Direct Member Mobilization

The first mechanism is direct mobilization via membership effects. Because organizations can mobilize their members through both the top-down and the bottom-up (Ahlquist and Levi Reference Ahlquist and Levi2013; Leighley Reference Leighley1996), it is useful to discuss both the incentives and roles of union leaders and benefits of members from the organizational structure of unions. First, because union leaders have strong incentives to achieve organizational goals, they need to directly contact their members and encourage them to take action (Rosenstone and Hansen Reference Rosenstone and Hansen2003). The traditional literature on group organization explains that group participation provides three different types of benefits: material, solidarity, and purposive benefits (Clark and Wilson Reference Clark and Wilson1961; Salisbury Reference Salisbury1969). While material benefits refer to personal and tangible rewards, solidarity and purposive benefits are intangible and suprapersonal. As union leaders are responsible for expanding their organizations and increasing policy benefits such as better rights and wages, they have strong incentives to directly mobilize their members for the latter benefits. It is indeed known that the presence of union leaders who are motivated to achieve their political objectives leads to more sustained and effective member mobilizations (Ahlquist and Levi Reference Ahlquist and Levi2013).

To raise solidarity benefits, leaders can hold frequent meetings and sponsor other types of gatherings where members share group membership and group identity. With the strengthened solidarity, they can in turn directly contact their members via mailings, phone calls, or mobilization drives such as door-to-door canvasses to achieve purposive benefits in electoral and/or policy-making process (Rosenstone and Hansen Reference Rosenstone and Hansen2003). They also can make a public endorsement of their preferred candidates. For example, the Retail, Wholesale and Department Store Union (RWDSU), which represents about 10,000 workers in New Jersey, publicly endorsed the state Governor, Phil Murphy, to encourage their members to go to the polls and vote for him.Footnote ¹ In short, union leaders mobilize their base to achieve policy goals that they pursue (purposive benefits)—not only monetary benefits of unions, but also sometimes social and political objectives (Ahlquist and Levi Reference Ahlquist and Levi2013)—by promoting group membership and identification among the rank and file (solidarity benefits) in elections.

Second, as the bottom-up process of the direct mechanism, union rank and file can enjoy benefits from the organizational structure. Labor unions, as typically federated organizations, can allow their members to contact each other and to share information and other resources across space and even time periods (Hertel-Fernandez Reference Hertel-Fernandez2018; Kim and Margarlit Reference Kim and Margarlit2017; Mutz and Mondak Reference Mutz and Mondak2006). Though various types of organizations including not only labor unions but also churches and voluntary associations provide members with a context for political interactions (Verba, Schlozman, and Brady Reference Verba, Schlozman and Brady1995), it is known that the organizational structure of labor unions is more likely to facilitate political discussions than other organizations because they encourage the discussion on issues related to their workplace via meetings, workshops, conventions, and other gatherings as noted earlier (Ahlquist and Levi Reference Ahlquist and Levi2013; Macdonald Reference Macdonald2019).

Indeed, unions can encourage rank-and-file members to devote their attentions to obtaining policy-related information (Ahlquist and Levi Reference Ahlquist and Levi2013). Also, because union membership is helpful to increase of political knowledge (Iversen and Soskice Reference Iversen and Soskice2015), union rank and file can reduce costs of voting and, accordingly, go to the polls more easily. Furthermore, given that the informing process can compensate for the lack of political resources, union mobilization via disseminating information and campaign activities is expected to be even more beneficial to electoral participation of the rank-and-file members, who usually lack resources otherwise (Kim Reference Kim2016; Lamare Reference Lamare2010; Leighley and Nagler Reference Leighley and Nagler2007; Rosenfeld Reference Rosenfeld2014; Verba, Schlozman, and Brady Reference Verba, Schlozman and Brady1995).

