1.1. Classifying early money measures

In Table 1, we classify existing approaches for operationalizing early money into two groups of measures.Footnote ² We identify a first group of “election-centered” measures that operationalize early fundraising based on an election-specific end date; all funds raised before that date are considered early. Some election-centered measures base their fundraising end date on months elapsed within an election cycle (e.g., Green and Krasno, Reference Green and Krasno1988).Footnote ³ Other measures base their fundraising end date on contest-level electoral timing, for example, ending some number of months before the primary (e.g., Leal, Reference Leal2003).Footnote ⁴ We identify a second group of “candidate-centered” measures that operationalize early fundraising based on a candidate-specific start date; all funds raised for a window after that date are considered early. The start date for early fundraising is unique to each candidate because it is based on that candidate’s fundraising behavior, such as the date she registers her campaignFootnote ⁵ with the Federal Election Commission (FEC) (e.g., Bonica, Reference Bonica2017, Reference Bonica2020) or the date she receives her first itemized donationFootnote ⁶ (e.g., Bonica and Grumbach, Reference Bonica and Grumbach2022; Porter and Steelman, Reference Porter and Steelman2023). A candidate’s early fundraising window is often specified as lasting 60 or 90 days after this start date; alternatively, some other work considers all fundraising reported in a candidate’s first quarterly report filed with the FEC to be early (e.g., Biersack et al. Reference Biersack, Herrnson and Wilcox1993; Thomsen, Reference Thomsen2023).Footnote ⁷

Table 1.

Summary of existing measures for early campaign fundraising

1.2. Variation in early fundraising estimates across measures

In Figure 1, we assess to what extent estimates of early fundraising vary within and across election- and candidate-centered measurement typologies. Average cross-measure correlations for early fundraising include all candidates who ran in elections for the U.S. House of Representatives from 2010 to 2020 (N = 6,001).Footnote ⁸ All measures are produced using itemized receipts provided by OpenSecrets and aggregate unitemized contribution totals drawn from FEC quarterly reports; for more detail on data preparation, see Appendix Section B. Figure 1(a) includes cross-measure correlations for incumbents who ran for reelection; Figure 1(b) includes cross-measure correlations for non-incumbent challengers. For ease of presentation, we restrict Figure 1 to include the most common early fundraising measures found in the extant literature, described in greater detail in Appendix Table A1.

Figure 1.

Average pairwise correlations for popular early money measures. (a) Incumbents and (b) non-incumbents.

Note: Standard errors for pairwise correlations present in Figure 1 are all < 0.001.

Average pairwise correlations in Figure 1(a) and (b) are consistently stronger within measurement typology than across typologies.Footnote ⁹ For instance, looking at candidate-centered approaches, the three measures presented (fundraising 90 days after a candidate registers with the FEC, fundraising 90 days after a candidate receives her first itemized contribution, and fundraising reported in a candidate’s first FEC quarterly report) have an average correlation of 0.75 for incumbents and 0.86 for non-incumbents.Footnote ¹⁰ Pairwise correlations between candidate-centered and election-centered measures are more modest ranging from 0.62 to 0.84 for incumbents and 0.39 to 0.59 for challengers.

We trace these weak correlations to the conceptual differences between candidate- and election-centered measurement approaches, particularly how these strategies define the early fundraising window. To illustrate, Figure 2 depicts fundraising window lengths and early fundraising estimates across various early money measures for Carolyn Maloney (Figure 2(a)) and Suraj Patel (Figure 2(b)), who both ran in the Democratic primary for New York’s 12th Congressional District in 2020. This primary was closely contested, with district incumbent Maloney besting Patel for the Democratic nomination by only 3.4 percentage points. The x-axis in Figure 2 is an election timeline expressing the number of days until New York’s Democratic primary election, which was held on June 23rd, 2020. Horizontal bars reflect the length of early fundraising windows across the same measures from Figure 1. Fundraising totals for each measure are included to the right of each horizontal bar.

Figure 2.

Timing of fundraising windows and estimates for early fundraising across extant measurement strategies. (a) Carolyn Maloney (D-NY) and (b) Suraj Patel (D-NY).

Note: Figure depicts the start and end date of early fundraising for Carolyn Maloney (Figure 2(a)) and Suraj Patel (Figure 2(b)) in the NY-12 Democratic primary in 2020 for the early money measures present in Figure 1. All fundraising totals are produced using itemized fundraising data from OpenSecrets and unitemized fundraising totals reported in FEC quarterly reports. See Appendix Section B for greater details.

