S TATES ’E XPANDING I NCOME G AP ? O C AMPAIGN D ONATIONS C ONTRIBUTE T O T HE U NITED M ONEY I N P OLITICS :D

M

ONEY

I

N

P

OLITICS

:

D

O

C

AMPAIGN

D

ONATIONS

C

ONTRIBUTE

T

O

T

HE

U

NITED

S

TATES

’ E

XPANDING

I

NCOME

G

AP

?

Author: Safan van der Gaauw*

Supervisor: Dr. P. Heijnen

Abstract

Campaign finance is becoming an increasingly controversial facet of American politics. In this article, I examine the effect of the political campaign finance system on income inequality in the United States. First, using a cross-country panel, the effect of campaign contributions to members of both federal- and state-level governmental bodies on several income inequal-ity and income share variables is estimated by means of Two Way Fixed Effects regressions. The contributions to federal policy-makers–specifically, members of Congress–are found to positively affect income inequality. This finding holds when controlled for potential reverse causality using System Generalized Methods of Moments. Additionally, a Spatial Durbin Model produces a negative spatial spillover effect for contributions to the House of Represen-tatives, indicating reversed spatial interactions are at play in the effects of campaign finance. This article’s outcomes bolster the central objection of critics of campaign finance, namely that it is biased in favor of the economic elite.

Keywords

Income Distribution · Campaign Finance · Political Action Committees

JEL Classifications

C33 · D31 · D72 · P16

Master Thesis Economics Hand-In Date: June 24, 2016

1

I

The United States’ political and societal spheres are showcasing an increased renouncement of the alleged influence or even dominance of money in politics. What is meant by money in poli-tics is the use of financial incentives to influence for instance, political outcomes, politicians, or election outcomes; it can be seen as a euphemism for lobbying, even though money is not nec-essarily involved in lobbying. American citizens, companies, and interest groups can lobby U.S. government officials in a variety of ways. One of those forms of lobbying always involves financial incentives, and has been given particular attention from critics, especially in the primaries for the 2016 U.S. presidential election. Financial donations to campaigns of politicians became one of the cardinal issues in the campaign of Democratic candidate Bernie Sanders and Republican candidate Donald Trump. The problem raised with respect to the political campaign finance sys-tem is that it is dominated by Special Interest Groups (SIGs) and the economic elites. As a result of the increasing importance of campaign contributions to not only Presidential candidates, but also members of Congress, the Supreme Courts, Governors offices, and so on, it is argued that U.S. politics misrepresents its society’s interests in favor of the affluent.

Lobbying as a political activity is enshrined in American politics ever since the federal repub-lic’s inception. The First Amendment to the Constitution of the United States prohibits Congress from instituting law “abridging the freedom of speech or of the press, or the right of the people peaceably to assemble, and to petition the Government” (U.S. Cons. amend. I). Lobbying, or more specifically, financial campaign donations have since been determined to be protected under the First Amendment by the U.S. Supreme Court, starting with the landmark 1976 case Buckley v. Valeo.1 This and several subsequent cases effectively construed the notion of campaign contribu-tions in such a way that it is a form of speech, whose freedom should not be impeded upon.

In spite of this firmly established position of lobbying in the American political system, it has become a more controversial issue recently, as evidenced by the aforementioned presidential election primaries. To be able to contribute to campaigns, money is needed, and the more one can donate, the better. This makes for a system wherein having more money enhances the ease with which an individual or SIG can advance their agenda. A campaign donor’s interests can be broadly categorized as, what Grossman and Helpman (1996) call, “electoral” or “influence” motives. The former refers to the SIGs’ aim of influencing the election in such a way that their preferred candidate wins, whereas the latter describes contribution aimed at steering the behav-ior of the politician into a preferred direction. Either way, it can be argued that those with most to contribute are the ones who gain the most from a system of political campaign finance. This could then exacerbate the difference between the rich and poor citizens. Hence, the American political campaign finance system may have an inherent skewness in favor of its population’s richest faction.

This alleged bias of the U.S. political system towards its biggest financial donors would add to another debate that has become progressively pervasive in the U.S. and other Western societies: the issues of growing income inequality. Inequality is on the rise in many Western economies and

the United States alike, which could explain the growing concern surrounding it. The protests targeted at the ‘one-percent’ in the immediate aftermath of the 2008 Financial Crisis were early eruptions of the growing discontent among Americans about the incomes of the richest percent-age of U.S. society. Since then, the incomes of the economic elite relative to that of the lower and middle classes have been a focal point for economists the likes of Joseph Stiglitz and Thomas Piketty. Though there is consensus in the economic literature about the fact that income in-equality has been rising for some successive decades, the question of what the effects of income inequality are, and whether these effects are detrimental or beneficial for key economic variables is still a contentious one.

Additionally, income inequality in the United States is studied among political scientists, who mostly equate it with political inequality. After juxtaposing the influence of economic elites and business interests-representing groups against that of average citizens and groups representing the interests of the masses, Gilens and Page (2014) find a significant difference, wherein only the former have a substantial impact on political outcomes. They conclude that American pluralism is biased, and that U.S. policy making is dominated by the economic elite. Similarly, Winters and Page (2009) argue that American political system can be regarded in terms of oligarchy, wherein extreme inequality allows for considerable political influence by a small, affluent faction of society. Interestingly, they suggest that factors such as lobbying and opinion shaping are likely to be pathways through which the “oligarchs” gain influence, though their research itself does not investigate these pathways. Whereas a possible bias in the American political system is researched by academics in economic and political contexts, the possible link and causation between inequality in the system of political campaign finance and in the income distribution has not yet been studied. This article will fill that gap in the literature on campaign finance.

inequal-ity by disproportionately serving the interests of the United States’ economic elite. Conversely, no significant effect of campaign contributions to state level governmental bodies (e.g. Governor) on income inequality is found. System-GMM regressions not only confirm these findings, but also indicate that it is robust for possible reverse causality between the campaign contribution-variables and inequality measures. Additionally, the spatial econometric inquiry indicates that the effects of campaign contributions to the House of Representatives spills over to bordering states. This indirect effect is negative, indicating that campaign contributions to members of the House of Representatives of a certain state bring about a lower income inequality in contiguous states.

All in all, there are two cardinal complementary findings in this article. First, income in-equality increases as a result of the political campaign contribution system. Secondly, the merits of campaign contributions are skewed in favor of America’s economic elite. In light of the fact that campaign contributions are an integral part of the American political system, these findings have two implications. First, the political campaign finance system has a polarizing effect on American society, by increasingly bifurcating the interests of the lower and upper classes. Secondly, this system appears to have a perpetuating dynamic, wherein the economic elite is able to further advance their interests and economic position through its disproportionate political influence.

The gravity of these findings and implications of course are contingent on certain political convictions, as is often the case with issues such as income inequality. If one would perceive current income inequality levels in the United States as excessive and wrong, which appears to be the tendency among a growing number of U.S. citizens, then this article’s findings shed a negative light on campaign contributions. Conversely, if income inequality is perceived to be acceptable, this article’s findings will likely be less of an issue.

