The data for this study is drawn from the 2000, 2004, and 2008 National Annenberg Election Survey (NAES 2000; 2004; 2008). These studies are national, rolling cross-sectional surveys fielded in the lead-up to the presidential elections with post-election surveys conducted among a smaller sample of the original population. These surveys interviewed 58,373; 81,422
and 57,967 respondents respectively. The Annenberg is uniquely suited to this study because it combines questions about political behavior, policy preferences, and demographic information, as well as questions that allow for a comparison between private and public sector union members. Moreover, the large survey populations include a large enough sample of public and private sector union members to make meaningful comparisons.
The predictor variables in this study are dummy variables for whether a respondent belongs to a public sector or private sector union. These variables were generated using several questions from the Annenberg surveys. The 2000, 2004, and 2008 National Annenberg Election Survey each asked questions that allow us to, with caveats, identify private and public sector union members. Both the 2000 and 2004 Annenberg Surveys ask respondents if they belong to a union household (Question cw29; cWB06) and 2004 also distinguished between union members and union households. Both years also ask respondents if they are employed by the federal, state or local government (Question cw13; cWB05). Thus, a respondent who reports that they belong to a labor union and says they are employed by the government is coded as a public sector member, whereas a respondent who says they are a union member but do not work for the government is coded as a private sector union member.41 The 2008 Annenberg asked
respondents if they belong to a union (Question WB05) and, if so, which union they belong to (Restricted Question WB06). Respondents were then coded public or private sector based on their union. Although unions these days often represent both public and private sector members,
41
The remainder of this chapter will refer to union members rather than union households and all of the analysis was run using union members rather than union household data. The exception is the 2000 Annenberg that does not distinguish between union members and union households. Further discussion will refer only to union members, but any data referencing the 2000 Annenberg survey actually can only speak to union households.
each union was classified as public or private based on the sector of the majority of their members.42
The Annenberg surveys limit our ability to wholly identify public sector union members because union members may not recognize they work for the government if they work in a school for instance. Furthermore, the 2008 Annenberg coding may be mislabeling members. For instance, members of the United Auto Workers are coded as private sector but, to give an example, UC Berkeley Postdoctoral Researchers comprise UAW Local 5810 with over 6,000 members. No data is sufficient and the findings of this paper should be seen in light of the limitations of identifying public and private sector union members and households. However, all of the issues inherent in the Annenberg surveys are likely to under-identify rather than over- identify public sector union members by leaving some public sector union members grouped with private sector union members: in the 2000 and 2004 data, public sector union members may not recognize they work for the local, state and federal government, and in the 2008 data, a small percentage of public sector union members are members of private sector unions that cannot be singled out. Thus, any statistically significant differences found in the preferences and behavior of public and private sector union members in this study are likely underestimated.
There are two key sets of dependent variables in this study: (1) self-reported turnout in the 2000-2008 presidential primary and general elections, and (2) public opinion variables that are detailed in Appendix B. A series of control variables are also included in the models. These variables are drawn from prior research that has found that they are likely to influence primary and general election turnout including dummies for gender, black, employed and self-reported measures of age, education, household income, and an individual’s strength of partisanship. Individuals have also been found to be less likely to turn out in primaries due to lower
registration rates among those who have recently moved so the models also control for how long a person has lived in their current residence (Highton 2000; Squire et al. 1987). Lastly, a set of controls is included to account for election specific influences on turnout. Given that the presence of statewide races may increase turnout, a dummy variable is included for whether there was a state governor race that year in the voter’s state. The state's vote margin—the percentage of a state‘s electorate voting Republican subtracted from the percentage voting Democratic in the election is included in order to control for party competitiveness, because closer races likely promote higher turnout. The final control is the average turnout percentage over the last three presidential elections of each state's voting eligible population (VEP), because, due to unique state characteristics, turnout can vary by state independently of all of these
individual and election specific characteristics.43
This analysis utilizes independent group t-tests that compare the difference of means of public and private sector union members on specific variables. These simple t-tests demonstrate whether, on any specific variable, the mean value of public sector union members is statistically different from the mean value of private sector union members. This does not tell us what is driving the difference between the two groups. Being a member of a public or private sector union as well as demographic differences—gender, education and income—are likely just as important, as are unknown variables. However, the value in using t-tests is to establish for the first time whether there are in fact statistically significant differences between public and private sector union members. Binary logit, OLS regressions, and coarsened exact matching is then used to determine what mechanisms—union membership, education, income and/or gender—are leading to different preferences and behaviors among public and private sector union members.
43 State level dummies were included with no effect on the results and thus are not displayed in these models.