CONGRESSIONAL. ELECTION PATTERNS- PARTY TRENES ANfl DE§TRECT RANK ORDER. RELATION$HWS, 1942-1956 Thesis for Hm Degree of M. A. MICHiGAN STATE UNIVERSITY Richard Lester Miiler 19 61 TH ESIS LIBRARY Michigan State University h¥4+._ _ _.._._____ ABSTRACT CONGRESSIONAL ELECTION PATTERNS--PARTY TRENDS AND DISTRICT RANK ORDER RELATIONSHIPS, 1942-1956 by Richard Lester Miller The study of congressional elections in the United States is a fruitful source of political information that few persons have chosen to investigate. For the most part congressional district elections have been treated as a by-product of the more glamorous Presidential elections. This investigation focused on the congressional election as a significant event in itself. The problem of locating and identifying the elements of partisan stability and variability in a series of congressional elections involved dealing with pertinent sub-problems. Analysis of aggregate election data for the period 1942- 1956 included a determination of the stability of partisan rank order of congressional districts, the identification of districts exhibiting low rank stability, and an examination of the cur- rently popular concept of “marginality” as a predictive device. Further, the problem of forecasting congressional election outcomes was investigated by experimenting with a new forecasting method. The basis data for this study consisted of votes cast for the major party congressional candidates in the United States for the period 1942 through 1956. These statistics were converted into index numbers, Stalemate Indexes, to facilitate subsequent statistical Operations. The Stalemate Index was defined as one-half the difference between the major party percentages of the total vote cast in a congressional dis- trict election. Abstract Richard Lester Miller The non-parametric rank correlation technique was employed to test congressional district rank stability. In addition, a new three-factor concept of political ”marginality" was proposed. Using this new concept, an experimental forecast of congressional election results was accom- plished by the projection of district partisan trend lines. Among the major findings of this study was the discovery that the concept of “marginality” currently accepted as a forecasting base is in- efficient. This concept, defined as the percentage spread between the two major party candidates at the last election, failed to forecast one— third of all party turnover cases that occurred where not expected. It failed to forecast at least 65% of non-party turnover cases where turn- over was expected. It was found that partisan rank orders of congressional districts are highly correlated between successive elections. Rank realignments occurred slowly over time, and no radical realignments were found even where an election produced a radical change in the partisan division of the House of Representatives. The Presidential year elections produced rank orders that closely reflected the rank ordersof the preceding mid- term elections. Another major finding was that those few districts which are largely reSponsible for a less than perfect rank correlation could have been predicted as liable to experience a large rank change prior to the election in which the change occurred. A projection of the 1948, 1950, 1952 and 1954 Stalemate Indexes of competitive congressional districts, fitted to a matrix of forecasted partisan rank positions, was accomplished. The projection was made by the method of linear regression. It produced a forecast of 1956 congressional election results that closely paralleled the actual results of the 1956 election. CONGRESSIONAL ELECTION PATTERNS--PARTY TRENDS AND DISTRICT RANK ORDER RELATIONSHIPS, 1942-1956 BY Richard Lester Miller A THESIS Submitted to the College of Business and Public Service of Michigan State University in partial fulfillment of the requirements for the degree of MASTER OF ARTS Department of Political Science 1961 S\~ ~ I 5/}.1/ 4 ACKNOWLEDGMENTS Through the Special efforts and indulgences of two people, the completion of this work has been accomplished. The writer is especially grateful to Dr. Ralph M. Goldman for his fatherly guidance at every stage in the development of this thesis. To my wife, Dee, whose contributions have been comparable to co-authorship, a belated but deserved acknowledgment for the sacrifices made in the production of these pages. >:< 3:: ::< >:< >:< >:< >2: >:< >1: >:< >:: ):< ii TABLE OF CONTENTS CHAPTER Page I. INTRODUCTION. . . . ................ 1 II. TECHNIQUES OF DATA GATHERING AND ORGANIZATION .................. 6 Selection of the Period for Study ..... . . . . 6 Structural Aspects of Congressional Districting . 8 Problems of Collecting Election Data . . ..... 13 Statistical Tools .................. 16 III. MARGINALITY AND THE SINGLE ELECTION FORECASTER................... 19 Defining Marginality ............ . . . . 20 Inadequacy of the Single-Election Forecaster . . . 26 Statistical Test on “Marginal" Districts ...... 36 Summary . q. . ............. . ..... 40 IV. COMPETITIVE DISTRICT RANK ORDER RELATION- SHIPS, 1942-1956 ................. 43 Hypothesis of Rank-Order Stability ........ 43 Computing Rank Order Correlation Coefficients . 47 Important Realignments Of Congressional District Rank Orders ................... 50 Reduction Of Ranks to Locate "Competitive" Districts ..................... 55 Rural-Urban Character Of Post-Step 3 Districts . 58 Analysis of Rank Order Relationships Excluding the One-Party, Composite, and Redistricted Districts. . . . . ..... . . . . ....... 60 Test Of Permanence and Rate of Realignment for Post-Step 3 Districts . . . . . . . . . . . . . . 63 Analysis Of Election Data for Competitive Districts ...... . . . . . . ......... 65 iii TABLE OF CONTENTS - Continued CHAPTER Test of Permanence and Rate Of Realignment for "Competitive” Districts ............. Party Victory in "Competitive" Districts Conclusions ................ V. DISTRICTS RESPONSIBLE FOR LOW RANK CORRE- LATION COEFFICIENTS ............ Identification of Error Districts . . . . . ..... Factors Related to Error District Occurrence Summary ................... VI. AN EXERCISE IN CONGRESSIONAL ELECTION FORECASTING ........ Three-Factor Marginality ............. Forecasting the 1956 Rank Order . . . . ..... Estimating Stalemate Indexes by Linear Regression .......... . . . Forecasting the 1956 Stalemate Indexes . . . . . . Combining Rank Order and Stalemate Index Fore- casts With the New Marginal Concept Efficiency Of Forecast . . ..... . . . ..... Summary ........... . . . . Proposals for Further Research. . . . . ..... BIBLIOGRAPHY . . . . APPENDICES.............. ........ A. Congressional District Stalemate Indexes, 1942-1956 B. Stalemate Index ...... , ............. C 1. Districts Removed in Step 1. One-Party Districts 2. Districts Removed in Step 2. Composite Districts 3. Districts Remaining After Step 4. Competitive Districts . ................ . Districts Removed in Step 3 by 1956 Rural- Urban Status. Redistricted Districts . . . . ........ iv Page 67 69 75 77 77 78 94 96 97 99 99 101 107 110 115 117 120 122 123 136 139 140 140 141 TABLE I. II. III . IV. VI. VII. .VIII . IX. LIST OF TA BLES Distribution of Congressional District Election Out- comes According tO Size Of Stalemate Index Change Between Successive Presidential-Year Elections. Distribution of Congressional District Election Out- comes According to Size of Stalemate Index Change Between Successive Mid-term Elections ..... Distribution of Congressional District Election Cases According to Size Of Stalemate Index Change Where Turnover Occurred at Second of Two Elections of the SameType. ...... . . . . .......... . . Distribution Of Pairs of Congressional District Elect- ion Cases According to Size of Stalemate Index at the First of Two Successive Elections Where Turnover Occurred at Second Election .......... . . Distribution Of Cases in Which Turnover Occurred Where Stalemate Index at Preceding Election was GreaterThan4.9 ....... ‘. . . . . . . . . . Distribution Of Congressional District Election Cases According to Size of Stalemate Index at First of Two Successive Elections (Where Out-Party Wins Second Election) ...... Distribution of Congressional District Election Cases According to Size of Stalemate Index at First of Two Successive Elections (Where Incumbent Party Candi- date Wins Second Election). . . . . . . . . . Composition Of the U. S. House Of Representatives by Party Designation, 1942-1956 Election Returns . . Rank Correlation Coefficients for Pairs of Congress- ional Elections in Sequences of Alternate Elections Rank Correlat1on Coefficients for Pairs of Successive Congressional Elections. . . . . . . ....... Page 28 29 30 30 32 3'7 37 44 49 49 LIST OF TABLES - Continued TABLE XI. XII. XIII . XIV. XV. XVI. XVII. XVIII. XIX. XX. XXI. Page Comparison Of Rank Order Correlation Coefficients Produced by Ranking Of Total Universe of Districts and Those Produced After Step 3 Removal. .' Comparison of Rank Correlation Coefficient Changes for Pairs of Election Pairs--Between Coefficients for All Districts and Coefficients for the Step 3 Districts ..... . . . . . . .......... Rank Correlation Coefficients for Districts Remaining After Steps 1, 2, and 3--Using 1942 as Base Year . . Rank Correlation Coefficients of Districts Remaining After the Step 4 Reduction of Ranks--For Consecutive Elections................ ....... Rank Correlation Coefficients of 22 Competitive Dis- tricts Remaining After the Step 4 Reduction of Ranks-- Using 1942 as Base Year. . . ............. Comparison Of Actual, Study, and Step 4 Districts-- Percentage of Districts Won by Democratic Party. . . Distribution of Partisan Changes in 22 Competitive Districts by Character Of Change for the Period 1942'1956 o o o o, ooooooooooooo o o 0 Comparison of Changes in Percentage of Seats Held by Democrats-—Actual Districts, Study Districts, Step 4 Districts. . . . . . . . ..... . . . . . . . Comparison of Rank Correlation Coefficients and Changes in the Partisan Distribution of Step 4 Dis- tricts .......................... Ten Largest Contributors of Rank Differences Be- tween the Elections Of 1942 and 1944 ........ Ten Largest Contributors of Rank Differences Between the Elections Of 1944 and 1946 ...... vi 61 62 64 66 68 71 72 74 74 79 79 LIST OF TABLES - Continued TABLE XXII. XXIII. XXIV . XXV. XXVI. XXVII. XX VIII. XXIX . XXX. XXXI. XXXII. XXXIII. XXXIV. Ten Largest Contributors of Rank Differences Between the Elections of 1946 and 1948 ....... Ten Largest Contributors of Rank Differences Between the Elections of 1948 and 1950 ........ Ten Largest Contributors Of Rank Differences Between the Elections of 1950 and 1952. . Ten Largest Contributors of Rank Differences Between the Elections Of 1952 and 1954. . ..... Ten Largest Contributors Of Rank Differences Between the Elections of 1954 and 1956 ..... Comparison Of House Composition and Partisan Trend Of Error Districts for Pairs of Successive Elections, 1942-1956. . . ..... Distribution of Error District Election Cases Accord- ing to Size Of Stalemate Index at the First of Two Successive Elections ............ Distribution of Special Factors Explaining Error District Occurrence. . . . ........ Competitive Districts' Projected Stalemate Indexes for 1956 Fitted to 1954 Rank Order . . . . ..... Projected Stalemate Indexes for 1956, Fitted to the 1954 Rank Order (Pivot Districts). . . Comparison of Changes in Number of House Seats Won by Democratic Candidates ............. Projected Stalemate Indexes for 1956, Trend Slopes, and ”Marginal" Classifications (Classified Districts) Comparison of Actual and Forecasted Stalemate Indexes and Rank Positions for l956--Step 4, Pivot, and Classified Districts. vii Page 80 80 81 81 82 84 84 92 100 105 105 109 111 LIST OF TABLES - Continued TABLE Page XXXV. Distribution of Differences Between Actual and Fore-— casted Stalemate Indexes of Step 4, Pivot, and Classified Districts for 1956. . . . . . . . . . . . . . 114 XXXVI. Comparison Of Actual and Forecasted Election Results,1956......................115 viii LIST OF CONTINGENCY TABLES TABLE Page A. Pairs of Presidential-Year Congressional Elections. . . 38 B. Pairs Of Mid-Term Congressional Elections. . . . . . . 38 C. Pairs of Successive Congressional Elections . . . . . . 38 ix CHAPTER I INTRODUCTION Each Congressman must take his voting record and party label before his electorate every two years. It has been said that the individual Representative, because of his short term Of office, spends the major share of his incumbency campaigning for the next election. Because Of the frequency of congressional elections in the United States, the partisan composition of the House of Representatives is Often viewed as a barometer Of party fortunes in the nation. In this study we will examine congressional election results for the period 1942 through 1956, giving particular emphasis to certain patterns of relationship. The approach to be used is directed toward a partial description of macro-politics in the United States. More specifically, we shall look into these general hypotheses: (l) Partisan rank orders of congressional districts remain sufficiently stable between elections to produce high rank correlation coefficients; (2) The prevailing use of the single election forecaster, in which case a future election outcome is forecasted by examining the most recent election result, is wasteful Of the election data which is available; and (3) a combination Of the rank order stability of congressional districts and a multi-election statistical history of each district provides an efficient use Of available election data in forecasting. The individual voter is considered here only as his ballot is a part Of the vote cast for a particular congressional candidate. Psycho- logical variables of electoral behavior are excluded, for the most part, not for reasons of irrelevance, but for reasons of selection. It is felt that the study of individual motivations in electoral behavior is largely dependent upon definitions of the context within which it is undertaken. The solutions tO questions Of causal relationships must necessarily wait upon adequate and accurate description which includes characteri- zations Of the elements of stability and variability in the phenomenon studied. The importance of time analysis on aggregate electoral decisions has been stated convincingly by Key and Munger: Perhaps the collective electoral decision, the people's choice, is merely the sum Of individual choices. If enough were under- stood about individual decisions by addition the collective political decision of the electorate would be comprehended. Yet when attention centers on the individual elector as he is led to decision by the compulsion Of his nonpolitical group, the " tendency is to lose sight of significant elements that both affect and relate individual decisions to the political aggregate. The study of electoral behavior then becomes only a special case Of the more general problem Of group inducement of individual. behavior in accord with group norms. As such it does not invariably throw much light on the broad nature Of electoral decision in the sense of decisions by the electorate as a whole. 1 One cannot ask to what extent, for example, voting is a process of rational dec1sion-making unless experience demonstrates that there is a phenomenon occurring which is called "voting" and that there is a psychological process which we call “decision-making" and, further, that there is more than one manner Of making these decisions. Also, certain characteristics of the way a decision is made allow us to classify the act as "rational. " In the same fashion, questions regarding individual voting behavior in congressional electorates are dependent upon meaningful description of these electorates and the generalizations that can be drawn from' the study Of this description. 1V. O. Key, Jr. and Frank Munger, "Social Determinism and Electoral Decision: The Case of Indiana, " American Voting Behavior, 6d. Eugene Burdick and Arthur J. Brodbeck (Glencoe, Illinois: The Free Press, 1959), 15, p. 281. From previous voting studies we have evidence supporting hypotheses of the persistence of partisanship as a variable of individual and group voting. We should strongly suspect that a person whose parent had sup- ported Democratic candidates would also support Democratic candidates. V. 0. Key, while making this Observation, has pointed out the pitfalls in making an assumption too inclusive: Party loyalties may extend from generation to generation. Party attitudes seem to be transmitted from father to son--not biologically, to be sure; community, family, and other influence play a part in fixing partisan attitudes .- . . . The notion of inheritance of partisan attitudes through family influences, of course, oversimplifies the process of acquisition of party affiliations by young persons. The young are subject to the same types Of community influences as are their parents and, in most instances, they look forward to a status in society similar to that Of their parents. Identity of outlook and interest probably has quite as much significance in the "inheritance" of party affiliation as does parental example.l Further, Key and Munger highlight the significance Of time analysis in relating group bases of electoral decisions to individual choices: Explicit attention to the time dimension Of electoral decision would probably bring to light a variety Of characteristics not readily perceptible by the observation of a single case. Illus- trative is the difficulty of Obtaining a satisfactory estimate of the nature and significance Of traditional or habitual partisan attach- ments by interviewing a sample at a particular point in time. Often electoral decision is not an action whose outcome is in doubt but a reaffirmation of past decisions, at least for the community as a whole. For generations the Democrats may carry this county and the Republicans may predominate in an adjacent county. 2 If these hypotheses can be verified, it would logically follow that a large portion of the electorate tends to exhibit persistence in partisan choices at the poms. 1V. O. Key, Jr., Politics, Parties, and Pressure Groups (New York: Thomas Y. Crowell Company, 1953, 3rd Edition), pp. 585-586. 2Key and Munger, _O_p_. cit” p. 282. It is this characteristic of voters and aggregates of voters upon which electoral studies have focused, utilizing the stability of party preference as a base for forecasting the results of future elections.1 More important for the discipline of political science, this stability has been used by students to illustrate the operation of the party system in the United States. 7‘ The bulk of aggregate voting studies, however, have dealt with congressional elections in summary fashion.3 There seems to many Observers to be only two possible ways to describe the election results for congressional districts, a Democratic victory or a Republican victory. Consequently, many Observers have been satisfied to summar- ize this aspect Of American politics with a two-party percentage break- down Of House membership. Congressional studies, for the most part, have been residual to presidential election analyses. The relatively small amount Of attention which-has been devoted to this area is not com- mensurate with its importance as an indispensable part of the party system and government of the United States. We do not argue that academic investigation should be apportioned according to an arbitrary weighting of prospective subject matter. But we do insist that congres- sional politics cannot be regarded as a by-product of presidential politics. A macro-political description Of congressional districts undertaken here includes a classification Of districts according to four essential 1See for example Louis H. Bean, Ballot Behavior: A Study of Presidential Elections (Public Affairs Press, American Council of Public Affairs, 1940). 2Julius Turner, Party and Constituency: Pressures on Congress (Baltimore: Johns Hopkins Press, 1951). 3See Cortez A. M. Ewing, Congressional Elections, 1896-1944 (Norman: University of Oklahoma, 1947); Malcolm Moos, Politics, Presidents and Coattails (Baltimore: The Johns Hopkins Press, 1952); and Louis H. Bean, How To Predict Elections (New York: Alfred A. Knopf, 1948). criteria. If we are to relate the results here to the functioning of the party system, the first criterion must be that of partisanship. The basic statistic must ShOW whether a plurality Of the district's electorate voted Republican or Democratic at each of the elections considered. The second essential criterion is that of volatility, or the likeli- hood that a district will or will not retain its present partisan character from one election to the next. Volatility refers to the fluctuation of a district's partisan voting percentages. The third criterion, trend lepe, is concerned with the long range direction of percentage shifts. The fourth criterion deals with various relationships, particularly rank orders, among all districts within the universe being studied. In the case of the first three criteria, each district is given a score or rated on a scale of absolute values. It is on this basis that the districts are finally ranked. What results from these analytical explorations are new indica- tions of the kind and extent Of stability in the universe of congressional districts that are the intervening geO-political variables between voters and party divisions in Congress. A further contribution of this study may be the techniques it introduces for the evaluation of electoral prospects in the House contests, a matter Of particular practical concern to the candidates and their campaign managers. CHAPTER II TECHNIQUES OF DATA-GATHERING AND ORGANIZATION Selection of the Period for Study The period selected for analysis in this study included congressional elections from 1942 through 1956. The use Of this particular period pre- sented some difficulties because of reapportionment action in many Of the states. The congressional districts of 1942 in such states were split up in subsequent redistrictings. The advantages of Spanning reapportion- ment years, however, made it necessary to attempt a solution to this problem rather than work around it. In the first place, democratic politics such as we think we have in the United States exhibit cyclical patterns in which the two major parties alternate as the ”party in power. " Although there has been much disagree- ment among political scientists on the causal factors affecting any .particular cycle, it certainly cannot be denied that irregular party cycles do occur. 