What can be said about these findings then? This outcomes seems to bolster what some political scientist have established, namely that the United States is becoming less democratic or, more extremely, no longer is a democracy. First, these findings do indicate that one faction of U.S. society enjoys disproportionate influence on politics through campaign finance, which can be seen as undemocratic.2 Secondly, basic political theory dictates democracy would logically serve the interests of the majority, but this article’s outcomes indicates that, in terms of the income distributions, this does not hold. Are Americans simply failing to vote in their own interest or can we speak of an undemocratic system of campaign finances? Clearly, the political campaign finance system does not promote and arguably even undermines the democratic values of the United States.

The outline of this research is as follows: Section two delve into the literature on lobbying in general, interest groups, and the system of political campaign contributions. In a similar vain, section three will also review the literature, this time on income inequality. At the end of the third section, the central hypotheses, founded on the findings of the literature review, will be presented. Thereafter, section four will present and discuss the data used. Section five outlines 2_{An influential and relevant historic text that addresses factional politics is written by one of America’s Founding}

the econometric methodologies, and will thus present three econometric specifications. Section six presents the results and deliberates on these outcomes. Section seven concludes.

2

L

: S

I

G

, C

C

,

P

A

C

Lobbying, the process of trying to influence government officials in their decision-making, is a political activity that is less on the forefront than most other political processes. Many different actors can engage in lobbying: corporations, individuals, associations, and interest or advocacy groups. These actors can lobby the government or government officials in a variety of ways. One way is direct lobbying, which is the process wherein a lobbyist is in direct contact with the political decision-maker who he tries to influence. Grassroots lobbying on the other hand, describes the act of trying to influence the legislature by mobilizing the constituents, in an effort to influence the decision-making government officials.3 Finally, an interest group can contribute to the political campaign of incumbent or challenging government officials.4 This way of influencing political decision-making is central in this article, and will therefore be the focal point in this section’s discussion of lobbying.

The lawfulness of lobbying differs heavily for each country, but often it is a relatively unreg-ulated activity. The United States, however, is an exception: lobbying is scrupulously organized and coordinated by law. This legal framework constructed around lobbying elicits a transparent and relatively comprehensible perception of lobbying activities in the United States. A probable underlying cause for the conspicuousness of lobbying in the United States–in the societal, polit-ical, and judicial sphere–may be the the importance of free speech and the right to petition. In fact, the 2006 Supreme Court Decision Randall v. Sorrell declared the Vermont limit on financial contributions to politicians as unconstitutionally low.5,6 Herein, Justice Antonin Scalia argued that it was not an issue of money, but of speech. Following this line of reasoning, petitioning or financially supporting politicians is a fundamental and constitutional right of American citizens. 3_{An exhaustive text on grassroots lobbying (what political scientists typically refer to as outside lobbying), is}

pro-vided by Kenneth Goldstein (1999). In his book Interest Groups, Lobbying, and Participation in America, he analyzes this channel of interest groups influencing the legislature, and finds, among other things, that interests groups that succeed in mobilizing mass participation possess extraordinary power. Another significant work on grassroots lobbying is Outside Lobbying. Herein, Kollman (1998) finds that interest groups usually capitalize on pre-existing preferences of the public.

4_{Clearly, this article treats the system of political campaign finance as a form of lobbying. This may be viewed as}

incorrect by some. Indeed, there are distinctions to be made between campaign finance and lobbying, especially when it comes to how the American laws treats both. Briffault (2008) discusses both the legislative distinctions, as well as the similarities. One thing he notes is that lobbying is less strictly regulated than campaign finance. Nevertheless, for the purpose of this article, written in an economic context, I think the similarities surpass the distinctions and the two can be viewed as two sides of the same coin. This assertion is validated by the fact that campaign contributions are shown to be the best predictor of lobbyists’ access to official makers (e.g. Ainsworth, 1993; Ansolabehere et al., 2002; Bertrand et al., 2014).

5_{Randall v. Sorrell, 548 U.S. 230 (2006).}

6_{For more on this, and other Court cases pertaining to campaign finances, please consult literature on U.S. law and}

Hence, lobbying in all its forms is therefore rather uncontroversial and prominent in the United States, largely due to its established connection to the right of free speech.

Since lobbying is such an integral part of American politics, it also affects its political econ-omy. After all, if lobbying influences American governance, it is likely to have effects on factors of particular interests within economic theory, such as production, trade, antitrust, and the distri-bution of wealth. Lobbying therefore is extensively debated among economists as well. As often is the case, when economist look at lobbying they critically assess whether it produces welfare-offsetting consequences, and whether lobbying distorts economic growth. In an influential essay published in 1957, which was followed up by the book An Economic Theory of Democracy, An-thony Downs already touched upon the role of lobbying in the political economy:

Men are much more likely to exert direct influence on government policy formation in their roles as producers than in their roles as consumers. In consequence, a demo-cratic government is usually biased in favor of producer interests and against con-sumer interests, even though the concon-sumers of any given product usually outnum-ber its producers. . . Government’s anticonsumer bias occurs because consumers ra-tionally seek to acquire only that information which provides a return larger than its costs. (Downs, 1957, p. 149)

Hence, Downs hints at some notion of political bias that arises as a result of lobbying, to wit, the idea that the group that is lobbying is skewed in favor of producers. In the tradition of Downs, albeit without the strict juxtaposition between consumer and producer, the concept of lobbying is often researched in terms of its consequences for political equality. The remainder of this section will look into economic literature covering lobbying and its effect, focusing on Special Interest Groups in general, and Political Action Committees more specifically.

2.1 Special Interest Groups (SIGs)

Special interest groups, or SIGs, can be defined in a variety of ways. In the context of this research, a SIG refers to an association of individuals or organizations that aim to influence the government in pursuance of a shared goal following their common demographic, cultural, or economic characteristic. In the economic literature there is a substantial and growing body of work on SIGs and their relation to the economy.

level can have influence that extends beyond U.S. Congress. Caldeira and Wright find “evidence of the efficacy of interest groups in lobbying the Senate during the fights over the nominations of Bork, Souter, and Thomas to the Supreme Court. At least for judicial nominations, interest group lobbying constitutes an important part of the legislative calculus” (1998, p. 520.) Hence, the influence of SIGs on a federal level reaches both the Congressional and Judicial branches.

In addition to this predominantly empirical body of literature, there is also a significant amount of theoretical economic research on SIGs and their mechanisms. These studies on in-terest groups have helped, through theoretical modeling, in gaining insight in what ways SIGs, PACs, or lobbyists might influence politicians and political decision-making. The seminal eco-nomic text on lobbying in general and SIGs in particular is Grossman and Helpman’s Special Interest Politics (2001). Herein, the authors theoretically assess the mechanisms through which SIGs affect policy-making.

It is important to note that campaign contributions can be part of different lobbying strate-gies. A SIG or PAC can try to influence elections in such a way that their preferred candidate wins the elections (electoral motive). Conversely, SIGs can contribute to multiple candidates or the one most likely to win in order to gain access to the (future) political realms and/or acquire political favors (influence motive). Research on the allocation of political contributions gives in-sight in the strategy of SIGs, albeit that the evidence is contradicting on this issue. Welch (1980) shows that campaign contributions from economic interest groups are predominantly going to likely winners of the elections, suggesting that political favors and access are preferred over in-fluencing the outcomes of elections. The opposite is found by Levitt (1998), who concludes that the “election-influencing channel" is in fact relatively important compared to political favors. Gross-man and HelpGross-man (1996) find sufficient evidence in that suggests that the “influence” motive is in fact more important than the “electoral” motive.