In a similar vein, labor unions can promote civic skills, social capital, and political efficacy. Union members can experience various intragroup activities such as electing leaders, making and following rules, and taking roles in the unions (Radcliff Reference Radcliff2001; Zullo Reference Zullo2004), which raise psychological resources such as civic skills for electoral participation. In these activities, members should interact with their colleagues and, accordingly, they can increase social capital as well (Putnam Reference Putnam1993). Likewise, labor unions affect members’ political efficacy (Campbell et al. Reference Campbell, Converse, Miller and Stokes1960). On the one hand, as the above mentioned intragroup activities allow members to experience democratic process, they can understand being a democratic citizen and its roles and thus increase internal efficacy. On the other hand, it is well known that unions have strong policy incentives, and also possess abundant political resources such as historical alliance with the Democratic Party (Frymer and Grumbach Reference Frymer and Grumbach2021; Rosenfeld Reference Rosenfeld2014). Thus, when they are successful in achieving their policy goals via lobbying activities in policy-making process (Baumgartner et al. Reference Baumgartner, Berry, Hojnacki, Kimball and Leech2009; Denzau and Munger Reference Denzau and Munger1986; Gordon Reference Gordon2001; Hall and Wayman Reference Hall and Wayman1990), it can directly enhance external political efficacy of the members to actively participate in political actions (Ahlquist and Levi Reference Ahlquist and Levi2013).

Indirect Nonmember Mobilization

In addition to the direct effects, there is another way that labor unions have an impact on voter turnout indirectly. Because organizations are a linkage between the personal world and the public world, the mobilization efforts of labor unions can extend far beyond their members via social networks. That is, “leaders need not communicate with every person directly. Instead, leaders contact their associates, associates contact their colleagues, colleagues contact their friends, families, and coworkers” (Rosenstone and Hansen Reference Rosenstone and Hansen2003, 27). Because union leaders seek to maximize their political and economic interests in policy-making process, they can enjoy reduced costs of mobilization by strategically engaging in nonmember mobilization using membership in social networks (Lamare Reference Lamare2010; Leighley and Nagler Reference Leighley and Nagler2007; Lyon and Schaffner Reference Lyon and Schaffner2020; Radcliff Reference Radcliff2001).

In particular, because family members are very interactive each other in our everyday lives, it is easy to share political and economic interests and deliver relevant information and sources in household (Sinclair Reference Sinclair2012). Traditional theories of political attitudes also explain that individuals begin to shape their political preferences in their household, and political resources for participation are transmitted through family members (Schlozman, Brady, and Verba Reference Schlozman, Brady and Verba2018). Thus, it is expected that, when a union member is aware of the roles of unions and what to do as a member in the coming election, his or her family members are more likely to learn about their roles and the issues in the election. As a result, they are more likely to participate in elections as a result of social networks (e.g., Lyon and Schaffner Reference Lyon and Schaffner2020).

In a similar vein, whenever union members need human and financial supports for their activities, they are likely to mobilize their families and close friends because they are easily contactable. Indeed, it is well known that organizing nonmembers such as families and friends has become a common strategy adopted by union-affiliated groups such as AFL-CIO’s Working America.

Indirect Aggregate Strength

As another indirect mechanism, labor unions can exert influence of indirect aggregate mobilization. This mechanism is based on their aggregate strength in the states. It is known that when unions are stronger and more pervasive in a given state, the state is likely to have less business-friendly policies (Witko and Newmark Reference Witko and Newmark2005), more liberal public policies (Radcliff and Saiz Reference Radcliff and Saiz1998), more equal income distribution (Bucci Reference Bucci2018; Rosenfeld Reference Rosenfeld2014; Volscho and Kelly Reference Volscho and Kelly2012), more equal political representation (Flavin Reference Flavin2018; Rosenfeld Reference Rosenfeld2014), and less poverty (Brady, Baker, and Finnigan Reference Brady, Baker and Finnigan2013). These findings suggest that labor unions may also have indirect state-level effects on political participation (Leighley and Nagler Reference Leighley and Nagler2007; Lyon and Schaffner Reference Lyon and Schaffner2020).

Specifically, the indirect aggregate strength of unions may arise through two ways. First, because unions pursue policy benefits, where labor unions are widespread, union-friendly issues are more likely to be salient and accordingly candidates and parties are likely to have stronger incentives to be responsive to them. Thus, during election campaigns, these issues are likely to attract not only union members but also nonunion members who share similar interests in states (Leighley and Nagler Reference Leighley and Nagler2007; Rosenstone and Hansen Reference Rosenstone and Hansen2003). In this sense, nonmembers can be mobilized as they can easily obtain relevant information about the issues and become interested in the elections (Kim Reference Kim2016; Lyon and Schaffner Reference Lyon and Schaffner2020; Macdonald Reference Macdonald2019).