1 Accepting the fact Of cyclical fluctuations and keeping the analysis within the realm of current politics, it was necessary to include suf- ficient. elections to increase the prospect of finding a recent "hump" or "trough" of the ”congressional cycle. " Taking a cue from presidential election results, it would appear that the latest change in the trend Of partisan politics occurred in the 1948-1952 period. At least one lHarold Gosnell, Grass Roots Politics, notes the irregularity of political cycles, page 9. Louis Bean, in How to Predict Elections, ’page 161, says ”The evidence is clear that the year 1947 will go down in our political history as marking the end of the downward trend Of the New Deal tide and the beginning of a new one. " experienced Observer placed the turning point even earlier than 1948. 1 Accepting Bean's conclusions, it was necessary to cover at least the period 1946-1956 to span the suSpected "trough" of the present cycle. Since we must go back to 1946 to include at least half Of a cycle, one apportionment period had to be spanned, that following the 1950 census. In view of the limited resources of this study, no additional difficulties of constituency structure were risked by running the analysis back over other apportionments. Another consideration in selecting the time period for study is the distinction between Off-year or midterm elections and those held in presidential election years. Each of these two types of congressional election will be treated separately in one part Of the analysis. However, the same number of elections of each type was included in order that a fair comparison between the two could be made. For this, the period 1942 to 1956 was suitable. Within the chosen length of time there occurred four Off-year elections and four presidential year elections. Since one purpose Of this study is an investigation Of trends associ- ated with successive elections, a time period including at least three elections was required. It was just as convenient tO Obtain data on four Off-year and presidential year elections, and this was done. Another reason for covering as long a period as would be manage- able was the desirability Of reducing the candidate-personality factor in the election trends. ’It is not uncommon for an incumbent Congress- men to be re-elected to his seat in the House. In fact some Congressmen are re-elected so many times that for all practical purposes they "own" the Office. A period that includes eight successive congressional elections increases the prosPect of a change of candidates for most districts, although it may not have eliminated the peculiar influence Of particular perennial congressional candidates on their districts' 192. <_:_i_t_. , Bean, p. 161. Bean sets the latest turning point in the national political cycle as 1947. election outcomes. This difficulty would not have been completely remedied by using ten or twelve elections either. The choice 'Of eight elections is a compromise dictated by convenience of data and the limits Of the reapportionment problem. Structural A8pects Of Congressional Districting This work was intended to be comprehensive in that the universe of electoral units to be studied includes every congressional district in the United States.1 Since it was necessary to carry the number of elections back beyond a period of reapportionment, some method had to be devised to deal with the reshuffling of geographicalareas and constituencies that accompanies reapportionment. Had a method for spanning the reapportiomnent periods not been devised, and if we were compelled to use only the congressional districtsicit reapportioned in the period 1942-1956, the universe would have included only districts in twenty-nine states. This is hardly a satisfactorily inclusive universe to constitute a national study. Another alternative would have been to carry the analysis, includ— ing the congressional districts of all states, back only as far as the most recent apportionment in any of the states involved. But since the most recent redistricting action occurred in 1952, we would have been limited to a period covering only three elections--two presidential year votes and one midterm contest. Again, this would hardly permit analysis Of the time series type, nor would it allow any illustration of a trend's cyclicality. ' In viewing the fluctuations of party fortunes in the electorates of congressional districts over a period of time, some attention must be given to the structural changes of the districts. A. period of fourteen 1This is subject to data-gathering difficulties which make it impossible to include certain districts, as explained later. years, including eight congressional elections, was selected for study. TO allow for a trend analysis Of the electoral behavior Of district con- stituencies during this period, it was necessary to reconstruct certain geographical areas so that the districts analyzed would be comparable geographically Over the entire period. If no reapportionment or redis- tricting had occurred during the period 1942-1956, the problem Of district boundaries could be ignored. Actually, nineteen Of the forty- eight states altered in some way the boundaries of their congressional districts between 1942 and 1956. A total of 196 districts were affected by these changes. The manner in which the vote for United States Representatives is reported by the states dictated the method of district reconstruction developed here. . The» county is the basic unit for which congressional election statistics are reported by all states except where there are many districts in one county. Since the alterations mentioned above eliminate the use Of the district figures where boundaries have not been permanent, it was necessary to re-collect the combination Of counties in the areas redistricted. ‘ For example, one district in New York consisted of Allegany, - Cattaraugus, and Chautauqua counties prior to a redistricting in 1952, at which time Livingston county was added to these three counties to form a new district. For the purposes Of this study these four counties are considered a single congressional district for the entire 1942-1956 period. 1 The boundaries in effect in 1956 are used in compiling election data for all eight elections. In this way we have been able to take advantage Of the permanence of county lines in creating a geographically stable universe of congressional districts. 1 1An exception to the stability of county boundaries is found in the Commonwealth Of Virginia where some cities have assumed ”county" status as local units Of government. This change does not prohibit using the same method Of retaining 1956 district boundaries over the entire period Of study, however. 10 While new political combinations may occur in a redistricted constituency, permanently affecting party balance, the shortness 'of this trend analysis minimizes secular considerations Of this kind. ‘ It would be interesting to see how the congressional district con.- stituencies would have voted had they maintained their 1942 geographical boundaries through 1956. The information would not possess as much .practical value, however,» as a study which shows the record Of a con- temporary constituency which will remain a single voting unit in the future. Since trend studies are in part a device for current forecasting, the use 1 Of 1956 district boundaries permits forecasting Of future voting patterns for the geographically stable congressional districts. After reconstructing a national aggregate of districts which were the same from one election to the next, it became necessary to give special attention to a small group Of "special-problem" districts smaller in area than a county. These were not readily handled by the method established above. All are located in the metropolitan centers Of the United States and are subdivisions Of counties. This made it necessary to use wards and precincts as the basic source of election statistics. Two factors, however, prohibited the reconstruction Of urban congressional districts by combining ward and precinct areas. The first is that wards and precincts in many Of the urban areas were changed during the 1942-1956 period. The secondis that reconstruction of such districts would usually require a block-by-block summary of voting returns. . This requirement was beyond the practical means available for this study. The votes cast by metropolitan constituencies are of course as important to this study as those cast by the more accessible non- metropolitan constituencies. TO eliminate these districts because their data cannot be collected in the same manner as the others would remove a large bloc Of the national congressional electorate from the analysis. This problem was dealt with by the invention of the "Composite District. " 11 Where reapportionment has led to redistricting of more than one district within or largely within a single county, all of the electorate within that county is treated here as a single constituency and is called a Composite District. For example, Hamilton County, Ohio, contains two congressional districts under the present apportionment. The geographic composition of Hamilton County's two districts was changed by the redistricting of 1952, however, although both remained entirely within the county. The two are combined for this study and Hamilton becomes a single congres- sional constituency. In this manner the voting data is not lost and the labor Of a block-by-block compilation of votes is avoided. Included in the 336 districts studied here are sixteen counties which became Composite Districts of this type. Eleven are in New York, two in Ohio, one in Missouri, and two are in Pennsylvania. Another form of Composite District is constructed in cases where the 1956 district includes more than one county, but whose boundaries are not coterminous with county boundaries. Since we are limited by the fact that election returns are reported by county or congressional district, whichever is smaller in area, we have combined all the counties that are found in the same 1956 district with each part of the county split by a district boundary. As an example, all Of Putnam County and part of Westchester County in New York form a 1956 district. The remainder of Westchester contains a single district, but the vote cast in this portion of the county prior to the 1952 apportionment is not readily accessible. Therefore, the two counties are combined to form a single district. Eight of this type of Composite District are used in this study. Composite Districts are as follows: 12 Composite State District Number Counties Contained in District Illinois 1C Cook, Lake Maryland 3C Baltimore City, Anne Arundel, Calvert, Charles, Howard, Prince Georges, and St. Marys Missouri 1C St. Louis (including St. Louis City) _ 4C Barton, Bates, Cass, Henry, Jackson (including Kansas City), Johnson, Lafayette and Vernon New York 1C Suffolk 2C Nassau 3C Queens 4C Kings 5C Richmond 6C New York 7C Bronx 8C Putnam, Westchester 9C Wayne 10C Monroe 11C Erie 12C Niagara 13C Orleans, Genesee and Wyoming 14C Albany, Rensselaer 15C Clinton, Essex, Warren, Saratoga, and Washington Ohio 1C Hamilton 11C Ashtabula, Geauga, Lake, Portage, Trumbull, and Mahoning 20C Cuyahoga Pennsylvania 1C Philadelphia city 27C Allegheny North Dakota>i< 1C New Mexico=i< 1C Arizona** 1C Arkansas’vt 1C >:CNorth Dakota and New Mexico each elect two Representatives at-large. For purposes of this study each Of these states is used as a single The vote recorded is that cast for the party candidate attract- ing the largest total vote. district. 2'): >:< Because Of data gathering problems explained in the next section, Arizona and Arkansas are considered one-district states. 13 Problems Of Collecting Election Data The election statistics employed in this study are presented in Appendix A. Compilation of such statistics for a study intended to be national in sc0pe has proved to be almost a complete project in itself. The problems of gathering data can be traced directly to the lack of any central source of election statistics, a lack that has been increasingly noted in the literature Of politics. ‘ Richard M. Scammon's America Votes has been an extremely valuable sourcebook for the voting figures of districts not affected by reapportionment. . The use Of Scammon's compilation was limited tO the years 1946 through 1954. Since we are concerned with the period 1942-1956, it was necessary to use Vote Cast in Presidential and Congressional Elections, 1928—1944, Department Of Commerce, United States Bureau Of the Census, to Obtain the 1942 and 1944 congressional voting figures. The 1956 statistics were obtained from Statistics of the Presidential and Congressional Election Of November 6, 1956, Ralph R. Roberts, Clerk of the House of Representatives. From these sources the necessary data was readily available, but for districts affected by apportionment the task involved many additional difficulties. State manuals, sometimes referred to as "blue books" or "red books, " are published by many Of the states. Many states include a public reporting Of county voting returns in these manuals. The Michigan State Library in Lansing possesses a collection of these publications, but it covers only a small minority of states. In fact, even for those states that are covered the collection is incomplete. Nevertheless, this source was used to the extent possible before resorting to direct correspondence with the states concerned. Where the county vote for United States Representative was needed, it was necessary to write to the separate state election authorities. 14 In most cases the reply to these requests for statistics was prompt. In some other cases, however, the reply did not contain the information requested and follow-up correspondence was required. Officials of sixteen states were contacted with requests for county voting figures. The following persons were also contacted for information the states were unable to provide: Professor Clarence A Berdahl, University of Illinois; Professor Cortez A. M. Ewing, the University of Oklahoma; Richard M. Scammon, Governmental Affairs Institute; Professor Malcolm Moos, Johns Hopkins University; Arthur A. Schwartz, Director Of the Ohio Legislative Reference Bureau; William B. Welsh, Research Director Of the Democratic National Committee; and Richard C. Bain, Research Associate of the Brookings Institute. The task of collect- ing the basic data was made much less difficult by the willing assistance Offered by these men. Kentucky had to be removed from the scope of this analysis because the Official records of county vote for 1942 through 1950 were destroyed by a fire. Mr. Scammon was able to furnish the 1946 statistics, but since a complete set Of voting returns could not be accumulated, the entire state had to be dropped from the study. ‘ All of the necessary statistics are available for the State Of California, but the cross filing system peculiar to that state does not allow the accurate illustration Of party vote required here. It is unfortunate that none of the California districts are included. The California electorate is especially important since we could expect its competitive status to be representative Of the nation as a whole if the views contained in DeGrazia's Western Public are accepted. Because the two basic statistics employed here are vote cast for the Republican candidate and vote cast for the Democratic candi- date, the vote for a person who is the nominee Of both parties is not easily bisected on the grounds of partisan intent. 1 Also, this "mixing" 1This distinction should not be confused with the vote cast for the party itself. If the party's candidate is supported by a third party, this vote is included. 15 of votes makes it impossible to accomplish trend cOmparisons from year to year. Only by a poll of the voters in California could it be determined accurately which party they were supporting in casting ballots for a candidate holding the nomination of both major parties. This requirement necessitates the omission Of California. Another problem was that state supplies of official voting returns were sometimes exhausted. In Oklahoma, the secretary of the State Election Board hired for the author a person to compile the statistics not previously published in a public report. The Director of Bureau of Commissions and Elections for Pennsylvania, Albert E. Eberman, fur— nished a photostated copy of the 1946 election returns in lieu of published returns not available. Following the suggestion of Professor Berdahl, the author requested and received from the University of Illinois Library reproduced voting figures which the Secretary Of State could not provide. The Secretary of State for Missouri extended a courtesy peculiar to that state alone. Because the supply of returns for public distribution was exhausted, he offered to loan the author his office copies for a short period of time. The Offer was accepted and the official voting figures were recorded from this source. In the cases of both Arizona and Arkansas, it was necessary to combine the total state vote for congressional candidates into a lumped figure and treat each state as a single district. ' Accessible information on Arizona and Arkansas showed only the total state vote cast for all congressional candidates. Despite repeated efforts to Obtain the necessary voting returns from these states, they were not available. The at-large congressional districts in existence at the time Of the 1956 elections are not employed here except where the whole state was the only district unit for congressional elections. TO Offer an example, Texas has twenty-one Congressmen elected from apportioned districts in addition tO a Representative elected from the state at-large. For our l6 purposes the at-large district was not included and Texas was limited to twenty-one districts. There are two specific reasons for excluding at-large districts found in states with regularly apportioned districts. The first is that the electorate for these at-large areas is the same electorate represented by all of the other districts in the state. . To include them would only repackage votes already selected for study. The second reason bears directly on one Of our primary Objectives, analysis of the single-member district. If we are to focus on this structural aspect Of inter-party conflict, the inclusion Of geographical areas repre- sented by more than a single congressman must be minimized. In addition, however, we have had to deal with cases such as Delaware which has a single Representative in Congress. Because it is not apportioned into more than one district, the State Of Delaware is con- sidered as a single congressional district. In addition to Delaware, the following are single-district states: Nevada, New Mexico, North Dakota, Vermont, and Wyoming. Of these six, New Mexico and North Dakota are actually entitled to two Representatives each, elected at-large under the 1956 apportionment. They are represented here, however, as single districts. The voting returns used are those case for each party candidate drawing the largest vote for a particular election. Statistical Tools The basic statistical unit used in this study has been called the "Stalemate Index. " It is selected as a convenient index number which is both adaptable to further mathematical operations and representative of the competitive status Of the Democratic Party y}: a y_i_§ the Republican Party for a particular election and a particular electoral unit. The Stalemate Index is defined as one-half the difference between the major party percentages of the total vote. ' In a case where the 17 Democratic candidate would draw 45% of the votes in a congressional district election and the Republican candidate would draw 35%, the Stalemate Index would be 5. O or one-half the difference between the two percentages. 1 To indicate which party is victorious, algebraic signs are attached to the absolute value of the index number. A positive Stalemate Index represents a Democratic victory and a negative index indicates a Republican victory. The complete representation of the above example, taking into account a Democratic plurality, would be +5. 0. We have used this particular manner Of summarizing voting returns to capitalize on two specific advantages as follows: 1. As a descriptive device, the Stalemate Index is superior to possible alternatives in that it represents in a single explicit number the outcome and the winner' 3 advantage in percentage. It is important to note, further, that this advantage is shown as one-half the difference between major party percentages of the total vote. From a politically practical point of view, the Index tells us what percentage Of the total vote the loser needed to attain a tie or stalemate. The major assumption in stressing the descriptive significance of the Index is that the losing major party, had it achieved a stalemate, would have had to cover the ground indicated. Since the percentages used to derive the Index are computed with total vote as denominator, the ”third” or minor parties are not completely disregarded while focusing on the major parties. 2. The Stalemate Index takes into account a reality of United States elections which has been neglected by other statistical descriptions-- elections are won by a plurality Of votes. A majority is unnecessary. If we were to employ the Democratic (or Republican) percentage of the total vote as our basic statistic, an election victory would be apparent from the statistic only when the Democratic (or Republican) candidate polled more than 50%. It is possible, however, that 45% Of the total vote 1See Appendix B for a complete explanation Of the Stalemate Index. 