DEFINING AND IDENTIFYING CANDIDATES WITH STEM BACKGROUNDS

To systematically study candidates with STEM backgrounds, it is first necessary to define who does (and does not) count as having a background in STEM. I considered candidates to have a STEM background if they have either an educational (master’s degree or higher) or employment background in the “natural” sciences (e.g., physics, chemistry, or medicine), technology (e.g., computer science), engineering, or mathematics. I also included candidates who have only a bachelor’s degree but only if they also are employed as scientists in a STEM field.

This definition, of course, is not the only way to operationalize whether a candidate has a STEM background. An alternative approach focuses on only those candidates who have graduate-level training in a science, technology, engineering, or mathematical field. Science Magazine used this approach in its recent reporting on STEM candidates (Koerth Reference Koerth2018). Given the subjective nature of identifying STEM backgrounds—and the importance of how we define “who counts” on the descriptive conclusions we draw—my definition errs on the side of inclusivity. By pursuing a more inclusive definition, researchers can use the data I make available along with this piece to subset or otherwise alter the data pursuant with alternative definitions (Motta Reference Motta2020).

It also is important to recognize that my definition is conceptually limited to candidates with backgrounds in the “natural”—as opposed to the “social”—sciences. As a social scientist, I want to be clear that it is not my intention to imply that social scientists are not worth studying. Indeed, we have recently had cause to mobilize on behalf of our politically relevant interests—for example, the now-infamous “Coburn Amendment” that temporarily limited National Science Foundation (NSF) political science funding (Mervis Reference Mervis2014).

Instead, I prefer to focus my conceptualization of STEM on the natural sciences because natural scientists have become the “public face” of mobilized science in recent years (Motta Reference Motta2018; Nyhan Reference Nyhan2017). Demonstrating this point, Brulle (Reference Brulle2017) and Fisher (Reference Fisher2017) showed that recent examples of mobilized science (e.g., the March for Science and the People’s Climate March) were largely a reaction to concerns about the Trump administration’s attempts to interfere with the role that scientists—especially climate scientists—play in shaping policies related to the environment.

Additionally, it is important to note that my definition of who counts as a STEM candidate is consistent with several existing political, governmental, and other research-related standards. For example, interest groups like the 314 PAC (i.e., an advocacy group devoted to recruiting, training, and endorsing candidates with STEM backgrounds) and VoteSTEM (i.e., an informational group that provides information to the public about STEM candidates) include only science, technology, engineering, and mathematics when defining STEM candidates in their official mission statements; so also does Conroy (Reference Conroy2018) in her analyses of who ran in the 2018 midterm elections. Moreover, whereas it certainly is true that some federal agencies (e.g., the NSF) define STEM to include social-scientific research, it is important to note that other agencies do not include the social sciences in their definitions (e.g., Department of Homeland Security and Immigration and Customs Enforcement).

THE “POLITICAL SCIENTISTS” DATASET

To facilitate the systematic study of these candidates, and working with my definition, I assembled a new dataset profiling 194 candidates with STEM backgrounds who ran for Congress in 2018 (see https://osf.io/84twz). Summary data for all variables included in the dataset are in table 1.

Table 1 Variable List and Summary Statistics in the Political Scientists Dataset

Notes: ^∗ Indicates that district-level variables are calculated for House candidates only (i.e., senators do not belong to districts). ⁺Indicates that although this variable is theoretically nominal, it is listed in the dataset as a numeric value, encoded as a “string.”

To initiate the process of identifying candidates whose backgrounds matched my definition, I first created a candidate list based on information from the nonpartisan group VoteSTEM.org. VoteSTEM considers a candidate to have a STEM background if that person has a bachelor’s and/or advanced degree in a STEM field or who works professionally in that field.

Recognizing that these data may be imperfect or incomplete, I supplemented this list by appending data from FiveThirtyEight’s list of Democrats running for Congress in 2018 (Conroy Reference Conroy2018). These data identified STEM candidates by noting whether they made some type of public statement regarding their scientific expertise. I cross-validated this list by researching candidate biographies on Ballotpedia.org and retaining only those candidates (N=37) whose scientific backgrounds matched the previous definition.

Of course, because FiveThirtyEight profiled only Democratic legislative candidates, under-coverage on the Republican side is possible. However, because Republicans comprised only a small percentage of observations in the VoteSTEM data, and because the Democratic Party tends to place a higher priority on scientific credentials than the Republican Party (Koerth Reference Koerth2018), I suspect that under-coverage issues were minor.

MEASURES

To systematically study candidates with STEM backgrounds, the Political Scientists dataset includes several measures profiling candidates’ social and political backgrounds. This section discusses how I operationalized each one of these measures.

DESCRIPTIVE ANALYSIS: THE FACE OF MOBILIZED SCIENCE IN THE 2018 MIDTERM ELECTIONS

Figure 1 summarizes several important descriptive findings about the types of candidates with STEM backgrounds who ran for Congress in 2018. First, concerning the candidates’ partisan affiliations, I found that the majority (about 80%) were affiliated with the Democratic Party (panel A). However, it is important to note that the candidates were not exclusively Democrats; approximately 15% were affiliated with the Republican Party and 5% were third-party candidates.