Nonetheless, in Special Interest Politics, Grossman and Helpman consider both the case of attempts to directly influence politicians and the case of electoral motives. The former situation is modeled in a variety of ways. The most parsimonious one molds the problem of a SIG and a politician in terms of trade-off of each party: Both the SIG and politician face a trade-off between policy and money–campaign funds for the politician, lobbying costs for the SIG. For the politician however, the policy outcome in turn affects the chance of reelection. Both seek to maximize a utility function given their respective trade-offs. A striking finding using such a simple model is that, in case of two rivaling interest groups, but of whom only one is organized, a politician is induced to treat the unorganized and non-contributing interest groups with reduced weight. This unequal concern of the politician produces an loss of aggregate welfare. Finally, as a result, we see that the politician’s political cost of his actions–that is, reduced chance of reelection–is what determines the contribution.

outcomes rest on certain limitations of the model though: There is no difference in power of the interest groups, and therefore the competition among SIGs is perfectly equal–the differences in outcomes rest predominantly in the underlying economic assumptions.7

From the setting of competing SIGs a few observations can be made. First, any equilibrium for such a setting of competition among SIGs hinges on political contributions which often are wasteful since SIGs members’ rents dissipate as a consequence. Indeed, many economists have argued that competition among advocacy groups is likely to cause a deadweight loss and in this way is wasteful on a societal level (Tullock, 1967; Krueger, 1974; Grossman and Helpman, 2001). Another thing to note is that competition among interest groups may not always be equal or per-fect. Dixit (1996) for instance observes a discrepancy between producers and consumers akin to Anthony Downs’ argument (1957): The incentive to cooperate and promote special interests is far stronger for producers than for consumers, since factors of production give rise to special interest much sooner than “distinctive tastes in consumption” (Dixit, 1996, p. 376). Hence, competition among SIGs itself, under the assumption of perfect competition, has adverse effects of dissipating rents. Moreover, the assumption of equal competition for influence among SIGs is one that may not be so accurate.

In addition to welfare-reducing effects of lobbying, academic theory on lobbying also often stresses the welfare enhancing effect of the provision of information by lobbyists to government officials. Lobbyists, in order to effectively promote their client’s interests, can become well-versed with the issues at stake, and sometimes assume the role of educator or advisor towards gov-ernment officials. Grossman and Helpman assert that due to the fact that lobbyists are knowl-edgeable on the issues at stake “in principle could allow the politicians to make better policy decisions” (2001, p. 105). However, the importance of the expertise of lobbyists may be some-what overstated. Bertrand, Bombardini and Trebbi (2014) find that it is more important for lobbyists to be well-connected than to be well-informed. Hence, mere access is more important than expertise after all. Additionally, Bennedsen and Feldmann (2006) find that lobbyists face an externality associated with becoming knowledgeable on the issues, which is the search cost of information. As a result, if an interest group is able to make campaign contributions instead, it will do so in favor of incurring search costs. However, information provisions can be useful in a competitive setting, since interest groups can provide information along with contributions to positively distinguish themselves from its competition.

2.2 Lobbying via Campaign Contributions

One of the most important forms of lobbying is through supporting politicians in their campaigns via financial contributions. Often referred to as campaign finance, this way of trying to influ-ence political outcomes has grown in popularity in the last two decades. Figure 1 depicts the development of campaign contributions to federal and state level government officials from 1990 onwards. Clearly, finances have increased significantly; financial donations to federal level gov-ernment bodies–comprising predominantly of donations to members of U.S. Congress–grew from little over a billion dollars in 1990 to more than nine times as much in 2012. For the state-level

Figure 1. Campaign Contributions (CCs) to Members of Federal and State Level

Government Bodies in the United States, 1990-2014

0 2000 4000 6000 8000 10000 C on tr ib u ti on s in m ln $ 1990 1995 2000 2005 2010 Year

CCs to Members of Congress CCs to State-level Government Officials

contributions, a similar tendency can be seen in Figure 1, albeit that in 2012, these contribu-tions accumulated to less than half of those to federal officials. Note that both lines exhibit a cyclic development. This cyclicity arises from the fact that every other period (that is, every four years) a presidential election takes place in addition to the elections of Figure 1. Apparently, the contributions to non-presidential elections increase in presidential years, perhaps because political engagement grows in these times. In the United States, an important group of actors in campaign finance are called Political Action Committees (PACs) and Super PACs. The PACs pool contributions from its members, which they subsequently donate to a political campaign. Unlike PACs, Super PACs are entities that are prohibited from contributing directly to, as well as coor-dinating with a political campaign, but can spend money on independent initiatives promoting or disproving a political candidate.

2.2.1 The Legal Framework of Campaign Finance and PACs

2010 Supreme Court case Citizens United v. FEC essentially removed all restrictions on non-profit corporations’ independent political expenses.8 Also in 2010, the U.S. District Court for the District of Columbia ruled in favor of SpeechNOW.org in the case SpeechNOW.org v. FEC, which

implied an end of contribution limits on independent groups.9 These two 2010 cases combined

brought about the inception of the so-called Super PACs: committees that can solely influence elections via political spendings that are independent from campaigns. The crucial differences between PACs and Super PACs therefore are: PACs can contribute a limited amount of money to a political campaign, whereas Super PACs can spend unlimited amounts of money on efforts to influence elections, but cannot directly contribute to a campaign.

2.2.2 The Sources of Campaign Contributions

Campaign contributions are crucially important in covering the costs of political campaigns. PACs account for a significant part of these costs, and therefore are often focused on in aca-demic research. An example of the effectiveness of PACs as lobbying instruments is provided by Brooks, Cameron and Carter (1998). They find that in the case of congressional votes on U.S. Sugar regulation that the votes have been bought via contributions of both the pro-sugar and anti-sugar coalitions. The found efficacy of PAC contributions “reflects the dominance of political pressure over ideological concerns” (Brooks et al., 1998, p. 441). Nevertheless, there is not a wide body of literature on campaign contributions affecting voting behavior or election outcomes. A possible explanation for this might be the endogeneity of campaign spending in relation to elec-tion outcomes (Gilens, 2012). Since an incumbent facing a tough and bright challenger will have a higher incentive to raise more campaign funds relative to one facing an incompetent challenger, and therefore higher spending can, in some cases, have an adverse effect on election outcomes.

The importance of PACs to campaign financing is itself also not undisputed. Ansolabehere et al. (2003) for instance assert that the contributions of PACs are not the most important ones: The individual contributions are still more voluminous than PAC contributions. In fact, they ar-gue, the marginal dollar of contribution is contributed by individual, small donors, which, accord-ing to Ansolabehere et al., debunks the common assertion that politicians rely in their campaign funding largely on interest groups. In turn, this assertion assumes that individual campaign con-tributions are mostly done by so-called small donors, which is highly contestable. Additionally, those that provide larger donations are more likely to be the contributor of the marginal dollar needed if they indeed cover most of the costs of the campaign.

