Overview

On Earth Day 2017, thousands of scientists and their supporters protested the Trump administration in Washington, DC as part of the March for Science; marches in other American cities drew tens of thousands [1]. Yet, other scientists express profound misgivings about such efforts to engage in partisan politics. They criticize their colleagues for having “crossed the imaginary line” from researcher to activist, worried that political advocacy might break the social contract underlying public science and research funding [2]. From this perspective, scientists’ credibility as impartial actors requires withdrawing from political life [3].

Other scientists defend colleagues’ political engagement [4]. These scientists believe passive information provision is insufficient in a world where objective facts are the subject of political debate [5, 6]. If political actors reject or discount severe risks to the public, it might be unethical for scientists to stay silent for fear that future research funding could be compromised [5].

These debates have spilled out into the public sphere. Many American political actors sharply question scientists’ motives. Some have asserted that scientists’ political activity reflects partisan preferences or self-interest in receiving government funding. For example, in recent decades, the Republican Party has criticized scientists, threatening to cut science funding and ignore evidence-based policies [7]. Similarly, American conservatives report over-time decline in trust in scientists [8]. After the March for Science, the percentage of conservatives who agreed that “scientists care less about solving important problems than their own personal gain” increased [9].

These critiques of scientists echo arguments about policy feedback effects. Researchers who study policy feedback argue that that government programs, such as subsidies for higher education, can affect their beneficiaries’ political attitudes and behaviors [10, 11]. For instance, the GI Bill increased the civic participation of the veterans who fought in WWII, particularly those who came from disadvantaged backgrounds [12]. Likewise, the National Defense Education Act of 1958, the Higher Education Act of 1965, and Title IX of the 1972 Education Amendments expanded higher education opportunities to women and thus increased their political participation [13]. Researchers suggest that government programs that support higher education created engaged citizens through both resource effects (i.e., increasing the recipients’ economic opportunities) and interpretative effects (i.e., making the recipients feel valued by the government) [14]. Could these theories extend to scientists, and thus validate claims that scientists’ political advocacy reflects vested interests in government resource provision?

Yet, despite contentious debates about the role of scientists in politics and policymaking, arguments about scientists’ political motives have rarely been examined empirically. In this paper, we use a natural experiment to test whether a major government subsidy for science education, the National Science Foundation Graduate Research Fellowship (NSF-GRF) Program, increased civic engagement by scientists. Contrary to partisan critiques of scientists’ political behavior, our study finds no robust evidence that scientists’ motivations to engage in science-related and non-science-related political advocacy are a function of individual-level government benefits provision. Scientists who did not benefit from the NSF-GRF, compared with scientists who benefited but are otherwise similar, were no less likely to vote, donate to political candidates, or participate in the March for Science. Consequently, our research provides credible empirical evidence against rhetorical arguments that link scientists’ political behavior with self-interested financial agendas. It also highlights the limited capacity of even generous government support to trigger positive policy feedbacks.

Materials and methods

Balance tests

Our research design depends on the presence of balance in subjects’ background characteristics between our treated and control groups. To evaluate this feature of our design, we analyzed the natural experiment as a block-randomized experiment (assuming complete randomization within block), where each block is a year/field-of-study. There are a total of 56 year/field-of-study blocks. For each background characteristic, we used the following inverse propensity-weighted (IPW) linear regression to estimate the differences in applicant characteristics between the award winners and non-winners: (1) where p_i is the probability that subject i is in the condition that they are in given their year/field-of-study, X_i is the background characteristic of subject i, and A_i is an indicator variable for subject i’s award status. is the estimate of the difference in the background characteristic between the winners and non-winners. We used IPW linear regression because subjects have different probability of getting an award depending on their year/field-of-study. These probabilities range from 0.17 to 0.83, with a mean of 0.45 and a median of 0.50.

We performed balance tests to assess whether the award winners are different from the honorable mentions along five metrics: 1) undergraduate university is an Ivy League, 2) undergraduate university is an Ivy League Plus (the Ivy League plus MIT, Caltech, Stanford, University of Chicago, Duke, and UC Berkeley), 3) graduate university (at the time of application) is an Ivy League, 4) graduate university (at the time of application) is an Ivy League Plus, 5) predicted gender of the applicant. The gender of the applicant was predicted using first name data from the U.S. Social Security Administration (SSA). We performed web searches for those applicants with gender-neutral names or names not within the SSA database to determine their gender.

