Importing Political Polarization? The Electoral ...pubdocs.worldbank.org/en/442901462973382109/DEC-Lecture-Series... · N=3503 district*county cells in column 1, N=2269 cells in districts

Importing Political Polarization?The Electoral Consequences of Rising Trade Exposure

David Autor David Dorn

Gordon Hanson Kaveh Majlesi

May 2016

Trade and Politics

Trade and Politics

The impact of trade on US workers has become a touchstoneissue in the 2016 presidential campaign

• Both among Republicans

“I would tax China on products coming in. I would do a tax, andthe tax, let me tell you what the tax should be... the tax should be45 percent.”

Donald Trump

• and Democrats

“I voted against NAFTA, CAFTA, PNTR with China. I think theyhave been a disaster for the American worker.”

Bernie Sanders

Widely Debated Hypothesis: Do the Economic Impacts ofTrade Favor Ideologically Far-Left and Far-Right Politicians?

Trump and Sanders Have a Point about Trade with China

Background: Rapid Growth of China’s ManufacturingExports since 1990...

02

46

810

Chi

na im

port

pene

tratio

n in

US

man

uf.

0

5

10

15

perc

ent

1991 1996 2001 2006 2011Year

China share of world manufacturing exportsChina import penetration in US manufacturing

...Contributed to Decline in U.S. Manufacturing

Economic Impacts of Import Competition from China

• Closure of manufacturing plants (Bernard Jensen Schott ’06),declines in employment (Acemoglu Autor Dorn Hanson Price ’16;Pierce Schott ’16) in more trade-exposed industries

• Lower lifetime incomes, greater job churning for workers in moretrade-exposed industries (Autor Dorn Hanson Song ’14)

• Lower employment, higher labor-force exit, higher long-rununemployment, greater benefits uptake in more trade-exposed locallabor markets (Autor Dorn Hanson ’13)

Impact on Trade Legislation and US Politics

Anti-trade views precede Trump and Sanders

• Congressional representatives from trade-exposed regions are morelikely to support protectionist trade bills (Feigenbaum Hall ’15; Che,Lu, Pierce, Schott, Tao ’15) and anti-China legislation (KleinbergFordham ’13; Kuk, Seligsohn, Zhang ’15)

• Our work studies whether the impacts of trade exposure extendbeyond voting on trade policy, and affect the ideologicalcomposition of Congress itself

Major Trend in U.S. Politics

Increasing partisanship in the US Congress

• Not due to a shift in vote shares going to the two major parties

• GOP has bicameral majorities, but nat’l vote shares are close to even• Voter identification with parties has become weaker, not stronger,

though persistence in county voting is greater

• Rather, the change is more polarized behavior among legislators• Poole-Rosenthal DW-Nominate scores of roll-call votes

• The ideological divide between the parties has been rising since themid-1970s and is now at an all-time high

• Although voters haven’t become more extreme, legislators have

• Also visible in polarized speech patterns in Congress (GentzkowShapiro Taddy ’15)

Polarization in Congress: DW-Nominate Scores-.4

-.20

.2.4

.6.8

Mea

n D

W-N

omin

ate

Sco

re

1964 1968 1972 1976 1980 1984 1988 1992 1996 2000 2004 2008 2012Year

Democrats Republicans

Mean Voting Behavior by Party in the House

-.4-.2

0.2

.4.6

.8M

ean

DW

-Nom

inat

e S

core

1964 1968 1972 1976 1980 1984 1988 1992 1996 2000 2004 2008 2012Year

Democrats Republicans

Mean Voting Behavior by Party in the Senate

Distribution of Democrats and Republicans on a 10-ItemScale of Political Values (Among Politically Engaged)

Distribution of Democrats and Republicans on a 10-item scale of political values, by level of political engagement

Among the politically engaged

Among the less engaged Source: Pew Research Center (2014).

Figure 8

As a starting point, we can return to the National Election Study. One of the longest-running

questions in the survey asks respondents how they view the Democratic and Republican parties on

a “thermometer” scale, where zero indicates feeling very “cold” toward them and 100 indicates

very “warm.” Partisans always feel more warmly toward their own party than toward the opposite

party. The magnitude of the gap, however, can give us a sense of the depth of the division.

Figure 9 shows trends in this gap over time. The y axis in the figure is the average thermome-

ter rating given to Democrats minus the average thermometer rating given to Republicans. The

blue line shows this difference among self-identified Democrats; these numbers are positive, since

Democrats on average feel warmer toward Democrats than Republicans. The red line shows the

difference among self-identified Republicans; these numbers are negative, since Republicans feel

less warm toward Democrats than Republicans. The dashed line shows the difference between the

two series—the “difference in differences.”

