American Political Science Review Vol.105.No.1 February 2011 doi:10.1017/S000305541000050X Costly Jobs:Trade-related Layoffs,Government Compensation, and Voting in U.S.Elections YOTAM MARGALIT Columbia University oes globalization's impact on the labor market affect how people vote?I address this question using a new dataset based on plant-level data that measures the impact of foreign competition on the U.S.workforce over an 8-year period.Analyzing change in the president's vote share,I find that voters were substantially more sensitive to the loss of local jobs when it resulted from foreign competition,particularly from offshoring,than to job losses caused by other factors.Yet,I also find that between 2000 and 2004,the anti-incumbent effect of trade-related job losses was smaller in areas where the government certified more of the harmed workers to receive special job training and income assistance. The findings have implications for understanding the impact of international economic integration on voting behavior,as well as for assessing the electoral effect of government programs designed to compensate the losers from globalization. penness to international markets has long been trend that further intensified during the 2008 elections a contentious domestic issue.Traditionally,a when the number of paid campaign ads concerned major source of opposition was the threat that with trade and offshoring increased almost sixfold.3 import competition will hurt local producers (Irwin This trend suggests that politicians sense that voters 2005:O'Rourke and Williamson 1999).In recent years, are anxious about globalization's impact on their em- however,sharp decreases in communication and trans- ployment prospects and also perceive this anxiety as portation costs have increasingly enabled firms to take an effective electoral issue,giving rise to two related advantage of cost disparities between countries by"off- questions:(1)does globalization's impact on the labor shoring"manufacturing operations and business func- market have a demonstrable effect on people's vot- tions to locations overseas.This offshoring trend has ing preferences?;and (2)are voters sensitive to job substantially broadened the range of local jobs that are losses per se,or is there a unique electoral response exposed to competition from abroad (Blinder 2006; when job losses are caused specifically by foreign Jensen and Kletzer 2008)and elevated the threat to competition? domestic workers posed by globalization into a salient The notion that voters'preferences are sensitive to electoral issue.Notably,political campaigns are not the impact of economic openness underlies a range only raising the specter of local job losses,but also of arguments in the literature.For example,some specifically highlighting foreign competition as a root scholars posit that trade liberalization leads to an cause (Public Citizen 2006). increase in social spending as governments seek to In the U.S.presidential elections of 2004,for exam- prevent a backlash by voters in the face of grow- ple,the Democrats made offshore outsourcing a staple ing employment risks from international competition issue of their campaign,charging that the president's (e.g.,Burgoon 2001;Cameron 1978;Garrett,1998: policies were rewarding corporations for"shipping U.S. Rodrik 1998;Ruggie 1982).Other studies explain gov- jobs overseas."1 In the midterm elections held 2 years ernments'choice of trade policy as a function of the later,foreign competition was again a ubiquitous elec- likely support the policy would obtain by the electorate toral issue used in 115 congressional campaigns,2 a if it were put to vote (e.g.,Dutt and Mitra 2002;Hillman 1989;Mayer 1984).Again,the logic is that govern- ments seek to avoid retribution from voters adversely Yotam Margalit is Assistant Professor,Department of Political Sci- affected by trade openness.Notably,both strands of ence,Columbia University,420 W.118th Street,New York.NY 10027 (ym2297@columbia.edu). research link citizens'preferences to policy outcomes For helpful comments and suggestions I thank Brian Burgoon. (on welfare expansion or on trade)by assuming that Jowei Chen,David Epstein,Robert Erikson,Judith Goldstein,Shi- the consequences of international economic integra- geo Hirano,Macartan Humphreys,Alex Kuo,David Laitin,Jeffrey tion affect how citizens vote.Yet,to date,systematic Lax,Neil Malhotra,Pablo Pinto,Oren Rigbi,Doug Rivers,Ken evidence to back up this key assumption has been Scheve,Robert Shapiro.Johannes Urpelainen.Yaniv Yedid-Levi, four anonymous referees and the APSR coeditors,as well as par- scant. ticipants at the American Empirical Series at Stanford University Several recent studies have made initial steps in ex- and the American Politics Workshop at Columbia University.Special amining the link between economic integration and thanks are also due to Patrick Chang and Khan Shing for outstanding voting by using survey data to reveal associations be- research assistance.Earlier versions of the paper were presented at the 2008 meetings of the International Political Economy Society tween self-perceived job insecurity and support for and at the 2009 Annual Meeting of the Political Science Association parties and candidates skeptical of globalization (e.g., I See Dan Balz and Paul Farhi,"Kerry,Edwards Attack Bush on Workers'Woes."Washington Post.February 26.2004. 2 See Public Citizen,"Election 2006:No to Staying the Course on 3 See Public Citizen,"Election 2008:Fair Trade Gets an Upgrade," Trade,"Global Trade Watch,November 8,2006,www.citizen.org Global Trade Watch,January 8,2009 (updated),www.citizen.org/ documents/Election2006.pdf (accessed October 24,2010). documents/ElectionReportFINAL.pdf(accessed October 24,2010) 166