In Fig 1, we find that award winners and non-winners are not statistically different when it comes to the quality of their undergraduate or graduate institutions, though winners are 5.7 percentage points (SE = 2.3) less likely to be predicted male compared with non-winners. This difference is perhaps not surprising because the NSF seeks to increase diversity in science education; as a result, application reviewers might favor women and ethnic minorities. To address this imbalance in background covariates, we conditioned on predicted gender (along with self-reported racial identity and place of birth in the survey analysis), along with the quality of education background covariates, in subsequent analyses.

Aside from gender, we find no statistical or substantive differences between in survey sample winners and sample non-winners in background covariates (e.g., quality of education, ethnicity, or U.S.-born). Furthermore, these background characteristics do not individually or jointly predict who won a fellowship (see S1 Data of S7 and S8 Tables. These results increase our confidence that treatment assignment (whether an individual received an NSF or honorable mention) is conditionally as-if random in our survey sample.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Comparing differences in background characteristics between winners and non-winners.

N = 2, 119. We find a slight imbalance between treated and control groups only in predicted gender; however, the groups are balanced on the quality of undergraduate university and quality of graduate university.

https://doi.org/10.1371/journal.pone.0230961.g001

Finally, as shown in Fig 2, there are no statistical or substantive differences between award winners and non-winners regarding missing email addresses or all other contact information. Those who were given an award were more slightly likely to have a “.edu” email address, suggesting that winners are more likely to work in academia.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Comparing differences in contact information between winners and non-winners.

https://doi.org/10.1371/journal.pone.0230961.g002

Results

Receiving a fellowship did little to change scientists’ political attitudes or self-reported behavior (see Fig 3). Interestingly, winners’ support for the March for Science is 3.5% lower than non-winners’ support; however, this difference is not statistically significant at the 5% level. Award winners do not indicate greater support for federal funding for the sciences, for the NSF, or for the NSF-GRF Program. Neither does winning the fellowship induce respondents to communicate their research findings with policymakers and journalists more frequently. Winning the fellowship do little to alter the respondents’ political party affiliation. Winners are somewhat more conservative in terms of political ideology, but this effect is not robust to an alternative model specification where we do not include background covariates (see S1 Data of S2 Table). Finally, in the donation experiment, respondents’ willingness to donate, whether in general or to organizations with a political agenda, is unaffected by their award status.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Effect of being awarded the NSF graduate research fellowship on political attitudes and behavior.

N = 408. The error bars represent 95% confidence intervals calculated from the heteroscedasticity-consistent standard errors. S1 Data of S1 Table contains the effect estimates, standard errors, two-sided p-values, and baseline estimates.

https://doi.org/10.1371/journal.pone.0230961.g003

We then directly evaluated individuals’ political participation using campaign donation data. This analysis was conducted across all individuals in our sampling frame (N = 2, 119), rather than simply across individuals who responded to the survey (N = 499). We do not find that applicants who were awarded the fellowship gave larger political donations to Democratic candidates or causes, as seen in Fig 4.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Effect of being awarded the NSF graduate research fellowship on political donations.

N = 2, 119. The error bars represent 95% confidence intervals calculated from the heteroscedasticity-consistent standard errors. S1 Data of S3 Table contains the effect estimates, standard errors, two-sided p-values, and baseline estimates.

https://doi.org/10.1371/journal.pone.0230961.g004

For a further robustness check on the impact of the fellowship on political donation amounts, for each outcome, we conducted a randomization inference test to test the sharp null hypothesis of no individual treatment effect. For our test statistic, we used the Wilcoxon rank statistic because it is robust to outliers [17]. For each outcome, we performed 1,000 simulations assuming the sharp null and the block randomization in our existing data to get the randomization distribution of the test statistic. Each one-sided p-value is the proportion of the randomization distribution that is more extreme than our observed test statistic. We find that the two-sided p-values (twice the one-sided p-values) are all above 0.90; therefore, we cannot reject the null hypothesis of no individual treatment effect for each outcome.