Up until the mid-1990s, these data showed little evidence of a growing partisan gap. Indeed,

these were among the data that Glaeser and Ward (2006, p. A126) used to conclude that growing

political divisions were a “myth.” They noted the slight divergence apparent in 2000 and 2004, but

surmised that this was “more of a George W. Bush effect than any ongoing move towards greater

partisan hostility.”

13

Many Subtleties in Public Opinion (Gentzkow ’16)

1 No growth of extremism on average• Distribution of views on issues mostly single-peaked, relatively stable

2 Rising correlations: Views across issues; between issues and party• Less likely for people to hold liberal views on some issues,

conservative views on others• Presidential votes increasingly predict citizens political views on

taxes, redistribution, social policy, gun control, the environment, etc.

3 Politics has become more personal and hostile• More likely to see other party’s supporters as selfish and stupid• 27% of Dems, 36% of Repubs agree: opposite party’s policies “are

so misguided that they threaten the nation’s well-being” (Pew ’14)• Less tolerant of cross-party marriage!

Explaining Polarization

Literature is large but little consensus on causal mechanisms

• Explanations shown to lack empirical support• Immigration, manipulation of blue-collar voters (Gelman et al. ’08)• Greater voter segregation, heterogeneity in voter attitudes (Glaeser

Ward ’06, Ansolabehere Rodden Snyder ’08, Abrams Fiorina ’12)• Gerrymandering, changes in election structure or congressional rules

(McCarty Poole Rosenthal ’09, Barber McCarty ’15)

• Explanations supported by circumstantial evidence• Tax/regulatory reform (Bartels ’10, Hacker Pierson ’10)• Stronger ideological sorting of voters by party (Levendusky ’09)

• Overall distribution of voter attitudes hasn’t changed but differencein distributions between Dem and GOP party members has

• Media partisanship (DellaVigna Kaplan ’07, Gentzkow Shapiro ’11)

Subject of this Paper: Trade and Political Outcomes

Has rising trade exposure in local labor markets contributed togreater political divisions in Congress?

• Anti-incumbency effect• Incumbents punished for bad outcomes• Fair (’78), Margalit (’11), Jensen Quinn Weymouth (’16)

• Party-realignment effect• Economic shocks change voter prefs — Leftward (Bruner Ross

Washington ’11; Che Lu Pierce Schott Tao ’16) or rightward(Malgouyres ’14, Dippel Gold Heblich ’15)

• Polarization effect• Economic shocks shift support from center to extremes• Failure of monotone likelihood ratio property: Dixit Weibull (’07),

Baliga Hanany Klibanoff (’13), Acemoglu Chernozhukov Yildiz (’15)

Agenda

1 Measuring Electoral Outcomes

2 Exposure to Import Competition from China

3 Empirical Specification

4 Anti-Incumbent, Party Realignment Effects

5 Polarization Effects

6 Heterogeneity in Polarization Effects

7 1990s versus 2000s

8 Conclusions

Challenge: Mapping Political to Economic Geography

Congressional districts can have extreme shapes that do not correspondto any definition of local labor market geography

An Extreme Example: District NC-12

NC-12 stretches over 100 miles and comprises parts of the Charlotte,Greensboro and Winston-Salem cities and Commuting Zones


The district closely follows Interstate 85, and at some points is barelywider than a highway lane


How should one deal with Davidson and Rowan counties, which bothpartly overlap with districts NC-5, NC-12, and NC-8?

Analysis at the County Level would be Problematic

The Davidson and Rowan Cty residents cast votes in three differentraces; the sum of votes across these races is hard to interpret

Observations Observation Weights?

1 Davidson Cty CZ Greensboro Davidson Cty NC-5?/NC-8?/NC-12? Total Votes? Population?2 Rowan Cty CZ Charlotte Rowan Cty NC-5?/NC-8?/NC-12? Total Votes? Population?

