Ted Kennedy, Orin Hatch, and Other Strange Bedfellows? A Network Explanation of Legislative Voting Jennifer N. Victor University of Pittsburgh [email protected] Gregory Koger University of Miami [email protected]
Dec 26, 2015
Ted Kennedy, Orin Hatch, and Other Strange Bedfellows?A Network Explanation of Legislative Voting
Jennifer N. VictorUniversity of [email protected]
Gregory KogerUniversity of [email protected]
Do Lobbyists “influence” legislators’ votes? The media say “yes:”
Sources of Campaign Finance in 2006
42%
10%
44%
3%
House Democrats
43%
9%
43%
2%4%
House RepublicansPACs
Small Individual Contributions
Large Individual Contributions
Self-Financing
Other
14%
20%
55%
3%8%
Senate Democrats
24%
11%56%
1%8%
Senate Republicans
Source: Center for Responsive Politics http://www.opensecrets.org/bigpicture/wherefrom.php?cycle=2006
Influences on Voting
Constituents/public opinion (Achen 1978; Hill and Hurley 1999; Miller and Stokes
1963)Representation of subgroups (Arnold 1990; Bartels 2008; Bishin 2000, 2009; Fenno
1978)Parties/Party loyalty (Cox and Poole 2002; Lebo, McGlynn and Koger 2007;
Lee 2008; Sinclair 2002)Organized Interests(Mansbridge 2003; Ansolahebere, de Figueiredo, and
Snyder 2003;
(A related question) Why Donate?Exchange/Access Theory
Campaign donations in exchange for votes or accessBut: Reneging? Small donations? Non-PAC organizations?
Information TheoryInformation persuades legislatorsBut: Why lobby allies?
Subsidy TheoryLobbyists subsidize legislatorsBut: Other resources? Why pay to play?
Lobbying, Networks, and Contributions
Legislators’ relationships with the lobbying community influence their voting behavior.
Emphasize the system of connections between legislators and lobbyist-donors, rather than the “transaction.”
Existing evidence that legislators and lobbyists desire long term relationships (Snyder 1990; Berry and Wilcox 2009).
Donations are observable evidence of relationships and common interests.
Expectations
Ceteris paribus, we expect legislators who are more connected through the lobbying-donation network (directly or indirectly) to be more likely to vote the same way.
Research DesignFederal donations by lobbyists in the 2006 election cycle (109th Congress)
Obtained from the Center for Responsive Politics20,639 donations by 1,225 lobbyistsRecipients
Candidates for Congress National Party PACsPACs, including Leadership PACs
9,751 dyadic observations of lobbyist donations to MCs.
The Lobbyist-Legislator Network
2-mode network
1-mode network
2
1
C
B
A
LegislatorsLobbyists
ORA B C1 2
Legislators
1 2B
Lobbyists
The Two-Mode Lobbyist-Legislator Network , 2006
Descriptive Statistics: Number of lobbyist-donors
Mean Median SD Min MaxPer MC 18 10 23.8 0 220Per Dyad 0.68 0 2 0 76Per House Dyad
0.42 0 1.1 0 32
Per Senate Dyad
3.4 1 6.5 0 76
Incumbent dyads with the most lobbyist-donors
Member 1 Member 2Number of Common