Turning first to Figure 2(a), election- and candidate-centered measures begin during similar periods of the election cycle—with the exception of the campaign registration-based measure.Footnote ¹¹ This symmetry in start dates across fundraising measures aligns with expectations; incumbents consistently begin fundraising within the first few months of an election cycle.Footnote ¹² Challengers are less consistent as to when they begin fundraising, as is evident in Figure 2(b). Because Patel and Maloney ran in the same primary election contest, their election-centered fundraising windows span the same campaign period. However, the fundraising windows for Patel’s candidate-centered measures fall much later in the election cycle than Maloney’s because Patel did not begin fundraising until September 2019. Consequently, election- and candidate-centered measurement approaches produce disparate estimates for Patel’s early fundraising. Additionally, Patel and Maloney’s fundraising windows cover fundamentally different windows of the election cycle, despite these two candidates appearing in the same race.

The examples presented here illustrate how differences in operationalization between election- and candidate-centered measurement approaches produce differing estimates of early fundraising. Appendix Figure A2 presents the average lengths and timing for fundraising windows across all incumbent and non-incumbent candidates who ran for the U.S. House of Representatives from 2010 to 2020. For incumbent candidates, average fundraising windows for candidate-centered measures fall within the windows of election-centered measures; variation across measurement typologies is primarily rooted in the length of fundraising windows (e.g., a 90-day candidate-centered window vs. a 12-month election-centered window). For non-incumbent candidates, the fundraising windows for election-centered and candidate-centered measures do not consistently overlap as they do for incumbents. These differences produce the modest cross-typology fundraising correlations for non-incumbents found in Figure 1(b).

1.3. How measurement choice impacts substantive conclusions

Different approaches for measuring early fundraising can produce different conclusions about money’s influence in elections. In Figure 3, we estimate a series of logistic regressions assessing candidate success in congressional primaries. Units of analysis include all candidates who ran in open seat, partisan primary elections for the U.S. House of Representatives from 2010 to 2020. We restrict our analysis to candidates running in open seats because the connection between fundraising and success is most pronounced in these contests, as no incumbent is running for reelection (Thomsen, Reference Thomsen2023). Moreover, the extant work about early fundraising focuses most often on these kinds of contests (e.g., Bonica, Reference Bonica2017, Reference Bonica2020; Porter and Steelman, Reference Porter and Steelman2023). Each model presented in Figure 3 employs a different estimation strategy for candidate-centered and election-centered early fundraising as a primary explanatory variable. We include other control variables in our models, including indicators for candidate partisanship, prior political experience, primary type, district seat safety, and the number of competitors in the race, as well as fixed effects by year and random effects by district partisan primary.

Figure 3.

Marginal differences in pairwise comparisons for early money coefficient in primary election success models.

Note: Estimated logistic regressions predict the effect of early fundraising on primary election success for U.S. House candidates running from 2010 to 2020. Each success model presented employs a different early money measure from Figure 1. The x-axis indicates the focal model used in pairwise model comparisons. The y-axis reflects the difference in effect size between the focal model and the comparison model.

We aim to assess how statistical inferences about the relationship between early fundraising and primary election success change conditional on the choice of early money measure. Because estimates for total early fundraising take on different ranges of values across each mode for measurement, we standardize all early money measures used in our models to have a mean of zero and a standard deviation of 1. Additionally, we follow the guidance of Mize et al. Reference Mize, Doan and Long(2019) to recover the covariance matrices for marginal differences to produce valid confidence intervals for these quantities. In this approach, we stack models using seemingly unrelated estimation to combine estimates from multiple models. We estimate a stacked model for each set of the pairwise model comparisons, producing a total of 10 model outputs. The coefficients estimated in these stacked models are identical to those produced by individual models; standard errors vary between stacked and individual models.

The marginal differences in early money coefficients across model pairwise comparisons are presented in Figure 3 with 95% confidence intervals. The x-axis indicates the focal model used in pairwise comparisons.Footnote ¹³ The y-axis indicates the difference in effect sizes across pairwise model comparisons. Confidence intervals that do not intersect with the horizontal gray line indicate the differences across pairwise model comparisons (i.e., marginal differences) are statistically significant. Plotted coefficients denoted with outlined shapes are election-centered measures; plotted coefficients denoted with shaded shapes are candidate-centered measures.

