Distributive Politics and Legislator Ideology ∗
Dan Alexander, Christopher R. Berry, and William G. Howell
Harris School of Public Policy - The University of Chicago August 13, 2014
∗ Direct correspondence to Dan Alexander:
[email protected]
Abstract This paper examines the relationship between legislative centrism (or conversely, extremism) and the distribution of federal outlays. A substantial body of theoretical work suggests that legislators closer to the chamber median are more attractive and willing candidates to engage in vote buying and hence should receive a disproportionate share of distributive bene ts. We investigate this prediction empirically with panel data covering 27 years of federal outlays. Our research design allows us to isolate the eect of changes in distance from the median on the receipt of outlays, and we nd a 9% decrease in outlays associated with a one-standard deviation increase in ideological distance from the median. We nd the eect of exogenous increases in legislative extremism on outlays to be robust across a wide variety of speci cations, and we take special care to distinguish this eect from those induced by potentially confounding covariates, most notably majority party status.
As scholars going back at least to Ferejohn (1974) have recognized, distributive politics amounts to more than just a contest among legislators for scarce federal resources. It also involves the deliberate use of these monies to, as Evans (2004) puts it, grease the wheels of the legislative process. Whether to buttress pre-existing support for a bill against the lobbying of an opposing party or faction, or to purchase the support of a member who, absent the side payment, would vote against a particular bill, policy entrepreneurs routinely make use of outlays to cultivate support for their legislative initiatives (Evans 1994, Evans 2004, Wiseman 2004, Cann & Sidman 2011). Who within Congress is most likely to bene t from vote buying, an activity that, as Richard
1 Who, that is, stands to reap a greater share
Neustadt once quipped, is as traditional as apple pie?
of federal outlays from successive eorts to build legislative coalitions through side payments? From Snyder (1991) to Dekel, Jackson, and Wolinsky (2008, 2009), a substantial body of theoretical work sheds light on the matter, suggesting that it is ideological moderates who represent the most likely candidates to be engaged by vote buyers. Because they are more likely to be ideologically indierent (or close to indierent) between policy alternatives, moderate members should be, in theory, more frequent targets of congressional vote buying activities. In this paper we investigate the eect of legislative centrism on the geographic distribution of federal outlays. Using a member xed-eects research design to analyze distributive outlays over a 27-year period, we observe a statistically and economically signi cant positive eect of increased legislative centrism on county-level outlays. Speci cally, a one standard deviation increase in a representative's
1 As
cited in Evans (2004).
1
ideological proximity to the House median leads to a 9% increase in outlays received by her constituents. These ndings, we show, are robust to a wide variety of alternative speci cations and do not appear to be an artifact of majority party status. The paper proceeds as follows. In the rst section, we review the relevant theoretical literature on vote buying as well as existing empirical studies of distributive politics. After discussing our identi cation strategy, model speci cation, and data, we then present our main results. We subsequently scrutinize the role of majority party status in distributive politics and subject our core analyses to a variety of robustness checks and placebo tests. We conclude by placing our ndings in the context of related, ongoing questions in the study of U.S. legislative and electoral politics.
Vote Buying Vote-buying models, which have made substantial advances of late, typically posit a legislative environment in which one or more lobbyists compete over two possible legislative outcomes.
2 Such lobbyists
might be conceived of as either traditional interest groups and thus unable to cast votes themselves, non-voting elected o cials such as the president, or as actual voting members or blocs in a collective decision-making body. In any case, these lobbyists oer side payments to legislators in exchange for their votes, with payments usually being conditional on support. What constitutes the payments in these models is generally left unspeci ed, though Baron (2006) writes that lobbying consists of providing politically-valuable resources to legislators, a criterion that budgetary outlays certainly satisfy. The payments compensate legislators for voting against their or their constituents' beliefs, thereby justifying the ideological or electoral compromise. Snyder (1991) initiates the modern literature on vote buying with a model of a single, pricediscriminating lobbyist.
The model predicts that a lobbyist will make payments to those initially
opposed to her favored position until majority support is procured. As Snyder puts it, the lobbyist pays the highest bribes to legislators whose ideal points are closest to the median of the legislature, but on the side of the median closer to the lobbyist's proposal. That is, a lobbyist does not bribe his close supporters...but rather his marginal opponents. Snyder goes on to recognize that most empirical work on money in politics neglects this result. Having ignored the ability of lobbyists to price discriminate, Snyder speculates, previous researchers had mischaracterized the distributive consequences of vote buying. Much congressional vote-buying may occur in a legislative environment populated by just one
2 This
stands in contrast to alternative formulations of legislative bargaining games, such as Baron & Ferejohn (1989)
and Baron (1991), which do not include a role for parties or other organizations within or outside of Congress, instead limiting proposal power to individual legislators.
2
lobbyist. However, as Snyder (1991) recognizes and as Wiseman (2004) reiterates, multiple lobbyists may compete over opposing legislative outcomes.
Recognizing this possibility, a stream of political
economists developed extensive-form vote-buying models with competing lobbyists, the rst being Groseclose & Snyder (1996). This two-stage bargaining model, further explored in settings with nite numbers of legislators by Banks (2000) and Groseclose & Snyder (2000), characterizes equilibria in which supermajorities are assembled in order to block threats from an opposing interest. While the Groseclose and Snyder model has had a profound impact on political scientists' thinking about votebuying activities, its power lies in demonstrating a strong second-mover advantage in a bargaining setting with an exogenously nite horizon. Its utility in assessing the distributive consequences of vote buying, however, is more limited.
3
Depending upon which equilibrium case is under consideration,
Groseclose and Snyder's model generates dierent predictions about the distribution of payments. Without a clear equilibrium selection mechanism, it is virtually impossible to distill clear, testable predictions. More recently, Dekel, Jackson, and Wolinsky (hereafter DJW) take up the tradition of competitive vote-buying models with a pair of companion papers that make important theoretical advances and generate clearer predictions. DJW (2008) examines vote buying in general elections, while DJW (2009), most relevant for our purposes, models vote buying within a legislature. In DJW (2009), careful use of a smallest unit of payment and an assumption about the irreversibility of bidding (that is both more reasonable and more enlightening than it may seem at rst) enable the authors to avoid the issues with ties and shifting strategies encountered in Blotto-like allocation games (see Roberson (2006)). Additionally, the use of a per-round bidding cost allows the games to be endogenously nite, capturing the dynamic nature of legislative negotiation without undue impositions on the number of bidding rounds. The central prediction of the DJW (2009) model on legislative vote buying is essentially identical to that of Snyder (1991).
When payments are made in equilibrium, they are directed to what the
authors refer to as near-median legislators,
distance from the median.
and payments are again decreasing in magnitude with
Such legislators, after all, are the cheapest that could be bought to secure
majority support, and minimum payments are made to secure the requisite support needed for a bill's passage. To illustrate these key predictions, we present an adaptation of this model in section A of the online appendix.
3 Where
equilibrium cases in the presence of competing vote buyers are determined by the relative and absolute
valuations of the two lobbyists.
3
Challenges to Deriving Empirical Predictions When using existing vote buying models to generate predictions about total federal outlays across geographic units, three basic challenges arise.
The rst concerns the aggregation of vote purchases
across bills. In every instance, the predictions of the single- or competing-vote buyer models re ect payments made in a single iteration of the game.
When aggregating across bills, as we do in the
empirical analysis, it is straightforward to see why we might expect overall payments to monotonically decrease as one moves further away from the median. Because payments, when made, always decrease in magnitude with distance from the median, the prediction of a single iteration of vote buying scales up to considering multiple independent iterations.
4
The second challenge concerns the existence of various types of budget constraints that could be a source of dependence among iterations of the game. Rather than imposing a balanced-budget requirement/divide-the-dollar structure on the game or considering lobbyist-speci c budgets that are likely to bind on each iteration of the game, the valuation-based version of the DJW (2009) model perceives of the groups seeking to in uence legislators as having signi cant discretionary funds from which they may allocate as much as they wish to a given legislative eort and distribute such funds as they see t. This view is borne out by the following story from a House Appropriations Committee staer as taken from Shepsle et al. (2009):
One House [of Representatives] Appropriations [Committee] sta member, for example, described a budget account that was explicitly divided into four with each partisan delegation in each chamber having authority over its share. Other interviews suggested this was the implicit norm for many of the most heavily earmarked accounts, although it was typically not explicitly codi ed.
However, DJW also investigate a version of the model that assigns constraints on each vote buyer's coers. Importantly, though, the distribution of payments in the equilibrium of a single round of the budget-constrained game is identical to that of the unconstrained, valuation-based version, under the conditions relevant to the budget-constrained version in which payments are made. We abstract away from the possibility that a budget would bind over multiple iterations of the game, a potential source of dependence between iterations.
4 This
feature of aggregation stands in contrast to some of the other vote-buying models, notably Groseclose & Snyder
(1996). When considering multiple independent iterations of that game, the aggregate distribution of payments relies heavily upon the assumed distribution of proposals and its implications for the size of ex ante legislator valuations and thus ex ante majority support. One could derive the result that payments decrease monotonically with distance from the median given certain assumptions and yet derive diering results given other assumptions. Such sensitivity to varying assumptions provides yet another reason to focus on the DJW model, which requires no further assumptions besides independence of iterations to derive predictions testable with aggregate data. Whether this prediction is as accurate as it is straightforward is, of course, an empirical matter.
4
Third, and nally, the result that payments are made by a party with ex ante minority support for its proposal might suggest that only the minority party uses outlays to achieve (its rare) legislative victories.
However, the procedural cartel model rst forwarded in Cox & McCubbins (1993) and
developed further in Cox & McCubbins (2005) suggests exactly the opposite. The authors argue for a theory of negative agenda control, in which the majority party colludes to prevent unwanted bills from reaching oor votes. Those bills that do see the light of day tend to move policy towards majorityparty centrists. While negative agenda control should mostly ensure passage of such bills, Cox and McCubbins stipulate that distributive bene ts may be needed on the margins to compensate majority members suering a policy loss
5 (Cox & McCubbins 2005, pp. 45-47, 159, ch. 10).
The implication is that, in eect, the majority party must from time to time partake in vote-buying, and that majority party members would be the primary bene ciaries. A couple of previous empirical analyses, discussed below, have already provided support for this implication of the procedural cartel model of party government. In an extension of our main analysis, we seek to demonstrate whether vote-buying with outlays bene ts moderate members of both the majority and minority parties, or whether it is applied to one group dierentially.
Previous Empirical Work on Distributive Politics and Legislator Ideology Though over the last couple of decades empirical studies on distributive politics have proliferated, only
6 Evans (1994, 2004) oers the most
a small portion of this work speci cally examines vote buying.
sustained empirical examination of the use of distributive side payments to achieve legislative objectives. As support for her case that federal monies (earmarks in her case, rather than the categorical grants that are our dependent variable) are used to purchase votes, Evans presents interview data as well as in-depth case studies, which include the legislation for the Federal Highway Program and the legislation for entry into the North American Free Trade Agreement. Evans also estimates regressions focused on these particular pieces of legislation. A central contribution of her work is showing that legislators' promises to vote in a certain way in exchange for particularized bene ts are in fact binding, as demonstrated by bill- and even vote-speci c changes in voting behavior. Canvassing evidence from the 107th through 110th Congresses, Cann & Sidman (2011) present
5 Centrists
if the model assumes an open rule, but potentially members on either extreme of the majority party if the
model assumes a closed rule.
6 Instead,
a
previous studies have scrutinized the importance of committee membership (Alvarez & Saving 1997 ),
majority party status (Balla, Lawrence, Maltzman & Spigelman 2002, Cox & McCubbins 1993, Levitt & Snyder 1995),
b
electoral competition (Alvarez & Saving 1997 , Stein & Bickers 1994), state size (Knight 2008, Lee 2000), majoritarian rules and universalism (Bickers & Stein 1997, Groseclose 1996, Shepsle & Weingast 1981), party alignment with various members of the executive branch or the president (Berry, Burden & Howell 2010, Bertelli & Grose 2009, Gordon 2011, McCarty 2000), partisan contributions (Cann & Sidman 2011), and the roles of local governments (Bickers & Stein 2004, Rich 1989).
5
evidence that parties reward their members with distributive bene ts in exchange for raising money for and consistently voting with their party. Members with higher party unity scores and who raise more money for their party tend to receive higher direct outlays, direct awards, and contingent liability rewards. Kriner & Reeves (2012), moreover, demonstrate that such distributive bene ts yield important electoral rewards of their own.
7 By channeling greater federal outlays to their districts,
Kriner and Reeves show, legislators substantially improve their vote margins in subsequent elections. Outlays, then, function as compensation for making an electorally unpopular (and highly observable) oor vote. Only Carroll & Kim (2010) and Jenkins & Monroe (2012) investigate the implications of vote buying, and in particular the predictions of Cox & McCubbins's (2005) theory, with explicit regard to ideological moderation and extremism. Carroll & Kim (2010) nd that members of the majority party with higher individual roll rates votes against bills that ultimately passed tend to receive greater shares of outlays (both in number of projects and dollar amounts). The importance of ideology (rather than roll rates) and the distributive consequences of ideology for members of the minority party are open questions. Along similar lines, Jenkins & Monroe (2012) present evidence that the majority party buys its negative agenda control with side payments to its centrist members (p. 910). The majority party leadership, they show, directs a greater share of their campaign contributions to centrist co-partisans than to extremists within their party. Consistent with the cartel theory, no such eects are observed within the minority party.
Whether budgetary outlays are deployed in a similar manner remains
unexplored.
Empirical Strategy and Data Our primary analysis matches county-level data on federal outlays with corresponding political and demographic variables at the county, district, and state level. The advantage of county-level data is that we can observe the same units over a long period of time, whereas the boundaries of most districts are redrawn decennially. However, we must exclude from our analysis counties that are divided into multiple congressional districts because we cannot cleanly match such counties to a single member of
8 These excluded counties disproportionately represent urban centers around the country,
congress.
and thus encompass a signi cant proportion of the total U.S. population. Indeed, 43% of the total
7 Such eects, though, vary across jurisdictions with dierent ideological leanings (see also Sellers (1997)). 8 County-to-district population correspondence data from http://mcdc.missouri.edu/websas/geocorr90.html, http:
//mcdc.missouri.edu/websas/geocorr2k.html,
and
http://mcdc.missouri.edu/websas/geocorr12.html.