18 represents a plurality sufficient for victory. The Index corrects this statistical deficiency while accurately representing the competitive status Of the defeated major party. In the case Of a 45—44-11 percentage split, the defeated major party was not 6% short of an election victory as would be the case if a majority were required. It lacked 1% plus 1 vote Of winning . CHAPTER III MARGINALITY AND THE SINGLE ELECTION FORECASTER Political parties function to win elections. When the decision-makers in the “out party" estimate that their chances of winning an election are great, they are willing to spend greater amounts of time and money on the campaign than when they estimate that their victory chances are small. Where the ”in party" perceives its chances of losing the next election to be great, we could expect that party to invest greater efforts in attracting campaign resources to save the incumbent. We would expect to find, for example, very little resource support for a Republican congressional candidate in Alabama's first congressional district. Information from past elections tells us that a Republican does not win this congressional seat. Therefore, we do not expect a Republican candidate to win. Nor do we expect that donors will step forward with large sums of money to support him. Campaign workers, having no real possibility of contributing to a successful campaign cannot be expected to volunteer in large numbers. On the other hand, a campaign in a congressional district where close elections have been common in the recent past and where the congressional seat has alternated between parties would be expected to attract a greater amount Of campaign ammunition. In short, much in the allocation of campaign resources hinges upon political estimates of the chances of winning. Willingness to expend resources for campaign purposes increases as the possibility Of an election turnover appears to increase, usually when a district is counted as 19 20 '1marginal." In the same manner, this willingness diminishes as the chances of a party turnover appear to diminish. There are so many variables involved in decisions on allocation Of campaign resources that these conditional generalizations are the closest approach to a valid "rule" of any kind. Defining Marginality The next question logically is: "How is it determined whether or not the possibility of an election turnover is strong? " Intuitive and personal factors are intimately entangled in answering such a question and probably would lead to as many answers as number of candidates questioned. An exhaustive study of how political candidates judge their chances of winning or losing an election would involve the endless array of variables necessary for studies of perception. Noting that this type of study would apply, but precluding it for our purposes here, we will proceed using a single, but general, variable of perception--the perceiver's judgment Of "marginality. " Those who are personally involved in a campaign are likely to evalu- ate the chances Of winning in a manner different from that of a political scientist making an Objective analysis. Opposition candidates in a close district are likely to carry identical high hopes for victory. Even in normally one-party districts, the hopelessly beaten underdog is likely to consider himself a sure upset victor until election day. Naive Optimism is a widespread characteristic among the ranks Of political candidates. Among the group of party strategists whose job it is to recommend the most efficient allocation Of campaign resources, the possibility Of an election victory in a particular district can not be forecasted on the basis of shallow Optimism. 21 Limited resources require that little be wasted on noble gestures. For purposes Of party morale expenditures might be necessary in districts where a loss is certain, but the greatest share of available money and materials will be funnelled to areas where success at the polls is at least possible. The allocation Of resources made available to the political parties and their candidates for campaign purposes is determined by the Special circumstances of each separate election battle. The personal wealth of a candidate may allow him to campaign freely. Friends, personal campaign committees and party campaign committees must attract these resources for their candidate's campaign. Also, a decision must be made as to how much Of the available resources might be used effectively in support of their candidate. Even the independently wealthy candidate, or his managers, must decide at what point further expenditures of time and money would or would not be worth the effort. V. 0. Key points out the relationship of competition and campaign expenditure levels in the following way: Campaigns that are not warmly contested are apt to evoke small expenditures. The presidential campaign Of 1924, for example, involved a relatively small expenditure. In many instances a sitting Official is re-elected with only token Opposition and with only slight expenditures. It is also apparent that campaign ex- penditures are influenced by the intensity of feeling about the issues. The keen sense Of competition, indeed the feeling that it is obligatory to upset any competitive balance, which is prevalent among active policital participants, has a great deal tO do with the allocation of campaign re- sources. Key sums it up as follows: ' The competitive factor induces high outlay. Each party or faction feels that it must attempt to match the expenditures and the showing 1Key, 52. <_:_i_t_., p. 532. 22 made by the other, for undoubtedly in many elections and prim- aries money helps mightily in gaining victory.l Investigation of available evidence on how peOple involved in political contests judge, orpre-judge, outcomes Of particular elections Shows that ”marginality" or "closeness" of recent elections is the determining factor. The more evenly matched were the two major party candidates at the previous election, the more “marginal" or "close" is the electoral district classified for the next election. Again turning to Key, we find that his Observations uphold this con- clusion: An elementary principle Of campaigning is that efforts should be concentrated where they will do the most good; usually in closely contested and doubtful states . . . Elsewhere the greatest outlay Of energy and of funds is likely to occur in the areas believed to be close.2 Key is doctrinaire in giving advice to politicians in regard tO their expenditure of campaign ammunition: The significance of the pattern of behavior for campaign strategy is plain. Campaign resources ought to be concentrated in close districts.3 But surely all people, and politicians are no less people than political scientists, do not read the same meaning into the words "marginality" and "close"--even if they agree that "marginality" and "closeness" are crucial in the selection of districts where a party turnover might occur. What do the terms "marginality” and "closeness" mean to political prognosticators? How does a national party organization and its research staff select the congressional districts in which campaign efforts will be concentrated? These are the questions that will be treated here. llbid., p. 537. 21bid., p. 494. 31bid., p. 524. 23 The Research Division of the Democratic National Committee in recent years has focused its attention upon the percentage of two-party vote cast for major party candidates in a congressional district.1 Each congressional district was classified as ”marginal" of "safe. " Where a Democratic candidate had won a congressional election by drawing less than 55% of the two-party vote, the district was classified as “marginal Democratic. " If this same candidate had won with more than 55% Of the two-party vote, the district became ”safe Democratic. " Districts won by Republican candidates were classified in the same manner. By this method the size of the percentage differential between the candidates at the last election was the crucial forecasting" device for the next election. This percentage differential size was supposed to tell us whether or not the district would be vulnerable to change at the next election. 2 The terms used by the Democratic National Committee to describe the degree of competition between the two major party candidates in a particular district, then, indicate their evaluation of the possibility that the district will be won by the Democratic candidate. A "safe" Democratic district is expected to re-elect a Democratic congressman. A “marginal" Democratic district is expected to be more vulnerable to Republican campaigning. 1Study Of Marginal Districts: 1952, 1954 and 1956; Research Division, Democratic National Committee, March 1, 1957. 2The Republican State Central Committee in Michigan, in a statistical analysis of state elections following the November 1958 General Election, has used "marginality" as a criterion for vulnerability to change. In their study, the Republican percentage of the two-party vote cast is used as the basic statistic. ”Marginal Congressional Districts" are those in which the winner's percentage of the major party vote at the last election was no greater than 5% more than the loser's percentage. "Semi-marginal Congressional Districts" are those in which the percentage difference between the two major parties at the last election was greater than 5% but less than 10%. Where the margin of victory at the last election was more than 10%, the Republican State Central Committee classifies the district as ”Safe. " 24 Statistical analyses of election results by both major parties employ this same method. Percentage spread between the two parties at the previous election is the crucial determinant of marginality. It is inferred that the prime target districts would be the ”marginal" Republican districts (for Democrats) and the districts demanding a strong defense would be the ”marginal" Democratic districts. Employing this type of classification would presume an emphasis upon channeling campaign resources to the "marginal" district contests. "Marginal” and "vulnerability to change" are used synonomously. Both parties in fact could be eXpected to follow the pattern. An ab- normal concentration Of men, material, and money in any particular "marginal" district by the out-party is likely to attract a counter concen- tration by the in-party. In the light of congressional election results for the period 1942 through 1956, is the "55% and less" district actually more likely to be won by the out-party than the "over 55%" district? And, more important, would complete success in using this definition Of marginality really be a success ? Would the definition provide the greatest possibility of selecting districts most vulnerable to turnover? Or does the assumed stability Of individual voting habits create voting trends that require analysis of many successive election results to determine the marginality Of any selected district? In Politics, Presidents and Coattails Malcolm Moos devotes an entire chapter to the "Marginal Congressional Districts. " His use of marginality emphasizes the same elements highlighted by the two party studies referred to above--percentage Of the two-party vote at the previous election. The Moos description is emphatic in precluding a great portion Of our cOngress— ional district elections from even the slightest significance in the compe- tition for political ascendency. 25 Moos insists that "legislative control of the Republic is determined by the elections in those districts where Republicans gather 45-55% Of the two—party vote. " He defines these districts as "Marginal. ”1 Moos establishes a separate classification Of "Critical Marginal" districts. These are the districts in which Republicans gather 48. 5% to 51. 5% Of the two-party vote. Although he does not establish rigid classifications as does Moos, V. O. Key sees "closeness" or "marginality" in the same terms as Moos. To Key, the percentage Spread between the two major-party candidates at the previous election is almost a determining factor in forecasting the result Of the coming election: Another analysis that throws some light on the nature of the mid- term decision is the identification Of the districts most likely to shift in partisan complexion at mid-term. Since voters' partisan attachments have a high degree Of persistence, it would be expected that those districts most likely to change at mid-term would be those with the closest results at the preceding presidential election. The Republicans might have a chance to pick up a district in 1950 that had gone Democratic by 51-49 per cent at the preceding polling, but their prospects in a district that had divided 65-35 would be much less . . . Party shifts occur in close districts, but Oddly enough the sentiment seems to move in the same direction in nearly all close districts.z In addition to the definition of marginality prOposed by Moos, Key and the party staff studies, it is suggested here that both the rate and direction of voting trends are necessary to evaluate effectively the possibility that a congressional district will be won by a party candidate at any given election. This statement presumes that the previous election result is not by itself adequate evidence for forecasting marginality. Marginality, employed as an indicator of vulnerability to change, would not be fully described by the closeness of the last election. For example, a district lMoos, loc. cit. ZKey, o_p_. c_i_t_., p. 522. 26 whose last election showed a 2—party vote percentage Split of 53—47 would be classified as a marginal district if closeness were the only criterion employed. If, however, further investigation showed that the results for the past four elections remained at a 53-47 split favoring the same party, we would be less inclined to classify the district as marginal. Inadequacy of the Single Election Forecaster Investigation of the election data gathered for this study will show that the "marginal" classification employed above is not an accurate description Of the congressional district's competitive status. Three types Of congressional election sequences have been examined-- successive elections, successive presidential year elections, and successive mid-term elections. Election sequences have been classified in this way because previous studies Of congressional contests have made claims that each type is not comparable with the other and should be investigated independently. l lMalcolm Moos, loc. cit.; Bean, loc. cit.; Ewing, loc. c_i_t_.; Bean, in fact, minces no word-Sin Stating that mesa-types Of eleOfions should be separated: ". . . congressional elections in mid-term are not strictly comparable with congressional elections in the more exciting presidential years. They must be studied separately. ” (p. 31). Bean explains these differences by Offering his coattail theory: ”Once it started upward, with the aid of the 1929-32 depression, the tide moved on a level of 6 or 7 per- centage points higher in presidential than in mid-term election years. This suggests that in presidential campaigns 6 to 7 per cent of the 435 Democratic congressional candidates were elected mainly by virtue of the fact that they were on the national ticket. In other words, about 26 to 30 congressmen thus appear to have ridden into Office on the President's coattail in 1932, 1936, 1940, and 1944" (p. 32). Moos, on the other hand, disagrees with Bean's coattail theory, but indirectly demands separate investigation Of the different types of congressional elections by postu- lating a relationship between presidential and congressional candidates: "If we find the presidential candidate running well ahead of his party's congressional ticket, we may assume that he helps the congressional candidates who trailed behind him" (p. 10). A predecessor to Moos, 27 Table I relates the Size Of change in voting percentages (as measured by the Stalemate Index) to the election results for four presidential year congressional contests (1944, 1948, 1952 and 1956). As an example, we can look at the election experience Of Michigan's Sixth congressional district. In all four elections (1944, 1948, 1952 and 1956) the Republican candidate won. This means that all of this district's cases (pairs of presidential-year elections) would be entered in the first row of Table I, the "Same Party Wins" row. In the election of 1944, the Stalemate Index for Michigan's sixth district was -5.4 and in 1948 it was -0. 3; the difference between the two election results is 5. 1 points. Therefore, the 1944-48 case for this district would be entered in column 2 of Table I, the 5. 0-9. 9 Stalemate Index Change column. The difference between the 1948 and 1952 elections was only 2.5 points. Therefore, this case would be entered in column 1 of Table I. The 1952-1956 difference would also put Michigan‘ 8 Sixth district in column 1 since the difference was 1. 9 points. This district's contribution to Table I, then, was a single occurrence entered in row 1, column 2., and two cases entered in row 1, column 1. If marginality were truly a function Of the "closeness" Of the last election Of the same kind in a district, we would expect to find the greater Cortez Ewing, from whom the former borrowed heavily, also implied the distinction between congressional election types by ranking presidential candidates by "efficiency, " or the percentage by which they led their con- gressional tickets. The evidence offered by these authors, however, is not sufficient ‘ to justify including or excluding separate analysis of the three types of congressional elections. Perhaps the sharpest Observation on the question was Offered by V. O. Key stating "the chances are that, by a process of elimination, one is pushed to the conclusion that the difference between the outcome of congressional elections in presidential and mid-term years must be attributed chiefly to the coattail effect operative in presidential years. " In view of the inconclusiveness of available evidence, the three types Of election sequence will be examined separately to determine whether or not there is any further justification for examining them separately. 28 number of other-party victories in the 0-4. 9 Stalemate Index Change column Of Table I, because these are the closest districts. Actually, however, we find the greatest number Of changes tO the other-party in those districts where the size of the Stalemate Index Change exceeds 5. 0 points. Table I. Distribution of Congressional District Election Outcomes According to Size of Stalemate Index Change Between Successive Presidential-Year Elections Election Size of Stalemate Index Change Result 0—4.9 5.0-9.9 10.0 plus Total Same Party 432 190 171 793 Wins Other Party 24 44 39 107 Wins Total 456 234 210 900 No Change Cases 2 107 In fact, there is an almost equal number Of change cases in the 5. 0- 9. 9 classification (44 cases) and the 10.0 plus classification (39 cases). The same type Of evidence is found in Table II which summarizes the data for congressional elections in four non-presidential years (1942, 1946, 1950 and 1954). Again, most congressional districts favor the candidate of the same party from one election to the next. But in those cases where there is a turnover, with the out-party winning the election, there are more occurrences in the 5.0 and over classifications. There is one point that should be stressed about these figures. Their real significance is in those cases where the out-party wins. 29 Table II. Distribution of Congressional District Election Outcomes Accord- ing to Size of Stalemate Index Change Between Successive Mid- term Elections Election Size Of Stalemate Index Change Results 0-4.9 5.0-9.9 10.0 plus Total Same Party 450 188 146 784 Wins Other Party 27 32 41 100 Wins Total 477 220 187 884 NO Change Cases = 124 If these findings contradict the predictive utility of the prevailing definition of marginality, those cases where the same party wins in successive elections must be disregarded. The mere fact that the greatest number of cases is found in the ”0-4.9, same-party" class does not alone contradict the prevailing definition. All of these cases could have occurred in heavily one-party areas where a small change means nothing. All, then, are vulnerable to the charge that they remained in the same-party column (deSpite a small c hange) because the difference in two-party vote percentages at the previous election was greater than 5%, or that the in-party candidate drew more than 55% Of the two-party vote. In Table 111, only the cases where the out-party won are re-examined and it becomes evident that an actual party turnover occurred in each case. Consequently, classification of cases into size of Stalemate Index change does offer sound evidence as tO whether or not the "closeness" of the previous election is a valid criterion for estimating turnover prospects. 30 If we find in a large number Of districts where turnover has occurred that the Stalemate Index change from one election to the next has been 5. 0 or greater, "marginality" as it is presently defined would be a poor guide to the most efficient allocation of campaign resources; district change Of greater than 5.0 points would be tOO common. If, on the other hand, we find that in most districts where there has been an unseating of the in- cumbent the Stalemate Index change was less than 5. 0, the evidence would support the prevailing predictive use of "marginality. " Table III. Distribution Of Congressional District Election Cases Accord- ing to Size of Stalemate Index Change Where Turnover Occurred at Second Of Two Elections of the Same Type Size of Stalemate Index Change 0-4.9 5.0-9.9 10.031115 Pres. Year Percent Elections 21.6% 39.6% 38.8% Of Cases Mid-term Elections 26.7% 31.7% 41.6% Totals 48.3% 71.3% 80.4% We see again that in cases where the district changed parties a greater percentage Of districts experienced Stalemate Index changes of greater than 5. 0. The next logical question is: "But did these districts really need to move more than 5. 0 for the out-party to win?” Perhaps all of them were in the 0. 0-4. 9 class before the turnover. The previous examination of district election statistics, Offered to point out the weakness in the prevailing concept of “marginality, " was based solely on the size Of the Stalemate Index change from election to election. This concept, however, included both the closeness Of the 31 previous election, _v_i_s__a_ y_i_s_ the relationship between the two parties, and the closeness Of the out-party to an election victory. We have shown that, according to size of electoral change, more districts move more than 5. 