Hostname: page-component-848d4c4894-x24gv Total loading time: 0 Render date: 2024-05-21T14:44:43.933Z Has data issue: false hasContentIssue false

Giving to the Extreme? Experimental Evidence on Donor Response to Candidate and District Characteristics

Published online by Cambridge University Press:  25 January 2024

Mellissa Meisels*
Affiliation:
Department of Political Science, Vanderbilt University, US
Joshua D. Clinton
Affiliation:
Department of Political Science, Vanderbilt University, US
Gregory A. Huber
Affiliation:
Political Science and Institution for Social and Policy Studies, Yale University, US
*
Corresponding author: Mellissa Meisels; Email: mellissa.b.meisels@vanderbilt.edu
Rights & Permissions [Opens in a new window]

Abstract

How does candidate ideology affect donors' contribution decisions in U.S. House elections? Studies of donor motivations have struggled with confounding of candidate, donor, and district characteristics in observational data and the difficulty of assessing trade-offs in surveys. We investigate how these factors affect contribution decisions using experimental vignettes administered to 7,000 verified midterm donors. While ideological congruence influences donors' likelihood of contributing to a candidate, district competitiveness and opponent extremity are equally important. Moreover, the response to ideology is asymmetric and heterogeneous: donors penalize more moderate candidates five times more heavily than more extreme candidates, with the most extreme donors exhibiting the greatest preference for candidates even more extreme than themselves. Republicans also exhibit a greater relative preference for extremism than Democrats, although partisan differences are smaller than differences by donor extremism. Our findings suggest that strategic considerations matter, and donors incentivize candidate extremism even more than previously thought.

Type
Article
Creative Commons
Creative Common License - CCCreative Common License - BY
This is an Open Access article, distributed under the terms of the Creative Commons Attribution licence (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted re-use, distribution and reproduction, provided the original article is properly cited.
Copyright
Copyright © The Author(s), 2024. Published by Cambridge University Press

The influence of political contributions and donors in elections is increasingly relevant for understanding politics in the United States. All eligible citizens have a single vote, but the means and desire to contribute financially to campaigns may magnify the voice of some above others. Concerns about the potential consequences of this asymmetry have led scholars to investigate how donors influence which candidates choose to run (Schlozman, Verba, and Brady Reference Schlozman, Verba and Brady2012; Thomsen Reference Thomsen2014), candidates' policy positions (Barber Reference Barber2016b; Canes-Wrone and Gibson Reference Canes-Wrone and Gibson2019; Kujala Reference Kujala2020), and the legislation that is ultimately passed (Barber Reference Barber2016c; Bartels Reference Bartels2008). Central to understanding these distortionary effects is the question of not just whether, but how, candidate ideology shapes donors’ willingness to provide financial support, as well the relative impact of candidate ideology vis-à-vis other (instrumental) considerations.

While the general importance of ideology in donation decisions has received a great deal of scholarly attention (Barber Reference Barber2016b; Barber, Canes-Wrone, and Thrower Reference Barber, Canes-Wrone and Thrower2017; Bonica Reference Bonica2014; Kujala Reference Kujala2020; La Raja and Schaffner Reference La Raja and Schaffner2015), current approaches are limited in their ability to identify how, and the relative extent, to which donors respond to candidate ideology. Studies using observed donation patterns to infer determinants of donors' decisions (for example, Ensley Reference Ensley2009; Gimpel, Lee, and Pearson-Merkowitz Reference Gimpel, Lee and Pearson-Merkowitz2008) struggle to disentangle the effects of election, candidate, and opponent characteristics, especially given the strategic entry and increasing use of donor lists to target giving in particular races (Hansen Reference Hansen2016; Koger, Masket, and Noel Reference Koger, Masket and Noel2009). For example, Jaime Harrison, the Democratic candidate for the Senate in South Carolina in 2020, raised a record-breaking $130 million in his raceFootnote 1, which may indicate that Democratic donors strongly supported his policy positions. However, it may have also been due to his personal character, the competitive nature of the election, or the presence of his prominent Republican incumbent opponent, Lindsey Graham. The confounding of features poses a prohibitive hurdle to systematically estimating donors' responses to candidate ideology from campaign receipts.

Given the identification challenges that arise from observational data, others use survey methods to directly question donors about factors affecting their giving decisions (Barber Reference Barber2016a; Francia et al. Reference Francia2003). While broadly informative, the difficulty of interpreting self-reported motivations and assessing trade-offs limits how much we can learn from direct survey reports of motivations. Donors consistently rank ideological agreement as important to their decisions, yet it is unclear whether this means they solely contribute to like-minded candidates (Bonica Reference Bonica2014), are less likely to support candidates as they grow ideologically distant (Barber, Canes-Wrone, and Thrower Reference Barber, Canes-Wrone and Thrower2017), prefer more extreme candidates to more moderate candidates (Patty and Penn Reference Patty and Penn2019), or vice versa (Hall Reference Hall2015). Likewise, existing survey analysis cannot assess the relative importance of multiple factors such as ideological agreement and influencing election outcomes, which donors report as important to their decisions. For example, we do not know whether donors prefer to give to an electorally secure, like-minded candidate or to a candidate running in a toss-up district with whom they disagree on some issues.

We administered a multifactorial vignette experiment to over 7,000 verified donors from the 2018 midterm elections to investigate how candidate ideology affects donor decisions. We identified the relative effects of several factors hypothesized to affect donation behaviour using an experimental approach that circumvents the potential endogeneity issues with observational studies and the limitations of self-reported donation motivations. Our research design combines the external validity of interviewing verified midterm donors with the internal validity provided by a randomized experiment, allowing us to estimate donors' willingness to give based on descriptions of candidates and their electoral conditions, which are difficult to recreate in surveys and isolate in observational studies.

Several important findings emerge. First, candidate ideology clearly matters to donors, but strategic electoral considerations – namely, the partisan lean of the district in which the candidate is running and her opponent's ideology – are equally consequential. Contrary to the behavioural assumptions of prominent donation-based measurement models (Bonica Reference Bonica2014; Hall and Snyder Reference Hall and Snyder2015), donors do not solely consider candidates' spatial proximity to themselves.Footnote 2 Indeed, the impact of running in a competitive district against an extreme opponent rivals the impact of ideological agreement on the likelihood of giving.

Second, donors' willingness to give decreases asymmetrically as candidates' views are further from their own. Donors are most likely to contribute to candidates who share their positions, all else equal, but they significantly prefer candidates who are more extreme than themselves to candidates who are more moderate (Thomsen Reference Thomsen2017). Candidates described as somewhat more extreme than the donor were 5 percentage points less likely to receive a contribution than candidates who share the donor's views, yet candidates described as somewhat more moderate than the donor were almost 20 percentage points less likely.

Third, this asymmetric response to candidate ideology is greatest among the most extreme donors. Although these donors are themselves more extreme than the average donor – who is already more extreme than the average voter (Bafumi and Herron Reference Bafumi and Herron2010; Barber Reference Barber2016c) – extreme donors have the largest relative preference for candidates more extreme than themselves. In fact, extreme donors are as willing to contribute to candidates who are more extreme than themselves as candidates who share their views.

Fourth, we uncover some partisan differences consistent with accounts of asymmetric polarization (Grossmann and Hopkins Reference Grossmann and Hopkins2016; Theriault Reference Theriault2006). Republican donors have a greater preference than Democratic donors for extreme candidates over moderates, but these cross-party differences are smaller than within-party effects of donor extremism and the common large penalty to candidate moderation. While extreme Democrats are less likely than non-extreme Republicans to support candidates described as somewhat more moderate than themselves, extreme Republican donors are the least likely to give to such candidates.

Donors' reported intent to support more extreme candidates over more moderate candidates – especially among extreme donors – suggests that contribution-induced incentives for extremism are greater than currently thought. While existing work argues that candidates financially benefit from adopting donors' (extreme) positions (for example, Kujala Reference Kujala2020), our results indicate that donors are much more willing to contribute to candidates with positions more extreme than theirs than candidates who hold more moderate positions. However, the large impact of contextual, strategic factors on donors' decisions may attenuate these ideological effects: donors are equally likely to support an extreme candidate in a less competitive district as a moderate candidate in a toss-up district. Overall, the patterns we uncover are consistent with forward-looking donors contributing to help move their legislative party's caucus in a more extreme direction (Cameron and Kastellec Reference Cameron and Kastellec2016; Kedar Reference Kedar2005; Krehbiel Reference Krehbiel2007).