6
9 although between 85% and 89% of counties
U.S. population remains after culling out these counties,
remain, varying slightly by year. In other words, we exclude a fairly small number of the most densely populated counties from our analysis. A supplementary analysis using district-level data helps to allay concerns that our results are somehow being driven by the exclusion of urban counties. A more substantive reason exists for only considering those counties represented by a single legislator. Representation of a single geographic or population unit by a group of legislators is innately a team-production problem. While it is likely the ideological moderation/extremism of the various members plays a role, the vote-buying theories we draw from provide no insight into how such a collection of ideologies would aect receipt of outlays. Similarly, considering the Senate (at the state-, district-, or county-level) runs into this same obstacle. We are fundamentally testing a prediction of vote-buying theories as we examine the role of ideology in the distribution of outlays, and these theories pertain to single-representative situations. To explain patterns in federal outlays in scal year September of year in year
t
t,
t,
which runs from October of year
we use political and demographic characteristics of year
is the result of the budget passed by congress in year
t − 1.
t − 1.
t−1
to
The money spent
As such, the distributive eects
of vote buying should be observed during the scal year after a budget passes Congress. The data on county-level outlays comes from the Consolidated Federal Funds Report (CFFR) over the years in which it was published ( scal years 1983-2010),
10 excluding FY1983 as we lack a county-
to-district correspondence for 1982. A strength of using CFFR is that it enables us to distinguish the set of programs most likely to be subject to political manipulation, non-formula grants, from those that are not, such as entitlements and formula-based grants, which we use for placebo tests. Honing in on non-formula grants makes our analysis more transparent than prior work that relied on a fairly ad-hoc distinction between low-variation and high-variation programs, which was based on an arbitrary threshold in a program's coe cient of variation across districts to determine a cut-point for exclusion from the analysis (e.g. Levitt & Snyder (1995) or Berry, Burden & Howell (2010)). In estimating the eect of a representative's ideological location on her district's receipt of federal outlays, we confront three potential sources of endogeneity. The rst includes the many other determinants of outlays that also correlate with a member's ideology. For instance, poorer districts may elect more liberal representatives and also receive more aid from federal programs targeting poverty alle-
9 With
the average population by county dropping from 84,035 individuals (across 83,031 total county-year observa-
tions) to 41,349 individuals (across 72,420 county-year observations).
10 Although
information on federal outlays by county is available through the present from other sources, for the
sake of consistency we construct our dataset using only outlays as reported in CFFR, of which publication ceased after scal year 2010 with the termination of the Federal Financial Statistics program. Data accessed Summer 2013 at
http://www.census.gov/govs/cffr/.
7
viation, generating a spurious correlation between distance from the chamber median and outlays.
11
Second, voters desiring more federal aid may intentionally chose more centrists representatives. Third, individual members desiring more aid may vote in a more centrist fashion in order to make themselves attractive candidates for vote trading.
In the second and third cases, we would be concerned that
greater centrism might be associated with other, possibly unobservable, eorts taken by the district or member to obtain more federal spending. Our research design and measurement strategies address each of these concerns. We use a member xed eects model, which relies exclusively on within-member changes in distance to the chamber median voter over time. Time-invariant attributes of the member including those associated with the geographies she represents are purged by the xed eects, obviating concerns that, for instance, members representing extremely liberal districts receive more outlays due to the xed characteristics of their constituents. Further, a range of covariates serve as controls for time-varying characteristics. In principle, redistricting could pose problems for our strategy by introducing an unwanted source of variation if a member's district comprises dierent counties over time.
However, we show that our
results are virtually unchanged when we use more restrictive member-by-county xed eects. The xed eects design addresses the rst and second endogeneity concerns outlined above. However, one might still be concerned that individual members moderate their own voting behavior during years in which they wish to bring home more federal spending−and that they also take other actions to obtain more funding at the same time.
To remove this potential source of endogeneity,
we rely upon a feature of Poole and Rosenthal's Common Space DW-NOMINATE scores (Poole & Rosenthal 1985, Poole & Rosenthal 1991), from which we use the rst-dimension to derive our measure of ideological distance from the median. In the Common Space scores, each legislator's score is xed
12 Because, we use scores that are constant across a member's career
over the course of her tenure.
in Congress, along with member-speci c xed eects, we only exploit variation in ideological distance caused by shifts in the location of the median voter. Thus, even if an individual member's roll call voting record did appear arti cially moderate due to vote-selling, this arti cial moderation could not generate uctuations in ideological distance due to changes in the median voter's location. Arti cial
13
moderation, if it existed, would only bias us against nding any results using our research design.
11 Indeed
if this particular example held true, it would bias us against nding evidence of the vote-buying mechanism
we seek to identify.
12 In
fact, when a legislator switches parties, they are assigned a new ICPSR ID number and thus allowed a new xed
ideological estimate in the calculation of Poole and Rosenthal's data. As a result, scores are xed over the course of a legislator's tenure in a given party a liation. This aects only a few cases.
13 Moreover,
it is likely that the number of votes a given legislator trades for outlays is small relative to the total
number of votes she casts, and that the Common Space score is based on a largely ideologically consistent body of decisions, with the few aberrations contributing little to the estimation of the score.
8
Table 1: Key County-Level Descriptive Statistics Absolute distance from median
Outlays (in 2010 $1,000's)
Overall Mean
0.318
9,310
Majority Party Mean
0.189
8,990
Minority Party Mean
0.501
9,580
W/in-Group Std Dev
0.122
5,990
is the absolute value of median-centered DWNOMINATE Common Space scores for the calendar years 1983-2009. Outlays are county-level grants (excluding formula grants), in ation adjusted to 2010 dollars, for scal years 1984-2010. Means are presented for the entire House as well as just the House majority and minority parties. W/in-Group Std Dev is the within-member standard deviation of each variable. The sample is restricted to only those counties represented by a single congressperson. Additional summary statistics may be found in Table 5 in the online appendix.
Notes:
Absolute distance from median
Using this research design and measurement strategy, we seek to explain variation in outlays received by members generated by the changing location of the chamber median over time.
Changes
in a member's distance from the median come only from changes in the composition of the chamber, which are almost surely exogenous with respect to the ideological position of any single representative. For example, consider a member with an ideal point of -0.50 who served in two congressional terms, one in which the median voter's ideal point was -0.25 and one in which the median voter's ideal point was 0.25. We ask whether the member received more outlays in the rst congress, when her absolute distance from the median was 0.25, than in the second, when her absolute distance was 0.75. The relative ideological location of the member changed between the two congresses composition of the chamber arising from elections in
other
only
because of changes in
districts the results of which we assume
to be exogenous making the comparison across terms a valid causal estimate of the eect of distance from the median voter on outlays. Our source of identifying variation is displayed in Figure 1, which shows the location of the House median voter, measured via DW-NOMINATE Common Space scores, from 1983 through 2009. There are two major swings in the median voter's location, which are associated with changes in majority party control in 1995 and 2007, as well as smaller year-to-year changes throughout our study period. Given the importance of changes in majority control in determining changes in the location of the median voter, it will be important to control for majority status, as well as the interaction of majority status and ideological distance, in our analysis. These issues receive sustained attention below. Formally, we estimate the following general model:
ln(outlaysit ) = β0 + αi + δt−1 + β1 |Distancei,t−1 | + Xi,t−1 Φ + εi,t−1 ,
where the dependent variable is the log value of county-level outlays in a given scal year
9
(1)
t.
The
1995 1980
1985
1990
Year
2000
2005
2010
Figure 1: Movement in the House Median by Year
-.2
-.1
0
.1
.2
Chamber median The horizontal axis represents the location of the median member of the House of Representatives as measured in DW-NOMINATE Common Space scores.
Notes:
xed eects,
αi ,
are generally member-speci c, though we examine the robustness to using county
and county-member xed eects instead. preceding calendar year and a constant,
All models also include year xed eects,
β0 .
δt−1 ,
for the
Our variable of interest is the absolute value of the
ideological distance to the oor median for the representative of a given county as calculated with DW-NOMINATE Common Space scores. The coe cient for this variable is
Xi,t−1
has corresponding coe cients given by
Φ.
β1 .
A vector of covariates
These covariates take on member-time speci c values
and include political as well as demographic characteristics.
10
When selecting time-varying political covariates, we take our cues from the existing empirical literature on the determinants of federal distribution.
We use dummy variables for membership in
the party of the president, majority status, party a liation (if using county rather than member xed eects), committee membership, and being a party leader, committee chair, or ranking minority
14 . A tenure variable tracks the number of terms served
committee member (Nelson, Stewart & Woon)
by a given representative, and we include an additional indicator for a representative's rst term. measure of party competitiveness,
Close election,
15 A
identi es those instances when a member receives
less than 5% of the two-party vote share in the last election. The last of the political variables, also electoral in nature, is the absolute value of the state-wide dierence in vote shares between the sitting president and the other major party candidate in the previous election.
16
This value decreases in
the competitiveness of the previous presidential election in a given area, while the close congressional election dummy identi es more competitive elections. Among the factors recognized by the existing empirical literature, majority party membership is most likely to confound the eect on the distribution of outlays of our variable of interest, absolute distance to the median. By de nition, the median voter will be a member of the majority party, and thus other members of the majority will tend to be found closer to the median voter on an ideological continuum than will their peers in the minority. As such, a representative with low ideological distance to the median will more likely be drawn from the majority party than the minority party. One might be concerned, then, that results indicating the importance of ideological distance re ect majority party in uences rather than the vote-buying mechanisms posited in our theory. Moreover, the implication of the majority party cartel theory put forward by Cox & McCubbins (1993) that Jenkins & Monroe (2012) test (using campaign contributions, rather than distributive outlays) represents yet another complication in considering the role of majority party status vis-à-vis ideological distance from the median. To disentangle their theoretical claims from those that emerge from the vote-buying literature, therefore, in the empirical tests that follow we include majority party status as both a control and, later, as an interaction with our variable of interest. Given our nearly 30-year sample, the lengthy tenure of many representatives, and the fact that some members in our sample represent multiple counties in our data, it is important to account for intra-county demographic changes over time. We therefore include the log values of county population and income as controls.
17 To account for correlation of the error term,
εit ,
both across and within
14 Both accessed Summer 2013 at http://web.mit.edu/17.251/www/data_page.html. 15 Information on individual legislators' ideology, political a liation, and tenure from http://voteview.com/dwnomin_
joint_house_and_senate.htm. 16 Both 17 Data
http://library.cqpress.com/elections/. http://bea.gov/iTable/index_regional.cfm.
electoral variables accessed Summer 2013 at accessed at
11
counties over time, we cluster standard errors at the state level. Further information about the data and the decisions made while compiling it may be found in section B of the online appendix.
Main Results Table 2: Absolute Distance & Rank from Median
Absolute distance from median
(1)
(2)
(3)
-0.091
-0.746***
-0.713***
(0.069)
(0.276)
(0.254)
Absolute rank from median (/100) Majority party President's party Committee chair Ranking minority member Party leader First term Tenure (# terms)
(4)
(5)
(6)
-0.016
-0.136***
-0.132***
(0.013)
(0.038)
(0.036)
-0.229**
-0.218**
-0.235***
-0.227***
(0.093)
(0.087)
(0.068)
(0.066)
0.005
0.009
0.018
0.021
(0.035)
(0.031)
(0.038)
(0.034)
-0.022
-0.051
-0.022
-0.051
(0.076)
(0.072)
(0.076)
(0.072)
-0.043
-0.065
-0.043
-0.066
(0.052)
(0.051)
(0.055)
(0.054)
-0.037
-0.097**
-0.044
-0.103**
(0.048)
(0.046)
(0.048)
(0.046)
0.006
-0.005
0.005
-0.004
(0.025)
(0.022)
(0.024)
(0.022) -0.250*
-0.177
-0.240*
-0.189
(0.123)
(0.133)
(0.122)
(0.134)
0.122***
0.119***
0.122***
0.120***
(0.033)
(0.029)
(0.033)
(0.029)
0.003
0.002
0.003
0.002
(0.002)
(0.002)
(0.002)
(0.002)
-0.452***
-0.456***
-0.458***
-0.452***
-0.455***
-0.457***
(0.118)
(0.119)
(0.120)
(0.118)
(0.119)
(0.120)
Log population
1.600***
1.604***
1.606***
1.599***
1.603***
1.605***
(0.124)
(0.126)
(0.126)
(0.124)
(0.126)
(0.126)
Constant
2.535***
2.611***
2.552***
2.540***
2.654***
2.602***
(0.328)
(0.473)
(0.448)
(0.326)
(0.421)
(0.406)
No
No
Yes
No
No
Yes
0.547
0.548
0.548
0.547
0.548
0.548
71147
70990
70990
71147
70990
70990
Close election State presidential margin Log income
Committee dummies Adj.
R2
N
Notes:
Standard errors clustered by state. Member and year xed eects used in all models. The dependent variable
is the log value of non-formula grants received by a given county in a given year. The outlays data span scal years
1984-2010 and are matched with explanatory variables from the previous calendar year. Absolute distance from median is the absolute value of the median-centered rst dimension of the DW-NOMINATE Common Space scores. Absolute rank from median (/100) is the rank ordering of the Absolute distance from median variable divided by 100 for scaling purposes; the higher the rank, the farther a legislator is ideologically from the median.
∗p
≤ 0.10 ∗∗ p ≤ 0.05 ∗∗∗ p ≤ 0.01
Table 2 presents our primary estimates of the eect of absolute distance from the chamber median on the distribution of federal outlays. All models include member xed eects, and as a result only take advantage of exogenous changes in each individual member's distance to the median.
Models
1-3 regress log outlays at the county-level on our measure of absolute distance from the median in NOMINATE space. Models 4-6 regress log outlays on the legislators' ranked absolute distance from
12
the median that is, the number of other representatives between the member and the median voter per 100 legislators. Models 1 and 4 regress the log value of county-level outlays on our main variable of interest, as well as the county-level demographic covariates, log values of population and income. As the dependent variable is in logs, the appropriate interpretation for a one unit change in the regressor would be, in the case of model 1, an approximately 9% decrease in the amount of outlays a district receives. In models 2 and 5, we add all of the political covariates except for the dummy variables representing membership on speci c committees. In doing so, we see the eect sizes of both
median
and
Absolute rank from median (/100)
Absolute distance from
increase markedly to -0.746 and -0.136, respectively,
and gain statistical signi cance. The positive and signi cant eect of being in a highly contested district, the
Close election
variable,
on outlays re ects the importance of electoral objectives in determining the distribution of outlays, but it also helps allay concerns that the signi cant negative eect of
Absolute distance on outlays stems from
electoral considerations rather than our proposed legislative mechanism. If electorally close districts are more likely to be represented by moderates, then absolute distance would have covaried inversely with our close congressional election measure.
distance
As such, it might have been the case that
re ected safer districts, less in need of targeted federal funds, and we would see a negative
eect size that in fact had nothing to do with our proposed vote-buying mechanism.