0 points than move 0-4. 9 points in any election. But it is entirely possible that the districts moving more than 5. 0 points actually would have had to move less than 5. 0 points for the out-party to win. Therefore, to complete the examination Of "marginality" we must look at the starting point, that is, the size Of the Stalemate Index at the last election, as well as the Size Of the shift from this starting point. If all districts which move less than 5. 0 points started from a Stalemate Index of less than 5. 0 and have moved to the out-party, then the prevailing use of marginality is essentially valid. But if an examination of the cases in which districts elected out-party candidates in the second Of two succes— sive elections shows that districts starting with a Stalemate Index of more than 5. 0 points produced as many victories as those with less than 5. O, the usefulness of ”marginality" is in serious question. Table IV classifies cases in which congressional districts elected out-party candidates in the second Of two successive elections. A quick inspection of the summarized statistics Shows that in a Sizeable, im- portant portion of the cases the out—party has overcome a Stalemate Index of more than 5. 0 to win the second of two successive elections of the same type. For presidential year elections, in 41. 7% Of the cases where the out-party won in the second of two consecutive elections, the Stalemate Index at the previous election fell into the four categories greater than 5. 0. In as many as 19. 4% of the cases, the previous election Stalemate Index was 10.0 or greater. For mid-term elections, in 35. 0% of the cases where the out-party won in the second of two consecutive elections, the Stalemate Index at the previous election was greater than 5. O. In 17.0%, the previous election Stalemate Index was 10. 0 or greater. 32 Table IV. Distribution of Pairs Of Congressional District Election Cases According to Size Of Stalemate Index at the First of Two Successive Elections Where Turnover Occurred at Second Election Size of 8.1. Pairs Of Pairs of Pairs of at the first Pres. Year Mid-term Successive Election Elections Elections Elections Cases Percent Cases Percent Cases Percent 0-4 9 63 58 3 65 65 O 148 68 5 5.0-9.9 24 22.2 18 18.0 41 19.0 10.0-14.9 12 11.1 8 8.0 20 9.2 15.0-19.9 6 5.5 3 3.0 4 1.9 20.0 plus 3 2.8 6 6.0 3 1.4 Totals 108 100.0 100 100.0 216 100.0 In 31. 5% Of the cases Of all election years where the out—party won in the second Of two consecutive elections, the Stalemate Index at the previous election was greater than 5. 0. In 12. 5% the previous election Stalemate Index was 10.0 or greater. Thus, in a majority Of cases where a party turnover has occurred in pairs of successive elections, the Stalemate Index Of the first election was less than 5. 0 points. This pattern holds true in all three types-- pairs of presidential year elections, pairs of mid-term elections, and pairs of successive elections. Each of the districts experiencing a party turnover in an election immediately following one where the Stalemate Index was greater than 4. 9 is listed in Table V. These are the districts that contributed to the totals in the ”Pairs of Successive Elections” column Of Table IV. It is signifi- cant that no section of the nation, or for that matter no state, has a Table V. 33 Distribution of Cases in Which Turnover Occurred Where Stale- mate Index at Preceding Election was Greater than 4. 9 District Pairs Of Election Years Mid- Term Pres. Year Consecutive Elections 42- 46- 50- 46 50 54 44- 48- 52- 48 52 56 42- 44 44- 46 46- 48 50- 52- 54- 52 54 56 Colo-3 Colo-4 Conn-l Conn-2 Conn-3 D D D D DU Del-l Fla-l Ida-l Ill-1C Ill-21 CW 51321 111-23 111-25 Ind-8 Ind-3 Ind-5 UU Ind-4 Iowa-6 Kan-1 Kan-5 Me- 2 CU Md-l Md-Z Md-3C Mass-2 Mass-4 Mass-1 Mich-13 Mich-l4 5990 Mich- 17 Minn- 3 Minn-4 Minn- 6 Minn- 8 DU Continued Table V - Continued 34 Pairs of Election Years Mid-Term 42- 46- 50- District 46 50 54 Pres. Year 44— 48- 48 52 56 52- Consecutive Elections 42- 44 44- 46- 48- 50- 52- 54- 46 48 50 52 54 56 Minn-9 D MO- 1C MO-6 MO-7 Mont-2 D D R Neb-l R Neb-Z Nev-1 N. J. -1l N. J. -4 CU UUU J.-6 D J.-14 .Y.-3C Y Y DU .-5C .-11C 2.22.2.2 .-14C R .-28 .-10 . .-1 Ohio-l8 D 2222 Uni-<14 Ohio-20C D Ohio- 16 Ohio- 14 Ohio-3 R Ohio- 15 R CO Ohio-1C Okla-1 Ore-2 Ore-4 Pa- 1C D CD CD CD Pa-14 D Pa-27C Pa-lO Pa-ZZ S. D. -l SUN Continued Table V - Continued 35 Pairs Of Election Years Mid-Term Pres. Year Consecutive Elections 42- 46- 50- 44- 48- 52- 42- 44- 46- 48- 50- 52- 54- District 46 50 54 48 52 56 44 46 48 50 52 54 56 Tex-5 R R R Utah-1 R Utah-2 R R R Va-6 R R R Va-9 R Va-lO R Wash-l R D D Wash-2 R R R Wash-3 D Wash-6 R R R Wisc-4 R R Wisc-5 D Wisc-9 D D D Totals ll 13 ll 16 13 14 7 15 23 l 5 7 8 32-States 9-R, l-R, 6-R 2-R, 12-R, 3-R, l-R, lZ-R,0-R, 1-R,4-R, Z-R,4-R, 82-Districts Z-D 12-D 5-D 4-D l-D ll-D 6-D 3-D 23-D 0-D l-D 5-D 4-D 36 monopoly on this type Of turnover, e. g. out-party overcomes a Stalemate Index greater than 4. 9 at previous election to win. Eighty-two districts in 32 states have experienced this phenomenon in the 1942-1956 period. The wide range in number Of cases of this type occurring at dif- ferent elections suggests that there may be a national factor more prevalent in one election than in another and that this factor enhances the vulner- ability of districts to turnover. In the case Of all successive elections, the number of turnover occurrences (Table V, column 3) for a single election ranges from 1 in the 1948—50 election pair to 23 in the 1946-48 election pair. Isolation of the factors which explain this variation certainly merits further study. A clearer picture Of the relationship between party turnover and the Size of the Stalemate Index at the first of each pair Of elections is illus- trated in Tables VI and VII. Table VI is concerned with the districts in which the out-party won in the second Of two successive elections. It Shows that at the preceding election a Stalemate Index of greater than 5. 0 existed for these turnover districts: (a) In nearly 42% Of the cases for Presidential- year sequences; (b) In 35% of the cases for mid-term sequences; (c) In 31% Of the cases for pairs of successive elections. Table VII is limited to those districts in which the incumbent party won in the second Of two successive elections. It shows, as most certainly would be expected, that a preponderance Of non-turnover cases occur where the Stalemate Index at the preceding election was greater than 5. 0. Statistical Test on ”Marginal" Districts That the "closeness” of the first of a pair of successive elections can not be ignored, however, in forecasting the vulnerability of a district to change is demonstrated by chi-square tests. By placing the data in contingency tables, the relationship between size of Stalemate Index at the last election and out-party victory in the second Of two elections is revealed. 37 Table VI. Distribution Of Congressional District Election Cases According to Size of Stalemate Index at First of Two Successive Elections (Where Out-party Wins Second Election) Number of Cases in Which Out-party Size of 8.1. Wins Second Election at First Pairs of Pairs of Pairs of Election Pres. Year Mid-term Successive Elections Elections Elections 0-4. 9 63 65 148 5. 0 plus 45 35 68 Totals 108 100 216 Table VII. Distribution of Congressional District Election Cases According ‘ to Size of Stalemate Index at First Of Two Successive Elections (Where Incumbent Party Candidate Wins Second Election) Number of Cases in Which In-party Size of 8.1. Wins Second Election t ' t 21:15:11 Pairs of Pairs of Pairs of Pres. Year Mid-term Successive Elections Elections Elections 0-4. 9 153 129 338 5. 0 plus 746 779 1797 Totals 899 908 2135 38 Contingency Table A: Pairs of Presidential-Year Congressional Elections Size of S.I. , Result Of Second Election At First 1 Election Turnover NO Turnover ‘ Total 0-4. 9 63 153 216 5.0 plus 45 ' , 746 791 Totals 108 ’ 899 1007 Contingency Table B: Pairs Of Mid-Term Congressional Elections Size of 8.1. ' Result Of Second Election At First Election Turnover No Turnover Total 0-4. 9 65 129 194 5.0 plus 35 779 814 Totals ’ i 100 908 1008 Contingency Table C: Pairs of Successive Congressional Elections Size of 5.1. Result of Second Election At First Election Turnover NO Turnover Total 0-4.9 148 338 486 5.0 plus . 68 1797 1865 Totals 216 2135 2351 39 The relationship between turnover and "closeness" may be tested by the chi-square test. The null hypotheses to be tested are as follows: 1. We do not have sufficient reason to say that party turnover in successive presidential-year congressional elections is dependent on whether or not the size Of the last Stalemate Index was above or below 5. 0 points. 2. We do not have sufficient reason to say that party turnover in successive mid-term congressional elections is dependent on whether or not the size of the last Stalemate Index was above or below 5. 0 points. 3. We do not have sufficient reason to say that party turnover in successive congressional elections is dependent on whether or not the size Of the last Stalemate Index was above or below 5.0 points. With a significance level Of . 05 the chi-square test of independence on each Of the contingency tables results in rejection of each Of the hypotheses. In other words, we cannot say that we do not have sufficient reason to say that party turnover is dependent on whether or not the size of the last Stalemate Index was above or below 5. 0 points. Although the null hypotheses Of independence must be rejected, indicating that the "closeness” of the first Of two elections is significant in guessing the outcome of the second, a second look at the contingency tables would indicate that the "closeness" does not tell the whole story. In the presidential-year contingency table, we see that in 45 out of the 108 cases where a turnover has occurred the Stalemate Index at the first election was greater than 5. 0. This means that 41. 7% of turn- over occurrences took place in districts which were not considered ”marginal" under the prevailing use Of the term. The mid-term statistics show that 35. 0% of the 100 cases where a turnover occurred would not have been classified as "marginal" prior to the election in which the party change took place. In the case of all successive elections, 31. 5% of the 40 districts followed this same pattern. In all three types of election sequences, then, approximately one-third of the party turnovers occurred in district elections which were not considered "marginal" or "close" in the first place. These changes could be called unexpected. On the other hand, the number of cases where a turnover might have been expected, but did not materialize, was high, enough to cast serious doubt on the validity Of the marginality criterion upon which the expectation Of turnover was based. In the presidential-year election sequences, 70. 8% Of the cases in which the previous Stalemate Index was less than 5. 0 did not change parties in the subsequent election. For mid-term election sequences, 66. 5% Of those cases in which a change might have been expected did not in fact turn over. For all cases of successive elections, 69. 5% of the cases where a district would be classified as "marginal” remained in the column Of the incumbent party. When the expectation of turnover is expressed as a function Of the "closeness" (measured in terms of percentage Spread between parties) of the last election, we found one-third of all turnover cases occurring where they were not expected and 65-70% of the non-turnover cases occurring where a turnover might be expected. To allocate limited campaign re- sources where the payoff probabilities are SO low is wasteful. When expectations Of party turnover are based upon criteria which result in as small a yield of success as the prevailing definition Of "marginality, " there is no reason for continuing to use it. Summa ry In this chapter we have examined the current measure of "marginality" widely employed by practicing politicians. Noting that ”marginal" districts have been located by examining only the results of the most recent election, we found that this Single—(election basis for forecasting is not efficient. 41 ' Examination of congressional election statistics, as summarized by the Stalemate Index, Shows that where the out-party wins, it does so by gaining more than 5. 0 Stalemate Index points between two consecutive elections more Often than it does by gaining less than 5. 0 points. This Observation holds for all three types Of election sequences--presidential year congressional elections, mid-term elections and consecutive elections. In cases where a congressional district elected the out-party candi- date at the second of two consecutive elections, its Stalemate Index at the first election was more Often smaller than 5.0 than it was greater. This Observation is pertinent to all three types Of election sequences. In a significantly large portion Of the cases where the out-party won, however (at least one-third of the cases, depending on the type of election sequence), the Stalemate Index at the first election was greater than 5. 0. Chi-square tests of significance on the relationship between size Of Stalemate Index at the first of two successive elections and the outcome Of the second election lead us to reject the null hypothesis that "We do not have sufficient reason to say that party turnover is dependent on whether or not the previous Stalemate Index was above or below 5. 0. " "Marginality, " or "closeness, " as currently used and as measured by percentage spread between the two major-party candidates at the last election does have some relationship with the outcome Of the next election. It is not sufficient explanation, however, for one-third of all turnover cases that occurred where they are not expected and 65-70% of non-turnover cases where a turnover might be expected. If “marginality" is to be used as an indicator of prospective party turnover in congressional elections, the size of the Stalemate Index at the last election ought to be included in determining “marginality. ” Other factors must also be considered in conjunction with size of Stalemate Index. Political experience Of a congressional district is not limited to 42 the "last" election. Many elections have been held and the outcome of each is a part Of the district's total experience. If the single-election forecaster is not really an effective fore- caster, the multi-election forecaster may be a more efficient manner of estimating future election results. - In other words, if all of a district's election results were charted on a graph, this experience could be represented by a regression line. Any series Of elections withdrawn from this historical representation could also be characterized by a trend line. This trend would be described by its direction Of movement and its rate of change (or degree of angle) over time. The employment Of a multi-election trend line in forecasting would add two kinds of information lacking in the Single-election forecasting base: direction and rate Of change. Having shown in this chapter that the result Of the "last" election is necessary but not sufficient evidence of prOSpective turnover, we now proceed to establish the value of rank order relationships as a forecasting aid (Chapter IV). In Chapter VI a multi-election (trend) forecasting base will be combined with demonstrated rank order relationships to suggest a forecasting procedure for future election results . CHAPTER IV COMPETITIVE DISTRICT RANK ORDER RELATIONSHIPS, 1942-1956 Hypothesis of Rank Order Stability Previous studies of individual voting habits Show the persistence of a person's tendency to vote for candidates of the same political party election after election. We are also aware Of the existence of SO-called one-party areas in which the candidates of a Single party monopolize election victories year after year. Assuming that voters tend to vote consistently for the candidates of the same party, we could picture the voters of an electoral district as consistently casting a majority Of their vote for the same party. This is in fact the case in the one-party districts of the "Solid South" where the real battle for election occurs in the Democratic primary. Such perfect consistency is not universal, however, as is evident from the changing partisan, character of elected bodies. The division Of Congress has never been the same from one election to the next; as illustrated by Table VIII. As a consequence Of the use Of Composite Districts, explained in Chapter II, this study covers a number Of congressional districts fewer than the legal total of 435. It can be seen in Table VIII, however, that the actual partisan con- trol of Congress is reflected in the study composition in seven out of the eight cases. In 1942, the exception, Republicans were actually in the minority, but the study composition shows a Republican majority. 43 .....i¢.¢.-A_ s) u.- sf... . cu.) .-- H | --‘~\.. an. Inn... 44 30303.0 3055 80.3 @3030 mc>flmucommnm0m mmm >30 0.8.5 0.8:» £03030 3%: 05 pohm .pm0> umflu .HOm 3?»me Hadowmmonmcoo June miwmcoomg‘r 80$ non—Home.“ 0.83 mandamus C0300? on 038003 .mofiocmOQr 03» 6mm; was 630G000> 030 new ~5th .2930: Ono :vmqfl $03589, 025. was ~3qu :HOfiO: 0G0 .NmoH $3300.? 000 was Kidd :HOCHOO: 0G0 .ommL trauma :MOAOO: 0:0 5me 300009» 000 was ~3an .1230: 0G0 .cvm; $9809? 000 paw 50353.9, paw m0>fimucmvmcnmom ~$nmm .1050: m0 035000, 033 ~§0>0 GM mmv no: one 3.30M; ~3qu :HOcSO: 03» €me 532500.? 0>E was 00>Emuc000Hmom >209 .1930: p.90.“ 0.33 0.8:“ Ni; 90h m.w¢ mém ma: 2% 0.34 o.mm com mmm ommfi .6154 bdm omH PK; v.34 m.mm mom NMN 4mm; m.mm New ova pm: w.om m.wv ANN :N mmoa m.:4 >.Nm omH N1: v.34 w.mm mm: 4mm 0mm: 0.34 v.~.m 2L mm; m.©m m.oo H: mcm mi: m.wm th om: ovH mOm Nimw mvN mm: 0:; 4.34 04m mm“: m: >64. ©.mm om; Nwm EAL mimm 5.34 o: hm: @414 Tom mom mam Nwofl .mcm .EOQ .Qom .EOQ diva .EOQ .mcm .EOQ 130% Mo unconcnm 1308 m0 ”.900qu use? GowfimOmEOO 35pm coflmeQEOO 3.3044 3150M cofioofim omadlmwod .Gofimcwflmofl ~3qu ~E m0>fid~d0m0na0m CO 0303 mmumum @333 05 mo COBROQEOO .HHH> 03mg. 45 If individual voting habits are persistent, this consistency should result in a relatively stable outcome of elections for each congressional district over time. For example, if congressional district A elected the Democratic candidate in 1942 and congressional district B elected a Republican candidate, we would place a higher probability on district A electing a Democrat in 1944 than district B electing a Democrat; it is more probable that most voters in both districts will vote the same in 1944 as in 1942 than that voters will change their votes. In the matter of balance of partisanship revealed by elections in each district, we would also expect a strong consistency. In other words, if the voters of district A cast 60% of their votes in the 1942 election for the Democratic candidate while the voters of district B were casting 60% of their votes for the Republican candidate, we would in turn expect the voters of district A to cast a larger percentage of their votes for the Democratic candidate in 1944 than would the voters of district B. Developing further this comparison between two hypothetical districts, each election outcome in each district places the district in a rank order relationship with all other districts, on the basis of the proportion of the voters that supported the major parties. In this study a Stalemate Index is computed for each congressional district and for each election from 1942 to 1956, inclusive. ‘ For each election the congressional districts have been ranked according to the size of the Stalemate Index. If all that has been said about the stability of voter preferences is valid, then it may be hypothesized that the Stalemate Index of each congress- ional district for each election, as it compares in size with the Stalemate Index of each of the other districts, will remain consistent enough so that the rank Order correlation coefficient between the rank orders of successive elections will be high. 1 lStatistical tests of significance would not be pertinent in this case. The non-parametric test of rank correlation coefficients tests the independ- ence of the correlated ranks. Here, we already know that successive elections in the same electoral district are not independent events. 46 The discovery of such a consistency in rank order relationships would be significant from a practical and a scientific point of view. - If the rank orders are sufficiently stable to serve as a basis for estimating future election prospects, the practicing politician could use such information in estimating which congressional districts are the most vulnerable to change and in allocating available resources accordingly. To the political scientist the discovery of high rank order correlations would be important largely as an addition to the existing body of descriptive information. It could serve as a sound exploratory test for more ”microsc0pic" investigation of individual voting habits. For example, assuming that high correlations between election years are common, the occurrence of a low rank order correlation coefficient between a particular set of elections would Show a realignment of congressional districts--and consequently a realign- ment of voters' preferences. This evidence could serve as sound justification for investigating further to discover the factors which cause realignment. A summary of election outcomes does not provide this type of descriptive information because it cannot isolate and identify a realignment in congress- ional electorate voting. A description of the mechanics Of change and the discovery of inconsistencies are possible only after locating the consistencies, and rank order relationships are one such measure of consistency and deviancy. , For example, a change in control of Congress from Democrats to Republicans at a particular election may have nothing at all to do with basic realignment in the electorate. The change might simply be the extension of an electoral trend over time. A low rank order correlation coefficient for the districts between this and the last election, however, would indicate a change involving something more than the prevailing trend. According to one political scientist, existing studies of voting behavior 1 may be classified into six basic types. These classifications include the 1Samuel J. Eldersveld, ”Theory and Method in Voting Behavior Research, " The Journal of Politics, Vol. 13 (February, 1951), pp. 70-87. 