ó Disappearing Democrats: Rethinking Partisanship Within Pennsylvania's Electorate Berwood Yost Franklin and Marshall College Pennsylvania has more registered Democrats than Republicans, yet Repub­ licans cOlltrol most 0/ the slalewide elecled offices and oath ella/llbers uf lhe biute legislature. How is it possible that the Democrats' registration advantage does not lead to more Democratic electoral success? Using a random survey ofPeT1n­ sylvania voters and comparative questions on party orientation and party regis­ tration, this article finds that many of the state's voters identify with a political party that is different from the party in which they are actually registered. Party identification is important because it is a better predictor of voting behavior than is party registration. The article finds that more registered Democrats in Pennsylvania ident~fy with the Republican Party than vice versa, making the Democrats'registration advantage an advantage i11 name only. Conducting surveys of Pennsylvania's electorate requires a researcher to ask a question of political partisanship different from the self-identifi­ cation measures used in the University of Michigan voting studies. The Michigan measure captures psychological identification,1 but it does not provide an accurate accounting of Democrats, Republicans, and inde­ pendents when compared with Pennsylvania's voter registration list. Producing a set of registration figures that accurately reflects regIstra­ tion is important in Pennsylvania. Unlike a national survey, the state's .voter registration figures are readily available for all to see, meaning that reporters who cover politics and politicians and who read the polls in­ variably ask about the partisan distribution measured by political sur­ veys. In statewide surveys, they quickly discount polls that do not pro­ vide a mirror image of the state's partisan distribution. Pollsters who have released polls in hotly contested election campaigns have been criti­ cized when their sample's partisan distribution did not approximate the state's known voter registration distribution. As a result of this criticism, the Keystone Poll measures actual party registration to ensure a partisan distribution of respondents that 111.atches the state's known registration figures. 2 Using a two-part question, the Keystone Poll has repeatedly obtained samples of registered voters that approximate within expected error ranges the state's actual registration figures. Recent developments in Pennsylvania politics call this approach into question. Since the 1950s, Pennsylvania has been considered a competi­ tive two-party state, but in the 1990s the state became dominated by Re- 77 publican officeholders. In 1998, registered Democrats outnumbered reg­ istered Republicans statewide by 445,000 voters, including a 500,000 voter registration margin in Philadelphia and a 300,000 voter registration mar­ gin in Allegheny County.3 Yet, Governor Torn Ridge was a Republican. Both of the state's two u.s. senators and two of its three row officers were Republicans. In the state legislature, Republicans held 103 of 203 seats in the House and 30 of 50 seats in the Senate. In the 1998 election cycle, Republican candidates for both the House and Senate collected more to­ tal votes than did Democrats.4 Thus, measuring actual voter registration may no longer provide an accurate picture of the state's electorate. How can the Democratic Party have such a decisive voter registration edge, yet manifest such weak electoral performance? Conversely, how can Republicans win so many elections when they trail so significantly in the number of registered voters? Some political analysts attribute the Democratic losses to lower voter turnout among Democrats. A greater percentage of Republicans than Democrats go to the polls, they say, erod­ ing the Democrats' numerical superiority. Other analysts theorize that a group of IIReagan Democrats," conservative on issues such as welfare spending and abortion limits, cross party lines to vote Republican. An­ other explanation is that the state Democratic Party suffers from greater heterogeneity among it registrants than does the state Republican Party, making Democrats less likely to maintain party unity, a trend that has been observed nationally (Mayer, 1996). An unspoken assumption un­ derlies these explanations: that voters faithful to the Democratic Party exist in greater numbers, sprawling across the Commonwealth like a sleeping tiger. Given the right Democratic candidate and the right is­ sues, this line of thinking goes, Democratic voters will awake, voting with the full strength of their numbers, and deliver victory to their party. But is this widely held assumption true? Are Democrats still the majority party in the state, as the voter registration rolls indicate'? This article has three objectives. First, it will quantify the differences that exist in the makeup of the state' 5 electorate by using two different measures of partisanship. Second, it will suggest which of these two mea­ sures of partisanship more accurately predicts election outcomes; Finany, it will identify the characteristics of voters who register with one party but identify with another. The Measurement of Partisan Identification Partisan identification is one of the most important and frequently used concepts in the study of voting behavior (Abramson and Ostrom 1994). An earlier generation of scholars considered party identification worthy of study for two reasons. First, they expected party identification to be stable over time. Second, they expected it to act as a cue by which 78 voters would evaluate political events (Campbell et a1. 