Literature

The literature on firms’ attempts to influence public policy is vast. The mainstay of this literature focuses on campaign contributions and, more recently, on lobbying. The literature on campaign contributions focuses on targets of campaign contributionsFootnote ¹⁰ and outputs of those efforts, including legislator effortFootnote ¹¹ and votes.Footnote ¹² Likewise, the literature on informational lobbying focuses on the targets of those effortsFootnote ¹³ and the effects of lobbying on policy outcomes.Footnote ¹⁴

This article examines an alternative mechanism that firms use to seek favorable public policy—EIAs. In the management literature, the idea that firms can strategically use operational activities to influence policy outcomes emanates from the work of Baron,Footnote ¹⁵ who introduced the idea of nonmarket strategy as a tool available to the firm to influence profitability. From a strategic management perspective, the literature has been developed by Funk and Hirschman,Footnote ¹⁶ who argue that a firm's market decisions could be used for political ends.Footnote ¹⁷ More formally, Bonardi and UrbiztondoFootnote ¹⁸ develop a specific mechanism of EIAs with a theoretical model of asset freezing, where firms time their investment and employment decisions in a manner favorable to local politicians’ reelection cycles. Raiha examines the theoretical microfoundations for EIAs in modeling the informational channels through which the economic/operational decisions of firms can provide political benefits to politicians and be used as an influence tool.Footnote ¹⁹ The model shows how a firm's strategy choices affects the state of the local economy and, in turn, the evaluations that voters make of the performance of an officeholder. Because of this, firms can use this capability to extract subsidies and policy favors from incumbent officeholders.

Empirically, many papers have demonstrated that firms in both developed and developing countries do seem to time their investment and employment decisions in ways that are consistent with electoral cycles to support of the firms’ favored legislators.Footnote ²⁰ There is also an emerging literature on the political geography of firms as it relates to EIAs. Pang et al. examine how firms change their business strategies in response to changes in representation on the House Transportation Committee.Footnote ²¹ They find that changes in committee membership induced airlines to offer more air routes within or nearby to committee members’ districts.Footnote ²² Bisbee and You study the 300 largest manufacturing firms in the United States and found that such firms open subsidiaries in congressional districts that are electorally competitive.Footnote ²³ Barber and Blake find that firms locate new establishments in districts that are politically similar to areas where the firm currently operates.Footnote ²⁴

In this article, we extend this literature on EIAs to examine a particular supply side mechanism for public policy—the legislative committee structure of the United States—and how EIAs can utilize committee membership in the US House and Senate to obtain public policy. Our article complements and extends previous work by focusing on the committee structure in Congress and developing a directional, causal, and dynamic theory of firm investments in congressional districts over time.

The literature on committees in the US Congress is also quite extensive.Footnote ²⁵ Here, we focus on the dynamics of firms’ investments in different geographies. It has long been known that firms locate in districts of committee members who are influential. Bingham and MierFootnote ²⁶ note merely one example:Footnote ²⁷

Stories such as this one have been told across a number of heavy industries.Footnote ²⁸

Despite this work, most studies of the relationship between committees and firms are confined to a single industry, or one slice of the economy. This limits our understanding of the scope of firm influence.

Moreover, the current literature, while making a substantial contribution to new and expansion of investment in various political geographies, does not address the dynamics of EIA. Previous papers usually show a positive relationship between committee representation and firm investment in a district, but usually does not describe the evolution of that investment over time, especially if the legislator exits the committee. Finally, while papers examining investment or employment decisions of firms in committee member districts show strong correlation between the economic and political variables, identifying directional effects of the congressional representation and firm investment in a district is often difficult.Footnote ²⁹

We attempt to address these shortcomings in the following theoretical and empirical sections of this article by developing a dynamic theory of EIAs, across the entire US economy, with a specific emphasis on industry entry, exit, expansion, and contraction in a focal geography, as it relates to a committee member's entry, seniority, and exit from a congressional committee. Moreover, as we describe later in this article, we use natural experiments to help our identification and rule out alternative mechanisms.

Theory

In this section we develop an intertemporal framework of EIAs and draw testable predictions for the mechanisms regarding the relationship between congressional committee membership and firm activity (employment and establishments). We use a bit of notation that will later be useful in describing the empirical model. We begin by assuming two types of actors: firms and politicians. One can think of these politicians as US senators or representatives. Each politician is assumed to represent a unique political district d, and serve on at least one committee j. The committee is assumed to have “oversight” or regulatory jurisdiction over at least one industry $\iota$. Thus, each legislator can be characterized by an unique $( {d, \;j, \;\iota } )$ triplet. What is particularly important is that a legislator who serves on committee j is uniquely positioned to provide favorable legislation or regulation to firms in industry $\iota$. The legislator is also assumed to be particularly concerned about reelection (or promotion in the political hierarchy). Each firm f that participates in industry $\iota$, can choose to locate in one or more districts d. Firms are assumed to seek long-term profits; they can forgo short-term profits to pursue long-term favorable regulatory environments. With these assumptions, we now turn to understanding the mechanism for EIAs.