Like PAC contributions, individual campaign contributions account for a large part of the costs of political campaigns, and their numbers have increased significantly. Whereas in 1980, 224,322 individual contributions were made, in 2012 this number had increased to 3,135,564 (Bonica et al., 2013). However, this increase in individual contributions is skewed, as observed by Bonica et al. (2013, p.111 ): “the share of total income received by the top 0.01 percent of households is about 5 percent but [...] the share of campaign contributions made by the top

0.01 percent of the voting age population is now over 40 percent.” This thus reflects a higher willingness and tendency of the most affluent of U.S. citizens to contribute to campaigns.

All in all, though PACs account for a significant amount of campaign expenses, individual con-tributions account for a larger part of concon-tributions, as noted by both Ansolabehere et al. (2003) and Bonica et al. (2013). PACs often channel the donations of interest groups and corporations, while individual contributions are a direct way of citizens participating. However, these individ-ual contributions are in turn dominated by the donations of the wealthy. Since this article looks into the campaign contributions as a whole, the distinction between contributions from PACs and from individuals is not necessarily important; more so, the possibly bias in the campaign contri-butions as a whole is the issue of interest here, and a bias appears to be ingrained in both PAC and individual contributions.

3

T

E

I

I

Inequality is covered extensively in economic literature, and, in recent years, more and more attention is devoted to this issue. This growing interest in inequality could very well stem from the fact that inequality is in fact rising in many Western societies. Moreover, among the Western economies, the United States has experienced a relatively strong increase in income inequality in recent decades. So much so, that it currently ranks among the OECD countries with highest income inequality (Cingano, 2014). Figure 2 depicts the development of inequality in the United States as measured by the Gini Coefficient and the income share of the top one percent from 1960 to 2013. Clearly, inequality has been almost exclusively on the rise in this time frame, starting at a Gini Coefficient below 0.45 in 1960, and approaching 0.65 in 2013. This graph clearly reinforces the notion that inequality in the U.S. is growing. This section will touch upon some of the vast body of–predominantly economic–literature devoted to income inequality in the United States, focusing specifically on the ‘one percent,’ political inequality, and hence, lobbying.

Figure 2. The Rising Inequality and Income Share of the Top 1% in the United States, 1960-2013 10 15 20 25 In com e S h a re (Per cen t) .45 .5 .55 .6 .65 G in i C oeffi ci en t 1960 1970 1980 1990 2000 2010 Year

Gini Coefficient Income Share Top 1%

note, those of the bottom 99 percent rose by 18.9 percent. More recently, the American economic elite appears to have profited most from the recovery of the 2008 financial crisis: Saez (2013) finds that between 2009 and 2012, the top one percent took hold of 95 percent of the income gains. Indeed, focusing just on those Americans that have the highest incomes already indicates a significant gap between the rich and poor is developing.

Two important books on the increasing inequality and what this might implicate were writ-ten by Joseph Stiglitz. In The Price of Inequality he states that “[w]idely unequal societies do not function efficiently, and their economies are neither stable nor sustainable in the long term” (Stiglitz, 2012, p. 56). In The Great Divide, Stiglitz further outlines the inefficiencies of high inequality. One example of how inequality hampers economic growth is the fact that a concentra-tion of money at the top will induce a decline of aggregate demand (Stiglitz, 2015). This is clearly one of the most important possible consequence of inequality for economists, which is reflected by the amount of research that has been done on the interplay between economic growth and income inequality.

di-rections have been found, albeit at times contradictory, it should nevertheless be included into this research.

In addition to its relation with economic growth, the issue of inequality is researched in rela-tion to various other economic variables. Other economic factors discussed in economic literature in relation to income inequality are, among others, political (in)stability (Alesina and Perotti, 1996), educational attainment or human capital (Glomm and Ravikumar, 1992; Gregorio and Lee, 2002), public spending (Afonso et al., 2010), constitutional structures (Birchfield and Crepaz, 1998), and financial development (Beck, Demirgüç-Kunt and Levine, 2007). Though not impor-tant to expand upon here, it should be noted that the economic literature on income inequality is characterized by dissonant outcomes. “It is remarkable how little agreement there is between the findings of various scholars attempting to isolate the most crucial predictors of income in-equality,” Birchfield and Crepaz (1998, p. 176) observe. Moreover, the issue of inequality is not universally perceived as a bad thing. The way in which one perceives inequality often depends largely on philosophical convictions. A common claim in defense of inequality is that it provides incentives to produce and innovate. Nevertheless, “the case for inequality seems to rest primar-ily on the claim that it promotes efficiency,” Jencks (2002, p. 64) observes, “and the evidence for that claim is thin.” Though a plethora of research does indicate inequality has harmful effects, normatively it is still a disputed topic.10

Besides the economic implications of income inequality, there also is a body of literature that looks into the relation between inequality and policy-making. Martin Gilens (2012) for instance, finds that policy outcomes in the United States are markedly skewed towards the preferences of the American upper-class. In his book Affluence and Influence, Gilens thus finds evidence for economic inequality spilling over into the political realm. This notion of economic inequality inducing or exacerbating political inequality is in fact found more often. In a similar vain as Gilens, but at the opposite end of the income spectrum, Rigby and Wright (2013) find no evidence for any response of political parties in the 50 American States to the preferences of the lowest income group, unless the preferences overlap with those of the high-income group. The strongest disparity they find is in responses pertaining economic issues. Analogously, an analysis of roll call voting behavior of U.S. Senators has provided evidence of Senators being much more responsive to their affluent constituents than to the poor constituents (Bartels, 2009). A caveat for this finding is that the higher classes often reflect a higher propensity to vote, contact, and influence their Senator. Solt (2008) finds that political engagement decreases as inequality increase, since the lower classes deem the democratic process unhelpful in advancing their preferences and interests. According to Bartels, this explains the disparity in the representation of the poor and affluent only in part. Income inequality thus has its effects on political outcomes, though this is not a one-way relationship.

As with inequality and growth, the relation between political and income inequality features a reverse causality. Naturally, political decision-making affects income inequality. For this rea-son, Bonica et al. (2013) insist we refrain from downplaying the role of politics in determining the 10_{Most notably, an overwhelming body of literature has found that income inequality hampers (long-run) economic}

causes of income inequality. One path in which politics can affect income inequality is the way in which it is organized. For example, the structure of the political realm can give rise to political inequality. Gilens and Page (2014) find evidence that American politics adheres to the theoret-ical notions of economic-elite domination and biased interest group pluralism. Using a study of 1779 American policy issues, they find that “interest-group alignments are not significantly related to the preferences of average citizens” (p. 576). Moreover, this study revealed that the impact of business-representing interest groups and the economic elite on government policies is considerable, as opposed to the average citizens and the interest groups that are aligned with their preferences, who have little, if any, impact.11 Much of what is written about this particular order of causality has to do with the right to petition the government, and is therefore already described in the previous section.