Second, when labor unions are stronger and more pervasive, it is also expected that electoral contests become more competitive. As labor unions emphasize pro-labor issues and promote pro-union candidates in elections, opposing candidates and parties who support business-friendly policies are likely to raise anti-labor issues and mobilize voters who have political and economic interests against labor unions. In other words, as labor unions not only emphasize pro-labor issues but also provide strong signals to individuals with anti-labor interests, they can help voters to recognize big differences between candidates or parties. Thus, the big differences in policies and candidate/party positions lead to higher voter turnout of both union members and nonmembers (Leighley and Nagler Reference Leighley and Nagler2014; Rosenstone and Hansen Reference Rosenstone and Hansen2003).

Based on the discussions above, scholars have provided empirical evidence that labor unions are positively associated with voter turnout rates (Leighley and Nagler Reference Leighley and Nagler2007; Radcliff Reference Radcliff2001; Radcliff and Davis Reference Radcliff and Davis2000). However, they have several analytical limitations. First, because Radcliff (Reference Radcliff2001) and Radcliff and Davis (Reference Radcliff and Davis2000) do not consider state-level effects in their individual-level analyses, their findings suffer from omitted variable bias. As Leighley and Nagler (Reference Leighley and Nagler2007) adequately point out, union strength can be correlated with unmeasured attributes of each state and, thus, the state or local level analysis may be more appropriate to examine the impact of labor unions on voter turnout. Moreover, given the fact that unions and their relationship with governments are different across industries, countries, and time periods (Ahlquist Reference Ahlquist2017), the American states, which embody an appropriate condition for “the controlled experiment” (Gray, Hanson, and Kousser Reference Gray, Hanson and Kousser2018, xi), would also provide analytical benefits to examine the relationship between unions and turnout.

Second, however, though Leighely and Nagler (Reference Leighley and Nagler2007) adopt more appropriate empirical models and provide more robust findings using the state-level data, they also suffer from misspecification. Specifically, they only consider union membership at the individual-level and union density at the state-level by just assuming that the latter “encompasses the nonmembership effects” (433). But, according to the theories above, there are three different mechanisms through which labor unions affect voter turnout: direct member mobilization (the individual-level), indirect nonmember mobilization (the individual-level), and indirect aggregate strength (the state-level). Previous studies do not carefully consider all of the three mechanisms, but examine only two of them (usually, the first and the third mechanisms). Therefore, I examine the three theoretically relevant mechanisms in my empirical models to provide more robust findings of the relationship between labor unions and voter turnout.

Before describing my data and model specification at both the state-level and the individual-level, I briefly discuss political ideologies of labor unions and how those ideologies have evolved in recent years. It is well known that the organized labor traditionally has close organizational ties with the Democratic Party (Frymer and Grumbach Reference Frymer and Grumbach2021; Greenstone Reference Greenstone1977; Rosenfeld Reference Rosenfeld2014). However, because labor unions have sometimes failed to successfully achieve their legislative goals, some describe the ties as one-sided partnership (Francia Reference Francia2010). Moreover, due to the election of Donald Trump, labor unions—especially, white working class—may be regarded as pro-Republicans at least in some regions (Devinatz Reference Devinatz2017; Francia Reference Francia2020).Footnote ² However, despite the recent dynamic evolution of labor unions’ political ideologies, I do not expect my analyses suffer from severe biases. First, theoretically, because I examine voter turnout, not vote choice, their political ideologies would not bias my analyses. Second, empirically, because I include the 50 states and multiple years in the analyses, I can control for state- and/or year-specific effects.

State-Level Analysis

For the state-level analysis, I collect data on voter turnout in all of the 50 states from 1980 to 2008.Footnote ³ The choice of time period is data-driven rather than theory-based. This yields a dataset of 750 elections from 50 states, with 15 elections in each state. My dependent variable is VEP turnout. Though there is a debate upon appropriate measurement for voter turnout, VAP turnout may lead to downward bias because it includes varying alien populations who cannot vote (Blais and Dobrzynska Reference Blais and Dobrzynska1998). In contrast, the VEP turnout purely calculates the percentage of eligible voters by excluding noneligible populations such as felons and noncitizens. Thus, it is regarded as the more accurate measure of turnout rates (Macdonald Reference Macdonald2021; McDonald and Popkin Reference McDonald and Popkin2001). Therefore, I use the VEP turnout rates in the models below.Footnote ⁴ I obtain the data from the website of United States Elections Project (McDonald Reference McDonald2020).