American Political Science Review Vol. 105, No. 1 February 2011 doi:10.1017/S000305541000050X Costly Jobs: Trade-related Layoffs, Government Compensation, and Voting in U.S. Elections YOTAM MARGALIT Columbia University Does globalization’s impact on the labor market affect how people vote? I address this question using a new dataset based on plant-level data that measures the impact of foreign competition on the U.S. workforce over an 8-year period. Analyzing change in the president’s vote share, I find that voters were substantially more sensitive to the loss of local jobs when it resulted from foreign competition, particularly from offshoring, than to job losses caused by other factors. Yet, I also find that between 2000 and 2004, the anti-incumbent effect of trade-related job losses was smaller in areas where the government certified more of the harmed workers to receive special job training and income assistance. The findings have implications for understanding the impact of international economic integration on voting behavior, as well as for assessing the electoral effect of government programs designed to compensate the losers from globalization. Openness to international markets has long been a contentious domestic issue. Traditionally, a major source of opposition was the threat that import competition will hurt local producers (Irwin 2005; O’Rourke and Williamson 1999). In recent years, however, sharp decreases in communication and transportation costs have increasingly enabled firms to take advantage of cost disparities between countries by “offshoring” manufacturing operations and business functions to locations overseas. This offshoring trend has substantially broadened the range of local jobs that are exposed to competition from abroad (Blinder 2006; Jensen and Kletzer 2008) and elevated the threat to domestic workers posed by globalization into a salient electoral issue. Notably, political campaigns are not only raising the specter of local job losses, but also specifically highlighting foreign competition as a root cause (Public Citizen 2006). In the U.S. presidential elections of 2004, for example, the Democrats made offshore outsourcing a staple issue of their campaign, charging that the president’s policies were rewarding corporations for “shipping U.S. jobs overseas.”1 In the midterm elections held 2 years later, foreign competition was again a ubiquitous electoral issue used in 115 congressional campaigns,2 a Yotam Margalit is Assistant Professor, Department of Political Science, Columbia University, 420 W. 118th Street, New York, NY 10027 (ym2297@columbia.edu). For helpful comments and suggestions I thank Brian Burgoon, Jowei Chen, David Epstein, Robert Erikson, Judith Goldstein, Shigeo Hirano, Macartan Humphreys, Alex Kuo, David Laitin, Jeffrey Lax, Neil Malhotra, Pablo Pinto, Oren Rigbi, Doug Rivers, Ken Scheve, Robert Shapiro, Johannes Urpelainen, Yaniv Yedid-Levi, four anonymous referees and the APSR coeditors, as well as participants at the American Empirical Series at Stanford University and the American Politics Workshop at Columbia University. Special thanks are also due to Patrick Chang and Khan Shing for outstanding research assistance. Earlier versions of the paper were presented at the 2008 meetings of the International Political Economy Society and at the 2009 Annual Meeting of the Political Science Association. 1 See Dan Balz and Paul Farhi, “Kerry, Edwards Attack Bush on Workers’ Woes,” Washington Post, February 26, 2004. 2 See Public Citizen, “Election 2006: No to Staying the Course on Trade,” Global Trade Watch, November 8, 2006, www.citizen.org/ documents/Election2006.pdf (accessed October 24, 2010). trend that further intensified during the 2008 elections, when the number of paid campaign ads concerned with trade and offshoring increased almost sixfold.3 This trend suggests that politicians sense that voters are anxious about globalization’s impact on their employment prospects and also perceive this anxiety as an effective electoral issue, giving rise to two related questions: (1) does globalization’s impact on the labor market have a demonstrable effect on people’s voting preferences?; and (2) are voters sensitive to job losses per se, or is there a unique electoral response when job losses are caused specifically by foreign competition? The notion that voters’ preferences are sensitive to the impact of economic openness underlies a range of arguments in the literature. For example, some scholars posit that trade liberalization leads to an increase in social spending as governments seek to prevent a backlash by voters in the face of growing employment risks from international competition (e.g., Burgoon 2001; Cameron 1978; Garrett, 1998; Rodrik 1998; Ruggie 1982). Other studies explain governments’ choice of trade policy as a function of the likely support the policy would obtain by the electorate if it were put to vote (e.g., Dutt and Mitra 2002; Hillman 1989; Mayer 1984). Again, the logic is that governments seek to avoid retribution from voters adversely affected by trade openness. Notably, both strands of research link citizens’ preferences to policy outcomes (on welfare expansion or on trade) by assuming that the consequences of international economic integration affect how citizens vote. Yet, to date, systematic evidence to back up this key assumption has been scant. Several recent studies have made initial steps in examining the link between economic integration and voting by using survey data to reveal associations between self-perceived job insecurity and support for parties and candidates skeptical of globalization (e.g., 3 See Public Citizen, “Election 2008: Fair Trade Gets an Upgrade,” Global Trade Watch, January 8, 2009 (updated), www.citizen.org/ documents/ElectionReportFINAL.pdf (accessed October 24, 2010). 166
American Political Science Review Vol.105,No.1 Mughan,Bean,and McAllister 2003;Mughan and Lacy nificant geographic variation in the electoral impact 2002).However.their reliance on observational survey of job losses stemming from trade,whereby the presi- data and the absence of objective measures of the la- dent's support was almost unaffected in some areas but bor market impact have made it difficult to establish dropped by upward of 4 percentage points in the hard- a causal link between economic integration and voting est hit counties.In Wisconsin,a state with 10 electoral outcomes.As a recent review of this literature con- votes,the electoral impact associated with job losses cludes,there is a notable paucity of empirical evidence due to foreign competition was in fact larger than the that globalization's consequences actually affect voting swing needed to overturn the election's outcome.These behavior(Kayser 2007).4 findings suggest that the localized electoral impact of To explore the electoral consequences of globaliza- trade-related job losses,although modest at the na- tion's labor market impact and address some of the key tional aggregate,could still have a significant