A “Blue Wave” swept through the United States in the 2018 midterm elections as the Democratic Party won control of the House of Representatives and races for governor in seven states that had been controlled by Republicans.
One of the more compelling media narratives of the 2018 election cycle explained this victory as the result of the mass mobilization of women frustrated by President Donald Trump’s disparaging remarks and conduct toward women. Well-publicized demonstrations like the Women’s March of January of 2017, in which thousands of people took to the streets across the country to protest the new president, appeared to illustrate the power of this mobilization. Women ran for office in record numbers. Observers found that women were more likely to vote Democratic in 2018 than in the 2016 presidential election. Additionally, women represented a larger share of the electorate than men according to exit polls.
In this post, we analyze data from the pre-election Grinnell College National Poll (GCNP) to evaluate whether women were more likely than men to engage in political activism prior to the election. We find that high levels of activism are correlated with an individual’s personal resources, specifically income and education, and by attitudes such as political ideology and identification as a feminist. We found no significant relationship between activism and an individual’s gender after controlling for these factors. Attitudes about women – not gender itself – helped to propel political activism in the 2018 midterms.
The Grinnell College National Poll was conducted August 29-September 2 for Grinnell College by Selzer & Co. of Des Moines, IA. It is based on telephone interviews with 1,002 U.S. adults ages 18 or older, including 779 likely voters in the 2018 general election. The poll included a unique battery of questions designed to measure the political activism of respondents prior to the election, including asking whether the respondent had recently attended a political protest or rally; contacted an elected official; given money to a political campaign; or, helped another person to register or vote. We combined these questions together into an index variable. The index variable ranges from 0 to 4, with 0 indicating the respondent had engaged in none of the listed activities “in the past few years” and 4 indicating the respondent had engaged in all of them during that time period.
We begin by presenting results from an Ordinary Least Squares (OLS) regression designed to measure the strength of the association between a standard set of demographic variables and our dependent variable, political activism. For ease of interpretation, we present the estimates graphically rather than numerically (Figure 1). Dots indicate coefficient estimates and lines indicate 95 percent confidence intervals. Confidence intervals that do not cross the center red line indicate a statistically significant relationship between the variable and political activism. We found that four variables have a statistically significant relationship with political activism: ideology, education, income, and self-identification as a feminist.
Ideology is structured in the dataset as a categorical variable. We analyze it here using dummy variables for Moderates and Liberals. For each, 0 indicates the respondent does not describe themselves using that label, and 1 indicates the respondent does. We use Conservatives as the reference group for this analysis. We find that there is no statistical difference between the activity level of Moderates and Conservatives, but self-identified Liberals were more politically active than Conservatives at a statistically significant level.
Education is scaled from 1 to 5, with 1 indicating a high school education or less and 5 indicating post-graduate education. We estimate a positive relationship between education and activism, indicating that respondents were more likely to engage in political activism as their level of education increased.
Income is scaled from 1 to 5, with 1 indicating an income $25,000 a year or under, and 5 indicating an income of $100,000 a year or over. We estimate a positive relationship between income and activism, indicating that respondents were more likely to engage in activism as their income increased.
“Feminist” is structured as a dummy variable with 0 indicating that a person did not identify as a feminist and 1 indicating they did. We estimate a positive relationship between self-identification as a feminist and political activism.
Our results are also notable for what we did not find. A respondent’s party identification, race and gender had no statistically significant relationship with political activism. In other words, knowing whether a respondent was a Republican or Democrat, or how they identify themselves in terms of race and gender, had no predictive power about their likelihood of engaging in political activism. This remained true when we estimated a separate regression in which we excluded the Feminist variable.
A few specific notes on decisions we made about how to handle our data: the Party ID and race variables are constructed as dummy variables for the OLS analysis. This approach requires us to leave one group out from our analysis as a reference group, or the point of comparison for the other groups. In this case, our analysis uses Republicans as the reference group for Party ID and whites as the reference group for race. The technical interpretation of our results is that there is no statistical difference in the level of activism between the reference group and the group that it is being compared with it. Additionally, our data includes very small sample sizes for some categories within the race variable, raising questions about the accuracy of those findings. We consolidated any group with a sample size below 50 into the “Additional” category for the purpose of this analysis.
Finally, we considered whether disaggregating our political activism index variable might lead to different results. For example, it might be reasonable to expect a relationship between gender and some forms of participation rather than others. We ran for separate OLS regressions using each form of activism (attending a protest, giving money, contacting an elected official, helping someone register to vote) as a dependent variable. In no case did we find a statistically significant relationship between the respondent’s gender and political activism. Interestingly, we did observe a reasonable strong (but not statistically significant) negative relationship between gender and donating money, indicating that men were more likely than women to engage in this form of participation.
