious" in terms of causal inference. For one thing,
much of the .nequality that existed among congressional districts in 1962 is
traceable to secular shifts in population concentration which are only
indirectly related to most of the variables included in this study. For
example, a finding that districts with high proportions of Catholic inhabi-
tants were ur~der-represented would not necessarily mean that state legis-
latures were guilty of purposeful religious discrimination, but more probably
would be a tunction of population stability over a long period of time in
heavily Protestant area s. Secondly, there is the problem of levels of
analysis. It is important to distinguish the kind of inquiry conducted here
from analysis in the field of comparative state politics of relationships be-
tween statewide attributes and severity of malapportionment in state legis-
latures.7 In those studies the units to which malapportionment scores are
assigned (that is, the states themselves) are the same as the units charged
with the constitutional responsibility for carrying out legislative districting.
When, as in the present study, the unit of'analysis is the individual con-
stituency, the situation is quite different. The adequacy of a given dis-
trict's representation depends primarily not on the behavior of its inhabi-
tants or their representatives, but upon the action or inaction of the legis-
lature of the state in which the district is located. Analysis of individual
constituencies' characteristics, therefore, can show which elements of
the population are discriminated against, but cannot provide a great deal
of information about the causes of such discrimination. Hence causal in-
ferences, and techniques appropriate to causal inference such as regression
analysis, will not be attempted; the objective pursued will be simply to
describe as iully as possible the relative degree of representation in Con-
gress enjoyed by various segments of society.
Apportionment is a many-faceted problem that could be approached
from several perspectives: gerrymandering, multiple-member versus single-
member districting, and so on. However, while recognizing that it is not
the only, no! even the most important, aspect of malapportionment, the
present inqu:ry will be confined solely to analysis of purely numerical in-
equality in repre sentation .
The index of malapportionment used will be the ratio of each
district's population to the average population of all 435 districts. Thus
the higher the index score the less well represented the district, with a
score of one indicating representation proportionate to population. In the
case of states which in the Eighty-eighth Congress elected some or all of
their representatives at large, it is necessary to modify this ratio in order
that citizens of such states not artifically-appear to be under-represented.
This adjustment is quite straightforward for the at-Jarge districts them-
selves. The population of such a district is simply divided by the total
number of seats allowed to the state in which the district is located. The
malapportionment index can then be computed. For non-at-large districts,
the modification is only a bit more complex. In such cases the population
of the district is adjusted by dividing it by the fraction of seats in the
district's state which are not elected at large.
A number of objections could be voiced against this method of
m ea during r ;re senta tion For one thing, some inequality wvill be produced
simply as a !esult of the fact the United States Constitution requires that
districts be apportioned by state and that each state receive at least one
re press e nta ti ,re. On the other hand, except for the inequality just cited,
it is obvionc that no numerical inequality will exist among at-large dis-
tricts.B Mo-e generally, when district types are compared, no inequality
will appear within a state in which all districts are of the same type. An
examination of the appendix will reveal that a large number of states are
composed entirely or predominantly of a single type of congressional dis-
The procedure adopted here seems defensible in view of the
descriptive nature of the present inquiry. The goal of this section is to
set forth the extent and nature of inter-constituency differences in represen-
tation, regardless of the factors causing or limiting such differences.
Needless to say, any attempt to go beyond this goal and to carry out a
causal analysis of malapportionment would have to begin by controlling
for essentially trivial causal factors such as outlined above.
Table 6 shows, for each district type, the mean malapportion-
ment score and the standard deviation of that score within the type. The
table gives the same information for all 435 districts taken together. At
the bottom of the table, the n12 coefficient is given. As noted in the pre-
vious chapter, this coefficient, also k~nowvn as the squared correlation
ratio, is computed by dividing between-group variance by total variance,
thus measuring the proportion of variance in a dependent variable (in this
case, malapportionment) explained by a nominal classification scheme.
It should be pointed out that the value of the coefficient depends in par',
on the number of categories in the independent variable. Hence the ratio
set forth here applies only to the five-fold typology selected for analysis.