Geographic Source of Variables

Data Structure: County-Level Analysis

Local Labor Market Shock

Demographic Composition Election Outcomes

Our Analysis is at the County-District Cell Level

Incorporates the overlapping structure of economic geography(CZ/county) and political geography (district)

Observations Observation Weight

1 Davidson Cty x NC-5 CZ Greensboro Davidson Cty NC-5 Cell Votes/District Votes2 Davidson Cty x NC-8 CZ Greensboro Davidson Cty NC-8 Cell Votes/District Votes3 Davidson Cty x NC-12 CZ Greensboro Davidson Cty NC-12 Cell Votes/District Votes4 Rowan Cty x NC-5 CZ Charlotte Rowan Cty NC-5 Cell Votes/District Votes5 Rowan Cty x NC-8 CZ Charlotte Rowan Cty NC-8 Cell Votes/District Votes6 Rowan Cty x NC-12 CZ Charlotte Rowan Cty NC-12 Cell Votes/District Votes

Data Structure: County-District Cell Analysis

Geographic Source of VariablesLocal Labor

Market ShockDemographic Composition

Election Outcomes

Empirical Strategy

We match local labor markets to congressional districts

• Divide US into county-by-congressional-district cells• Attach each county to its corresponding commuting zone (CZ)• Weight each cell by its share of congressional-district votes• Result is a mapping of CZ shocks to district political outcomes• Use CZ trade shocks from Acemoglu Autor Dorn Hanson Price (’16)

• Examine electoral outcomes over 2002 to 2010• Because of redistricting, we can only examine intercensal periods• Helpfully, these are non-presidential election years• Our time period spans the rise of the Tea Party

Data Sources

1 Voting behavior of congressional representatives

• DW-Nominate scores (Poole & Rosenthal ’85, ’91, ’97, ’01)• Estimated for each legislator in each Congress• Tag 2003-2005 score to winning legislator in 2002 election,

2011-2013 score to winning legislator in 2010 election

2 Vote shares by party in House elections

• Dave Leip’s Atlas of US Presidential Elections• Vote counts for each party by county-district cell

Congressional Districts Included in Sample

No. Districts % of Total(1) (2)

Total Districts in U.S. Congress 435.0 100%

Excluded States 4.0 1%AK 1.0HI 2.0VT 1.0

Inconsistently Observed Cells 14.7 3%TX 9.3GA 5.4

Total Districts in Sample 416.3 96%

The sample excludes Alaska and Hawaii due to complications in definining Commuting Zones in those states, and Vermont, whose only district was represented by a congressman without party affiliation during the sample period. It also excludes county-district cells that are not continuously observed over time due to rezoning in the states of Texas and Georgia. The omitted areas correspond to about 1/3 of the districts in each of these states.

Poole-Rosenthal DW-Nominate Score

Based on spatial model of voting

• PR use roll-call (recorded) votes to estimate model across time

• Each legislator is assumed to have an ideal point in the 2-D plane

• Chooses ’yea’ or ’nay’ on each bill to maximize utility (which is anexponential function of distance plus iid stochastic term)

• Estimated parameters are the 2-D coordinates of each legislator’sideal point and weighting parameters on each dimension

• Since 1980s, nearly all explanatory power is in 1st dimension, whichis interpreted as a liberal-conservative scale (from −1 to 1)

• Our Nominate data comprise 1st dimension of DW-Nominate scorefor each legislator in each congressional term

Sample Nominate Scores in the US Senate

Rankings of 717 US senators who served between 1964 & 2014

• Rand Paul (0.95): 2nd most conservative

• Ted Cruz (0.88): 4th most conservative

• Marco Rubio (0.58): 33rd most conservative

• Barrack Obama (−0.38): 596th most conservative

• Hillary Clinton (−0.40): 605th most conservative

• Bernie Sanders (−0.53): 714th most conservative

Nominate Scores, Win Margins by Party in House

Parties are winning with more extreme candidates and narrower victories

.4.4

5.5

.55

.6.6

5.7

.75

6065

7075

1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012

Election Year

Republican vote share [%, left scale]Nominate Score [right scale]

-.38

-.36

-.34

-.32

6065

7075

1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012

Election Year

Democrat vote share [%, left scale]Nominate Score [right scale, inverted axis]

Big Story is Polarization in Nominate Scores, Not Vote Shares

-.8-.4

0.4

.81.

2

1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012Election Year

Republican mean Democrat mean

Republican min/max Democrat min/max

Republican 5th/95th percentile Democrat 5th/95th percentile

Agenda








8 Conclusions

Mapping Industry Import Shocks to Commuting Zones

• Observed ∆ in industry import penetration from China

∆IP j ,τ =∆Mcu

j ,τ

Yj ,91 + Mj ,91 − Ej ,91

∆Mcujτ is 4 in China imports over ’02-’10 in US industry j ,

Yj ,91 + Mj ,91 − Ej ,91 is industry absorption in ’91 (pre-China shock)