By including
the measure of the electoral competitiveness of a district, we would then expect the eect of
distance
Absolute
Absolute
to fall. In fact, the eect size increases, lending further credence to our core theoretical claims.
Lastly, in models 3 and 6, we add indicator variables for membership on all of the standing committees (variables not shown).
This appears to make little dierence to both the eect sizes and
signi cance levels of our variables of interest. While the variables are jointly signi cant, the F-statistic
18 We omit the committee dummies
is remarkably small given the number of variables we are testing.
in subsequent analyses, although our results are robust to their inclusion. Given the increase in polarization over our sample period, documented at length in McCarty, Poole & Rosenthal (2006), it is possible that new members entering the chamber tended to make their senior colleagues relatively more moderate. We might worry about con ating seniority (and the distributive bene ts it may confer) with moderation. By controlling for a member's number of terms in o ce with the
Tenure
variable, we are able to account for the eect of seniority.
When interpreting the magnitude of the estimates, it is important to remember that the location of
18 Only
a couple of these variables display statistical signi cance, which is an expected result given the number of tests
being performed, even if all of the null hypotheses of zero eect held true.
13
a legislator relative to the median simply never changes by the equivalent of one unit in NOMINATE space. As a result, it makes more sense to consider a standard deviation's worth of change in
distance.
Absolute
For model 2, the within-member standard deviation for the sample used in estimation is
19 We would then associate a 9.1% decrease in outlays with a one standard deviation increase
0.122.
in distance from the median. Put dierently, a one-standard deviation increase in distance from the median leads to an approximately $850,000 decrease in outlays (at the county level, in 2010 dollars), or a loss of $20 per capita for the average county. For
Absolute rank, the within-group standard deviation is
0.710, implying a 9.7% decrease in outlays associated with a one-standard deviation increase in ranked distance from the median in model 5.
distance
In subsequent analyses, we continue with only the
Absolute
variable, though again, consistent results are recovered from both measurement strategies.
Numerous auxiliary analyses were performed, all of which may be found in section C of the online appendix. Several bear mentioning here. First, we explored other xed eects strategies, including the
20 Comparing the
inclusion of no xed eects, county xed eects, and county-member xed eects.
dierent xed eects strategies, we see that the member xed eects models produce estimates largely identical to the even more stringent county-member xed eects models, both of which preserve our identi cation strategy. As such, we adopt model 2, which uses member xed eects, as a baseline for comparisons in the foregoing analyses. A variety of other characterizations of the
Absolute distance
variable were tried, including the
addition of higher and lower order terms as well as the natural log of
Absolute distance.21
Only the
speci cation in levels proved consistent across models and speci cations of the dependent variable. We also carried out an analysis using splines, in which
Absolute distance
was segmented and allowed
22 We
to take on dierent slopes on either side of a knot point placed at gradually increasing points.
found that the eect of absolute distance from the oor median tended to be concentrated around, but not limited to, near-median legislators, as vote-buying theory would suggest. Running our baseline model with a variable measuring absolute distance from
party
mean, we
observe a positive and statistically signi cant estimate of the coe cient for this new variable.
23 One
standard deviation's worth of within-member variation in absolute distance from party mean is so small, however, that we are reluctant to conclude that intra-party extremists receive more in distributive outlays. To explore the possibility that tendencies to vote with or against one's party re ect general ex-
19 See 20 See 21 See 22 See 23 See
Table 1 above. Table 6 in the online appendix. Table 7 in the online appendix. Table 8 in the online appendix. Table 9 in the online appendix.
14
tremism or moderation, we used party unity scores in lieu of absolute distance from the median (as
24 In none of such models did we nd that party unity had a
well as alongside the distance variable).
consistent relationship with the distribution of federal outlays. Lastly, using CVP scores (Fowler & Hall 2012),
25 an alternate roll call-based measure of legislator
ideology, yielded similar patterns of signi cance, magnitude of eect size, and consistently negative coe cients on the estimates of the eect of absolute distance from the median on log outlays.
26 We
take the similarity of the results as encouraging, and also point out that CVP scores feature an ease of interpretation that NOMINATE scores lack. The estimates from regressions using CVP scores may be read as the percent increase in outlays associated with a given increase in the probability of voting conservatively relative to the median member of the chamber.
Majority Party Status Majority status merits a deeper discussion than most of the other controls and covariates. While several models and empirical studies suggest that majority status is an important determinant of the distribution of federal outlays, the relationship between majority status and outlays is theoretically ambiguous. Theories in which the majority party extracts rents for itself would suggest that majority status leads an individual representative to receive more in federal outlays. Alternative theories, in which the majority uses parliamentary prerogatives (such as proposal power) to achieve legislative victories, might suggest that the eect of majority status on outlays could even be negative. Theories of negative agenda control, meanwhile, predict that the majority moves the agenda to the center of its party's distribution and pays o moderate members of its own party to compensate them for policy losses. If indeed this theory only applies to the majority party, with its dominant ability to set the House agenda, then we might expect to nd a discernible relationship between legislator ideology and federal outlays only among members of the majority party. Finally, from a purely empirical standpoint, majority party status plays a key role with regard to
Absolute distance.
Distance from the oor median will be inversely correlated with majority party
status for the simple reason that the chamber median lies within the majority party (given that the parties have ceased to overlap in NOMINATE space). Accounting for this correlation is essential for identifying the eect of each variable individually. Given these various sources of ambiguity, we explore the separate and joint contributions of legislative extremism and majority status in Table 3. Model 1 includes
Absolute distance
24 See Table 10 in the online appendix. 25 CVP scores available at http://www.andrewbenjaminhall.com/papers/. 26 See Table 11 in the online appendix.
15
without the variable
Table 3: Interacting
Absolute Distance (1)
Absolute distance from median
and
Majority Party
(2)
-0.121 (0.073)
Majority party
(3)
(4)
-0.746***
-0.772**
(0.276)
(0.308)
0.021
-0.229**
-0.260*
(0.023)
(0.093)
(0.138)
Absolute distance x majority
0.123 (0.226)
President's party Committee chair Ranking minority member Party leader First term Tenure (# terms)
0.021
0.021
0.005
0.004
(0.041)
(0.040)
(0.035)
(0.035)
-0.030
-0.021
-0.022
-0.023
(0.073)
(0.078)
(0.076)
(0.076)
-0.014
-0.022
-0.043
-0.044
(0.071)
(0.068)
(0.052)
(0.051)
-0.034
-0.033
-0.037
-0.038
(0.045)
(0.045)
(0.048)
(0.047)
0.009
0.011
0.006
0.006
(0.024)
(0.024)
(0.025)
(0.025)
-0.165
-0.169
-0.177
-0.176
(0.123)
(0.122)
(0.123)
(0.125)
0.124***
0.121***
0.122***
0.121***
(0.033)
(0.032)
(0.033)
(0.033)
0.003
0.003
0.003
0.003
(0.002)
(0.002)
(0.002)
(0.002)
-0.457***
-0.457***
-0.456***
-0.456***
(0.119)
(0.119)
(0.119)
(0.119)
Log population
1.605***
1.605***
1.604***
1.604***
(0.126)
(0.126)
(0.126)
(0.126)
Constant
2.280***
2.223***
2.611***
2.625***
(0.389)
(0.370)
(0.473)
(0.493)
0.548
0.548
0.548
0.548
70990
70990
70990
70990
Close election State presidential margin Log income
Adj.
R2
N
Notes:
Standard errors clustered by state. Member and year xed eects used in all models. The
dependent variable is the log value of non-formula grants received by a given county in a given year. The outlays data span scal years 1984-2010 and are matched with explanatory variables from the previous calendar year.
Absolute distance from median
is the absolute value of the
median-centered rst dimension of the DW-NOMINATE Common Space scores. This variable is interacted with the dummy variable for majority party status in model 4.
∗p
≤ 0.10 ∗∗ p ≤ 0.05 ∗∗∗ p ≤ 0.01
for majority status but with the remainder of the controls. The coe cient is negative, as expected, but imprecisely estimated. Model 2 includes only majority status along with the controls, leaving out our measure of absolute distance to the median. Although insigni cant, the positive coe cient estimated here supports either a rent-seeking theory or, to the extent that majority party covaries inversely with
Absolute distance,
accounts of vote buying.
Model 3 includes both absolute ideological distance to
the median and majority status, along with the other controls we have employed throughout. coe cient estimate for
Absolute distance
The
is statistically signi cant and approximately ve times larger
than that in a comparable model that did not include majority party status. Here, the coe cient for
16
majority status is now both signi cant and negative. Model 4 adds an interaction term between
Absolute distance
Majority party
and
Absolute distance.
The eect of
appears to be stronger within the minority party, but we cannot reject that there
is no dierence in the eect of absolute distance from the chamber median between the majority and minority parties.
Most importantly, however, we see no evidence that our results for the eect of
ideological distance to the median are an artifact of majority status. Moreover, the fact that
distance
Absolute
remains signi cant in the presence of an interaction variable suggests that moderate minority
members are also bene ciaries of vote buying (whether from their own party or the majority party cobbling together coalitions). Outlays emerge as a broad instrument of coalition building, extending beyond negative agenda control's stricter conception of being used by the majority party on majority party members. The negative coe cient on the indicator variable for membership in the majority party in models 3 and 4 may seem counter-intuitive, especially in light of the positive coe cient in model 2.
How
can the average majority party member obtain more outlays than the average minority member, as implied by model 2, while bringing home less than a minority member located at the same ideological distance from the chamber median, as implied from by models 3 and 4?
The answer is that the
average majority party member is much closer to the chamber median than is the average minority party member. Speci cally, the average absolute distance from the median for members of the majority party is 0.189, while the average absolute distance from the median for members of the minority party
27
is 0.501.
Figure 2 presents a visual representation of the estimates from model 4.
The negative
relationship between ideological distance and outlays holds within both the majority (dashed lines) and minority (solid lines) parties, which is our main point of interest here. In addition, at any given distance from the median, a minority party member is predicted to garner more outlays than a majority party member at the same location. However, given their generally closer proximity to the median, the average member of the majority party (vertical dashed line) acquires slightly more (indeed, 2% more, as indicated in model 2) overall outlays (horizontal dashed line) than the average member of the minority party (vertical solid line) receives in outlays (horizontal solid line). Still, the graph emphasizes one crucial point: minority party members located closest to the median have the highest expected outlays. This pattern is consistent with a model in which marginal members of the opposition are the most likely targets of vote buying. As an additional, related set of tests, we interact an indicator variable for periods of uni ed government (de ned as instances when the Senate and House majorities as well as the president are all of the
27 See
Table 1 above.
17
Distance from Median
and
Majority Party
2.6
Figure 2: Understanding the Interaction between
2.4
Minority
1.8
2
Log Outlays 2.2
Majority
0 Notes:
.2
.4 .6 Distance to Median
.8
1
The dotted lines correspond to the majority party, while the solid lines correspond to the minority
party. The downward-sloping lines are the outlays awarded to members of each party as a function of the members' absolute distance to the median in DW-NOMINATE Common Space scores. These are based on the estimates from model 4 of Table 3.
The vertical lines represent the average absolute distance
from the median for each party, as in Table 1. The horizontal lines pass through the intersection of each pair of diagonal and vertical lines.
The horizontal line for the majority party (2.242) lies just above
the horizontal line for the minority party (2.238), indicating the average member of the majority party receives 2% more than the average member of the minority party, as found in model 2 of Table 3.
same party) with the
Absolute distance
and
Majority party
variables.
28 If legislation is more di cult
to pass during periods of divided government, as some have found (Howell et al. 2000), we may observe a greater reliance on vote buying and/or agenda setting using federal outlays as the instrument. While the
Uni ed government, Absolute distance,
and
Majority party
variables remain signi cant as the re-
gressions progress towards full saturation, the interaction terms do not retain signi cance throughout this process. Two other analyses, one which interacts the size of the majority party with other variables of interest
29 and one which focuses on majority members on the far side of the oor median from the 30 yield similarly inconclusive insights.
majority of their party
28 See 29 See
Table 12 in the online appendix. Table 13 in the online appendix.
30 See
http://history. http://www.senate.gov/pagelayout/history/one_
General information on parties in Congress from
house.gov/Institution/Party-Divisions/Party-Divisions/ item_and_teasers/partydiv.htm. Table 14 in the online appendix.
18
and
Robustness Checks and Placebo Tests Table 4: Robustness & Placebo Tests Robustness
Absolute distance from median Majority party President's party Committee chair Ranking minority member Party leader First term Tenure (# terms) Close election State presidential margin Log income Log population Constant Adj.
R2
N
Notes:
Placebo
(1) No Formula
(2) All Grants
(3) Formula Only
-0.746***
-0.357***
-0.204***
(4) Dis. & Ret. 0.023
(0.276)
(0.089)
(0.075)
(0.029)
-0.229**
-0.098***
-0.051**
0.014
(0.093)
(0.030)
(0.025)
(0.013)
0.005
0.000
-0.004
-0.001
(0.035)
(0.011)
(0.011)
(0.003)
-0.022
-0.003
0.000
-0.011
(0.076)
(0.023)
(0.024)
(0.013)
-0.043
0.021
0.012
-0.017
(0.052)
(0.024)
(0.022)
(0.012)
-0.037
-0.052
-0.036
0.009
(0.048)
(0.039)
(0.040)
(0.019)
0.006
-0.016
-0.028**
0.003
(0.025)
(0.014)
(0.012)
(0.003)
-0.177
-0.077
-0.018
-0.078*
(0.123)
(0.054)
(0.031)
(0.044)
0.122***
0.039***
0.022
-0.005
(0.033)
(0.014)
(0.013)
(0.004)
0.003
0.003***
0.004***
-0.000
(0.002)
(0.001)
(0.001)
(0.000)
-0.456***
-0.845***
-1.117***
0.084
(0.119)
(0.101)
(0.110)
(0.067)
1.604***
1.837***
2.112***
0.879***
(0.126)
(0.104)
(0.112)
(0.069)
2.611***
6.742***
7.065***
6.217***
(0.473)
(0.296)
(0.285)
(0.182)
0.548
0.792
0.827
0.943
70990
71199
71194
71198
Standard errors clustered by state. Member and year xed eects used in all models. The dependent variable
is the log value of outlays received by a given county in a given year. Outlays are de ned as all non-formula grants in model 1, all grants in model 2, only formula grants in model 3, and disability and retirement payments to individuals in model 4. The outlays data span scal years 1984-2010 and are matched with explanatory variables from the previous calendar year.