Also, it should be noted that as the time period in the study is from 1980 to 2008, it includes both presidential elections and midterm elections. Previous studies suggest that presidential and midterm elections should be treated separately (Grummel Reference Grummel2008; Jackson Reference Jackson1997; Smith Reference Smith2001; Tolbert and Smith Reference Tolbert and Smith2005). This is because factors that affect voter turnout are different in presidential and midterm elections (Rosenstone and Hansen Reference Rosenstone and Hansen2003). That is, as presidential races already provide voters with a strong incentive to vote, turnout in presidential elections is higher than in midterm elections. Even though adding mobilization factors driving voter turnout should increase benefits of voting, the pattern of diminishing returns appears in presidential elections. That is, “the larger the benefits already gained from voting, the weaker the stimulus to turnout” (Smith Reference Smith2001, 701). Therefore, I provide three separate models: midterm election, presidential election, and pooled models. I expect that the variable of interest, union mobilization, will be weakly or not significant in pooled models and presidential election models, while it will be significant in midterm election models.Footnote ⁵

My independent variable is union mobilization. For its measurement, I draw data on union membership density from the website of Union Membership and Coverage Database from the CPS (Hirsch and Macpherson Reference Hirsch and Macpherson2020).Footnote ⁶ Additionally, I include controls that have been found to affect voter turnout in American contexts. First, I include institutional and political variables in models. Following Grummel (Reference Grummel2008) and Tolbert, Grummel, and Smith (Reference Tolbert, Grummel and Smith2001), I measure registration requirements as the number of days before the election a voter can register to vote in each state over time. For example, I coded 0 if states have the Election Day registration, and coded 30 if states require voter registration a month before the election.

I also control for electoral competition in states. It is expected that highly competitive elections incentivize voters to go to the polls because they can expect their votes to be more decisive and valuable (Engstrom Reference Engstrom2012; Jackman Reference Jackman1987; Powell Reference Powell1986). Holbrook and Van Dunk (Reference Holbrook and Van Dunk1993) offered the measure of electoral competition based on district-level state legislative elections in 1980s, and Shufeldt and Flavin (Reference Shufeldt and Flavin2012) updated the data based on elections since 1990s. Thus, I obtain the data from both Holbrook and Van Dunk (Reference Holbrook and Van Dunk1993) and Shufeldt and Flavin (Reference Shufeldt and Flavin2012) for time period before and after 1990, respectively. It ranges from 0 to 100, and higher values mean greater competition. As direct democracy influences voter turnout, I control for effect of initiative and referendum. I obtain the information on initiative and referendum from Initiative and Referendum Institution at the University of Southern California. States with statewide initiative or referendum are coded 1, and otherwise 0. Because professionalized state legislatives are likely to address higher political stakes which drive voter turnout, I control for professionalization score developed and measured by Squire’s (Reference Squire2017) index. It ranges from 0 to 1.

To control for economic and sociodemographic factors, I first include economic condition of states, which is measured by Gross State Product (GSP) growth rate from Bureau of Economic Analysis. Second, to account for education, the proportion of the population with a high school degree or higher in each state is included, which is from U.S. Department of Commerce. Third, the proportion of population residing in urban areas and median age of population in each state are included (Grummel Reference Grummel2008; Jackson Reference Jackson1997). Fourth, because race is another important factor (Hill and Leighley Reference Hill and Leighley1999; Leighley and Nagler Reference Leighley and Nagler2014), I control for minority diversity index, which has been used by prior studies (Grummel Reference Grummel2008; Tolbert and Hero Reference Tolbert and Hero1996). Lastly, the models contain the following dummy variables: southern state, presidential election, senate election, gubernatorial election. The summary statistics of all variables in the analysis is presented in Supplementary Appendix A.

