Overview

Our analysis proceeds in five main steps: (1) compiling and validating raw campaign finance data from the FEC; (2) identifying elected House representatives for each two-year cycle; (3) extracting member-to-member contributions via their affiliated committees; (4) constructing directed money flow networks (MFNs); and (5) fitting and analyzing Exponential Random Graph Models (ERGMs) to evaluate the role of exogenous and endogenous predictors. Detailed implementation procedures are provided in S1 Method.

Data sources and processing

We analyze campaign contributions between U.S. House members using publicly available data from the Federal Election Commission (FEC) covering seven two-year election cycles from 2009–2022 [14]. The analysis draws on four core FEC datasets: the Candidate Master file, which records each candidate’s name, identifier, party, and principal campaign committee; the Candidate—Committee Linkage file, which records all political committees officially associated with each candidate; the Committee Master file, which records every registered political committee and its classification; and the Contributions from Committees to Candidates and Independent Expenditure file, which reports all committee-to-candidate contributions and independent expenditures by political committees. Because these datasets are released separately for each election cycle, we compiled and processed 28 files in total. Each dataset was cross-checked by candidate and committee identifiers to ensure that all records referred to valid, active entities during the corresponding cycle.

For each two-year cycle, we next identified the 435 seats filled through regular general elections to the U.S. House of Representatives and matched them to the winning candidates who began serving in the subsequent Congress (for example, winners of the 2020 general election who entered the 117th Congress in January 2021). We focused on elected members at the end of each cycle because campaign funds raised and distributed during that period are intended to support service in the upcoming Congress. The elected members were identified using FEC election results [35] and verified against official House rosters published by the Clerk of the U.S. House [36]. In most cases, the two sources listed the same members. When discrepancies occurred—for instance, when a candidate died, resigned, or declined to assume office before being seated—we retained the individual who originally won the election rather than the successor who later filled the seat, since fundraising during a given cycle was directed toward the elected candidate’s campaign, not their replacement.

Finally, we identified contributions made by one member of Congress to another through their affiliated political committees using the Contributions from Committees to Candidates and Independent Expenditure file, which records all contributions by political committees to candidates and all independent expenditures made in support of or opposition to a candidate. Under FEC rules, members of Congress may contribute funds to one another through committees formally linked to them, including their principal or secondary campaign committees and their Leadership Political Action Committees (Leadership PACs) [37]. These committees are subject to contribution limits and public reporting requirements, and their affiliations are specified in the Candidate—Committee Linkage file. When a member was associated with multiple committees, all linked committees were aggregated by the member. We excluded personal donations made by one representative to another in an individual capacity because the dataset does not reliably identify the donor’s congressional status and would require uncertain name matching. Such personal donations are generally small and symbolic rather than strategic. For these reasons, the analysis includes only verified transfers made through officially registered political committees affiliated with elected House members.

Detailed procedures for data compilation and validation are provided in S1 Method.

Network construction

We constructed separate directed money-flow networks (MFNs) for Democratic and Republican members in each election cycle, based on both theoretical and empirical considerations. Contributions across parties are extremely rare—often absent in several cycles—whereas within-party transfers serve as the principal channel through which members coordinate fundraising, allocate resources, and organize electoral support [11,38,39]. Partitioning the data by party also improves comparability across cycles and reduces estimation difficulties that can arise from extreme sparsity, near-disconnected components, or unstable ERGM behavior, as discussed in the modeling section below [33,40–42].

Accordingly, all contributions between members of different parties were excluded. The few cross-party contributions observed involved members who subsequently changed party affiliation. At the time these contributions occurred, the members were still officially registered with their original party and are coded accordingly in the dataset. Such cross-party contributions constitute less than 0.50% of all records and are insufficient for reliable statistical estimation. Although the House permits candidates to serve as Independents, none were elected during the study period. Two members of the 116th Congress—Justin Amash (MI) and Paul Mitchell (MI)—were elected as Republicans and later declared Independent status. Each is coded as a Republican for the cycle in which they were elected, and neither sought nor won re-election as an Independent.

