|
|
Methodological AppendixMy goal in writing Congressional Elections was to evaluate systematically and comprehensively congressional election campaigns. I pursued ten major objectives in carrying out the research that forms the basis of the book:
Although each edition of the book has taken into account new developments in election law, campaign finance, and the conduct in elections, the generalizations gleaned from the research have remained consistent throughout each edition. Data SourcesAddressing the first eight objectives required a major data collection effort. I conducted roughly 100 semi-structured personal interviews with congressional candidates and campaign aides in the 1992, 1994, 1996, 1998, 2000, and 2002 elections to learn about specific campaign strategies. Students in my undergraduate honors seminars on congressional elections supplemented this information with materials collected for another 110 campaigns, which they collected while working on their course projects. Each student tracked both the Democratic and Republican campaigns conducted in one House election. In the course of their research, they interviewed candidates, campaign managers, other campaign aides, and journalists. They also systematically collected press clippings, issue papers, and other materials that were relevant to the elections they researched. Some, but not all of these materials were gathered from the Internet. All of the interviews were guided by a common questionnaire.[1] I further supplemented these materials with information I had collected during the 1984, 1986, and 1988, and 1990 elections. The information on party strategy is based mainly on over 100 interviews that I conducted during the 1992, 1994, 1996, 1998, 2000, and 2002 elections. National party officials were interviewed about their organizations' goals, expectations, and strategies on several occasions over the course of these elections. The interviews enabled me to recognize how these organizations formulated their initial strategies and adjusted their tactics in response to changing conditions. As was the case with the information collected from candidates and campaign aides, this information was supplemented with materials collected during prior elections. The information on PAC and interest group decision making is based on numerous interviews that I conducted over the course of the aforementioned elections and on interviews that were conducted by some colleagues and graduate students as part of related studies of PACs and interest groups.[2] Other information was gathered from newspapers, periodicals, candidates' Web sites, and discussions with a wide range of political operatives with whom I spoke at conferences, strategy meetings, fundraising events, and more informal occasions. As the study progressed, parts of the research were made available to some of the participants, allowing for the establishment of a continuing dialogue. Feedback from the participants was occasionally used to illuminate some of the observations I had made earlier. The information collected from the interviews and informal conversations, and my own participation on political campaigns, enabled me to describe congressional elections from the perspectives of the individuals who are most deeply involved in the electoral process. I turned to a variety of sources for information about different aspects of congressional election campaigns. Information on the backgrounds of candidates who ran prior to 1998 was culled from issues of CQ Weekly; the 1998 data were obtained directly from CQ's database. These data were supplemented with information from files kept by various party organizations and PACs, the Joint Center for Political and Economic Studies, the National Association of Latino Elected Officials, the Native American Caucus; from Asia Week, various editions of Who's Who in American Politics, Politicis in America, and The Almanac of American Politics; and from interviews with candidates and state party committee staffs. For the 2002 elections, I turned to similar sources and to various candidate, political party, and other political web sites, as well as to newspaper articles searched on Lexis-Nexis and other online newspaper indexes. Campaign finance data were supplied by the Federal Election Commission, the Center for Responsive Politics, Political Money Line, and Public Citizen. I relied on the most recent versions of campaign finance data available when doing the research; these data are continuously being “cleaned” and updated. The 2002 Congressional Campaign Study, a survey of candidates and campaign aides who competed in the 2002 House and Senate elections, was an important source of information about the candidates and campaign organizations, budgets, strategies, issue positions, and related topics (see Figure A-1). Data were collected using a questionnaire that was distributed through the mail, over the telephone, via fax, using e-mail, and in person. An e-mail was sent on October 21, 2002 to alert House campaigns to the questionnaire’s arrival. It included the questionnaire to allow campaigns that were comfortable completing it before election day to do so. Additional questionnaires were e-mailed October 29, 2002, November 6, 2002, and December 13, 2002. Paper copies of the questionnaire were sent via U.S. mail to campaign headquarters on