1964). This is not to say that party identification would not or could not change over time, but only that it would not change easily. Furthermore, not all voters were expected to align themselves with a party; a sufficiently large number of voters would never have strong party attachments, making them subject to short-term political forces that could affect electoral outcomes (Campbell et a1. 1964). Party identification was thought to be the major device with which most people would deCipher politics. The most stud­ ied, commonly used, and reliable measure of partisan identification was developed by the Survey Research Center (SRC) at the University of Michigan (Abramson and Ostrom 1994). The expectations for the party identification measure as originally designed are best summarized by its creators: The initial selection of a party may often be a response to nonpo­ litical pressures; once made, partisan choice tends to be maintained long after its nonpolitical sources have faded into oblivion. Cur­ rent pressures arising outside the political order continue to affect the evaluation process, and from time to time they may contrib­ ute to a critical margin of political victory. Yet for most of the people most of the time such contemporary forces turn out to be the mi­ nor terms in the decision equation. (Campbell et a1. 1964,66) Whether party identitication oUers such stability has been questioned in recent years. Researchers who doubt the stability of party attachments believe that a great deal of short-term variability takes place in reaction to short-term forces, such as consumer confidence and presidential per­ formance, that is missed because of the timing of the Michigan studies.s Using another long-running series of survey data collected by the Gallup Organization, these authors have found that partisanship is subject to wide swings as a result of short-term forces. Recent studies, however, refute this notion. Some authors have found that short-term forces do not have a large impact on partisanship as measured by the Michigan question.6 Most scholars now believe that party identification is a very stable concept, as originally thought, and that the instability of partisan identification found by some authors was the result of measurement er­ ror created by question wording? Mayer (1996, 97) summarizes current thinking on the subject in saying that most Americans "develop a psy­ chological tie to one of the parties in their late adolescence or early teen years. And once formed, party identifications tend to be remarkably stable over the rest of a person's life." As a measurement tool, the Survey Research Center's party identifi­ cation question has proven to be highly reliable, with only one-ninth of its variance resulting from measurement error (Shickler and Green 1997). 79 Although the Michigan guestion is considered a valid way to measure an important and stable concept, using it to measure the makeup of a single state's electorate has been problematic. The actual number of Re­ publican and Democratic registered voters within a state often differs from what is (ouflJ uy surveys using the SRC measure. According to Epstein (1986t this could mean that some people express a party prefer­ ence in registering to vote that they do not actually hold. It also means that some registered partisans adhere to their party label less strongly than do other registrants. Why would this happen? There are several situations where one would expect to obtain registration figures that dif­ fer from those found in sample surveys. Possible reasons include social pressures, job holding or seeking, or the desire to vote in the primary of a locally dominant party (Epstein 1986). The desire to vote in primary elections is presumed to be especially important in a state like Pennsyl­ vania, which has a closed primary system. Epstein (1986, 247) reports, U \:;tCltes that most strongly encourage party registration, by firm1y dos­ ing their primaries to unaffiliated voters, record not only higher party registration percentages but also higher percentages of party identifiers than surveys in states with more permissive registration procedures." Besides suggesting that measuring party identification is not the same thing as measuring party registration, the different figures resulting from party identification and party registration in some states raise some in­ triguing questions. If there are differences between party identification and party registration, do they change the state's partisan balance? Which measure is a better predictor of voting behavior? What are the character­ istics of voters whose party registration and party identification differ? The remainder of this article seeks