• Exposure of commuting zone i to trade with China

∆IPcuiτ =

∑j

LijtLit

∆IPcujτ

where Lijt/Lit is share of industry j in employment of CZ i in ’00

Change in Import Penetration from China for CongressionalDistricts over 2002-2010

All DistrictsDistrict won by R in 2002

District won by D in 2002

(1) (2) (3)

Mean 0.71 0.72 0.71

25th Percentile 0.40 0.42 0.40

Median 0.57 0.62 0.53

75th Percentile 0.89 0.95 0.89

P75 - P25 0.49 0.53 0.49

N=3503 district*county cells in column 1, N=2269 cells in districts that elected Republicans in the 2002 election in column 2, N=1234 cells in districts that elected Democrats in the 2002 election in column 3. Industry import penetration is the growth of annual imports from China 2002-2010, divided by an industry's U.S. domestic market volume in 1991. The Commuting Zone average of import penertration weights each industry according to its 2000 share in total Commuting Zone employment.

• In AADHP ’16 “Great Sag,” estimate that each 1pt of importpenetration reduces working age adult emp/pop by 1.89pts

• 90/10 district-country ∆ in import exposure is 1.28pts

Isolating the Supply Shock Component of China Imports:Instrumental Variables Approach

Problem• US import demand ∆′s may contaminate estimation

Instrumental variables approach• IV for US imports from China using other DCs (Austria, Denmark,Finland, Germany, Japan, New Zealand, Spain, Switzerland)

• Assumption: Common component of ∆ in rich country importsfrom China is China export supply shock

∆IPcoit =

∑j

Lijt−10

Luit−10∆IPco

jτ

where ∆IPcoit = ∆Mco

jτ / (Yj ,88 + Mj ,88 − Ej ,88) is based on changein imports from China in other high-income countries

Imports from China in the US and Other DevelopedEconomies 1991 – 2007

United States Japan Germany Spain Australia

∆ Chinese Imports (Bil$) 303.8 108.1 64.3 23.2 21.5No. Industries with Import Growth 385 368 371 377 378Correlation w/ U.S.-China Imports 1.00 0.86 0.91 0.68 0.96

Finland DenmarkNew

Zealand Switzerland

∆ Chinese Imports (Bil$) 234.7 5.7 4.7 3.8 3.3No. Industries with Import Growth 383 356 362 379 343Correlation w/ U.S.-China Imports 0.92 0.58 0.62 0.92 0.55

Imports from China in the U.S. and Other Developed Economies 1991 - 2007 (in Billions of 2007$), and their Correlations with U.S.-China Imports

8 Non-US Countries

Correlations of imports across 397 4-digit industries are weighted using 1991 industry employment from the NBER Manufacturing database.

Agenda








8 Conclusions

Empirical Specification

We examine the impact of trade shocks on political outcomes

• Party orientation of a congressional district• Change in party, change in party vote shares

• Ideological positioning of elected representatives• Changes in nominal, absolute Nominate scores of legislators• Changes in likelihood liberal, moderate, or conservative is elected

• Heterogeneity of trade impacts across districts• By initial party in power in 2002• By vote shares in Bush–Gore election in 2000• By racial composition (white majority, non-white majority)

Empirical Specification

Primary specification

∆Yjkt = γd + β1∆IPcujt + X

′jktβ3 + Z

′jtβ2 + ejkt

• ∆Yjkt is ’02-’10 change in electoral outcome for county j , district k

• ∆IP jt is ∆ in import exposure in CZ for county j (IV using ∆IPcoit )

• Xjkt is vector of control variables, γd is census division dummy

• Pol. conditions in ’02 for county-district jk (winning party, winner’svote share, whether winner unopposed, winner’s Nominatescore—interacted w/ GOP dummy)

• Econ. conditions in ’00 for CZ containing county j (manuf. emp.share, routine-task intensity, offshorability index)

• Demogr. composition in ’00 in county j (pop. shares by age,gender, education, race, ethnicity, nativity groups)

• Weight by jk vote share in district k , cluster by CZ and by district

Agenda








8 Conclusions

Effect of Trade Exposure on Pr[Change in Party in Power]Note: Mean of Dependent Variable 12.5%

Trade exposure is weakly related to changes in party in power

(1) (2) (3) (4) (5) (6) (7) (8)

4.64 8.48 8.42 8.73 7.54 ~ 8.40 8.73 7.71(2.89) (5.40) (5.38) (5.38) (4.00) (9.09) (8.37) (8.26)

Estimation: OLS 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS 2SLS

F-statistic first stage 33.47 ** 33.45 ** 34.93 ** 35.04 ** 13.76 ** 13.67 ** 11.87 **

Control Variables:2002 Elected Party yes yes yes yes yes yes2002 Election Controls yes yes yes yes yes2002 Nominate Controls yes yes yes yes2000 Ind/Occ Controls yes yes yes2000 Demography Controls yes yesCensus Division Dummies yes

N=3503 County*District cells. Observations are weighted by a cell's fraction in total votes of its district in 2002, so that each district has an equal weight in the regression, and standard errors are two-way clustered on CZs and Congressional Districts. ~ p ≤ 0.10, * p ≤ 0.05, ** p ≤ 0.01.