Clearly, there is some sort of interplay between the process of political decision-making and income inequality, wherein each perpetuates the other. The remainder of this article will there-fore scrutinize a part of this causal nexus, by looking at the effect of one aspect of the (unequal) process of political decision-making–that is, campaign contributions as a proxy for lobbying–on income inequality. The simultaneous causality of politics and income inequality indicates that endogeneity of the campaign contribution variables is something to take into account in the up-coming sections.

To substantiate the finding in this section and the previous one that there is some interplay between the–at times–unequal political decision-making process in the United States and income inequality, the remainder of this article will explore the effect of lobbying, proxied by campaign contributions to American politicians, on income inequality. The findings presented in this sec-tion indicate that campaign contribusec-tions mainly come from the affluent Americans and large corporations, and aim to promote the interest of these factions of U.S. society. For this reason, I hypothesize:

Hypothesis 1: The effect of campaign contributions on income inequality is positive. In addition to this key hypothesis, some subordinated hypotheses can be formulated. First, as was established in this section, the interplay between political inequality and economic inequality can very well be characterized by reverse causation. This could potentially be problematic, and should be controlled for in the remainder of this study. Additionally, an explicit assumption about the direction of causality will be made.

Hypothesis 2: Regarding the relationship examined in this study, the causality runs from the system of campaign finance to the income distribution.

11_{Note that the theoretic construct of Grossman and Helpman (2001) assumes perfect competition between SIGs in}

Finally, this article will look at the spatial interactions surrounding the interplay of campaign finance and income inequality. More specifically, since this study looks at the central government in which Representatives and Senators represent both their state’s and America’s interests as a whole, spatial dimensions could be in play as well. The 50 U.S. states have rather strong economic and political ties, which could indicate that income inequality itself will exhibit some spatial interdependencies, as well as that a state’s campaign contributions can have an effect not only on its own income inequality, but also be associated with income inequality in bordering states.

Hypothesis 3: Campaign contributions have a positive spatial spillover (indirect) effect on income inequality, implying that the effect of campaign contributions to government officials representing

state i are positively associated with income inequality in state j ( j 666=== i).

In what follows, these hypotheses will be tested. In order to do so, several steps have to be taken. The next section will present the data and variables that will be used to test this hypothe-sis. Most importantly, this section will outline the campaign contribution-variables, as well as the various measures of income inequality. Section 5 will then outline the econometric specifications with which statistical inference will be done. Section 6 will subsequently present the results of the various regressions, and discuss what these results entail and implicate. Finally, section 7 will conclude by summarizing the main findings and accentuate the ramifications thereof.

4

D

This section will discuss and describe the data used for the empirical research. First, the data on campaign contributions will be outlined. Subsequently, the variables that will be used as depen-dent variables, that is, the several variables that measure income inequality, will be discussed. Thereupon, the control variables shall be discussed. Summary statistics can be found in Table 1, and the cross-correlations in Table 9 (Appendix C).

4.1 Data on Campaign Contributions

The data on campaign contributions are available due to the enactment of the 1995 Lobbying Disclosure Act. This and other laws regarding campaign finance disclosure laws require the recipient of a contribution to report all those in excess of $200 to the Federal Election Commision (FEC). The report of the recipient ought to contain the donor’s name, occupation, address, and employer. These reports are collectively provided to the public by the FEC, albeit that the data is not cleaned or standardized, which makes the data somewhat incomprehensible.

Though the datasets as provided by Influence Explorer are clean relative to how it is released by the FEC, they are still rather raw. Data on campaign contributions to state-level governmental bodies comprises of five different governmental bodies for each state: Upper Houses (i.e. Senate), Lower Houses (i.e. House of Representative, House of Delegates, etc.), the Judiciary branch, and two variables from the executive branch, to wit, the Governor and Offices. On federal level, the dataset comprises of both bodies of the United States Congress: the House of Representatives and the Senate.12 Since all the members of these two federal governmental bodies represent a state, the data of federal campaign contributions can, just like the state-level data, be categorized according to the state13. The dataset for Federal Campaign contributions contains more than 28 million observations–each observation being one filing report by a recipient of a contribution. Likewise, the state-level data comprises of more than 29 million contribution filings. Table 7 (Appendix A) provides the summary statistics of both datasets per election cycle.

In order to do any statistical inference with this data, further cleaning and categorizing is needed. Both datasets have been altered as follows: The millions of observations and correspond-ing contribution amounts have been accumulated accordcorrespond-ing to the state the recipient represents, the governmental body (e.g. Senate, House of Representatives, Governor) and the election cycle year. This results in seven variables, one for each of the government bodies. Each of these vari-ables have the total amount of campaign contributions for state i, where i = 1,2,... N, and for year t, t = 1,2,... T. Of these seven variables, five are the state level variables, Jud, O f f , Gov, U p, and Low, and two variables capture the federal campaign contribution variables, namely H oR and S en. Table 1 presents the summary statistics for all variables described in this section (As well as the variables to be discussed hereafter). In light of the fact that these campaign con-tribution variables have high values, while other variables, most notably the income inequality indices, have very low values, the campaign contribution variables are expressed in millions of dollars so that they are more easily interpreted in the remainder of this article.

The federal level lobbying data is available for the years 1990 to 2014, whereas the state-level data ranges from 1990 to 2012. Since elections are held every two years in American governmen-tal bodies, it is best to assume a biannual time-span. The contributions in off-election years are destined for the election in the year after, and since the contributions increase the sooner the election takes place, an annual time-interval will result in an election year-off election year sea-sonality that will distort econometric inference. The state-level variables have several missing values, whereas the federal-level variables are fully balanced without gaps. This results in 474 observations for the state-level variables, and 650 observations for the federal level variables.

As can be seen in the summary statistics (Table 7) in Appendix A, both databases include var-ious negative campaign contributions. A negative campaign contributions comes across rather counterintuitive, as no politician is expected to hand out contributions to for example lobbying firms. According to Influence Explorer, this can mean either of two things: Either (part of) the contribution is being refunded by the recipient, or other, more complex accounting maneuvers are

12_{Figure 1 is compiled using these two state- and federal-level datasets.}

13_{Whereas each State has two Senators in Senate, the number of Representatives for each state in the House of}

the reason for a negative contribution. Whereas refunded contributions should be incorporated, other accounting maneuvers may cause an erroneously low cumulative amount of campaign con-tributions, which implies that all seven campaign contributions will have lower values. However, it is not possible to distinguish between the different reasons for the negative contributions, and therefore, they can only be collectively kept or removed. To this end, the negative contributions will not be omitted from the dataset with which inference will be done. However, as a robustness check, the same calculations will be done with data from which the negative contributions are omitted.