Absolute distance from median
is the absolute value of the median-centered rst dimension of the
DW-NOMINATE Common Space scores. Models 1-3 are marked Robustness as we expect to observe some, perhaps attenuated, eect of
Absolute distance
on outlays even across dierent cuts of the grants category. Model 4 is labeled
Placebo as we do not expect to observe any eect of
Absolute distance
on outlays for disability and retirement
payments to individuals.
∗p
≤ 0.10 ∗∗ p ≤ 0.05 ∗∗∗ p ≤ 0.01
In Table 4, we present the results of additional robustness tests and a placebo test for which we expect to nd no eects.
The rst set of models, 1-3, consider alternative segmentations of federal
grants. The rst column is our baseline model, culling out formula-based spending from the grants category of outlays (as in the models in Table 2) - the spending that should be most susceptible to political manipulation.
31 The second column utilizes the entire grants category. The third column
looks at only formula-based grants, where we would expect to see the smallest, if any, eects of vote
31 See Table 15 in the online appendix for models which use all grants as the dependent variable.
Patterns of signi cance
are largely similar, though as seen in Table 4, results attenuate vis-à-vis non-formula based grants.
19
buying. We see that the signi cance of the negative coe cient estimate for absolute distance to the oor median is robust to all of these speci cations, but as expected, the eect size and signi cance attenuate as the focus moves to what we would presume would be less manipulable, formula-based grants. We see the same pattern in the
Close election
variable. Again, it is commonly held that marginal
seats are targeted with funds to bolster the incumbent's strength in the coming election. The ideal funds for such purposes would presumably be the more manipulable, non-formula based grants. Indeed we see that as we introduce formula-based spending and then consider only formula-based spending across models 1-3, the positive coe cient on the
Close election
variable steadily attenuates.
Model 4 uses an altogether dierent CFFR category, namely direct disability and retirement payments, which are the major categories of entitlements. Because they should not be subject to votebuying activities, these payments provide a useful placebo test. Reassuringly, the estimated coe cient for
Absolute distance
is no longer statistically signi cant and even switches sign.
Seeing no eect
where we should not, this last model serves as encouragement that our signi cant negative coe cient estimate in the baseline model re ects the vote-buying mechanism as predicted by theory. To further assess the robustness of our core ndings, we employ an altogether dierent dataset: the
32 The FAADS maps
Federal Assistance Award Data System (FAADS) data on district-level outlays.
federal outlays into congressional districts, and thereby avoids the challenges associated with ensuring a clear correspondence between a county and a single congressperson. This feature of the data allows us to use all districts, without excluding the most urban areas as we had to do in the county-level analysis. As with the CFFR data, it is reasonably straightforward to identify and cull out formula-based spending.
When using FAADS, however, new complications arise.
To begin with, we must remove
redistricting years, which introduce a mismatch between districts in which outlays are distributed and the political and demographic variables associated with their authorization. Additionally, if using member-district xed eects, we must employ redistricting-speci c member xed eects for each period in which dierent district boundaries apply, to take account of the fact that members may represent substantively dierent geographies following decennial redistricting. As such, we signi cantly reduce the temporal variation present in this model vis-à-vis the county-level analysis that spanned nearly thirty years within units. Using only non-formula grants as the dependent variable and member xed eects, the coe cient of the absolute distance variable does not appear signi cantly dierent than zero and occasionally
32 Our
data utilize and extend the data as documented in Bickers & Stein (1991). See Berry, Burden & Howell (2010)
for a previous use and extension of the FAADS data.
20
displays a sign contrary to our expectation. Models employing all grants as the dependent variable display consistently negatively signed coe cients. Models regressing non-formula grants against CVP scores as the independent variable consistently retain the expected negative sign, almost identical eect
33 In other words, the district-level results are uneven and inconclusive
size, and statistical signi cance.
when using member xed eects, and even more so when further limiting within-unit variation by using redistricting period-speci c member xed eects. Most importantly, though, in none of these analyses are the results stronger when restricting the sample to only those districts represented in the countylevel analysis. This suggests that our county-level results were not driven by the exclusion of the most populous counties (i.e. those containing more than one district).
Conclusion This paper presents the most compelling evidence to date that ideological moderates receive more distributive outlays than do ideological extremists within Congress.
The estimated 9% decrease in
outlays associated with an exogenously derived one-standard deviation increase in ideological distance from the median remains signi cant both statistically and substantively, resistant to false positives in placebo tests, and robust to various speci cations and measures of both absolute distance to the median and outlays. Due to the inherent unobservability of side-payments, we cannot be completely sure that we have captured the unique eects of vote buying on receipt of outlays. Our research design rules out objections related to the taste of legislators for pork barrel relative to legislative activities and other similarly rst-order threats to our causal claim. Considering more nuanced questions raised around the validity of our ndings in the context of other ongoing questions in the literature, though, suggests potentially productive ways forward on those challenges.
34
The ndings in this paper also suggest that the eect of majority party status is more di cult to track empirically than to predict theoretically, where it is widely held to be an important determinant in legislative bargaining and distributive politics. The analysis performed herein suggests that this may be due to the covariation and interaction of majority status with ideological distance to the median, as well as a result of the possible concurrence of negative and positive agenda control.
A deeper
understanding of these interplays may help future work better account for majority party in uence.
33 We 34 For
omit results here and refer the reader to Tables 16-19 in the online appendix. example, one promising line of investigation appears in forthcoming work by Halberstam and Montagnes, which
suggests that ideologically more extreme representatives are ushered into o ce on presidential coattails, relative to the ideology of those elected in midterm elections. Their theoretical ndings prompt one to ask whether these extremists, more electorally vulnerable than their margins of victory would suggest, would be assisted by their parties in reelection eorts with (possibly disproportionate) awards of outlays.
21
Finally, the team production issues that prevented a straightforward application of vote-buying theory to the Senate and to counties served by multiple representatives deserve attention in their own right. How might theories of vote-buying and agenda setting be modi ed or adapted to consider such cases? While the ndings in this paper provide compelling evidence for the role of ideology in outlays received by a single representative for a given area, its applicability to cases of overlapping political jurisdictions remains unanswered.
22
References Alvarez, R. Michael & Jason L. Saving. 1997a. Congressional Committees and the Political Economy of Federal Outlays.
Public Choice
92(1):5573.
Alvarez, R. Michael & Jason L. Saving. 1997b. De cits, Democrats, and Distributive Bene ts: Congressional Elections and the Pork Barrel in the 1980s.
Political Research Quarterly 50(4):809831.
Balla, Steven J., Eric D. Lawrence, Forrest Maltzman & Lee Spigelman. 2002. Partisanship, Blame Avoidance, and the Distribution of Legislative Pork.
American Journal of Political Science
46(3):515525.
Banks, Jerey S. 2000.
Review
Buying supermajorities in nite legislatures.
American Political Science
94(3):677681.
Baron, David P. 1991.
Majoritarian Incentives, Pork Barrel Programs, and Procedural Control.
American Journal of Political Science Baron, David P. 2006.
35(1):5790.
Competitive Lobbying and Supermajorities in a Majority-rule Institution.
The Scandinavian Journal of Economics
108(4):607642.
Baron, David P. & John A. Ferejohn. 1989.
Science Review
Bargaining in Legislatures.
The American Political
83(4):11811206.
Berry, Christopher R., Barry C. Burden & William G. Howell. 2010. The President and the Distribution of Federal Spending.
American Political Science Review
104(04):783799.
Bertelli, Anthony M. & Christian R. Grose. 2009. Secretaries of Pork? A New Theory of Distributive Public Policy.
The Journal of Politics
71(03):926945.
Bickers, Kenneth N. & Robert M. Stein. 1991.
Federal Domestic Outlays, 19831990: A Data Book.
Aramonk, NY: M. E. Sharpe.
Bickers, Kenneth N. & Robert M. Stein. 1997. Building Majority Coalitions for Sub-Majority Bene t Distributions.
Public Choice
91(3):229249.
Bickers, Kenneth N. & Robert M. Stein. 2004. Interlocal Cooperation and the Distribution of Federal Grant Awards.
The Journal of Politics
66(3):800822.
Cann, Damon M. & Andrew H. Sidman. 2011. Exchange Theory, Political Parties, and the Allocation of Federal Distributive Bene ts in the House of Representatives. 73(04):11281141.
23
The Journal of Politics
Carroll, Royce & Henry A. Kim. 2010. Party Government and the Cohesive Power of Public Plunder .
American Journal of Political Science
54(1):3444.
Legislative Leviathan: Party Government in the
Cox, Gary W. & Matthew D. McCubbins. 1993.
House.
Berkeley and Los Angeles: University of California Press, Ltd.
Cox, Gary W. & Matthew D. McCubbins. 2005.
in the U.S. House of Representatives.
Setting the Agenda: Responsible Party Government
Cambridge: Cambridge University Press.
Dekel, Eddie, Matthew O. Jackson & Asher Wolinsky. 2008. Vote Buying: General Elections.
of Political Economy
Journal
116(2):351380.
Dekel, Eddie, Matthew O. Jackson & Asher Wolinsky. 2009. Vote Buying: Legislatures and Lobbying.
Quarterly Journal of Political Science
4(2):103128.
Evans, Diana. 1994. Policy and Pork: The Use of Pork Barrel Projects to Build Policy Coalitions in the House of Representatives.
Evans, Diana. 2004.
Congress.
American Journal of Political Science
38(4):894917.
Greasing the Wheels: Using Pork Barrel Projects to Build Majority Coalitions in
Cambridge: Cambridge University Press.
Ferejohn, John A. 1974.
Pork Barrel Politics: Rivers and Harbors Legislation, 1947-1968.
Stanford:
Stanford University Press.
Fowler, Anthony & Andrew Hall. 2012. Conservative Vote Probabilities: An Easier Method for the Analysis of Roll Call Data.
unpublished - Available at SSRN 2120720
.
Gordon, Sanford C. 2011. Politicizing Agency Spending Authority: Lessons from a Bush-era Scandal.
American Political Science Review
105(04):717734.
Groseclose, Tim. 1996. An Examination of the Market for Favors and Votes in Congress.
Inquiry
34(April):320340.
Groseclose, Tim & James M. Snyder, Jr. 1996. Buying Supermajorities.
Review
Economic
American Political Science
90(2):303315.
Groseclose, Tim & James M. Snyder, Jr. 2000. Vote Buying, Supermajorities, and Flooded Coalitions.
American Political Science Review Halberstam, Median
Yosh Voter:
&
B.
Pablo
Senator
94(3):683684.
Montagnes.
Selection
in
2014. U.S.
Presidential Elections.
http://individual.utoronto.ca/halberstam/coattails_extremism.pdf 24
Coattails
unpublished .
-
versus
Available
The
at
Howell, William, Scott Adler, Charles Cameron & Charles Riemann. 2000. Divided Government and the Legislative Productivity of Congress, 1945-94.
Legislative Studies Quarterly
25(2):285312.
Jenkins, Jeery A. & Nathan W. Monroe. 2012. Buying Negative Agenda Control in the U.S. House.
American Journal of Political Science
56(4):897912.
Knight, Brian. 2008. Legislative Representation, Bargaining Power and The Distribution of Federal Funds: Evidence From The US Congress.
Kriner, Douglas L. & Andrew Reeves. 2012. Elections.
The Economic Journal
118(532):17851803.
The In uence of Federal Spending on Presidential
American Political Science Review
106(02):348366.
Lee, Frances E. 2000. Senate Representation and Coalition Building in Distributive Politics.
ican Political Science Review
94(1):5972.
Levitt, Steven D. & James M. Snyder, Jr. 1995. Outlays.
Political Parties and the Distribution of Federal
American Journal of Political Science
39(4):958980.
McCarty, Nolan M. 2000. Presidential Pork: Executive Veto Power and Distributive Politics.
ican Political Science Review
Amer-
94(1):117129.
McCarty, Nolan M., Keith T. Poole & Howard Rosenthal. 2006.
Ideology and Unequal Riches.
Amer-
Polarized America: The Dance of
Cambridge: The MIT Press.
Nelson, Garrison. 2013. Committees in the U.S. Congress, 1947-1992: House Committees, 98th-102nd Congresses..
Poole, Keith T. & Howard Rosenthal. 1985.
American Journal of Political Science
A Spatial Model for Legislative Roll Call Analysis.
29(2):357384.
Poole, Keith T. & Howard Rosenthal. 1987. Analysis of Congressional Coalition Patterns: A Unidimensional Spatial Model.
Legislative Studies Quarterly
12(1):5575.
Poole, Keith T. & Howard Rosenthal. 1991. Patterns of Congressional Voting.
Political Science
American Journal of
35(1):228278.
Poole, Keith T., Howard Rosenthal & Kenneth J. Koford. 1991. On Dimensionalizing Roll Call Votes in the U.S. Congress.
Rich, Michael J. 1989.
The American Political Science Review
85(3):955976.
Distributive Politics and the Allocation of Federal Grants.
Political Science Review
83(1):193213.
25
The American
Roberson, Brian. 2006. The Colonel Blotto Game.
Economic Theory
29(1):124.
Sellers, Patrick J. 1997. Fiscal Consistency and Federal District Spending in Congressional Elections.
American Journal of Political Science
41(3):10241041.
Shepsle, Kenneth A. & Barry R. Weingast. 1981. Political Preferences for the Pork Barrel: A Generalization.
American Journal of Political Science
25(1):96111.
Shepsle, Kenneth A., Robert P. Van Houweling, Samuel J. Abrams & Peter C. Hanson. 2009. The Senate Electoral Cycle and Bicameral Appropriations Politics.
Science
American Journal of Political
53(2):343359.
Snyder, Jr., James M. 1991. On Buying Legislatures.
Economics and Politics
3(2):93109.
