Electoral and Ideological Change in the South by maqboolshahin

VIEWS: 2 PAGES: 19

									Electoral and Ideological Change in the South: The Case of the U.S. House of Representatives, 1952-2000

Dennis M. Simon Department of Political Science Southern Methodist University dsimon@mail.smu.edu

Presented at the Annual Meeting of the Southern Political Science Association, New Orleans, Louisiana. 8-11 January 2004.

Electoral and Ideological Change in the South: The Case of the U.S. House of Representatives, 1952-2000 Introduction: “But Not in the South”
During the 1940s and 1950s, the discipline of political science experienced what numerous scholars called a behavioral revolution. A variety of quantitative methods were imported from other disciplines and used to analyze survey and aggregate data on political behavior in the United States.1 One feature of the literature produced during this period is that the south was, more often than not, treated as an exception. The region was so solidly and consistently Democratic that the dependent variables that were the subject of attention in national studies – party identification, aggregate votes for federal offices, and the partisan composition of legislatures – exhibited little variation. Thus, the corollary or caveat “but not in the South” often tempered the conclusions drawn from empirical analyses. The situation has changed dramatically in the last fifty years. The south has evolved from a bastion of the Democratic Party of the New Deal era to become the “home base” for the Republican Party. The “community of political science” has devoted considerable attention to the changing south. There is work that documents the shifting partisan ties of southern voters (e.g., Nadeau and Stanley, 1993; Stanley, 1988). Other work focuses upon the importance of the south in presidential elections and its growing clout in the Electoral College (e.g., Black and Black, 1992). There are studies of what some call the “southern realignment” and how the battles over the Civil Rights Act of 1964 and the Voting Rights Act of 1965 triggered this realignment (Carmines and Stimson, 1989). Others, in the tradition of V.O. Key (1984;1944), address questions pertaining to the character of democracy and representation in the south (Aldrich, 2000). An especially dramatic aspect of the changing south involves the elections to the U.S. House of Representatives.2 These elections are the focus of this paper. The analysis consists of three parts. I highlight the major changes in House elections since 1952 as well as trends in the ideology of southern members in Part I. In addition, I illustrate how the population of congressional districts, including the black populations, changed as a result of redistricting. This overview of change forms the basis of four design principles that are used to specify, estimate, and evaluate a model of electoral outcomes in House elections. This is the task undertaken in Part II. In Part III, the same design principles
1

Lazarsfeld, Berelson, and Gaudet (1944) and Berelson, Lazarsfeld, and McPhee (1954). The classic behavioral study of voting in national elections is Campbell, Converse, Miller, and Stokes (1960). The volume is an in depth analysis of survey data collected during and after the 1952 and 1956 presidential election campaigns. The most thorough analysis of House elections in the South is Black and Black (2002).

2

2

and essential variables are used to formulate a model of member roll call voting on the floor of the House. Overall, the analysis shows that the history of redistricting in the south has produced a markedly skewed distribution of black and hispanic residents. As a result, the impact of these groups on electoral outcomes has been diluted. The changes also have influenced the roll call voting of members and contributed to the polarization of southern Democrats and Republicans.

Part I Partisan and Ideological Change in the South: The Case of the U.S. House of Representatives
The electoral fortunes of southern Republicans seeking election to the U.S. House of Representatives have increased substantially since 1952. Figure 1 presents two measures of these changing fortunes over time: the proportion of the two-party vote won by Republicans in southern congressional districts and the proportion of southern seats won by Republicans.3 In 1952, Republican candidates for the House won only 21.3% of the two-party vote and only 7.6% of the seats. This outcome is particularly illustrative of the electoral politics of the era in that even a popular Republican nominee for president had no coattails in the region. Dwight Eisenhower, the Republican standard bearer in 1952, received 48.8% of the two-party vote in the south and won a majority of the vote in 39% of southern House districts. A similar outcome prevailed in 1956. President Eisenhower won a majority of the popular vote in the south, 56%, and majorities in 44.2% of the congressional districts. Despite this performance, Republican candidates for the House won only 27.3% of the two-party vote and captured only 8.4% of the seats. --- Please Refer to Figure 1 --The figure illustrates that there are two prominent periods of Republican gains in House elections. Between 1960 and 1972, the proportion of the two-party vote won by Republican candidates increased from 26.0% to 40.2%. The seats won by Republicans during this period increased from 7% of the delegation to 31.3%. The second period of growth occurred in the 1990s. In the midterm election of 1990, Republicans won 43% of the popular vote and 30.6% of the seats. In 1994, Republicans – for the first time since the Reconstruction Era – won majorities of both the two-party vote and seats. These majorities then increased slightly during the remainder of the decade. The southern Republicans thus traveled a path from a weak minority in the 1950s to a solid majority in the last four national elections.

--- Please Refer to Figure 2 ---

3

In this analysis, the south is defined as the eleven states of Confederacy along with Kentucky and Oklahoma. For the source of these data, see Table 2.