At each in point in time t, the firm makes a decision to expand its operations. It can choose to locate or expand in the most profitable district from an operational standpoint, or the firm can choose to locate or expand in a district that is less profitable from an operational standpoint, but in a location that offers long-term political gains that will result in higher long-term profit. The political gains we envision is the ability to influence the political environment for the industry. The way in which the firm attempts to influence the political environment is to locate its new facility or expand its incumbent facility in the district of a politician who has oversight of the industry. This location decision increases the employment (and in the case of new investment, establishments) of industry $\iota$ within district d.

Recall the example in the “Literature” section. A defense contractor has to decide where to locate a factory to make avionics. In the absence of any political motive, the firm would locate its facility in California—California being the most efficient location for the facility for a variety of economic reasons (e.g., labor force). However, assume that a senator on the Armed Services Committee represents Kansas. The advantage to locating in Kansas is that the firm has a chance to influence the Kansas senator on defense policy. That policy will affect not only the Kansas facility but also the firm's facilities throughout the country. The firm is influential because it has created jobs and revenue in the senator's district. Hence the firm, in making the location decision, weighs the utility from locating in the most efficient location in the country against the gains from locating in a suboptimal location from an operational perspective, but gaining influence over a legislator with policy-making power over the entire industry by virtue the legislator's position on the Armed Services Committee.

Banking call centers provide another example. It may be most efficient to locate a banking call center in South Dakota because of the state's labor force, land costs, and geographic proximity to both coasts. However, in the 1980s, many new banking call centers were relocated to San Antonio, Texas. Although the location was inferior to South Dakota on many dimensions, the call centers were located in the district of Henry Gonzalez, the Chairman of the House Banking Committee. The facilities, and the numerous jobs they created in Gonzalez's House district, gave the banks access and influence to a powerful political figure, access that was not granted to these banks before.Footnote ³⁰

It is useful to point out that firms may face a collective action problem. While it might be collectively rational for firms in an industry to invest in a district, it may not be individually rational for those firms to ramp up investment unless there is an individual positive net benefit. As the call center example highlights, firms and industries do seem to overcome that collective action problem. Indeed, we often observe examples where the largest firms with the highest benefits from favorable industry legislation enter the politician's district first.

The aerospace industry's growth in Alabama is an example where multiple firms within the same industry have overcome the collective action problem at an industry level to increase employment and their economic footprint to gain political capital. While Alabama has had an aerospace sector for decades, recent major growth began with the entry of European aerospace titan Airbus in 2013, which constructed a new major assembly plant in Mobile.Footnote ³¹ It was acknowledged that the move by Airbus to build an assembly plant well outside its traditional operational geography was to gain political allies in the United States even at the expense of operational efficiency. Since the initial Airbus plant announcement in 2012, other American industry players such as GE Aviation, Lockheed Martin, Northrop Grumman, and Raytheon Technologies have expanded or located production or operations in Alabama.Footnote ³² It is notable that US senator, Richard Shelby, has served on the Senate Appropriations Committee for more than a decade, as well as on the Appropriations Defense Subcommittee.

Like the campaign contributions literature, our theory of EIAs does not take a stand as to whether these investments are requested by the politician—colloquially known in the literature as threats or “shakedowns”—or whether a firm initiates the investment to contemporaneously or subsequently influence the legislator. The theory only notes that at least some component of the investment is driven by political considerations.

What makes EIAs different from campaign contributions, though, is when a legislator exits a position of power (or retires from congress) campaign contributions cease from the interest group. However, it is extremely uneconomic for a firm to close a facility when there is turnover in Congress, given the fixed and sunk cost of setup and operation. This sunk cost raises the cost of exit; we argue that the firm will not close its current facilities, but it will cease to increase investment and expansion in that district, shifting its future expansion to districts of new powerful legislators.

From a research perspective, how we measure campaign contributions and EIAs is also quite different. Campaign contributions are, by definition, a flow variable. By contrast EIAs, measured by employment or establishments are, by definition, stock variables. This difference also yields different expected patterns in the relationship between committee membership and firm political influence activities, namely because stocks accumulate over time while flows do not.