Stein, Robert M. & Kenneth N. Bickers. 1994. Congressional Elections and the Pork Barrel.
of Politics
Journal
56(2):377399.
Stewart, III, Charles & Jonathan Woon. 2013. Congressional Committee Assignments, 103rd to 112th Congresses, 19932011: House Committees..
Wiseman, Alan E. 2004. Tests of Vote-Buyer Theories of Coalition Formation in Legislatures.
Research Quarterly
57(3):441450.
26
Political
Online Appendix To be made publicly available upon publication
A Adaptation of Dekel, Jackson & Wolinsky's (2009) VoteBuying Model While Dekel, Jackson & Wolinsky (2009) explore a range of political variables and outcomes, our interest is limited to the equilibrium distribution of payments across the ideological spectrum of voters. To elucidate this as quickly as possible, we present just the key features of the model necessary to motivate an understanding of the equilibrium solution and the distribution of payments. We adapt the notation with trivial modi cations when convenient, and we adjust the variables to the setting of the distribution of federal projects as manipulated by political parties in Congress. We refer the interested reader to the presentation in DJW (2009) for the full details. We suppose an odd number of legislators,
i ∈ N,
and two parties,
R
and
L.
At the start of the
game, the party that lacks ex ante support for its preferred outcome has the opportunity to make the rst bid (in the example below, this would be party
R),
where a bid consists of a promise of outlays to
one or more legislators, and payments to dierent legislators need not be the same size. The parties then alternate bids, while incurring a non-zero cost to bidding each round. Oers by a party to a legislator cannot be less than previous oers made by the same party to the same legislator, an assumption that oers are irreversible. While oers are binding, if party
R,
a higher oer to a particular legislator than party party
R
L makes
that legislator will now be voting for party
L,
so
is free to reallot the funds it had previously promised to the legislator. This feature becomes
important in motivating what DJW refer to as the shadow price of buying a legislator's vote: the legislator's support itself costs money, but there is an additional cost of having freed up the promises of the opposing party. A legislator
i
receives utility
pki + Vik ,
where
pki
is the payment she receives from party
the utility she receives from voting for the outcome preferred by party
k.
35 Legislator
for the outcome proposed by party Denote the median legislator by
35 Based
L
i
and
Vik
These ex ante valuations,
are assumed to be exogenously xed throughout. We arrange the legislators such that is a non-increasing function in i, as in Figure 3.
k
is
Vik ,
v(i) = ViL − ViR
i follows her dominant strategy and votes
L R R pL i + Vi > pi + Vi .
m = (N + 1)/2.
on Figure 1 in DJW (2009).
27
We asssume without loss of generality that party
Figure 3: Visualizing the Equilibrium Distribution of Payments 8
7
6
5
ViL - ViR
4
3
2
_
1
V
0
1
2
3
4
5
6
7
(m)
8
9
10
11
(n)
-1
-2
-3
Legislators i∈N
L
has ex ante majority support, such that
ante the proposal that party
V¯ =
Pn
i=m
vi ,
L
v(m) > 0.
prefers be denoted by
Let the rightmost legislator that supports ex
n = arg max{i : vi > 0}.
Then we may de ne
admittedly omitting non-trivial nuances regarding the smallest units of payment.
is this quantity
V¯
36 It
that will be particularly important in characterizing the payments in equilibrium.
Note that, as depicted visually in Figure 3 (as well as Figure 1 of DJW (2009)),
V¯
represents an
area that is a right triangle of sorts, with its hypotenuse sloping (by steps) downwards away from the median (to the right, if we assume
L
has ex ante majority support). Lastly, let the valuations of the
parties for the passage of their preferred proposal be denoted as
WL
and
W R.
Suppose that the party with ex ante minority support for its preferred proposal begins the bidding; there is no need for the party with ex ante majority support to preempt in a potentially in nite game as there will always be an opportunity to respond if necessary.
Proposition 3 in DJW (2009) then
states that, for su ciently small per-round bidding costs and given our additional assumption that party
36 In
R
always makes the rst move, one of two cases obtains in any equilibrium:
fact, the existence of a smallest unit of payment (equal to one in the gure) is the reason that the shaded area
extends above the valuation lines in Figure 3. Our omission of this detail in the exposition in no way drives the core result regarding equilibrium payments.
28
1. If
W R > W L,
party
R
wins at cost
V¯
paid to the legislators
m
(median) through
n
(almost-
indierent).
2. If
W R ≤ W L,
party
L
wins at no cost, as party
In equilibrium, we either observe party
L
R
will not initiate bidding.
win and make no payments, or we observe party
and make payments that correspond to the region
V¯ .
R
win
These payments are approximately the sum of
the valuations of each legislator from the median to the farthest-right legislator that ex ante preferred party
L's
preferred outcome. Party
R
must procure the support of the median and all legislators to
the right of the median, a winning coalition in a simple majority-rule setting. The non-zero cost to bidding discourages parties from bidding for legislators' votes when they know they will not have to deliver on their promises. Given the setting of complete and perfect information, each party applies backwards induction, bidding only if they must in order to win. Recall that in this example party voter), so party
L
has ex ante support for its preferred outcome (i.e. has the support of the median
R has the choice of making the rst bid.
preferred outcome more than party that party
R
L
R
Intuitively, then, if party
L values winning its
values winning its preferred outcome, then party
R
is willing to bid higher and will thus stay out of the bidding entirely. If, however, party
values the bill more, it will procure the necessary votes to obtain a majority and party
party
R's
R
L,
realizing
willingness to outbid, will make no promises to any legislators.
The legislators whose valuations are represented by the area party
will realize
V¯
are the cheapest legislators that
could have bought to form this winning coalition. When payments occur, therefore, they are
directly exclusively to near-median legislators. Moreover, past a certain point, more extreme legislators receive no payments whatsoever.
29
B Data: Sources & Variables Created The data sources are introduced more less in the order in which they are invoked in the .do le. Data and .do le are available upon request, as is the regression replication .do le.
B.1 General Data on Federal Government Using the sources below, a le was created with facts about the broad political landscape. Variables include year and congress (e.g. 98th, 111th) identi ers as well as the size of each party in each chamber, the party with the majority in each chamber, the party of the president, whether it was a period of uni ed government (Senate and House majorities of the same party as the president), and an in ation multiplier (with 2010 set as the base year).
http://history.house.gov/Institution/Party-Divisions/Party-Divisions/ http://www.senate.gov/pagelayout/history/one_item_and_teasers/partydiv.htm http://www.bls.gov/data/inflation_calculator.htm To this was merged a le with state names, FIPS and ICPSR codes, and years, such that the resulting le had year-state observations from 1983-2009.
B.2 Ideology Data B.2.1
NOMINATE
Poole & Rosenthal's NOMINATE data (Poole & Rosenthal 1985, Poole & Rosenthal 1987, Poole & Rosenthal 1991, Poole, Rosenthal & Koford 1991) are used not only for our ideology variables but for the information they provide on a number of other legislator characteristics. As explained in the paper, DW-NOMINATE Common Space scores were used.
http://voteview.com/dwnomin_joint_house_and_senate.htm First a variable tracking the tenure of legislators is created, simply counting the number of terms the legislator has served in Congress. Note that a legislator's ICPSR ID number follows her between chambers, but not if she switches party. Observations for seats that were split by two representatives in a single Congress were dropped.
This covers retirements and deaths, in which a special election
or appointment was made to ll the seat (assuming the replacement cast enough votes to receive a NOMINATE score for that congress) as well as representatives switching parties, receiving a dierent ID number mid-congress.
In the nearly 30 years of our sample, around 100 cases fell into one of
those categories. Two primary alternatives to dealing with split seats were both tried, with eectively identical results: 1) The entire congress could have been assigned to the member who cast the greatest
30
number of votes, as in Berry, Burden & Howell (2010); or 2) Each year could be assigned to the member that served the majority of that year, favoring slightly those members that served the earlier part of the year in which the budget was under construction. To reiterate, though, all methods were tried and results never varied signi cantly.
B.2.2
CVP
For an additional measure of ideology, Fowler & Hall's (2012) Conservative Vote Probabilities (CVP) were used.
http://www.andrewbenjaminhall.com/papers/ B.2.3
Party Unity
Party Unity scores were also used, where a party unity vote is de ned as one where at least 50 percent of Democrats vote against at least 50 percent of Republicans.
http://pooleandrosenthal.com/party_unity.htm B.2.4
Derived Ideology Measures
From the three basic measures of ideology discussed above, a number of derivative variables were created. For NOMINATE and CVP scores, the central variable was absolute distance from the House median in ideology scores. Absolute rank from the median was created similarly. Using these distance measures and the party unity scores, a number of interaction variables (e.g. with majority/minority and Democratic/Republican parties) and other functional forms (e.g. log, square, square root) were tested.
B.3 Electoral Margins From vote shares for the two major parties in congressional elections, a dummy variable was created equal to 1 if the margin of victory was within 5%. For all years, the most recent past electoral outcomes were utilized.
http://library.cqpress.com/elections/ From these data and the availability of the geographic correspondences described below, a variable denoting whether there exists a geographic correspondence for a district since its last redistricting was created. In practice, however, Congressional Quarterly appears to designate a district as having been redistricted if any district in the state was redistricted.
31
Excluding all districts for which there was
not an updated geographic correspondence since the last redistricting date seemed overly conservative and created a signi cant amount of unevenness in our sample, so the most current geographic correspondence data were employed, catching all major redistrictings and most if not all of the minor redistrictings. Presidential election results at the state level mark the last of the electoral data. The key variable created is, for each year and each state, the absolute value of the state-wide vote margin in the previous presidential election.
B.4 Committee Membership Data From Nelson (2013) and Stewart & Woon (2013) come data on committee membership.
The key
variables were dummies for being a party leader (though in practice many Speakers of the House lack enough votes to be given NOMINATE scores and are thus
de facto
excluded from our dataset),
being a committee chair, or being the ranking minority member. Additionally, dummy variables for membership on all of the individual standing committees were generated, although these were left out of the baseline model in the paper. (The member xed eects strategy leaves little variation for detecting the eect of a representative's tenure on a given committees.)
http://web.mit.edu/17.251/www/data_page.html Poole and Rosenthal continually update their ID numbers, but the committee data re ects what appear to be un-updated ID numbers. As a result, corrections in ICPSR ID numbers in the committee le were required, and appear in an auxiliary .do le.
B.5 County-Level Outlays Data on county-level outlays come from the Consolidated Federal Funds Report (CFFR), and as discussed in the paper, data from scal year
t
are matched with data from calendar year
t − 1.
CFFR
data span from 1983-2010, when the Federal Financial Statistics program remained in operation. The year 1983 is dropped, as there does not exist geographic correspondence data for 1982, the last year of a previous redistricting period. The corresponding calendar years for the explanatory variables is then 1983-2009. Outlays are in ation adjusted, putting all amounts in 2010 dollars.
http://www.census.gov/govs/cffr/ All negative entries, which likely re ect an accounting exercise, are dropped. The outlays are then split into dierent types of transfers and are summed by county, state, year, and the type of transfer. From the FAADS data (discussed below), a correspondence was created between program ID num-
32
bers and all of the types of transfers (or types of grants) funds were issued under for a given program ID number. After merging this le into the outlays data, formula grants could be culled out of the grants category, leaving block grants, project grants, and cooperative agreements. The entire grants category as well as only formula grants were also separately analyzed. For all characterizations of outlays, funds were summed, leaving one county-state-year observation for each of the characterizations.
B.6 Geographic Correspondences (County-to-District & District-to-District) A county-to-district crosswalk was created, containing all of the relevant population correspondences among counties and congressional districts, using each raw correspondence le repeatedly for all of the years until a new le is available. As mentioned above, this seems to cover all of the major redistrictings and most if not all of the minor redistrictings. Even for the occasional district that might have been redistricted and that lacks an updated le for a couple years, the correspondence is still likely quite accurate. The les may be found at the links below.
http://mcdc.missouri.edu/websas/geocorr12.html http://mcdc.missouri.edu/websas/geocorr2k.html http://mcdc.missouri.edu/websas/geocorr90.html These data identify counties that were represented by only one district. Additionally, with these data one can determine what proportion of a district's population remains in the district across redistricting. Using varying thresholds of population carryover, it is possible to look at within-member variation that spans redistrictings.
Further explanation of the use of this method appears in the
footnotes for Tables 17, 18, and 19.
B.7 County-Level Demographics Income and population at the county level come from the Bureau of Economic Analysis (BEA).
http://bea.gov/iTable/index_regional.cfm CFFR does not parse apart New York City, which is not a problem for this study as it would be dropped anyway under the restriction that counties be contained in only one district. Additionally, several Alaskan boroughs and Virgina cities (Virginia has a number of cities not within counties) appear in BEA but not in CFFR, or at least not using the same codes. These instances do not ultimately, then, appear in the analysis.
33
B.8 District-Level Outlays The FAADS data on district-level outlays are compiled as per Bickers & Stein (1991) and extended in the same manner as outlined in Berry, Burden & Howell (2010), and the reader is referred to both of those papers for more detail. The only real departure in this study is the breakout of grants by type, rather than coe cient of variation. The no-formula category is constructed to mirror the CFFR no-formula category, as it excludes only formula-based spending and includes block grants, project grants, and cooperative agreements. The all-grants measure includes formula-based spending.
34
C Supplementary Analyses Table 5: Additional Summary Statistics at the County Level Obs
Mean
Std. Dev.
Min
Max
Absolute distance from median
Variable
83237
.323
.21
0
1.172
Absolute rank from median (/100)
83237
1.886
1.227
.01
4.32
Absolute distance x majority
83237
.113
.141
0
.865
Distance from median (sq)
83237
.149
.162
0
1.374
Distance from median (sqrt)
83237
.53
.206
0
1.083
Distance from median (ln)
83026
-1.479
1.054
-7.601
.159
Distance x Republican
83237
.211
.241
0
1.172
Distance x Democrat
83237
.113
.183
0
.964
Distance from within party mean
83237
.123
.097
0
.642
Party unity score
83237
84.787
12.948
9.36
100
Uni ed government
83237
.258
.437
0
1
Majority party size
83237
243.83
17.74
220
270
Distance (CVP) from median
83205
.201
.16
0
.631
Rank (CVP) from median (/100)
83205
1.935
1.234
.01
4.32
Maj on far side of median
83237
.092
.289
0
1
Majority party
83237
.585
.493
0
1
President's party
83237
.497
.5
0
1
Committee chair
83093
.042
.201
0
1
Ranking minority member
83093
.052
.222
0
1
Party leader
83093
.003
.055
0
1
First term
83237
.145
.352
0
1
Tenure (# terms)
83237
5.039
3.81
1
28
Close election
83139
.076
.266
0
1
State presidential margin
83237
13.878
9.935
0
55.94
Log income
83031
13.122
1.541
7.555
19.85
Log population
83031
10.184
1.393
4.007
16.097
Log grants (no formula)
84340
14.706
1.962
.405
22.979
Log grants (all)
84565
16.371
1.72
4.078
24.245
Log grants (formula only)
84556
15.963
1.824
3.717
23.97
Log retirement and disability (all)
84553
17.361
1.516
4.404
23.627
Notes: Absolute distance from median
is the absolute value of median-centered ideal point estimates
in DW-NOMINATE Common Space scores.