October 30, 2002 to ensure they would arrive by election day. Another wave of questionnaires was mailed on November 22. Questionnaires were faxed to campaign headquarters between December 5 and December 20, 2002, at which time telephone interviews also were conducted. Personal interviews were conducted on December 13, 2002. The process netted a total of 322 usable observations from House campaigns: 41.9 percent of the respondents exercised the mail option, 29.2 percent completed the survey over the telephone, 18.9 percent responded via e-mail, 8.7 percent responded by fax, and 1.2 percent participated in a personal interview. The response rate for the House survey was 45.1 percent, and the sample closely represents the underlying population of candidates on important characteristics, such as party, incumbency, and election outcomes (see Table A-1). A slightly different questionnaire was distributed to Senate campaign organizations using similar methods. However, because Senate elections are less central to the book than House elections, less effort was given to coaxing Senate campaigns into completing the questionnaires after the campaigns were initially contacted. Because of a smaller sample size, the data from the Senate campaigns was analyzed with greater caution, using less powerful statistical techniques, and mainly to draw loose comparisons with House campaigns. The 2002 Congressional Campaign Study replicates many parts of The 1998 Congressional Campaign Study, The 1992 Congressional Campaign Study, and The 1984 Congressional Campaign Study. It confirms the vast majority of the generalizations gleaned from its predecessors. However, because it was not possible or desirable to replicate each study in its entirety, a small portion of the information contained in this edition of the book, including some of the generalizations about Senate campaigns, is based on information collected in previous studies.[3] Quantitative Data AnalysisNo distinctions were drawn between the responses of candidates, campaign managers, and other campaign aides when analyzing the data collected in The 2002 Congressional Campaign Studyor its predecessors.[4] There are several reasons for treating these responses equally. First, because campaign decision making is a cooperative endeavor involving a small group of decision makers, these individuals can be expected to have similar perceptions of the important factors in and determinants of the campaign.[5] Second, during pretests of the questionnaires, several members of the same campaign provided responses that were nearly identical. Third, in about a dozen cases a copy of the survey was received from a candidate and a campaign adviser of the same campaign, and, as was the case in the pretests, the responses of the two campaigners were extremely similar. The data analyzed in Chapter 2 were for all major-party primary contestants from 1978 through 2002. Descriptive statistics, including sums, means, and percentages, provide the foundation for most of the data analysis in Chapters 2 through 8. The analyses in Chapters 3 through 9 are based on major-party contested general election campaigns, excluding incumbent-versus-incumbent House races because they have an atypical dynamic. Minor-party and independent candidates were omitted from the quantitative analysis. Following standard practice, all current officeholders were classified as incumbents, candidates who ran against them as challengers, and the remainder as open-seat candidates. This means that challengers who defeated incumbents during the nominating phase of the election were classified as challengers during their primary races and as open-seat candidates during the general election. Similarly, their general election opponents were also classified as open-seat candidates. Competitive campaigns include campaigns in which a challenger defeated an incumbent or the victor defeated the second-place finisher by 20 percent or less of the vote; the rest are classified as uncompetitive. This is a fairly large victory margin, but it is appropriate given the heightened level of uncertainty that surrounds contemporary congressional elections. A narrower competitiveness measure, such as a 15 percent margin, would have eliminated campaigns that were competitive for part of the election but were ultimately decided by more than 15 percent of the vote. The 20 percent measure enabled me to contrast the campaign organizations, budgets, strategies, and campaign contributions and services received by campaigns that were competitive at some point during the election with those that were never competitive. It corresponds to the measures used by many academics and political reporters. Slightly changing the boundaries for the competitiveness measure does not significantly change the results.