To illustrate our theoretical framework, and contrast it to other forms of political influence, we present figures 1 and 2. The intertemporal predictions for campaign contributions are illustrated in figure 1. Before the legislator joins the powerful committee (region A), campaign contributions are set at some level. While on the committee, the legislator will receive an increasing amount of contributions as she rises in the committee hierarchy (region B). However, when the legislator exits the committee, the campaign contributions revert to a lower level (region C).

Figure 1: Committee service and contributions.

Our theory of EIAs, however, has different predictions that are described in figure 2. Legislators who are not on the committee should expect to see firm investment in the industrial sectors governed by the committee to be standardized at some constant level (or macroeconomically trending level) (region A). However, in districts where members join the committee, the amount of investment and employment in those industries overseen by the committee will rise as the seniority of the legislator rises on the committee (region B). The increase in employment may come from a firm creating a new establishment in the district, or from redistributing its current employees from facilities outside the district into facilities inside the district. Note, in industries which are not overseen by the industry, the level of investment and employment in the industry will not rise from the trending level (bottom, no committee line). As noted earlier, unlike the campaign contribution theory of political influence, our theory of EIAs predicts that when the legislator exits the committee, the number of jobs and amount of investment in the district will stay roughly constant (region C). The firm will not exit the district, but will instead keep the current levels of jobs.

Figure 2: Committee service and employment.

In addition to the increasing relationship between accumulated committee service and industry employment, figure 2 also depicts this relationship as concave—that is, accumulated committee service is associated with increasing employment, but at a decreasing rate. There are two primary theoretical reasons that suggest concavity may arise, emanating from work of Bonardi and UrbiztondoFootnote ³³ and Raiha.Footnote ³⁴ First, the electoral benefits of EIAs diminish as they grow. Initial boosts to the local economy and employment improve the electoral popularity of the incumbent, but as employment grows the incumbent's electoral fortunes become more certain—thus incremental increases in employment carry smaller political benefits. Second, once there is a critical mass of employment within an industry, the incumbent's interests become aligned with the interests of the firm. This is because voters are increasingly stakeholders of the firm—either directly as employees, or indirectly through economic multipliers. As stakeholders with an economic dependence on the firm, their view of the politician will depend on how well the firm is treated by policy makers. Thus, once employment rises sufficiently, the incumbent politician is incentivized to have the interests of the firm in mind, even without subsequent firm investment. Relatedly, empirical work finds a curvilinear relationship in the related area of employment and campaign contributions,Footnote ³⁵ which further informs our theory.Footnote ³⁶

Overall, our theory of EIAs argues that firms make operational decisions that consider not only the economic efficiency of the operations but also their opportunity to gain political influence. Firms will locate some of their facilities to optimize their operational and political value. This will manifest itself in the location or expansion of facilities and employment in districts of congressional committee members who have oversight of the firm's industry. Figure 2 encompasses the theoretical predictions. Later in the section “Robustness” we consider the possibility of reverse causation and other alternative explanations.

Empirical Strategy

Variables

Our main dependent variables are the number of people employed, and the number of establishments, aggregated by SIC industry within a congressional district, in a given year. These variables measure a particular industry's economic presence and footprint within a congressional district.

As mentioned in the “Theory” section, there are three predictions of our theory, each corresponding to a region in figure 2. To distinguish between these regions we require two different types of independent variables that capture the time accrued after a particular event has occurred. Because these variables measure time accrued and are, as a consequence, (monotonically) nondecreasing we call these measures duration clocks.

Our main independent variable is a duration clock of the number of two-year terms a particular congressional district's representatives have accumulated on congressional committees of relevance to the given SIC industry. For example, consider the four-digit SIC industry 5144—Poultry and Poultry Products—for whom the House Agriculture Committee and Senate Agriculture, Nutrition, and Forestry Committee are the only relevant committees.Footnote ⁴² Also consider two particular congressional districts. Prior to 1983, in Illinois's 17th district, no House representative had served on the House Agriculture committee (dating back to 1947), until Rep. Lane Evans served three terms on the committee between 1982–88. As a result the relevant House duration clock for SIC 5144 in Illinois's 17th district begins the sample period with a value of 1 (in 1983), rises to 3 (in 1987), and remains at 3 (in 1989). By contrast, by 1983, Washington's 5th district Rep. Tom Foley had accumulated ten terms of service on the House Agriculture Committee (even serving as chair from 1975 to 1981). So the House relevance duration clock for SIC 5144 in Washington's 5th district begins the sample period with a value of 10 (in 1983), rises to 11 (in 1985), and remains constant once Foley left the committee. Because this particular duration clock measures the accumulated time a district's representative has spent while in a committee relevant to the SIC industry, we will refer to this variable as DClockIn.