Had a different number of district types been selected, a different ratio
would of course have been obtained.
The table indicates that the most under-represented type is the
Prosperous-growth district, with a mean population which is 15 per cent
above the national average. The most over-represented type is the Un-
developed district, with a mean population 12 per cent under the national
average. Population Center districts and Undeveloped-deprived districts
are also over-represented, though to a lesser degree, while Manufacturing
districts are slightly under-represented.
Another way of looking at these data is to compute the number of
seats in Congress that each type would have if each were assigned seats
in proportion to its total population (as adjusted above), and to compare
this figure with the number of seats actually held by each type.
MVALAPPORTIONMENT SCORES, BY DISTRICT TYPE
Manu- Pros- Popu- Unde- Unde- All
fa ctur- perous- nation veloped veloped- Dis-
ing Growth Center Deprived tricts
1.03 1.15 .93 .88 .95 1.00
0 .21 .28 .17 .19 .22 .23
REPRESENTATION STRENGTH OF~ DISTRICT TYPES
ACTUAL VERSUS PROPORTIONATE
Manu- Pros- Popu- Unde- Unde-
factur- perous- Jation veloped veloped-
ing Gr owth Center Deprived
No. of seats
Actual 136 80 42 94 83
Proportionate 140 93 39 84 79
Gain or Loss +4 +13 -3 -10 -4
CORRELATIONS OF MALAPPORTIONMENT SCORES
WITH CONSTITUENCY ATTRIBUTES
Home Ownership .05
B. INDIVIDUAL CHARACTERISTICS
Pop. Gro. 29
Density -. 02
For. Stk. .06
Priv. Ed. .14
Coll. Ed. .31
Low Inc. -.28
High Inc. .25
Wht. Col. .35
Own. Occ. .08
This information is shown in Table 7. The table indicates that a net total
of seventeen changes in group strength would have resulted from such a
hypothetical reapportionment. Most of those changes would have taken
the form of gains by Prosperous-growth districts (with thirteen additional
seats) at the expense of Undeveloped districts (which would have lost ten
seats). Gains and losses among the other types would have been relatively
It should be noted that the differences discussed so far are
generally not very dramatic. One widely accepted guideline for judging
equality in representation suggests that deviation in population of up to
15 per cent from the mean district population might be considered accept-
able. Using this rule of thumb, it can be seen from Table 6 that, despite
very great disparities in representation am ang many individual districts,
only one type as a whole deviates from the mean by as much as 15 per
cent, and just barely so. Table 7 shows t.iat, while the actual strength
in Congress of two district types differs by a fairly sizable margin from
what might be expected on the basis of population, relatively few seats
overall would have changed hands among types under a "one man-one
vote" apportionment formula.
An even clearer indication of the relative lack of inter-type
differences in malapportionment is the low n12 coefficient of .16 shown in
Table 6. This figure, which is analagous to a Pearson r of .40 for
associations between interval level measures, indicates that only 16
per cent of the variance in malapportionment is inter-type variance.
This is not to say, of course, that the inequalities shown in the tables
are not politically significant. Certainly the over-representation of Un-
developed districts and the lack of proportionate representation of Pros-
perous-grow~th areas might easily be so classified. The point being made
here is simply that, as a whole, the constituency typology does not
sharply differentiate districts with respect to malapportionment.
In part this lack of differentiation among groups is due to the
fact that, as noted earlier, many states are comprised predominantly or
entirely of one type of district. To the extent that states are internally
homogeneous with respect to district type, variance in malapportionment
will of necessity be within-group, except for that malapportionment re-
sulting from constitutionally mandated apportioning procedures.
Another reason for the relatively small inter-type differences
is revealed by the data in Table 8. This table shows, for each constitute ncy
factor and individual characteristic, the zero-order correlation of that
factor or characteristic with malapportionment. Al1 of the correlations
shown are low or at most of moderate size, with the highest being only .35.