Our results also have implications for how political scientists study political donors. A comparison of donors' self-reported motivations for giving and the experimental effects of various factors on giving decisions reveals differences that are difficult to reconcile. In particular, donors who report caring more about candidates' positions and donors who report caring more about candidates' chances of winning weigh hypothetical candidates' ideologies and electoral environments no differently in their contribution decisions. The failure of self-reported measures to predict differences in experimental responses highlights the challenges of using direct survey questions to characterize the trade-offs donors make when deciding who to support in complex multidimensional choice environments. The relationships we uncover between donor ideology, candidate ideology, and likelihood of giving also suggest complications for scaling approaches that interpret observed donations as an expression of donors’ ideologies (for example, Bonica Reference Bonica2014; Hall Reference Hall2015; Hall and Snyder Reference Hall and Snyder2015). In so far as donors respond asymmetrically and heterogeneously to deviations in candidate ideology from their own, using donations to ‘bridge’ candidates and donors into a common space runs the risk of substantial bias.

We substantiate our characterization of donor motivations as follows. Section 1 explores prominent approaches to identifying the motivations of political donors, which grounds and guides our experimental design. Section 2 introduces the experimental multifactorial vignette design we use to identify the effect of different considerations on donors' likelihood of contributing to several hypothetical candidates. Section 3 presents donors' responses to variations in candidate ideology relative to factors such as district competitiveness and opponent extremity and examines whether ideologically extreme and non-extreme donors respond differently. We also use survey questions to test whether direct elicitation predicts the effects we identify in our experimental vignettes. Section 4 concludes by discussing the implications of our findings for both our understanding of donors' impact on contemporary politics and how we study donors and interpret patterns of observed contributions.

Why Donors Give

A large body of work argues and empirically demonstrates that candidate ideology is relevant to individual donors' contribution decisions (Barber Reference Barber2016a; Barber Reference Barber2016b; Barber, Canes-Wrone, and Thrower Reference Barber, Canes-Wrone and Thrower2017; Bonica Reference Bonica2014; Brown, Powell, and Wilcox Reference Brown, Powell and Wilcox1995; Ensley Reference Ensley2009; Francia et al. Reference Francia2003; Gimpel, Lee, and Pearson-Merkowitz Reference Gimpel, Lee and Pearson-Merkowitz2008; Kujala Reference Kujala2020; Magleby, Goodliffe, and Olsen Reference Magleby, Goodliffe and Olsen2018), but it is challenging to identify precisely how ideology affects donors' willingness to give to candidates.

The extant literature is divided on how donors use candidate ideology when deciding who to support. Some argue that individuals contribute such de minimis amounts relative to the possible marginal political gains from influencing elections that donors must support candidates who share their views for the consumption value associated with expressing their views via an additional mode of political participation (Ansolabehere, de Figueiredo, and Snyder Reference Ansolabehere, de Figueiredo and Snyder2003).Footnote 3 If donors primarily give to express support for like-minded candidates, the distribution of donations across candidates will reflect donors' ideologies rather than the importance of strategic considerations such as the competitiveness of the election.Footnote 4 This behavioural model of ideologically expressive giving is the basis of donation-based measurement models of contributor and recipient ideology (for example, CF scores [Bonica Reference Bonica2014] and those developed by Hall and Snyder [Reference Hall2015]).

Others claim that while donors are most likely to contribute to ideologically aligned candidates, they also give to misaligned candidates with decreasing probability as the ideological distance between them grows (Barber Reference Barber2016b; Barber, Canes-Wrone, and Thrower Reference Barber, Canes-Wrone and Thrower2017; Kujala Reference Kujala2020). Donors are commonly assumed to be indifferent between candidates who diverge in either direction, but they may also care about the direction of the divergence. On the one hand, donors may allocate greater resources to candidates more moderate than themselves if moderates are expected to be more electorally successful than extremists (Hall Reference Hall2015; but see Utych Reference Utych2020). Conversely, the greater electoral vulnerability of extremists, all else equal, may lead donors to believe that they are in greater need of financial support than moderates. Donors who give with an eye toward the subsequent lawmaking environment may also prefer to support more extreme candidates who are less likely to compromise with the opposition party (Groseclose and McCarty Reference Groseclose and McCarty2001; Lee Reference Lee2016) and most likely to move the legislative party median closer to donors' extreme preferences (Barber Reference Barber2016c; Cameron and Kastellec Reference Cameron and Kastellec2016; Krehbiel Reference Krehbiel2007; Patty and Penn Reference Patty and Penn2019).

Understanding the importance of candidate ideology relative to other considerations is important for contextualizing the potential effects of donations on the larger political system. Given donors' extreme preferences, those who give solely on the basis of ideological alignment may contribute to elite polarization (Barber Reference Barber2016b; Kujala Reference Kujala2020; La Raja and Schaffner Reference La Raja and Schaffner2015). However, if donors prioritize strategic factors, such as whether a candidate is running in a competitive district due to the importance of majority control for passing legislation (Cox and McCubbins Reference Cox and McCubbins1993), the incentives for candidate extremism may be more muted. Indeed, prior work has found that donors report caring about affecting electoral outcomes (Barber Reference Barber2016a) and are more likely to contribute to Senate candidates in competitive races (Barber, Canes-Wrone, and Thrower Reference Barber, Canes-Wrone and Thrower2017) and give to close races out-of-state (Gimpel, Lee, and Pearson-Merkowitz Reference Gimpel, Lee and Pearson-Merkowitz2008). Issues of endogeneity have made it difficult to assess how important district competitiveness is to donors' decisions compared to the ideology of candidates,Footnote 5 and it is also unclear whether donors prefer giving to candidates running in districts that lean toward or against their party.Footnote 6

Donors may also respond to the characteristics of a candidate's opponent (Barber Reference Barber2016a; Magleby, Goodliffe, and Olsen Reference Magleby, Goodliffe and Olsen2018). For example, those facing an incumbent opponent may be less likely to receive contributions because incumbents are historically more difficult to defeat (Abramowitz, Alexander, and Gunning Reference Abramowitz, Alexander and Gunning2006; Thomsen Reference Thomsen2023; but see Jacobson Reference Jacobson2015). At the same time, donors may prefer giving to candidates challenging incumbents because defeating an incumbent of the opposing party shifts the overall chamber margin by two seats. An ideologically extreme opponent may also increase donors' willingness to contribute, similar to how voters turn out in greater numbers for candidates running against extreme opponents (Hall and Thompson Reference Hall and Thompson2018). This would be consistent with a concern that the election of an extreme opponent will push the opposing party's policy even further away from the donor than the election of a more typical opponent (Hill and Huber Reference Hill and Huber2017).

A final set of potential considerations for donors emerge from the large literature on ‘quality challengers’. Donors may rely on characteristics related to prior legislative experience, fundraising, and interest group support to identify candidates who are more likely to be electorally successful (Biersack, Herrnson, and Wilcox Reference Biersack, Herrnson and Wilcox1993; Box-Steffensmeier Reference Box-Steffensmeier1996; Jacobson Reference Jacobson1989; Maestas and Rugeley Reference Maestas and Rugeley2008). If donors contribute strategically to help their party win or retain seats, they may prefer to support candidates who are thought to be viable based on district enthusiasm, endorsements, strong fundraising, or favourable previous electoral performance. Although such traits have been associated with winning in the past, so-called ‘newcomers’ may also be increasingly seen as viable given contemporary patterns of candidate recruitment (Bawn et al. Reference Bawn2012; Porter and Steelman Reference Porter and Steelman2023).

One complication is that the importance of these considerations may vary across donors. Donors differ in many ways, but a question of particular interest in understanding their potential impact on political polarization is whether those who are the most extreme behave differently than those who are more moderate. Even if all donors prefer giving to extreme candidates over moderate candidates to help pull their party caucus closer to the donors' comparatively extreme preferences, extreme donors have the most to gain (and lose) from changes in majority control because of their relative extremity (Barber Reference Barber2016c). Consequently, it is unclear whether they support more extreme candidates to try and shift the ideological makeup of the party caucus or more moderate candidates better aligned with rank-and-file voters, presumably making them more electable (and majority status more likely).

Research on polarization at the mass and elite levels, which shows greater polarization among Republicans than Democrats (Hacker and Pierson Reference Hacker, Pierson and Persily2015; Mann and Ornstein Reference Mann and Ornstein2016; Theriault Reference Theriault2006), also suggests that the importance of these considerations may vary by partisanship. While it is impossible to know whether differences reflect variation in partisans' priorities – for instance, recent work argues that the parties differ in the relative importance of ‘issues’ versus ‘identity’ considerations (Grossmann and Hopkins Reference Grossmann and Hopkins2016) – or their political context – donors in 2018 were giving to candidates following the 2016 election, where the Republicans took control of both the presidency and Congress – arguments about asymmetric polarization make it important to determine whether Democratic and Republican donors respond similarly to candidate and electoral characteristics.