However, because we also need to capture any different patterns in employment after a district's representative has left all committees of relevance, we need a second duration clock. This variable, which we will refer to as DClockOut, measures the number of terms that have transpired since a particular district's representative served on a committee of relevance to the industry. This variable allows us to separately identify the portion of the committee curve in region C of figure 2, as distinct from the committee curve in region B, and the noncommittee curve in region C.

Like with the variable DClockIn, this second duration clock is also nondecreasing—even if the district's representative subsequently joins a committee of relevance, DClockOut simply remains constant, and does not increase. Moreover, DClockOut does not begin to increase until a district's representative has left all committees of relevance—so districts whose representatives have never served on a committee relevant to a particular SIC (i.e., those whose DClockIn = 0) have a DClockOut equal to zero. Like the variable DClockIn, DClockOut accounts for time transpired since departing relevant committees incurred from before the sample period (dating back to 1947). For instance, if a district had a representative on the House Agriculture committee until 1977 (but not after), then the DClockOut for SIC 5144 in that district would be 3 in 1983, as it would have been three terms since the industry had representation on a relevant committee in that district.

Table 1 presents basic descriptive statistics. Because our unit of observation is a four-digit SIC-congressional district-year (SIC-CD-Y), the average numbers of employees, establishments, and sales are quite small, though the standard deviations are quite high. An average SIC-CD-Y has just more than 300 employees, 3.26 establishments, and approximately $400,000 in sales.Footnote ⁴³

Table 1: Descriptive statistics and sample comparison.

Confidence levels: * = 10%; ** = 5%; *** = 1%.

Note: This table reports the descriptive statistics and comparison for the full sample of SIC-CD-Y observations. In the top portion of the table, the first column reports the mean value of each key variable, while the second column provides the standard deviations. The middle portion compares the employment, establishments, and sales of the sample between SIC-CD-Ys whose representative is a House Committee member (on a relevant committee) or not, and the bottom portion between SIC-CD-Ys either of whose senators is a Senate Committee member (on a relevant committee) or not. In the middle and bottom portions, the first two columns compare means, the second two columns compare standard deviations, while the third column is the T-statistic from a test comparing the equality of means between the groups.

We see that for about 14 percent of the SIC-CD-Ys the local House member is on a committee of relevance, while for about 39 percent of the SIC-CD-Ys at least one of the state's senators is on a committee of relevance.Footnote ⁴⁴ As was mentioned, both duration clocks account for accumulated time spend on/off committees of relevance from before the sample period, dating back to 1947. This is why the variables DClockIn and DClockOut range from zero to twenty-two.

Table 1 presents a comparison of SIC-CD-Ys whose representative(s) are on committees of relevance and those who are not. Though this simple comparison involves no control variables, we see that there is a statistically significant difference in the level of employment, establishments, and sales between the two groups. Quite clearly, SIC-CD-Ys whose representative(s) are on either a House or Senate committee of relevance, feature 26.1 percent higher employment, 4.3 percent more establishments, and 10.3 percent higher sales than those who are not.

Empirical Strategy

Table 1 shows greater economic activity in industries whose representatives are on committees of relevance to the industry. Thus, we follow the literature that uses variation in congressional committee membership as a means of empirical identification (e.g., Bertrand et al., Reference Bertrand, Bombardini, Fisman and Trebbi2020; Fouirnaies and Hall, Reference Fouirnaies and Hall2018; Powell and Grimmer, Reference Powell and Grimmer2016). However, we would like to see whether the relationship holds controlling for factors that affect the local level of economic activity, and whether or not greater accumulated committee service is a predictor of economic activity.

To do this, we estimate the following regression specification:

(1)$$\eqalign{ln( y_{idt}) & = \beta _0 + \beta _1DClockIn_{idt} + \beta _2DClockIn_{idt}^2 \cr & \quad+ \beta _3DClockOut_{idt} + \beta _4DClockOut_{idt}^2 \cr & \quad+ \alpha _1\iota + \alpha _2\delta + \alpha _3\tau + {\epsilon }} $$

where y _idt is the number of employees or establishments in industry i in district d in year t, while $\iota$, δ, and τ are industry, district, and year fixed effects, respectively.Footnote ⁴⁵

This equation not only distinguishes between the sections of the curves in figure 2, but allows for nonlinearity in the sections. Given our hypothesized relationship, we expect employment to be increasing in DClockIn but at a decreasing rate—thus we expect β ₁ to be positive and β ₂ to be negative (but smaller in magnitude). The theory predicts firms do not systematically alter employment or cease to make further investments once a representative has departed committees of relevance, thus we expect the coefficients on DClockOut, β ₃ and β ₄ to be statistically insignificant.