Discussion

In sum, the null results of award receipt on campaign donations and survey responses suggest that receiving the award did little to change respondents’ political attitudes or behavior. More broadly, we conclude that individual government benefits do not appear to shape scientists’ willingness to engage in either science-related or non-science-related political advocacy. NSF fellowship winners are no more likely than non-winners to support and participate in the March for Science or make political donations.

In the S1 Data of S1 File, we also report the results of two experiments embedded in our survey. In Experiment 1, we randomly divided the survey sample into three groups. The first group received a politically neutral statement about the 2017 March for Science. The second group received the same message with explicit noting that “many March participants wanted to resist Republican party attacks on science and protest federal science policy’s general direction.” A third group received a message that framed the perceived attacks as coming from the Trump administration rather than the Republican Party. Results are reported in S1 Data of S14 Table. Support for the March for Science is equivalent across all conditions. This suggests our sample’s support for political advocacy is relatively non-responsive to partisan cues.

In Experiment 2, we presented respondents with a simple vignette describing the climate advocacy behavior of a hypothetical scholar. We varied both the scholar’s discipline (atmospheric chemistry vs. economics) as well as the nature of the scholar’s advocacy (communication about atmospheric climate science, public endorsement of a carbon tax, public endorsement of a carbon tax plus endorsement of political candidates who support carbon taxes). Results are reported in S1 Data of S2 Fig. We find that scientists’ comfort with political advocacy does not shift as a function of advocacy content. Scientists are as comfortable with a peer advocating for a carbon tax and endorsing candidates who support climate policies, as they are with a peer who limits their advocacy to climate science narrowly.

These results, taken together with our core political advocacy and donation findings, offer an empirical basis to doubt criticism of scientists as being motivated by self-interested political or financial agendas. Other potential mechanisms might contribute to this null finding. While the size of the benefit is significant (amounting to over $200,000 in recent years) and the program is highly visible (recipients receive regular communication from the NSF), its duration is only three years. Furthermore, the program’s beneficiaries might not view the fellowship as a government social program but rather as just reward for their hard work—much like the many Americans who failed to recognize that Social Security is a government social program [18]. As a result, the recipients might not have acknowledged the need to engage in politics to defend the program or the NSF. Alternatively, scientists may fear the loss of future government funding. The March for Science instructed protesters not to wear attire that would identify their universities, use government-funded websites for advertising the protest, or communicate about the March using their institutional email accounts [19]. Because participating in partisan politics could jeopardize their careers, scientists may be reluctant to speak out.

One likely explanation for these findings is the high baseline levels of political engagement in our sample. Evidence from our survey and other surveys of scientists suggest that scientists, as a whole, are politically engaged; interventions to increase their political engagement is likely to hit a ceiling. As the baseline (non-winner) estimates in S1 Data of S1 and S2 Tables show, non-winners in ours survey sample support increasing government funding for science, endorse the March for Science, and overwhelmingly identify as Democrats. More than one-third of the survey sample participated in the March for Science, a costly form of political participation. Regarding a less costly but common form of civic participation, over 98% of both sample winners and non-winners reported they have registered to vote. We also note that government benefits received at an early stage in an individual’s career may have different long-term effects than benefits received at later stages. Follow-up work could investigate the causal effects of other government grant-making programs.

However, none of these potential explanations provide evidence in support of critics’ key claim: that scientists’ political behavior reflects individual financial self-interest. Even at a moment of heightened science politicization, we cannot find empirical evidence in support of the proposition scientists’ political behavior is shaped by having received individual-level government benefits provision. Scientists and their critics should look elsewhere to make sense of scientists’ political interventions.

Supporting information

Acknowledgments

The authors thank Parrish Bergquist, Mark Buntaine, Devin Caughey, Alexander Hertel-Fernandez, Peter Howe, Jennifer Marlon and participants in the UCSB Environmental Politics Seminar and the Midwest Political Science Association 2018 conference in Chicago, IL for comments on earlier drafts of this manuscript.