Because I use time-series cross-sectional data, which covers 50 states from 1980 to 2008, ordinary least squares (OLS) regression is not appropriate due to the data structure. This is because error terms in regression equations cannot be independent, which violates an OLS regression assumption and causes autocorrelations. Moreover, since the data include more panels (N = 50) than time points (T = 15), statistical issues of heteroskedasticity and serial correlation across panels should be addressed. Even though estimated coefficients may not be biased under OLS regressions, the analyses will suffer from inefficient and biased standard errors. In order to address the issues, I utilize panel corrected standard errors (PCSE) as Beck and Katz (Reference Beck and Katz1995) recommend. At the same time, I also include a lagged dependent variable (LDV) to allow for temporal dynamics with AR(1) process. In particular, the inclusion of LDV is beneficial both theoretically and methodologically. In a theoretical sense, variation of voter turnout is a dynamic process such that the past turnout in a state is related to the current turnout in the state. In addition, methodologically, the inclusion of LDV can make models avoid omitted variable bias and mitigate problems related to unit heterogeneity. Indeed, when specifying dynamic models for dynamic theories, it is more appropriate to include LDV in models (De and Keele Reference De Boef and Keele2008; Keele and Kelly Reference Keele and Kelly2006). However, in addition to this specification, I also report results from two-way fixed effects (2FE) models with robust standard errors clustered by state. Because it is expected that the 2FE models make the standard errors a little bit larger and remove all observed and unobserved time-invariant differences across states, they can be a much stronger specification.Footnote ⁷

State-Level Findings

Table 1 shows empirical findings from the three separate state-level models (midterm election model, presidential election model, and pooled model) as well as from the 2FE models.

Table 1. State-level analysis: Labor union and voter turnout (VEP)

Note. Nonstandardized coefficients and panel corrected standard errors are presented in Model 1–3. Nonstandardized coefficients and robust standard errors clustered by state from two-way fixed effects models are reported in Model 4. Significance level: **p < 0.01 and *p < 0.05 (two-tailed).

Model 1 depicts the relationship between union mobilization and voter turnout in midterm elections. Because voters tend to have lower information and interests in midterm elections due to narrower scope of conflicts compared to presidential elections (Grummel Reference Grummel2008; Schattschneider Reference Schattschneider1960; Tolbert and Smith Reference Tolbert and Smith2005), it is expected to observe a statistically significant and positive coefficient estimate on the union membership density variable. As expected, the coefficient is in the expected direction and statistically significant. This clearly reconfirms that, while controlling for other variables, labor union membership density is positively associated with voter turnout (Leighley and Nagler Reference Leighley and Nagler2007; Radcliff Reference Radcliff2001).

Substantively, this effect indicates that if the union membership density is 32.1% (e.g., highest; HI in 1982) compared to 3.3% (e.g., lowest in the data; NC in 2006 and SC in 2006), the voter turnout increases by about 7% point in midterm elections given the previous level. Also, if the density is 21.3% compared to 7.7% (one standard deviation above and below the mean value, 14.5%, in midterm elections), the voter turnout is expected to increase by about 3.3% point given the previous level.

While there is a significantly and substantively significant impact of the labor union membership density in midterm elections, labor unions are not related to voter turnout rates in presidential elections. Model 2 displays the association between union mobilization and voter turnout in presidential elections. Following the discussion above, it is expected that the mobilizing force of unions may be insignificant due to the diminishing returns, which suggest “the larger the benefits already gained from voting, the weaker the stimulus to turnout” (Smith Reference Smith2001, 701). Indeed, the coefficient of the union membership density is not statistically significant in Model 2. The similar pattern is observed in Model 3, where both midterm and presidential elections are pooled. As the statistically and substantively significant coefficient of the presidential election dummy variable shows, union mobilization does not have additional driving forces for electoral participation when voters already have strong incentives to participate in presidential elections, which have higher stakes of interest through broader scope of conflicts (Schattschneider Reference Schattschneider1960).

In order to visually display the findings in Table 1, I pool both midterm and presidential elections in one model by creating an interaction term between union membership density and presidential election dummy variable. The models are reported in Supplementary Appendix D, from which I draw Figure 1. The vertical axis in the figure indicates the predicted values of turnout measured by VEP (left panel) and VAP (right panel) across different levels of union membership density on the horizontal axis.Footnote ⁸ I again confirm the patterns of the relationship that are different between midterm and presidential elections. Both VEP and VAP increase as union membership density increases in midterm elections, whereas the pattern is not observed in presidential elections. Thus, it reveals that the mobilizing force of labor unions is limited to the former.

Figure 1. Labor union and voter turnout (VEP and VAP).