influence methodological issues hampering past research,I as- on a government's trade agenda by potentially risking sembled a novel dataset based on plant-level data from the president's chances in specific closely fought states the U.S.Department of Labor (DOL).This dataset or by threatening the electoral chances of some of includes every application made over an 8-year pe- the administration's allies in Congress,thus depriving riod(1996-2004)to the DOL's Trade Adjustment and the president's trade policy of crucial support in the Assistance (TAA)division to request compensation legislature. to workers whose employment was harmed by trade- A final notable result is that during the first Bush Ad- related competition.I use these data to generate geo- ministration,the anti-incumbent effect of trade-related graphically detailed measures of the impact of various job losses was smaller in those counties where more forms of trade openness on layoffs in each U.S.county. of the harmed workers were made eligible to receive Matching these objective measures with electoral data trade adjustment assistance from the government.This I exploit the variation over time and across geographic result ties into the ongoing debate over the policies units to estimate the localized effect of trade-related needed to sustain public support for globalization(e.g., job losses on vote choice.In addition,by incorporating Kletzer 2001:Marcal 2001:Scheve and Slaughter 2007) data on the TAA's actual certification decisions.I also and suggests that a government-funded compensatory explore the link between government compensation scheme can do more than help workers readjust in the schemes and the sensitivity of voters to the labor mar- labor market.It can also serve as a political tool for ket consequences of globalization. politicians that want to advance trade liberalization A central finding of this article is that trade-related but fear its electoral repercussions job losses,predominantly those resulting from off- This article contributes to the literature on the po- shoring,have a significant negative effect on sup- litical economy of trade (e.g.,Hillman 1989;Hiscox port for the incumbent that is above and beyond the 2002;Rogowski 1987)by estimating the electoral im- generic"electoral effect associated with local reduc- pact of a key political aspect of trade liberalization, tion in employment.This effect is found across multiple namely,its effect on domestic job losses.It also adds election cycles,under both Republican and Democratic to the literature on economic voting that holds that presidents.Between 2000 and 2004,the electoral cost vote choice reflects citizens'retrospective assessment to the incumbent of a marginal job lost due to foreign of the economic conditions (e.g.,Kinder and Kiewiet competition was,on average,more than twice as large 1979;Lewis-Beck and Stegmaier 2000).Indeed,eco- as the effect of a job loss resulting from other causes nomic voting models typically include employment (e.g.,domestic competition).I demonstrate that this as one of the factors that influence vote choice.The effect holds across a wide range of specifications and finding that layoffs due to foreign competition have cannot be explained by measurement issues or by the a uniquely large electoral impact,particularly when method used to account for employment changes in caused by offshoring,demonstrates that vote choice the county.I also test my identification strategy with is not responsive to all job losses in an equal fash- a placebo specification and offer further evidence that ion.Rather.the electoral effect varies as a function the finding is not spurious. of the cause of the job loss,indicating that the retro- I find that the overall national electoral effect of spective assessments of voters are perhaps more dis- trade-related job losses in the 2004 elections was about criminating than assumed in most models of economic a 0.2 percentage point drop in support for the in- voting cumbent,equivalent to the electoral effect of an in- The remainder of the article is organized as follows. crease of half a percentage point in counties'unem- The next section provides a theoretical overview and ployment rates between the elections.Notably,my background.I then explain the empirical strategy,de- analysis suggests that this effect was probably offset scribe the data,and review the main measurement is- by the electoral gains associated with trade's contri- sues.The results of the analysis ensue.I begin by exam- bution to the national economy.Yet,I also find sig- ining the impact of trade-related job losses on voting. Next,I explore the political implications of the finding by focusing on the differential effects of offshoring and 4 In a review of the literature dealing with globalization's impact on on the electoral impact of government compensation domestic politics,Kayser notes that"The sheer volume of literature for trade-affected workers.The final section investi- in this area has made it easy to overlook an important fact:Very little of it addresses the effect of economic globalization on actual politics. gates the mechanism underlying the unique electoral understood more narrowly as electoral politics"(341). response to trade-related job losses. 167
American Political Science Review Vol. 105, No. 1 Mughan, Bean, and McAllister 2003; Mughan and Lacy 2002). However, their reliance on observational survey data and the absence of objective measures of the labor market impact have made it difficult to establish a causal link between economic integration and voting outcomes. As a recent review of this literature concludes, there is a notable paucity of empirical evidence that globalization’s consequences actually affect voting behavior (Kayser 2007).4 To explore the electoral consequences of globalization’s labor market impact and address some of the key methodological issues hampering past research, I assembled a novel dataset based on plant-level data from the U.S. Department of Labor (DOL). This dataset includes every application made over an 8-year period (1996–2004) to the DOL’s Trade Adjustment and Assistance (TAA) division to request compensation to workers whose employment was harmed by traderelated competition. I use these data to generate geographically detailed measures of the impact of various forms of trade openness on layoffs in each