3

Figure 2 is designed to illustrate that when electoral and partisan change occurs, ideological change follows. The figure is based on the roll call voting scores obtained from the analysis by Poole and Rosenthal (1997). These scores are calculated by estimating a spatial model of floor votes in the U.S. House. The data used to estimate this model include all nonconsensual votes on the floor of the House in a given Congress. The scores range from –1.00 (most liberal) to 1.00 (most conservative).4 The scores are comparable across the Congresses in a given era and thus can be used to track ideological change among southern representatives.5 Figure 1 tracks the median scores of Republicans, white Democrats, and beginning in 1972, black and hispanic Democrats on the first or economic dimension of the model. The figure reveals a subtle trend in these scores. Beginning with the 90th Congress elected in 1966, the median position of white Democrats begins to move, albeit slowly, in a more liberal direction while Republican scores move in a more conservative direction. Especially noteworthy is the comparison of median scores for the 93rd Congress (elected in 1972) and the 107th Congress (elected in 2000). In the 93rd Congress, the difference between white and minority Democrats is greater than the difference between white Democrats and Republicans.6 These relative differences are reversed over time such that by the 107th Congress, white and minority Democrats have grown closer in ideology with the gap between white Democrats and Republicans widening. Over time then, southern white Democrats and Republicans have become more polarized with the ideology of white Democrats growing more similar to minority Democrats and national Democrats in general.7

4

The model developed by Poole and Rosenthal (1997) is a two-dimensional spatial model. In essence, the first dimension is economic and redistributive. As Poole and Rosenthal (1997: 46) explain, the first dimension “picks up” the conflict between “rich and poor.” Their analysis shows that this dimension, by far, is the most important for both predicting and understanding the roll call behavior of House members. Further, “except for brief periods,” the authors note, this dimension “divides the two major parties” (46). The second dimension largely captures divisions within the parties (e.g., slavery, civil rights). Its salience and explanatory power vary quite widely over time. In particular, the salience of this dimension has declined since the 1970s (see especially Chapter 11 entitled “The Unidimensional Congress,” pp. 227-232). Because of this variability in salience, my analysis relies on the first dimension. See http://voteview.uh.edu/page2a.htm. As Poole notes in this description, the “DW-NOMINATE scores in one Congress are directly comparable with scores in another Congress. However, cross-Congress comparisons should be conducted only between Congresses occurring during one of the stable 2-party periods of American History.”

5

In 1972, two blacks were elected to the House for the first time since Reconstruction: Andrew Young representing the 5th District of Georgia (a score of –0.63 in the 93rd Congress) and Barbara Jordan representing the 18th District of Texas (a score of –0.50 in the 93rd Congress). The other minority members elected were Eligio “Kiki” de la Garza from the 15th District of Texas (a score of –0.15) and Henry Gonzalez from the 20th District of Texas (a score of –0.55). Note that by the 107th Congress, the number of Black and hispanic representatives from the south had increased to 22 and constituted nearly 38% of all southern Democrats.
7

6

On the growing polarization of the national parties in Congress, see Bond and Fleisher (2000).

4

The South and the Changing Rules of the Electoral Game. My discussion of Figure 1 and Figure 2 highlight several key points in the evolution of the south. The movement of white Democrats in a more liberal direction begins after the 1966 election and the first blacks are elected to the House in 1972. Both of these developments are the initial manifestations of changes in the basic rules of the electoral game. To use the parlance of time series analysis, the electoral politics of House elections in the south were influenced by two, nearly simultaneous “interventions” in the 1960s: the Supreme Court decision in Wesberry v. Sanders in 19648 and the passage of the Voting Rights Act of 1965. The changes triggered by these interventions altered the traditional structure of House elections in the south. Wesberry v. Sanders. Consider, first of all, the impact of the Wesberry decision. In 1962, the U.S. Supreme Court ruled in the decision of Baker v. Carr9 that a challenge to the apportionment of seats in the Tennessee General Assembly was a justiciable issue. The standard established by this landmark case is often described as “one person, one vote” and held that disparities in population across legislative districts are unconstitutional. In practice, the decision reduced the dominance exercised by representatives of under-populated rural districts in many state legislatures. On February 17, 1964, the Supreme Court announced its decision in Wesberry v. Sanders, a case that challenged the congressional district boundaries in Georgia. Here, the Court applied the precedent from Baker and held “that, construed in its historical context, the command of Article I, Section 2, that Representatives be chosen ‘by the People of the several States’ means that as nearly as is practicable one man's vote in a congressional election is to be worth as much as another's.”10 --- Please Refer to Figure 3--The problem of malapportionment in the south was, as Figure 3 demonstrates, substantial. The figure presents a measure of relative deviation in the population of congressional districts in the south. The measure is constructed as follows. The ideal population of a district is calculated as the total population divided by the number of districts in a state. The ideal population is then subtracted from the actual population of a district to obtain a measure of absolute deviation. A positive result means that the district is larger than ideal while a negative result denotes a district is under-populated. This absolute deviation is then divided by the ideal population to express the disparity as a percentage (Butler and Cain, 1992: 46-47). Figure 3 presents the distribution of these disparities for 1962, the last pre-Baker reapportionment and redistricting. Of the 110 districts in the south,11 41 (37.3%) were substantially under-populated with disparities of