In addition to the fact that donors of different ideologies and parties may weigh considerations differently when deciding whether to support a candidate, it is also plausible that these motivating considerations interact with one another. Estimating the average effect of a factor while holding all else equal may obscure the conditional nature of the relationship between different considerations, especially among strategic donors. In particular, donors' willingness to give to candidates who are more extreme or more moderate than themselves may depend on whether the candidate is running in a more or less safe district (Baron Reference Baron1994) or against an extreme opponent (Woon Reference Woon2018). Understanding whether the impact of certain factors depends on others is important for evaluating both the extent of donors' impact on the political environment and the types of research designs that can be supported in the presence of such selection effects.

Research Design and Experimental Vignettes

Research on individual donors' contribution decisions largely relies on either aggregate-level donation patterns (for example, Barber Reference Barber2016b; McCarty, Poole, and Rosenthal Reference McCarty, Poole and Rosenthal2006; Stone and Simas Reference Stone and Simas2010) or self-reported motivations from surveys (for example, Canes-Wrone and Gibson Reference Canes-Wrone and Gibson2019; Rhodes, Schaffner, and La Raja Reference Rhodes, Schaffner and La Raja2018). Such studies provide important insights, but both approaches are limited in their ability to isolate the relative effects of multiple factors on donors' decisions. We seek to complement existing findings by providing more specific interpretations of their general patterns and by identifying new patterns in donor behaviour that are difficult or impossible to capture using extant approaches.

Consider, for example, the meaning of the above-average level of funding we observe in competitive districts. Does this reflect ideology-driven donors giving to like-minded candidates who happen to run in competitive races? Or are donors giving to help their party win close races, regardless of the candidates' positions? Alternatively, perhaps donors are choosing to support high-quality candidates who are more likely to be recruited to run in pivotal races. It is even possible that donors know nothing about the candidate they are supporting, given contemporary fundraising practices that encourage contributions to races about which an individual may know little. Making strong inferences about motivations using observed donation patterns is – to put it lightly – challenging due to inevitable problems of confounding, omitted variables and endogeneity caused by the strategic behaviour of candidates, donors, and parties.

Given the difficulties with observational data, some have chosen to directly question donors about their motivations for contributing (for example, Barber Reference Barber2016a; Barber, Canes-Wrone, and Thrower Reference Barber, Canes-Wrone and Thrower2017; Francia et al. Reference Francia2003). In a survey of 2012 Senate donors, Barber (Reference Barber2016a), for example, finds that ideologically extreme donors were more likely to report that a candidate's ideology, the opponent's ideology, and a chance to shape the election outcome were all important reasons for giving. While this work clearly demonstrates that donors consider multiple factors when giving (Magleby, Goodliffe, and Olsen Reference Magleby, Goodliffe and Olsen2018), the relative importance of each factor and how donors trade-off between competing considerations is unclear. Moreover, it can be difficult to interpret what it means for a consideration to matter in donation decisions. If donors report valuing a candidate's ideology, for example, it could mean that they (1) only contribute to like-minded candidates, (2) are willing to give to candidates with whom they disagree, but less so as candidates grow ideologically distant, or (3) have a general preference for supporting more extreme candidates over more moderate candidates, or vice versa.

To both examine the relative importance of candidate, opponent, and district characteristics in donors' willingness to contribute and to avoid ambiguity associated with interpreting survey responses, we combine the advantages of interviewing verified donors with the causal identification provided by the randomization of candidate and electoral characteristics in a multifactorial experimental vignette. We randomly selected nearly 69,000 verified donors from the 2018 midterm elections using Targetsmart's national database of itemized donors, including all congressional donors who gave over $200, as well as those who contributed to campaigns that voluntarily disclosed smaller donations.Footnote 7 Each selected donor was sent a letter, and half were also sent a follow-up reminder postcard inviting them to participate in an online survey in return for a $1 charitable contribution made on their behalf. Ultimately, 7,335 verified donors completed the survey (10.6 per cent), which included vignettes gauging respondents' likelihood of contributing to a hypothetical same-party candidate whose ideology, viability, electoral context, and opponent characteristics were experimentally manipulated.

Table 1 presents the weighted and unweighted demographics of our donor sample. As expected, given prior studies of donor demographics, donors tend to be older, highly educated, very wealthy, and the majority are male (Hill and Huber Reference Hill and Huber2017). To ensure that our respondents are representative of our sampling frame, we construct individual respondent weights to match the distribution of respondent demographics to the demographics of the sampling frame (which was itself a random sample of the population of midterm donors).Footnote 8 Democratic donors are overrepresented in the sampling frame because more Democrats contributed during the 2018 midterm election than Republicans and, consistent with larger non-response patterns in polling (Clinton Reference Clinton2020), Democrats were more likely to respond to the survey than Republicans.Footnote 9 To account for partisan differences, we also conducted separate analyses for Democrats and Republicans in Appendix D.

Table 1. Donor self-reported demographics

The table includes all respondents who finished the survey and self-identified with a party. The weighted sample is smaller due to missing survey weights based on voter file information. Appendix G reports the raw sample demographics in terms of voter file demographics.

In addition to directly asking about motivations for giving (see Appendix A), our primary analysis examines donors' likelihood of contributing to hypothetical same-party candidates that are described with particular sets of randomized traits, opponents, and electoral environments. Our experimental vignette approach has three key advantages. First, the independent randomization of candidate and election characteristics in each vignette allows us to isolate their effects without dealing with the issues of omitted variable bias, collinearity, and endogeneity that are present in analyses of observed donations. Second, because we present each donor with several randomized multifaceted vignettes, we can directly characterize the relative importance of every factor for different donors. This also allows us to estimate the effects of one consideration conditional on randomized changes in another. In so doing, we are careful to minimize potential concerns about design-induced demand effects by providing donors with an entirely new randomized combination of characteristics each time they evaluate a hypothetical candidate. Third, to avoid the difficulties associated with trying to locate candidates and donors on a comparable and commonly understood ideological scale, we explicitly define candidates' ideology relative to the donor's own ideology.

Of course, no research design is perfect, and relying on an experimental vignette introduces a few limitations. First, we must measure self-reported intent to give rather than actual giving. Even if this leads respondents to overestimate their actual likelihood of giving, it will likely bias the relative effects of each consideration rather than the intercept. Second, because we present respondents with multiple vignettes composed of randomized descriptions, the vignettes are necessarily generic and abstract. On the one hand, this is desirable because it avoids problems associated with omitted variables arising from information respondents have about real candidates, as well as bias that might originate if respondents felt external pressure to report supporting specific candidates. On the other hand, hypothetical candidates lack other features that may affect donations, such as public reputation, professional ties, or campaign style – considerations that are more likely salient when donating to in-state candidates. Together with the vignette description of hypothetical candidates as running in a state other than the respondent's, the design generalizes better to out-of-state giving than to in-state giving.Footnote 10

All randomized vignette features are described in Table 2. We introduced the vignettes as follows: with text in << >> indicating randomly assigned features and text in [[ ]] indicating features based on prior survey responses:

We will next present you with 5 different [[OWN PARTY]] candidates who are likely to be running for DIFFERENT House races in the next election cycle. Suppose you were approached by each candidate. How likely would you be to donate to their campaign during the <<RACE TYPE>> election?

Table 2. Randomized vignette features

Each respondent was randomly assigned either five primary or five general election vignettes. All features were randomly assigned with equal probability in each vignette.

Although respondents are instructed to imagine being solicited by the candidate, the candidates are presented using neutral descriptions rather than in the language candidates would typically use in their own appeals, making the design more akin to donors choosing to give of their own initiative rather than reacting to fundraising solicitations.Footnote 11 While donors who did not identify with either of the two major political parties were randomly assigned OWN PARTY as either ‘Democratic’ or ‘Republican’, we focus on the 94 per cent of donors who identified as one or the other. For each donor, all five vignettes were randomly assigned <<RACE TYPE>> of either ‘Primary’ or ‘General’ in order to ease the cognitive load and minimize confusion. Asking about both general and primary elections allows us to determine whether donors' calculations differ when deciding to contribute to candidates in intra-party versus inter-party contests. Each vignette had the following structure:

Candidate #1 [NAME WITHELD] is a <<RACE TYPE>> election candidate in [ANOTHER STATE]. The district <<DISTRICT LEAN>>. Your party's candidate <<VIABILITY>>. They hold policy positions that are <<OWN PARTY IDEOLOGY>>. They will likely face <<OUT PARTY IDEOLOGY>> <<OUT PARTY>> <<OUT PARTY INCUMBENCY>>. What are the chances you would contribute to this candidate?

The outcome of interest is the donor's reported likelihood of contributing to the hypothetical candidate on a five-point labelled scale: I would almost certainly NOT contribute (0–10 per cent), Not very likely (10–35 per cent), Close to even (35–65 per cent), Very likely (65–90 per cent), and I would almost certainly contribute (90–100 per cent). For ease of interpretation, we present results with a binary outcome of whether the donor was very likely or almost certain to contribute (1) or not (0), with parallel specifications using the linear scale reported in Appendix E.