U.S. county. Matching these objective measures with electoral data, I exploit the variation over time and across geographic units to estimate the localized effect of trade-related job losses on vote choice. In addition, by incorporating data on the TAA’s actual certification decisions, I also explore the link between government compensation schemes and the sensitivity of voters to the labor market consequences of globalization. A central finding of this article is that trade-related job losses, predominantly those resulting from offshoring, have a significant negative effect on support for the incumbent that is above and beyond the “generic” electoral effect associated with local reduction in employment. This effect is found across multiple election cycles, under both Republican and Democratic presidents. Between 2000 and 2004, the electoral cost to the incumbent of a marginal job lost due to foreign competition was, on average, more than twice as large as the effect of a job loss resulting from other causes (e.g., domestic competition). I demonstrate that this effect holds across a wide range of specifications and cannot be explained by measurement issues or by the method used to account for employment changes in the county. I also test my identification strategy with a placebo specification and offer further evidence that the finding is not spurious. I find that the overall national electoral effect of trade-related job losses in the 2004 elections was about a 0.2 percentage point drop in support for the incumbent, equivalent to the electoral effect of an increase of half a percentage point in counties’ unemployment rates between the elections. Notably, my analysis suggests that this effect was probably offset by the electoral gains associated with trade’s contribution to the national economy. Yet, I also find sig- 4 In a review of the literature dealing with globalization’s impact on domestic politics, Kayser notes that “The sheer volume of literature in this area has made it easy to overlook an important fact: Very little of it addresses the effect of economic globalization on actual politics, understood more narrowly as electoral politics” (341). nificant geographic variation in the electoral impact of job losses stemming from trade, whereby the president’s support was almost unaffected in some areas but dropped by upward of 4 percentage points in the hardest hit counties. In Wisconsin, a state with 10 electoral votes, the electoral impact associated with job losses due to foreign competition was in fact larger than the swing needed to overturn the election’s outcome. These findings suggest that the localized electoral impact of trade-related job losses, although modest at the national aggregate, could still have a significant influence on a government’s trade agenda by potentially risking the president’s chances in specific closely fought states or by threatening the electoral chances of some of the administration’s allies in Congress, thus depriving the president’s trade policy of crucial support in the legislature. A final notable result is that during the first Bush Administration, the anti-incumbent effect of trade-related job losses was smaller in those counties where more of the harmed workers were made eligible to receive trade adjustment assistance from the government. This result ties into the ongoing debate over the policies needed to sustain public support for globalization (e.g., Kletzer 2001; Marcal 2001; Scheve and Slaughter 2007) and suggests that a government-funded compensatory scheme can do more than help workers readjust in the labor market. It can also serve as a political tool for politicians that want to advance trade liberalization but fear its electoral repercussions. This article contributes to the literature on the political economy of trade (e.g., Hillman 1989; Hiscox 2002; Rogowski 1987) by estimating the electoral impact of a key political aspect of trade liberalization, namely, its effect on domestic job losses. It also adds to the literature on economic voting that holds that vote choice reflects citizens’ retrospective assessment of the economic conditions (e.g., Kinder and Kiewiet 1979; Lewis-Beck and Stegmaier 2000). Indeed, economic voting models typically include employment as one of the factors that influence vote choice. The finding that layoffs due to foreign competition have a uniquely large electoral impact, particularly when caused by offshoring, demonstrates that vote choice is not responsive to all job losses in an equal fashion. Rather, the electoral effect varies as a function of the cause of the job loss, indicating that the retrospective assessments of voters are perhaps more discriminating than assumed in most models of economic voting. The remainder of the article is organized as follows. The next section provides a theoretical overview and background. I then explain the empirical strategy, describe the data, and review the main measurement issues. The results of the analysis ensue. I begin by examining the impact of trade-related job losses on voting. Next, I explore the political implications of the finding by focusing on the differential effects of offshoring and on the electoral impact of government compensation for trade-affected workers. The final section investigates the mechanism underlying the unique electoral response to trade-related job losses. 167
Costly Jobs February 2011 THEORETICAL OVERVIEW more likely than the lower half"(Blinder 2009,77). AND BACKGROUND This implies that up to 38 million U.S.jobs could po- A large body of research examines how the distributive tentially be offshored in the coming years.Jensen and Kletzer (2008)estimate the figure to range between consequences of globalization,particularly its impacts 15 million and 20 million jobs,still a sizable portion on the labor market,affect various political outcomes. of the U.S.labor force.In addition,offshoring could For example,some scholars attribute the growth in social spending to a shift in the public's preferences adversely affect workers whose jobs are not outsourced on welfare policy wrought by economic openness (e.g., by suppressing wages of certain low-skilled sectors Burgoon 2001;Cameron 1978;Rodrik 1998).Others in large economies (Grossman and Rossi-Hansberg assign an important role to the preferences of the con- 2008).Voters may thus be more electorally sensitive to job losses that result from offshoring because it is per- stituents in shaping the policy positions that elected ceived as an emerging threat to a broad segment of the officials take on trade(e.g.,Dutt and Mitra 2002;Mayer population. 