We find statistically significant differences across five of the six unique pairwise comparisons made in Figure 3 between election and candidate-centered measures. To place differences between measures in substantive terms, we turn to predicted probabilities. When employing a candidate-centered measure of early money, such as all fundraising reported in a candidate’s first FEC quarterly report, an increase from one standard deviation below the mean to one standard deviation above the mean increases the predicted probability of success from 0.08 to 0.56. Conversely, when employing an election-centered measure of early money, such as all fundraising garnered 9 months before a candidate’s primary, an increase in fundraising from one standard deviation below the mean to one standard deviation above the mean increases the predicted probability of success from 0.14 to 0.43. The difference-in-difference between these two measures is 0.19.Footnote ¹⁴

Scholars have traditionally treated early money measures as interchangeable, offering little theoretical rationale for their measurement choices under the assumption that extant strategies produce similar fundraising estimates. Our preceding analyses indicate that this ambivalence is a threat to inference. We demonstrate that election-centered and candidate-centered measurement approaches produce differing estimates of early fundraising. Moreover, we show that these measurement differences have downstream consequences, impacting the substantive conclusions drawn from statistical analyses. To address this disconnect, we argue that scholars should leverage conceptual differences between election- and candidate-centered approaches to better align their empirical measurement of early fundraising with their theoretical motivations.

2.1. Theoretical motivations for measuring early fundraising

A candidate-centered measure is well-suited to address research questions centered on the initial financial backing a candidate’s campaign receives. Recall that this family of approaches uses the start of a candidate’s fundraising campaign to define the beginning of her early fundraising window. As such, a candidate-centered approach should reflect those first receipts raised by a candidate. This approach stands in contrast to an election-centered strategy that defines a candidate’s fundraising windows based on the election calendar, which may fail to capture a candidate’s initial fundraising.Footnote ¹⁵ By capturing initial donations, a candidate-centered measure should reflect the “seed” funding that helps to get a candidate’s campaign off the ground (Jacobson, Reference Jacobson1980; Biersack et al., Reference Biersack, Herrnson and Wilcox1993). The quicker a candidate can raise seed money, the sooner she can put these resources to use and make herself known to potential voters and donors. The political action committee EMILY’s List embodies in its name the principle that early contributions have an outsized impact on future fundraising, affirming that “Early Money Is Like Yeast” because it “makes the dough rise.” To that end, a body of research seeks to measure the qualities or quantity of a candidate’s early fundraising to identify the political influencers providing financial support to candidates in the “invisible” or “money” primary. For example, Bonica and Grumbach Reference Bonica and Grumbach2022 use early campaign contributions to show that the “selectorate” of donors providing seed funding to candidates skew older and ideologically conservative. Porter and Steelman Reference Porter and Steelman(2023) similarly employ early money to assess the kinds of political influencers active in the “invisible primary” and demonstrate that early monetary support from out-of-district donors is especially predictive of future fundraising success.

An election-centered measure is well-suited to addressing research questions about the interplay between early campaign fundraising and electoral competition. Recall that this family of approaches holds the fundraising window constant for all candidates running in the same electoral contest, allowing for direct comparisons across competitor fundraising at a fixed point in time. In this way, an election-centered approach encapsulates the dynamic of a candidate tailoring her behavior to her competitors’ fundraising performance. This stands in contrast to a candidate-centered strategy where there is no guarantee that early fundraising estimates reflect funds raised at the same point in the election cycle for all candidates in a race.Footnote ¹⁶ An election-centered measure should capture the kind of fundraising data that journalists, pollsters, and voters use when making cross-candidate comparisons to glean who is “ahead” or “behind” in a race (Raja, Reference Raja2007; Anderson et al., Reference Anderson, Burden, Butler, Harbridge-Yong and Ryan2023). This dynamic is the subject of research assessing how fundraising disparities influence electoral competition. Epstein and Zemsky Reference Epstein and Zemsky(1995), for instance, argue that challenger entry decisions are, in part, a function of incumbents’ financial signaling—where a strong show in early incumbent fundraising will deter challenger entry. Similarly, Thomsen Reference Thomsen2023 shows that congressional candidates today are increasingly likely to drop out of elections when they are at an early fundraising disadvantage relative to their primary election competitors.