8

376 U.S. 1 (1964). 369 U.S. 186 (1962). http://caselaw.lp.findlaw.com/scripts/getcase.pl?court=us&vol=376&invol=1#f9

9

10

11

The total number of districts in the south following the reapportionment based on the 1960 census was 119. For the election of 1962, Alabama dissolved its district boundaries and chose its eight-member

5

–10% or more; 34 of the districts (30.9%) were over-populated with disparities of 10% or more; only 13 districts (11.8%) fell within the range of +/- 2%. --- Please Refer to Table 1 -The Wesberry decision triggered several rounds of redistricting in the south. The process is summarized in Table 1, which displays for each state, the number of districts redrawn prior to the congressional elections from 1962 to 1972. Consider first the 1962 round. Less than half of the southern districts were redrawn, a typical practice prior to the Baker and Wesberry rulings. Six states left their boundaries untouched. As the atlarge seat elected in Texas illustrates, it was not atypical for a state gaining a seat to elect the member at-large for one or two elections until the state legislature “got around” to redrawing district lines.12 Table 1 shows that the efforts to comply with Wesberry occurred prior to the midterm election of 1966 when 67.2% (80 of 119) of southern districts were redrawn and prior to the 1968 election when 47.1% (56 of 119) of all districts were redrawn. Between 1962 and 1970, all states redrew at least once. Six states redrew one or more of their district boundaries twice (Alabama, Louisiana, North Carolina, South Carolina, Tennessee, and Texas). Florida undertook the process three times (1962, 1966, and 1968). The reapportionment of seats after the 1970 census and the first post-Wesberry redistricting produced changes in 117 of the 121 districts in the south (96.7%). --- Please Refer to Figure 4 --The ultimate results of the Wesberry case and the redistricting it triggered are illustrated in Figure 4. The figure presents the distribution of relative disparities following the reapportionment and redistricting prior to the 1972 election. It shows that the relative disparities evident in 1962 disappeared with the distribution virtually collapsing around the point of ideal population. 93.4% (113/121) of the districts had disparities within 2% of the ideal.13 The Voting Rights Act of 1965. The history and consequences of the Voting Rights Act are well known. In the wake of the act, black registration and turnout increased in the south with a concomitant increase in the number of black officeholders (see, for example, Carmines and Stimson, 1989: 49; Stanley and Niemi, 2003: 60). Of
delegation in a statewide at-large election. In the same year, Texas was awarded a new seat and used an atlarge election to fill the new seat.
12

Perhaps the best illustration of this practice is the case of Oklahoma. On the basis of the 1930 census, Oklahoma was awarded a ninth congressional seat. Will Rogers was elected to fill this seat as an at-large representative in the elections of 1932, 1934, 1936, 1938, and 1940. In the reapportionment after the 1940 census, the state lost this seat. In the 1983 case of Karcher v. Daggett [426 U.S. 725 (1983)], the Court essentially ruled that any population disparity among districts is unacceptable (see Butler and Cain, 1992: 166).

13

6

particular importance here is that the act imposed a pre-clearance requirement on numerous jurisdictions in the south. This requirement prohibited any change in election laws without prior approval of the U.S. Department of Justice.14 This pre-clearance requirement includes the congressional districting laws of a state. These provisions essentially married the Voting Rights Act to the Wesberry decision. The equalization of district populations required by Wesberry was linked to the concerns of those enforcing the Voting Rights Act – the distribution and re-distribution of the black population. --- Please Refer to Figure 5 and Figure 6 --To illustrate this connection, I use the proportion of residents in a congressional district who identify their race as African-American.15 Given these data, I then generated the distribution of the black population across the congressional districts of the south after each census, reapportionment, and redistricting. Figure 5 displays the distribution after the 1962 round. The median black population is 20.3% and the measure of skewness is modest, 0.48.16 Approximately one-third of the districts have a black population greater than 25% but less than 50%. Figure 6 displays the distribution that was put in place for the 1972 election cycle. Compared to 1962, the black population is more dispersed among the districts; the median drops to 16.3%. There is little change in the skewness of the distribution. --- Please Refer to Figure 7 and Figure 8 --There is a more noticeable shift in the distribution after the 1980 round. The median drops slightly from 16.3% to 15.8%. However, the skewness of the distribution doubles from 0.44 to 0.88. The number of districts with a black population between 26% and 49% drops from 36 (29.8%) in 1972 to 32 (24.8%) in 1982. At the same time, the number of districts with a black population of 15% or less increases from 55 (45.4%) in 1972 to 62 (48.0%) after the 1982 round. In effect, the distribution is shifting to the right. In the literature on congressional elections, there exists a vigorous debate on the effects of the redistricting that followed the 1990 census (see, for example, Canon, 1999; Petrocik and Desposato, 1998; Hill, 1995). As Figure 8 illustrates, the change in the distribution is dramatic. The median of the distribution drops and the skewness increases substantially. Visually, the distribution looks like someone stepped on the middle of the an earlier distribution and forced the cases either to a handful of majority-minority districts on the right side or to a larger group of districts with a small proportion of blacks
14