To investigate the effect of candidate ideology on giving, we randomly assign the hypothetical same-party candidate to either share the donor's views on policy, hold views that are somewhat more extreme, hold views that are much more extreme, or hold views that are somewhat more moderate than the donor's.Footnote 12 Recognizing that Democratic and Republican respondents react differently to candidates described as more liberal or conservative than themselves, we also vary the description of the candidate's ideology depending on the donor's partisanship, as outlined in Table 2. For example, for Republican donors, we describe the candidate's policy positions as either ‘about the same as yours’, ‘somewhat more conservative than yours’, ‘much more conservative than yours’, or ‘somewhat more liberal than yours’, with the last option corresponding to being more moderate than the donor. In addition to explicitly describing the candidate's ideology in reference to the donor's, using the terms ‘liberal’ and ‘conservative’ rather than ‘extreme’ and ‘moderate’ helps minimize confusion for donors who consider themselves moderate or ideologically misaligned with their party.Footnote 13

Figure 1 presents the basic relationship between candidate ideology and donors' willingness to give. Consistent with prior findings, the largest percentage of donors (39 per cent) report wanting to contribute when the candidate is described as holding policy views that are ‘about the same’ as their own. More novel, however, is the fact that the willingness to give decreases only slightly (to 36 per cent) for candidates with somewhat more extreme views than the donor's and slightly more (to 30 per cent) for much more extreme candidates, while the largest decline is present for candidates described as somewhat more moderate than the donor (only 21 per cent). Far fewer donors report being willing to support a candidate who is more moderate than themselves relative to a candidate who is more extreme, providing preliminary evidence of donors penalizing moderation more than extremism.

Figure 1. The proportion of donors wanting to give by candidate ideology. The horizontal axis is randomized candidate ideology, described relative to the donor's own positions. The vertical axis is the percentage of donors who indicated being very likely or almost certain to contribute to the candidate.

As Table 2 details, we also randomize other features of the race. This allows us to evaluate the average impact of candidate ideology on donors' decisions relative to the impact of other considerations, as well as whether donors respond differently to candidates' ideologies depending on these other race features. The partisan ‘lean’ of the district in which the candidate is running is said to lean toward the respondent's party, the opposition party, or be a pure toss-up.Footnote 14 We also vary the extremity and incumbency status of the candidate's opponent. Finally, the same-party candidate is described as either a political newcomer, having received key endorsements, raising a good deal of money, and having barely lost a prior election, or running in a district that is enthusiastic about the candidate.

To allow for non-linear treatment effects, we estimate the relationship using indicator variables for every value of the randomized traits. For expositional ease, we focus on an additively separable specification that models the likelihood of donor i contributing to the candidate described in vignette v using:

$$\eqalign{Pr( Y_{iv} = 1) & = \alpha + \beta _1\;{\rm Candidate}\;{\rm Ideolog}{\rm y}_{iv} + \beta _2\;{\rm District}\;{\rm Lea}{\rm n}_{iv} \cr & \quad + \beta _3\;O{\rm pponent}\;{\rm Extrem}{\rm e}_{iv} + \beta _4\;{\rm Opponent}\;{\rm Incumben}{\rm t}_{iv} \cr & \quad + \beta _5\;{\rm Candidate}\;{\rm Viabilit}{\rm y}_{iv} + \gamma {\rm Primary}\;{\rm Rac}{\rm e}_i + \delta _v + {\rm \epsilon }_i.} $$

where Y iv = 1 denotes whether respondent i reported that they were either very likely or certain to contribute to the candidate as described in vignette v. In accordance with Table 2, Candidate Ideology represents separate indicators for whether the candidate is described as being somewhat more moderate, somewhat more extreme, much more extreme, or has about the same ideology (base condition) as respondent i, allowing for non-linear effects. For District Lean, the district either leans toward the respondent's party, leans toward the other party, or is a toss-up (base condition). Opponent Extreme indicates whether the opponent is extreme rather than typical (base condition) and Opponent Incumbent designates whether they are an incumbent rather than a new candidate (base condition). Candidate Viability is a set of indicators for whether the candidate has either received key endorsements, raised money, believes that the district is enthusiastic about the candidate, barely lost last time, or is a newcomer (base condition). We include vignette order fixed effects (δ v)Footnote 15 and cluster standard errors at the respondent level.Footnote 16

The specification also includes an indicator for whether respondent i was assigned primary or general election vignettes to allow for differences in the willingness to contribute by election type (γ). Although campaigns may believe donors' motivations differ in the primary versus general elections (Hassell Reference Hassell2011), Appendix D reveals similar average estimated conditional effects for primary and general election vignettes.Footnote 17 Donors were less likely to report a willingness to contribute to candidates running in primary elections overall, but the estimated incremental effects of the randomized considerations on donors' willingness to give did not vary across election types. Consequently, we pool analyses across primary and general election vignettes.

Although we focus on an additively separable specification, there is reason to think that the effects of various factors are interdependent. Existing research suggests that donors' response to candidate ideology may have an interactive effect on district competitiveness (Baron Reference Baron1994) or the presence of an extreme opponent (Woon Reference Woon2018). To account for strategic interactions, we investigate differences in the effects of candidate ideology by district and opponent treatments in Appendix D. Regardless of whether the candidate was described as running in more-or-less competitive districts or facing an extreme opponent, donors reacted similarly to the same-party candidate's relative ideology.

Results

Figure 2 plots the average effect of each consideration relative to the baseline category on donors' reported likelihood of contributing to a same-party candidate. To help compare the relative effects of candidate ideology, district competitiveness, and opponent extremism, Table 3 summarizes the predicted probabilities.

Figure 2. The average effect of vignette manipulations on the likelihood of contributing. Whiskers are 95 per cent confidence intervals. The outcome is 1 if ‘Very Likely’ or ‘Almost Certain’ to contribute and 0 otherwise. The intercept is 0.34.

Table 3. Predicted likelihood of giving

The baseline represents the model intercept. The percentage point change is the raw difference between the baseline and the likelihood of giving to the candidate with the stated feature. The percentage change is the percentage point change divided by the baseline.

If donors are only willing to give to candidates who are ideologically aligned with themselves (the baseline category), we should observe large negative coefficients for other values of candidate ideology in Fig. 2. While donors are most willing to contribute to candidates who share their views, all else equal they also express some willingness to contribute to candidates of every ideological description, including those who are much more extreme than themselves. Table 3 helps interpret the magnitude of these effects. The probability of expressing a willingness to give to a like-minded candidate with all other factors at their baseline is 0.34 on the 0–1 scale, a 34 per cent probability. All else equal, the probability of giving to a somewhat more moderate candidate is only 16 per cent – 18 percentage points lower. Donors are, therefore, 54 per cent less likely to support a candidate who is somewhat more moderate than themselves relative to an ideologically aligned candidate.

By contrast, donors' predicted probability of giving to a candidate described as somewhat more extreme than themselves is still 31 per cent – a 3 percentage point penalty (8 per cent decrease). This suggests that, on average, donors penalize candidates more moderate than themselves five times more heavily than candidates more extreme than themselves. Finally, although donors are less willing to contribute to a much more extreme candidate (25 per cent), this difference relative to a like-minded candidate is only 9 percentage points – less than half of the penalty associated with a somewhat more moderate candidate.

These results also reveal that considerations beyond candidate ideology strongly influence contribution decisions. If donors were solely motivated by ideological alignment, we would observe null effects of candidate viability, district competitiveness, and opponent characteristics. Instead, donors have a strong preference for giving to candidates running in toss-up districts relative to districts that lean toward one party. A candidate is 7 percentage points (or 20 per cent) less likely to receive a contribution if they are in a district that leans in either direction than in a toss-up district. In fact, the estimated difference in giving between more and less competitive districts rivals the difference between ideologically-aligned and much more extreme candidates and is far larger than the difference between somewhat more extreme and ideologically-aligned candidates. This suggests that electoral context is as influential on donors' decisions as candidate ideology.

We also find a substantial effect of the opponent's ideology on the decision to contribute: donors are about 10 percentage points more likely to give to a same-party candidate running against an ideologically extreme opponent. The relative magnitudes associated with same-party and opposing candidate ideology imply that donors are as willing to support a same-party candidate who is much more extreme than themselves but running against an extreme opponent as they are to a candidate who shares their ideology but is facing a typical opponent. Additionally, Fig. 2 shows that opponent incumbency and common proxies for candidate viability have modest effects on donors' decisions.