Import Exposure and Congressional Election Outcomes 2002-2010.Dependent Variable: 100 x Dummy for Change in Party

Δ CZ Import Penetration

Change in Party, Election 2010 vs 2002

Effect of Trade Exposure on Change in Party Vote Shares

Trade exposure reduces vote shares for party in power butdoesn’t realign vote shares in favor of any one party

(1) (2) (3) (4) (5) (6)

-6.98 * 1.97 0.31 -2.28 -13.49 -13.04 ~(2.75) (2.71) (2.88) (1.79) (12.23) (6.68)

Δ 2002 - 2010 -8.48 1.20 -1.28 0.08 -11.81 -12.03Level in 2010 62.06 50.16 46.69 3.15 14.71 6.15


N=3503 County*District cells. All regression include the full set of control variables from Table 1. Observations are weighted by a cell's fraction in total votes of its district in 2002, so that each district has an equal weight in the regression, and standard errors are two-way clustered on CZs and Congressional Districts. ~ p ≤ 0.10, * p ≤ 0.05, ** p ≤ 0.01.

Change in Voting Outcomes 2002-2010 % Vote for Party that

Won in 2002

Repub-lican Vote

Share

Demo-cratic Vote

ShareOther Vote

Share

Pr(Winner gets >75% of Vote)

Pr(Winner is Unopposed)

A. Point Estimates

B. Summary Statistics

Agenda








8 Conclusions

Effect of Trade Exposure on Change in Nominate ScoresNote: Level in 2002 = 13.9, Level in 2010 = 21.3

Trade exposure induces shift away from center and net shift toright in legislator voting—due to turnover not within-person ∆′s

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

18.41 * 14.15 * 10.69 * 3.46(7.93) (6.09) (5.30) (2.32)

20.13 ** 15.61 ** 12.14 * 3.47(7.86) (5.95) (5.15) (2.33)

N=3503 County*District cells. Panel B replaces the Nominate scores of the 2010 election winners with their Nominate score from the 108th (2003-2005) congress or the first subsequent congress to which they were elected. This eliminates a within-person change in the Nominate score for districts that elected the same representative in 2002 and 2010. Observations are weighted by a cell's fraction in total votes of its district in 2002, so that each district has an equal weight in the regression, and standard errors are two-way clustered on CZs and Congressional Districts. ~ p ≤ 0.10, * p ≤ 0.05, ** p ≤ 0.01.


2002-2010 Change in Political Position

A. Between and Within Person Change of Nominate Score

B. Between Person Change of Nominate Score Only

Decomposition of Change in Absolute Nominate Score


Nominate Score

Absolute Nominate

Score

Shift to

Right

Shift to

Left

Interpreting Magnitudes

Consider two congressional districts that are at the 25th and 75th

percentile of change in trade exposure, respectively

• More trade-exposed district would have:

• change in Nominate score that is 0.18 (18.41× (0.89− 0.40)/49)standard deviations higher

• change in distance from political center that is 0.36(14.15× (0.89− 0.40)/19) standard deviations greater

Decomposing Changes in Nominate Scores and RepublicanPercentage of Two-Party Vote Share

Big changes are between legislator (both D to R and R to R)

(1) (2) (3) (4) (5) (6)

30 22 104 42 95 123

94.75 -72.51 14.89 -2.94 6.00 -1.50

6.83 -3.80 3.73 -0.30 1.37 -0.44

29.61 -18.28 -10.31 10.88 -1.18 6.22

2.14 -0.96 -2.58 1.10 -0.27 1.84

N=3503 County*District cells. Observations are weighted by a cell's fraction in total votes of its district in 2002. Values in Panel C sum to the average change in Nominate score of 7.39 for the whole sample. Values in Panel E sum to the average change in Republican two-party vote percentage of 1.27 for the whole sample.