Jurisdictions in nine southern states are subject to these pre-clearance requirements: Alabama, Florida, Georgia, Louisiana, Mississippi, North Carolina, South Carolina, Texas, and Virginia. In addition, jurisdictions within six states are covered by bilingual language requirements: Florida, Louisiana, Mississippi, North Carolina, Oklahoma, and Texas (Stanley and Niemi, 2003: 78). These data were obtained from the web site of Scott Adler (see Table 2).

15

16

Skewness is calculated as the third moment about the mean of a distribution. Positive values indicate that the distribution is imbalanced or skewed to the left while negative values indicate the distribution is “lopsided” to the right.

7

on the left. Compared to the 1972 round, for example, the median drops from 16.3% to 11.6%; skewness increases from 0.44 to 1.55. The proportion of districts with a black population greater than 25% but less than 50% drops from 32.8% to 5.8% (8/137); finally, the proportion of districts with a black population of 15% or less increases from 45.4% (55/121) in 1972 to 53.3% (73/137) in 1992. The changes depicted in this section are substantial and have altered the landscape on which candidates compete for seats in the U.S. House. The south has changed from a malapportioned set of districts in which blacks, largely disenfranchised, were distributed in a relatively unskewed manner to an equally apportioned set of districts in which the black population is distributed in a markedly skewed manner. Two questions thus arise. Have these changes influenced the outcomes of House elections and, if so, how?

Part II Explaining the Outcomes of House Elections
The objective in this section is to specify, estimate, and evaluate a model of electoral outcomes that allows me to assess (1) the virtual equalization of population after the 1972 round of districting and (2) the changing distribution produced by the decennial restricting plans. This model is based upon a set of four design principles. --- Please Refer to Table 2 --Principle 1: The outcomes of House elections are a function of the “demography” of the congressional districts. Studies that use the congressional district as the unit of analysis typically posit that the socio-economic profile and another demographic features of the district will influence the vote.17 My analysis follows in this tradition. I have gathered and compiled for each election in the south from 1952-2000 the following measures: percent of black residents in the district, percent of residents living in urban areas, the percent of unionized workers,18 and the median income of the district.19 The specific measures and the sources from which I obtained all data are presented in Table 2. Beginning in 1970, the census included the percent of residents identifying themselves as hispanic and I have included this measure for all election from 1972 to 2000.
17

For example, Black and Black (2003) use district proportions of blacks and urban dwellers to define what they call districts of the “old south” and “new south.” This measure is based on state level data. In other words, it measures the proportion of unionized workers statewide. This performance of this measure is superior to the district-level data on the proportion of blue-collar workers. Simply using the median income in each district could be problematic given the upward trend in the measure over time. To control for this drift, I calculated the overall median income for the region in each year and then divided the district median by this “grand median.” Thus the measure expresses median income in the district as a proportion of the “grand median.” Values greater than 100 represents districts over the median and values less than 100 are those below the median. The magnitude of the measure conveys the degree to which a district is “rich” or “poor” in a relative sense.

18

19

8

Principle 2: The outcomes of House elections will be influenced by the partisan leanings of the district. In the absence of systematic data on party registration or party identification at the district level, analysts typically rely upon the presidential vote in the district. Accordingly, the model includes the proportion of the two-party vote won by the Republican candidate for president. In addition, I include two binary variables designed to capture the effects of incumbency. One assumes a value of one if a Republican incumbent is running for reelection in the district; the other is set to one if there is no incumbent candidate in the election. Districts with a Democrat incumbent seeking reelection represent the suppressed category. Finally, because the distinction between deep and peripheral southern states has a rich tradition in the analysis of southern politics (Key, 1984; Black and Black, 1992, 2002; also see Ardoin and Garand, 2003), the model includes a binary variable that assumes a value of one for the “deep” southern states of Alabama, Georgia, Louisiana, Mississippi, and South Carolina. --- Please Refer to Figure 9 --Principle 3: The model should be specified in a manner that affords the opportunity to evaluate the impact of the near-equality in population first achieved in the redistricting following the 1970 census. In addition to this development, the election of 1972 was noteworthy not only for the election of the first black representatives since reconstruction but also for the success of President Richard Nixon’s “southern strategy.” This is demonstrated in Figure 9, which presents the proportion of southern districts in which the Republican candidate for president won the district with a majority of the twoparty vote. The figure shows that Nixon carried 118 of the 121 districts in 1972 (93.7%). Note, too, that prior to 1972, the Republican candidate won a majority of the two-party vote only once.20 From 1972 forward, the Republican nominees for president captured a majority of southern districts in seven of the eight contests. The question here is whether there was a structural shift in southern electoral politics such that the relationships embedded in the model are different for the period after 1970. Examining this possibility is straightforward. First, I define two binary variables. One assumes a value of one for the elections from 1952 to1970; the second is set to one for elections from 1972 to 2000. Second, I multiply the independent variables, except for the proportions of blacks and hispanics (see below), by each binary variable. This will then enable me to estimate the impact of each variable for both the 1952-1970 and 1972-2000 periods. Principle 4: The model should be specified in a manner that affords the opportunity to evaluate the impact of redistricting eras and the substantial redistribution of the black population across the congressional districts of the south. Satisfying this principle is a variant of the strategy used for Principle 3. With respect to the distribution of black residents, there are five time periods of interest. The years from 1952 through
20