Table 1. Summary statistics

(1) (2) (3) (4) (5)

VARIABLES Obs Mean Std. Dev. Min Max

HoR 650 14.196 19.967 .111 193.539 Sen 650 10.672 16.045 -.001 123.831 Jud 474 .857 1.971 0 13.504 Off 474 3.434 7.663 -.003 71.107 Gov 474 10.38 22.502 -.167 253.51 Up 474 6.207 8.821 0 49.274 Low 474 9.498 14.518 0 93.233 Gini 600 .586 .036 .526 .711 Theil 600 .776 .183 .47 1.498 Atkin05 600 .266 .038 .2 .411 RMeanDev 600 .818 .051 .721 1.013 Top 10% 600 42.356 5.106 32.240 61.945 Top 1% 600 16.334 4.583 8.720 34.422 Top 0.1% 600 7.353 3.1873 2.861 22.457 Top 0.01% 600 3.283 1.978 .9134 14.198 HighSchool 637 .62 .046 .464 .738 College 637 .172 .043 .079 .304 PCPI 585 31869.21 10438.02 13288 69838 UR 585 5.645 1.797 2.3 13.533

4.2 Data on Income Inequality

The data on Income inequality is drawn from a database provided by Mark Frank and used in his articles on inequality (Frank, 2009, 2014). Frank’s dataset includes several state-level income inequality measures: the Gini Coefficient, the Theil (Entropy) Index, the Atkinson Index, and the Mean-Deviation Inequality Index. Furthermore, in addition to inequality indices, income share measures will be used to proxy the development of the incomes of America’s economic elite. These income share measures were prepared for the World Top Income Database by Mark Frank, Estelle Sommeiller, and Mark Price, which is based on Internal Revenue Service’s income data.14 14_{Please consult Frank (2014) and Sommeiller and Price (2015) for the specifics regarding the construction of these}

4.2.1 Income Inequality Indices

This paper employs four different variables as inequality measures. This section will concisely describe these four indices.15 As can be seen in Table 9 in the appendix, these four inequality all have a relatively high correlation. The Gini Coefficient and Relative Mean Variation and the Theil and Atkinson indices are particularly similar according to their cross-correlations.

The Gini Coefficient

The Gini coefficient is the income inequality measure that is, presumably, used most often. So-ciologist Corrado Gini introduced this measure. The coefficient’s value ranges between zero and one, where zero indicates perfect equality–when x percent of income goes to x percent of the population–and one marks perfect inequality. Note that Figure 2 depicts this inequality mea-sure.

The Theil Index

The Theil Index, which belongs to the familiy of Generalized Entropy measures, is an inequality index that uses ratios of income percentiles. According to Frank (2009), it is the most attractive measure in analyzing inequality, because this index is decomposable. It is named after econome-trician Henry Theil. The index includes a sensitivity parameterα, which reflects the sensitivity

of higher incomes. The Theil Index data as provided by Mark Frank’s database, has an αwith

value 1, which means that it is more sensitive to the higher incomes than to the lower incomes in determining inequality. As with the Gini coefficient, the Theil index takes higher values for higher inequality. The Theil Index has, unlike the other inequality measures used, no upper bound, as can also be seen in Table 1, given that its maximum value is 1.498. Furthermore, following the mean values, the Theil index takes higher values relative to the Gini coefficient.

The Atkinson Coefficient

The next inequality measure is one that is grounded in social welfare theory. British economist Antony Atkinson created this measure, and on making an inequality measure he writes: “a com-plete ranking of distributions cannot be reached without fully specifying the form of the social welfare function” (1970, p. 262). Adding a social welfare function allows this index to be grounded in value judgments regarding social welfare; it adds a normative facet to this inequality measure. His index should therefore serve as an alternative to measures such as the Gini coefficient, which according to Atkinson, do not specify a social welfare function. Another reason why a measure such as the Gini coefficient is not always adequate according to Atkinson is that this measures features a constant inequality aversion: the degree of people preferring equality is invariable16. The Atkinson index explicitly incorporates the preference of equality through its inequality aver-sion parameter (_ε). The database used features an inequality aversion parameter of 0.5, which describes a rather low inequality aversion. A somewhat low inequality aversion parameter is

probably a good description for the United States in light of the U.S.’ libertarian character and– relative to most other Western economies–limited welfare system.

Like the Gini coefficient, the Atkinson Index ranges between zero and one, and a higher value indicates more inequality. Furthermore, the Atkinson has an attractive analytic property, as De Maio notes: “Atkinson values can be used to calculate the proportion of total income that would be required to achieve an equal level of social welfare as at present if incomes were perfectly distributed. For example, an Atkinson index value of 0.20 suggests that we could achieve the same level of social welfare with only 1-0.20 = 80% of income” (2007, p. 850). As can be seen in the summary statistics, the value of the Atkinson Index is significantly lower than those of the Theil index and Gini coefficient, with a mean of 0.266.

Relative Mean Deviation

Finally, the Relative Mean Deviation. This measure captures the average distance between the population’s mean income and each individual’s income. It is has a lower- and upper bound of zero and two, and takes higher values for higher inequality. Unlike the three inequality mea-sures outlines above, the Relative Mean Deviation violates the Pigou-Dalton principle in case of transfers in income between two individuals on the same side of the mean income. A transfer in income from the poorest individual in a population to one that has on income only slightly under the average income should alter income inequality, but does not necessarily change the Relative Mean Deviation. For this reason, Frank (2009) refers to this inequality measure as the one that has least analytical value. This measure is nevertheless incorporated for it can serve as a robust-ness check. Table 1 indicates that the mean value of the Relative Mean Deviation is higher than all other inequality indices.

4.2.2 Top Income Shares

In additional to the inequality indices, this article also employs the income shares of those people with an income that places them in the top 10, 1, 0.1, or 0.01 percent of the United States’ income spectrum. These variables thus express the income of the top 10, 1, 0.1, or 0.01 percent in terms of the income of the bottom 90, 99, 99.9, or 99.99 percent. These variables can serve both as additional measures of inequality and as proxies for the American economic elite. Figure 2 depicts the income share of the top one percent variable for the United States as a whole; for this research the state-level values of this and the other percentiles are used. As can be observed in Table 1, these variables are expressed in percentage points, and therefore bound between zero and a hundred rather than between zero and one.

4.3 Control Variables

Human Capital is a variable often used in macroeconomic literature in order to explain eco-nomic growth, but also income inequality. Educational attainment, for instance, often determines for a large part the wages and income of individuals. Human capital will be incorporated with the inclusion of two variables: High school attainment (HighSchool) and College attainment (College). Both variables are from the same database as the inequality and top income shares are from, and used for instance in Frank (2009). In this article, Frank finds an insignificant effect of Highschool on income inequality, and a significant and negative effect of college attainment. Both human capital variables are annually and on state level. The human capital levels are not available for New Mexico, and therefore this state will drop out of regressions including human capital as control variables.

Another variable often utilized in predicting income inequality is (per capita) GDP growth (Income or production). This article will employ Per Capita Personal Income (PCPI). Next, the Unemployment Rate (UR). It is likely to assume that the majority of the unemployed demographic will have income that are relatively low, and in this way the unemployment rate is likely to be positively associated with income inequality, since a high unemployment rate implies a larger share of the population having a low-income.

5

E

S

This section will outline and describe the estimation methods used in order to pervasively scruti-nize the effect of campaign contributions on income inequality. First, the Ordinary Least Squares (OLS) methodology will be presented, followed by Dynamic Panel Estimation using the Gener-alized Method of Moments (GMM) panel estimation. Finally, a spatial econometric section will provide a more in depth instrument to examine the relationship.