B. Average Change in 100*Nominate Score by Type of District

A. Number of Districts

I. Party Change II. Representative Change III. No Change

C. Contribution to Overall Change in Average Nominate Score

D. Change in Republican Percentage of Two-Party Vote by Type of District

E. Contribution to Overall Change in Pct Republican Two-Party Vote

Democrat to

Republican

Republican to

Democrat

Republican to

Republican

Democrat to

DemocratRepublican

PersistsDemocrat

Persists

Effect of Trade Exposure Ideological Position of Winners

Trade exposure hurts moderates, helps conservative Republicansand Tea Party

(1) (2) (3) (4) (5) (6)

-37.66 ** 0.27 -23.69 ** -13.97 37.38 ** 24.44 ~(13.95) (7.11) (8.72) (9.58) (14.04) (12.77)

Δ 2002 - 2010

-19.64 2.64 -4.61 -15.03 17.00 11.74Level in 2002 56.78 19.92 27.01 29.77 23.31 6.15Level in 2010 37.13 22.56 22.40 14.74 40.31 17.89


N=3503 County*District cells."Liberal Democrats", "Moderates" and "Conservative Republicans" are defined as politicians whose Nominate scores would respectively put them into the bottom quintile, middle three quintiles, or top quintile of the Nominate score in the 107th (2001-2003) congress that preceeds the outcome period. A Tea Party Member is defined as a representative who was a member of the Tea Party or Liberty Caucus during the 112th (2011-2013) Congress. ~ p ≤ 0.10, * p ≤ 0.05, ** p ≤ 0.01.

A. Point Estimates

B. Means

Change in Probability 2002-2010 that Winner has Given Political Orientation

ModerateLiberal

DemocratModerate Democrat

Moderate Repub-

lican

Conserv-ative

RepublicanTea Party Member

Note on Alternative Specifications

How we define the change in the ideological position of winners

• Previous table examines change in outcome (eg, whether moderateelected in 2010 minus whether moderate elected in 2002)

• Given initial values on are RHS, we could have used the ’10 level onLHS instead of the ’02-’10 change

• Results are robust to:

• Using ’10 levels, rather than ’02-’10 changes, on LHS• Controlling for quadratic in or bin sizes of ’02 Nominate scores• Defining liberals and conservatives cardinally as outside [−0.5, 0.5]

Specification Checks: Using 2002-2010 First-Difference

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

0.54 -24.06 * -14.53 38.05 **(6.78) (11.57) (10.59) (14.61)

0.54 -24.08 * -14.50 38.03 *(6.79) (9.58) (10.07) (15.68)

0.70 -24.28 ** -14.78 38.36 **(7.03) (8.84) (9.55) (14.31)

0.27 -23.69 ** -13.97 37.38 **(7.11) (8.72) (9.58) (14.04)

1.48 -23.54 ** -14.56 36.62 **(6.73) (8.58) (9.69) (13.22)

4.63 -29.50 ** -8.39 33.26 *(6.31) (8.95) (7.59) (13.55)

7.45 -29.47 ** -7.93 29.95 *(5.14) (8.87) (7.49) (11.73)

N=3503 County*District cells. Classifications of candodate ideology is as in Table 4. Observations are weighted by a cell's fraction in total votes of its district in 2002. Standard errors are two-way clustered on CZs and Congressional Districts. ~ p ≤ 0.10, * p ≤ 0.05, ** p ≤ 0.01. ~ p ≤ 0.10, * p ≤ 0.05, ** p ≤ 0.01.

G. Linear x Party + 4 Categories

A. No Nominate 2002 Control

B. Linear Nominate

C. Quadratic Nominate

D. Linear Nominate x Party (Primary Spec)

E. Quadratic Nominate x Party

F. 4 Nominate Categories

Liberal Democrat

Moderate Democrat

Moderate Republican

Conservative Republican

Specification Checks: Using 2010 Outcome (Level)

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

9.03 -32.55 ** -5.81 29.33 *(9.61) (9.44) (7.55) (14.06)

9.01 -32.54 ** -5.82 29.35 *(7.90) (9.63) (7.50) (11.59)

8.60 -32.19 ** -5.68 29.26 *(5.81) (9.95) (7.32) (11.87)

9.78 ~ -33.20 ** -6.12 29.53 *(5.63) (10.17) (7.35) (11.94)

9.36 ~ -31.43 ** -6.69 28.75 **(5.66) (9.04) (7.46) (10.83)

4.63 -29.50 ** -8.39 33.26 *(6.31) (8.95) (7.59) (13.55)

7.45 -29.47 ** -7.93 29.95 *(5.14) (8.87) (7.49) (11.73)

G. Linear x Party + 4 Categories

N=3503 County*District cells. Classifications of candodate ideology is as in Table 4. Observations are weighted by a cell's fraction in total votes of its district in 2002. Standard errors are two-way clustered on CZs and Congressional Districts. ~ p ≤ 0.10, * p ≤ 0.05, ** p ≤ 0.01. ~ p ≤ 0.10, * p ≤ 0.05, ** p ≤ 0.01.