The year was 1968 and Nixon defeated Democrat nominee Hubert Humphrey in 64 of the 119 districts. When the Wallace candidacy is taken into account, Nixon won the three-party contest in 50 of the districts (42.0%), Wallace carried 48 districts (40.3%), and Humphrey won in 21 districts (17.6%).

9

1964 predate both the Wesberry case and the Voting Rights Act. Next is the interim period from 1966 to 1970, after the Voting Rights Act but before the near equalization of district populations for the election of 1972. Finally, there are the three decennial reapportionment and redistricting cycles: 1972-1980, 1982-1990, and 1992-2000. Accordingly, I define a binary variable for each period and multiply each by the measure of black population in the congressional districts. The same procedure is used to define separate measures of the hispanic population in the three redistricting cycles (1972-80, 1982-90, 1992-2000) for which there is data. --- Please Refer to Figure 10 --The Dependent Variable and Estimation. The choice of a dependent variable to represent the outcome of elections is not straightforward. The confounding feature is the incidence of uncontested elections in the south. The frequency of these “non-elections” is displayed in Figure 10.21 The figure reveals that uncontested elections reached a peak in the period from 1952-1960 at 61.3% (359/586) of all House elections. Methodologically, the frequency of uncontested elections introduces the problem of censoring or censored samples (Greene, 1997; Achen, 1986). In such instances, the distribution of the dependent variable, the two-party vote for Republican candidates, ranges from zero (no Republican runs) to 100 (Republican faces no opposition) with a more than trivial number of cases clustering at the upper and lower limits of the scale. Simply excluding these contests is not advised because much information about the relationships of interest is lost. There are two remedies. The first is to estimate the model using tobit, a maximum likelihood technique that specifically incorporates and accounts for the constraints at the upper and lower bounds of the variable (Greene, 1997: 962-974). In effect, tobit simultaneously estimates the impact of the independent variables on the probability of running and, among those who run, the proportion of the vote. Operationally, the estimated value of the dependent variable generated by tobit is based on a scale not bounded by zero and 100. Estimated values less than zero are taken as “predictions” that no Republican runs and values greater than or equal to 100 as “predictions” that no Democrat runs. The second approach involves two steps (Grier, Munger, and Roberts, 1994). The first is to use a probit model and obtain estimates of the probability that a Republican runs in each district.22 This enables one to ascertain whether not running varies systematically across congressional districts and, if so, estimate the impact of the

21

I have added the 1932-1940 and 1942-1950 periods to provide a more complete picture of the trend.

22

Technically, two probit models should be estimated – one for Republicans and one for Democrats. The Democrat model is not presented because it performs very poorly, indicating that there is not a systematic component underlying uncontested Republican wins. The coefficients are not significant and the model does a poor job correctly classifying the cases. These cases were included in the two-limit tobit procedure. While tobit does a satisfactory job classifying the Republican cases, the results for the Democrats are poor and parallel the Democratic results using the probit technique.

10

independent variables on that probability. In the second step, ordinary least squares are used on the truncated sample that excludes the uncontested races.23 --- Please Refer to Table 3 --For comparative purposes, I estimated the model of electoral outcomes using both strategies.24 The results are reported in Table 3. It should be emphasized that comparing the results should focus more on the patterns of statistical significance among the estimates rather than upon their magnitudes since the techniques assume different underlying scales. The estimates can be compared in two ways. First, comparing the tobit and probit results (columns 2 and 3) highlights the impact of the variables on the probability that a Republican will run in the district. Comparing tobit to the OLS results (columns 2 and 4) focuses upon the impact of the independent variables on the proportion of the two-party vote won by the Republican candidate. The estimates reveal, first of all, that there is a systematic component underlying the probability that a Republican will run in the district, especially in the 1952-1970 period when the incidence was high (see Figure 10). During the years from 1952 to 1970, there is a positive relationship between the probability of running and the two-party vote for Republican presidential candidates; the proportion of blacks, urban dwellers, and unionized workers in the district influences the probability of running negatively. In addition, Republicans were less likely to run in deep southern states. The probit estimates illustrate that Republicans were less likely to run when facing a Democrat incumbent; the tobit results are similar in that Republican incumbents were more likely to run and incumbency adds to the predicted vote share. With respect to the vote shares of Republican candidates, comparing the tobit results with the truncated OLS estimates reveals that more often than not, the patterns of statistical significance are similar. This holds true for the presidential vote in the district, median income from 1972-2000, the urban population as well as the incumbency and open seat measures. The most noteworthy results pertain to the impact of black and hispanic residents across the congressional districts. Here, the pattern of statistical significance is essentially invariant with respect to method. It is dramatic as well. Between 1952 and 1990, the proportion of blacks in a district is negatively related to both the probability that a Republican will run and the share of the vote won by Republicans who run.
23