Main Results

Our main results are contained in tables 2 and 3. In each table, models A through D as well as models E through H present four different regression specifications, each leading up to the full specification from equation 1. Models A and E have DClockIn and DClockOut as the independent variables. Models B and F include both the linear and quadratic DClockIn variables. Models C and G presents estimates from the full regression equation (1). All the specifications include congressional district, four-digit-SIC industry, and year fixed-effects. Models D and H further include control variables for the Democratic Party vote share, from the most recent presidential election, within the congressional district, as well as the state unemployment rate. Models A through D have the log of industry employment as the dependent variable, while models E through H have the log of industry establishments as the dependent variable. Table 2 examines the impact of Senate committees—that is, both DClockIn and DClockOut measure accumulated service time on (and off) Senate committees. Table 3 examines the impact of House committees—both DClockIn and DClockOut measure accumulated service time on (and off) House committees. T-statistics are reported in parentheses. Throughout we use standard errors clustered at the industry-state level.Footnote ⁴⁶

Table 2: Full sample regressions—employees/establishments and Senate committees.

Confidence levels: * = 10%; ** = 5%; *** = 1% (t-values).

Note: This table presents estimates of eight regression models, of accumulated committee service in the Senate on relevant industry employment and establishments, using the full sample. In specifications A–D the dependent variable is the log of industry employment, while in specifications E–H the dependent variable is the log of industry establishments. All specifications include industry, district, and year fixed effects. The F-statistic of the test of the joint significance of all explanatory variables, is reported below each column. T-statistics are reported in parentheses. Standard errors are clustered at the four-digit SIC-state level.

Table 3: Full sample regressions—employees/establishments and House committees.

Confidence levels: * = 10%, ** = 5%, *** = 1% (t-values)

Note: This table presents estimates of eight regression models, of accumulated committee service in the House of Representatives on relevant industry employment and establishments, using the full sample. In specifications A–D the dependent variable is the log of industry employment, while in specifications E–H the dependent variable is the log of industry establishments. All specifications include industry, district, and year fixed effects. The F-statistic of the test of the joint significance of all explanatory variables, is reported below each column. T-statistics are reported in parentheses. Standard errors are clustered at the four-digit SIC-state level.

Table 2 presents regression results on the impact of Senate committee membership on employment. The estimates reported in model A show that accumulated service on Senate committees of relevance is associated with higher industry employment—each additional (two-year) term of committee service translates into an increase in employment of about 0.84 percent. Model A also shows a modest increase in employment following the departure of a relevant Senate committee member. Model B shows that there appears to be a nonlinearity in the relationship between committee service and employment. The coefficient on DClockIn is positive and statistically significant, while the coefficient on DClockIn-squared is of much smaller magnitude, negative, and statistically significant. This indicates that Senate committee service is associated with increasing employment, but at a decreasing rate—as hypothesized.Footnote ⁴⁷ The magnitudes of the coefficients indicate that employment would peak after thirteen terms of accumulated Senate committee service. In contrast to model A, the statistical significance of the coefficient on DClockOut vanishes in models C and D where we account for nonlinearity in the relationship between our independent variables and employment. The coefficients on DClockIn are nearly identical to those in model B, while the coefficients on DClockOut are both statistically insignificant. Furthermore, the magnitude of the coefficients (i.e., point estimates) on DClockOut are both considerably smaller than for DClockIn.Footnote ⁴⁸ Overall, these results appear consistent with our hypotheses that committee service should translate into increased employment (at a decreasing rate), but time after committee departure produces no significant change in employment.Footnote ⁴⁹

Models E through H in table 2, are analogous to models A through D, and present regression results on the impact of Senate committee membership on the number of establishments. Model E shows that accumulated service on Senate committees of relevance is associated with a higher number of establishments, while time accrued postcommittee has no statistically significant association with establishments. The magnitude, however, is smaller than the same coefficient in model A—each additional term of committee service translates into an increase in establishments of about 0.46 percent. Model F, further, shows that there appears to be a nonlinearity in the relationship between committee service and establishments. Once again, the direction and statistical significance of the coefficients on DClockIn are similar to their counterparts from the regression on employment. Thus, it appears that Senate committee service is associated with increasing establishments, but at a decreasing rate. The magnitudes of the coefficients on establishments indicate that establishments would also peak after approximately thirteen terms of accumulated Senate committee service. The results in models G and H are qualitatively similar to their counterparts in models C and D.Footnote ⁵⁰ The coefficients on DClockIn are nearly identical to those in model B, while the coefficients on DClockOut are both statistically insignificant. Overall, the results in table 2 indicate that there is a similar relationship between establishments and Senate committee service, as there was between employment and committee service, though the magnitude is smaller for establishments than for employment.Footnote ⁵¹