Impact of Key Resources and Attitudes
Next, we go into more depth about the relationships we found between political activism and the four variables that are associated with it. Overall, our data show that high levels of political activism were uncommon. Over 44 percent of respondents – a plurality – engaged in no political activity prior to the election. Nearly 24 percent participated in one of the activities under study, 16 percent in two, 10 percent in three, and just 6 percent in four. Who was more likely to be active? The four variables we found significant can be separated broadly into two categories: resources and identity. The income and education variables represent two forms of resources that impact political participation. Political ideology and self-identification as a feminist are both forms of political identity.
Research consistently shows that people with high levels of income and education are more likely than those without such resources to vote and to engage in other forms of political participation, including trying to persuade others to vote, doing campaign work, and donating money to a campaign (Leighley and Nagler pg. 27, Rosenstone and Hansen pg. 132). Our analysis aligns with these findings. Both income and education level were positively correlated with political participation (Figure 2).
As a reminder, our political activism variable ranges from 0-4, and indicates the number of election-related activities engaged in by the respondent. Respondents with a high school education or less had a mean score of .78, while respondents with postgraduate work or degree had a mean score of 1.58. Similarly, respondents with an annual household income under $25,000 had a mean score of .65 while respondents with an income of $100,000 and over had a mean score of 1.51 (Figure 3). Those with the highest levels education and household income were more than twice as likely to participate in electoral politics in the run up to the 2018 midterms than those with the lowest levels.
Political ideology also correlates with participation. Previous research on voting rates by ideology indicates that conservatives are generally overrepresented in the electorate compared to both moderates and liberals (Leighley and Nagler pg. 160). This contrasts with GNCP data, which indicates that ideological liberals were significantly more likely to be active preceding the 2018 midterm elections. Liberals had a mean political activism score of 1.62 while conservatives had a mean political activism score of 1.02. The midterm election mobilized liberals more than it did conservatives.
In contrast, party identification was not correlated with political activism after controlling for other variables. There may have been a ‘blue wave,’ but it was made up of ideological liberals rather than partisan Democrats.
Self-identification as a feminist also was correlated with political activism. Respondents who identified as feminists had a mean political activism score of 1.43, while respondents who did not identify as feminists had a mean score of .92. How widespread is identification as a feminist? According to our data, 36 percent of those polled described themselves as a feminist. Of those who identify as feminist, 60 percent are women and 40 percent are men. Feminists are concentrated in the Democratic party. A total of 56 percent of Democrats identified as feminist compared to 15 percent of Republicans and 38 percent of independents. Among Democrats, feminists were more politically active than non-feminists, with a mean activism score of 1.56 compared to 1.01, respectively (Figure 4).
Perhaps more interesting is what was not significant: gender. Controlling for other factors, women were no more politically active than men according to our data. Attitudes about women mattered more than a respondent’s gender in predicting political activism in the lead up to the 2018 midterms.
Our findings add important nuance to the discussion of the role that gender played in the 2018 midterm elections. Specifically, the political mobilization prior to the substantial Democratic victories in 2018 was not a general mobilization by women or even by Democrats. Instead, it was a mobilization by women and men who leaned ideologically left, valued equal treatment for women, and who were more likely to possess personal resources in the form high levels of education and income. While our findings do not diminish the role that gender played in other aspects of the election, we show that attitudes toward gender played a critical role in mobilizing voters.
While the August/September GCNP offers a useful measure of political activism, continued refinement of the question battery could improve its effectiveness. For instance, the question introducing the political activism battery was phrased “For each [activity], please tell me if this is something you have done in the past few years, have ever done in your past but not recently, or something you have never done before,” followed by a list of activities. This phrasing left open the possibility that respondents could have based their response on activism prior to the 2016 election rather than the 2018 election. Asking about participation “since the 2016 election” would have allowed us to more clearly analyze the change in political activism since Donald Trump’s election as president.
Our research could be extended by repeating the battery of participation questions in future polls. This would allow us to compare activity levels of women over time in addition to comparing them to men. Repeating the participation battery would be especially interesting leading up to the 2020 presidential election. While the 2018 midterms were in many ways a referendum on President Trump, the 2020 election will actually determine if he has enough support to remain in office.
Leighley, Jan E. and Jonathan Nagler. 2014. Who Votes Now? Demographics, Issues, Inequality, and Turnout in the United States. Princeton: Princeton University Press.
Rosenstone, Steven J. and John Mark Hansen. 2003. Mobilization, Participation, and Democracy in America. USA: Longman.
Peter Hanson is associate professor of political science at Grinnell College.
Georgia Rawhouser-Mylet ’21 is a political science major at Grinnell College.