Before discussing the various econometric specifications, an intermediate step must be taken. In order to see whether the variables can be added to the regressions in level form, all variables used in the following specifications should not have a unit root. If certain variables do have a unit root, they cannot be used for statistical inference as such, and possibly have to be differenced in order to use them. In order to check the stationarity of the variables described in this section, the existence of a unit root has been tested using the Levin-Lin-Chu, the Augmented Dickey-Fuller, and the Phillips-Perron tests.17 The results can be found in Table 8 in Appendix B. The null hypothesis of nonstationarity is rejected unanimously for all variables, which leads to the conclusion that all variables can be used in levels.

5.1 Ordinary Least Squares Methodology with Unobserved Effects

Of the panel data described in the previous section, the various income inequality and income share variables will serve as dependent variables, and will be regressed on the campaign contri-bution variables and some control variables. This gives rise to the following model:

Yit= X0_i,t−1β+ Z0_itγ+νit, (1)

where

νit=α+εit. (2) Herein, Yit denotes the dependent variable for state i at time t. This term can denote either the Gini Coefficient, Theil index, Atkinson index, Relative Mean Deviation, or the income share of the Top 10, 1, 0.1, or 0.01 percent. X0_itis a vector of the various campaign contribution variables. For the federal level, we have that

Xit= ( H oRit S enit ) ,

while for the state level variables, we have that

Xit=                    J udit O f f_it G ovit U pit Lowit                    .

Given that it is unlikely that campaign contributions have a contemporaneous effect on income inequality, the campaign contribution variables are lagged one period–which is, given that the data is biannual, two years. Hence, either of these Xitvectors enter the model with a lag of one period. In this way, the model measures the effect of campaign contributions of two years ago on the current income inequality. The vector Z0

it contains the control variables H i ghS choolit, Coll e geit, PCP Iit, and U Rit. Finally,αis the constant term andεitdenotes the error term.

It is important to note here that in the vector of coefficients_βwill lie the validation of the cen-tral hypothesis (Hypothesis 1) of this article, as outlined at the end of section three. If the effect of campaign contributions on income inequality is–in accordance with the hypothesis–positive, we have that, for any coefficient inβ, the following should hold:

βk> 0.

That is, the coefficient for campaign contributions variable k should be positive.

The next step is to add unobserved effects to the model. These unobserved effects can be treated either as random effects (uncorrelated with the explanatory variables) or as fixed effects (possibly correlated with the regressors). Furthermore, one can distinguish between state- and time-unobserved effects, and add either one or both. These unobserved effects can be added to Equation 1 by redefining_νit(Equation 2). First, adding a state systematic component gives

νit=α+ψi+εit,

where ψi is the state-unobserved effect parameter, and νit the error term. Conversely, adding time- instead of state-unobserved effects gives rise to

where ξt captures the time systematic component. Finally, adding both state-and time-effects yields:

νit=α+ψi+ξt+εit,

In the discussion of the results, it will first be described which unobserved effect should be added, and whether the unobserved effects are best incorporated as fixed or random effects, and hence, whether they are best estimated with the Fixed Effects (within) estimator or with the Random Effects estimator.

5.2 Generalized Method of Moments

Several cross-country panel studies of income inequality have utilized, either solely or as a com-plement to OLS, the GMM methodology (e.g. Beck et al., 2007; Dollar and Kraay, 2002; Heckel-man and Wilson, 2014). The main reason to employ the GMM estimator (in addition to OLS) is that it can control for the possible endogeneity of some of the explanatory variables. This prob-lem could also be addressed with the Two-Staged Least Squares estimator in combination with instrumental variables. However, with this methodology it will be extremely difficult to find at least one valid, exogenous, and quantifiable instrument for the campaign contribution-variables. With GMM, we can control for endogeneity of the explanatory variables of interest without hav-ing to find instrumental variables, because GMM utilizes internal instruments instead.

There is a possibility that the campaign contribution variables are subject to endogeneity or weak exogeneity. These variables may for instance be subject to reverse causality: besides campaign contributions affecting income inequality, one could argue that higher income inequal-ity in turn induces campaign contributions to increase, and hence, the causalinequal-ity to run in both ways. It may be the case that when income inequality increases, people on either end of the income-spectrum feel the need to protect their higher incomes or fight for higher incomes. Hence, the possibility of reverse causality between lobbying in the form of campaign donations and in-come inequality might lead to the campaign contribution variables suffering from endogeneity. To this end, system-GMM will be applied. This dynamic panel data estimator features a lagged dependent variable among the explanatory variables, such that Equation 1 can be rewritten as

Yit=λYi,t−1+ X0i,t−1β+ Z0itγ+α+ψi+εit. (3)

This model can be estimated with the system-GMM estimator18. This system-GMM panel

esti-mator, is outlined in Arellano and Bover (1995) and Blundell and Bond (1998), and founded on the difference-GMM estimator by Arellano and Bond (1991). What it entails is that a system of Equation 3 is estimated both in differences and in levels. The lagged values of the levels are used as instruments for the regression in differences, and, conversely, the lagged values of the differences are used as instrument for the level regression. In this way, the system-GMM esti-mator distills the exogenous element of the relations between the explanatory variables and the regressand.

Typically, the potentially endogenous variables are used as instruments–both in differences and in levels–with two lags and higher. Due to the somewhat low number of States in this research, the number of instruments exceeds the number of cross-section units under a specifica-tion with multiple lags of the campaign contribuspecifica-tion variables, which can lead to biased results, as Roodman (2006) accentuates. Therefore, the number of lags to be used is restricted depend-ing on how many instruments it produces. In this way, the system-GMM estimator will allow for endogeneity or weak exogeneity of the campaign contributions, and give consistent results. Furthermore, its first-differencing component ensures that the unobserved individual effect (ψi) is removed along with the omitted-variable bias associated with it. The unobserved individual components are included via the use of cluster-robust standard errors. Hence, the system-GMM analysis will include regressors that may be endogenous (the lagged dependent variable and the campaign contribution variables). Because the campaign contribution variables are the main focal point of this article, the control variables are assumed to be predetermined, and these are therefore used as instruments in level-form only. Finally, in light of some gaps that arise in the data with the inclusion of the economic control variables, the loss of data is minimized by using forward orthogonal deviations transform instead of first-differencing.

Finally, a note on the internal instruments being appropriate or not. In order to determine the validity of the instruments for the endogenous variables, two tests are important. First, the Hansen test for over-identifying restrictions, of which the null hypothesis should not be rejected. Secondly, the Arrellano-Bond estimator second-order autocorrelation should indicate that there is no second order autocorrelation present in the differenced residuals. If this is the case, deeper lags–and hence, deeper serial correlation tests–can produce appropriate instruments. If an ef-fect of campaign contributions on income inequality is found, and the internal instruments are proven to be valid, hypothesis 2 is validated insofar that the result obtained by the system-GMM regressions are robust to possible reverse causality.