Lobbyist Donors
Cantwell Clinton 76
Santorum Allen 67
Conrad Cantwell 61
Carper Cantwell 60
Cantwell Nelson, Ben
59
Menendez Clinton 56
Nelson, Bill Clinton 56
Nelson, Bill Cantwell 55
Conrad Clinton 52
Kennedy Cantwell 51
Point Connectivity
We aren’t just interested in the number of common donors legislators share, but how legislators are connected through the network.Ties come in different forms:
Lobbyists [A,B] indirectly connect legislators [1,3]
B
A
3
2
1
LegislatorsLobbyists
Point Connectivity
We aren’t just interested in the number of common donors legislators share, but how legislators are connected through the network.Ties come in different forms:
Lobbyists Reinforce Cleavages
CB
321Legislators
LobbyistsA D
4
Point Connectivity
We aren’t just interested in the number of common donors legislators share, but how legislators are connected through the network.Ties come in different forms:
Lobbyist Ties Link Legislators
CB
321Legislators
LobbyistsA D
4
Distribution of Point Connectivity
Distribution of Point Connectivity
Top Incumbent Recipients (by chamber), by Point Connectivity
Senate House
Member 1 Member 2(ln) Point
Connectivity
Number of Common Lobbyist-donors
Member 1 Member 2(ln) Point
Connectivity
Number of Common Lobbyist-donors
Cantwell (D-WA)
Nelson (D-NE)
5.945 59DeLay (R-TX 22)
Bonilla (R-TX 23)
5.724 22
Santorum (R-PA)
Nelson (D-NE)
5.894 17Lewis (R-CA 41)
Bonilla (R-TX 23)
5.724 10
DeWine (R-OH)
Santorum (R-PA)
5.889 43Lewis (R-CA 41)
DeLay (R-TX 22)
5.717 9
DeWine (R-OH)
Cantwell (D-WA)
5.878 15Lewis (R-CA 41)
Menendez (D-NJ 13)
5.697 4
DeWine (R-OH)
Nelson (D-NE)
5.875 11DeLay (R-TX 22)
Menendez (D-NJ 13)
5.694 0
Conrad (D-ND)
Cantwell (D-WA)
5.866 61Menendez (D-NJ 10)
Bonilla (R-TX 23)
5.694 0
Conrad (D-ND)
Nelson (D-NE)
5.864 46Lewis (R-CA 41)
Pombo (R-CA 11)
5.68 7
Santorum (R-PA)
Cantwell (D-WA)
5.861 10DeLay (R-TX 22)
Pombo (R-CA 11)
5.677 16
Santorum (R-PA)
Allen (R-VA)5.855 67
Pombo (R-CA 11)
Bonilla (R-TX 23)
5.677 8
DeWine (R-OH)
Allen (R-VA)5.852 35
Pombo (R-CA 11)
Menendez (D-NJ 13)
5.673 1
Measures—Dependent Variables
Voting AgreementThe probability legislator a voted the same as legislator b, given that they both voted.House: mean = 0.69, range: 0.1-1Senate: mean= 0.65, range: 0.26-0.98
Regression, Inference, and Network Data
Analysis of Social Network data requires particular attention to:
Sampling Autocorrelation
We want to model the relationships between observations.Use a mixed model: (legislators nested in dyads).
Dyads (level 1, i); Legislators (level 2, j).Include a legislator-specific random intercept, ζ1j, to capture unobserved heterogeneity between observations.We assume the random intercept and residual are normally distributed ζj ~N(0, ψ); εij ~N(0,θ)
ExpectationsLegislators who are more connected through the lobbyist-donors network are more likely to vote together.