[6] Tables 9-1 through 9-3 were created using ordinary least-squares (OLS) regressions. The full regression equations are presented in the notes in Chapter 9, and the analysis used to test the effects of campaign expenditures on television, radio, and other methods of campaign communications in challenger campaigns are presented in Tables A-1 and Chapter 9, note 27.The equations are the product of an extensive model-building process that tested the impact of numerous variables using a variety of statistical techniques. The variables that were tested in earlier versions of the models presented in Chapter 9 include the candidates' background characteristics and political experience, their party affiliations, the professionalism of their campaign organizations, and various combinations of variables representing different targeting approaches and issue stances. Numerous regressions were tested prior to selecting the final equations. The final models were selected for reasons of statistical fit, parsimony, and ease of interpretation. They are statistically robust; adding, subtracting, or transforming the variables does not lead to major changes in the results or the generalizations they support. The models provide support for generalizations that are similar to those developed in previous editions using the data collected in The 1992 Congressional Campaign Study and The 1998 Congressional Campaign Study. The models also were replicated to the extent possible, given the limited availability of campaign-related data, for other elections to verify that the basic relationships between the variables held across elections. Predictors of House Incumbents' Vote SharesThe estimates in Table 9-1 (and in Chapter 9, note 8) are for the impact of incumbent campaign activity on incumbents' vote shares while controlling for conditions that preceded the general election campaign (redistricting, the partisan bias of the district, the ideological strength of the incumbent, and incumbent scandal) and other factors. The information for the preexisting conditions and whether the incumbent received the most local media endorsements is drawn from questionnaires received from incumbent campaigns. The variable for Republican national partisan tide is a dummy variable for party identification that was created to capture the affect of the candidates’ party affiliations. The information for challengers’ and incumbents’ spending on campaign communications was compiled from data provided by Political Money Line using a computer program that coded campaign expenditures according to their use.[7] Incumbent spending on campaign communications was included in the regression analysis and failed to reach statistical significance. It was excluded from Table 9-1 to simplify the interpretation of the results for readers. The candidates' personal characteristics, party affiliation, and many of the campaign organization, strategy, and communications variables discussed in Chapters 2, 3, 7, and 8 were not included in the incumbent equation because they did not approach statistical significance. The findings for incumbent campaigns that reported that political parties, interest groups, or other outside groups tried to influence the campaign agenda in their election (discussed at the end of the section on incumbent campaigns and presented in Chapter 9, note 20) were generated using the identical equation that was used to generate the estimates presented for all incumbent campaigns. The findings differ from those for all incumbent candidates in that only those campaigns that responded that outside groups tried to set the campaign agenda to Question 5 of The 2002 Congressional Campaign Study were included in the analysis. The operationalization of the independent variables is as follows: District is very different or completely new: this information on redistricting was collected using question III.4 of The 2002 Congressional Campaign Study (see Figure A-1). Partisan bias: the percentage of registered voters of the incumbent's party minus the percentage of registered voters of the challenger's party. This information was collected in question III.1 of The 2002 Congressional Campaign Study. Incumbent’s ideological strength: this information was collected using question I.4 of The 2002 Congressional Campaign Study. The seven-point scale used to collect the information was folded and recoded so that moderate candidates are coded 0, the most liberal or conservative candidates are coded 3, and all other candidates are coded 1 or 2, depending on whether they responded “slightly liberal” or “slightly conservative” or “liberal” or “conservative.” Incumbent implicated in scandal: this information was collected using question VIII.1 of The 2002 Congressional Campaign Study. The information was recoded so that campaigns that stated that a scandal was moderately, very, or extremely important were coded 1 for scandal and 0 if otherwise. Incumbent spending on campaign communications (per $100,000): spending on television, radio, direct mail, campaign literature, newspaper advertisements, the Internet, field activities (mass telephone calls, get-out-the-vote drives, campaign literature, billboards and yard signs), and other communications. This figure was compiled from data provided by Political Money Line. Challenger spending on campaign communications (per $1,000): spending on television, radio, direct mail, campaign literature, newspaper advertisements, the Internet, field activities (mass telephone calls, get-out-the-vote drives, campaign literature, billboards and yard signs), and other communications. This figure was compiled from data provided by Political Money Line. Incumbent received most endorsements from local media: coded 1 if the incumbent's campaign responded to question I.8 of The 2002 Congressional Campaign Study that their candidate received the most endorsements and 0 if