After applying these exclusions, we defined each party-specific MFN as a directed graph G_p = (V_p, E_p), where . V_p is the set of nodes representing all elected House members from party p in a given cycle. A directed edge from node i to node j is present if representative i, through an affiliated committee, contributed to representative j during that cycle. Together, the Democratic and Republican networks include all 435 elected members in each cycle. To maintain consistency across parties and time, all members, including those who neither gave nor received contributions, are retained as nodes in the corresponding networks.

Detailed procedures for network construction, including data validation and aggregation rules, are provided in the S1 Method.

Exponential Random Graph Models (ERGMs)

We model tie formation in the money-flow networks (MFNs) using Exponential Random Graph Models (ERGMs) [34]. ERGMs specify a probability distribution over all possible networks of a given size, where the likelihood of observing a particular network depends on specified structural and covariate-based features. Formally, the probability of observing a network y is:

where s(y) is a vector of sufficient statistics representing the network features included in the model, is the corresponding vector of parameters, and is a normalizing constant that ensures probabilities sum to one over all possible networks. Each term k corresponds to a network statistic s_k(y) with parameter , indicating the strength and direction of its association with the probability of observing y.

Factors in an ERGM are either exogenous, derived from attributes external to the network, or endogenous, defined by structural dependencies among ties (e.g., reciprocity, triadic closure, and in-degree centralization). In the MFNs constructed earlier, an example of an exogenous factor is state homophily, which tests whether two House members from the same state are more likely to exchange contributions than those from different states. This term is exogenous because it derives from a nodal attribute rather than the network structure, and dyadic because it is defined for pairs of nodes. Node-level factors are also included, such as a representative’s leadership status, which tests whether individual characteristics are associated with the probability of giving or receiving contributions. The ERGMs estimated in this study include both node-level and dyadic terms, exogenous and endogenous, to model the formation of money-flow networks across all seven election cycles.

Analogous to logistic regression, each coefficient represents the change in the log-odds of a tie forming associated with a one-unit increase in the corresponding statistic s_k(y), conditional on all other factors in the model. A positive value of indicates that the presence or greater magnitude of factor k increases the probability of observing a tie, whereas a negative value indicates a decrease in that probability. For example, a positive estimate for the state homophily term implies that representatives are more likely to contribute to co-partisans from the same state, whereas a negative estimate implies a tendency toward cross-state giving. Hence, ERGM coefficients can be interpreted as conditional log-odds ratios for tie formation, analogous to logistic regression parameters.

ERGMs have become increasingly common in political network analysis [41,43,44]. Much of this work has examined international relations [45,46] and policy networks [47–49]. For example, Henry et al. [49] used ERGMs to analyze how policy beliefs and social capital influence collaborative ties among policymakers. For an applied example, Desmarais and Cranmer (2012) introduce a resampling approach for MPLE uncertainty and a temporal ERGM, and apply it to U.S. Senate cosponsorship networks [50]. Despite these advances, to our knowledge, ERGMs have not yet been used to study contribution ties within the U.S. House of Representatives.

Endogenous terms

All models include four endogenous structural terms: edges, mutual, gwesp(0.50), and gwidegree(0.50), representing baseline tie propensity, reciprocity, triadic closure, and in-degree centralization, respectively. Each term quantifies a specific structural dependence that influences the probability of a tie. Below, we describe the interpretation of each term and its relevance for the money-flow networks (MFNs) analyzed in this study.

Edges: This term serves as the model intercept and represents the baseline log-odds of a tie forming between two randomly selected representatives, conditional on all other factors. A strongly negative coefficient, common in empirical applications, indicates a sparse network, meaning that the observed number of ties is far smaller than in a random graph of equivalent size. Because the edges coefficient is typically large and negative in such sparse settings, its corresponding odds ratio is well below one; other coefficients are therefore interpreted relative to this low baseline probability of a tie.

Reciprocity (mutual). This term tests whether directed ties are reciprocated. In MFNs, it measures whether a contribution from member i to member j increases the probability of a contribution from j to i. Prior studies show that reciprocal exchanges often occur in leadership contests or among active fundraisers seeking to reinforce alliances [51]. In this model, an odds ratio (OR) below 1 for the mutual term indicates lower conditional odds of reciprocation; the presence of a contribution in one direction reduces the conditional odds of a reciprocal contribution.