All variations of this variable are explained in the
footnotes of the results tables in which they are used. Further explanations here are given mostly
Majority party is a dummy variable equal to 1 if a legislator is in the majority President's party is a dummy variable equal to 1 if a legislator shares the same party as the sitting president. Committee chair is a dummy variable equal to 1 if a legislator is the chair of any standing House committee. Ranking minority member is a dummy variable equal to 1 if a legislator is the ranking minority member on any standing House committee. Party leader is a dummy variable equal to 1 if a legislator has a party leadership role. First term is a dummy variable equal to 1 for a legislator's rst term in Congress. Tenure (# terms) counts the number of Congresses in which a legislator has served. Close election is a dummy variable equal to 1 if the margin of victory in the previous congressional election was 5% or less. State presidential margin is absolute value of the state-wide vote margin in the previous presidential election. Log income is the natural log of countylevel income. Log population is the natural log of county-level population. Log grants (no formula) for control variables. party.
is the log value of non-formula grants received by a given county in a given year, as measured in the Consolidated Federal Funds Report. The other measures of outlays are explained in the footnotes of the results tables in which they are used.
35
Table 6: Experimenting with Fixed Eects
Absolute distance from median Majority party President's party Committee chair Ranking minority member Party leader Party First term Tenure (# terms) Close election
(1)
(2)
(3)
(4)
-0.086
-0.169*
-0.746***
-0.647**
(0.200)
(0.090)
(0.276)
(0.307)
-0.042
-0.013
-0.229**
-0.190*
(0.078)
(0.035)
(0.093)
(0.100)
0.029
0.030
0.005
0.020
(0.029)
(0.029)
(0.035)
(0.032)
0.146
-0.071
-0.022
0.023
(0.171)
(0.045)
(0.076)
(0.073)
0.094
-0.017
-0.043
-0.037
(0.129)
(0.055)
(0.052)
(0.057)
-0.386**
0.165**
-0.037
0.093
(0.185)
(0.079)
(0.048)
(0.088)
-0.129
0.011
(0.077)
(0.035)
0.030
0.009
0.006
0.013
(0.036)
(0.022)
(0.025)
(0.023)
0.001
0.005
-0.177
-0.145
(0.008)
(0.004)
(0.123)
(0.097)
0.063
0.075**
0.122***
0.125***
(0.048)
(0.031)
(0.033)
(0.030)
0.012***
0.001
0.003
0.003
(0.004)
(0.002)
(0.002)
(0.002)
-0.006
-0.062
-0.456***
-0.108
(0.180)
(0.146)
(0.119)
(0.152)
1.046***
-0.108
1.604***
-0.033
(0.201)
(0.161)
(0.126)
(0.203)
2.458***
14.657***
2.611***
14.525***
(0.418)
(1.514)
(0.473)
(1.449)
Fixed Eects
None
County
Member
Cnty-Member
R2
0.590
0.350
0.548
0.278
70990
70990
70990
70990
State presidential margin Log income Log population Constant
Adj. N
Notes:
Standard errors clustered by state.
additional xed eects are used.
Year xed eects used in all models.
In model 1, no
In model 2, county-level xed eects are used.
Model 3 is our
baseline model (model 2 from Table 2 of the paper), in which we use member xed eects. Model 4 uses county-by-member xed eects. Models 3-4, which condition on each member, do not admit
Party as an additional control, as party is constant within all legislator ICPSR ID numbers; legislators
who switch parties receive a new ID number. The dependent variable is the log value of non-formula grants received by a given county in a given year.
The outlays data span scal years 1984-2010
and are matched with explanatory variables from the previous calendar year.
from median
Common Space scores.
∗p
Absolute distance
is the absolute value of the median-centered rst dimension of the DW-NOMINATE
≤ 0.10 ∗∗ p ≤ 0.05 ∗∗∗ p ≤ 0.01
36
Table 7: Variations on Absolute Distance
Absolute distance from median Distance from median (sq)
(1)
(2)
(4)
(5)
-0.794***
-0.781*
(3)
-0.776**
-1.150***
(0.221)
(0.392)
(0.299)
(0.350)
(6)
0.068 (0.233)
Distance from median (sqrt)
0.035 (0.180)
Distance from median (ln)
-0.025
0.003
(0.019)
(0.014)
Distance x Republican
1.453*
0.303
(0.768)
(0.488)
Distance x Democrat
-1.150*** (0.350)
Majority party President's party Committee chair Ranking minority member Party leader First term Tenure (# terms)
-0.228**
-0.229**
-0.011
-0.234**
-0.113
-0.113
(0.092)
(0.094)
(0.036)
(0.092)
(0.070)
(0.070)
0.005
0.005
0.021
0.006
0.012
0.012
(0.035)
(0.035)
(0.039)
(0.036)
(0.039)
(0.039)
-0.022
-0.022
-0.024
-0.023
-0.022
-0.022
(0.076)
(0.076)
(0.077)
(0.076)
(0.075)
(0.075)
-0.044
-0.043
-0.025
-0.043
-0.047
-0.047
(0.052)
(0.052)
(0.064)
(0.052)
(0.049)
(0.049)
-0.037
-0.037
-0.031
-0.038
-0.036
-0.036
(0.048)
(0.048)
(0.046)
(0.048)
(0.046)
(0.046)
0.006
0.006
0.010
0.006
0.004
0.004
(0.025)
(0.025)
(0.024)
(0.025)
(0.025)
(0.025)
-0.178
-0.177
-0.170
-0.177
-0.182
-0.182
(0.123)
(0.123)
(0.121)
(0.123)
(0.124)
(0.124)
0.122***
0.122***
0.122***
0.122***
0.116***
0.116***
(0.033)
(0.033)
(0.032)
(0.033)
(0.033)
(0.033)
0.003
0.003
0.003
0.003
0.003
0.003
(0.002)
(0.002)
(0.002)
(0.002)
(0.002)
(0.002)
-0.456***
-0.456***
-0.454***
-0.454***
-0.454***
-0.454***
(0.119)
(0.119)
(0.117)
(0.117)
(0.119)
(0.119)
Log population
1.604***
1.604***
1.602***
1.601***
1.602***
1.602***
(0.126)
(0.126)
(0.124)
(0.124)
(0.126)
(0.126)
Constant
2.615***
2.604***
2.206***
2.629***
2.256***
2.256***
(0.463)
(0.458)
(0.363)
(0.487)
(0.398)
(0.398)
0.548
0.548
0.547
0.548
0.548
0.548
70990
70990
70792
70792
70990
70990
Close election State presidential margin Log income
Adj.
R2
N
Notes:
Standard errors clustered by state. Member and year xed eects used in all models. The dependent variable
is the log value of non-formula grants received by a given county in a given year. The outlays data span scal years
1984-2010 and are matched with explanatory variables from the previous calendar year. Absolute distance from median is the absolute value of the median-centered rst dimension of the DW-NOMINATE Common Space scores. Distance from median (sq) is Absolute distance from median squared. Distance from median (sqrt) is the square root of Absolute distance from median. Distance from median (ln) is the natural log of Absolute distance from median. Distance × Republican/Democrat is Absolute distance from median interacted with a dummy variable for membership in each party.
∗p
≤ 0.10 ∗∗ p ≤ 0.05 ∗∗∗ p ≤ 0.01
37
Table 8: Spline Analysis 0.1
0.3
0.5
0.7
0.9
1.1
(1)
(2)
(3)
(4)
(5)
(6)
Distance from median - spline 1
-0.825*
-0.772***
-0.747***
-0.757***
-0.746***
-0.746***
(0.466)
(0.234)
(0.249)
(0.263)
(0.276)
(0.276)
Distance from median - spline 2
-0.740**
-0.728**
-0.741*
-0.429
-0.814
2.007
(0.281)
(0.325)
(0.434)
(0.756)
(0.622)
(6.056)
-0.228**
-0.228**
-0.229**
-0.230**
-0.229**
-0.228**
(0.093)
(0.092)
(0.092)
(0.091)
(0.093)
(0.093)
0.005
0.005
0.005
0.005
0.005
0.005
(0.035)
(0.035)
(0.035)
(0.035)
(0.035)
(0.035)
Majority party President's party Committee chair Ranking minority member Party leader First term Tenure (# terms)
-0.023
-0.022
-0.022
-0.022
-0.022
-0.022
(0.076)
(0.076)
(0.075)
(0.076)
(0.076)
(0.076)
-0.043
-0.044
-0.043
-0.043
-0.043
-0.043
(0.052)
(0.052)
(0.053)
(0.052)
(0.052)
(0.052)
-0.037
-0.037
-0.037
-0.037
-0.037
-0.037
(0.048)
(0.049)
(0.048)
(0.048)
(0.048)
(0.048)
0.006
0.006
0.006
0.006
0.006
0.006
(0.025)
(0.025)
(0.025)
(0.025)
(0.025)
(0.025)
-0.177
-0.177
-0.177
-0.181
-0.177
-0.179
(0.122)
(0.122)
(0.123)
(0.124)
(0.125)
(0.123)
0.122***
0.121***
0.122***
0.122***
0.122***
0.122***
(0.033)
(0.033)
(0.033)
(0.033)
(0.033)
(0.033)
0.003
0.003
0.003
0.003
0.003
0.003
(0.002)
(0.002)
(0.002)
(0.002)
(0.002)
(0.002)
-0.456***
-0.456***
-0.456***
-0.456***
-0.456***
-0.456***
(0.119)
(0.119)
(0.119)
(0.119)
(0.119)
(0.119)
Log population
1.604***
1.604***
1.604***
1.604***
1.604***
1.604***
(0.126)
(0.126)
(0.126)
(0.126)
(0.126)
(0.126)
Constant
2.617***
2.615***
2.611***
2.609***
2.612***
2.609***
(0.472)
(0.463)
(0.468)
(0.473)
(0.475)
(0.474)
0.548
0.548
0.548
0.548
0.548
0.548
70990
70990
70990
70990
70990
70990
Close election State presidential margin Log income
Adj.
R2
N
Notes:
Standard errors clustered by state. Member and year xed eects used in all models. The dependent variable
is the log value of non-formula grants received by a given county in a given year. The outlays data span scal years 1984-2010 and are matched with explanatory variables from the previous calendar year.
Distance from median
is
the absolute value of the median-centered rst dimension of the DW-NOMINATE Common Space scores and is allowed to take two slopes, one on either side of a knot point. connecting spline 1 to spline 2. Spline 1 estimates the eect of while spline 2 estimates the eect of
Distance from median
The number above each model denotes the knot
Distance from median
from zero to the knot point,
from the knot point and higher.
The knot point is
allowed to increase across the models, such that spline 1 covers an increasingly large share of the estimation, while spline 2 covers less.
∗p
≤ 0.10 ∗∗ p ≤ 0.05 ∗∗∗ p ≤ 0.01
38
Table 9: Distance from Party Mean (1)
(2)
(3)
Absolute distance from median Distance from within party mean
(4)
(5)
(6)
-0.144**
-0.698***
-0.668***
(0.072)
(0.245)
(0.226)
1.349***
1.527***
1.520***
1.554***
1.449***
1.449***
(0.475)
(0.512)
(0.519)
(0.490)
(0.528)
(0.530)
0.043*
0.042
-0.192**
-0.183**
(0.024)
(0.025)
(0.085)
(0.080)
0.019
0.022
0.004
0.008
(0.038)
(0.034)
(0.034)
(0.030)
Majority party President's party Committee chair Ranking minority member Party leader First term Tenure (# terms)
0.000
-0.027
-0.002
-0.029
(0.077)
(0.072)
(0.074)
(0.069)
0.006
-0.018
-0.016
-0.037
(0.062)
(0.058)
(0.047)
(0.047)
0.025
-0.037
0.018
-0.043
(0.044)
(0.049)
(0.048)
(0.048)
-0.001
-0.010
-0.005
-0.015
(0.024)
(0.023)
(0.024)
(0.022) -0.265**
-0.194
-0.261**
-0.201
(0.119)
(0.127)
(0.120)
(0.129)
0.120***
0.117***
0.121***
0.118***
(0.032)
(0.028)
(0.032)
(0.028)
0.003
0.003
0.003
0.003
(0.002)
(0.002)
(0.002)
(0.002)
-0.453***
-0.457***
-0.460***
-0.452***
-0.456***
-0.459***
(0.117)
(0.119)
(0.120)
(0.118)
(0.119)
(0.120)
Log population
1.601***
1.606***
1.608***
1.600***
1.604***
1.607***
(0.124)
(0.126)
(0.126)
(0.125)
(0.126)
(0.127)
Constant
2.361***
2.025***
1.987***
2.385***
2.398***
2.341***
(0.329)
(0.374)
(0.371)
(0.333)
(0.477)
(0.454)
No
No
Yes
No
No
Yes
0.547
0.548
0.548
0.547
0.548
0.549
71147
70990
70990
71147
70990
70990
Close election State presidential margin Log income
Committee dummies Adj.
R2
N
Notes:
Standard errors clustered by state. Member and year xed eects used in all models. The dependent variable
is the log value of non-formula grants received by a given county in a given year. The outlays data span scal years
1984-2010 and are matched with explanatory variables from the previous calendar year. Absolute distance from median is the absolute value of the median-centered rst dimension of the DW-NOMINATE Common Space scores. Distance from within party mean is the absolute value of the mean-centered rst dimension of the DW-NOMINATE Common Space scores, by party, thus measuring each legislator's ideological distance from her party's mean.