There is a third alternative, the Heckman two-stage technique (Greene, 974-983). A two-equation model is specified; the first is a probit equation in which the probability of running, in this case, is estimated along with a variance adjustment parameter. This parameter enters into the second stage regression with the proportion of the vote as the dependent variable. The problem in this instances is that same independent variables operate on both equations. In effect, the system is underidentified. Note that there is one change in the specification for the probit routine. The incumbency variable in the tobit and OLS equations assumes a value of one if the incumbent is a Republican. In the probit equation, the variable takes on a value of one if the incumbent is a Democrat.

24

11

Following the redistricting based on the 1990 census, the relationships disappear. The coefficients essentially fall to zero in all three equations. A similar pattern holds for distribution of hispanics. In the 1972-1980 period, all coefficients are statistically significant. In the 1992-2000 period, all coefficients are indistinguishable from zero. --- Please Refer to Table 4 and Table 5 --Before examining these relationships in more detail, it is useful to evaluate the performance of the models. Table 4 and Table 5 are designed for this purpose. Table 4 evaluates performance of the tobit and probit models in classifying or predicting whether a Republican runs in the district for the entire time frame of the analysis as well as for the 1952-1970 and 1972-2000 periods. Each row of the table presents the overall proportion of cases that are classified correctly, the null prediction,25 and the proportion of correct predictions within each category of the dependent variable. Overall, the performance of the models is satisfactory. With one exception, the models outperform the null prediction and do better in classifying cases where a Republican runs (dependent variable equal to 1) than in cases where no Republican runs (dependent variable equal to 0). Nonetheless, the results do reinforce my earlier observation that there is a systematic component to uncontested elections in the 1952-1970 period. Tobit correctly classifies 59.1% (300/508) of these cases and probit does even better with a rate of 77.4% (393/508). The predictive performance of both models drops in the 1972-2000 period indicating that uncontested races were less systematic than in the previous era. Table 5 presents a similar comparison for the estimation of the two-party vote. The performance of the tobit and truncated OLS models are similar and reasonably strong.26 All classifications far surpass the null prediction. There is no substantial difference in the performance of the models across the time periods. While the models do better in classifying Democrat wins than Republican wins, the proportions of Republican wins correctly classified all exceed 70%. --- Please Refer to Table 6 --Table 6 is designed to provide more detail about the impact of the variables on the Republican share of the vote. It is based upon the estimates from the truncated OLS procedure. In addition to the estimates and median values (or proportions) in the data set, the table presents the impact range of each variable. This is calculated by multiplying the estimated coefficients by the minimum and maximum values of each independent variable. The results thus illustrate the range of impact, expressed as the addition or subtraction of vote percentages, for each variable. Finally, the table presents the results
25

The null model prediction is obtained by classifying all cases into the most frequently occurring category of the binary variable. For example, if Republicans do not contest 60 of 100 districts in a given year, the null prediction is that Republicans do not run in any district. This would produce a correct prediction rate of 60%. To obtain the proportions correctly classified, I used the estimated values of the dependent variable. Estimates greater than 50 were categorized as “predicted wins” and those below 50 as “predicted losses.” I then used the actual vote to create a win-loss variable and compared it to the predicted value.

26

12

of the F-test for the equality of coefficients. This is used to test whether there was a structural change in the impact of the variables between the 1952-70 and 1972-2000 periods. In the case of the black and Hispanic measures, the test pairs a given time period with the preceeding time period. The results show a significant change in the impact of the presidential vote and median income in the districts. The value of Republican incumbency increases, as does the likelihood of Republicans winning open seats. The table also reinforces the conclusions that there was a substantial change in the relationship between the black population of congressional districts and the Republican vote. The impact range is greatest in the first period, 1952-1964. This suggests that Democrat incumbents were most entrenched in districts with large numbers of blacks. This is consistent with the tobit and probit results indicating that Republicans were deterred from running in such districts. In the next three periods, from 1966 until 1990, there is no significant change in the coefficients, as evidenced by the F-tests. The median value, however, declines and thus the median impact drops. After the 1990 round, the electoral impact completely disappears. The proportion of black residents in a district exerts no effect on the share of the Republican vote. There is a similar effect for the distribution of hispanic residents. The relatively modest impact in the 1972-80 period disappears in the two subsequent periods.