A. No Nominate 2002 Controls

B. Linear Nominate

C. Quadratic Nominate

D. Linear Nominate x Party (Primary Spec)

E. Quadratic Nominate x Party

F. 4 Nominate Categories

Liberal Democrat

Moderate Democrat

Moderate Republican

Conservative Republican

Agenda








8 Conclusions

Heterogeneity in Effects

Trade exposure raises likelihood of within-party transitions inlegislators for the GOP

7.71 15.94 -23.65 *(8.26) (11.45) (10.63)

29.88 ~ -5.21 -24.67(17.82) (18.05) (18.37)

-13.95 ~ 41.12 * -27.17 ~(7.69) (16.31) (13.91)

C. Initially Republican DistrictΔ CZ Import Penetration

B. Initially Democratic DistrictΔ CZ Import Penetration

N=3,503 County*District cells in Panel A, N=1,234 in Panel B, N=2,269 in Panel C. All regression include the full set of control variables from Table 1. Observations are weighted by a cell's fraction in total votes of its district in 2002, so that each district has an equal weight in the regression, and standard errors are two-way clustered on CZs and Congressional Districts. ~ p ≤ 0.10, * p ≤ 0.05, ** p ≤ 0.01.

A. All DistrictsΔ CZ Import Penetration

(1)Different Rep

(2)Same Rep

(3)

No Change in PartyChange in Party

Heterogeneity in Effects: Initial Party in Power

Losses of centrists compensated by gains on the left and right(initially Dem districts), or right only (initially GOP)

Nominate Score Moderate

Liberal Dem

Moderate Dem

Moderate Repub

Conserv-ative

RepubTea Party Member

(1) (2) (3) (4) (5) (6) (7)

17.13 -45.67 * 15.61 -45.48 * -0.19 30.07 31.46(15.06) (21.04) (18.85) (18.96) (6.61) (19.14) (23.35)

Mean in 2002 57.56 42.44 57.56 0.00 0.00 0.00Δ 2002 - 2010 12.99 -17.42 5.63 -21.00 3.58 11.79 5.39

12.19 ~ -34.64 * 0.00 -13.95 ~ -20.69 34.64 * 16.53(7.11) (17.54) . (7.69) (14.62) (17.54) (15.67)

Mean in 2002 56.09 0.00 0.00 56.09 43.91 11.58Δ 2002 - 2010 1.71 -21.61 0.00 9.88 -31.49 21.61 17.36

N=1,234 County*District cells in Panel A, 2,269 County*District cells in Panel B. All regression include the full set of control variables from Table 1. Observations are weighted by a cell's fraction in total votes of its district in 2002, so that each district has an equal weight in the regression, and standard errors are two-way clustered on CZs and Congressional Districts. ~ p ≤ 0.10, * p ≤ 0.05, ** p ≤ 0.01.


A. Initially Democratic District


B. Initially Republican District


Heterogeneity in Effects: Racial Composition

Trade exposure helps conservative GOPers in white-majoritydistricts, liberal Dems in non-white-majority districts

(1) (2) (3) (4) (5) (6)

20.98 * -0.01 -27.22 ** -15.64 42.87 ** 25.28(8.69) (7.94) (9.88) (11.61) (16.17) (15.47)

Mean in 2002 16.09 24.74 33.60 25.57 6.32Δ 2002 - 2010 8.44 2.15 -4.14 -17.57 19.57 13.31

-7.03 26.88 * -25.33 * 11.36 -12.91 1.58(8.60) (12.87) (12.02) (7.17) (10.20) (8.10)

Mean in 2002 40.02 38.97 9.59 11.42 5.25Δ 2002 - 2010 1.89 5.24 -7.08 -1.67 3.51 3.52

A. Counties >1/2 of Voting Age Pop is Non-Hispanic WhiteΔ CZ Import Penetration

B. Counties ≤ of Voting Age Pop is Non-Hispanic White

N=3241 County*District cells covering 349.8 weighted districts in Panel A, N=262 County*District cells covering 66.5 districts in Panel B. All regression include the full set of control variables from Table 1. Observations are weighted by a cell's fraction in total votes of its district in 2002, so that each district has an equal weight in the regression, and standard errors are two-way clustered on CZs and Congressional Districts. ~ p ≤ 0.10, * p ≤ 0.05, ** p ≤ 0.01.