2.2. How to measure candidate-centered early fundraising

Because a candidate-centered measure of early money seeks to measure the initial contributions a candidate receives, the quantity of interest is raw dollars raised by a candidate.Footnote ¹⁷ Moreover, for a candidate-centered measure to aptly capture a candidate’s initial “seeds” of fundraising, it should not reflect a protracted period of her fundraising campaign. Based on empirical conclusions drawn from Appendix Sections C and D, we suggest that researchers define a fundraising window as beginning when a candidate’s first itemized contribution is received and set a window length of 90 days. This start date ensures a candidate’s earliest receipt is captured in early fundraising measurement.Footnote ¹⁸ Additionally, using a 90-day fundraising window ensures that a candidate’s first FEC quarterly reporting deadline is included in candidate-centered early money measurement. As shown in Appendix Figure A6, deadlines for reporting fundraising to the FEC clearly influence the data generation process for campaign receipts. Extant work finds that candidates are mindful of the vital role fundraising plays in shaping perceptions of campaign viability and, accordingly, ratchet up their fundraising efforts proximate to FEC reporting deadlines (Smidt and Christenson, Reference Smidt and Christenson2011; Magleby et al., Reference Magleby, Goodliffe and Olsen2018). Candidate-centered measures that do not capture a candidate’s first FEC reporting deadline will under-report early fundraising.

When employing a candidate-centered approach, researchers should impose a cutoff date for early money measurement such that funds raised after a specific date are no longer deemed early.Footnote ¹⁹ A sizable minority of candidates begin fundraising only a short time before their election; over 10% of all non-incumbent candidates who ran for the U.S. House between 2010 and 2020 began fundraising within 90 days of their primary.Footnote ²⁰ It is undoubtedly the case that campaign fundraising garnered months before the primary versus mere days before the election have differentiated impacts on campaign success. It is important to explicitly acknowledge, however, that imposing such a cutoff will result in some candidates possessing observed fundraising periods that include fewer than 90 days. For instance, if a candidate begins raising money a month prior to a selected cutoff date, their candidate-centered fundraising will only capture a 30-day period.

We propose using state-level filing deadlines to inform a cutoff for early fundraising. Filing deadlines fall at least 2 months before each state’s primary elections. Over 95% of non-incumbent general election winners received their first itemized contribution on or before their state’s filing deadline. Importantly, a cutoff specification employing filing deadlines accounts for state-level variability in primary election timing. Specifying a fixed fundraising cutoff across all elections will not account for this variability.Footnote ²¹ To illustrate, in the context of congressional elections, primaries are held across 8 months, the earliest occurring in February of the election year and the latest in September.Footnote ²² Specifying an early fundraising cutoff of January 1st would fall only a month from February primaries but over 250 days from September primaries. An early money measure capturing these different campaign phases for February and September primaries will surely fail to measure candidate fundraising consistently. One potential limitation in employing state-level filing deadlines as an early fundraising cutoff is that these deadlines vary across states, inducing disparate fundraising window lengths. Importantly, though, we find no relationship between the number of days in the window of our proposed measure and the number of days before the election that filing deadlines occur. This provides empirical evidence that variation in the timing of state-level filing deadlines will have limited influence on cross-candidate comparisons. For reference, we graphically depict our proposed candidate-centered measure of early money in Appendix Figure A5 using the same motivating example as Figure 2.

2.3. How to measure election-centered early fundraising

Given that an election-centered measure of early money is interested in a candidate’s comparative fundraising performance at a fixed point, our quantity of interest here is not raw dollars raised but, instead, relative fundraising. Research questions seeking to assess whether a candidate is over- or under-performing early fundraising expectations should rely on Bonica’s (Reference Bonica2020) approach for calculating normalized fundraising relative to contest-level averages. For candidate i in primary contest j at time t, let f_ijt be a candidate’s fundraising total at a fixed point in the primary and n_jt be the number of candidates competing in the primary contest at that time, such that $\widehat{Normalized \ Fundraising} = \frac{f_{ijt}}{(\sum \frac{f_{jt}}{n_{jt}})}$. Using this approach, a value of greater than one indicates that a candidate is raising more than the contest average in early fundraising; a value of less than one indicates that a candidate is raising less than the contest-level average. Research questions seeking to assess how a candidate’s early money compares to a specific competitor should rely on Thomsen’s (Reference Thomsen2023) approach for calculating a fundraising share. For candidate i in primary contest j at time t, let f_ijt be a candidate’s fundraising total at a fixed point in the primary and m_jt be the fundraising total for the comparison candidate (e.g., lead fundraiser or incumbent) at the same fixed point, such that $\widehat{Fundraising \ Share} = \frac{f_{ijt}}{m_{jt}}$. A value of greater than one indicates that a candidate is raising more than the comparison candidate in early fundraising; a value of less than one indicates that a candidate is raising less than the comparison candidate.