47 hypothesis-testing exploratory study, the mas s-tabulation case study, the comparative statistical survey, the single-hypothesis trend study, the hypothesis-testing factorial analysis, and the community dynamics type. We are attempting here the fourth type, the single-hypothesis trend study, more specifically described by Eldersveld as follows: A fourth category of research is the single-hypothesis trend study, in which the investigator, advancing a single proposition or an interpretation of one aSpect of voting behavior, explores its validity over a considerable span of elections and in many different electoral 1 units. Although Professor Eldersveld points out that the contribution of this type of study to theory construction has been negligible, he also points out that this approach in combination with the survey approach has been valuable: The trend- survey approach, however, is essentially valid, given hypotheses which are based on some objective facts, and systematic- ally pursued in a variety of research situations, with a rigorous technique. 7‘ As stated earlier, this is the objective of this study--t0 provide a valid basis upon which a survey investigation Of the realignment of voting preferences could be made. Computing Rank Order Correlation Coefficients There are 336 congressional districts, many of them Composite Districts, included in this study. The Stalemate Index has been computed for each district for each election from 1942 to 1956. Subsequently, each district has been ranked for each election according to the value of its Stalemate Index. For the period 1942 to 1956, inclusive, then, there re- sults eight arrays of congressional districts in which all are ranked from 1 through 336. For each election the district with the highest positive l1161.1. zlbid. 48 Stalemate Index (largest Democratic percentage) is ranked number 1 and the district with the lowest Stalemate Index (lowest Democratic percentage margin) is ranked number 336. The formula used for computation of the rank correlation coefficient is that devised by M. 0.~ Kendall. 1 Rank correlation coefficients were computed for three types of congressional election sequences--(a) for successive elections, (b) for successive presidential-year elections, and (c) for successive mid-term congressional elections. The correlation coefficients that resulted here suggest that, as far as the partisan rank order of congressional districts is concerned, there is not sufficient justification for examination of mid-term and presidential year elections as different types of election. .Tables IX and X Show the rank correlations between elections for all types of sequences to be uniformly high. If the rank correlation coefficients had been higher between successive mid-term elections, or between successive presidential year elections, than between successive elections, it would appear necessary to treat each sequence of four year intervals separately. This evidence would have indi- cated that the constituency groupings of voters do not react to mid-term 1M. G. Kendall, Rank Correlation Methods (New York: Hafner Pub- lishing Company, 1955), p. 38, Formula (3.8). . The formula is: 1/6 (N3-N) - 5(d2) - T'-U' P JF{1/6(N3-N)-2T') {1/6(N3-N)-2U')] where T' = 1/12 2: (t3—t) and U' = 1/12 § (u3-u) t and S(d2) = sum of squares of rank differences t = number of occurrences of tied ranks in first ranking u : number of occurrences of tied ranks in second ranking 49 Table IX. Rank Correlation Coefficients for Pairs of Congressional Elections in Sequences of Alternate Elections Presidential-year Elections ' Mid-term" Elections Rank Correlation RaEfCorrelation Year Coefficient Year Coefficient 1944-48 . 9105 1942-46 . 8859 1948-52 .9255 1946-50 .9217 1952-56 . 8822 1950-54 . 8949 Table X. Rank Correlation Coefficients for Pairs Of Successive Congressional Elections “—- m Rank Correlation Rank Correlation Year Coefficient Year Coefficient 1942-44 . 9100 1950-52 . 9286 1944-46 . 9234 1952-54 . 9369 1946-48 . 9478 1954-56 . 9184 1948-50 . 9307 congressional contests in the same manner as they react to presidential year elections. Such evidence would support Bean's thesis that “con- gressional elections in mid-term are not strictly comparable with congressional elections in the more exciting presidential years . . . they must be studied separately. " At least insofar as the present rank order evidence goes, whatever variation between on-year and off-year congress- ional elections does exist occurs fairly evenly spread among all districts thereby leaving rank order relatively unaffected. 50 Change in the rank order lineup of districts, however, is a very slow, plodding type of phenomenon. In fact, the coefficient is higher for each successive pair of elections than it is for each four-year interval pair of elections. The partisan lineup of districts changes less between 1942 and 1944 than it does between 1942 and 1946. The partisan lineup Of districts changes less between 1944 and 1946 than it does between 1944 and 1948. This pattern is constant for all elections in the period covered. There is in fact no instance in the period studied in which the change in rank orders between a successive pair of mid-term elections produced a higher co- efficient than the change between the first mid-term election and the succeed- ing presidential year election. Similarly, there is no instance in which the change in rank orders between a successive pair of presidential year elections produced a higher coefficient than the changes between the first presidential year election and the succeeding mid-term election. The stability in rank order relationships Shown by the high coefficients of Table X, then, is not a phenomenon in which on-year and off-year types of elections are involved. For our purposes here the coefficients indicate that there is only one type of election so far as ranking is concerned. Important Realignments of Congressional District Rank Orders The dimensions of change and stability in the partisan rank relation- ships of congressional districts are described here by correlation co- efficients and changes in the over-all composition of the House of Repre- sentatives. The coefficients point out certain basic voting consistencies and realignments in district rank orders. The over-all political compo- sition of Congress, of course, is evidence which major party has profited from any realignment of congressional constituencies. 51 For the period 1942 through 1956 there-were two elections in which the alignment of districts changedagreat deal.1 The lowest correlation co- efficient occurred for the rank changes between 1942 and 1944. , This coefficient was . 9100, indicating that there was an important realignment of districts in 1944. The result of the 1944 election in terms of the districts used for this study was the election of 173 Democrats and 162 Republicans (see Table VIII). wActual composition of the House in 1944 was 242 Democrats and 190 Republicans. The realignment of 1944, then, was to the advantage Of the Democratic Party. The other low correlation pair, 1954 to 1956, produced a coefficient of . 9184, second lowest of the seven pairs. The result was the election Of 173 Democrats and 163 Republicans (study district composition of the House). The primary significance to be drawn from these apparent realignment election years is the fact that they Show a greater than normal change within the universe of congressional districts. Certain of the districts within this universe assumed a political posture different from that displayed prior to the realignment year. And, more important, a sufficient number of these districts have altered their partisan complexion so that the total universe of districts is significantly different. If, because of the great stability in rank orders, the rank order of congressional districts are used as a basis for forecasting the possible outcomes of future elections, a low correlation pair of elections would indicate a period of realignment in the electorate. Rank orders previous to the realignment year would be less serviceable as a reliable forecasting base. 1Since the range of coefficients in Table X runs from . 9100 to .9478, all relatively high correlations, even the election pairs with the lowest coefficients must be viewed as having high consistency in ranks. 52 The 1942-44 election pair produced a low coefficient of . 9100 and a gain in the number of seats held by Democrats. The following election pair, 1944-46, produced a higher coefficient of . 9234 and a large gain in the number of House seats held by Republicans; in 1946 House composition stood at 196 Republicans and 140 Democrats. This combination of factors would indicate that, following the 1944 realignment, the districts remained in a relatively stable rank order, but there was a uniform movement of the entire array of districts toward the Republican party. The greatest stability in rank order for the period studied occurred between the elections of 1946 and 1948. For this election pair, the co- efficient was . 9478. Democrats gained seats to hold a plurality of 193 to 143 for the Republicans. Again, the rank order remained consistently firm while there was a relatively uniform movement among the districts toward the Democratic party. Another high correlation Was produced in the 1948-1950 election pair. With a coefficient of . 9307 the Democratic majority was reduced to 18 - seats. The party alignment in the House was 177 Democrats to 159 Republi- cans. There had been very little realignment Of districts and only a moderate movement toward the Republican Party. - In 1952 the trend of the previous election toward the G. O. P. was con- tinued. Republicans took control of the House by a margin of 179 to 159. The realignment of districts was again relatively small, however, as shown by a coefficient of . 9286. From 1952 to 1954 the composition of the House was completely re- versed. ' In the latter year the Democrats regained control of the House by a margin of 177 to 159. The coefficient of correlation produced by this pair of elections was the second highest of all the years studied. A co- efficient Of . 9369 is evidence that 1954 was not a year of realignment, but one in which there was an across-the-board gain for the Democratic party. 53 The final election in the study period, 1956, produced the second greater than normal realignment. A relatively low coefficient of . 9184 coupled with a relatively stable composition of the House meant that the reshuffling of districts was now the beginning phase of a major shift away from one, the other, or both parties. The Democratic party still retained its majority in seats by a 173 to 163 count. Although, in summary, the Democrats had a net loss of three seats to Republicans in 1956, this in itself is not sufficient evidence to conclude that the realignment favored the Republican party. By recalling that the 1944 realignment, which was accompanied by a Democratic gain, was followed by the election of a Republican-controlled House in 1946, .we cannot assume that the 1956 election results point to a mounting Republican trend. In fact, the correlation coefficients produced by the 4-year interval election sequences indicate just the opposite. The changes in the rank order of congressional districts between 1952 and 1956 were sufficient to produce a coefficient of . 8822, the lowest of any in the 4-year interval comparison (see Table IX). Tliis realignment between presidential election years was accompanied by a Democratic gain of 16 seats in the House, good evidence that the 1956 election was not a harbinger of happy days ahead for the Republican party. (The previous description of 1944 and 1956 as important realignment years is Supported by the data showing 4-year interval correlations. The lowest correlation coefficient for (any of the mid-term election com- parisons was produced by the 1942-1946 pair (.8859). The lowest co- efficient for presidential year pairs came between 1952 and 1956 (. 8822). Evidence to support the hypothesis that this type of realignment is not a phenomenon of shorttenure is also present. The low correlation between 1942 and 1944 was not followed by a return to the 1942 rank order. We see this in the low coefficientproduced by comparison of 1942 and 1944 finished at the time'of the 1944 election. A higher, but still relatively low, 54 coefficient of . 9234 between 1944 and 1946 indicated that the realignment continued on after the 1944 election. ~ In the 1956 realignment, the rank changes were not Spread out over a four-year period as was the case in 1944. Whereas the correlation be- tween 1952 and 1956 produced a very low coefficient of . 8822, the rank changes between 1952 and 1954 produced a high coefficient of . 9369. The big change or realignment, then, came in the two yearsibetween 1954 and 1956 as the relatively low coefficient (. 9184) indicates.1 The rank order analysis of all the congressional districts used in this study has illustrated a phenomenon which would elude any trend study of congressional elections based upon the rise and fall of party fortunes in Congress. Where studies which use party membership in the U. S. House of Representatives as an indicator of partisan trends fail to highlight re- alignments in the universe of congressional districts, the correlation of rank order relationships does provide a measuring rod for district realign- ments. The analysis in this section has isolated election pairs that we have accepted as “important realignments. " Consistently high rank correlations have made it difficult, however, to accept these realignments as basic realignments. The relatively small difference between the highest and lowest coefficients point to the absence of any basic rank change between election pairs. . In the next section we will employ a reduction of ranks to further investigate the possible occurrence of basic realignments. It has been shown that the partisan rank order of congressional dis- tricts is very stable from election to election. For purposes of illustration, this finding could be simplified by picturing a permanently fixed rank order of units over time. In this illustration the unit ranked number 1 always 1This interpretation of the 1956 realignment as one in which important changes occurred in a two-year period depends, of course, on the corre- lation between 1956 and 1958. It could be possible that the 1956-1958 coefficient of correlation was also low indicating a continuation of the 1956 realignment. 55 rates higher than unit number 2 on the basis of the criterion of classifi- cation. Similarly, unit number 2 always rates higher than unit number 3 over time. In other words, were these units correlated over consecutive time periods, the coefficient would always be 1. 000, indicating perfect correlation. The criterion of classification employed here is the Stalemate Index. If we were to assume perfect correlation of congressional district rank orders over time, the dimensions of change in congressional elections would involve only the movement of all districts in a single, monolithic shift up and down the fixed array of districts. We have seen from the correlation coefficients in Tables IX and X that the assumption of perfect correlation is not valid. It has been Shown, however, that almost perfect correlation between district rank orders does exist. The greatest number of districts retain a stable rank order in relation to the other districts. Only a few of the districts exhibit any radical change in position. It is to these districts that we will look in the next chapter to locate the causes of congressional electorate realignment. Reduction of Ranks to Locate "Competitive" Districts In the preceding section, the correlation coefficients of consecutive pairs of elections were examined. The purpose was to locate that pair, or pairs, of elections during which a greater than normal partisan realign- ment of districts occurred. Low rank order correlation coefficients were used as indicators of such realignments. Examination of the coefficients showed so little fluctuation in rank correlation that realignment years so identified were suspect. An appar- ent cause of this correlation stability was the inclusion of non-competitive districts that retained a relatively stable rank in each of the eight elections. These districts were responsible for a low level of rank differences and consequently for a high coefficient. 56 In this section a reduction of the universe of districts will be accomp- lished in order to isolate more certainly the realignment years. Reduction of ranks will make the correlation coefficient a more sensitive instrument in detecting the changes in partisan alignments of congressional districts. Such a reduction can be accomplished by removing on logical grounds certain groups of districts from the rank orders in successive steps. There are distinct groupings of the congressional districts studied here. Some of the districts have remained unchanged by redistricting through the entire period, 1942-1956. Others have been arbitrarily held constant according to their 1956 boundaries. Still others, the Composite Di. stricts, are actually "artificial" districts employed to prevent waste of available data. All of these districts fall into distinct groups because of redistricting, or lack of it, during the study period. There are two other obvious classifications which separate the universe of districts. These classifications are produced by the actual election re- sults. Some districts can readily be classified as ”one-party, " such classification thereby dividing the universe into two groups. The distinct groups of districts, described below, will be removed from the rankings in the following order by steps:l Step 1. . Remove one-party districts. The definition of a one-party district used here is “a district that has had a Stalemate Index of 20. 0 or greater in at least five of the eight elections. An additional condition is that one party must have won all eight of the elections in that district. " In this step, 93 districts are removed, 243 districts remain. Step 2. Remove all Composite Districts. In this step, 25 additional districts are removed, 218 districts remain. Step 3. Remove all districts that have been changed by‘redistricting during the 1942-1956 period. These districts have been previously treated 1Appendix C contains a roster of districts removed in each of the first three steps. Those remaining after step 4 are listed in Appendix D. 57 as unredistricted by reconstructing their actual 1956 makeup back through the whole period. In this step, 64 additional districts are removed, 154 districts remain. Step 4. To determine influence of highly competitive districts, remove all except those that have been won’tby one of the major parties not more than five of the eight elections. In this step, 132 additional districts are removed, 22 districts remain. To isolate the factors contributing to realignment years more clearly than was done in the preceding section, those districts with high rank stability must be identified and removed. To find districts with high rank stability, rank order correlations must be computed before and after removal of each group of districts. A decrease in the computed coefficient for the districts remaining after removal of a group will mean that the eliminated group had been contributing to high rank stability for the uni- verse in that election pair. The end result of this removal process will be to "give the, data a chance"--a chance to show when the districts are actually realigning in rank order. After the removal process the districts remaining should be those that have shown a partisan flexibility, those actual districts in which true competition and change have occurred during the eight elections. The eXperience of elections in these districts is that either major party can win an election. By definition, the one-party districts do not qualify as districts in which competition occurs to any great degree. In Step 1, these districts were removed, and an illustrative spot-check made of the effect of their removal. A rank order correlation coefficient was computed and found to be . 802 for the 1942-1944 pair of elections, after the one-party districts had been removed. ‘ In Chapter III it was Shown that the rank order corre- lation coefficient for the 1942-1944 was . 910, with all one-party districts included. By removing one-party districts, then, the coefficient 58 was reduced from .910 to .802. The one-party districts, as expected, contribute much to increase rank order correlation, and their removal is justified. In Step 2, the Composite Districts are also removed. The computed coefficient after the removal of these districts is . 792, or just slightly less than the . 802 produced in Step 1. The removal of these districts, then, is only moderately justified; the relatively minor reduction from . 802 to . 