Part III Explaining Member Ideology
The question addressed in this section flow from the observation that ideological change follows in the wake of electoral and partisan change. My objective is to ascertain whether the changes shown to influence the outcomes of elections also exercise an effect on the roll call voting of House members elected from the south. To do so, I follow the logic of the design principles developed in the previous section to develop a model of the ideology in roll call voting. Thus, the model includes pairs of variables, representing the 1952-70 and 1972-2000 periods, for the presidential vote, median income, urban dweller and unionized proportions in each congressional district. The model also includes a pair of binary variables for Republican incumbents in the two periods and separate binary variables for black and hispanic members who served in the House between 1972 and 2000. To account for maturation or aging effects, I include pairs of variables – one pair for Democrats and one for Republicans – that measure the number of terms served in the House. These variables effectively control for those situations in which a member drifts leftward or rightward over the course of their career. To assess the impact of the district proportion of black residents, I retain the fiveperiod distinction used in the electoral analysis. The key to this analysis is to determine whether the proportion of blacks in the district influences Democrat and Republican members in different ways over time. Accordingly, the model includes separate sets of variables for Democrat and Republican members. The measures of hispanic population are specified in the same manner. The model thus includes a set of three hispanic population variables for Democrats and Republicans.

13

--- Please Refer to Table 7 --The dependent variable in the analysis is the Poole-Rosenthal DW-Nominate score of each House member.27 Because the variable is a continuous measure, OLS is an appropriate estimation technique.28 The results of estimating the equation are presented in Table 7. The table displays the estimated coefficients, their standard errors, and a summary of the performance of the model. The summary indicates that the overall performance of the model is satisfactory; the F-statistic is significant and the model explains 68% of the variance in scores. Turning to the estimated coefficients, note that a positive coefficient indicates that the variable influences the scores in a conservative direction while a negatively signed coefficient represents an influence in the liberal direction. The estimates first show that there are sources of continuity across the time periods. The Republican proportion of the presidential vote is a case in point. It exercises a conservative influence. As the presidential vote rises across districts, so too does the conservatism of the representative. The proportion of urban residents and unionized works exert their influence in a liberal direction. The estimates for the binary variables denoting the party and ethnicity of members conform to expectations as well. In both periods, Republicans members are more conservative than the average white Democrat member; hispanics are more liberal than white democrats, and blacks are more liberal than hispanics. --- Please Refer to Table 8 --By far, the most dramatic results pertain to the impact of black and hispanic residents on member ideology. This can been seen in the estimates themselves and in Table 8 which displays the estimates along with the medians, the impact range, and the F-tests for a structural shift in the coefficients. Consider white Democrats first of all. In both the 1952-64 and 1966-70 periods, the proportion of blacks in a district influenced Democrats in a conservative direction. This relationship changes abruptly in the wake of the 1972 round of redistricting. The sign of the coefficient switches denoting a structural shift in influence from conservative to liberal (an F value of 51.4). As illustrated by the magnitude of the coefficients and impact ranges, the strength of the relationship increases after both the 1982 and 1992 redistricting cycles. There are structural shifts in Republican districts as well. Prior to 1972, the coefficients for the proportion of black residents, while positively signed, are statistically insignificant, indicating a negligible influence on ideology. The relationship changes after the 1972 redistricting. The proportion of black residents influences Republican roll
27

The scores from the first (economic, redistributive) dimension are used. The second dimension, which taps civil rights and north-south divisions, is interesting but problematic to use. As Poole and Rosenthal (1997) note, the clarity and salience of this dimension declined during the 1970s. This predates the changes examined in this analysis. To aid with the interpretation of the coefficients, I multiplied the DW-nominate scores by 100 to place the variable on a scale ranging from –100 to 100.

28

14

call voting in a liberal direction with an impact range of approximately 17 points. This relationship remains intact after the 1982 round of redistricting with an impact range over 12 points. Note, however, the decline in the median proportion of blacks residing in Republican districts. The proportion reaches its low of point of 8% following the 1992 round of redistricting. Importantly, there is a significant change in relationship between the proportion of blacks in a district and the roll call voting of Republican members. The coefficient becomes positively signed, indicating that larger black populations in Republican districts now increase the conservatism of the representative. Thus, over the time frame of the study, there are structural changes in the relationship between the proportion of blacks in a district and member ideology within both parties. At present, as the proportion of black residents in a district increase, Republican members grow more conservative while Democrat members grow more liberal. As Table 8 demonstrates, the same relationship holds for the hispanic population as well.