This finding of relatively slight associations between malap-
portionment and a variety of social, economic, and demographic measures
has important normative implications. Much of the drive for more equal
apportionment during the 1950s and early 1960s was based on the assumption
that malapportionment served to discriminate against certain segments of
society by denying them a fair share of legislative representation. The
data presented here suggest that, at least for the districts of the Eighty-
eighth Congress, this is true to only a very limited degree. Whatever the
inequalities suffered by some districts, in the Congress as a whole no
truly striking differences occur, whether constituency is described in
terms of types, attribute dimensions, or individual attributes. Such a
finding complements the conclusions of a number of researchers, at both
the national and state levels, who have treated malapportionment as an
independent variable and have found it not to be an important causal factor
in determining outputs of the political system.
Such associations as are found in Table 8 show that it is the
more affluent districts, that is, those whose residents might in other
respects be expected to possess the greatest potential political power,
who fair worst in terms of representation in Congress. Conversely, dis -
tricts with high proportions of low income families and/or high rates of
unemployment tend to fair well. This pattern has been noted by other ob-
servers, and has sometimes been used as a defense for malapportionmen'.
practices.12 Districts with proportionately high numbers of elderly resi-
dents also tend to be well represented.
The table also indicates that population growth has one of the
higher correlations shown. This is hardly surprising since much of mal-
apportionment was the result of inertia on the part of state legislative
bodies which failed to redistrict congressional delegations despite pop-
ulation shifts. The fact that the correlation is not considerably higher
than it is suggests that in examining the roots of malapportionment one
must look for relatively early migration and natural growth patterns not
closely associated with those occurring during the 1950s.
The figures in Tables 6 through 8 suggest the difficulty of ex-
plaining patterns of malapportionment in terms of traditional urbanism con-
cepts. This is noteworthy since so much of the literature on malapportion-
ment has been cast in terms of urban-rural conflict. Agriculture is directly
related to over-representation but the correlation is not high. Other at-
tributes that might be considered to be conceptually related to urbanism
or ruralism showY even lower associations with malapportionment. Compari-
sons among district types lead to similar conclusions. Undeveloped dis-
tricts, which, as noted in Chapter III, constitute the most "rural" district
type, are the most over-represented. The second most over-represented
type, however, is the Population Center, which, as noted earlier, is mos~t
closely related to the classic definition of urbanism. On the other hanc',
those researchers who have viewed malapportionment in terms of "subur;,an"
rather than "urban" under-representation come much closer to accurately
portraying the situation found here, i.e., under-representation is relate 1
to high socioeconomic states and to districts in the Prosperous-growth
The remainder of this chapter will be devoted to an examination
of the behavior of the electorate in the 1962 congressional elections.
First, an analysis of levels of political participation will be made. Once
this is completed, a study of patterns of party preference by voters and of
inter-party competition will be undertaken.
While level of political participation and patterns of partisan-
ship will be examined separately, it should be kept in mind that the two
are interrelated If certain segments of the potential electorate are dis-
franchised, or fail to exercise their franchise, the relative strength of
the Republican and Democratic parties, and hence also the degree of inter-
party competition, may well be altered. Conversely, it is likely that voter
turnout will be directly related to the perceived closeness of electoral out-
comes. These considerations should be kept in mind especially with re-
spect to analysis of electoral behavior in the South, where much of the
Black population has been effectively disfranchised, where voter turnout
even among whites is typically very low, and were inter-party competition
has generally been Jacking.
Previous research, at both the individual and aggregate levels,
has demonstrated that differential levels c; participation are associated
with social, economic, and demographic variables similar or identical to
those included in the present study. Researrch has also shown that turnout
is higher in "urban" and metropolitann" areas than in "rural" places, and
considerable regional variation in turnout has also been noted. The present
analysis will seek to add to this body of researchl4 by relating level of
turnout in the 1962 congressional elections to constituency dimensions and
to individual constituency attributes, by comparing turnout in each of the
five district types, and by examining the possible effect upon turnout of
Southern political culture.
Table 9 shows the mean and standard deviation in voter turnout
for each district type in the 1962 congressional elections. The clearest
difference shown is that between Undeveloped-deprived districts and the