792 means that these districts probably have little effect on degree of rank correlation. It will be recalled that Composite Districts consist of many actual districts lumped together because collection of election data for the com- ponent districts was ,not possible. The apparent result of this lumping was to cancel out the more extreme partisan voting elements of the lumped districts leaving the artificial districts stable in rank. In any event, the Composite Districts are not actual districts. Their removal seems additionally justified as a precautionary measure. To continue to include them could, because of their "artificiality, " color the validity of the final results. In Step 3, all districts changed by redistricting have been removed in addition to those withdrawn in Steps 1 and 2. These Step 3 districts are the ones altered by redistricting but treated as if their 1956 makeup had been constant Since 1942. Rural-Urban Character of Post-Step 3 Districts It is significant to note what kind of district it is that remains after the Step 3 removal. If only predominately rural areas are retained in the analysis following the removal process, the results of the analysis itself would have to be viewed with certain limitatiOns. In such a case, it would have to be concluded that rural districts generally escape redistricting, 59 influence heavily the analysis of unredistricted congressional districts, and limit our post-Step 3 analysis to a small rank correlation study of rural congressional districts. Fortunately, it was discovered that the post-Step 3 group of districts contained a fair distribution of all types of composition as described by the rural-urban character Of the districts. Classification of the districts remaining after Step 3 removal according to rural-urban status shows the types of districts with which we are dealing. Identification of the rural-urban character of each of the 154 remaining districts is provided in Appendix D. The classification employed here is the same as that used by the Congressional Quarterly. 1 1Congressional Quarterly Almanac, Vol. XII, 1956, p. 788. Accord- ing to the 1950 Population Census definition, the urban population includes all persons living in places of 2, 500 or more inhabitants, and in the densely settled urban fringe around cities of 50, 000 or more. The re- maining population is rural. Because the term urban is applied indiscriminately to tiny villages and huge metropolitan centers, CQ'S classification of Congressional dis- tricts takes into account city size as well as the percentage of urban residents. They are defined as follows: Class 1. Rural district. General characterization: predominantly rural. Specifically: (a) At least two-thirds rural; or (b) One-half to two- thirds rural, with no city of 25, 000 or more population. Class II. Small-town district. General characterization: substantial rural population but with one or more cities of 25, 000 to 50, 000 population. Specifically: (a) One-third to one-half urban, and with a city of 25, 000 to 50, 000; or (b) More than one-half urban, but with no city of 50, 000 or more; or (c) One-third to one—half urban with a city of 50, 000 or more, the city having less than one-third the total district population. Class III. Mid-urban district. General characterization: sub- stantially influenced by a city of 50, 000 to 200, 000. Specifically: (a) More than one-half urban, and with a city of 50, 000 to 200, 000; or (b) One-third to one-half urban and with a city of more than 50, 000, the city having more than one-third of the total district population; or (c) One-half to two-thirds urban and contains or is partly contained in a city of 200, 000 or more. 60 The summary figures Show that there are 28 rural districts (Class I), 48 small town-districts (Class II), 44 mid-urban districts (Class III), and 34 metropolitan districts (Class IV) left after the Step 3 removal. These districts remaining after Step 3 were all untouched by redistricting from 1942through1956.l Analysis of Rank Order Relationships Excluding the One-Party, Composite, and Redistricted Districts The correlation coefficients for consecutive pairs of elections, following the Step 3 removal, were computed for Table XI. These coeffic- ients used all districts except: (a) one-party districts, (b) Composite Districts, and (c) redistricted districts. Comparison of the 1942-1944 coefficient after Step 3 removal with the coefficient after Step 2 removal shows a further decrease in rank correlation. The new coefficient is .747. The step 2 coefficient was . 792; the Step 1 coefficient was . 802; and the coefficient for all districts was . 910. Further comparison of the Step 3 coefficients with those computed prior to the reduction shows a decrease in the size of the coefficient for every pair of elections (see Table XI). Class IV. MetrOpolitan district. General characterization: pre- dominantly “big-city“--including metrOpolitan suburbs. ‘ Specifically: (a) More than two-thirds urban, contains or is partly contained in a city of 200, 000 or more; or (b) More than two-thirds of the population lives in the urbanized area of a city of 200, 000 or more. A number of one-party districts, those removed in Step 1, also were not redistricted. Of the unredistricted, one-party districts, 24 were Class I, 21 were Class II, 15 were Class III, and 12 were Class IV. Of all the 226 unredistricted districts, there were 52 Class I, 69 Class II, 59 Class III, and 46 Class IV districts. While Class II or "small town" dis- tricts were the largest in number of the unredistricted cases, the distribu- tion of cases among the four classes is clear evidence that no one class of district escapes redistricting significantly more than the other classes. llbid. 61 Table XI. Comparison of Rank Order Correlation-Coefficients Produced by Ranking of Total Universe of Districts and Those Produced After Step 3 Removal"< 1 Pairs of Rank Order Correlation Election Coefficients Years All Districts » Step 3 Districts Change 1942-1944 .910 .747 -.163 1944-1946 .923 .802 -.121 1946-1948 .948 .865 -.083 1948-1950 .931 .851 -.080 1950-1952 .929 .841 -.088 1952-1954 .937 .838 -.099 1954-1956 .918 .736 -.182 *Following the Step3 removal only those districts which are not classified as one-party, composite, or which have not been redistricted during the study period remain. This leaves a total of 154 districts remaining in the universe. The differences between the coefficients computed after Step 3 and those for the whole universe highlight. a variable effect of the removed districts on rank stability. In every case the one-party districts, the Composite Districts and the redistricted districts served to keep the rank order correlation high. But the effect was not uniform for every election pair (see Table XI). Between 1942-1944 and 1944-1946, the original coefficient increased . 013 and the Step 3 coefficient increased .055. This would mean that the removed districts actually kept the original coefficient higher than it would have been without them. Between 1944-1946 and 1946-1948, the same relationship held true. Between 1946-1948 and 1948-1950, the original coefficient dropped .017 while the Step 3 coefficient dropped only .014. The conclu- sion here is that the removed districts caused the original coefficient to 62 be lower than it would have been without them. Between 1948-1950 and 1950-1952, just the Opposite was true. Removed districts kept the original coefficient from dropping more than it did. Between 1950-1952 and 1952-1954, a situation existed where the original coefficient was increasing by .083 while the Step 3 coefficient dropped . 003. The removed districts, then, were highly rank stable between this pair of election pairs. The districts removed in Step 3 were largely responsible for the original high correlations. When they were included the coefficient increased. When removed, the coefficient dropped. Between the final pair of election pairs, 1952-1954 and 1954-1956, the original coefficient drOpped .019. The Step 3 coefficient dropped . 102. The post-Step 3 group of districts, then, was experiencing an internal shuffling of ranks more drastic than that experienced by the original group of districts, indicating that Step 3 districts were heavy contributors to the relatively low coefficient of the original group. The post-Step 3 rank order correlation coefficients in Table XI cluster around an average . 810, indicating that the removal of districts by Step 3 was not sufficient to magnify evidence of basic realignments, if any actually did occur, from 1942 to 1956. Where the lowest coefficients were produced on the total set of data, the lowest coefficients were pro- duced in Step 3. But the changes in the coefficient from one pair of elections to the next remained just about as small as when the coefficients were computed using all districts. The three-step reduction of ranks has provided supporting evidence for the findings produced by rank correlation analysis of the original group of 336 districts. The low coefficient between 1942 and 1944 for both the original and the Step 3 correlation indicates an atypical realignment of districts in 1944. The finding of a low coefficient for the 1954-1956 election pair in the post-Step 3 analysis also supports the findings of the original (336 district) analysis that the 1956 election was one in which an important realignment of districts occurred. 63 Table XII whichcompares the Step 3 and the original changes in coefficient size over time shows the close relationship between rank changes within the universe of post-Step 3 districts and rank changes within the original universe of 336 districts. Table XII. Comparison of Rank Correlation Coefficient Changes for Pairs of Election Pairs-—Between Coefficients for all Districts and Coefficients for the Step 3 Districts Pairs of Rank Correlation Coefficient Changes Election Pairs All Districts Step 3 Districts 1942-44 to 1944-46 + .013 + .055 1944-46 to 1946-48 + .025 + .063 1946-48 to 1948-50 - .017 - .014 1948-50 to 1950—52 - .002 - .010 1950-52 to 1952-54 + .008 - .003 1952-54 to 1954-56 .019 — . 102 The distinction between the removed districts' responsibility for high rank correlations and their responsibility for fluctuations (or changes) in size of coefficients should be carefully considered, however. In every pair Of elections, the removed districts were responsible for keeping the rank correlation coefficients above . 900. Generally, the one-party, Composite and redictricted districts are a stabilizing influence on the partisan rank order of congressional districts. Test of Permanence and Rate of Realignments for Post-Step 3 Districts A different basis of comparison was applied to the 154 districts remaining after Step 3 reduction of ranks. Correlation coefficients were computed to measure the rank stability between 1942 and each of the other 64 elections. This procedure leads to the development of a coefficient trend line which has 1942 as its source. The advantage of employing such a procedure was that it provided an indication as to the rate of change in coefficients from election to election. It also provided information as to whether or not successive rank changes represent temporary or perman- ent realignments of congressional districts. The coefficients produced by the correlations between 1942 and the other elections are contained in Table XIII. Table X111. Rank Correlation Coefficients for Districts Remaining After 1 Steps 1, 2, and 3--Using 1942'as Base Year Pairs of Election Years Rank Correlation Coefficient 1942-1944 . 747 1942-1946 . 656 1942-1948 . 572 1942-1950 .643 1942-1952 . 578 1942-1954 .458 1942-1956 . 306 The regression of the coefficients computed, using 1942 as the base year, has but one "hitch" in it. While the coefficient dropped from . 656 in 1942-1946 to . 572 in 1942-1948, the correlation moved back up to . 643 in 1942-1950. After the 1942-1950 pair, the regression continued smoothly downward once again. This “hitch” in the regression indicates that the realignment of districts from 1946 to 1948 was only temporary because the 1950 rank order was almost identical to that of 1946. The 1948 realignment took on 65 permanent characteristics in 1952 since there was a difference of only . 006 between the coefficients of these two years. In summary, Table XIII Shows the existence of relatively high and stable correlations until 1954. Overall, the 1942 rank order pattern experienced a slow but steady disintegration until big changes began to appear in 1954 and 1956. It is apparent from the series of midterm election coefficients that the 1942 alignment was almost completely dis- integrated following the 1954 election. . Focusing on presidential year elections, the 1942-1956 comparison shows a big rank change which must be recognized as a real departure from the political alignments of congressional districts in the era preceding the election of 1956. Analysis of Election Data for Competitive Districts Step 3 removals did not produce clear evidence where basic realign- ments of congressional districts occur. As a result, a further reduction of ranks was undertaken. In Step 4 of the reduction of ranks, all remaining districts have been removed except those in which neither major party has more than five election victories in the eight elections covered. This means that the computation Of rank correlation coefficients in Step 4 will be concerned only with an arbitrary but defensibly defined group of "competitive" dis- tricts in which one major party (Democratic or Republican) won at least three but not more than five elections. There are twenty-two competitive- districts left after Step 4 removals (see Appendix C-3 for description). The rank correlation coefficients produced by the array of com- petitive districts are presented in Table XIV, which also includes a comparison with the coefficients computed using all districts. The "change" column shows the difference between the original coefficients and those produced by Step 4 districts; that is, the relative "amounts" of the original coefficients attributable to the four groups of removed districts. 66 Table XIV. Rank Correlation Coefficients of Districts Remaining After the Step 4 Reduction of Ranks-~For Consecutive Elections Rank Correlation Pairs of Mid-term Pres. to Coefficients Election Years to Pres. Mid-term All Districts Step 4 Changes 1942-1944 X .910 .888 -.022 1944-1946 X .923 .353 -.570 1946-1948 X . 948 .668 -. 280 1948-1950 X .931 . 302 -.629 1950-1952 X .929 . 349 -.580 1952-1954 X .937 .767 -. 170 1954-1956 X .918 .801 -.117 Analysis of the Step 4 correlations shows a phenomenon which was not present in the Step 3 analysis--a distinction between mid-term and presidential year rank orders. It also highlights a phenomenon to which many political scientists have directed their studies, that of the presidential coattail effect. Low correlations from midterm to presidential year are an indication that the presidential coattail effect was a significant factor in the congressional elections. Employing the guideline presented in the preceding paragraph, Table XIV shows that the coattail phenomenon operates in some elections, but not in others. For the 22 competitive districts, the midterm to presidential year coefficients were high for 1942-44, 1946-48, and 1954- 1956. The conclusion would have to be that the coattail-pulling power was weak in 1944, 1948, and 1956. In 1952, however, as shown by the low correlation of . 349 between 1950 and 1952, the power of the presidential candidate to alter the regular pattern of district realignment is readily apparent. 67 The findings presented in Table XIV Show that the coefficients of the 22 competitive districts follow an alternating pattern of change and stability. This alternating pattern is one in which the mid-term to presidential year coefficient is high, but followed by a low coefficient from presidential year to mid-term. The exception to the general pattern found is between 1950 and 1952. The previous coefficient (1948-1950) was found to be . 302, showing a relatively low rank correlation. ‘ If the 1950-1952 coefficient had fit into the pattern Shown by all other mid-term to presidential year coefficients, it would have risen. The mid-term rank order would have remained quite stable over to the following presidential year. On the contrary, the 1950-1952 coefficient was . 349, or almost equivalent to the previous coefficient (. 302). Out of the entire array of coefficients produced by competitive district rankings, then, only the 1950-1952 comparison was atypical. Not only was the coattail of Dwight D Eisenhower sturdy enough in 1952 to carry Republicans into control of the House of Representatives, but it also was strong enough to cause an important partisan realignment of congressional district rankings. In general, this Step 4 analysis of what we have defined as "competitive" districts shows that it would be better to forecast from the mid-term election results than from presidential-year election results, if rank order realignment is to occur in the mid-term election year. When a significant realignment of our “competitive" district occurs at the presidential-year election, it indicates an unusual reshuffling of partisan realignments of the congressional districts which can be attributed to the coattail power of a presidential candidate. Test of Permanence and Rate of Realignments-- Competitive Districts The rank correlation between 1942 and each of the other election Years illustrates two atypical realignments in the 11942-56 period (see Table XV). The regression of coefficients did retain a fairly stable 68 pattern of drOp-off except between the 1942-1946 and 1942-1948 pair and between the 1942—1954 and 1942-1956 pair. Table XV. Rank Correlation Coefficients of 22 Competitive Districts Remaining After the Step 4 Reduction of RankS--Using 1942 as Base Year Pairs of Election Years Rank Correlation Coefficient 1942-1944 .888 1942-1946 .217 1942-1948 .215 1942-1950 -.015 1942-1952 -.490 1942-1954 -.803 1942-1956 -.682 The coefficient for the 1942-1946 correlation was . 217. The co- efficient for the 1942-1948 correlation was . 215, indicating that there was very little rank realignment between 1946 and 1948. Recalling that the previous Table (Table XIV) showed a relatively high correlation between 1946 and 1948, further support is added to the assumption of rank stability between these two years. Between the 1942-1954 and 1942-1956 pair, the coefficient actually jumped from -. 803 to -. 682. This change is definitely atypical, being the only increase between consecutive pairs of election pairs. It could be hypothesized from this occurrence that the election of 1956 was a major realignment of rank order for these Step 4, ”competitive" districts. It would seem that the correlation with 1942 had reached its low point in 1954, Showing an almost complete negative correlation, and had begun a 69 return to the 1942 rank alignment in the election of 1956. Unfortunately, in terms of testing this hypothesis, the 1956 election is the last of the period studied and it is impossible to see if future elections would follow this upward trend. The constant regression in the "1942-Base" coefficients of Step 4 competitive districts points to the conclusion that the mid-term to presidential year realignments found in "consecutive-election" corre- lations are relatively permanent realignments. While the partisan rank— ing of competitive districts is constantly changing, as compared to the 1942 base year, the general pattern is high stability from mid-term to presidential year congressional elections. The original hypothesis of this section was that basic realignments of the partisan rank order of congressional districts do occur. Low rank order correlations were to be indicative of such basic realignments, presumed to occur suddenly and all at once. But just the opposite was found to be true. The "normal” pattern found was a continuing realign- ment going on in successive elections. A constant change in the rank alignment of these competitive districts should be expected, according to Table XV. A period of stable rank positions is, in fact, atypical. Party Victory in Competitive Districts Rank relationships and the fluctuation of rank orders of districts through time present only a partial picture of party competition. Whether or not a district actually elects a Democrat or a Republican is in the final analysis the most important fact of the election. I Examination of the election outcomes for Step 4 competitive dis- tricts Shows that had Congresses been elected only by Step 4 districts during the 1942-1956 period, these Congresses would have been in almost every case, controlled by the same party winning a majority of all 70 districts used in this study, that is, the ”Study Districts” (see Table XVI). The actual control of Congress is reflected in the Step 4 outcomes in all but two cases, 1942 and 1956. In 1944, for example, Democrats actually held 55. 6%of all House seats while 63. 7% of the 22 competitive districts were won by Democrats. In 1952, when Democrats actually won only 48. 5% of the House seats, 40. 9% 0f the competitive districts went Democratic. In all cases, however, the changes in party strength in Congress were accurately reflected by changes in party fortunes among Step 4 districts. A closer examination of partisan changes in the competitive dis- tricts is presented in Table XVII. Between 1942 and 1944, Democrats picked up strength ”across the board. ” In 1944 they won 4 seats held by Republicans in 1942 and increased their margins of victory in 7 districts having a Democratic majority in 1942. The Democrats even gained strength in six districts won by the Republicans in 1944. In 1946, the G.O.P. regained its 1944 losses, taking 9 seats previously held by the Democrats. But in 1948 the Democratic Party consolidated its 1944 gains and won a victory from which the Republicans had not recovered by 1956. The Democrats took 16 competitive districts from the Republicans in 1948. The 1952 election was a big one for the Republican party. It took 10 districts from the Democrats and picked up strength in 5 districts retained by the Democrats. In the next 4 elections the Republicans were able to take only 14 districts from the Democrats. But during the same period, the Democrats took 5 districts from the G. O. P. So, the net G.O. P. gain for the four elections was only 9 districts. Although the Republicans gained strength in 20 districts won by the Democrats from 1950 to 1956, their gains were not resulting in election victories. 71 Table XVI. Comparison of Actual, Study and Step 4 Competitive Districts--Percentage of Districts Won by Democratic Party Percentage of Seats Democratic Actual Study Step 4 Election Districts Districts Districts Year (435) (336) (22) 1942 50.1 46.7 45.4 1944 55.6 51.6 63.6 1946 43. 2 41.7 22.7 1948 60.5 57.4 95. 5 1950 53.8 52.7 86.4 1952 48. 5 46.7 40.9 1954 53. 