Triadic closure (gwesp(0.50)). This term tests whether two members are more likely to form a tie when both are linked to a common third member, thereby forming a closed triad. In MFNs, it measures whether representatives who share a contribution partner are also more likely to exchange contributions. The gwesp(0.50) statistic (geometrically weighted edgewise shared partners, with a fixed decay of 0.50) quantifies how the probability of a tie increases with the number of shared partners while allowing the marginal effect of each additional partner to decline geometrically [40]. Comparable transitive structures have been identified in legislative cosponsorship networks, where shared partners are associated with a higher probability of tie formation [50].

In-degree centralization (gwidegree(0.50)). This term measures concentration in the in-degree distribution (receiver centralization) by up-weighting configurations in which some members receive contributions from many co-partisans, with diminishing marginal influence as in-degree increases [40,52]. In sparse directed networks, this term helps represent heterogeneity in incoming ties. Because gwidegree is a curved term, its coefficient does not admit a simple one-to-one interpretation in odds-ratio units across networks; we therefore interpret its direction and relative magnitude qualitatively and report the estimate primarily as a structural control rather than a stand-alone substantive effect.

For curved terms (gwesp(0.50)/gwidegree(0.50)), ORs reflect a one-unit change in the change statistic and are not comparable to simple covariates; we report them for completeness and interpret direction/magnitude qualitatively.

Exogenous covariates

Exogenous covariates are included at both the node and dyadic levels to represent members’ institutional positions, electoral environments, and financial linkages within the campaign finance system.

State Homophily (dyadic). This binary variable equals one if two members represent the districts within the same state and zero otherwise. It tests whether representatives are more likely to contribute to representatives from their own state, given that members sharing a state often draw on overlapping donor networks, participate in joint fundraising events, and maintain shared political alliances [38,53].

Seniority (House Terms Served) (node). The data used for constructing this variable was obtained from the congress-legislators dataset maintained by The United States Project [54], which provides authoritative historical records of U.S. Congress members, including bioguide identifiers, term start dates, and complete service histories. This source was selected because of its comprehensive coverage, consistent maintenance, and direct linkage between Congressional records and FEC candidate identifiers, enabling reliable matching of representatives to their career service data. Each representative’s seniority is defined as the cumulative total number of House terms served throughout their entire congressional career, as of the election cycle in question. This count includes both consecutive and non-consecutive terms served, regardless of when they began service. For example, in the 2021–2022 cycle, a member who first served in 1955 and completed 27 terms (including any gaps in service) would have a seniority value of 27. Importantly, this definition is invariant across election cycles: seniority always represents cumulative career service measured on the same scale, differing only in realized values as members accrue additional terms over time. Separate sender and receiver effects (Seniority_Sender and Seniority_Receiver) test whether longer-serving members are more likely to give or receive contributions. Prior research shows that senior members generally have greater fundraising capacity and institutional visibility [16,50].

Leadership (node). Binary indicators identify whether a member holds or will hold a leadership position, including Speaker, party leader, whip, caucus or conference chair, and committee chair or ranking member. Leadership positions are determined in party caucus elections held after each general election, and members’ chances of selection generally depend on fundraising performance and established relationships from prior cycles [44,51]. Leaders were identified from official caucus and committee rosters and cross-checked against Congressional archival sources to confirm timing and officeholders.

Two sets of variables are included. Leadership(Next)_Sender and Leadership(Next)_Receiver indicate members who will hold a leadership position in the next Congress (the Congress convening after the election cycle), while Leadership(Current)_Sender and Leadership(Current)_Receiver indicate those who hold such positions during the same Congress. The “Next” indicators test whether members who later enter leadership contribute more before assuming office, whereas the “Current” indicators test whether sitting leaders contribute more or receive fewer contributions while in office. Because many leaders serve consecutive terms, the two sets are highly collinear; therefore, they are estimated in separate model specifications. We report results for the “Next” specification in the Results section, as most redistribution occurs in the cycle preceding leadership selection.

