Labor Unions and White Democratic Partisanship

The Supplementary Appendix and all data and Stata code to replicate the main findings are available at the Political Behavior Dataverse (https://​dataverse.​harvard.​edu/​dataverse/​polbehavior).

Union membership has declined dramatically in the private sector while it has remained largely stable in the public sector (http://​unionstats.​com/​). Even though public sector workers make up a large share of union members (nearly half as of 2018, according to CPS data), far more Americans work in the private sector (130 million) than the public sector (22 million), according to data from the Bureau of Labor Statistics (BLS) (https://​www.​bls.​gov/​news.​release/​empsit.​t17.​htm). As such, private sector union decline is especially consequential and has certainly had a large impact on declining overall union membership.

Not all labor unions are staunch supporters of the Democratic Party, nor is their support constant across elections and candidates (https://​news.​bloomberglaw.​com/​daily-labor-report/​gop-candidates-labor-unions-make-strange-bedfellows). For instance, the Teamsters endorsed Richard Nixon in 1972 (https://​www.​nytimes.​com/​1972/​07/​18/​archives/​meany-stand-on-mcgovern-spreads-labor-dissension.​html), while Ronald Reagan and Donald Trump both received a sizable proportion of the union vote (https://​www.​washingtonpost.​com/​news/​the-fix/​wp/​2016/​11/​10/​donald-trump-got-reagan-like-support-from-union-households/​). Although certain labor unions may occasionally endorse Republicans, organized labor is, in general, among the strongest organizational supporters of the Democratic Party (https://​www.​opensecrets.​org/​industries/​indus.​php?​Ind=​P).

See Ahlquist (2017) and Rosenfeld (2019) for fuller reviews.

See Aldrich et al. (2018, Chap. 5) and Roper exit poll data (https://​ropercenter.​cornell.​edu/​data-highlights/​elections-and-presidents/​how-groups-voted) for greater detail.

Union membership peaked in 1953 at nearly 35% (Goldfield and Bromsen 2013). This was driven primarily by the private sector (with membership at nearly 43%) as public sector unionization was very limited at the time (Anzia and Moe 2016; Flavin and Hartney 2015). Today, according to Current Population Survey (CPS) data, the public sector makes up a near majority of union members, while just 6.4% of private sector workers belonged to a labor union (http://​unionstats.​com/​). Private sector union membership has also declined among both whites and Blacks (Rosenfeld and Kleykamp 2012).

See also for example, the AFL–CIO’s website highlighting the actions (broad and specific) that labor unions take to better working conditions for their members and for the broader working/middle classes (https://​aflcio.​org/​what-unions-do).

It also seems likely that labor unions’ efforts to inform provision efforts will also reach, albeit to a more limited degree, non union members who live in union households. One way that this could occur is via political discussions that occur in labor union households. For example, data from the Cumulative ANES shows that just 22% of Americans report that they never discuss politics with family and friends. This could also occur because these individuals (non union members in a union household) are likely somewhat reliant upon labor unions and the benefits they provide the household, i.e., for some portion of their economic well-being. Data from the Cumulative ANES seems to support this. Union household members (who are not in a union themselves) give labor unions a mean rating of 63 out of 97 (on the feeling thermometer scale), compared to a mean rating of 51 for people who are not at all union-affiliated.

I look at two dependent variables: (1) the canonical seven-point scale, and (2) a difference in feeling thermometer ratings between the Democratic and Republican Parties. I do this because partisanship is conceptualized as consisting of both identification and affect (Campbell et al. 1960).

This demographics-only model may be subject to criticism as it does not include controls for issue attitudes and retrospective evaluations. I am confident, however, that this relationship is robust to their inclusion, given the findings of past research. For example, Hajnal and Rivera (2014, Table 1) show that union membership is significantly associated with white Democratic party identification, using data from the 2008 ANES that includes controls for demographics, ideology, issue positions, retrospective economic evaluations, presidential approval, and racial and immigration attitudes. Zingher (2018, Table 2) uses ANES data over several decades, showing that white union members are significantly more likely to identify as and vote Democratic when controlling for demographics, and economic/social policy attitudes. Though labor unions are not either paper’s main explanatory variable, I rely on these two papers’ findings (Hajnal and Rivera 2014; Zingher 2018) to show that the relationship between union affiliation and white Democratic partisanship is robust to controls beyond the demographics included here.

It is important to note that the results here display the “average” influence of union affiliation on white partisanship. It is likely that the relationship is stronger for people who feel closer to/identify more strongly with labor unions (Campbell et al. 1960, Chap. 12; Lewis-Beck et al. 2008, Chap. 11).