to answer these questions. Methodology The data presented her come from a state-wide survey of 451 Penn­ sylvania voters (sampling error of ± 4.6 percent) conducted between July 8 and 28, 1998. Respondents for the survey were selected at random us­ ing a two-stage process. First, telephone households were randomly se­ lected using a random-digit-dialing sampling method (Chummings 1979). Once a residential telephone number was identified, a respondent within each household was selected at random using the last birthday method of respondent selection (Salmon and Nichols 1983). This article focuses on responses to three survey questions. The first question, involving nominal party registration, asked the following: "Many people are registered to vote; however, many others are not. How about you? Are you currently registered to vote at your present address?" All registered voters were asked, II Are you currently registered as a Demo­ crat, a Republican, an independent, or something else?"8 80 Table 1 Actual Party Orientation and Nominal Party Registration Actual Party Nominal Party Orientation Orientation Rep. Ind. / Other Dem. Total Republican / Lean Rep. 90% (165) 49% (22) 11%(21) 50% (208) True Independent 4% (8) 24% (11) 7% (14) 8(Yo (33) Democrat / Lean Dem. 6%(11) 27% (12) 81 % (154) 42% (177) Total 44% (184) 11%(45) 45% (189) The second question, involving actual party orientation, used the tradi­ tional Michigan formulation: "In politics, as of today, do you think of yourself as a Republican, a Democrat, or an independent?" Those who said they were either a Republican or a Democrat were then asked, "Would you call yourself a strong (Republican / Democrat), or a not very strong (Republican/Democrat)?" Those who said they were indepen­ dent were asked, "Do you think of yourself as closer to the Republican or Democratic Party?" The party orientation measure was recoded for analysis; strong, weak, and leaning partisans were grouped together to produce a trichotomous measure. This recoding seems reasonable be­ cause independent leaners are attitudinally and behaviorally similar to weak partisans and different from "pure" independents (Smith et al. 1995). A similar grouping strategy has been employed by others.9 The final question measured respondents' voting tendencies in re­ cent elections. The question asked: "Thinking about the last few state and national elections, which best describes how you voted: straight Democrat, mostly Democrat, a few more Democrats than Republicans, about equally for both parties, a few more Republicans than Democrats, mostly Republican, or straight Republican?" In the tables that follow, responses were recoded to reflect a preference for more Democratic can­ didates, more Republican candidates, or about equal preferences. Findings The survey reveals very different pictures of the electorate depending upon which measure of partisanship is employed. Table 1 shows the sig­ nificant difference between nomina] party registration and actual party orientation. The party registration totals (bottom row) reflect the Demo­ cratic Party's nominal voter registration advantage. In contrast, the party orientation totals (right column) show the actual voler advanlage of the Republican Party. 81 Table 2 Votes in Recent Elections by Actual Party Orientation Recent Votes More Democrats Equal Actual Party Orientation Rep. Ind. / Other Dem. 5% (10) 17% (34) 14% (4) 61 % (17) More Republicans 78% (158) 25% (7) 78(10 (134) 20% (34) 2% (4) A= .60 A2 = 290.476, df = 4, p. < .001 Ninety percent of Republicans identify with the Republican Party, 81 % of Democrats are oriented to the Democratic Party, and 24% of indepen­ dents are tndy independent. Conversely, a few Republicans (10%), more Democrats (19%) and most independents (76%) do not adhere to the po­ litical party indicated by their registration. Finding that a large number of independent voters do not adhere to theil- registraLion runs COU1lter tu expectations. If, as Epstein (1986) suggests, people register with a major party in order to be able to vote in primaries, one might expect registered independents to be least likely to adhere to a major party label and most committed to their independent status. This simply is not the case. Table 1 reveals a phenomenon that can be called the misaligned voter­ those who have registered in one party but now see themselves as more closely attached to another. Twenty-one percent of voters are misaligned - on the rolls as members of a party that they no longer ~l1prort. TWlrP as many misaligned voters are registered as Democrats (40%) or indepen­ dents (39%) than as Republicans (22%) [A2 = 93.896, df = 2, p. < .001]. When respondents indicated how they had voted in recent elections, actual party orientation (Table 2) proved a better indicator of their choices as measured by the statistic lambda particularly for Democrats and ind~pend~nts - than did nominal party registration (Table 3).10 From a demographic standpoint, who are the misaligned? Table 4 summarizes the results of a discriminant function analysis used to clas­ sify misaligned voters. The goal of discriminant analysis is to predict membership in two or more mutually exclusive groups from a set of predictor variables. The utility of discriminant function analysis rests with its ability to accurately classify subjects into groups. This information is conveyed by the classification rate, which indicates the percentage of subjects for each analysis that are successfully assigned to the group to which they actually belong using the relative discriminant function. Dis­ criminant analysis identifies the variables that are most useful for pre­ dicting group membership. 