3 52.7 50.0 1956 53.6 51.5 54.5 (The tendency of Competitive districts is to swing in the same di- rection as the totalyuniverse of districts, but to swing further. Thus, in 1952, Competitive districts gave Democrats 45.4% of the 22 seats and in 1944 the Democrats won 63.6%. This was an increase of 18. 2 percentage points between the two elections. Between the same pair of elections, the actual composition of Congress moved from 50. 1% to 55. 6% Democratic, a gain of 5. 5 percentage points. And the composition of Congress as shown by all of the Study districts (336 districts) showed an increase of 4. 9 percentage points for the Democrats. The relationship between rank correlation coefficients and the changes in party majority in Congress (as shown by Step 4 Competitive districts), is one in which the party that wins the presidency gains seats when it wins the presidency; high correlations occur between mid-term to presidential year election pairs except when a presidential candidate is able to interject a strong coattail influence on congressional election “vi-u....nv us. 1- roe an.-. .1...iv >1. 19 r. , stupemvu Im,4wt\u~ ~v-flv~pvflfl .v-- y~«..\ .- sue-An pvtl-s.sru..~a!-..w.v .11. .u. .,L.-u{-..-HV A....h.-.-..An ~a» -.~».~m.-.A§---eg.-R~ lhfi) . Yr‘ liv~nv§91~ / 72 h o w. H N N H o ommHuwmoH m w H m. o m 0 0H HumoHtNmmH N H4 m 0 OH O NH H4 NmoHtomoH H m OH O N o MH «0 ommHtwvoH o m H H 0 0H N oN wvoHtovoH m H H4 H o 0 0H 0 oon #4me N N. m o c He. m 5H won tNHsoH .mom .EoQ .mom .EoQ .mom .EOQ mchU 3500 93.0%. 2830on UHOHIH 3000.314 ~3qu MOHHHO >3 no? ~3pdnH .HOHHHO .nHom .Eofl H0 muHmnH mHOHnumHQ SH :30 HoHuumHQ Rom .GHmO Eonh modem CH3 mommoiH .Ho mGHmO 3.5mm H0 monbmh. 1 m0 odoupfio 00 H0 .HOnHESZ oqutNHAoH HooHuonH of .HOH owfldflnv HO HQHUNHNHU >3 muUwHumwQ ®>wufiu®QEOU NN Gun m0Mfld£U Edmwuhmm HO COfl—Sflahumaa .HH>VA QHQMH. 73 outcomes; and low correlation occurs between presidential to mid-term election pairs except when the preceding mid-term to presidential year pair correlation is low (see Table XIX). In each of the three election pairs, 1942-44, 1946-48, and 1952-54, a high correlation occurred with Democratic gains. In each of the three election pairs, 1944-46, 1948-50, and 1950-52, a low correlation occurred with Republican gains. The 1954-56 election pair did not Show conclusive changes. The correlation was very high (. 801) and Republicans gained strength in 13 districts while Democrats gained in 9 districts. But each party took two seats from the other. In short, the period 1954-56 was one of very little change for congressional politics. The rank order remained stable. There was very little change in the partisan composition of Congress. And both parties remained fairly invulnerable in districts previously held. All of this indicates an unusual period in view of the changes exhibited during other 2-year periods. Every change in the actual party composition of Congress is, there- fore, substantially contributed to by the party division of competitive districts (see Table XVIII). The agreement in partisan direction of the changes is important, and suggests the political importance of these dis- tricts. For example, if one could forecast the election results of the competitive districts, the agreement between Actual and Step 4 changes in Table XVIII suggests that this forecase would provide a basis for esti- mating the actual party composition of Congress. In summary, then, rank realignment of the array of competitive districts has been in favor of the Republican party while subsequent rank stability has been characterized by Democratic gains. The 1956 election is the only one of the series in which increased rank stability was not accompanied by Democratic gains. The Republican victory in the presi- dential election of 1956 could well explain such variation from the normal pattern. 74 Table XVIII. Comparison of Changes in Percentage of Seats Held by Democrats--Actual Districts, Study Districts, Step 4 Districts Pairs of Change in Percentage of Districts Held Election Actual Study Competitive Years Districts Districts Districts 1942-1944 + 5.5 + 4.9 +18.2 1944-1946 -12.4 - 9.9 -40.9 1946-1948 +17.3 +15.7 +72.8 1948-1950 - 6.7 - 4.7 - 9.1 1950-1952 - 5 3 - 6.0 -45.5 1952-1954 + 4.8 + 6.0 + 9.1 1954-1956 + 0 3 - 1.2 + 4.5 Table XIX. Comparison of Rank Correlation Coefficients and Changes in the Partisan Distribution of Step 4 Districts Pairs of Coefficient of Number of Occurrences Election Consecutive Dem. Rep. Years Elections Gains Gains 1942-1944 . 888 17 5 1944-1946 .353 6 16 1946-1948 . 668 20 2 1948-1950 . 302 9 13 1950-1952 . 349 4 17 1952-1954 . 767 16 6 1954-1956 .801 9 l3 75 Conclusions The hypothesis that the partisan rank order of congressional dis- tricts is highly correlated between successive elections has been found to be valid. The fact that rank correlations between election years have been shown to be consistently high casts serious doubt on our original con- tention that low correlations would emerge upon occasions of basic party realignments in the congressional electorate. A reduction of ranks was employed to make the correlation co- efficient a more sensitive indicator of important realignment years. In successive steps it was shown that the one-party, the Composite, the redistricted, and "non-competitive" districts were contributors to high rank correlations. While the groups of districts removed in the reduction Of ranks contributed to high correlation in every case, each group had a variable effect on the changes that occurred between successive corre- lation coefficients. In‘some cases a group kept the coefficient relatively stable between successive election pairs. In other cases a group of dis- tricts was reSponsible for significant changes between successive com- puted coefficients of correlation. A series of correlations which compared the rank order of 1942 with that of each of the other elections provided evidence of the permanence of rank realignments. In each successive correlation except one, the correlation coefficient decreased-~the gradual realignments were not temporary diversions from the 1942 alignment. The single exception occurred between the 1942-1946 and 1942-1948 comparisons. While the 1948 rank order more closely paralleled the 1942 alignment than did the 1946 rank order, the gradual realignment continued on its course in 1950. The step 4 reduction of ranks, which eliminated one-party, Composite, redistricted and "non-competitive" districts, revealed a 76 distinction between mid-term and presidential year election rank orders. Comparisons showed a high mid-term to presidential year correlation and a lower presidential to mid-term correlation. This analysis indicates that forecasting based on rank orders is apt to be more accurate from mid-term to presidential year elections than the reverse. To be successful the forecast must be made on the basis of historical political facts, in this case the recent partisan rank orders of congressional districts. In addition, the evidence discovered here that presidential-year rank orders tend to reflect closely the preceding midterm year rank orders is a good indication that such a reflection will hold true in the future. - Such evidence adds sound support for rank order forecasts. The congressional district rank correlation device has provided new information relative to a Special facet of congressional politics, the presidential coattail effect. It has been shown that this method can locate the elections in which such an effect has been influential. A low rank correlation from mid-term to presidential year has been shown to indicate the existence of coattail power. Also, it has been discovered that the existence of this coattail phenomenon affects not only the partisan division of congressional seats won, but also the stability of the entire partisan rank order of congressional districts. Actual changes in partisan composition of Congress are found to be reflected in the partisan division of competitive districts. This is good evidence that the ”competitive" districts comprise a fulcrum for the entire array of congressional districts. Estimating future congressional election results by making trend analyses of competitive districts obviously is strategically sound from a research as well as practical point of view. In Chapter VI experimentation with forecasting election results of com- petitive districts will be attempted against the background of the rank order phenomena discovered here. CHAPTER V DISTRICTS RESPONSIBLE FOR LOW RANK CORRELATION COEFFICIENTS Rank correlation coefficients have added new information to the description of congressional elections. Composition of the House by party designation is not of itself sufficient evidence to tell the complete story of congressional partisan trends. Rank orders and their changes over time have isolated the outline of electoral realignments. A more complete description of the mechanics of change should result from an examination of those districts that contribute most to a less-than-perfect correlation. The success of all efforts at forecasting depends upon reducing the unexpected to the expected. Further insight into the details of change in the voting decisions of congressional electorates may also provide new guidelines to investigation of the political process. Identification of the Error Districts Which are the districts that have experienced such a great change in rank position between successive elections that they have contributed the largest source of error to the correlation coefficients? These are the districts that hold the secrets of change in rank orders. In Tables XX through XXVI, the ten largest contributors of rank differences between pairs of successive elections are listed. For convenience Of communi- cation we shall call these districts Error Districts. Included in the tables is each district's Stalemate Index and an indication whether a party turnover occurred in the second of the paired elections. 77 78 Error Districts are not the sole possession of any section of the United States. Nor does any state or district monopolize occurrences in the category of largest contributors of error in the computation of rank correlation coefficients. In all of the election pairs covered, only fifteen districts out of a total 55 Error Districts appeared more than once. Of these fifteen, fourteen appeared twice and one appeared three times (Maine-1). A common occurrence in the case of Error Districts is the phenomenon of "counteraction. " Of those districts that appeared twice, twelve that experienced a large rank change moved back in the Opposite direction at the second election pair in which they experienced a large rank change. In three of the twelve cases, this counteraction was the result of one major party not having a candidate at one of the elections. This meant that the voters of that party, no matter how few, could not register their votes at one of the elections, but could at the other election of the pair being compared. Five of the twelve counteracting cases were the result of actual partisan turnovers within the districts--or unexpected occurrences. The remaining four counteracting cases were Composite Districts. The rank changes in these cases, also, were the result of actual partisan realignments within the district's electorate. Three districts that twice classified as Error Districts moved in the same direction both times (Neb. -1, Tex. -18, Mass. -5). Factors Related to Error District Occurrence Examination of the Error Districts indicates that there is no single, readily discernible characteristic, common to all of the group which separates them from the other districts. They don't all move with the prevailing partisan trend. Nor do they all move against the prevailing trend. 79 Table XX. Ten Largest Contributors of Rank Differences Between the Elections of 1942 and 1944 State and Number of Rank 1942 1944 1944 Error Places Shifted Stalemate Stalemate Party District 1942 to 1944 Index Index Turnover? Maine-l -109.0 - 7.0 -18.8 No Minn.-3 120.5 -15.5 1.0 Yes Minn. -4 172.5 —27.7 1.8 Yes Minn.—8 128.0 -20.5 - 1.9 No Neb.-l 222.0 18.3 -19.9 Yes N.Y.-5C - 95.5 3.0 - 7.2 Yes N.Y.-12C 107.0 -22.3 - 5.3 No N.C.-9 - 94.0 50.0 8.8 No N.D.-l 123.5 -10.4 11.2 Yes Ohio-l3 -ll6.0 - 7.4 -23.9 No Table XXI. Ten Largest Contributors of Rank Differences Between the Elections of 1944 and 1946 State and Number of Rank 1944 1946 1946 Error Places Shifted Stalemate Stalemate Party District 1944 to 1946 Index Index Turnover? Maine-l 100.0 -18.8 - 9.6 No Nev.-l -101.0 13.1 - 8.8 Yes N.D.-l -186.0 11.2 -21.5 Yes Ohio-13 92.0 —23.9 -ll.9 No Pa.-7 - 96.0 - 1.5 -l6.5 No Tenn.-2 - 93.0 - 7.1 -34.0 No Wash.-l -106.0 3.4 -l3.8 Yes Wisc.-l 135.0 -37.4 - 6.9 No Wisc.-6 89.0 -17.1 - 9.8 No Wisc.-10 108.5 -13.2 - 5.3 NO 80 Table XXII. Ten Largest Contributors of Rank Differences Between the Elections of 1946 and 1948 _ —' State and Number of Rank 1946 1948 1948 Error Places Shifted Stalemate Stalemate Party District 1946 to 1048 Index Index Turnover? 111.-1C -101.0 - 1.9 - 6.1 NO Maine-2 - 94.5 -10.7 -17.2 NO Mich.-12 - 89.0 - 4.6 - 6.9 No N.J.-8 107.5 -21.2 - 0.1 No N.Y.-5C 83.0 —16.0 - 0.1 No Ore.-3 - 82.5 - 6.7 - 9.2 No Va..-1O -116.0 26.6 0.1 NO Wash.-1 81.5 -13.8 2.0 Yes Wash.-6 - 95.5 - 3.9 - 7.1 No Wisc.-10 - 81.5 - 5.3 - 6.6 No Table XXIII. Ten Largest Contributors of Rank Differences Between the Elections of 1948 and 1950 State and Number of Rank 1948 1950 1950 Error Places Shifted Stalemate Stalemate Party District 1948 to 1950 Index Index Turnover? Maine-1 97.0 -12.5 - 4.0 No Maine-2 87.5 -l7.2 - 7.7 No Mass.-5 242.0 -49.7 -26.1 No Mass.-10 85.5 -19.5 - 8.2 No Neb.-2 -105.5 1.4 -l3.5 Yes N.J.-2 91.5 -12.1 - 4.3 No N.J.-8 - 94.5 - 0.1 -13.8 No N.Y.-1C -133.5 -l5.8 - 0.9 No N.Y.-10C - 84.5 - 1.4 -14.1 NO Tex.-18 - 86.5 38.7 2.5 No 81 Table XXIV. Ten Largest Contributors of Rank Differences Between the Elections of 1950 and 1952 State and Number Of Rank 1950 1952 1952 Error Places Shifted Stalemate Stalemate Party District 1950 to 1952 Index Index Turnover? Fla.-l 122.0 50.0 0.7 No Fla.-7 89.5 50.0 6.3 No Kan.-l -163.5 -l6.5 1.5 Yes Mass.-5 237.0 -26.1 -26.1 NO Mich.-1l - 94.5 -l6.7 - 9.3 No Neb.-l 114.5 - 4.5 -22.0 No Neb.-2 - 89.0 -l3.5 - 6.1 ’No Tenn.-2 121.5 - 2.2 -18.9 No Tex.-18 -121.5 2.5 50.0 No Va.-6 103.0 49.7 - 1.6 Yes Table XXV. Ten Largest Contributors of Rank Differences Between the Elections of 1952 and 1954 State and Number of Rank 1952 1954 1954 Error Places Shifted Stalemate Stalemate Party District 1952 to 1954 Index Index Turnover? N.J.-6 122.0 -l4.3 4.6 Yes N.Y.-1C - 83.5 - 9.8 -l4.6 No N.Y.-12C - 82.0 -10.8 -16.5 NO Ohio-15 -113.5 14.3 - 4.0 Yes Okla.-4 83.0 9.0 50.0 NO Tex.-5 -168.5 50.0 - 2.9 Yes Tex.-8 - 82.0 50.0 12.4 No Va.-6 -127.0 - 1.6 -12.5 NO Wash.-3 -130.5 - 3.4 -14.9 No Wisc.-9 133.5 -15.2 5.4 Yes 82 Table XXVI. Ten Largest Contributors of Rank Differences Between the Elections of 1954 and 1956 State and Number of Rank 1954 1956 1956 Error Places Shifted Stalemate Stalemate Party District 1954 to 1956 Index Index Turnover? Fla.-5 -l30.0 50.0 1.4 No Fla.-6 -102.5 50.0 4.7 No Iowa-6 99.0 -10.3 0.1 Yes Kan.-5 147.0 -l4.9 0.5 Yes M0.—9 99.5 9.0 50.0 No Neb.-3 142.5 -15.2 - 0.1 NO Ore.-l 98.5 -l3.0 - 4.7 No S.D.-1 101.0 - 8.0 2.4 Yes Va.-l -104.0 49.9 0.8 No Wash.-4 103.0 -ll.0 - 0.4 NO Focusing on the party turnover experience of the Error Districts, we see that in 19 out of the 70 cases a party turnover occurred. Consequently, there were 51 cases, or approximately 73% of all cases, in which the incumbent party was successful while the district was experiencing an unusually large change in rank position. Such a majority of cases in the non-turnover category Shows that the unusual rank changes of the Error Districts were not accompanied by district swings to the "out" party--at least not to the extent that the "out" party wins. Large average Stalemate Index change figures for non-turnover cases support the indication that the districts involved were, for the most part, districts in which the majority party enjoyed a large margin prior to occurrence of the large rank change. For the period covered, the average Stalemate Index change between pairs of elections for non-turnover cases of Error Districts ranges from a low of 9. 28 (for the 1946-48 pair) to a high of 31.13 (for the 1954-56 pair). 83 The second lowest average change for non-turnover cases is 26.07 (1950-52) and the remaining average change figures are 15. 53 (1944-46), 21.05 (1942-44), 18.58 (1952-54) and 15.36 (1948-50). The fact that must be considered along with these averages is that despite relatively large changes in both their Stalemate Indexes and their rank positions, these Error Districts returned the incumbent party to office. They were not retaining a stable partisanship while the other, non-Error Districts were experiencing a uniform partisan change. By and large, the un- usually high changes in rank position between successive elections were the result of the districts' own partisan changes. Further, there is no clear relationship between Error Districts and any over-all partisan trend, as shown by changes in party composition of the House of Representatives. For example, the 1942-1944 pair of elections, while Democrats were gaining 24 House seats, five Error Districts were experiencing a Democratic trend and five a Republican trend (according to the study composition of the House, Democrats gained 15 seats at the 1944 election). Each of the other six election pairs, also, indicates no over-all relationship between partisan changes in Error Districts and partisan changes in the composition of the House (see Table XXVII). Examination of the Stalemate Indexes of Error Districts Shows that most would not be considered “close" districts. Applying for the moment the standard definition that a district election is "close" if the absolute value of its Stalemate Index is less than 5.0, we find most Error Districts not experiencing ”close" elections immediately prior to the election during which they made radical rank changes (see Table XXVIII). In fact, 67% of Error District election cases experienced a Stalemate Index of 10.0 or greater at the first election Of the pair in which they made big rank changes and 80% experienced a Stalemate Index of greater than 5. 0. To summarize, there is no evidence to indicate that the partisan ”closeness" of a district by itself bears any relationship to its potentiality for experi- encing a large rank change. 84 Table XXVII. Comparison of House Composition and Partisan Trend of Error Districts for Pairs of Successive Elections 1942- 1956 Democratic Gain or , Number of Error Loss--House Seats Districts Election Actual Study Moving Moving Pair Composition Composition Democratic Republican 1942-44 +24 +15 5 5 1944-46 -54 -33 5 5 1946-48 +75 +53 3 7 1948-50 —29 - l6 6 4 1950-52 - 23 - l9 4 5 1952-54 +21 +19 3 7 1954-56 + l - 3 7 3 Table XXVIII. Distribution of Error District Election Cases According to Size of Stalemate Index at the First of Two Successive Elections Party Turnover Cases- Non-Turnover Cases- Stalemate Index at Stalemate Index at Election First Election First Election Pair 0-4.9 5.0-9.9 10.0+ 0-4.9 5.0-9.9 10.0+ 1942-44 1 0 4 0 2 3 1944-46 1 0 2 1 1 5 1946-48 0 0 1 3 2 4 1948-50 1 0 0 2 0 7 1950-52 0 0 2 3 0 5 1952-54 0 0 4 2 2 2 1954-56 0 1 2 0 1 6 Totals 3 l 15 11 8 32 85 On the contrary, the greatest share of Error Districts would have been classified as "safe" districts prior to their radical rank changes. A closer investigation of the circumstances in which each Error District occurred shows that most occurrences can be explained by special factors. Knowledge of these special factors was available prior to the elections and the resulting change in rank order could be expected. Forty-three of the total 70 occurrences of Error Districts can be explained by these special factors (see Table XXIX). The ten Error Districts of the 1942-1944 election pair may be accounted for as follows. Three cases are Composite Districts. Three cases were the result of third party influences. In two cases one of the major parties offered no candidate in one of the two elections. In two cases the district bucked the national trend. In 1942 there was a Democratic candidate in Ohio's 13th district, but there was no Democratic candidate in 1944. North Carolina's 9th district had no Republican candidate in 1942, but a Republican ran and lost in 1944. These two districts appeared as Error Districts because there was no two-party competition. North Dakota 1C, New York 5C, and New York 12C are Composite Districts. The third party influence caused Minnesota's 3rd, 4th, and 8th districts' to be Error Districts for the 1942-1944 pair of elections. In the case of each district, there was a large third party vote in 1942 with the Republican candidates winning in all three districts. In 1944, however, the Democratic-Farmer-Labor Party entered the picture. The 3rd party vote subsequently disappeared, absorbed by the new Democratic-Farmer-Labor coalition. AS a result, the D-F-L candidates won in the 3rd and 4th districts. The Republican candidate beat the D-F-L candidate in the 8th district, but the Republican margin was cut from 36, 000 in 1942 to 4, 000 in 1944. r1“ 86 Only Maine's lst district and Nebraska's lst district of the ten Error Districts for 1942-1944 are not readily explained. Both of these districts are persistently Republican and the GOP margin of victory was increased in both between 1942 and 1944 despite a Democratic trend nationally. It appears that the two reacted to a Democratic trend by increasing the Republican vote. The ten Error Districts of the 1944-1946 election pair include one Composite District, North Dakota 1C. In three cases one major party offered no candidate at one of the two elections. There was no Democratic candidate in 1944 in Wisconsin's lst district and Ohio‘s 13th district. In both cases the Democrats Offered a candidate in 1946, but lost. Tennessee's 2nd district had a losing Democratic candidate in 1944, but only a Republican candidate in 1946. The remaining Six Error District occurrences of the 1944-1946 election pair are not as easily explained. In three cases, those of Washington 1, Nevada 1, and Pennsylvania 7, the large rank change could be accounted for by the national Republican trend. All three districts showed Republican gains, with two actually experiencing a party turnover. Wisconsin's 10th, Maine's lst and Wisconsin's 6th, however, experienced an increase in Democratic strength--moving against the national trend. It could be that Maine‘ 5 lst district, which had bucked the national trend in 1944 to register a Republican gain, was just returning to normal in 1946. Between the 1946 and 1948 elections, two Error Districts were Composite Districts--Illinois 1C and New York 5C. Virginia's 10th district appeared as an Error District through the process of recon- structing, because it was not actually brought into existence by redistrict- ing until 1952. Two districts, New Jersey 8 and Washington I, experienced Democratic gains in 1948 over 1946. In both cases the Republican 87 incumbent ran in 1948. The incumbent in Washington 1 was beaten in 1948, but the incumbent won in New Jersey 9 although his margin was cut from 34, 000 to 148. The voters of Wisconsin 10 and Washington 6 moved against the national Democratic trend to increase the winning margin of the Republican incumbent in each case. Michigan's 12th district remained fairly stable in the face of the national trend, however, to re-elect the Republican in— cumbent with approximately the same margin in 1948 as in 1946. The Progressive Party exerted some influence on the major party balance in 1948 for Oregon's 3rd district. The Progressive candidate attracted 13, 171 votes. While the Republican incumbent was increasing his margin of victory between 1946 and 1948, most of this increase could be attributed to normally Democratic votes swinging temporarily to the Progressive candidate. In the remaining 1946-1948 Error District, Maine 2 seems to have expressed its gratitude for the departure of the incumbent. Republican incumbent Margaret C. Smith did not run for re-election in 1948. Her successor, Republican Charles P. Nelson, won in 1948 and accumulated a bigger margin than Mrs. Smith did in l946--deSpite a national Democratic trend. Two Composite Districts fell into the Error classification in the 1948-1950 election comparison, New York 1C and New York 10C. Also, there was a Single case, that of Massachusetts 5, where the Democratic Party offered a candidate in 1950, but not in 1948. The Republican candidate won despite opposition in 1950. Four Error Districts moved against the Republican trend from 1948 to 1950. In Maine 1, the GOP incumbent was re-elected, but with a reduced plurality. In New Jersey 2 and Maine 2, the same thing happened. Partisan changes in the district hold the secrets of rank changes for these districts. 