CONTROLS: Service on the same committeesConstituent PreferencesParty membership (same party)Being from the same stateBeing electorally vulnerableBeing a party/committee leaderTerms servedDemographics
Coeff. Z Pr>|z| Coeff. Z Pr>|z| Coeff. Z Pr>|z|0.0026 0.0050(0.0001) (0.0003)
0.0014 0.0016 0.0013(0.0004) (0.0004) (0.0015)
-0.0033 -0.0032 0.0034(0.00002) (0.00002) (0.0001)
0.3535 0.3539 0.3532(0.0006) (0.0005) (0.0019)
0.3507 0.3540 0.3278(0.0005) (0.0005) (0.0019)
0.0111 0.0130 -0.0001(0.001) (0.001) (0.0035)
0.0005 -0.0005(0.0026) (0.0025)
0.0257 0.0274 0.0178(0.0006) (0.0006) (0.0019)
-0.0045 -0.0034 0.0102(0.0005) (0.0005) (0.0016)
-0.0009 -0.001 0.0019(0.0005) (0.0005) (0.0018)
-0.0034 -0.0025 0.0125(0.0006) (0.0006) (0.002)
0.0001 0.0000 -0.0000(0.00001) (0.00003) (0.0001)
0.5538 0.5410 0.5773(0.0008) (0.0014) (0.0026)
N 191,386 163,612 27,774Number of Groups 95,693 81,806 13,887Log Likelihood 702371.19 604889.38 82329.5
671.66 0.000
Woman (at least one) -7.40
-4.41
0.000
0.000 0.000
0.000 0.000
0.058 -1.96 0.050
0.20 0.843 -0.22
-161.13 0.000
Both Democrats 637.67 0.000 631.03 0.000
0.000Constituency Preferences (cd presidential vote difference)
-166.04
Both Republicans 670.84
Total Donors 4.53
Senior (at least one) -6.06
Black (at least one)
-9.98
Party Leader (at least 1) -1.90
Both in Competitive CDs
Dyad in Same State
0.826
41.85 0.000 42.69 0.000
Committee Service Coincidence 3.28 0.001 3.49 0.000
MODEL I - Full House MODEL II - Connected Dyads
(ln)Connectivity (via common lobbyist-donors)
19.41 0.000 15.51 0.000
9.16 0.000
-52.14 0.000
182.56 0.000
177.11 0.000
-6.24 0.000
1.08 0.281
-6.38 0.000
TABLE 5 HOUSE VOTING AGREEMENT: Random Intercept Mixed Regression Model, 2006 election cycle
-
- - -
-0.04 0.968
MODEL III - Unconnected Dyads
- -
0.85 0.397
11.19 0.000 13.03 0.000
Numbers in parenthese are robust s tandard errors
-0.46 0.000
224.50 0.000
0.000 2.14 0.032
Constant 670.54 0.000 380.14 0.000
Coeff. Z Pr>|z| Coeff. Z Pr>|z| Coeff. Z Pr>|z|-0.001 0.0040(0.0007) (0.0017)
0.0040 0.0051 0.0007(0.0017) (0.0018) (0.0038)
-0.005 -0.0046 -0.0064(0.0002) (0.0002) (0.0005)
0.3805 0.3807 0.3767(0.0035) (0.004) (0.0077)
0.3752 0.3732 0.3778(0.0031) (0.0033) (0.0077)
0.0341 0.0202 0.0718(0.0128) (0.0141) (0.0286)
0.0204 0.0184(0.0056) (0.0054)
0.0069 0.0040 0.0367(0.0093) (0.0096) (0.0286)
0.0039 0.0054 -0.002(0.003) (0.0032) (0.007)
-0.0047 -0.0000 -0.014(0.0035) (0.0041) (0.0069)
0.0191 0.0078 0.0575(0.0042) (0.0047) (0.0096)
0.0001 -0.0000 0.0001(0.00002) (0.0000) (0.00007)
0.4947 0.4768 0.4843(0.005) (0.0086) (0.011)
N 10,098 7,656 2,442Number of Groups 5,049 3,828 1,221Log Likelihood 25796.165 19428.573 5188.0659
Constituency Preferences (cd presidential vote difference)
-26.56 0.000 -22.57 0.000
Committee Service Coincidence 2.40 0.016 2.81 0.005
Both Republicans 122.88 0.000 114.82 0.000
Both Democrats 107.54 0.000 96.12 0.000
Both in Cycle 3.67 0.000 3.38 0.001
Dyad in Same State 2.66 0.008 1.43 0.153