Variation by Self-Reported Motivation

Perhaps the average effects discussed above mask important variations among donors who are motivated to contribute for different reasons.Footnote 18 To explore this possibility, we re-examine the effect of the vignette considerations while allowing effects to vary based on donors' self-reported motivations for giving. Specifically, we leverage donors' responses to a survey question asked prior to the experimental vignettes:

Which better characterizes your decision to contribute to a specific <<OWN PARTY>> House candidate?

  • I care more about the candidate's positions

  • I care more about the candidate's chances of winning the election

Consistent with prior work on the importance of ideology in contribution decisions, Fig. 3 reveals that donors report caring more about the candidate's positions than the candidate's chances of winning by nearly a 2–1 margin (62 per cent versus 38 per cent). To evaluate whether different self-reported motivations predict different responses to factors implied by those motivations, we separately estimate the effects of the vignette manipulations on the likelihood of contributing among donors who report caring more about the candidates' positions versus the candidates' chances of winning.

Figure 3. Self-reported motivation for giving. Weighted proportions of responses to, ‘which better characterizes your decision to contribute to a specific <<OWN PARTY>> House candidate?’

Figure 4 plots regression results by self-reported motivation and reveals that they are nearly identical to the pooled results in Fig. 2. Donors who report caring most about candidates' positions are not significantly less responsive to district characteristics than donors who purport to prioritize winning. Although we might expect donors who care more about candidates' positions to penalize incongruence more than donors who care more about winning, this is not borne out. If anything, issue-motivated donors only further penalize moderation, and the difference between issue-driven and winning-driven donors is statistically and substantively insignificant. Whereas the one-third of donors who report caring most about winning are less willing to contribute in primaries than general elections overall compared to issue-based donors, the overall pattern of results suggests that donors react to candidate ideology and other factors similarly, regardless of their self-reported motivations for giving.

Figure 4. The average effect of vignette manipulations on the likelihood of contributing by self-reported donation motivation. Hollow circles report prioritizing winning over issues when contributing, and filled circles prioritize issues over winning. Whiskers are 95 per cent confidence intervals. Outcome is 1 if ‘Very Likely’ or ‘Almost Certain’ to contribute, and 0 otherwise. Intercept is 0.38 for Winning and 0.35 for Issues.

These results show a lack of mapping between donors' self-reported priorities and responses to the factors relevant to such priorities. Donors who report caring about candidates' positions more than their chances of winning do not respond more strongly to candidates' ideologies or district context in their decisions than donors who report caring about candidates' chances of winning more than positions. As explicated previously, the interpretation of directly elicited motivations is frequently unclear. For instance, donors who prioritize funding ideologically aligned, more extreme, or more moderate candidates may all report issues as more important than winning. The inability of direct motivations to predict variations in relative effects of key ideological and strategic contribution considerations highlights the difficulty of capturing complex choice environments with straightforward survey items. Although experimental vignettes are likewise limited, they better capture the trade-offs associated with donation decisions, thus complementing and enhancing the interpretation of existing results on donors' self-reported motivations.

Variation by Donor Ideological Extremity

Because donors are, on average, more extreme than voters (Bafumi and Herron Reference Bafumi and Herron2010) and incumbent senators (Barber Reference Barber2016c), investigating whether those who are relatively more and less extreme make their contribution decisions differently is important for understanding the potential incentives they create for candidates. In particular, identifying the extent to which the most extreme donors give on the basis of candidates' ideologies, as well as how these donors respond to candidates of different ideologies, can provide insight into how they could exacerbate elite polarization. If the most extreme donors have a preference for candidates who are even more extreme than themselves, this essentially implies a preference for the most extreme candidates in the entire political system.

Expectations about the giving behaviour of more extreme versus less extreme donors are unclear. On the one hand, extreme donors may be willing to support candidates more moderate than themselves if they are aware that their views are out of line with those of voters, and more moderate candidates may be required for the party to win elections (Hall Reference Hall2015). On the other hand, extreme donors may prefer supporting candidates who are even more extreme than themselves in hopes of moving their party's platform and caucus composition closer to their own more extreme positions (Patty and Penn Reference Patty and Penn2019).

We employ several measures of extreme ideology to assess whether more and less extreme donors respond differently to election features when making their decisions. First, we classify donors as extreme if they self-identified as either ‘Extremely Liberal’ (for Democrats) or ‘Extremely Conservative’ (for Republicans) on a standard 7-point ideology scale. Nearly 20 per cent of donors in our sample identified as such.

To alleviate concerns that each donor may use the 7-point ideology scale differently, we also perform principal component analyses using forty-nine issue questions asked elsewhere in the survey to construct issue-based ideology scores. Following Broockman and Malhotra (Reference Broockman and Malhotra2020), we create a summary measure using all issue questions, as well as dimension-specific measures using eleven questions related to domestic social issues, thirteen questions related to domestic economic issues, and thirteen questions dealing with globalism issues.Footnote 19 For comparability, we classify the Democrats with the most liberal scores and the Republicans with the most conservative scores to match the proportion of self-reported extremists in each party.Footnote 20

Although ideological scales may reflect both issue extremity and consistency (Broockman Reference Broockman2016), the fact that donors' views are more constrained than the average voter suggests that the variation we recover is related to extremity rather than consistency. In Appendix F, we detail this estimation and demonstrate the robustness of the results to using social, economic, or globalism issue-specific measures where consistency across questions is more easily satisfied, suggesting that the summary measure captures meaningful ideological variation. To demonstrate that measurement differences are not driving our results, Fig. 5 plots the proportion of donors self-identifying as extreme across different values of the continuous issue-based extremism score. As expected, very few donors in the lower half of the issue extremism scale identify as extreme, and donors in the upper half become increasingly more likely to identify as extreme as their issue positions become more extreme. However, consistent with some individuals interpreting ideology in non-issue-based ways (Ellis and Stimson Reference Ellis and Stimson2012; Hopkins and Noel Reference Hopkins and Noel2022), not all donors with the most extreme issue positions identify as extreme.

Figure 5. The relationship between issue-based ideology and self-reported ideology by party. The horizontal axis is an absolute scaled issue-based PCA score. The vertical axis is the proportion of donors in the bin who identified as ‘Extremely Liberal’ or ‘Extremely Conservative’. Bin intervals span 0.1 on the absolute PCA scale.

Figure 6 plots the estimated effects of the aforementioned vignette features on the likelihood of contributing separately for self-reported (left) and issue-based (right) extreme and non-extreme donors. Across both classifications, two differences between extreme and non-extreme donors are especially noteworthy. First, even though extreme donors are, by definition, among the most extreme individuals participating in politics, they are less likely than non-extreme donors to support a candidate described as somewhat more moderate than themselves. Second, extreme donors are actually more likely than non-extreme donors to support candidates who are described as more extreme than themselves. In fact, by both measures of extremism, extreme donors are at least as willing to support a candidate who is somewhat more extreme than themselves as they are to support a candidate who shares their views.

Figure 6. The average effect of vignette manipulations on the likelihood of contributing by ideological extremism. Left: Extreme identified as extremely liberal or conservative. The baseline is 0.39 for Extreme and 0.33 for Non-Extreme. Right: Extreme falls in equivalent quantiles of issue-based PCA scores. The baseline is 0.40 for Extreme and 0.33 for Non-Extreme. Whiskers are 95 per cent confidence intervals. Outcome is 1 if ‘Very Likely’ or ‘Almost Certain’ to contribute, and 0 otherwise.

Despite already holding views that are more ideologically extreme than those of other donors, politicians, and voters in their party, these results suggest that extreme donors are also the least willing to penalize those more extreme than themselves while harshly penalizing those who are more moderate. To quantify the magnitude of extreme donors' relative preference for extremism over moderation, we can compare their reported behaviour to that of non-extreme donors. Compared to the likelihood of contributing to a candidate who shares the donor's ideology, non-extreme donors are 17 percentage points less likely to support a somewhat more moderate candidate and 4 percentage points less likely to support a somewhat more extreme candidate. In contrast, extreme donors are 24 percentage points less likely to support a somewhat more moderate candidate and at least 2 percentage points more willing to support a somewhat more extreme candidate. Given the differences in effect sizes, these estimates suggest that extreme donors are 26 points less likely to give to a more moderate candidate relative to a more extreme candidate – double the 13 point relative penalty to more moderate candidates among less extreme donors.

Variation by Donor Ideological Extremity and Partisanship

Beyond differences due to donor extremity, are there partisan differences in how donors give? Recent work has highlighted asymmetric partisan polarization at both the mass and elite levels (Grossmann and Hopkins Reference Grossmann and Hopkins2016; Theriault Reference Theriault2006; Thomsen Reference Thomsen2014), which suggests that Republican and Democratic donors may respond to candidate characteristics differently. In fact, some have argued that contemporary Republican donors and other party elites have a lower tolerance for moderation (Hacker and Pierson Reference Hacker, Pierson and Persily2015). To investigate variation by donor partisanship, we re-estimate the relationships in Fig. 6 separately for extreme and non-extreme Democratic and Republican donors. Figure 7 reports the estimated effects of candidate ideology on donors' willingness to contribute by party and extremity. To formally test for partisan differences using a nested model, Table 4 reports partisan interactions from a pooled model.