Table 3 is analogous to table 2—it presents regression results on the impact of House committee membership on employment and establishments. However, unlike table 2, the regression estimates in table 3 fail to find the same statistically significant association. The coefficients on DClockIn are not statistically significant, though the direction of the coefficients is as hypothesized—that is, the coefficient on DClockIn is positive, while the coefficient on DClockIn-squared is of much smaller magnitude and negative. Further, little changes when DClockOut is included in the regressions—the coefficients on DClockIn remain statistically insignificant, while the coefficients on DClockOut are, for the most part, also statistically insignificant.Footnote ⁵² Overall, table 3 does not indicate the same association between House committee service and employment or establishments as we observed in the Senate.

In sum, our main results are consistent with the predictions of our hypotheses for Senate committees, but not for committees in the House of Representatives. Although we expected to find similar results in the House and Senate, we can identify a number of institutional reasons why one might find this difference between the two chambers.

First, on a per representative basis, senators are more powerful than House members. There are only 100 senators compared to 435 House members. Furthermore, congressional committees in the Senate are smaller than those in the House. This difference in power may make it more attractive for firms to target key senators compared to House committee members. The more limited power of individual House members may make them less ideal targets for (arguably) more expensive influence activities such as EIAs.

Second, senators each serve six-year terms as opposed to the two-year terms served by House members. If EIAs are intended to influence and establish a longer-term relationship between an industry and its representatives, the longer terms of senators allows for less risk for firms in making (more) permanent economic investments.

Third, the jurisdictions of House members are geographically smaller than the jurisdictions of senators. This makes senators easier targets for EIAs because the specific location of employment and establishments does not need to be as accurate as it would need to be in targeting House members’ districts.

Fourth, House districts are subject to periodic redistricting, while state boundaries are not altered. This geographic uncertainty might also make senators relatively more attractive for location-based EIAs than House members.

Fifth, in concentrated urban areas, the people who work in a particular congressional district may not live in the same district, but rather commute from another district. Even if jobs are created in a specific congressional district, those jobs may instead benefit workers from other districts. This same leakage is less common across state boundaries.

Together, these reasons provide possible explanations for the difference in the House and Senate results. However, more work is needed to more fully explore the factors that drive the differences in the results.

Robustness

Though our main results support our theory that firms use EIAs to influence key congressional committee members in the Senate, there are several alternative explanations that could instead explain the observed results. In this section we discuss three possible alternative stories that could account for our findings. Because our main results indicated no systematic relationship between committee service and economic activity in the House, our remaining results in this section focus on Senate committees.

Committee assignments chasing economic activity

One alternative explanation is that firms aren't responding to congressional committee membership and seniority, but rather the reverse—legislators are joining committees of relevance when they observe certain industries growing in their districts. Then, once the representative anticipates the industry growth is finished, she chooses to leave the relevant committees. This story assumes legislators are making these choices to join as well as leave approximately in conjunction with industry growth commencing or terminating.

Discerning between this alternative explanation and our hypothesis, based on EIA, is challenging because a legislator's decision both to join and leave a committee is endogenous. However, there are circumstances in which a legislator leaves congressional committees not by choice, but rather by exogenous circumstances—for instance, due to death or scandal.

We use the exogenous departures of senators, as described in figure 3, to test whether the alternative reverse story holds. Models A and C in table 4 report regression estimates that are analogous to models C and G in table 2, but have some key differences in the sample. While results in table 2 are obtained from regressions run on the full sample, the results in table 4 are obtained from regressions only including SIC-CD-Ys where the senator is currently on a committee of relevance or, if not on a committee, where the departure was due to an exogenous event. The exogenous departures we consider are departures due to death, resignation related to scandal, or promotion to a higher office (e.g., vice president). In this way, the regressions in table 4 are comparing the increasing and concave curve regions from B and C of figure 3, against the “depart” (constant) curve from region C of figure 3. The alternative explanation would predict the trend in employment and establishments for SIC-CD-Ys that retain committee members of relevance is different from trends for SIC-CD-Ys whose committee members departed exogenously.

Figure 3: Exogenous committee exit and employment.

Table 4: Robustness regressions—Senate committees.