0.093
Black (at least one) 0.74 0.457 0.42 0.676
Woman (at least one) 1.33 0.183 1.72 0.085
0.18 0.854
Total Donors 1.84 0.066 -0.04 0.972
Party Leader (at least 1) -1.34 0.181 -0.01 0.990
Senior (at least one) 4.53 0.000 1.68
TABLE 6 SENATE VOTING AGREEMENT: Random Intercept Mixed Regression Model, 2006 election cycleMODEL III - Unconnected Dyads
- - -
MODEL I - Full Senate MODEL II - Connected Dyads
(ln)Connectivity (via common lobbyist-donors)
-1.47 0.143 2.40 0.016
1.28 0.199
-14.14 0.000
49.12 0.000
48.89 0.000
2.51 0.012
- - -
-0.25 0.803
-2.06 0.040
5.98 0.000
1.90 0.058
43.98 0.000
Numbers in parenthese are robust s tandard errors
Constant 99.36 0.000 55.76 0.000
HOUSE ResultsHOUSE VOTING AGREEMENT: Random Intercept Mixed Regression Model, 2006 election cycle MODEL I - Full House
Coeff. Z Pr>|z|(ln)Connectivity (via common lobbyist-donors)
0.0026 19.41 0.000(0.0001)
Committee Service Coincidence 0.0014 3.28 0.001(0.0004)
Constituency Preferences (cd presidential vote difference) -0.0033 -166.04 0.000
(0.00002)
Both Democrats 0.3535 637.67 0.000(0.0006)
Both Republicans 0.3507 670.84 0.000(0.0005)
Dyad in Same State 0.0111 11.19 0.000
N 191,386 Number of Groups 95,693 Log Likelihood 702371.2
SENATE ResultsTABLE 6 SENATE VOTING AGREEMENT: Random Intercept Mixed Regression Model, 2006 election cycle
MODEL I - Full Senate MODEL II - Connected Dyads
MODEL III - Unconnected Dyads
Coeff. Z Pr>|z| Coeff. Z Pr>|z| Coeff. Z Pr>|z|
(ln)Connectivity (via common lobbyist-donors)
-0.001-1.47 0.143
0.00402.40 0.016 - - -(0.0007) (0.0017)
Committee Service Coincidence
0.0040 2.40 0.016 0.0051 2.81 0.005 0.0007 0.18 0.854(0.0017) (0.0018) (0.0038)
Constituency Preferences (cd presidential vote difference)
-0.005-26.56 0.000
-0.0046-
22.57 0.000-0.0064
-14.14 0.000(0.0002) (0.0002) (0.0005)
Both Democrats 0.3805 107.54 0.000 0.3807 96.12 0.000 0.3767 49.12 0.000(0.0035) (0.004) (0.0077)
Both Republicans 0.3752 122.88 0.000 0.3732 114.82 0.000 0.3778 48.89 0.000(0.0031) (0.0033) (0.0077)
N 10,098 7,656 2,442 Number of Groups 5,049 3,828 1,221
Interpretation of Results
0.6
0.62
0.64
0.66
0.68
0.7
0.72
min 5% 25% 50% 75% 95% maxPred
icte
d Pr
obab
ility
of V
oting
A
gree
men
t
Connectivity via Common Lobbyists' Donations
Predicted Probability of Co-voting Among MCs Connected by Lobbyists' Donations
House Voting Coincidence Senate Voting Coincidence
House Mean Co-voting
Senate Mean Co-voting
Visualization of Results
Random Senators (N=38)Actual Data: Most Central Senators in Lobby-Donor Network (N=38)
Senate 38 random senators, opacity of tie indicates voting agreement, color party, squares are in-cycle, circles are not.
Senate 38 most central actors (those with greater than mean degree centrality), opacity of tie indicates voting agreement, color indicates leadership, squares are in-cycle, circles are not. Compared to random data: more GREEN, more dark ties, more SQUARES, and LARGER nodes.