SuperPAC Activity (node). Four node-level covariates measure whether the number of SuperPACs making independent expenditures for or against a representative is associated with their contribution activity. Committees are classified as SuperPACs when identified by the FEC as independent-expenditure-only committees, consistent with Advisory Opinions 2010–09 and 2010–11. Support and opposition are identified via transaction type codes 24E (support) and 24A (oppose) in the independent-expenditure records (detailed in S1 Method). For each election cycle, we count the number of distinct SuperPACs making expenditures supporting or opposing each member. The covariates and test whether representatives supported by more SuperPACs are more likely to give or receive contributions, respectively, and and test the same for oppositional spending. These variables evaluate whether independent expenditures signal electoral strength or vulnerability in ways that influence internal fund allocation [17,55–57]. In a sensitivity analysis, we replace SuperPAC counts with total expenditure amounts (in $100,000) to assess robustness to the use of spending levels rather than counts.

Model variants

We estimate three ERGM specifications (one main model and two variants) that differ in their inclusion of exogenous covariates (Table 1). All three models include the four endogenous structural terms described above, and each also includes State Homophily as a control variable.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. ERGM model terms included in Models (1), (1a), and (1b). A ✓ indicates inclusion of a term.

https://doi.org/10.1371/journal.pcsy.0000104.t001

Model (1) serves as the main specification and incorporates factors for both House terms served and leadership. For each, we estimate separate sender and receiver effects, as the likelihood that a member gives contributions may differ from the likelihood of receiving them. Model (1) uses the next Congress leadership specification (Leadership(Next)_Sender and Leadership(Next)_Receiver), testing whether contribution activity is associated with attaining leadership in the next Congress.

Two sensitivity analyses extend this baseline model. Model (1a) replaces the leadership covariates with the current Congress specification (Leadership(Current)_Sender and Leadership(Current)_Receiver), testing whether sitting leaders differ from other members in how they give and receive contributions. Model (1b) retains the next Congress leadership terms but replaces the number of distinct SuperPACs with the total dollar amount of SuperPAC expenditures for each representative, testing whether the results are robust when using expenditure amounts instead of counts.

Unless otherwise noted, the Results section reports findings from Model (1). Results from Models (1a) and (1b) are referenced explicitly where relevant to assess the robustness of the main findings.

Estimation and analysis

All models were estimated using the ergm package in R [33] with Markov Chain Monte Carlo Maximum Likelihood Estimation (MCMLE). MCMLE is required for models with structural dependence terms such as mutual, gwesp, and gwidegree. The method approximates the full likelihood by drawing samples from the distribution of networks implied by the ERGM and computing parameters that maximize the expected log-likelihood.

Goodness-of-fit (GOF) diagnostics were computed for all fitted models using the ergm package. These diagnostics compare networks simulated from the fitted parameters with the observed data across key auxiliary statistics, including in-degree, out-degree, geodesic distance, and edgewise shared partners. Each model was evaluated using 100 networks simulated from the fitted parameters. Across election cycles, the fitted models reproduce the sufficient statistics closely; while some deviations appear in auxiliary distributions (e.g., higher-order shared-partner counts), the observed statistics generally remain within the range of simulated values [58]. We distinguish between in-model statistics—those explicitly included as sufficient statistics for parameter estimation—and auxiliary diagnostics, which characterize residual features not directly constrained by the model [59]. While our inferences rely on the stable recovery of the former, we report the latter in S1 Fig for transparency. To ensure model validity, we monitor for degeneracy, defined as the near-concentration of probability mass on extreme graphs or MCMC behavior that fails to explore the distribution, which was not observed in any of the final models reported here [60,61].

Our model specification is theory-driven: endogenous terms control for well-documented structural dependencies in directed networks, while exogenous covariates correspond directly to the hypotheses under investigation. We prioritize a consistent specification across all seven election cycles to preserve the cross-cycle comparability of effect estimates that is central to our analytical design. We nevertheless evaluated alternative structural specifications aimed at improving auxiliary GOF, including additional degree-based terms and specifications without mutual. Across cycles, these alternatives frequently produced poor MCMC mixing or unstable MCMLE updates and were therefore unsuitable for a comparative analysis requiring reliable estimation in every cycle. The retention of the current specification was not driven by selection on substantive results, but by the requirement that the model converge reliably across all seven cycles. We use GOF diagnostics to verify that the model adequately recovers its sufficient statistics and is free from degeneracy, rather than as a criterion for specification search. Details of these sensitivity checks are reported in S2 Table.