Figure 7. Effect of same-party candidate ideology by party and extremism. Models include all vignette manipulations. Left: Extreme identified as ‘Extremely Liberal’ or ‘Extremely Conservative’. Right: Extreme falls in equivalent quantiles of issue-based PCA scores. Whiskers are 95 per cent confidence intervals. Outcome is 1 if ‘Very Likely’ or ‘Almost Certain’ to contribute, and 0 otherwise.

Table 4. Party-interacted effect of candidate ideology on giving

*p < 0.05, **p < 0.01, ***p < 0.001. Models include all vignette manipulations. Respondent-clustered standard errors are in parentheses. The outcome is 1 if ‘Very Likely’ or ‘Almost Certain’ to contribute and 0 otherwise.

The results of Fig. 7 reveal that extreme and non-extreme donors do not differ substantially by party. In general, extreme Democratic donors and extreme Republican donors are less willing to support more moderate candidates and more willing to support candidates who are somewhat more extreme or much more extreme than themselves. The cross-party similarities between extreme and non-extreme donors suggest that donors in both parties likely play a part in incentivizing extremism, and within-party differences in ideological extremism appear more consequential than between-party differences for donor behaviour.

Although there are similarities between more and less extreme donors across parties, there is also some evidence that Republican donors are less willing than Democratic donors to support candidates more moderate than themselves. Consistent with arguments suggesting that ideological moderates are less welcome in the Republican Party than in the Democratic Party, the partisan interaction terms reported in Table 4 for the pooled specification (1) reveals that Republican donors are a few percentage points more likely to report wanting to contribute to candidates who are more extreme than themselves than are Democratic donors, and they may be even less willing to support more moderate candidates as well.

To further probe the partisan asymmetry in relative preference for extremism, we separately estimate the interaction model by donor extremism. First, comparing the differences in main effects between Specification 2 versus Specifications 3 and 4 versus Specification 5 makes clear that extremity explains donors' response to candidate ideology better than partisanship. In general, partisan differences captured by the Republican interaction terms in each model are smaller than differences between the same coefficients in extreme versus non-extreme models. Second, there is some evidence that Republicans are willing to support more extreme candidates than Democrats, even conditional on donor extremity. While not always distinguishable from zero, the interaction effects estimated in Table 4 reveal that Republicans are generally less likely to support moderates than Democrats and more likely to support somewhat or much more extreme candidates than similarly situated Democrats.

Conclusion and Implications

Understanding what affects donors' decisions to contribute to a candidate is a prerequisite for understanding the nature of donor influence in contemporary politics. A substantial body of work has established the general importance of candidate ideology to contribution decisions, but identifying its importance relative to other considerations is challenging. Because of the difficulty of disentangling characteristics of candidates, districts, and donors in observational studies and the ambiguity inherent in direct survey questions that fall short of capturing the complex choice environment, our current knowledge of how candidate ideology shapes donors' decisions is limited.

We contribute to this important effort by conducting the largest-ever survey of verified midterm donors by administering a multifactorial vignette experiment of over 7,000 donors to estimate the impact of the candidate, opponent, and district characteristics on the likelihood of supporting a hypothetical candidate. Independently randomizing each of these factors in every vignette allows us to identify their relative effects, as well as whether those effects vary across donor traits and election contexts.

Our findings provide compelling evidence that donors respond strongly to a same-party candidate's ideology, the competitiveness of the district in which they are running, and the extremity of their opponent. All else equal, donors are significantly more likely to contribute if a candidate is running in a toss-up district, facing an extreme candidate, or shares their views. Donors are about as willing to give to an ideologically divergent candidate running in a competitive district against an extreme opponent as they are to an ideologically aligned candidate running against a typical opponent in a district that leans toward one party. The fact that multiple considerations affect donors' decisions means that we cannot easily interpret patterns of observed donations as simply a reflection of donors' policy positions.

Although donors prefer to support like-minded candidates, donors also strictly prefer to support more extreme candidates over more moderate candidates. Moreover, this asymmetry is largest among the most extreme donors, despite the fact that they are already more extreme than others in their party. Republican donors also appear less likely to support moderate candidates, but the effect of ideological positions in each party is more consequential than the partisan differences we detect. While Republicans may have a greater relative preference for extreme over moderate candidates compared to Democrats, the difference in relative preferences is greater between extreme and non-extreme donors across the parties.

While we cannot directly test why donors vastly prefer contributing to candidates who are more extreme than themselves over more moderate candidates, the finding is consistent with strategic, forward-looking behaviour. Akin to the results regarding voter preferences from Patty and Penn's (Reference Patty and Penn2019) formal model, donors may likewise have an ‘induced taste for extremism’ (744) due to successfully elected candidates' ability to play only a small part in influencing final policies. Because most representatives can do little more than vote on agenda items, donors may give to candidates who are extreme, running in key districts, or facing extreme opponents in order to help ‘move the median’ in their preferred (extreme) direction and prevent the out-party from doing the same (Cameron and Kastellec Reference Cameron and Kastellec2016; Kedar Reference Kedar2005; Krehbiel Reference Krehbiel2007).

Our analyses also have implications for the study of contributions and donor motivations. The asymmetric and heterogeneous relationships between donor ideology, candidate ideology, and the likelihood of giving suggest complications for scaling approaches that interpret observed donations as an expression of donors' ideologies (for example, Bonica Reference Bonica2014; Hall Reference Hall2015; Hall and Snyder Reference Hall and Snyder2015). In addition to responsiveness to strategic factors, donors' asymmetric preference for extreme candidates over moderate candidates implies that donors' locations are not simple functions of the locations of candidates whom they support. Moreover, extreme donors' greater preference for extreme candidates suggests that donors' utility functions vary with their ideologies. Because donors respond to equidistant candidates differently, depending on their relative locations, and donors' responses also vary by the donors' own extremism, using observed contributions to place candidates and donors in a common space likely requires a more complicated behavioural – and, therefore, statistical – model than those currently used.Footnote 21

Likewise, the inability of self-reported donation motivations to predict the patterns we identify in our experimental vignettes is concerning. Although surveys are exceptionally valuable for characterizing donors' policy views, their usefulness for determining the motivations for and implications of donors' giving may be more limited due to the complexity of donors' choice environment – a complexity that our vignettes seek to capture, albeit imperfectly. While we cannot directly translate the effects from our hypothetical, experimental setting into the implied effect of actual giving, the persistence and robustness of the overall patterns of effects we find across the multiple within-donor vignettes are reassuring.

Looking forward, subsequent work can build on our approach in several ways. One is to examine whether the patterns uncovered here persist over time. Given the modest partisan differences we find, future studies can try to better understand whether partisan donors react differently depending on the larger political context they face. In 2018, for example, Republicans controlled both legislative chambers and the presidency, which may have affected those who donated and the relative weight partisans gave to different factors in choosing among candidates. Another possibility is to add elements to the choice environment we lay out in the experimental vignettes, including other candidate features such as race or gender. Because our study presents respondents with neutral descriptions of candidates, questions also arise as to how campaign appeals may affect donors' willingness to give. Can moderate candidates use messaging to increase their attractiveness to extreme donors? Building on the sentinel evidence we collect can help to further enhance our knowledge of the relative influence of these factors and more. Finally, our combination of purposive sampling of donors and a vignette approach is portable to other offices, levels of government, and potentially countries to help interpret observational findings.

Supplementary material

The supplementary material for this article can be found at https://doi.org/10.1017/S0007123423000650.

Data availability statement

Replication data for this article can be found in Harvard Dataverse at: https://doi.org/10.7910/DVN/Q1X3RZ.

Acknowledgements

The authors thank Larry Bartels, Andy Hall, Seth Hill, Cindy Kam, Neil Malhotra, John Sides, and audiences at Vanderbilt University, UC San Diego, and APSA 2021 for helpful feedback and reactions to earlier versions.

Financial support

None.

Competing interests

None.

Footnotes

2 While Bonica (Reference Bonica2013) allows for the systematic incorporation of non-spatial considerations, scaling models that include only individual donors do not.

3 Ansolabehere, de Figueiredo, and Snyder (Reference Ansolabehere, de Figueiredo and Snyder2003) name multiple factors that may affect donors' giving decisions, but scholars have since interpreted donation-as-consumption to mean ideology driven.

4 In the stark donation-as-consumption story, a donor exhausts her budget on candidates with whom she fully agrees with before contributing to candidates with whom she disagrees. This interpretation has been challenged in the context of corporate giving by scholars arguing that strategic giving is calibrated to expected needs and political returns (Gordon, Hafer, and Landa Reference Gordon, Hafer and Landa2007).