Size of node = $ contributionsColor of node = Non-leader
= LeaderShape of node = in cycle = not up
Strange Bedfellows, HouseRepresentative Representative
Vote Agreement(μ = 0.69)
Ideological Difference(μ = 15.0)
Point Connectivity
(μ = 96.4)Edolphus Towns
(D-NY)Roy Blunt (R-MO) 0.45 54 246
Edolphus Towns (D-NY)
Chet Edwards (D-TX)
0.811 56 242
Edolphus Towns (D-NY)
John Carter (R-TX)
0.447 53 235
Edolphus Towns (D-NY)
Joe Barton (R-TX) 0.441 52 234
Charles Rangel (D-NY)
Chet Edwards (D-TX)
0.80 60 225
Charles Rangel (D-NY)
Tom DeLay (R-TX)
0.432 54 225
Charles Rangel (D-NY)
John Sullivan (R-OK)
0.412 55 225
Charles Rangel (D-NY)
John Carter (R-TX)
0.444 57 225
Charles Rangel (D-NY)
Eric Cantor (R-VA)
0.43 52 225
Charles Rangel (D-NY)
John Boehner (R-OH)
0.452 55 225
Strange Bedfellows, SenateSenator Senator
Vote Agreement(μ = 0.65)
Ideological Difference(μ = 9.7)
Point Connectivity
(μ = 95.7)Edward Kennedy
(D-MA)Orin Hatch (R-UT) 0.359 35.94 319
Edward Kennedy (D-MA)
Ben Nelson (D-NE)
0.583 29.26 318
Orin Hatch (R-UT)Hillary Clinton (D-
NY)0.408 32.37 304
Orin Hatch (R-UT)Richard Durbin
(D-IL)0.38 33.42 245
Orin Hatch (R-UT)Lincoln Chafee
(R-RI)0.626 33.42 245
Thomas (R-WY)Lincoln Chafee
(R-RI)0.608 30.35 241
Craig Thomas (R-WY)
Hillary Clinton (D-NY)
0.374 29.3 240
Edward Kennedy (D-MA)
Craig Thomas (R-WY)
0.324 32.87 240
Patrick Leahy (D-VT)
Craig Thomas (R-WY)
0.35 29.87 215
Patrick Leahy (D-VT)
Orin Hatch (R-UT) 0.39 32.94 215
ConclusionsOur innovations on the question of how/whether lobbyists influence legislators:
Look at lobbyists’ personal donations, not PACs
Use network analysis.
We find that, ceteris paribus, the stronger the connection between legislators in the lobbying network, the more likely the are to vote together.
Effect is stronger in the House than the Senate
Conclusions
At the very least, lobbyists’ donation are indicative of legislators latent policy preferences.Our data are also consistent with the relatively unsupported claim that lobbyists buy votes.
Future Work
RepresentationWhich has more explanatory power: donations or constituents?
PowerWho is most central in the legislator network?
TiesCan we predict who will donate/receive?
If lobbyists primarily seek relationships, there will be evidence of ties over time.
Why Donate?
Prof. Jennifer N. Victor
EXTRA SLIDES
Measures—Dependent Variables
Voting Agreement--House0
12
34
5D
ensi
ty
0 .2 .4 .6 .8 1House Coincidence of Voting
Measures—Dependent Variables
Voting Agreement--Senate0
12
34
5D
ensi
ty
.2 .4 .6 .8 1Senate Coincidence of Voting
The Network Approach
Why networks, and why now?
Not inconsistent with methodological individualism.
Network analysis considers the unit of analysis to be a relationship rather than the individual.
Politics is naturally about relationships.
Technology now makes it possible.
