Why does import competition favor republicans? Localized trade shocks and cultural backlash in the US

Figures & data

Figure 1. Alternative causal pathways linking trade shocks and candidate selection.

Figure 2. Congressional district exposure to Chinese import competition (1991–2016).

Table 1. Import exposure and sentiments towards ethnic and racial minorities.

	(1)	(2)	(3)	(4)
	OLS	TSLS	TSLS	TSLS
Panel A: Dep. Var.: Feeling Thermometer – Hispanics
$\Delta IP$	–1.32***	–1.57***	–1.14***	–0.73**
	(0.33)	(0.36)	(0.32)	(0.36)
Observations	10,922	10,922	10,633	10,633
R^2	0.013	0.013	0.092	0.097
Panel B: Dep. Var.: Feeling Thermometer – African Americans
$\Delta IP$	–0.20	–0.21	–0.07	0.02
	(0.26)	(0.29)	(0.25)	(0.31)
Observations	10,926	10,926	10,637	10,637
R^2	0.009	0.009	0.109	0.110
Panel C: Dep. Var.: Feeling Thermometer – Asians
$\Delta IP$	–0.67**	–0.82**	–0.74***	–0.29
	(0.29)	(0.32)	(0.29)	(0.34)
Observations	10,880	10,880	10,595	10,595
R^2	0.009	0.009	0.039	0.043
Panel D: Dep. Var.: Feeling Thermometer – Whites
$\Delta IP$	0.47**	0.57**	0.57**	0.85***
	(0.23)	(0.25)	(0.24)	(0.30)
Observations	10,928	10,928	10,639	10,639
R^2	0.002	0.002	0.022	0.024
Demo Controls			✓	✓
Geo Controls				✓

Notes. This table presents the estimated effects of district-level exposure to import competition on sentiments towards Hispanics, African Americans, Asians, and Whites. All models allow for election year fixed effects. Observations are weighted by ANES sampling weights. Column (1) reports OLS estimates. In columns (2) through (4), $\Delta I{P}_d$ is instrumented with $\Delta I{P}_d^{IV},$ a similar measure that replaces US imports from China with Chinese manufacturing exports to 8 high-income countries. The model in column (3) controls for gender, age, education, race, and ethnicity. The model in column (4) controls for the 1990 district share of white and foreign born population, the 1990 district share of manufacturing employment, as well as the district-level two-party Republican vote share in the 1992 presidential election. Standard errors clustered at the district-year level are presented in parentheses. ^*** p < 0.01, ^** p < 0.05, ^* p < 0.1.

Table 3. Heterogeneity by individual gender identity.

	Gender identity of respondent
	(1) Male	(2) Female
Outcome variable
Feeling thermometer: Hispanics	–1.199**	–0.268
	(0.533)	(0.473)
Feeling thermometer: White	0.824*	0.829*
	(0.440)	(0.460)
Feeling thermometer: Muslims	–1.418***	–0.642
	(0.441)	(0.609)
Feeling thermometer: Christians	0.942*	1.088**
	(0.561)	(0.483)
Feeling thermometer: Gay men and lesbians	–1.485**	0.071
	(0.577)	(0.585)
Gay and lesbian couples allowed to adopt = 1	–0.023**	–0.018*
	(0.009)	(0.010)
Sample size	6,547	5,743

Notes. This table presents the estimated effects of district-level exposure to import competition on attitudes towards racial, ethnic, religious and sexual in-groups and out-groups by gender identity. Observations are weighted by ANES sampling weights. All models allow for election year fixed effects and control for gender, age, education, race, ethnicity, the 1990 district share of white and foreign born population, the 1990 district share of manufacturing employment, as well as the district-level two-party Republican vote share in the 1992 presidential election. Standard errors clustered at the district-year level are presented in parentheses. ^*** p < 0.01, ^** p < 0.05, ^* p < 0.1.

Figure 3. Import exposure and attitudes towards religious groups and sexual minorities.

Notes: This coefficient plot illustrates the estimated effects of district-level exposure to import competition on sentiments towards religious out-groups and in-groups in the left-hand panel, and opinions about sexual minorities in the right-hand panel. The leftmost (lightest) points of each triplet are TSLS estimates from models with no controls; middle points are TSLS estimates from models controlling for gender, age, education, race, and ethnicity; rightmost (darkest) points are TSLS estimates from models including the full set of controls used in Table 1. All models allow for election year fixed effects. Observations are weighted by ANES sampling weights. Standard errors are clustered at the district-year level. 95% confidence intervals are indicated with thin bars, while 90% intervals are indicated with thick bars.

Figure 4. Import exposure and attitudes towards trade and redistribution.

Notes: This coefficient plot illustrates the estimated effects of district-level exposure to import competition on sentiments towards international trade in the left-hand panel, and opinions about redistribution in the right-hand panel. The leftmost (lightest) points of each triplet are TSLS estimates from models with no controls; middle points are TSLS estimates from models controlling for gender, age, education, race, and ethnicity; rightmost (darkest) points are TSLS estimates from models including the full set of controls used in Table 1. All models allow for election year fixed effects. Observations are weighted by ANES sampling weights. Standard errors are clustered at the district-year level. 95% confidence intervals are indicated with thin bars, while 90% intervals are indicated with thick bars.

Table 2. Heterogeneity by individual race and ethnicity.

	Racial/Ethnic group of respondent
	(1)	(2)	(3)	(4)
	White	Black	Hispanic	Asian
Outcome variable	Non-Hispanic	Non-Hispanic
Feeling thermometer: Hispanics	–0.539	–0.688	–0.926	–3.792**
	(0.429)	(1.107)	(1.097)	(1.546)
Feeling thermometer: White	0.698**	1.604	1.161	–0.593
	(0.338)	(1.122)	(0.895)	(1.195)
Feeling thermometer: Muslims	–1.237***	–0.167	0.160	–3.375**
	(0.458)	(1.294)	(0.863)	(1.608)
Feeling thermometer: Christians	1.377***	0.556	0.339	–1.849
	(0.447)	(1.095)	(0.884)	(1.428)
Feeling thermometer: Gay men and lesbians	–1.439***	1.944	0.927	0.813
	(0.516)	(1.305)	(0.985)	(1.822)
Gay and lesbian Couples allowed to adopt = 1	–0.028***	–0.005	–0.007	0.046
	(0.009)	(0.022)	(0.014)	(0.028)
Sample size	7,583	1,973	1,957	756

Notes. This table presents the estimated effects of district-level exposure to import competition on attitudes towards racial, ethnic, religious and sexual in-groups and out-groups by individual race and ethnicity. Observations are weighted by ANES sampling weights. All models allow for election year fixed effects and control for gender, age, education, race, ethnicity, the 1990 district share of white and foreign born population, the 1990 district share of manufacturing employment, as well as the district-level two-party Republican vote share in the 1992 presidential election. Standard errors clustered at the district-year level are presented in parentheses. ^*** p < 0.01, ^** p < 0.05, ^* p < 0.1.