1984:Milner and Kubota 2005).These studies rely on the assumption that citizens will electorally punish Intensity of Media Coverage.The very visible na- candidates that ignore these preferences induced by economic openness.Yet,as noted,very little empirical ture of offshoring (i.e.,of a local plant relocating its operations abroad)might be drawing greater media evidence has been put forward to date to establish this attention to instances of job losses due to this specific key assumption. cause.Because the media has substantial influence over For trade openness to have such an effect on vot- the issues that voters use to assess politicians (e.g. ing behavior,two conditions must be met:(1)voters need to be able to make a connection between trade Iyengar and Kinder 1987;Miller and Krosnick 2000), openness and its impact on their well-being,and(2) more frequent reporting on job losses when caused by offshoring may bring about a stronger shift in citizens the impact needs to be sufficiently meaningful to in- vote choice in localities hurt by this form of foreign fluence their voting preferences.So far,the empirical competition literature has predominantly offered evidence on the first condition.By examining individual-level survey data,a number of studies find that people's preferences Severity of Consequences.Empirical analyses show that displaced workers in import-competing manufac- on trade policy correlate with the labor market conse- turing are on average older and less educated,tend quences of trade (e.g.,Mayda and Rodrik 2005;Scheve to suffer lower reemployment rates,and experience and Slaughter 2001).Baker(2005)offers cross-national more sizable earning losses postplacement than the evidence that people's attitudes on trade policy reflect average displaced worker in the non-manufacturing the effect of trade not only on employment,but also on sector (Kletzer 2001,2004).7 Voters in communities consumption options.Taken together,these empirical hurt by trade might therefore be more electorally re- regularities suggest that people do make a connection sponsive because the consequences of trade-related job between openness to trade and its impact on their ma- displacements on the workers and their families are terial well-being. However,evidence on the second condition (i.e.,on more damaging in objective terms. the electoral consequences of openness)is a lot more Clarity of Responsibility.Voters tend to punish tentative.In fact,some studies argue that if govern- elected officials for adverse outcomes in conditions ments are perceived by voters to be constrained by when assignment of responsibility is clearer(e.g.,Pow- the international market,economic integration may decrease the prevalence of economic voting (Hellwig ell 2000:Powell and Witten 1993).8 If job losses due to trade are perceived by the electorate as a direct 2001:Hellwig and Samuels 2007).Yet,other findings in the literature point to several reasons why jobs lost due outcome of government policy (e.g.,due to the signing to economic openness might bring about a uniquely of a trade agreement),then people's vote choice might be more affected than when the cause of the job loss is strong electoral response. less clear. An Emerging Threat.Offshoring,particularly in the Economic Nationalism and Ethnocentricism.A services sector,is a relatively new phenomenon that is predicted to accelerate dramatically over the next number of studies find a close empirical relationship two decades.As Blinder(2006)notes,"We have so far barely seen the tip of the offshoring iceberg,the even- 6 Furthermore.even if voters themselves are shielded from the threat tual dimensions of which may be staggering"(114) of offshoring but perceive it as detrimental to the broader U.S.labor According to his estimates,"offshoring encompasses force,sociotropic concerns may also heighten voter responsiveness between 22%and 29%of all the jobs in the 2004 U.S in the face of trade-related job losses (e.g.,Feldman 1982;Sears and workforce,with the upper half of that range perhaps Funk1990). 7 The experiences of workers that lost their job due to trade-related competition are quite similar to those of workers in the nontradeable S Of course,one must keep in mind that the aggregate number of manufacturing sector. jobs in an economy is primarily a function of the size of the labor The argument about clarity of responsibility typically centers on force.Therefore,when discussing trade-related job losses,the issue the political context (e.g.,the electoral system in place).Yet,it is also is not a decrease in the aggregate number of jobs in the economy but likely to apply to the extent to which voters can attribute a specific rather a change in the sectoral composition of employment. outcome to government policy. 168
Costly Jobs February 2011 THEORETICAL OVERVIEW AND BACKGROUND A large body of research examines how the distributive consequences of globalization, particularly its impacts on the labor market, affect various political outcomes. For example, some scholars attribute the growth in social spending to a shift in the public’s preferences on welfare policy wrought by economic openness (e.g., Burgoon 2001; Cameron 1978; Rodrik 1998). Others assign an important role to the preferences of the constituents in shaping the policy positions that elected officials take on trade (e.g., Dutt and Mitra 2002; Mayer 1984; Milner and Kubota 2005). These studies rely on the assumption that citizens will electorally punish candidates that ignore these preferences induced by economic openness. Yet, as noted, very little empirical evidence has been put forward to date to establish this key assumption. For trade openness to have such an effect on voting behavior, two conditions must be met: (1) voters need to be able to make a connection between trade openness and its impact on their well-being, and (2) the impact needs to be sufficiently meaningful to in- fluence their voting preferences. So far, the empirical literature has predominantly offered evidence on the first condition. By examining individual-level survey data, a number of studies find that people’s preferences on trade policy correlate