The Network Approach
Network tools are particularly useful when we want to understand:
Flow of information i.e., voter contagion: Nickerson APSR 2008
Coordination and cooperationi.e., collective action problems: Siegel AJPS 2009
Informal institutionsi.e., Caucuses: Victor & Ringe 2009
Multiple levels of organizations i.e., international capitalism: Lazer 2005
The Network Approach
Senate Co-sponsorship (Fowler 2006)
The Network Approach
2004 A-list Bloggers (Adamic and Glance 2005)
The Network Approach: An Increasing Trend
1971-
1997
1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 200902468
101214161820
Number of Papers Published in Major Political Science Journals with the word "Network" in the Title
(APSR, AJPS, JOP, IO, LSQ, BJPS, APR, PSQ, PA, PS, PC)(Data Complied by author)
Num
ber o
f Pub
licati
ons
Anecdotal Support for Network Perspective
Quotes from lobbyists:
‘I don't usually give out my personal money unless I know the person and I feel like I've got some kind of respect and relationship with that person’
- Republican lobbyist Richard F. Hohlt as quoted in Carney 2007.
Anecdotal Support for Network Perspective
Quotes from lobbyists:
‘I do not give for the purpose of having access. Virtually everyone I deal with in representation of a client I know personally and I have known personally for 10, 15, 20 years. So, when I enter, I enter on the basis of my credibility and the issues at hand, and not based upon the fact that I have contributed to an individual and am seeking access to that individual.’
-Former Rep. Tom Loeffer (R-TX) quoted in Carney 2007.
Anecdotal Support for Network Perspective
Quotes from lobbyists:
Tony Podesta says that personal relations, not a desire for access, drive his donations. ‘In every case, they are people I know, people who are friends, people I have a relationship with,’ he says. ‘It’s not a door-opener kind of thing. It’s rather an effort to keep in office or send to office people who are doing a good job.’
- Tony Podesta, Democratic lobbyist as quote in Carney 2007.
Coeff. Z Pr>|z| Coeff. Z Pr>|z|-0.490 0.244(0.229) (0.313)
0.500 1.328(0.107) (0.145)
-0.322 -0.285(0.233) (0.339)
-0.029 -0.148(0.010) (0.227)
0.724 0.388(0.149) (0.158)
0.008 -0.064(0.012) (0.042)
2.523 3.065(0.103) (0.156)
-0.394296 -0.394629(0.089) (0.194)
0.674154 0.673930(0.060) (0.131)
N 437 101Log Restricted-Likelihood -1522 -454robust standard errors in parentheses, clustered on state
Negative Binomial Predicting Number of Lobbyist-Donors, 2006 cycleHouse Senate
Number of Lobbyist-Donors Number of Lobbyist-Donors
Competitive District/In Cycle
4.47 0.000 9.19 0.000
Distrance from Median -2.14 0.032 0.78 0.435
Woman -0.29 0.771 -0.65 0.514
African-American/Minority -1.38 0.167 -0.84 0.401
0.000
Party/Committee Leader 4.87 0.000 0.014
Terms Served 0.65 0.516 -1.51 0.132
ln(alpha)
alpha
2.45
19.67Constant 24.43 0.000
Measures—Independent Variables
Common Lobbyist-DonorsCommittee Coincidence
House: mean = 0.2, range: 0-3Senate: mean = 0.73. range: 0-4
Ideological DistanceHouse: mean = 0.5, range: 0 – 1.9Senate: mean = 0.5, range: 0 – 1.9
Same State: 0 (139,457) or 1 (4,996)Electoral Vulnerability
House (Cook Competitive District):
Measures—Independent VariablesElectoral Vulnerability, at least 1
House (Cook Competitive District): 0 (81,406); 1 (14,297)Senate (in cycle 2006): 0 (2,628); 1 (2,422)
Leadership (party, committee, cardinal) , at least 1
House: 0 (69,378); 1(26,325)Senate: 0 (1,275); 1 (3,775)
Senior, at least 1 greater than mean terms served
House: 0 (16,117); 1 (79,586)Senate: 0 (828); 1 (4,222)
Measures—Independent VariablesAfrican-American, at least 1
House: 0 (79,401); 1 (16,302)Racial Minority, at least 1
Senate: 0 (4,465); 1 (585)Woman, at least 1
House: 0 (69,378); 1 (26,325)Senate: 0 (3,741); 1 (1,309)