Summary and Conclusions
The south presents a dramatic and interesting story about political change in American politics. In this analysis, I have attempted to examine political change as it pertains to House elections. My analysis consists of three parts and leads to a number of conclusions. The first part of the analysis is largely descriptive and based upon examining the distribution of the general population and black residents across the congressional districts of the south. Essentially, it leads to the conclusion that the landscape of House elections in the south has changed dramatically. In 1952, the south was a region characterized by a malapportioned set of congressional districts with a relatively even distribution of black residents. After 1992 in particular, it was a region of equally and legally apportioned districts with a malapportioned black population. This change has an important implication. If, as Black and Black (2003) suggest, the key to Democratic electoral fortunes in the south involves the formation of bi-racial coalitions, the trend in “packing” the black population into a handful of majority-minority districts reduces the number of districts in which there is an opportunity to build such coalitions. The second part of the analysis is based upon a model of electoral outcomes in House elections. There are two major conclusions from this exercise. First, I have shown that the absence of competition, a characteristic of the “old south,” can be understood in a systematic manner. The probability of a Republican running in a district, particularly in the era from 1952 until 1970, varied in a systematic manner with the demographic characteristics of the district, including the distribution of the black population. Second, I have shown that after the 1990 round of redistricting, the distribution of the black population across the congressional districts in the south was no longer a significant factor in accounting for the outcomes of these elections. The creation of majority-minority districts appears to involve a trade off. Without a doubt, these districts enhance black and hispanic influence and representation. Ironically, when viewed from a regional and partisan perspective, their creation concentrates and thereby dilutes the influence of these groups.

15

The last part of the analysis identifies a number of structural changes in the relationship between the voting ideology of southern representatives and distribution of the black and hispanic populations. Among Democrat members, increases in the proportion of black constituents now influence roll call voting in a liberal direction whereas before 1972, the influence was in the conservative direction. This is a major change that is useful for distinguishing the “old south” from the “new south.” The relationship among Republicans has changed as well. In fact, it has changed twice – from no influence prior to 1972 to a liberal influence from 1972 until 1990 and then to a conservative influence from 1992 forward. When viewed in tandem with the changes among Democrats, we can conclude that since 1952, the landscape of southern politics has changed from consensual and anti-egalitarian to one that is more egalitarian but more polarized as well.

16

References
Achen, Christopher. 1986. The Statistical Analysis of Quasi-Experiments. Berkeley: University of California Press. Aldrich, John. 2000. “Southern Politics in State and Nation.” Journal of Politics 62: 643-670. Ardoin, Phillip and James Garand. 2003. “Measuring Constituency Ideology in U.S. House Districts: A Top Down Simulation Approach.” Journal of Politics 65:1165-1189. Berelson, Bernard, Paul Lazarsfeld, and William McPhee. 1954. Voting. Chicago: University of Chicago Press. Black, Earl, and Merle Black. 1992. The Vital South: How Presidents Are Elected. Cambridge, Mass.: Harvard University Press. Black, Earl, and Merle Black. 2002. The Rise of Southern Republicans. Cambridge, Mass.: Harvard University Press. Bond, Jon R. and Richard Fleisher, eds. 2000. Polarized Politics: Congress and the President in a Partisan Era. Washington D.C.: Congressional Quarterly. Butler, David and Bruce Cain. 1992. Congressional Redistricting. New York: Macmillan. Campbell, Angus, Phillip Converse, Warren Miller, and Donald Stokes. 1960. The American Voter. Chicago: University of Chicago Press. Canon, David. 1999. Race, Redistricting, and Representation: The Unintended Consequences of Black Majority Districts. Chicago: University of Chicago Press. Carmines, Edward and James Stimson. 1989. Issue Evolution: Race and the Transformation of American Politics. Princeton, N.J.: Princeton University Press. Greene, William. 1997. Econometric Analysis. 3rd Edition. Upper Saddle River, N.J.: Prentice Hall. Grier, Kevin, Michael Munger, and Brian Roberts. 1994. “The Determinants of Industry Political Activity, 1978-1986.” American Political Science Review 88: 911-926. Hill, Kevin. 1995. “Does the Creation of Black Majority Districts Aid Republicans?” Journal of Politics 57: 384-401. Key, V. O. Jr. 1984 (1944). Southern Politics in State and Nation. Reprint; Knoxville:

17

University of Tennessee Press. Lazarsfeld, Paul, Bernard Berelson, and Hazel Gaudet. 1944. The People’s Choice. New York: Columbia University Press. Nadeau, Richard and Harold Stanley, “Class Polarization in Partisanship among Native Southern Whites.” American Journal of Political Science 37 (August 1993): 900-919. Petrocik, John and Scott Desponsato. 1998. “The Partisan Consequences of MajorityMinority Districting in the South, 1992 and 1994.” Journal of Politics 60: 613-633. Poole, Keith and David Rosenthal. 1997. Congress: A Political-Economic History of Roll Call Voting. New York: Oxford University Press. Stanley, Harold. 1988. “Southern Partisan Changes: Dealignment, Realignment, or Both?” Journal of Politics. 50: 64-88.

18

19


								
To top