The date a researcher defines as the end of election-centered early fundraising should be tied to her question of interest.Footnote ²³ For example, following Epstein and Zemsky Reference Epstein and Zemsky(1995), if seeking to understand how the size of an incumbent’s campaign war chest deters challenger entry, a fundraising end date early in the election season should be specified. This same end date would not be suitable for examining a different kind of candidate–fundraising relationship, such as the dropout decisions of candidates who are poor fundraisers (Thomsen, Reference Thomsen2023). When choosing an early fundraising end date, it is also important to be mindful of state-level variability in the timing of elections. Researchers should specify a fundraising end date tied to contest-level electoral conditions. For example, Leal Reference Leal(2003) considers receipts early if they are raised 9 months before the primary.Footnote ²⁴ It is important to explicitly note that, unlike our candidate-centered approach proposed above, this approach uses the same observation period across candidates in the same race because it is based on election timing rather than candidate fundraising behavior.

For many applications, the filing deadline for a candidate’s election is an appropriate end date for early fundraising. All federal, state, and local elections require candidates to submit paperwork by a specific date to appear on the ballot. Once a state’s filing deadline passes, the ballot is set, and no more candidates can enter the race. A filing deadline early fundraising cutoff will, therefore, include all ballot-eligible candidates; this is desirable if a researcher seeks to use an election-centered measure to make cross-candidate fundraising comparisons in advance of the election. It is worth noting that while the filing deadline for some states is closer to the primary election than others, this is not problematic for an election-centered measure. Our election-centered measure is calculated as a share rather than raw totals. Therefore, the measure allows for valid comparisons across states given all candidates within a race are evaluated on the same criteria, and all election-centered measures reflect a comparable share of fundraising. For reference, we graphically depict our proposed election-centered measure of early money in Appendix Figure A5 using the same motivating example as Figure 2.

2.4. Data considerations for early fundraising measurement

Campaign contributions data are often reported in two formats: raw itemized contributions from bulk data files and aggregated totals from fundraising disclosure reports. Using each of these sources for data involves trade-offs. The benefit of employing raw contributions is that these data allow for the fine-grained measurement of early money. Because receipt dates are available for itemized contributions, fundraising windows can be precisely and flexibly specified to include contributions garnered across specific dates. The drawback of working with raw data files is that it is computationally intensive. Raw contributions data can be downloaded in bulk from the FEC website, or specific data requests can be queried to the FEC API. Resulting files often contain tens of millions of observations; processing these data and manually calculating candidates’ contribution totals requires data-wrangling skills and knowledge of campaign finance reporting requirements. Additionally, consistent receipt dates are unavailable for unitemized contributions—even in a raw data format (Alvarez et al., Reference Alvarez, Katz and Kim2020). If researchers wish to incorporate small-dollar donations into their early money estimates, they must still rely on aggregate totals provided in disclosure reports.Footnote ²⁵

The benefit of employing contributions data from disclosure reports is tied to ease of use. Accessing disclosure reports (e.g., quarterly reports to the FEC) is straightforward, and pre-aggregated contribution totals are reported in disclosure reports—no manual calculations are required. Moreover, at the federal level, FEC quarterly reports include itemized and unitemized receipts, giving a complete picture of candidate fundraising across a consistent window. The drawback of employing these data is that fundraising totals are pre-aggregated across a fixed reporting period that may not align with a researcher’s specified early fundraising window. Moreover, disclosure reports may not reflect a consistent campaign period (e.g., 90 days in FEC quarterly reports). For example, in Appendix Section E, we demonstrate that about 6% of U.S. House candidates who ran between 2010 and 2020 had a first FEC quarterly reporting period that spanned more than 90 days, and 21% had a first reporting period that spanned less than 45 days.

To aid researchers, we make several resources available to measure early money. First, we provide a new dataset of state filing deadlines for U.S. congressional elections between 2004 and 2020. Second, we produce and make available off-the-shelf measures for early fundraising in U.S. House elections. For our candidate-centered measure, we calculate early money raised 90 days after receipt of a candidate’s first itemized contribution. We cut off a candidate’s early fundraising window if it intersects with her state’s filing election deadline. For our election-centered measure, we calculate early fundraising totals with an end date based on contest-level filing deadlines. Additionally, we calculate and make available election-centered early fundraising totals at the 3-, 6-, and 9-month marks from a candidate’s primary election. All estimates are produced using raw data on itemized campaign receipts and aggregated unitemized receipt totals. For greater detail, see Appendix Section B. Estimates are created based on the recommendations outlined in our preceding discussion. For reference, our candidate-centered estimates correlate with extant candidate-centered measures from Figure 1 at 0.86 and the election-centered estimates correlate with extant election-centered measures at 0.93.