Note. Predicted values of voter turnout (solid lines) are from pooled interaction models in Supplementary Appendix D. Dashed lines indicate the 90% confidence intervals.

When it comes to the LDV and control variables in Table 1, there is little difference between midterm and presidential elections. Previous turnout, the LDV, is statistically significant in the expected positive direction, which shows its temporal dependence. Voter registration requirements have significant and negative coefficients, suggesting that, as previous studies demonstrate, voter turnout becomes higher as a state has generous registration rules such as the Election Day registration (Alvarez and Ansolabehere Reference Alvarez and Ansolabehere2002; Leighley and Nagler Reference Leighley and Nagler2014). Electoral competition and direct democracy are also statistically significant and positive as expected. That is, higher level of electoral competition, especially in the district-level, leads to higher voter turnout in states. Likewise, states with statewide initiative or referendum are more likely to have higher turnout. Interestingly, though legislative professionalism is statistically significant, it is negative in midterm elections but positive in presidential elections. Due to the inconsistent directions of the impact, it is not significant in the pooled models.

Regarding economic and sociodemographic variables, GSP growth rate has significant and negative coefficients in the three models. This supports the negative relationship between economic conditions and voter turnout (Kern, Marien, and Hooghe Reference Kern, Marien and Hooghe2015). In other words, American voters are more likely to go to the polls in economic hardships. While education and age are not significant, urbanization and minority diversity have statistical significance. But, the former has a negative and significant coefficient only in midterm elections, whereas the latter does only in presidential elections.

In additional analysis with alternative specifications, I checked the robustness of the findings by using the 2FE models, operationalizing the dependent variable using the VAP instead of VEP, and dropping the LDV and instead including state dummy variables. None of these alternative specifications, as reported in Model 4 and Supplementary Appendices B and C, produced qualitatively different findings.

Individual-Level Analysis

As previous studies have demonstrated (Kim Reference Kim2016; Lamare Reference Lamare2010; Leighley and Nagler Reference Leighley and Nagler2007; Radcliff Reference Radcliff2001), my state-level analysis in the previous section clearly shows that labor unions are positively associated with voter turnout. However, it reveals that the impact of labor unions is observed only in midterm elections, but not in presidential elections, because of the pattern of diminishing returns (Smith Reference Smith2001). In this section, more importantly, I further examine whether the impact is direct via member mobilization or indirect via nonmember mobilization. As noted earlier, though previous studies find a positive relationship between labor unions and political participation, they do not carefully examine whether the relationship comes primarily from direct member mobilization (the individual-level), indirect nonmember mobilization (the individual-level), or indirect aggregate strength (the state-level).

To address the existing analytical limitations, I test the individual-level mechanisms using the cumulative CCES data (2006–2019) (Kuriwaki Reference Kuriwaki2020).Footnote ⁹ Despite its online survey format, the primary reason that I use the CCES, rather than the ANES, is that union capacity to mobilize voters relies more on the state or local context than on the national context (Leighley and Nagler Reference Leighley and Nagler2007). Thus, I expect the data to provide analytical benefits with more respondents from each state who are surveyed around November of each year.

My dependent variable in the individual-level analysis is voter turnout. To measure respondents’ participation in the elections, pollsters directly asked whether they participated in the elections or not. If a respondent participated, I score him or her 1 on this measure, and 0 if otherwise. To examine different mechanisms of the impact that labor unions have on voter turnout, three measures are employed as independent variables: (a) whether a respondent is currently a member of a union (direct member mobilization); (b) whether any member of a respondent’s household is currently a member of a union (indirect nonmember mobilization)Footnote ¹⁰; and (c) state union membership density (indirect aggregate mobilization). The first and second individual-level variables are measured as binary variables, while the third state-level variable again comes from Union Membership and Coverage Database from the CPS as (Hirsch and Macpherson Reference Hirsch and Macpherson2020).

Apart from the independent variables, political and sociodemographic covariates are included in the analysis: political ideology (1 = “very liberal”; 3 = “very conservative”); partisan strength (1 = “independent”; 3 = “strong partisan”); income (1 = “first quintile”; 5 = “fifth quintile”); education (1 = “≤high school graduation”; 2 = “two-year college”; 3 = “four-year university”; 4 = “graduate”); gender (1 = “female”; 0 = “male”); age and its squared term; race (White, Black, Hispanic, and others); marital status (1 = “married”; 0 = “not married”). I also include the presidential election dummy variable. The descriptive statistics of all variables in the analysis is presented in Supplementary Appendix E.