To summarize effects across election cycles, we conducted random-effects meta-analyses using the metafor package in R [62]. For each model term, we extracted estimated log-odds and standard errors across the seven cycles as inputs to independent random-effects models. A random-effects framework is adopted because true effect sizes may vary across cycles due to differences in political context, membership composition, and external influences [63]. For each term, the meta-analysis estimates the pooled effect size, standard error, 95% confidence interval, and p-value. In Results, we report odds ratios (exponentiated coefficients) with 95% confidence intervals.

Full parameter estimates, standard errors, and model diagnostics are provided in S5 Table–S10 Table. Detailed procedures for data construction and model estimation are provided in S1 Method.

Ethics statement

This study did not involve human subjects research as defined by institutional review board standards. All data analyzed were publicly available campaign contribution records from the Federal Election Commission. No identifiable private or sensitive information was collected, and the research adhered to all relevant ethical standards regarding the use of public data.

Structural characteristics of contributions between members

We examined four structural properties of contribution networks: edges (how frequently contribution ties formed overall), reciprocity (whether contributions tended to be returned), triadic closure (whether members who shared contribution partners were more likely to contribute to one another), and in-degree centralization (whether incoming ties were concentrated among a small number of recipients). Throughout the Results section, “contributing” and “receiving” refer to the presence of a contribution tie between two members (i.e., whether any contribution was made). For triadic closure and in-degree centralization, the odds ratios do not have the same straightforward interpretation as those for simple covariates such as seniority or leadership; we report them to compare direction and relative magnitude across cycles. The edge term implies very low baseline tie probabilities in both parties (see S5 Table–S6 Table), consistent with highly sparse networks.

Contribution ties between House members were infrequent in both parties and generally unidirectional. Among Republicans, reciprocity—the tendency for a member who received a contribution to give one in return, measured as the conditional odds of a reciprocal tie—declined from 0.70 (95% CI: 0.45–1.11) in 2009–10 to 0.03 (95% CI: 0.01–0.09) in 2021–22, with some fluctuation across intermediate cycles. Among Democrats, reciprocity remained consistently low, with odds ratios ranging from 0.13 to 0.34 across cycles (see S5 Table–S6 Table for cycle-level estimates).

Contribution ties among Republicans were more likely to form within locally interconnected groups—a structural property known as triadic closure, in which members who shared contribution partners had higher conditional odds of forming a tie with one another. This tendency was present in both parties but stronger among Republicans (pooled OR = 1.91, 95% CI: 1.63–2.23) than among Democrats (pooled OR = 1.36, 95% CI: 1.15–1.62).

In-degree centralization reflects whether incoming contribution ties were concentrated among a small number of recipients. (Values further below 1 indicate stronger centralization; values closer to 1 indicate weaker centralization.) Among Republicans, in-degree centralization weakened overall from 2009–10 (OR = 3.90 × 10⁻³, 95% CI: 2.60 × 10⁻³–5.80 × 10⁻³) to 2021–22 (OR = 1.08, 95% CI: 0.63–1.87), where the estimate was not statistically different from 1, with some fluctuation across intermediate cycles. Among Democrats, in-degree centralization remained strong across all cycles (OR consistently far below 1), indicating persistent concentration of incoming ties among a small subset of recipients. Full estimates are provided in S5 Table–S6 Table.

In summary, reciprocity among Republicans generally declined over time, and incoming ties shifted from being concentrated among a few recipients toward a more even distribution. Republicans also showed stronger evidence of triadic closure. Among Democrats, ties remained infrequent and mostly unidirectional, with persistent concentration of incoming ties among a small subset of recipients across cycles.