otherwise. Republican national partisan tide: Coded 1 for Republican candidates and 0 for Democrats. Predictors of House Challengers' Vote Shares and the Impact of Issues Agreement on the Ballots Cast for House ChallengersThe estimates in Table 9-2 (and in Chapter 9, note 22) are for the impact of incumbent campaign activity on challengers’ vote shares while controlling for conditions that preceded the general election campaign (the partisan bias of the district and whether the candidate’s primary was contested or not) and other factors. The information for the partisan bias of the district, voter targeting, the focus of the challenger’s campaign communications, and whether the challenger received the most local media endorsements is drawn from questionnaires received from challenger campaigns. The information on the primary (contested or not) is compiled from FEC data on all challenger campaigns. The information for challengers’ and incumbents’ spending on campaign communications was compiled using the same method as in the incumbent equation presented in Table 9-1.[8] The same procedure was used to generate the findings for challenger expenditures on individual campaign communication techniques, except spending for each communication activity was substituted for overall communications expenditures (see Table A-2) for the results. The same procedure was used to generate the findings for challenger campaigns that reported that political parties, interest groups, or other outside groups tried to influence the campaign agenda as was used in the analysis of incumbents. (This is discussed at the end of the section on challenger campaigns and presented in Chapter 9, note 25). As in the incumbent equation, many of the variables discussed in preceding chapters were not included in the challenger equation because they did not approach statistical significance or because of multicollinearity. The variables are operationalized in ways that are similar to those in the incumbent equations or have definitions that are self-evident, with the following exceptions: Contested primary: coded 1 if a challenger had to compete with one or more primary opponents for the party nomination and 0 if otherwise. Targeted party members, independents, or both: was coded 1 if a challenger campaign’s response to question III.2 of The 2002 Congressional Campaign Study campaign indicates a campaign targeted voters identified with one of these groups and 0 if otherwise. Advertising focused on challenger or incumbent’s issue positions: coded 1 if a challenger campaign’s response to question I.3 of The 2002 Congressional Campaign Study indicates the challenger focused primarily on the candidate’s issue positions, the major opponent’s issue positions, or both and 0 if otherwise. The estimates in Table 9-3 (and in Chapter 9, note 26) are for the impact of open-seat candidates and their opponents (also open-seat candidates) on the candidates’ vote shares while controlling for conditions that preceded the general election campaign (the partisan bias of the district) and other factors that influence an open-seat candidate’s vote shares. The variable for Democrats who ran on Democratic issues was included the regression analysis and failed to reach statistical significance. It was excluded from the table to simplify the interpretation of the results for readers. Again, many of the variables discussed in preceding chapters were not included in the open-seat equation because they did not approach statistical significance or because of multicollinearity. The findings for campaigns that reported that political parties, interest groups, or other outside groups tried to influence the campaign agenda in their election (discussed at the end of the section on incumbent campaigns and presented in Chapter 9, note 29) were generated using the identical equation that was used to generate the estimates presented for all open-seat campaigns. The operationalization of the independent variables is similar to that in the previous equations or have definitions t hat are self-evident, with the following exceptions: Republican ran on Republican issues: coded 1 if the campaign responded to question I.2 of The 2002 Congressional Campaign Studythat the most important issue in its campaign was the War on terrorism/situation with Iraq, taxes, other defense-related issues, or taxes and 0 if otherwise.[9] Democrat ran on Democratic issues: coded 1 if the campaign responded to question I.2 of The 2002 Congressional Campaign Studythat the most important issue in its campaign was the Social Security, the economy and jobs, education, health care/prescription drugs, the environment, or corporate fraud and 0 if otherwise.