Because the dependent variable is a binary variable, binary logistic regression models should be utilized. And, the year fixed effects should be considered to control for the influence of unique variables of each year that might affect the dependent variable though not included in the models. However, more importantly, I try to examine the effect of labor unions operationalized by three different measures. Yet, the state union membership density does not vary across individuals in a given state. This implies that there are two nested levels in the data: individuals (level 1) and states (level 2). If the empirical models fail to recognize the multilevel data structure, the results suffer from severe bias due to the violation of the assumption of independent errors (Raudenbush and Bryk Reference Raudenbush and Bryk2002; Steenbergen and Jones Reference Steenbergen and Jones2002). To avoid the bias, I estimate hierarchical binary logistic regression models. In addition to the multilevel approach, I here again report results from 2FE models with robust standard errors clustered by state.

Individual-Level Findings

Is the impact of labor unions direct via member mobilization or indirect via nonmember mobilization? Table 2 presents the results from the multilevel models. Model 5 and Model 6 examine whether being a union member or having family members who belong to the unions is more likely to participate in elections, respectively. As shown in the models, the empirical findings provide the evidence that both have mobilizing effects. The coefficients of the variables of interest are statistically significant in positive directions while controlling for the state-level union density, which also has statistical significance in a positive direction as expected (indirect aggregate strength).

Table 2. Individual-level analysis: Direct and indirect mobilization (CCES)

Note. Coefficients and standard errors from multilevel logistic regression models are presented in Model 5–7. Coefficients and robust standard errors clustered by state from binary logistic regression models with two-way fixed effects are reported in Model 8. Significance level: **p < 0.01 and *p < 0.05 (two-tailed).

Substantively, according to Model 5, when an individual is a member of labor unions, the probability of his or her voting increases by about 4.7%. Likewise, in Model 6, when having family members who belong to labor unions, the predicted probability of voting increases by about 6.6%. Therefore, it seems that the impact of labor unions on political participation comes from both direct member mobilization (the individual-level) and indirect nonmember mobilization (the individual-level).

However, the third column of Table 2, where the two variables of interest are included at the same time in one model, shows interesting results. In Model 7, while having family members who belong to labor unions remains still significant in a positive direction, being a labor union member does not hold statistical significance anymore. The state-level union density is still controlled for in the model and, it has a positive and significant coefficient. In other words, when controlling for the indirect nonmember mobilization and the indirect aggregate strength of labor unions, the direct member mobilization does not drive voter turnout.

The results contradict the findings in previous studies that both the direct member mobilization and the state-level union membership density boost electoral participation (Kim Reference Kim2016; Leighley and Nagler Reference Leighley and Nagler2007; Radcliff Reference Radcliff2001). However, while they cannot directly compare the impacts of member and nonmember mobilization because of the omission of the indirect nonmember mobilization in their empirical models, the results in Model 7 of Table 2 enable the comparison. And it clearly shows that being a member of labor unions is not associated with turnout when indirect nonmember mobilization is appropriately controlled for. Instead, having family members who belong to labor unions is positively related to voter turnout. That is, substantively, when at least one of family members is a union member, the probability of voting increases by about 5.8%. This suggests that labor unions’ positive impact does not come from the direct member mobilization, but from the indirect nonmember mobilization and the indirect aggregate strength of labor unions. Put differently, the primary reason that declining union membership reduces the voter turnout rates in the US is not direct mobilization, but indirect mobilization.Footnote ¹¹ And I obtain similar results from Model 8 which includes state and year fixed effects (2FE) with robust standard errors clustered by state.

The control variables in the models of Table 2 have significant coefficients, and they are not qualitatively different across the models. As individuals have liberal ideologies, they are more likely to participate in elections. As expected, partisan strength, income, education, and marriage are positively associated with voter turnout and, racial minorities such as blacks and Latinos are less likely to participate than whites. Also, female is less likely to go to the polls compared to men, and both age and its squared term are statistically significant in the expected direction. As the state-level findings show in the previous section, both presidential election and state-level union membership density have a positive association with voter turnout.