5.3 Spatial Econometric Analysis

Whereas the main focus of the OLS and GMM methodologies described above is the direct effect of campaign contributions on income inequality, this section will outline an econometric method with which an indirect effect can be tested. Given the relatively close economic and political ties between a lot of American states, it is interesting to investigate the existence of a similar inter-connection between campaign contributions to Senators and Representatives of a certain state with the income inequality in the states bordering the state in question. In order to investigate such regional interdependency among states, and hence, test the third hypothesis as outlined at the end of Section 3, this section will turn to spatial econometric methodology.

In order to be able to compute such spatial interactions, a so-called spatial weight matrix (W) can be used. The spatial weight matrix captures the spatial relations among the states. For the purpose of this article, the W matrix will be a row-normalized binary contiguity matrix. First, a binary contiguity matrix for the United States is constructed. Herein, each element w_{i j} is equal to one if the ith and jth states (i 6= j) are contiguous and wi j = 0 otherwise. This symmetric binary contiguity matrix is subsequently row-normalized, which means that every element wi jis divided by the sum of the ith row. Consequently, the sum of each row of the W matrix is equal to one. The various spatial econometric outlined here each incorporate W in a different way.

Several spatial econometric formulations have been developed, however, not all are suitable here since some do not allow for the estimation of a spatial spillover effect. In order to see what is the best point of departure, it is advisable to turn to that specification that is most appropriate according to the theory. Since the theoretic literature on campaign contributions has to date not yet touched upon spatial interactions, this is not possible here. In this case, Halleck Vega and Elhorst (2015) advocate to take the Spatial Lag of X (SLX) model as point of departure. The main reason to take the SLX model as point of departure is that this model–unlike for instance the often used Spatial Autoregressive (SAR) and Spatial Autoregressive combined (SAC) models– produces clear, flexible, and easily observable spillover effects, which is the main objective of spatial econometric research. First, the SLX will therefore be outlined, followed by the Spatial Durbin Error Model (SDEM) and the Spatial Durbin Model (SDM). The former model is estimated using the OLS estimator, whereas the SDEM and SDM models are computed with Maximum Likelihood Estimation.19

5.3.1 The Spatial Lag of X (SLX) Model

In order to obtain the Spatial Lag of X model, the spatial weight matrix W is added to Equation 1, such that

Yit= X0itβ+ Z0itγ+ WX0itθ+ WZ0itφ+α+εit. (4) This is the SLX model, which means that, with respect to Equation 1, we have added a spatial lag of all explanatory variables by premultiplying them with W. This means that the effect of the explanatory variables now has two different coefficients: the lobbying variables will have an element of the vector of coefficients β as well as an element of the vector of coefficientsθ. The_βcoefficients will represent the direct effects of the campaign contributions, whereas the_θ coefficients will capture the indirect, or spatial spillover effect.

The next step in the spatial econometric method is to check whether the starting point–the SLX model–is the best fitting specification, or whether alternative specifications are more suit-able. To this end, several alternative specifications will be considered: The Spatial Durbin Error Model (SDEM), as well as the Spatial Durbin Model (SDM).

5.3.2 The Spatial Durbin Error Model

The Spatial Durbin Error Model has, in addition to the spatially lagged regressors, an interac-tion effect among the error terms included. That is to say, the error term of state i depends on other state’s (j 6= i) error terms. Hence, this terms captures unobserved shocks that follow a spatial pattern, thereby including possible spatial autocorrelation of omitted determinants of the regressand. This model takes the following form:

Y_{it= X}0_{itβ+ Z}0_{itγ+ WXit}0 _{θ+ WZ}0_{itφ+α+ uit}, (5) where

u_it=λW u +εit. Hence,λcaptures the interaction effect among the errors.

As with the SLX model, the direct and indirect effects of the SDEM are captured by the coefficients of the ‘normal’ and spatially lagged explanatory variables. That is, the coefficients of β and γ will capture the direct effects of the campaign contribution variables and control variables, whereas the coefficients of_θand_φwill denote the spatial spillover effects.

5.3.3 Spatial Durbin Model

The Spatial Durbin Model (SDM) adds an endogenous interaction effect to the SLX specification. The endogenous interaction effect is the spatial lag of the dependent variables, and thus implies an effect of the dependent variable of state i on the dependent variable on another state (j 6= i). This gives rise to the following expression:

Yit=ρWYit+ X0_itβ+ Z0_itγ+ WX0_itθ+ WZ0_itφ+α+εit, (6) whereρis the coefficient of the endogenous interaction effect. The direct and indirect effects as measured by the SDM model are not simply captured by theβ,γ,θ, andφ parameters, as was the case for the SLX model and SDEM. To illustrate this, we rewrite Equation 6 such that:

Yit= (I −ρW)−1(X0itβ+ Z0itγ+ WX0itθ+ WZ0itφ+α+εit). (7) Subsequently, the direct and indirect effects can be derived by differentiating this expression with respect to the variable of interest. For instance, the matrix of partial derivatives for Yitwith respect to H oRitthus becomes:

h _∂Y it ∂HoRit . . . ∂YN T ∂HoRN T i = (I −ρW)−1        β1 w12θ1 . . . w1nθ1 w21θ1 β1 . . . w2nθ1 .. . ... . .. ... w_n1θ1 w_n2 . . . _β1        ,

6

R

& D

This section will present the results of the various econometric tests outlined in the previous section. First, the Two Way Fixed Effects results will be presented. After that, the Generalized Method of Moments outcomes will be discussed, followed by the spatial econometric results. Sub-sequently, some robustness checks will be outlined and the results thereof presented. Thereafter, the main results from these different methods will be further discussed altogether, focusing on the campaign contributions’ effect on income inequality, as well as the implications of the findings presented here.

6.1 Empirical Results

6.1.1 Two Way Fixed Effects

Before discussing the results, the right specification for our model and data must be determined. In essence, this implies that first, as a point of departure, Equation 1 must be calculated. Tests should then describe the behavior of the errors of the model. Next, the validity of including the unobserved effects must be tested, as well as the appropriate estimator with which these effects should be estimated. These three extended models must subsequently be estimated using both the fixed effects (within) estimator and the random effects estimator, and a test should be performed to see which estimator is most appropriate. Additionally, tests must indicate whether state- and/or time-unobserved effects must be added.

First, upon computation of Equation 1, tests can give insight on the behavior of the errors of the model. This is important since the OLS, Fixed effects, and Random effects estimators are biased in case the errors suffer from heteroskedasticity or serial correlation. To this end, several tests to the various specifications have been done. All results of the tests described here can be found in Table 10 (Appendix D). For Pooled-OLS regressions (Equation 1) the Breusch-Pagan test for heteroskedasticity rejects the null of homoskedastic errors soundly and across all independent variables. This suggests that the conventional errors are probably not suitable to use. Similarly, the Modified Wald test, based on Greene (1997), rejects the null of homoskedasticity for all re-gressands. This tests, unlike the Breusch-Pagan test for homoskedasticity, tests for groupwise heteroskedastic errors in the fixed effects regression model. Hence, this indicates that the model suffers from heteroskedastic errors with and without the inclusion of fixed effects. Furthermore, the Woolridge test for autocorrelation indicates that the errors also exhibit serial correlation. This implies that the results will be biased as long as the conventional errors are used. To this end, heteroskedasticity and autocorrelation robust standard errors are used instead.