Confidence levels: * = 10%; ** = 5%; *** = 1% (t-values).

Note: This table presents estimates of five regression models, of accumulated committee service in the Senate on relevant industry employment or establishments. Specifications A and C use a subsample of SIC-CD-Ys for whom either (1) DClockOut is zero, due to an exogenous departure, or (2) DClockOut is positive, and the representative is on committee. Specifications B and D use a subsample of SIC-CD-Ys for whom DClockIn is nonzero. Specification E uses the full sample. In specifications A and B the dependent variable is the log of industry employment, in specifications C and D the dependent variable is the log of industry establishments, and in specification E the dependent variable is the log of industry sales. The F-statistic of the test of the joint significance of all explanatory variables, is reported below each column. T-statistics are reported in parentheses. Standard errors are clustered at the four-digit SIC-state level.

The results presented in models A and C of table 4 are qualitatively similar to our main results. Accumulated Senate committee service is associated with increases in employment and establishments. Moreover, the increases to both employment and establishments occur at a decreasing rate. The magnitudes of the coefficients are also similar to our main results.

More important are the coefficient estimates on the DClockOut variables, where the departures are due only to exogenous events. We see in models A and C that growth in employment and establishments levels off after the departure of the senator from a committee of relevance, even though the departure is exogenous. This is inconsistent with the alternative explanation. In light of these findings on exogenous departures, this alternative explanation seems unlikely. Similar to arguments made by Bertrand et al. (Reference Bertrand, Bombardini, Fisman and Trebbi2020), who also examined congressional committee departures in establishing causality in strategic targeted corporate philanthropy, the fact that growth in employment and establishments levels off after committee departure suggests that at least part of the initial firm investments are politically motivated.

Extensions

In this section, we explore extensions to our theory and main results. Although the results in this section are not definitive, they do establish what we believe are interesting facts that can be extended by future work.

Political heterogeneity

In the main section of the paper, we include district fixed-effects in all regressions to control for time-invariant district heterogeneity. However, the political competitiveness of a district may change over time and cause different EIA intensity by firms. The recent literature does not contain a single prediction as to how electoral competitiveness might affect EIA intensity. On one hand, tight political contests might result in higher job creation ratesFootnote ⁵⁵ and, on the other hand, wide election margins by a politician might attract private investment to a legislator who might be a long-term attractive ally.Footnote ⁵⁶

To examine the role of vote margin on EIA, we estimate models that include an election vote margin variable as a measure of safety. In particular, we add Vote Margin to the employment specification in equation (1), both directly and interacted with the DClockIn variables.Footnote ⁵⁷

The results including Vote Margin are presented in table 5. In the Senate models A and B we see that our original findings are unchanged by the inclusion of vote margins. The coefficient for DClockIn is still positive, statistically significant, and of a similar magnitude to our main results in table 2. By itself, the coefficient on Vote Margin is positive and statistically significant. The positive magnitude indicates that states with higher average win margins exhibit higher employment. In model B we find no coefficients on the interaction terms between Vote Margin and DClockIn (or its square) are statistically significant. Overall, inclusion of Vote Margin has little effect on our original Senate results.

Table 5: Win margins results—Senate and House committees.

Confidence levels: * = 10%; ** = 5%; *** = 1% (t-values).

Note: This table presents estimates of four regression models, of accumulated committee service in the Senate and House on relevant industry employment, using the full sample. In all models the dependent variable is the log of industry employment. In models A and B the committee service is in the Senate, while in models C and D the committee service is in the House. All models include industry, district, and year fixed effects. The F-statistic of the test of the joint significance of all explanatory variables, is reported below each column. T-statistics are reported in parentheses. Standard errors are clustered at the four-digit SIC-state level.

Models C and D in table 5 show, however, the win margin results are not the same in the House. In models C and D, the direct effects of DClockIn and its square continue to be statistically insignificant, as in the original results. In addition, in model C, the direct effect of Vote Margin is positive and statistically significant, as in the Senate models in the table. But in model D, we find that the interactions between win margin and DClockIn are both statistically significant. This finding, in model D, provides the first glimpse of a result in the use of EIA for House committees, and indicates that increased committee seniority is associated with increased employment only in electorally safe House districts. The result is worth further future investigation.