Nominate Score

Liberal Dem

Moderate Dem

Moderate Repub

Conserv-ative


Agenda








8 Conclusions

Polarization of House Ongoing Since Late 1970s

Do We Find Same Trade Exposure Effects in the 1990s?

Comparable effects on vote shares, turnover to 2000s. But noeffect on Nominate scores

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

-10.33 * -3.79 5.24 -14.81 ~ -4.62(4.68) (4.09) (4.38) (8.30) (5.18)

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

7.52 16.18 -23.70 ~ -0.12 0.47(11.12) (12.70) (14.06) (9.51) (6.90)

A. Change in Voting Outcomes% Vote for Party that

Won in 2002Republican Vote Share

Democrat Vote Share

Pr(Winner gets >75% of Vote)

Pr(Winner Unopposed

)

N=3523 County*District cells covering 421.7 districts (all except districts omitted states AK and HI, two districts in VT and VA that elected independents, and 8.3 districts with cells that are not continuously observable to rezoning, primarily located in LA, GA, NC, VA).

C. Change in Nominate Index



B. Change of Party and Incumbent

Change in Party

Same Party, Different

RepSame Party, Same Rep

Nominate Score

Absolute Nominate

Score

Do We Find Same Trade Exposure Effects in the 1990s?

No significant effect on ideological orientation of those elected

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

-8.23 0.57 -1.78 -6.45 7.66(12.80) (6.24) (11.20) (7.71) (12.36)


N=3523 County*District cells covering 421.7 districts (all except districts omitted states AK and HI, two districts in VT and VA that elected independents, and 8.3 districts with cells that are not continuously observable to rezoning, primarily located in LA, GA, NC, VA). Observations are weighted by a cell's fraction in total votes of its district in 1992, so that each district has an equal weight in the regression, and standard errors are two-way clustered on CZs and Congressional Districts. ~ p ≤ 0.10, * p ≤ 0.05, ** p ≤ 0.01.

Change in Probability 1992 - 2000 that Winner has Given Political Orientation

ModerateLiberal

DemocratModerate Democrat

Moderate Republican

Conserv-ative

Republican

Agenda








8 Conclusions

Discussion

Rising political polarization is striking but not well understood• Coincidence with widening income inequality leads naturally toconjecture that economic shocks are behind greater partisanship

2016 U.S. prez primary doesn’t appear anomalous in retrospect• US manufacturing decline—which voters see as due in part toglobalization—likely to have political consequences

• Was affecting 2002-2010 House elections (we now know)

Why would trade shocks contribute to political polarization?• Divergent responses to common shocks based on differences in priorbeliefs (Dixit Weibull ’07)

• Voter attitudes are stable in aggregate, but Dem and GOP beliefsabout policy, ideology, partisanship are sharply diverging

Counterfactual Calcs: Dialing Back the Trade Shock by 50%

(1) (2) (3) (4) (5) (6) (7)

-8 8 11 -19 -63 71 49

4 -4 7 -3 -56 52 36[-3,11] [-11,3] [2,12] [-10,4] [-63,-50] [42,63] [25,46]

-12 12 4 -16 -6 18 13[ -5, -19 ] [ 19, 5 ] [ 9, -1 ] [ -9, -23 ] [ 0, -13 ] [ 29, 8 ] [ 24, 3 ]

Total Repub

I. Actual Change 2002-2010

II. Counterfactual Change 2002-2010: 50% Lower Import Growth

The counterfactuals are constructed using the coefficient estimates in Panel B1 and C1 of Table 6, and columns 3-7 of Table 7. They subtract from the actual seat changes the product of 50% x the mean growth of import penetration (Appendix Table 2) x the r2 of the 2SLS first stage (0.38) x the start-of-period number of seats in the indicated category x the regression coefficient divided by 100. The numbers in brackets in Panel II indicate a confidence interval for the counterfactul, which is obtained by computing the counterfactul using the point estimates from Tables 6 and 7 plus/minus one standard error.

Change in Number of House Seats by Category in 416-District Sample

Total DemLiberal Dem

Moderate Dem

Moderate Repub

Conserv-ative


III. Actual minus Counterfactual (Net Effect of 50% Shock)

New York Times Graphic (4/26/16)

Documents

Importing Political Polarization? The Electoral ...pubdocs.worldbank.org/en/442901462973382109/DEC-Lecture-Series... · N=3503 district*county cells in column 1, N=2269 cells in districts