5 Because candidates can strategically enter particular races or strategically adapt their public ideology for the electoral setting, it is likely that candidates' ideologies are systematically related to district lean (and opponent characteristics). Importantly, the relationship between candidate ideology and district lean likely varies by candidate quality – making it difficult to pinpoint, for instance, whether donors eschew extremists running in competitive districts because of their ideology or because they are low-quality.

6 While greater instrumental returns to giving in toss-up districts is clear, donors have a greater chance to shape their party's legislative coalition in districts that lean toward the party, whereas districts that lean toward the opposition present a chance to gain a seat.

7 Donors who contributed under $200 according to the FEC constitute 18 per cent of our sample. Given mixed evidence regarding the differences between small and itemized donors (Alvarez, Katz, and Kim Reference Alvarez, Katz and Kim2020; Magleby, Goodliffe, and Olsen Reference Magleby, Goodliffe and Olsen2018), we compare vignette results among both groups in Appendix D and find negligible differences.

8 Appendix G details our inverse propensity score and iterative raking weighting approaches. Our results are robust to the exclusion and different choices of weights (Miratrix et al. Reference Miratrix2018).

9 Because donor partisanship is only known in party registration states, it is impossible to construct respondent weights based on the partisanship of the sampling frame (see Appendix G).

10 In the Appendix, we separately analyze results for donors who only contributed to in-state races and those who only contributed out-of- state, with similar effects across groups.

11 While our experimental framework could be extended to investigate how donation decisions vary with campaign fundraising strategies, we first provide evidence about donors' responses to objective descriptions of candidate, district, and opponents. As such, our results may serve as a benchmark for future studies measuring the impact of campaign messaging strategies on donors' proclivity to give.

12 We did not include an option corresponding to the candidate being ‘much more liberal’ than a Republican donor or ‘much more conservative’ than a Democratic donor so as to focus our limited statistical power on variation in extremism in light of recent increases in candidate extremism (e.g., Utych Reference Utych2020). Moreover, we focus on a unidimensional, liberal-conservative dimension because issue sorting among candidates suggests that issue-based giving would be unlikely to induce bias in our estimates. For example, donors who prioritize the issue of abortion would undeniably associate pro-choice with ‘liberal’ and pro-life with ‘conservative’, making our descriptions informative and relevant, even for donors only thinking about a specific issue.

13 For instance, describing a candidate as holding positions ‘more moderate than yours’ may be interpreted by a Democratic donor who is ideologically conservative as either more liberal or more conservative than themselves. While real-world candidates may not willingly identify as ideologically misaligned with their party, the media and observers frequently use such language. We are more interested in how donors respond to independent descriptions of candidates than to candidates' strategic campaign appeals.

14 One concern with randomized vignettes is that certain combinations of vignette features may be implausible, such as extremists running in swing districts. Empirically, however, other work finds that extremists run with some frequency in swing districts: in 2018, about one-third of general election nominees in competitive districts were extreme, compared to about 40 per cent in safe districts (Meisels et al. Reference Meisels, Clinton and Huber2023).

15 This accounts for baseline changes in the likelihood of giving across the vignette order. Separate models in Appendix D show that effects do not vary substantially across vignette order.

16 We use a linear probability model for ease of interpretation, but using a logit or probit (or even an ordered logit on the full scale) reveals substantively similar conclusions with the added interpretative costs.

17 Whereas we might expect donors to penalize ideologically misaligned candidates more in a primary, we find that, if anything, the opposite is true. This may reflect donors' dual lack of enthusiasm about the prospect of electing a candidate more moderate than themselves and skepticism about the chances of successfully electing a candidate more extreme than themselves.

18 In Appendix D, we also investigate differences by donors' total contribution number, amount, and wealth, finding substantively similar results.

19 Appendix F demonstrates strong relationships between all four measures, the weakest correlation being 0.71.

20 To ensure comparability between measures, we classified as issue-based extremists the same proportion of partisans who self-identified as extremists (~15 per cent of Republicans and ~19 per cent of Democrats).

21 This suggests that even incorporating asymmetric preference for candidate extremism into donors' utility functions would not fully address the scaling challenges implied by our results. In particular, the implication that donors' utility functions vary by their ideal points poses a formidable hurdle to estimation, as donors' ideal points are themselves an unobserved parameter to be estimated.