∗p
≤ 0.10 ∗∗ p ≤ 0.05 ∗∗∗ p ≤ 0.01
39
Table 10: Party Unity
Party unity score
(1)
(2)
(3)
(4)
0.004
0.001
0.004*
0.001
(0.002)
(0.002)
(0.002)
(0.002) Party unity x majority
0.007**
0.006*
(0.004)
(0.003)
Absolute distance from median
-0.741***
Majority party
Ranking minority member Party leader
-0.648**
-0.251***
-0.628**
(0.027)
(0.307)
(0.089)
(0.270)
0.026
0.012
0.010
0.006
(0.042)
(0.036)
(0.037)
(0.036)
-0.027
-0.032
-0.028
-0.031
(0.074)
(0.072)
(0.072)
(0.072)
-0.023
-0.041
-0.045
-0.049
(0.065)
(0.050)
(0.050)
(0.048)
-0.031
-0.032
-0.036
-0.034
(0.046)
(0.041)
(0.049)
(0.043)
First term Tenure (# terms)
(0.138)
-0.004
President's party Committee chair
-0.411***
(0.240)
0.006
0.003
0.001
0.001
(0.024)
(0.024)
(0.025)
(0.025)
-0.182
-0.189
-0.191
-0.192
(0.122)
(0.124)
(0.124)
(0.125)
0.124***
0.122***
0.124***
0.122***
(0.032)
(0.031)
(0.033)
(0.032)
0.003
0.003
0.003
0.003
(0.002)
(0.002)
(0.002)
(0.002)
-0.456***
-0.456***
-0.455***
-0.456***
(0.119)
(0.119)
(0.119)
(0.119)
Log population
1.604***
1.605***
1.603***
1.604***
(0.126)
(0.126)
(0.126)
(0.126)
Constant
1.938***
2.225***
2.330***
2.372***
(0.355)
(0.446)
(0.424)
(0.438)
0.548
0.548
0.548
0.548
70990
70990
70990
70990
Close election State presidential margin Log income
Adj.
R2
N
Notes:
Standard errors clustered by state.
Member and year xed eects used in all models.
The dependent variable is the log value of non-formula grants received by a given county in a given year. The outlays data span scal years 1984-2010 and are matched with explanatory variables from the previous calendar year.
Absolute distance from median
is the absolute value
of the median-centered rst dimension of the DW-NOMINATE Common Space scores.
unity score
Party
is the percentage of votes in which at least 50% of Democrats vote against at least
50% of Republicans that a legislator voted with her party. This variable is also interacted with the dummy variable for a member's being in the majority party.
∗p
≤ 0.10 ∗∗ p ≤ 0.05 ∗∗∗ p ≤ 0.01
40
Table 11: Absolute Distance & Rank from Median - CVP
Distance (CVP) from median
(1)
(2)
(3)
-0.102
-0.434*
-0.431*
(0.072)
(0.257)
(0.237)
Rank (CVP) from median (/100) Majority party President's party Committee chair Ranking minority member Party leader First term Tenure (# terms)
(4)
(5)
(6)
-0.017
-0.068**
-0.064**
(0.011)
(0.030)
(0.028)
-0.108
-0.107
-0.119*
-0.109*
(0.081)
(0.076)
(0.064)
(0.060)
0.009
0.013
0.017
0.020
(0.035)
(0.031)
(0.037)
(0.033)
-0.026
-0.056
-0.027
-0.057
(0.077)
(0.072)
(0.074)
(0.070)
-0.038
-0.060
-0.042
-0.063
(0.057)
(0.054)
(0.054)
(0.053)
-0.030
-0.092*
-0.021
-0.083*
(0.047)
(0.047)
(0.048)
(0.046)
0.009
-0.002
0.011
0.000
(0.025)
(0.022)
(0.024)
(0.022) -0.232*
-0.166
-0.231*
-0.168
(0.122)
(0.132)
(0.123)
(0.133)
0.119***
0.116***
0.114***
0.112***
(0.032)
(0.028)
(0.033)
(0.029)
0.003
0.003
0.003
0.002
(0.002)
(0.002)
(0.002)
(0.002)
-0.451***
-0.456***
-0.458***
-0.451***
-0.454***
-0.456***
(0.118)
(0.119)
(0.120)
(0.118)
(0.119)
(0.120)
Log population
1.598***
1.604***
1.606***
1.598***
1.602***
1.605***
(0.125)
(0.126)
(0.126)
(0.125)
(0.126)
(0.126)
Constant
2.523***
2.388***
2.344***
2.540***
2.452***
2.400***
(0.326)
(0.447)
(0.429)
(0.329)
(0.443)
(0.427)
Close election State presidential margin Log income
Committee dummies Adj.
R2
N
Notes:
No
No
Yes
No
No
Yes
0.547
0.548
0.548
0.547
0.548
0.548
71115
70958
70958
71115
70958
70958
Standard errors clustered by state. Member and year xed eects used in all models. The dependent variable
is the log value of non-formula grants received by a given county in a given year. The outlays data span scal years 1984-2010 and are matched with explanatory variables from the previous calendar year.
median
Distance (CVP) from
is the absolute value of legislators' Conservative Vote Probabilities (CVP scores), which are already median-
centered.
Rank (CVP) from median (/100)
is the rank ordering of the
Distance (CVP) from median
variable,
divided by 100 for scaling purposes; the higher the rank, the farther a legislator is ideologically from the median.
∗p
≤ 0.10 ∗∗ p ≤ 0.05 ∗∗∗ p ≤ 0.01
41
Table 12: Uni ed Government
Uni ed government
(1)
(2)
(3)
(4)
4.777***
4.973***
5.167***
5.022*** (1.618)
(1.581)
(1.602)
(1.600)
0.018
-0.158**
-0.256***
-0.118
(0.037)
(0.077)
(0.078)
(0.133)
-1.030***
-1.031***
-1.098***
(0.327)
(0.318)
(0.322)
Unif. gov't x majority Absolute distance from median Unif. gov't x distance
-0.279**
-0.092
(0.139)
(0.148)
Absolute distance x majority
0.231 (0.222)
Unif. gov't x maj'ty x distance
-0.372 (0.338)
Majority party
0.014
-0.258***
-0.263***
-0.328***
(0.032)
(0.082)
(0.081)
(0.121)
0.014
0.061
0.056
0.053
(0.040)
(0.050)
(0.050)
(0.047)
President's party Committee chair Ranking minority member Party leader
-0.021
-0.026
-0.026
-0.024
(0.078)
(0.074)
(0.074)
(0.074)
-0.021
-0.055
-0.057
-0.056
(0.068)
(0.049)
(0.048)
(0.047)
-0.033
-0.037
-0.028
-0.028
(0.045)
(0.046)
(0.045)
(0.044)
First term Tenure (# terms)
0.011
0.003
0.005
0.007
(0.024)
(0.025)
(0.025)
(0.025)
-0.169
-0.177
-0.180
-0.178
(0.122)
(0.124)
(0.123)
(0.125)
0.122***
0.119***
0.114***
0.115***
(0.032)
(0.032)
(0.032)
(0.032)
0.003
0.003
0.003
0.003
(0.002)
(0.002)
(0.002)
(0.002)
-0.457***
-0.455***
-0.455***
-0.456***
(0.119)
(0.119)
(0.119)
(0.120)
Log population
1.605***
1.603***
1.604***
1.604***
(0.126)
(0.126)
(0.126)
(0.127)
Constant
2.229***
2.698***
2.701***
2.737***
(0.368)
(0.480)
(0.475)
(0.488)
0.548
0.548
0.548
0.548
70990
70990
70990
70990
Close election State presidential margin Log income
Adj.
R2
N
Notes:
Standard errors clustered by state.
Member and year xed eects used in all models.
The dependent variable is the log value of non-formula grants received by a given county in a given year. The outlays data span scal years 1984-2010 and are matched with explanatory variables from the previous calendar year.
Absolute distance from median
is the absolute value
of the median-centered rst dimension of the DW-NOMINATE Common Space scores.
government
Uni ed
denotes years in which the House and Senate majorities were both of the same party
Uni ed government with the dummy Absolute distance, model 3 interacts Uni ed
as the president. All models include a variable interacting variable for majority status. Model 2 reintroduces
government Absolute distance, and model 4 fully saturates the regression by interacting all terms with the dummy variable for majority status.
∗p
≤ 0.10 ∗∗ p ≤ 0.05 ∗∗∗ p ≤ 0.01
42
Table 13: Majority Party Size
Majority party size Majority x maj pty size
(1)
(2)
(3)
(4)
-0.077***
-0.078***
-0.080***
-0.080***
(0.021)
(0.021)
(0.022)
(0.022)
-0.000
-0.002
-0.000
-0.000
(0.002)
(0.002)
(0.002)
(0.002)
-0.774***
-1.430*
-1.529**
(0.272)
(0.741)
(0.733)
Absolute distance from median Abs dist x maj pty size
0.003
0.003
(0.004)
(0.003)
Abs dist x majority x maj pty size
0.000 (0.001)
Majority party President's party Committee chair Ranking minority member Party leader First term Tenure (# terms) Close election State presidential margin Log income Log population Constant Adj.
R2
N
Notes:
0.076
0.128
-0.132
-0.212
(0.472)
(0.482)
(0.513)
(0.503)
0.021
0.005
0.007
0.007
(0.039)
(0.035)
(0.035)
(0.034)
-0.021
-0.024
-0.023
-0.023
(0.078)
(0.076)
(0.076)
(0.076)
-0.022
-0.046
-0.043
-0.042
(0.068)
(0.052)
(0.050)
(0.050)
-0.033
-0.038
-0.037
-0.038
(0.045)
(0.047)
(0.047)
(0.046)
0.011
0.006
0.005
0.006
(0.024)
(0.025)
(0.025)
(0.025)
-0.168
-0.175
-0.180
-0.179
(0.121)
(0.121)
(0.122)
(0.124)
0.121***
0.120***
0.120***
0.120***
(0.032)
(0.032)
(0.032)
(0.032)
0.003
0.003
0.003
0.003
(0.002)
(0.002)
(0.002)
(0.002)
-0.457***
-0.456***
-0.456***
-0.456***
(0.120)
(0.120)
(0.120)
(0.120)
1.605***
1.605***
1.604***
1.604***
(0.126)
(0.127)
(0.126)
(0.127)
23.112***
23.725***
24.274***
24.283***
(5.595)
(5.606)
(5.748)
(5.780)
0.548
0.548
0.548
0.548
70990
70990
70990
70990
Standard errors clustered by state. Member and year xed eects used in all models. The
dependent variable is the log value of non-formula grants received by a given county in a given year. The outlays data span scal years 1984-2010 and are matched with explanatory variables from the previous calendar year.
Absolute distance from median
is the absolute value of the median-centered
rst dimension of the DW-NOMINATE Common Space scores. number of members in the majority party in a given Congress.
∗p
measures the
This variable is interacted with
In model 2, Absolute distance is reintroduced, in model 3 Absolute distance is Majority party size, and in model 4 a nal, three-way interaction term is included.
majority status. interacted with
Majority party size
≤ 0.10 ∗∗ p ≤ 0.05 ∗∗∗ p ≤ 0.01
43
Table 14: Majority Members on Far Side of Median
Maj on far side of median
(1)
(2)
(3)
(4)
-0.140
-0.077
-0.014
-0.010
(0.091)
(0.059)
Maj far side of med X abs dist Absolute distance from median
(0.046)
(0.046)
-1.912**
-1.896**
(0.796)
(0.766)
-0.618***
-0.217
-0.242
(0.214)
(0.211)
(0.214)
Absolute distance x majority
0.062 (0.214)
Majority party President's party Committee chair Ranking minority member Party leader First term Tenure (# terms)
0.035
-0.178**
-0.036
-0.057
(0.026)
(0.074)
(0.085)
(0.099)
0.023
0.009
0.016
0.016
(0.040)
(0.036)
(0.039)
(0.038)
-0.019
-0.021
-0.023
-0.024
(0.080)
(0.077)
(0.074)
(0.073)
-0.029
-0.044
-0.053
-0.053
(0.059)
(0.051)
(0.048)
(0.048)
-0.038
-0.039
-0.038
-0.039
(0.044)
(0.047)
(0.044)
(0.044)
0.011
0.007
0.006
0.006
(0.024)
(0.025)
(0.025)
(0.025)
-0.180
-0.182
-0.180
-0.179
(0.123)
(0.124)
(0.125)
(0.127)
0.118***
0.120***
0.115***
0.115***
(0.032)
(0.033)
(0.033)
(0.033)
0.003
0.003
0.003
0.003
(0.002)
(0.002)
(0.002)
(0.002)
-0.456***
-0.455***
-0.454***
-0.454***
(0.119)
(0.119)
(0.120)
(0.120)
Log population
1.604***
1.603***
1.602***
1.602***
(0.126)
(0.126)
(0.126)
(0.127)
Constant
2.212***
2.539***
2.340***
2.353***
(0.375)
(0.453)
(0.409)
(0.435)
0.548
0.548
0.548
0.548
70990
70990
70990
70990
Close election State presidential margin Log income
Adj.
R2
N
Notes:
Standard errors clustered by state. Member and year xed eects used in all models. The
dependent variable is the log value of non-formula grants received by a given county in a given year. The outlays data span scal years 1984-2010 and are matched with explanatory variables from the previous calendar year.
Absolute distance from median
is the absolute value of the
median-centered rst dimension of the DW-NOMINATE Common Space scores.
of median
Maj on far side
is a dummy variable equal to one if a legislator is a member of the majority party but
has an estimated ideal point in the rst dimension of DW-NOMINATE Common Space scores that is on the far side of the median ideal point from the majority of the majority party. This variable is interacted in models 3 and 4 with
∗p
≤ 0.10 ∗∗ p ≤ 0.05 ∗∗∗ p ≤ 0.01
Absolute distance.