82 Table 3 Votes in Recent Elections by Nominal Party Orientation Recent Votes Actual Party Orientation Rep. Ind. / Other Dem. More Democrats 6% (10) 20'10 (10) 72% (138) Equal 16<10 (29) 37% (18) 24% (46) More Republicans 79% (143) 43% (21) 4°/c) (8) A= .51 A2 = 249.974, df = 4, p. < .001 Table 4 shows that party registration and region of residence are the best predictors of being misaligned. Using these variables to predict whether a person is misaligned yields 40% fewer errors than what would be expected if respondents were classified by chance alone (62 real errors versus 104 expected errors). Variables such as age, education, political ideology, religion, marital status, and gender do not add any classifica­ tory power to the analysis. As Table 4 indicates, there are more misaligned voters in southeast­ ern Pennsylvania (Bucks, Chester, Delaware, and Montgomery counties) than in any other area of the state. Misalignment for all parties is higher there than in the rest of the state and it is not confined to one party. Table 4 Results of Discriminant Function Analysis for Misaligned Votersll Nominal Party Registration Misaligned Not Misaligned H,epublican Democrat Independent Region of State Southeast Rest of State tau =.40 Press's Q = 199.50, P < .01 10% 19% 76% 30% 19% Percent of cases correctly classified == 85 83 90% 81% 24% 70% 81% Conclusions Clearly, these findings should be considered provisionat for they rep­ resent only a single point-in-time survey that calls for replication. The find­ ings do, however, point to some conclusions that are worth considering. Pirst, for some people, actual party orientation is different from nomi nal party registration. If replicated, this finding may have enormous im­ plications for how we observe and make sense of electoral behavior. In an era of near universal registration as a result of motor-voter legisla­ tion, party registration may have less relevance for making sense of po­ litical behavior. Second, more of the state's voters are oriented to the Republican Party than the voter registration rolls suggest. If state voter registration num­ bers are adjustpd to Tpf]pct political orientation, Republicans outnumber Democrats by a 250,000 voter margin. This is a profound difference in partisan alignment that by itself can explain the anomaly in Pennsylva­ nia of Republican hegemony despite Democrat registration superiority. Third, actual party orientation is a better predictor of voting behavior than is nominal party registration, although both measures perform well. Unless both measures are considered, large differences in correlations with vote choice may be overlooked. Moreover, because actual party ori­ entation is a better predictor of behavior it provides a more complete picture of the electorate. This conclusion strongly suggests that pollsters should consider using both measures when possible. Fourth, some intriguing geographic patterns show up in the data. In particular, southeastern Pennsylvania may be an area worthy of addi­ tional study since more misaligned voters appear there than anywhere else in the statc. Why this particular region differs from others is not clear. A number of demographic factors including suburbanization and migration may be operating here. If so, they are important to understand­ ing better the role region may play in these findings. Finally, the large number of independent voters not adhering to their registration confounds expectations. If people register with a major party in order to vote in primaries, as Epstein (1986) contends, then registered independents ought to be less likely to adhere to a major party label and mORt committed to being independent. Yet they are not. The disjunction between independents and voter registration looms as one of the more perplexing findings reported. Since they mostly vote for major party can­ didates, why do independent voters not register with that party? Per­ haps their voting behavior is determined in large part by the paucity of independent candidates running for office in Pennsylvania. Additional research is needed to confirm these findings and to ex­ plore further other underlying questions. Among the more tantalizing 84 questions are these: When and why did misaligned voters change their political orientation? Why have they not changed their registration to reflect their new orientation? Is misalignment a permanent or temporary condition? Notes 1. The development and utility of the Michigan measure is discussed in great detail in Campbell et al. (1964, 67-96). 2. The Keystone Poll is produced at Franklin and Marshall College on behalf of the Philadelphia Daily News, the Harrisburg Patriot News, and the Pittsburgh Tribune Review. The Keystone Poll is conducted four times each year among the state's voting age population. The pan was produced at Millersville University between 1991 and 2002. 