with the labor market consequences of trade (e.g., Mayda and Rodrik 2005; Scheve and Slaughter 2001). Baker (2005) offers cross-national evidence that people’s attitudes on trade policy reflect the effect of trade not only on employment, but also on consumption options. Taken together, these empirical regularities suggest that people do make a connection between openness to trade and its impact on their material well-being. However, evidence on the second condition (i.e., on the electoral consequences of openness) is a lot more tentative. In fact, some studies argue that if governments are perceived by voters to be constrained by the international market, economic integration may decrease the prevalence of economic voting (Hellwig 2001; Hellwig and Samuels 2007). Yet, other findings in the literature point to several reasons why jobs lost due to economic openness might bring about a uniquely strong electoral response.5 An Emerging Threat. Offshoring, particularly in the services sector, is a relatively new phenomenon that is predicted to accelerate dramatically over the next two decades. As Blinder (2006) notes, “We have so far barely seen the tip of the offshoring iceberg, the eventual dimensions of which may be staggering” (114). According to his estimates, “offshoring encompasses between 22% and 29% of all the jobs in the 2004 U.S. workforce, with the upper half of that range perhaps 5 Of course, one must keep in mind that the aggregate number of jobs in an economy is primarily a function of the size of the labor force. Therefore, when discussing trade-related job losses, the issue is not a decrease in the aggregate number of jobs in the economy but rather a change in the sectoral composition of employment. more likely than the lower half” (Blinder 2009, 77). This implies that up to 38 million U.S. jobs could potentially be offshored in the coming years. Jensen and Kletzer (2008) estimate the figure to range between 15 million and 20 million jobs, still a sizable portion of the U.S. labor force. In addition, offshoring could adversely affect workers whose jobs are not outsourced by suppressing wages of certain low-skilled sectors in large economies (Grossman and Rossi-Hansberg 2008). Voters may thus be more electorally sensitive to job losses that result from offshoring because it is perceived as an emerging threat to a broad segment of the population.6 Intensity of Media Coverage. The very visible nature of offshoring (i.e., of a local plant relocating its operations abroad) might be drawing greater media attention to instances of job losses due to this specific cause. Because the media has substantial influence over the issues that voters use to assess politicians (e.g., Iyengar and Kinder 1987; Miller and Krosnick 2000), more frequent reporting on job losses when caused by offshoring may bring about a stronger shift in citizens’ vote choice in localities hurt by this form of foreign competition. Severity of Consequences. Empirical analyses show that displaced workers in import-competing manufacturing are on average older and less educated, tend to suffer lower reemployment rates, and experience more sizable earning losses postplacement than the average displaced worker in the non–manufacturing sector (Kletzer 2001, 2004).7 Voters in communities hurt by trade might therefore be more electorally responsive because the consequences of trade-related job displacements on the workers and their families are more damaging in objective terms. Clarity of Responsibility. Voters tend to punish elected officials for adverse outcomes in conditions when assignment of responsibility is clearer (e.g., Powell 2000; Powell and Witten 1993).8 If job losses due to trade are perceived by the electorate as a direct outcome of government policy (e.g., due to the signing of a trade agreement), then people’s vote choice might be more affected than when the cause of the job loss is less clear. Economic Nationalism and Ethnocentricism. A number of studies find a close empirical relationship 6 Furthermore, even if voters themselves are shielded from the threat of offshoring but perceive it as detrimental to the broader U.S. labor force, sociotropic concerns may also heighten voter responsiveness in the face of trade-related job losses (e.g., Feldman 1982; Sears and Funk 1990). 7 The experiences of workers that lost their job due to trade-related competition are quite similar to those of workers in the nontradeable manufacturing sector. 8 The argument about clarity of responsibility typically centers on the political context (e.g., the electoral system in place). Yet, it is also likely to apply to the extent to which voters can attribute a specific outcome to government policy. 168
American Political Science Review Vol.105,No.1 between opposition to trade openness and nationalist sourcing,became a central campaign theme.It rose to and ethnocentric sentiments(e.g.,Mansfield and Mutz national prominence following a statement in the 2004 2009;Margalit 2006;O'Rourke and Sinnott 2001).If a Economic Report of the President,which asserted that perception that foreigners are taking away "our jobs" "When a good or a service is produced more cheaply is stirring such sentiments,then trade-related job losses abroad,it makes more sense to import it than to make may instigate a strong electoral backlash.Lou Dobbs' or provide it domestically"(229).Seizing on this state- "Exporting America,"a television series dedicated to ment,Senator Kerry criticized the president as being exposing local businesses that are "sending Ameri- intent on "export[ing]more of our jobs overseas"and can jobs overseas,"is a prominent example of this introduced new legislation that proposed to eliminate sentiment. the deferral of corporate tax on some overseas earn- These different factors suggest that the adverse la- ings of U.S.multinationals.Soon after,the president bor market effects of economic openness,particularly responded by warning of "politicians in Washington" those resulting from offshoring,may generate a strong whose response is to "build a wall around this country electoral response.Indeed,several studies offer survey- and to isolate America from the rest of the world."10 based evidence that the consequences of economic Figure 1 presents the monthly frequency of media re- openness are associated with shifts in voting prefer- ports that discuss either offshoring