44
Table 15: Absolute Distance & Rank from Median - All Grants
Absolute distance from median
(1)
(2)
(3)
-0.061*
-0.357***
-0.337***
(0.030)
(0.089)
(0.076)
Absolute rank from median (/100) Majority party President's party Committee chair Ranking minority member Party leader First term Tenure (# terms)
(4)
(5)
(6)
-0.009*
-0.049**
-0.052***
(0.006)
(0.020)
(0.016)
-0.098***
-0.090***
-0.071*
-0.075**
(0.030)
(0.025)
(0.036)
(0.029)
0.000
-0.003
0.007
0.003
(0.011)
(0.011)
(0.012)
(0.012)
-0.003
-0.005
-0.003
-0.004
(0.023)
(0.023)
(0.023)
(0.022)
0.021
0.003
0.024
0.004
(0.024)
(0.024)
(0.025)
(0.025)
-0.052
-0.080*
-0.054
-0.082*
(0.039)
(0.043)
(0.040)
(0.043)
-0.016
-0.015
-0.016
-0.014
(0.014)
(0.011)
(0.013)
(0.011) -0.091*
-0.077
-0.087
-0.080
(0.054)
(0.052)
(0.056)
(0.053)
Close election
0.039***
0.041***
0.039***
0.041***
(0.014)
(0.014)
(0.014)
(0.014)
State presidential margin
0.003***
0.004***
0.003***
0.003***
(0.001)
(0.001)
-0.845***
-0.845***
(0.001)
(0.001) -0.845***
-0.839*** (0.101)
(0.101)
(0.100)
(0.101)
(0.101)
(0.101)
Log population
1.831***
1.837***
1.838***
1.831***
1.837***
1.837***
(0.104)
(0.104)
(0.103)
(0.104)
(0.104)
(0.103)
Constant
6.716***
6.742***
6.728***
6.717***
6.711***
6.720***
(0.256)
(0.296)
(0.287)
(0.256)
(0.299)
(0.286)
Committee dummies Adj.
R2
N
Notes:
No
No
Yes
No
No
Yes
0.792
0.792
0.792
0.792
0.792
0.792
71357
71199
71199
71357
71199
71199
Standard errors clustered by state.
variable is the log value of
all
-0.839***
-0.844***
Log income
Member and year xed eects used in all models.
The dependent
grants in the Consolidated Federal Funds Reports' grants category received by a given
county in a given year. The outlays data span scal years 1984-2010 and are matched with explanatory variables
Absolute distance from median is the absolute value of the median-centered rst Absolute rank from median (/100) is the rank ordering of the Absolute distance from median variable divided by 100 for scaling purposes; the higher the rank, the farther
from the previous calendar year.
dimension of the DW-NOMINATE Common Space scores. a legislator is ideologically from the median.
∗p
≤ 0.10 ∗∗ p ≤ 0.05 ∗∗∗ p ≤ 0.01
45
Table 16: Additional Summary Statistics at the District Level Obs
Mean
Std. Dev.
Min
Max
Absolute distance from median
Variable
11579
.367
.22
0
1.19
Majority party
11579
.561
.496
0
1
President's party
11579
.47
.499
0
1
Committee chair
11555
.049
.217
0
1
Ranking minority member
11555
.048
.213
0
1
Party leader
11555
.007
.085
0
1
First term
11579
.135
.342
0
1
Tenure (# terms)
11579
5.443
4.014
1
28
Close election
11569
.065
.246
0
1
State presidential margin
11579
12.881
8.964
0
55.94
FAADS Grants (no formula - B,P,CA) (ln)
11562
17.849
1.464
9.41
22.535
FAADS Grants (all - B,P,CA,F) (ln)
11571
19.436
1.209
9.679
23.644
Notes: Absolute distance from median
is the absolute value of median-centered ideal point estimates in
Majority party is a dummy variable equal to 1 if a legislator is President's party is a dummy variable equal to 1 if a legislator shares the same party as the sitting president. Committee chair is a dummy variable equal to 1 if a legislator is the chair of any standing House committee. Ranking minority member is a dummy variable equal to 1 if a legislator is the ranking minority member on any standing House committee. Party leader is a dummy variable equal to 1 if a legislator has a party leadership role. First term is a dummy variable equal to 1 for a legislator's rst term in Congress. Tenure (# terms) counts the number of Congresses in which a legislator has served. Close election is a dummy variable equal to 1 if the margin of victory in the previous congressional election was 5% or less. State presidential margin is absolute value of the state-wide vote margin in the previous presidential election. FAADS Grants (no formula - B,P,CA)(ln) is the log value of
DW-NOMINATE Common Space scores. in the majority party.
non-formula grants (block grants (B), project grants (P), and cooperative agreements (CA)) received by a given district in a given year, as measured in the Federal Assistance Award Data System.
(all - B,P,CA,F)(ln)
adds formula grants (F) to the preceding measure of outlays.
46
FAADS Grants
Table 17: District-Level Analysis - No Formula Full Sample
Absolute distance from median Majority party President's party Committee chair Ranking minority member Party leader First term Tenure (# terms) Close election State presidential margin Constant Adj.
R2
N
Notes:
County-Analysis Sample Only
(1) CQ
(2) 100%
(3) 80%
(4) CQ
(5) 100%
-0.081
-0.014
0.007
0.134
0.212
(6) 80% 0.158
(0.183)
(0.193)
(0.173)
(0.297)
(0.321)
(0.278)
-0.027
-0.006
0.004
0.044
0.057
0.035
(0.075)
(0.080)
(0.073)
(0.124)
(0.132)
(0.115)
-0.025
-0.018
-0.017
-0.027
-0.017
-0.019
(0.020)
(0.020)
(0.021)
(0.024)
(0.024)
(0.025)
-0.014
-0.017
-0.015
0.011
0.017
0.018
(0.059)
(0.054)
(0.054)
(0.084)
(0.082)
(0.081)
0.043
0.021
0.027
0.017
-0.013
-0.009
(0.058)
(0.056)
(0.055)
(0.054)
(0.057)
(0.056)
0.324
0.344
0.350
0.257
0.257
0.262
(0.215)
(0.223)
(0.219)
(0.176)
(0.169)
(0.168)
-0.039
-0.023
-0.021
0.024
0.049
0.050
(0.033)
(0.032)
(0.031)
(0.033)
(0.033)
(0.033)
-0.240***
-0.206**
-0.205**
0.021
0.072*
0.075**
(0.077)
(0.088)
(0.088)
(0.063)
(0.038)
(0.035)
0.044
0.037
0.033
0.067
0.062
0.052
(0.041)
(0.041)
(0.042)
(0.042)
(0.045)
(0.046)
-0.000
-0.000
0.000
0.001
0.002
0.003
(0.002)
(0.002)
(0.002)
(0.002)
(0.002)
(0.002)
16.268***
16.252***
16.239***
16.323***
16.296***
16.327***
(0.144)
(0.149)
(0.141)
(0.185)
(0.192)
(0.164)
0.643
0.614
0.614
0.674
0.635
0.633
10276
10851
11182
5230
5469
5657
Standard errors clustered by state. Member and year xed eects used in all models. The dependent variable
is the log value of non-formula grants received by a given district in a given year. The outlays data span scal years 1984-2010 and are matched with explanatory variables from the previous calendar year.
median
Absolute distance from
is the absolute value of the median-centered rst dimension of the DW-NOMINATE Common Space scores.
The column heads speak to redistricting, where the data on outlays from the rst year after redistricting goes into eect must be excluded from the analysis.
CQ
takes Congressional Quarterly's designation of redistricting, where
redistricting of any districts in a state is counted as redistricting of all districts in that state.
100
uses a requirement
that 100% of a district's population be carried over across the redistricting for the district-year observation not to be dropped from the sample.
80
uses a requirement that 80% of a district's population be carried over across
the redistricting for the district-year observation not to be dropped from the sample. Models 1-3 consider the full remaining sample, while models 4-6 analyze only those districts that appeared in the county-level analysis after counties represented by more than one congressperson were dropped. These last few columns serve to verify that the county-level results were not driven by the exclusion of densely populated areas from the sample.
∗p
≤ 0.10 ∗∗ p ≤ 0.05 ∗∗∗ p ≤ 0.01
47
Table 18: District-Level Analysis - All Grants Full Sample
Absolute distance from median Majority party President's party Committee chair Ranking minority member Party leader First term Tenure (# terms) Close election State presidential margin Constant Adj.
R2
N
Notes:
County-Analysis Sample Only
(1) CQ
(2) 100%
(3) 80%
(4) CQ
(5) 100%
-0.134
-0.092
-0.142
-0.145
-0.053
(6) 80% -0.135
(0.145)
(0.175)
(0.160)
(0.169)
(0.200)
(0.175)
-0.038
-0.024
-0.042
-0.045
-0.017
-0.048
(0.056)
(0.064)
(0.059)
(0.067)
(0.076)
(0.067)
-0.009
-0.003
-0.007
-0.013
-0.005
-0.010
(0.014)
(0.015)
(0.014)
(0.014)
(0.014)
(0.013)
-0.013
-0.021
-0.018
-0.001
-0.009
-0.007
(0.025)
(0.026)
(0.025)
(0.027)
(0.031)
(0.032)
0.029
0.009
0.008
0.014
-0.009
-0.012
(0.037)
(0.034)
(0.032)
(0.029)
(0.029)
(0.029)
0.158**
0.159**
0.159**
0.091
0.083
0.079
(0.071)
(0.073)
(0.073)
(0.066)
(0.062)
(0.058)
-0.029**
-0.020
-0.018
-0.021
0.005
0.005
(0.012)
(0.013)
(0.013)
(0.015)
(0.019)
(0.019)
-0.266***
-0.252***
-0.250***
-0.062
-0.058
-0.058
(0.046)
(0.052)
(0.052)
(0.077)
(0.058)
(0.059)
0.031
0.021
0.017
0.013
0.000
-0.010
(0.025)
(0.030)
(0.030)
(0.032)
(0.041)
(0.042)
-0.002
-0.001
-0.001
0.000
0.001
0.002
(0.001)
(0.001)
(0.001)
(0.001)
(0.001)
(0.001)
17.731***
17.723***
17.749***
17.807***
17.722***
17.768***
(0.101)
(0.119)
(0.112)
(0.097)
(0.101)
(0.089)
0.855
0.794
0.793
0.858
0.798
0.793
10281
10859
11191
5231
5470
5659
Standard errors clustered by state. Member and year xed eects used in all models. The dependent variable
is the log value of
all
grants in the Consolidated Federal Funds Reports' grants category received by a given district
in a given year. The outlays data span scal years 1984-2010 and are matched with explanatory variables from the previous calendar year.
Absolute distance from median
is the absolute value of the median-centered rst dimension
of the DW-NOMINATE Common Space scores. The column heads speak to redistricting, where the data on outlays from the rst year after redistricting goes into eect must be excluded from the analysis.
CQ
takes Congressional
Quarterly's designation of redistricting, where redistricting of any districts in a state is counted as redistricting of all districts in that state.
100
uses a requirement that 100% of a district's population be carried over across the
redistricting for the district-year observation not to be dropped from the sample.
80
uses a requirement that 80%
of a district's population be carried over across the redistricting for the district-year observation not to be dropped from the sample. Models 1-3 consider the full remaining sample, while models 4-6 analyze only those districts that appeared in the county-level analysis after counties represented by more than one congressperson were dropped. These last few columns serve to verify that the county-level results were not driven by the exclusion of densely populated areas from the sample.
∗p
≤ 0.10 ∗∗ p ≤ 0.05 ∗∗∗ p ≤ 0.01
48
Table 19: District-Level Analysis - No Formula Full Sample
Distance (CVP) from median Majority party President's party Committee chair Ranking minority member Party leader First term Tenure (# terms) Close election State presidential margin Constant Adj.
R2
N
Notes:
County-Analysis Sample Only
(1) CQ
(2) 100%
(3) 80%
(4) CQ
(5) 100%
-0.400**
-0.388**
-0.366**
-0.062
-0.025
(6) 80% -0.036
(0.181)
(0.177)
(0.177)
(0.249)
(0.239)
(0.234)
-0.125*
-0.123*
-0.113*
-0.021
-0.023
-0.029
(0.065)
(0.065)
(0.064)
(0.095)
(0.093)
(0.089)
-0.033
-0.027
-0.026
-0.032
-0.024
-0.024
(0.021)
(0.021)
(0.021)
(0.025)
(0.024)
(0.025)
-0.015
-0.018
-0.016
0.011
0.018
0.018
(0.059)
(0.054)
(0.054)
(0.085)
(0.083)
(0.082)
0.040
0.017
0.023
0.013
-0.018
-0.013
(0.058)
(0.055)
(0.054)
(0.053)
(0.057)
(0.056)
0.324
0.343
0.350
0.258
0.257
0.263
(0.216)
(0.224)
(0.220)
(0.176)
(0.168)
(0.167)
-0.043
-0.027
-0.025
0.022
0.047
0.048
(0.033)
(0.032)
(0.031)
(0.033)
(0.033)
(0.032)
-0.240***
-0.206**
-0.205**
0.020
0.072*
0.074**
(0.076)
(0.088)
(0.088)
(0.063)
(0.039)
(0.035)
0.043
0.036
0.032
0.067
0.063
0.053
(0.041)
(0.042)
(0.042)
(0.041)
(0.045)
(0.046)
-0.000
-0.000
0.000
0.002
0.002
0.003
(0.002)
(0.002)
(0.002)
(0.002)
(0.002)
(0.002)
16.381***
16.399***
16.388***
16.420***
16.421***
16.426***
(0.104)
(0.109)
(0.110)
(0.116)
(0.105)
(0.100)
0.644
0.615
0.615
0.674
0.635
0.633
10274
10849
11180
5228
5467
Standard errors clustered by state.
Member and year xed eects used in all models.
5655 The dependent
variable is the log value of non-formula grants received by a given district in a given year. The outlays data span scal years 1984-2010 and are matched with explanatory variables from the previous calendar year.
from median
Distance (CVP)
is the absolute value of legislators' Conservative Vote Probabilities (CVP scores), which are already
median-centered.
The column heads speak to redistricting, where the data on outlays from the rst year after
redistricting goes into eect must be excluded from the analysis.
CQ
takes Congressional Quarterly's designation of
redistricting, where redistricting of any districts in a state is counted as redistricting of all districts in that state.
100
uses a requirement that 100% of a district's population be carried over across the redistricting for the district-year observation not to be dropped from the sample.
80
uses a requirement that 80% of a district's population be carried
over across the redistricting for the district-year observation not to be dropped from the sample. Models 1-3 consider the full remaining sample, while models 4-6 analyze only those districts that appeared in the county-level analysis after counties represented by more than one congressperson were dropped. These last few columns serve to verify that the county-level results were not driven by the exclusion of densely populated areas from the sample.
∗p
≤ 0.10 ∗∗ p ≤ 0.05 ∗∗∗ p ≤ 0.01
49