3. An registration figures are from Official Voter Registration Statistics (1998). 4_ Tota1ing those races where two candidates werlO! running yielded the following results: State House, Democrats = 1,174,147 votes, Republicans = 1,317,571 votes; State Senate, Democrats = 607,634 votes, Republicans = 705,301 votes. Calculations by author from data published by the Pennsylvania Department of State. S. The most commonly cited source for this argument is MacKuen, Erikson, and Stimson (1989). 6. See, for example, Box-Steffensmeier and Smith (1996); Green, Palmquist, and Schickler (1998); Green and Shickler (1993); McAllister and Wattenberg; (1995); Rice and Hilton (1996); and Schickler and Green (1997). 7. Abramson and Ostrom (1994) and Green, Palmquist, and Shickler (1998) found that the instability attributed to partisan identification was a result of the wording of the Gallup question. Because of its wording, the Gallup macropartisanship measure is much more susceptible to changes resulting from short-term trends. There is a sizable literature related to the effect of question wording on survey response. Schuman and Presser (1996) provide an excellent introduction to this literature. S. The nominal registration item produced a partisan distribution comparable to that found within the state, according to registration statistics. The survey produced a sample with 200 Democrats (44%), 190 Republicans (42%) and 61 independent/ other voters (14%). The actual registration figures for Pennsylvania put the distribution of voters at 48.6% Democrat, 42.4% Republican, and 8.9% other. 9. This grouping strategy appears in Abramowitz and Saunders (1998) and Box-Steffensmeier and Smith (1996). 10. Lambua (A) ~how~ the proportion by which error in predictin~ the value of the dependent variable is reduced by knowing the value of the independent variable (Weisberg, Krosnkk, and Bowen 1996, 274). 11. Statistical significance for each analysis is measured by Press's Q statistic and Tau. Press's Q is a measure of the classificatory power of the discriminant function when compared to the results expected from chance (Hair et a1. 1992). Tau is a proportional reduction in error statistic that yields a :,tandardized measure of classificatory improvement (Klecka 1980). 85 References Abramowitz, Alan 1., and Kyle L. Saunders. 1998. "ldeologkal Realignment in the u.s. Electorate." Journal of Politics 50 (3): 634-652. Abramson, Paul R., and Charles W. Ostrom. 1994. "Question Wording and Partisanship." Public Opinion Quarterly 58: 21-48. Box-Steffensmeier, Janet M., and Renee M. Smith. 1996. "The Dvnamics of Aggregate Partisanship." American Political Science Review 90 (3): 567-581. Campbell, Angus, Philip Converse, Warren Miller, and Donald Stokes. 1964. The American Voter, An Abridgement. New York: John Wiley & Sons. Cummings, Michael K. 1979. "Random Digit Dialing: A Sampling Technique for Telephone Surveys" Public Opinion Quarterly 47: 233-244 Epstein, Leon. 1986. Political Parties in the American Mold. Madison, WI: University of Wisconsin Press. Green, Donald P., Bradley Palmquist, and Eric Shickler. 1998. "Macropartisanship: A Replication and Critique." American Political Science Review 92 (4): 883-899. Green, Donald P., and Eric. Shickler. 1993. "Multiple-measure Assessment of Party Identification." Public Opinion Quarterly 57: 503-535. Hair, Joseph, Rolph Anderson, Ronald Tatham, and William Black. 1992. Multivariate Data Analysis with Readings. MacmHIan: New York. Klecka, William R. 1980. Discriminant Analysis. Sage University Paper 19, from the Quantitative Applications in the Social Sciences series. MacKuen, Michael B., Robert S. Erikson, and James A. Stimson. 1989. "Macropartisanship." American Political Science Review 83: 1125-1142. Mayer, Wi11iam G. 1996. The Divided Democrats. Boulder, CO: Westview Press. McAllister, Ian, and Martin P. Wattenberg. 1995. "Measuring Levels of Party Identification: Does Question Order Matter?" Public Opinion Quarterly 59 (2): 259-269. Official Voter Registration Statistics. 1998. Commonwealth of Pennsylvania. Department of State. August. Rice, Tom W., and Tracey A. Hilton. 1996. "Partisanship Over Time: A Comparison of United States Panel Data. Political Research Quarterly 60 (1): 191-201. Salmon, Charles, and John Spicer Nichols. 1983. "The Nex't Birthday Method of Respondent Selection." Public Opinion Quarterly 47: 270-276. Schuman, Howard, and Stanley Presser. 1996. Questions and Answers i1l Attitude Surveys. Thousand Oaks, CA: Sage Publications. Shickler, Eric, & Donald P. Green. 1997. "The Stability of Party Identification in Western Democracies: Results from Eight Panel Surveys." Comparative Political Studies 30 (4), 450-484. Smith, Andrew, Alfred Tuchfarber, Eric Rademacher, and Stephen Bennett. 1995 . . "Partisan Leaners Are Not Independents." The Public Perspective: October / November, 9-12. Weisberg, Herbert, Jon Krosnick, and Bruce Bowen. 1996. An Int1'Oductiol1 fo Survey Research, Polling, and Data Analysis. 3rd ed. Thousand Oaks, CA; Sage Publications. 86