Many commentators view Trump’s victory and the rise of other anti-establishment parties in the Western world as a backlash against economic globalization. Indeed, Donald Trump campaigned heavily on claims that “U.S. jobs fled the country and headed to China and Mexico.”

We looked more closely at this issue and found that counties facing stronger import competition from Mexico — and especially China — logged more votes for Trump compared with other Republican candidates in recent U.S. presidential elections.

Yes, import competition matters to voters

From an economic standpoint, Trump’s victory is puzzling. The United States has low and declining unemployment, so a platform based on a protectionist agenda should make little sense. But focusing on the role of international trade helps solve part of the enigma.

Economists David AutorDavid Dorn and Gordon Hanson provide convincing evidence that imports from China led to fewer manufacturing jobs in the United States. In a recent paper with Kaveh Majlesi, they found that districts with industries more vulnerable to Chinese imports tended to elect congressional representatives taking more polarized stances, both among Republicans and Democrats. As reported in the Washington Post, new research by this group found that import penetration from China boosted support for Trump.

And economists J. Bradford Jensen, Stephen Weymouth and Dennis P. Quinn point out that incumbent parties are more likely to lose votes when imports increase and exports decrease, particularly in swing states where low-skilled manufacturing workers face competition from imports.

Other political economists show international trade affects U.S. voting. For example, Yotam Margalit shows that trade-related job losses — more than other job losses — make it harder for incumbents to win elections.

So what about 2016 voting patterns?

We collected county-level data on employment by industry in the U.S. manufacturing sector from the Quarterly Census of Employment and Wages (QCEW). Also, we considered county-level election results, drawn from the CQ Voting & Elections Collection. Our third data source, for international trade statistics, was the UN Comtrade Database.

We used an approach first introduced by Autor, Dorn and Hanson to gauge county-level exposure to import competition. Economist Italo Colantone and Piero Stanig, a political scientist, also used this method to study the Brexit vote.

1) Here’s what we measured: The indicator uses each county’s industry composition at the beginning of a specific time frame, as well as variations in national imports from China and Mexico during the same period. So a high indicator for a county means local industries competed more heavily with Chinese and Mexican imports. This competition typically concentrates in electronics and textiles from China, or machinery and crude petroleum from Mexico.

For Mexico, we considered variations in trade between 1993 and 2016, to measure trade flows after the North American Free Trade Agreement (NAFTA) came into force. We looked at China-related data after 2001, when China joined the World Trade Organization (see Figure 1). We found 1993 employment data publicly available for 2,220 counties, and 2001 data for 2,588 counties.

Next, we performed a regression analysis of the effect of import competition on the difference between the Republican vote share in the 2016 presidential election and the average share of votes for Republicans from 1996 (the first national election after the signing of NAFTA) through 2012 (see Figure 2). We weighted for each county’s population and added controls for changes in ethnic composition and educational attainment between 2000 and 2015.

 2) Here’s what we found: Counties experiencing greater import shocks are the same counties where votes for the Republican presidential candidate jumped more in 2016. This is especially true for Chinese imports (see Figure 3). On average, a one-point increase in the indicator of import competition from China is associated with a 2.9 percent increase in support for Donald Trump vis-à-vis the county’s average support for Republican candidates over the past 20 years.

For Mexican imports there is a slightly weaker relationship. The average increase in support for Republicans associated with import shocks from Mexico is around 2 percent.

Comparing the two scatter plots shows two interesting facts. First, counties faced on average tighter competition from China than Mexico. Second, the correlation between import competition and support for Trump is stronger for the Chinese indicator than the Mexican one.

Our model looks at trade exposure in a way that is independent from the state of the domestic economy, and the results hold even after accounting for changes in county demographics and educational attainment. We used a regression model that incorporates import variations experienced by the other G20 countries from the same trade partners and over the same time frame. This “instrumental variable” approach, similar to the one adopted by Autor, Dorn and Hanson, addressed the possibility that unmeasured factors might drive both import competition and voting behavior.

Does this explain the appeal of protectionism?

We think that economic globalization — and its consequences at the local level — has fostered demands for protection, especially among the most vulnerable categories of workers. In turn, this has underpinned Trump’s anti-trade agenda for the United States. Ultimately, our analysis points out that import competition may have played a role in the presidential vote.

While the backlash against globalization may be one of the common denominators of the rise of populism in the Western world, we caution that there is more to this story. We need more research and more accurate theories to explain the timing of this phenomenon and its context-specific effects, which seem to be very dependent upon political, institutional and historical contingencies.

Therefore, we cannot presume that Donald Trump’s electoral success revolves entirely on domestic conflicts triggered by international economic dynamics. But we claim that deeper scrutiny of these dynamics is important to make full sense of the changing directions of our times. The data suggest so, too.

Andrea Cerrato and Francesco Ruggieri are research professionals at the University of Chicago Booth School of Business. Federico Maria Ferrara is a PhD candidate in political science at the University of Geneva. This post was submitted  Nov. 20 and election results are provisional to that date.