These are pooled data (across five decades) and thus may mask the changing relationship between several demographics, such as gender, education, and religiosity, and whites’ partisanship. See Abramowitz and Saunders (2006), Mason (2018, Chap. 3), and Zingher (2014, 2019) for greater detail on the over-time partisan allegiances of different demographic groups. See Margolis (2018) for a broader treatment on the complex relationship between religion and partisanship in the United States.

Furthermore, the ANES data only asks about current union affiliation. Thus, it is possible that the “non union affiliated” group includes former union members; this could further underestimate the magnitude of the union-party relationship among white Americans.

Several of the relevant demographic controls were not available in the 1960s. As such, I opted to run simple bivariate models instead. Furthermore, the pooled analyses in Table 1 demonstrate that the union-party relationship is robust to a battery of additional demographic controls.

The sample size for each decade ranges between 1688 and 4252. The election years for each decade are as follows: 1960s = (1964/1968); 1970s = (1972/1976); 1980s = (1980/1984/1988); 1990s = (1992/1996); 2000s = (2000/2004/2008); 2010s = (2012/2016).

This doesn’t mean that partisan control of government is inconsequential, but rather that changes in white macropartisanship do not appear to have directly driven changes in labor union membership.

These two datasets allow me to examine whether a respondent’s union affiliation changed between 1973/1982 and 2010/2012, respectively. However, it does not the exact year in which a respondent joined a union nor if they moved from a RTW state to a NRTW state during the two panel waves, or vice versa. To ensure that peoples’ incentives to join a union remained the same, i.e., that they did not move from a NRTW state to a RTW state and then decide to join a union because of their partisanship, I restrict both samples to respondents whose RTW status did not change in between 1973/1982 and 2010/2012. This means that they could have moved from North Carolina to South Carolina (both RTW) or from New York to New Jersey (both NRTW) but not from California to Florida (NRTW to RTW) or from Iowa to Pennsylvania (RTW to NRTW), for example. I also drop observations from Louisiana (1973/1982) and Indiana (2010/2012) as their right to work status was in flux, i.e., these two states enacted such legislation in the middle of the two panel waves (https://​www.​ncsl.​org/​research/​labor-and-employment/​right-to-work-laws-and-bills.​aspx). This results in approximately 10% of the sample (in both analyses) being dropped. The results are substantively similar if the full sample is examined.

Cumulative ANES data from 1964 to 2016 shows a similar pattern. A simple bivariate probit model shows that white union members are 14 percentage points more likely to identify as Democrats than are non union members in non right to work (NRTW); they are 15 percentage points more likely to identify as Democrats in right to work (RTW) states. This substantively small difference (in the union-party relationship across RTW and non RTW states) is not statistically significant  (p = 0.418).

Further analysis of the 1965–1997 Youth Socialization Panel (ICPSR 4037) from 1973 to 1997 shows that there is relatively little movement in/out of a labor union over peoples’ working lives. For instance 87% of peoples’ union membership status was the same in 1973 (age 26) and 1982 (age 35); this number was 88% from 1982 (age 35) to 1997 (age 50). In short, most people who join a union early on in life tend to remain in one throughout their working lives. Those who do choose to join/leave a union do not appear to be systematically doing so for partisan reasons.

I examine mean differences in Table 5. In Supplementary Appendix B, I run a series of models that include, for each demographic subgroup, a battery of additional demographic controls similar to those in the ANES analyses. The results show that the union-party relationship is robust for each of these subgroups.

The CCES data include categories for: (1) currently a union member, (2) formerly a union member, and (3) never a union member. These data also include analogous categories for union household residence. Here, I simply compare the first and third categories (current member vs. never a union member).

Union decline is likely to be especially consequential in the industrial Midwest. States in this region, including Michigan, Ohio, Pennsylvania, and Wisconsin, have seen 25–30 point declines in union membership, are politically competitive, and have large white populations, relative to the national average. However, even in the former Confederacy, the average decline in union membership (across the 11 states) from 1964 to 2018 was nearly 10 percentage points (http://​unionstats.​gsu.​edu/​MonthlyLaborRevi​ewArticle.​htm). This is not drastically different from the national average, which was approximately of 16 percentage points.

The number of Americans working in the private sector (https://​fred.​stlouisfed.​org/​series/​USPRIV) has consistently dwarfed the number who work in the public sector (https://​fred.​stlouisfed.​org/​series/​USGOVT), even as labor unions grew increasingly strong in the public sector.

At the time of this writing, West Virginia’s right to work law was undergoing court challenges.