References

Abramowitz, AI, Alexander, B, and Gunning, M (2006) Incumbency, redistricting, and the decline of competition in U.S. House elections. Journal of Politics 68(1), 7588.Google Scholar
Alvarez, RM, Katz, JN, and Kim, S-yS (2020) Hidden donors: the censoring problem in U.S. federal campaign finance data. Election Law Journal 19(1), 118.Google Scholar
Ansolabehere, S, de Figueiredo, JM, and Snyder, JM (2003) Why is there so little money in U.S. politics? Journal of Economic Perspectives 17(1), 105–30.Google Scholar
Bafumi, J and Herron, MC (2010) Leapfrog representation and extremism: A study of American voters and their members in Congress. American Political Science Review 104(3), 519–42.Google Scholar
Barber, M (2016 a) Donation motivations: Testing theories of access and ideology. Political Research Quarterly 69(1), 148–59.Google Scholar
Barber, MJ (2016 b) Ideological donors, contribution limits, and the polarization of American legislatures. Journal of Politics 78(1), 296310.Google Scholar
Barber, MJ (2016 c) Representing the preferences of donors, partisans, and voters in the US Senate. Public Opinion Quarterly 80(S1), 225–49.Google Scholar
Barber, MJ, Canes-Wrone, B, and Thrower, S (2017) Ideologically sophisticated donors: Which candidates do individual contributors finance? American Journal of Political Science 61(2), 271–88.Google Scholar
Baron, DP (1994) Electoral competition with informed and uninformed voters. American Political Science Review 88(1), 3347.Google Scholar
Bartels, LM (2008) Unequal Democracy: The Political Economy of the New Gilded Age. Princeton: Princeton University Press.Google Scholar
Bawn, K et al. (2012) A theory of political parties: Groups, policy demands and nominations in American politics. Perspectives on Politics 10(3), 571–97.CrossRefGoogle Scholar
Biersack, R, Herrnson, PS, and Wilcox, C (1993) Seeds for success: Early money in congressional elections. Legislative Studies Quarterly 18(4), 535–51.Google Scholar
Bonica, A (2013) Ideology and interests in the political marketplace. American Journal of Political Science 57(2), 294311.Google Scholar
Bonica, A (2014) Mapping the ideological marketplace. American Journal of Political Science 58(2), 367–86.Google Scholar
Box-Steffensmeier, JM (1996) A dynamic analysis of the role of war chests in campaign strategy. American Journal of Political Science 40(2), 352–71.Google Scholar
Broockman, DE (2016) Approaches to studying policy representation. Legislative Studies Quarterly 41(1), 181215.Google Scholar
Broockman, D and Malhotra, N (2020) What do partisan donors want? Public Opinion Quarterly 84(1), 104–18.CrossRefGoogle Scholar
Brown, CW, Powell, LW, and Wilcox, C (1995) Serious Money: Fundraising and Contributing in Presidential Nomination Campaigns. Cambridge: Cambridge University Press.CrossRefGoogle Scholar
Cameron, CM and Kastellec, JP (2016) Are Supreme Court nominations a move-the-median game? American Political Science Review 110(4), 778–97.Google Scholar
Canes-Wrone, B and Gibson, N (2019) Does money buy congressional love? Individual donors and legislative voting. Congress & the Presidency 46(1), 127.Google Scholar
Clinton, J (2020) Task Force on 2020 Pre-Election Polling: An Evaluation of the 2020 General Election Polls. Alexandria, VA: American Association for Public Opinion Research.Google Scholar
Cox, GW and McCubbins, MD (1993) Legislative Leviathan. Berkeley: University of California Press.Google Scholar
Ellis, C and Stimson, JA (2012) Ideology in America. New York: Cambridge University Press.Google Scholar
Ensley, MJ (2009) Individual campaign contributions and candidate ideology. Public Choice 138(1–2), 221–38.Google Scholar
Francia, P et al. (2003) The Financiers of Congressional Elections: Investors, Ideologues, and Intimates. New York: Columbia University Press.Google Scholar
Gimpel, JG, Lee, FE, and Pearson-Merkowitz, S (2008) The check is in the mail: Interdistrict funding flows in congressional elections. American Journal of Political Science 52(2), 373–94.Google Scholar
Gordon, SC, Hafer, C, and Landa, D (2007) Consumption or investment? On motivations for political giving. Journal of Politics 69(4), 1057–72.Google Scholar
Groseclose, T and McCarty, N (2001) The politics of blame: Bargaining before an audience. American Journal of Political Science 45(1), 100119.Google Scholar
Grossmann, M and Hopkins, DA (2016) Asymmetric Politics: Ideological Republicans and Group Interest Democrats. Oxford: Oxford University Press.Google Scholar
Hacker, J and Pierson, P (2015) Confronting asymmetric polarization. In Persily, N (ed.), Solutions to Political Polarization in America. Cambridge: Cambridge University Press, 5968.Google Scholar
Hall, AB (2015) What happens when extremists win primaries? American Political Science Review 109(1), 1842.Google Scholar
Hall, AB and Snyder, JM (2015) Candidate Ideology and Electoral Success. Public Choice 176(1–2), 175–92Google Scholar
Hall, AB and Thompson, DM (2018) Who punishes extremist nominees? Candidate ideology and turning out the base in US elections. American Political Science Review 112(3), 509–24.Google Scholar
Hansen, JM (2016) Mobilization, participation, and political change. Party Politics 22(2), 149–57.Google Scholar
Hassell, HJG (2011) Looking beyond the voting constituency: A study of campaign donation solicitations in the 2008 presidential primary and general election. Journal of Political Marketing 10(1–2), 2742.Google Scholar
Hill, SJ and Huber, GA (2017) Representativeness and motivations of the contemporary donorate: Results from the merged survey and administrative records. Political Behavior 39(1), 329.Google Scholar
Hopkins, DJ and Noel, H (2022) Trump and the shifting meaning of “conservative”: Using activists’ pairwise comparisons to measure politicians’ perceived ideologies. American Political Science Review 116(3), 1133–40.CrossRefGoogle Scholar
Jacobson, GC (1989) Strategic politicians and the dynamics of U.S. House elections, 1946–86. American Political Science Review 83(3), 773–93.Google Scholar
Jacobson, GC (2015) It's nothing personal: The decline of the incumbency advantage in US House elections. Journal of Politics 77(3), 861–73.Google Scholar
Kedar, O (2005) When moderate voters prefer extreme parties: Policy balancing in parliamentary elections. American Political Science Review 99(2), 185–99.Google Scholar
Koger, G, Masket, S and Noel, H (2009) Partisan webs: Information exchange and party networks. British Journal of Political Science 39(3), 633–53.Google Scholar
Krehbiel, K (2007) Supreme court appointments as a move-the-median game. American Journal of Political Science 51(2), 231–40.Google Scholar
Kujala, J (2020) Donors, primary elections, and polarization in the United States. American Journal of Political Science 64(3), 587602.CrossRefGoogle Scholar
La Raja, R and Schaffner, B (2015) Campaign Finance and Political Polarization: When Purists Prevail. Ann Arbor: University of Michigan Press.Google Scholar
Lee, FE (2016) Insecure Majorities: Congress and the Perpetual Campaign. Chicago: University of Chicago Press.Google Scholar
Maestas, CD and Rugeley, CR (2008) Assessing the “experience bonus” through examining strategic entry, candidate quality, and campaign receipts in U.S. House elections. American Journal of Political Science 52(3), 520–35.Google Scholar
Magleby, DB, Goodliffe, J, and Olsen, JA (2018) Who Donates in Campaigns? The Importance of Message, Messenger, Medium, and Structure. Cambridge: Cambridge University Press.Google Scholar
Mann, TE and Ornstein, NJ (2016) It's Even Worse Than It Looks: How the American Constitutional System Collided with the New Politics of Extremism. New York: Basic Books.Google Scholar
McCarty, N, Poole, K, and Rosenthal, H (2006) Polarized America: The Dance of Ideology and Unequal Riches. Cambridge: MIT Press.Google Scholar
Meisels, M, Clinton, JD, and Huber, GA (2023) “Replication Data for: ‘Giving to the Extreme? Experimental Evidence on Donor Response to Candidate and District Characteristics.’” https://doi.org/10.7910/DVN/Q1X3RZ, Harvard Dataverse, V1.Google Scholar
Miratrix, LW et al. (2018) Worth weighting? How to think about and use weights in survey experiments. Political Analysis 26(3), 275–91.CrossRefGoogle Scholar
Patty, JW and Penn, EM (2019) Are moderates better representatives than extremists? A theory of indirect representation. American Political Science Review 113(3), 743–61.CrossRefGoogle Scholar
Porter, R and Steelman, TS (2023) No experience required: Early donations and amateur candidate success in primary elections. Legislative Studies Quarterly 48(2), 455466.Google Scholar
Rhodes, JH, Schaffner, BF, and La Raja, RJ (2018) Detecting and understanding donor strategies in midterm elections. Political Research Quarterly 71(3), 503–16.Google Scholar
Schlozman, KL, Verba, S, and Brady, HE (2012) The Unheavenly Chorus: Unequal Political Voice and the Broken Promise of American Democracy. Princeton: Princeton University Press.Google Scholar
Stone, WJ and Simas, EN (2010) Candidate valence and ideological positions in U.S. House elections. American Journal of Political Science 54(2), 371–88.Google Scholar
Theriault, SM (2006) Party polarization in the US Congress: Member replacement and member adaptation. Party Politics 12(4), 483503.Google Scholar
Thomsen, DM (2014) Ideological moderates won't run: How party fit matters for partisan polarization in Congress. Journal of Politics 76(3), 786–97.Google Scholar
Thomsen, DM (2017) Opting Out of Congress: Partisan Polarization and the Decline of Moderate Candidates. Cambridge: Cambridge University Press.Google Scholar
Thomsen, DM (2023) Competition in congressional elections: Money versus votes. American Political Science Review 117(2), 675691.Google Scholar
Utych, SM (2020) Man bites blue dog: Are moderates really more electable than ideologues? Journal of Politics 82(1), 392–6.Google Scholar
Woon, J (2018) Primaries and candidate polarization: Behavioral theory and experimental evidence. American Political Science Review 112(4), 826–43.Google Scholar
Figure 0

Table 1. Donor self-reported demographics

Figure 1

Table 2. Randomized vignette features

Figure 2

Figure 1. The proportion of donors wanting to give by candidate ideology. The horizontal axis is randomized candidate ideology, described relative to the donor's own positions. The vertical axis is the percentage of donors who indicated being very likely or almost certain to contribute to the candidate.

Figure 3

Figure 2. The average effect of vignette manipulations on the likelihood of contributing. Whiskers are 95 per cent confidence intervals. The outcome is 1 if ‘Very Likely’ or ‘Almost Certain’ to contribute and 0 otherwise. The intercept is 0.34.

Figure 4

Table 3. Predicted likelihood of giving

Figure 5

Figure 3. Self-reported motivation for giving. Weighted proportions of responses to, ‘which better characterizes your decision to contribute to a specific <> House candidate?’

Figure 6

Figure 4. The average effect of vignette manipulations on the likelihood of contributing by self-reported donation motivation. Hollow circles report prioritizing winning over issues when contributing, and filled circles prioritize issues over winning. Whiskers are 95 per cent confidence intervals. Outcome is 1 if ‘Very Likely’ or ‘Almost Certain’ to contribute, and 0 otherwise. Intercept is 0.38 for Winning and 0.35 for Issues.

Figure 7

Figure 5. The relationship between issue-based ideology and self-reported ideology by party. The horizontal axis is an absolute scaled issue-based PCA score. The vertical axis is the proportion of donors in the bin who identified as ‘Extremely Liberal’ or ‘Extremely Conservative’. Bin intervals span 0.1 on the absolute PCA scale.

Figure 8

Figure 6. The average effect of vignette manipulations on the likelihood of contributing by ideological extremism. Left: Extreme identified as extremely liberal or conservative. The baseline is 0.39 for Extreme and 0.33 for Non-Extreme. Right: Extreme falls in equivalent quantiles of issue-based PCA scores. The baseline is 0.40 for Extreme and 0.33 for Non-Extreme. Whiskers are 95 per cent confidence intervals. Outcome is 1 if ‘Very Likely’ or ‘Almost Certain’ to contribute, and 0 otherwise.

Figure 9

Figure 7. Effect of same-party candidate ideology by party and extremism. Models include all vignette manipulations. Left: Extreme identified as ‘Extremely Liberal’ or ‘Extremely Conservative’. Right: Extreme falls in equivalent quantiles of issue-based PCA scores. Whiskers are 95 per cent confidence intervals. Outcome is 1 if ‘Very Likely’ or ‘Almost Certain’ to contribute, and 0 otherwise.

Figure 10

Table 4. Party-interacted effect of candidate ideology on giving

Supplementary material: File

Meisels et al. supplementary material

Meisels et al. supplementary material
Download Meisels et al. supplementary material(File)
File 2.8 MB
Supplementary material: Link

Meisels et al. Dataset

Link