Industry heterogeneity: Breadth of industries

While our main findings indicate that firms appear to engage in EIA toward relevant Senate committee members, we examine here the extent of EIA across industries. To do this, we run regressions with separate DClockIn variables for each two-digit SIC industry. We estimate the following regression specification:

(2)$$\eqalign{ln( y_{idt}) & = \beta _0 + \beta _1DClockIn_{idt}^{SIC = 01} + \ldots + \beta _{97}DClockIn_{idt}^{SIC = 97} \cr & \quad+ \beta _3DClockOut_{idt} + \alpha _1\iota + \alpha _2\delta + \alpha _3\tau + {\epsilon }} $$

where $y_{idt}$ is employment, and $\iota$, δ, and τ are defined as before. $DClockIn_{idt}^{SIC = X}$ is the industry-specific DClockIn variable. Equation (2) allows us to measure the importance of Senate committee seniority differently across different industries. We use two-digit SIC industries because this involves only seventy-four DClockIn coefficients.Footnote ⁵⁸ Also, to simplify the analysis and reduce the number of coefficient estimates, we do not include squared terms.

We estimate a total of seventy-four two-digit industry coefficients.Footnote ⁵⁹ A total of nineteen of the industry DClockIn coefficients are positive and statistically significant, while twenty-nine of them are positive but not statistically significant. Another twenty coefficients are negative and not statistically significant, and six of them are negative and statistically significant. Consistent with our earlier findings, the coefficients on the DClockOut variables are not statistically significant. The binomial probability of at least nineteen coefficients being statistically significant by chance is less than 0.1 percent, even at a 10 percent significance level.

Table A2, in the appendix, presents the coefficient estimates from the industries that had positive and statistically significant coefficients. A broad variety of industries are represented ranging from agriculture to oil and gas, to eating and drinking places, industrial machinery and equipment, as well as hotels and other lodging. The scope of industries ranges from heavy industries studied by other researchers,Footnote ⁶⁰ to more service-oriented industries often overlooked in the literature.

Industry heterogeneity: Responsiveness of investment

In a final extension, we explore the responsiveness of investment entry to committee membership and seniority. While some industries may face significant lag times and start-up costs when locating new establishments, or reorienting operations and employment, others might have varying degrees of flexibility and ease in creating new operations.

We attempt to quantify the ease with which firms in an industry can make new investments in EIA, by examining one potential measure of operational flexibility and time horizon—the length of the product development cycle (PDC). The PDC refers to the industry average time it takes to develop and bring new products to market.Footnote ⁶¹ Some industries, such as chemicals and pharmaceuticals, face long PDCs as it typically takes years of development, research, and operational readjustment to create new products. By contrast other industries, such as retail or printing and publishing, can develop and reorient operations to develop new products in a shorter period of time. As such an industry's typical PDC is likely correlated with its ease of operational flexibility.

In the context of EIAs, we believe that firms in industries with greater operational flexibility and ease of entry are more likely to employ EIAs. To test this, we employ a dataset of average PDCs at the industry level developed by National Academy of EngineeringFootnote ⁶² and Bushman et al.Footnote ⁶³ We define the variable Short PDC as those industries with a PDC of less than four years.Footnote ⁶⁴ While the set of industries measured by this dataset is not exhaustive, it covers more than one-third of our samples’ district-industry-year observations.

To test our hypothesis, we add interactions to the specification in equation (1), so that the DClockIn and DClockOut variables appear on their own, as well as interacted with Short PDC.

The results are presented in table 6. Models A through D are analogous to the same labeled models in table 2. In all models we find that the coefficient on the interaction between DClockIn and Short PDC is positive and statistically significant, which indicates that committee membership and seniority is associated with greater employment particularly for industries with short PDCs. Furthermore, the interaction between Short PDC and the squared DClockIn is also statistically significant and consistent with our main results. We generally find little evidence of any statistically significant interaction between Short PDC and DClockOut. In models C and D we also observe no significant association between DClockIn and employment for industries with long PDCs. The direct effect of the clock variables maintain their original signs, though at times do not reach statistical significance, perhaps because of the lower power of the tests with the fewer observations. These findings suggest that firms with potentially greater operational flexibility and ease of entry are more likely to engage in EIA. These results, and the role of different kinds of fixed and sunk costs, do warrant further investigation in the future, as the coefficients of the DClockOut parameters do meander in and out of statistical significance.

Table 6: Industry heterogeneity table—Senate committees.

Confidence levels: * = 10%; ** = 5%; *** = 1% (t-values).

Note: This table presents estimates of four regression models, of accumulated committee service in the Senate on relevant industry employment, using the subsample of industries for which the product development cycle variable is available. In all models the dependent variable is the log of industry employment. All models include industry, district, and year fixed effects. The F-statistic of the test of the joint significance of all explanatory variables, is reported below each column. T-statistics are reported in parentheses. Standard errors are clustered at the four-digit SIC-state level.