or international ences.Mughan and Lacy (2002)use survey data of U.S. trade and its impact on jobs in five major news outlets. voters from the 1996 presidential elections to show that The graph illustrates the sharp spikes in the media's Americans that reported higher levels of job insecurity interest in offshoring during the year preceding the were more likely to vote for the third-party candidate 2004 elections.11 Ross Perot.They attribute this vote choice to Perot's Throughout the year preceding the elections,the opposition to the North American Free Trade Agree- threat posed by international competition to the U.S. ment (NAFTA)and his appeal to the "widespread workforce was a highly contentious topic.However, sense of job insecurity rooted in the perception that despite the Democrats'success in elevating the issue free trade...was depleting the stock of good jobs in the public debate,President Bush won the 2004 for Americans"(515).Mughan,Bean,and McAllister elections by a fairly comfortable margin of 2.4%.In (2003)study voting in the 1998 Australian general elec- the months immediately following the elections,off- tions and find that a sense of personal job insecurity in- shoring and trade policy faded from the media spot- creased support for restrictionist immigration policies, light.Whether the threat of foreign competition had an which in turn led to a higher probability of voting for effect on vote choice remains open for debate.In the the populist One Nation Party.In a similar vein,Walter following sections,I explore this issue in some detail. (2010)uses survey data from Switzerland to show that employment in a sector exposed to competition from overseas (a proxy for globalization-induced job inse- EMPIRICAL STRATEGY curity)was associated with greater support for welfare Studying the electoral impact of trade openness poses expansion,which in turn increased the probability of a substantial empirical challenge.First and foremost, voting for the Social Democratic Party. the difficulty lies in measuring the effect of trade open- These studies make headway by offering evidence ness on the voters.Simply assuming that all individuals that links economic openness,job insecurity,and shifts with a certain skill set or that all those employed in in voting preferences.Although certainly plausible,the a given industry were harmed by trade liberalization evidence put forward is either indirect or based on is grossly inaccurate.12 As a way to address this mea- survey data that have significant limitations.In partic- surement issue,I collected data from the DOL's TAA ular,most studies (with the exception of Walter)rely division,which is in charge of certifying workers hurt by solely on respondents'self-assessments of job insecu- trade-related competition to receive government com- rity and are therefore vulnerable to the possibility that pensation.This data cover every DOL review of job survey respondents are rationalizing their vote choice ex post:After voting for a party that made a certain issue salient,respondents may seek to "justify"their As Bhagwati,Panagariya.and Srinivasan(2004)note,although the vote choice by reporting that issue (in this case,job two terms are not the same,they are often used interchangeably in insecurity)as a major concern.A second and more the public discourse.The key aspect of "offshoring"is the move- ment of domestic production and related jobs overseas."Offshore serious limitation is that in using only cross-sectional outsourcing"is a subset of the offshoring phenomenon:it refers only survey data without temporal or spatial variation,these to cases in which domestic producers procure certain components of observational studies are constrained from identifying their product or aspects of their service from foreign-based suppliers, a causal effect. regardless of whether those suppliers are owned by the domestic In this article.I offer new evidence that helps ad- oducer. See CNN.com "Transcripts:Bush Addresses Employment dress the deficiencies described previously and sheds Issues in Ohio,"March 10,2004,http://transcripts.cnn.com/ new light on the link between globalization's labor TRANSCRIPTS/0403/10/lol.02.html(accessed October 24,2010). market impact and vote choice.I study the effect of 11 Figure 1 is an extension of data presented in Mankiw and Swagel trade-related job losses on voting in U.S.presidential (2006).Note that the sharpest spike in the frequency of mentions of elections between 1996 and 2004.During the latter offshoring in the news occurred in February and March,immediately following the release of the ERP. election.a relatively new form of international trade. 12 See Kletzer(2001)for a detailed discussion of the conceptual and commonly referred to as offshoring or offshore out- empirical difficulties in the measurement of trade-related job losses. 169
American Political Science Review Vol. 105, No. 1 between opposition to trade openness and nationalist and ethnocentric sentiments (e.g., Mansfield and Mutz 2009; Margalit 2006; O’Rourke and Sinnott 2001). If a perception that foreigners are taking away “our jobs” is stirring such sentiments, then trade-related job losses may instigate a strong electoral backlash. Lou Dobbs’ “Exporting America,” a television series dedicated to exposing local businesses that are “sending American jobs overseas,” is a prominent example of this sentiment. These different factors suggest that the adverse labor market effects of economic openness, particularly those resulting from offshoring, may generate a strong electoral response. Indeed, several studies offer surveybased evidence that the consequences of economic openness are associated with shifts in voting preferences. Mughan and Lacy (2002) use survey data of U.S. voters from the 1996 presidential elections to show that Americans that reported higher levels of job insecurity were more likely to vote for the third-party candidate Ross Perot. They attribute this vote choice to Perot’s opposition to the North American Free Trade Agreement (NAFTA) and his appeal to the “widespread sense of job insecurity rooted in the perception that free trade ... was depleting the stock of good jobs for Americans” (515). Mughan, Bean, and McAllister (2003) study voting in the 1998 Australian general elections and find that a sense of personal job insecurity increased support for restrictionist immigration