88 Republican Christian Herter of Mas sachusetts' 10th district was re-elected in 1950, but the Democratic candidate in 1950, Francis X. Hurley, actually out Herter's 1948 margin despite a national Republican trend. This is another case where local factors seem to hold the key to unexplained changes in rank order. New Jersey 8 followed the national trend from 1948 to 1950 as it did from 1946 to 1948. The Republican incumbent was re-elected in 1950 with a wider margin of victory than he had had in 1948. Texas' 18th district elected a Democrat in 1948, but the incumbent. did not run in 1950 and the Democratic plurality was down. A combination of the incumbent not running and a national Republican trend could account for Texas 18 appearing as an Error District in the 1948-1950 election pair. Between 1950 and 1952 two Error Districts are explained by the absence of competition. In Texas 18 the Republican party ran a candidate in 1950 and lost, bur offered no candidate in 1952. Virginia's 6th district had no Republican candidate in 1950, but in 1952 a Republican candidate ran and defeated the Democratic incumbent. The election results in Virginia 6 point to the influence of the national Republican trend in 1952. Florida 1 and Florida 7 were changed by redistricting between the 1950 and 1952 elections. The reasons for their appearances as Error Districts, then, are hidden by the reshuffling of geographic areas that occurs with redistricting. Massachusetts' 5th district exhibited little internal partisan changes between 1950 and 1952. A long-time Republican incumbent won in 1952 and there was no unusual increase in Republican strength that could be traced to the influence of a national trend. In short, it appears that Massachusetts 5 was an Error District because it retained a fairly stable party division of the vote while other districts in the nation were swinging with a Republican trend. Nebraska 1 followed the 1950 to 1952 national swing by increasing the victory margin of the Republican incumbent Curtis. 89 The voters of Nebraska 2 moved against the national swing from 1950 to 1952. A partial explanation is afforded by the fact that the Republican incumbent did not run in 1952. The new GOP candidate carried the district, but the 1950 Republican margin of 30, 000 was cut to 17, 000 in 1952. The Republican candidate in Tennessee's 2nd district in 1950 ran for the first time and was elected. As the incumbent in 1952 he was re- elected with a larger margin than in 1950. The Democratic vote remained stable between 1950 and 1952. - It appears that both the national swing and incumbency contributed to an above-average rank change here. I In Kansas 1, the voters bucked the national trend from 1950 to 1952 to defeat the Republican incumbent in 1952. The large rank change of Kansas 1 seems to have been caused by local factors peculiar to that district in 1952. A local factor also was responsible for the wide ,rank fluctuation of Michigan 11 between 1950 and~1952. In 1950, Charles Potter, Republi- can, was elected. Potter ran for the U. S. Senate in 1952, however, and a new Republican candidate, Victor Knox ran in 1952. Although Knox was pOpular in that district, having been elected State Representative many times and having served as Speaker of the Michigan House of Repre- sentatives, his Democratic Opponent was a strong ”name candidate. " Prentiss Brown, Jr. , the son of a former Michigan U.. S. Senator with the same name, cut into the 1950 Republican margin despite the national swing to GOP candidates. Two Composite Districts, New York 1C and New York 12C, appeared as Error Districts between 1952 and 1954. In Texas 5 and Texas 8 there was no Republican candidate in 1952. The Democratic incumbent in Texas 5 did not run in 1954 and the Republicans were able Itoiwin the district. The Republican candidate in Texas Sin 1954 cut the Democratic margin from 1952 drastically but the Democratic incumbent won. Both of 90 these Error Districts were moving against a national Democratic trend in 1954. The large rank change for Oklahoma 4 between 1952 and 1954 is accountable to the failure of the Republican Party to offer a candidate in 1954. There was a GOP candidate in 1952. In Wisconsin's 9th district and New Jersey's 6th district the Republican incumbents did not run for re-election in 1954. In both cases, the voters of these districts elected Democrats, indicating vulnerability to the national trend. Also, in both cases, the "retiring" Republican candi- dates were popular individuals. In Wisconsin 9, Merlin Hull had been elected many times prior to 1952 and Clifford Case, the New Jersey 6 incumbent, ran successfully for the U. S. Senate from New Jersey in 1954. Bucking the 1954 national trend were districts 3 of Washington and 15 of Ohio. In Ohio 15, the Democratic incumbent was not a candidate in 1954 and the Republican candidate won the election. In Washington 3, the Republican incumbent of many terms actually added to his 1952 margin of victory in 1954. The voters of the remaining 1952-1954 Error District (Virginia 6) elected the Republican incumbent in 1954, but the increased Republican margin (against a national Democratic trend) could be attributed to a change in Democratic candidates. In 1950, Democrat Clarence Burton ran unopposed. Burton was beaten by Republican Poff in 1952, but a new Democratic candidate ran against Poff in 1954. Between 1954 and 1956 three districts appeared in the Error classi- fication because there was no Republican candidate in 1954 but a Republican candidate was offered in 1956. In all three cases the Democratic candidates won in 1956. These districts were Florida 5, Virginia 1, and-Florida 6. In Missouri 9, the GOP offered no opposition candidate to Democrat Clarence Cannon in 1956. There was a Republican candidate in 1954, so the lack of Opposition in 1956 could explain the large rank fluctuation of Missouri 9 between 1954 and 1956. 91 Six of the 1954-1956 Error Districts swung with what little national trend there was between the two elections by moving Democratic. Oregon 1 re-elected a long-time Republican incumbent, but his 1954 margin of 40, 000 was cut to 19, 000 in 1956. The long-time GOP incumbent in Kansas 5 was not a candidate for re-election in 1956 and the Democratic candidate won by a narrow margin. In Nebraska 3, the Republican incumbent won in 1956, but the GOP margin in 1954 was cut drastically in 1956. The same situation occurred in Washington 4 where the Republican incumbent barely nosed out his Democratic opponent. An upset occurred in Iowa 6 where the Republican incumbent of many years lost by 198 votes out of 129, 000 votes cast. And in South Dakota 1, a 4-term Republican incumbent was upset by a Democrat. The 1954-1956 Error Districts, then, Show a strong swing to Democrats (7 out of 10 cases) despite the fact that the Democratic Party actually experienced a net gain of only one seat from the Republicans. The high rank correlations of the total array of study districts discovered in the preceding chapter suggested that forecasting of con- gressional district rank orders could be possible with substantial accuracy. Location and understanding of the deviant cases (Error Districts) actually strengthens this prospect. In the relatively cursory survey of the last few pages more than half (43) of the 70 cases of Error Districts were explained by Special factors which were either forecastable prior to the election or would have meant that the district did not belong in the rankings in the first place (see Table XXIX). In the category of special factors which were known prior to the election and which appear determinative of the election results, we find the following distribution: (1) There were sixteen cases in which one of the two major parties offered no candidate at one of the pair of elections for which the district qualified as an Error District; (2) in six cases the 92 NH o m 0 0H mHmuoB o o o H H4 omuwmoH N o o m H» HsthmmH o o N H N NmtomaH N o o o H omuwme N H H H o wwtcme H N o o m £14va m m o o N wwanoH HOHSmHQ 030.9ch .034 OOHSmHQ :OHHO0H0 mGOHuO0HM 03H. 0.900% 0OHwOQEOU ~3pdfl U0H00HU wcHuoHSme 10m .HOH and “Oz H0 0:0 H0 00.3qu GOHHO0HM HOMER. t0mtt0oc0hhdooO 0000 E09552; #032 03H. H0 0G0 Ho 0.30m .HOHHmH H0 0EHB H0 .HOH 0363500 OZ H00m “OZ OOHSmHQ whouomh H0H00thum0od0hp5000 H0 H0QESZ 0200:5000 uOHSmHQ uOupmH wcHQHoHonMH whooomh H0H00nHm H0 COHHSQHSmHQ .XHXN 03.09. 93 incumbent did not run for re-election; and (3) in six cases a strong third party candidate upset the major party balance in the district. In the category of special factors that would have kept the district out of the ranking in the first place, we find the following: (1) There were twelve cases in which Error Districts were Composite Districts--dis- tricts that could not be forecasted for future elections because they are not "actual" districts now; and (2) in three cases the districts were not "actual" districts at the time of Error classification, being created at a later date by redistricting. If all of these Error Districts explained by special factors had been removed from the rankings prior to computation of the rank correlation coefficients, as they should have been, the coefficients would have been even higher. The rank orders could have been forecasted more accurately without these districts than with them. There were 26 Error District occurrences not explained by the special factors outlined in Table XXIX. Almost half of these Error cases, however, were associated withia change in candidates for the incumbent party. In 5 of the 26 cases the "out" party beat the "in" party when the incumbent did not run for re-election (Wisconsin 9, 1952-54; New Jersey 6, 1952-54; Ohio 5, 1952-54; Kansas 5, 1954-56; and Iowa 6, 1954-56). In 5 cases the incumbent party held the district deSpite the departure of its incumbent candidate (Maine 2, 1946-48; Texas 18, 1948-50; Nebraska 2, 1950-52; Tennessee 2, 1950-52; and Michigan 11, 1950-52). Although not quite as strong an indicator as the Special factors in Table XXIX, the absence of an incumbent candidate in a congressional election appears to be a good indication of impending rank change greater than normal. Removing these ten no-incumbent-candidate cases from the group of apparently unforecastable rank changes means that out of the 70 Error District cases only 16 were actually cases in which the possibility of 94 significant rank fluctuation could not have been expected--on the basis of present knowledge and hypotheseS--further support for the proposition that congressional district rank orders can be accurately forecasted. Summary In this chapter we focused on the deviant cases, Error Districts, attempting to find out some of the more significant reasons why the rank correlations of the previous chapter were less than a perfect 1.000. By examining the election circumstances of each Error occurrence we dis- covered that in 54 of the total 70 cases of Error occurrence the large rank changes actually could have been expected. These occurrences were explained by obvious circumstances, such as no candidate Opposi- tion, departure of a long-time incumbent, and third party influence. Had we been attempting to forecast the rank order at the next election, these obvious and special circumstances would have been justification for removing the districts with which they were associated from the rank order. The removal of these districts would have in- creased the rank correlation between election pairs and consequently strengthened the accuracy of the rank order forecast. This discovery will be put to use in the next chapter where pro- cedures for making rank order forecasts will be tested. A history of perfect rank correlations for all districts between successive elections would indicate that the rank order of the election to be forecasted would be the same as that of the previous election. For each individual district, a perfect rank correlation would mean that a trend analysis of each district in the rank order would tell us whether the whole rank order was moving in a Democratic or Republican direction. A trend analysis of those districts having a history of follow- ing the national swings closely should provide strong evidence for a valid forecast of the next election's results. 95 The trend analysis of partisan changes in individual districts will be combined with the rank order stability found in Chapter IV to produce a forecast in Chapter VI. CHAPTER VI AN EXERCISE IN CONGRESSIONAL ELECTION FORECASTING In the preceding chapters several discoveries pertaining to con- gressional election results have been presented. These findings will be employed in this final chapter to experiment with a new method of forecasting election results. Employing all of these discoveries, we will essay a forecast of the 1956 congressional election results. The procedure will follow these steps: 1. Set up a new operational concept of "marginality" to classify districts according to their vulnerability to a party turnover at the next election. 2. Adopt the 1954 rank order as an estimate of the 1956 rank order. 3. Accomplish a projection to 1956 of the Stalemate Indexes for each Step 4 Competitive district by fitting a linear regression to the 1948, 1950, 1952, and 1954 Stalemate Indexes for each Step 4 Com- petitive District. 4. Fit the forecasted Competitive District Stalemate Indexes into the estimated 1956 rank order to locate the rank at which all districts above it are Democratic and all below it are Republican--the "cut-off" point. 5. Correct the expected partisan exaggeration of the forecasted Step 4 Competitive District election outcomes. 96 97 Three- Factor Marginality The new concept of marginality to be employed in this study will classify districts on three criteria, all three of which are related to a forecast of the Stalemate Index of the district. being classified. The first criterion is the direction in which the partisan trend of a district is moving. The second is the Size of the Stalemate Index forecasted by extending the trend line. The third criterion is whether or not the Slope of the district's trend line is steep enough to result in a forecast of party turnover in the district. For purposes of rating the probability of party turnover in a dis- trict, those districts whose slopes indicate a Republican gain will be classified "Republican" and those with a slope moving toward a greater Stalemate Index will be classified as ”Democratic. ” Where the fore- casted Stalemate Index indicates that a district could elect the out-party candidate, the district will be classified as ”critical, " regardless of the size of the forecasted Stalemate Index. Where the forecasted Stale- mate Index indicates that a district could elect the in-party candidate, the district will be classified as ”marginal" or "close" depending on the size of the forecasted Stalemate Index. If the forecasted Stalemate Index is between 0 and4.9, the rating will be "marginal. " If it is between 5.0 and 9. 9, the rating will be "close. ” If the forecasted Stalemate Index indicates no turnover and a resulting Stalemate Index 10. 0 or greater, the district is classified as "non-competitive. " The following chart lists all possible combinations of the above factors and the district classification which would result from each c ombination: 98 1' Size (Absolute Value) Of Forecasted Stalemate Index 0-4.9 5.0-9.9 10.0 plus Democratic Turnover Critical Critical Critical Trend Forecasted Democratic Democratic Democratic (Positive No Turnover Marginal Close Non- Slope) Forecasted Democratic Democratic Competitive Republican Turnover Critical Critical Critical Trend Forecasted Republican Republican Republican (Negative No Turnover Marginal Close Non- Slope) Forecasted Republican Republican Competitive The new concept of marginality can best be illustrated by the use of an example. By linear regression the Stalemate Index of Wisconsin-5 is estimated to be -0.6 in 1956 (see Table XXX). Since the actual Stalemate Index in 1954 was 2. 3, a party turnover is forecasted. The computed trend Slope for Wisconsin-5 is negative, or moving Republican (b = -. 57), indicating a Republican trend. Wisconsin-5 would be "Critical Republican. " For 1956, then, the classification of Our new concept of marginality, like the Old, classifies a district according to a judgment of the probability that the district will elect the out-party candidate in the next election. The prevailing concept of marginality classifies a district on the basis of the result of the last election. upon a statistical forecast of the next election result. The concept used here, however, demands a judgment based More important, the forecast upon which our marginality classification depends is based upon more than one previous election. Because the forecast is based upon a consideration of a district's electoral history, our marginality rating also is determined by a district's trend line, the slope or rate of change of that trend line, and the direction in which the trend is moving. Briefly, then, there are four basic classifications--Critical, Marginal, Close, and Non-competitive. We would estimate that the probability of a party-turnover in Critical Districts is greater than in 99 Marginal Districts. Similarly, it is estimated that a party turnover is more likely to occur in Marginal Districts than in Close Districts. We expect no turnover in the Non-competitive Districts. Forecasting the 1956 Rank Order The Step 4, or "competitive, " districts are a microcosm of all congressional districts. We will work here with the group of 22 Com- petitive Districts, forecasting their 1956 partisan character. Calling upon the rank order stability demonstrated in Chapter III, we can justifiably estimate that the rank correlation of all districts between 1954 and 1956 will produce a correlation coefficient greater than . 9000. We shall accept this correlation as evidence that the 1954 rank order is the best possible estimate of the 1956 rank order. Acceptance of the 1954 rank order as an estimate of the 1956 rank order allows the assignment of a 1956 rank position to each of the 22 Competitive Districts. These rank positions are listed for each Com- petitive District in Table XXX. Except for Virginia-6 which is ranked 296, most of the array of Competitive Districts are bunched in the middle of the 1-to-336 ranking. The tOp-ranked district is New Jersey-10 which has a rank position of 109. Other than Virginia-6, North Carolina-10 has the lowest rank, that of 247%. The first part of our forecast for 1956 is completed by the assign- ment of forecasted ranks of Step 4 Districts for 1956. Estimating Stalemate Indexes by Linear Regression A rank order forecast is not a complete estimate of the over-all results of an impending national congressional election. In addition, it is necessary to know which party will control Congress as a result of the election. This estimate is aided by a forecast of election results in at least some of the individual districts. 100 Table XXX. Competitive Districts' Projected Stalemate Indexes for 1956 Fitted to 1954 Rank Order"< Estimated 1956 State and Rank 1956 Forecasted District (Based on 1954) Stalemate Index New Jersey-10 109 15.95 Minnesota-6 114 12.10 Massachusetts-2 121 8.10 New Jersey-4 129 5. 00 New Jersey-11 132 8.65 Ohio-18 1332'- 8.15 Connecticut-l 138 6. 35 Colorado-4 1545- 2.60 West Virginia-1 158 0 50 Indiana-8 163 - 1. 95 Wisconsin-5 . 163 -0. 60 Pennsylvania— 11 172 —0. 10 Montana-2 178%- -3. 60 Washington-2 188-g)- -10. 75 Washington-1 192%- -4. 40 Connecticut-3 195 -4. 30 Utah-l 204 -13. 15 Nevada-l 214 -5. 05 Indiana- 11 220 - 10. 40 Utah-2 . 245 - -12. 10 North Carolina- 10 24717 -16. 10 Virginia-6 ' 296 -20. 80 I :I