Effects of Seniority and leadership status on internal contributions

Seniority (measured as cumulative House terms served) showed different associations with giving and receiving across parties. Among Democrats, each additional term was associated with higher conditional odds of contributing (OR = 1.05, 95% CI: 1.04–1.06, p < 0.001) and lower odds of receiving (OR = 0.95, 95% CI: 0.93–0.97, p < 0.001). Among Republicans, seniority was also negatively associated with receiving (OR = 0.95, 95% CI: 0.93–0.96, p < 0.001), but the pooled association with contributing was not statistically significant (OR = 0.99, 95% CI: 0.96–1.03, p > 0.05). Cycle-specific sender estimates were generally below 1 from 2009–10 to2019–20 and exceeded 1 in 2021–22, indicating heterogeneity over time in the seniority–giving association.

Beyond seniority, leadership status—defined as holding a formal position in the next Congress (Speaker, majority or minority leader, whip, caucus or conference chair, or committee chair or ranking member)—was strongly associated with giving and receiving, particularly among Republicans. Republican leaders had higher odds of contributing than non-leaders (OR = 2.11, 95% CI: 1.81–2.46, p < 0.001) and lower odds of receiving (OR = 0.71, 95% CI: 0.66–0.77, p < 0.001). Among Democrats, leaders also had higher odds of contributing on average (OR = 1.49, 95% CI: 1.07–2.06, p = 0.02), but showed no significant pooled association with receiving (OR = 0.96, 95% CI: 0.87–1.05, p > 0.05).

A key question is whether members gain leadership positions by building support through contributions, or whether leaders use contributions to sustain influence once in office. To address this, we re-estimated Model (1a) with leadership status defined for the same Congress in which contributions occurred, so that leadership reflects positions held at the time of contributing. While this does not establish causality, the results show that leaders in both parties had higher conditional odds of contributing. Republican leaders had lower odds of receiving (OR = 0.63, 95% CI: 0.49–0.82, p < 0.001) and higher odds of contributing (OR = 2.26, 95% CI: 1.74–2.93, p < 0.001). Democratic leaders had higher odds of contributing (OR = 1.45, 95% CI: 1.17–1.80, p < 0.001) and did not differ significantly in receiving (OR = 1.01, 95% CI: 0.95–1.07, p > 0.05). Full results for both specifications are reported in S7 Table–S8 Table.

As a complementary descriptive analysis, we compared centrality distributions by leadership status (Fig 3A–B). Fig 3C illustrates within-cycle directionality: leaders frequently give to non-leaders, whereas contributions from non-leaders to leaders are rare.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. A comparison of centrality distributions and network structure for leaders and non-leaders.

(A) Comparison of out-degree (top pane) and in-degree (bottom pane) centrality distributions between leaders and non-leaders for each election cycle. Using a one-sided Mann-Whitney U test, *, **, and *** indicate a significant shift in the centrality distributions between leaders and non-leaders at the 10%, 5%, and 1% levels, respectively. In each boxplot, outliers were excluded, and whiskers extend to 1.5 times the interquartile range from the lower and upper quartiles. Detailed statistical test results underlying these significance markers are provided in S3 Table. (B) The same data as in (A), but with the centrality distributions shown separately for representatives of the Democratic (DEM) and Republican (REP) parties. Similar to (A), we compute and report significant differences in the centrality distributions between out-degree and in-degree for leaders and non-leaders, but for each party separately. Statistical tests for party-specific comparisons are reported in S3 Table. (C) A network illustration of contributions between Republican representatives during the 2013–2014 cycle (left pane) and Democratic representatives during the 2019–2020 cycle (right pane), when they each controlled the House. The left figure in each pane shows contributions from leaders to non-leaders, and the right figure shows contributions from non-leaders to leaders. Nodes not participating in contributions (singletons) are included only if they are leaders. Grey nodes represent non-leaders, while red and blue nodes represent leaders.

https://doi.org/10.1371/journal.pcsy.0000104.g003

Effects of SuperPAC activity on internal contributions

Representatives targeted by a larger number of distinct SuperPACs making independent expenditures—supportive or oppositional—had lower conditional odds of contributing and higher conditional odds of receiving. Substantively, members targeted by more SuperPACs were more likely to receive contributions from co-partisans and less likely to contribute to others. For Democrats, each additional supportive SuperPAC was associated with a 47% decrease in the odds of contributing (OR = 0.53, 95% CI: 0.32–0.87, p = 0.01) and a 25% increase in the odds of receiving (OR = 1.25, 95% CI: 1.12–1.41, p < 0.001). For Republicans, each additional supportive SuperPAC corresponded to a 24% decrease in the odds of contributing (OR = 0.76, 95% CI: 0.72–0.79, p < 0.001) and a 19% increase in the odds of receiving (OR = 1.19, 95% CI: 1.14–1.24, p < 0.001).