Figure 1 A Descriptive Profile of Congressional Candidates with STEM Backgrounds

Second, panels B and C profile the candidates’ scientific backgrounds. Almost half of these candidates have a doctoral degree (49%; see panel B)—43% of whom hold a doctoral degree in a STEM field and an additional 6% of whom hold at least a bachelor’s degree in a STEM field with a doctoral degree in another field. Moreover, panel C shows that whereas the candidates’ range of scientific expertise was broad, a majority came from only three fields: engineering (28%), medicine (25%), and computer science (11%).

Figure 1 also displays a prominent gender gap (panel D). About one quarter of the STEM candidates who ran for Congress in 2018 were women (26%). This stands in notable contrast to the number of women who sought elected office more generally in 2018. For example, a recent study found that almost half (48%) of all Democratic nominees for federal and gubernatorial races were women (Conroy Reference Conroy2018).

In addition to describing the background of STEM candidates who ran for Congress in 2018, these data provide the opportunity to ascertain whether they were successful. Figure 1 (panel E) shows that less than one third of the STEM candidates who ran for Congress in 2018 advanced beyond the primaries. Although some (9%) were eliminated before primary contests, most lost in primary elections (62%). Only 12% ultimately won election to Congress.

Finally, panel F in figure 1 profiles the campaign histories of STEM candidates who ran in 2018. Consistent with journalistic reporting on the race, I found that the overwhelming majority of STEM Democrats who ran in 2018 never sought elected or congressional office before Trump’s presidency. Three fourths (75%) of the STEM candidates who ran for office in 2018 had no previous experience with electoral politics, and one fifth (21%) had run for congressional or statewide office in the past. An additional 5% previously ran for local office.

Table 2 presents the results of several logistic-regression models that regress a binary indicator of advancement beyond the primaries on candidate- and district-level factors that could potentially influence electoral success. We might expect, for example, that candidates who receive campaign contributions from the 314 PAC may gain a visibility and/or funding advantage that contributes to their success in primary or general-election contests. District-level factors also could influence candidate success if candidates with STEM backgrounds in less-educated districts—or those that more strongly supported Trump over Clinton in 2016—have difficulty finding an audience for pro-science campaign messages. However, the disproportionate share of STEM candidates who are men, Democrats, and/or first-time candidates makes it difficult to predict how gender, partisanship, and campaign histories might factor into electoral success. Because these candidate-level traits are better represented overall, they also may be well represented among those who experience success. Alternatively, because most candidates I observed did not win reelection, it could be the case that these factors have no discernible bearing on election success—or are even negatively associated with it.

Table 2 The Effect of House Candidate and District-Level Factors on Electoral Success (Democrats and Non-Incumbents Only)

Notes: *p<0.05, two-tailed. Logistic-regression coefficients presented with robust (i.e., clustered) standard errors are in parentheses. Models are restricted to House races and Democratic candidates only. All models also exclude incumbent candidates (i.e., all incumbents in the dataset won their respective races). Additional information about the measurement of each variable is in table 1. All variables were recoded to range from 0 to 1, including district-level factors (in which observed minima were assigned to a value of zero, maxima were assigned to a value of 1, and all other items were rescaled proportionately to fall somewhere in between.)

Before turning to the results, five caveats warrant mentioning. First, due to the low frequency of Senate candidates (N=11), I focused my analyses on only House races. Second, because some states and districts featured multiple candidates with STEM backgrounds, I accounted for the possibility of geographically correlated errors by clustering standard-error estimates at both the district (models 1 and 2) and state (models 3 and 4) levels. Third, because all candidates who received 314 PAC funding were Democrats, and out of an abundance of caution regarding the possibility of undercounting Republicans, I included only Democratic candidates in these analyses. Fourth, to avoid collinearity concerns, I modeled district-level educational attainment (models 2 and 4) and support for Trump in the 2016 (models 1 and 3) because these two were highly correlated (r=-0.51). Fifth, I excluded incumbents (N=11) from these models. Because all incumbents in the dataset won reelection, incumbency perfectly predicts advancing to the general election. Consequently, it is important to note that conceptually, of all results observed in table 2, none are as strong as the effects of being an incumbent on electoral success.

These results suggest that receiving support from the 314 PAC was associated with a positive and statistically significant increase in the likelihood of advancing beyond the primary (row 1). Substantively, 314 PAC endorsement (N=27) was associated with a 40% to 42% increase in the likelihood of advancement, across modeling strategies.

Because these models necessarily exclude incumbents, it is interesting that no evidence that first-time Democratic candidates were less successful than those who had previously run for Congress or local office. Similarly, I found no consistent evidence that women candidates with STEM backgrounds were any less or more effective at winning congressional office than men. In fact, in some modeling strategies, both groups at times appear more likely to be successful than their counterparts.

Finally, regarding district-level factors, I found that—contrary to what might be expected—Democrats from more-educated districts were less likely to advance beyond the primaries, whereas candidates from districts that more strongly supported Trump were more likely to advance. A potential explanation for this phenomenon could be that Democrats strategically only ran in Trump-supporting districts—which somewhat (yet imperfectly) reflects educational attainment—if they are of particularly “high quality” and therefore more likely to win (e.g., Maestas et al. Reference Maestas, Fulton, Sandy Maisel and Stone2006). Future research should systematically unpack STEM candidates’ motivations for seeking congressional office.