[10] Open-seat candidate spending on campaign communications: The natural log of candidate spending on campaign communications: this measure was used in lieu of the untransformed version of this variable in order to control for the diminishing effects of campaign spending on open-seat candidates’ vote shares. Opponent spending on campaign communications: this measure was used in lieu of the untransformed version of this variable in order to control for the diminishing effects of campaign spending on open-seat candidates’ vote shares. Senate CampaignsThe analytical techniques that were used to examine Senate campaigns were constrained by the small sample size and the other factors discussed in the book. The analysis for the Senate was conducted along the same general lines as that for the House and based on data collected in The 2002 Congressional Campaign Study and The 1992 Congressional Campaign Study, but bivariate tests were often substituted for the multivariate tests used for the House. Wherever possible, the statistical findings for the Senate were compared to those reported in other studies of Senate elections. Because of the limitations of the data, the generalizations developed for Senate campaigns were advanced tentatively and received less coverage than those for the House. Most of the discussion of Senate campaigns is comparative and is meant to highlight similarities and differences between campaigns for the upper and lower chambers of Congress. ------------------------------------------------------------------------------------------------------------------------------ Footnotes[1] The questionnaire is included with the materials for student research projects that appear elsewhere on this Web site. [2] See the case studies in Robert Biersack, Paul S. Herrnson, and Clyde Wilcox, eds., Risky Business? PAC Decisionmaking in Congressional Elections (Armonk, N.Y.: M.E. Sharpe, 1994) and Robert Biersack, Paul S. Herrnson, and Clyde Wilcox, After the Revolution: PACs, Lobbies, and the Republican Congress (Boston: Allyn and Bacon, 1999). [3] See the first, second, and third editions of Congressional Elections and Paul S. Herrnson, Party Campaigning in the 1980s (Cambridge, Mass.: Harvard University Press, 1988), Appendix. [4] The respondents to The 2002 Congressional Campaign Study were as follows: 52.7 percent candidates, 24.3 percent campaign managers, 5.0 percent communications aides, 18.0 percent other senior advisers, including field coordinators, fundraisers/treasurers, policy advisers, and candidate spouses. [5] See, for example, Robert Agranoff, The Management of Election Campaigns (Boston: Holbrook Press, 1972), 176-211; and Xandra Keydon, Campaign Organization (Lexington, Mass.: D.C. Heath, 1978), 147-148. [6] A variety of competitiveness measures were considered prior to selecting the measure based on the 20-pecent margin. Measures based on previous election margins were rejected because they depend in large part on the campaigns waged by candidates who ran in 2000 and because redistricting resulted in many congressional districts undergoing significant change following that election cycle. Measures based on the lists of competitive elections compiled by party organizations, lead PACs, and publications such as CQ Weekly, Roll Call, the Cook Political Report, the Rothenberg Political Report, and other political publications were rejected because of the differences among them and because they were based on a variety of criteria and could not be replicated. Moreover, lists that were compiled early in the election cycle--when candidates, parties, and PACs were making some important strategic and contribution decisions--classified more campaigns as competitive than lists that were published later. Nevertheless, it should be noted that the 20-point classification used here produces similar results to the forecasts of political journalists. When the categories for seats the Rothenberg Political Report assesses as leaning or tilting toward one party are combined into one category with toss-up races, its May 8, 2002 forecast classifies 88 percent of the races identically to the 20-point classification used here. Similarly, collapsing the Cook Political Report’s May 28, 2002 lean, likely, and toss-up races into one category also results in an 88 percent overlap with the races categorized as competitive using the 20-point classification. The Rothenberg Political Report’s July 29, 2002 and CQ Weekly’s June 4, 2002 forecasts overlap 87 percent and 85 percent, respectively, with the 20-point classification scheme. [7] Expenditures include only those reported by the candidate's principal campaign committee (and exclude leadership PACs and other affiliated organizations). Thirty-three of challenger campaigns and one of the open-seat campaigns included in the study did not report their expenditures electronically to the FEC. The itemized disbursements for the twenty challengers and one open-seat candidates were obtained by inspecting electronic images of their Schedule B disclosure forms. Another twenty challengers campaign spent less than $10,000 on their campaigns and did not itemize their disbursements. The communication expenditures for these candidates were approximated by multiplying the average percentage of expenditures made by similar candidates spent on communications (36 percent) by the candidates' total expenditures. There also were 76 challengers who reported no expenditures to the FEC because they raised less $5,000 and thus were exempt from the FEC’s disclosure requirements. Communication expenditures for these candidates were approximated at $500. [9] A few operationalizations of this variable were tested, including some that included the federal budget among the Republican issues. This did not change the results substantively. [10] A few operationalizations of this variable were tested, including some that excluded education from among the Democratic issues. This did not change the results substantively. |
|||
 |
|
|
|
|