Counts of distinct SuperPACs making oppositional expenditures were likewise associated with lower contributing and higher receiving in both parties. For Democrats, each additional opposing SuperPAC was associated with a 46% decrease in the odds of contributing (OR = 0.54, 95% CI: 0.38–0.77, p < 0.001) and a 41% increase in the odds of receiving (OR = 1.41, 95% CI: 1.24–1.61, p < 0.001). For Republicans, the corresponding estimates were a 42% decrease in the odds of contributing (OR = 0.58, 95% CI: 0.42–0.79, p < 0.001) and a 31% increase in the odds of receiving (OR = 1.31, 95% CI: 1.22–1.41, p < 0.001).

We find no direct correspondence between representatives’ likelihood of giving or receiving contributions and the overall rise in SuperPAC activity during the study period. In fact, these internal contribution rates declined slightly in the final cycle even as SuperPAC activity remained high over the study period. Fig 4 shows that in 2009–10, only about 16–17% of members in each party received SuperPAC support, but this proportion grew rapidly, exceeding 80% for Republicans by 2015–16 and nearly 90% for Democrats by 2017–18. Since then, more than half of all members in both parties have consistently received SuperPAC support. Fig 4B further distinguishes between opposing and supporting expenditures. Since 2011–12, Democrats have consistently had more SuperPACs supporting them than opposing them—a gap that has widened in recent cycles. In contrast, Republicans experienced more SuperPACs opposing them than supporting them in 2009–10, 2011–12, and 2019–20.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. (A) Counts of “Super” Political Action Committees (SuperPACs) making independent expenditures for Democratic (DEM) and Republican (REP) representatives in each cycle.

The total number of elected representatives from each party is shown at the top of each bar. Error bars represent 95% confidence intervals (CIs) for the share of representatives in each party for whom at least one SuperPAC made expenditures (nonzero SuperPAC count). Using a two-sample z-test of proportions, we find that starting from the 2013–2014 cycle, the proportions of Democratic and Republican representatives with nonzero SuperPAC counts are significantly different at the 99% confidence level (p < 0.01), indicated by ***. In the other two cycles, significant differences were not observed at the 90% confidence level or above. Detailed breakdowns by party and SuperPAC count and the underlying z-test statistics are provided in S4 Table. (B) Disaggregation of SuperPAC expenditures into those supporting or opposing candidates. For each election cycle, the number of SuperPACs spending money to support or oppose Democratic (DEM) and Republican (REP) representatives is displayed separately. Different textures differentiate between support and opposition expenditures. The corresponding tabular data stratified by expenditure type and party affiliation are provided in S4 Table.

https://doi.org/10.1371/journal.pcsy.0000104.g004

As a sensitivity analysis, we re-estimated Model (1b) using total SuperPAC expenditure (rescaled to $100,000 units) instead of counts of distinct SuperPACs and pooled the estimates across cycles. The expenditure-based results matched the main specification in direction. For both parties, higher SuperPAC expenditure was associated with higher odds of receiving (Democrats: OR_support = 1.07, 95% CI: 1.065–1.073; OR_oppose = 1.03, 95% CI: 1.024–1.032; Republicans: OR_support = 1.05, 95% CI: 1.04–1.07; OR_oppose = 1.07, 95% CI: 1.068–1.079). Higher expenditure was also associated with lower odds of contributing (Democrats: OR_support = 0.81, 95% CI: 0.79–0.82; OR_oppose = 0.99, 95% CI: 0.987–0.993; Republicans: OR_support = 0.67, 95% CI: 0.57–0.78; OR_oppose = 0.94, 95% CI: 0.937–0.948). All pooled associations were statistically significant (p < 0.001). As in the main specification, the direction was consistent across cycles while magnitudes varied. Full results for this analysis are reported in S9 Table–S10 Table.