policies, which in turn led to a higher probability of voting for the populist One Nation Party. In a similar vein, Walter (2010) uses survey data from Switzerland to show that employment in a sector exposed to competition from overseas (a proxy for globalization-induced job insecurity) was associated with greater support for welfare expansion, which in turn increased the probability of voting for the Social Democratic Party. These studies make headway by offering evidence that links economic openness, job insecurity, and shifts in voting preferences. Although certainly plausible, the evidence put forward is either indirect or based on survey data that have significant limitations. In particular, most studies (with the exception of Walter) rely solely on respondents’ self-assessments of job insecurity and are therefore vulnerable to the possibility that survey respondents are rationalizing their vote choice ex post: After voting for a party that made a certain issue salient, respondents may seek to “justify” their vote choice by reporting that issue (in this case, job insecurity) as a major concern. A second and more serious limitation is that in using only cross-sectional survey data without temporal or spatial variation, these observational studies are constrained from identifying a causal effect. In this article, I offer new evidence that helps address the deficiencies described previously and sheds new light on the link between globalization’s labor market impact and vote choice. I study the effect of trade-related job losses on voting in U.S. presidential elections between 1996 and 2004. During the latter election, a relatively new form of international trade, commonly referred to as offshoring or offshore outsourcing, became a central campaign theme.9 It rose to national prominence following a statement in the 2004 Economic Report of the President, which asserted that “When a good or a service is produced more cheaply abroad, it makes more sense to import it than to make or provide it domestically” (229). Seizing on this statement, Senator Kerry criticized the president as being intent on “export[ing] more of our jobs overseas” and introduced new legislation that proposed to eliminate the deferral of corporate tax on some overseas earnings of U.S. multinationals. Soon after, the president responded by warning of “politicians in Washington” whose response is to “build a wall around this country and to isolate America from the rest of the world.”10 Figure 1 presents the monthly frequency of media reports that discuss either offshoring or international trade and its impact on jobs in five major news outlets. The graph illustrates the sharp spikes in the media’s interest in offshoring during the year preceding the 2004 elections.11 Throughout the year preceding the elections, the threat posed by international competition to the U.S. workforce was a highly contentious topic. However, despite the Democrats’ success in elevating the issue in the public debate, President Bush won the 2004 elections by a fairly comfortable margin of 2.4%. In the months immediately following the elections, offshoring and trade policy faded from the media spotlight. Whether the threat of foreign competition had an effect on vote choice remains open for debate. In the following sections, I explore this issue in some detail. EMPIRICAL STRATEGY Studying the electoral impact of trade openness poses a substantial empirical challenge. First and foremost, the difficulty lies in measuring the effect of trade openness on the voters. Simply assuming that all individuals with a certain skill set or that all those employed in a given industry were harmed by trade liberalization is grossly inaccurate.12 As a way to address this measurement issue, I collected data from the DOL’s TAA division, which is in charge of certifying workers hurt by trade-related competition to receive government compensation. This data cover every DOL review of job 9 As Bhagwati, Panagariya, and Srinivasan (2004) note, although the two terms are not the same, they are often used interchangeably in the public discourse. The key aspect of “offshoring” is the movement of domestic production and related jobs overseas. “Offshore outsourcing” is a subset of the offshoring phenomenon; it refers only to cases in which domestic producers procure certain components of their product or aspects of their service from foreign-based suppliers, regardless of whether those suppliers are owned by the domestic producer. 10 See CNN.com, “Transcripts: Bush Addresses Employment Issues in Ohio,” March 10, 2004, http://transcripts.cnn.com/ TRANSCRIPTS/0403/10/lol.02.html (accessed October 24, 2010). 11 Figure 1 is an extension of data presented in Mankiw and Swagel (2006). Note that the sharpest spike in the frequency of mentions of offshoring in the news occurred in February and March, immediately following the release of the ERP. 12 See Kletzer (2001) for a detailed discussion of the conceptual and empirical difficulties in the measurement of trade-related job losses. 169
Costly Jobs February 2011 FIGURE 1.Media Reports on Offshoring or on Trade and Jobs in Major News Outlets (January 1996-March 2005) 100 90 Offshoring .....Trade and Jobs 80 60 50 40 20 10 0 AA % Note:This figure reports the monthly number of media reports that discuss offshoring and its variants or trade and jobs in five media outlets:New York Times,USA Today,Washington Post,Houston Chronicle,Los Angeles Times. dislocations claimed to be a result of trade openness. toral consequence to the fact that job dislocations are With this detailed data.I generate measures for each caused specifically by international competition,the U.S.county of the proportion of its workforce whose model also controls for the level of unemployment in employment was hurt by trade-related competition.I the county,as well as the change in unemployment rate then estimate the electoral effect of those job dislo- in the year preceding the elections and in the 4-year pe- cations on the change in the president's vote share in riod between the elections.13 To ensure that the results each county between the two elections are representative of the average voter,and because I begin the analysis by examining the shift between the precision of the county vote share decreases the the 2000 and 2004 elections because data on the exact smaller the number of votes,I weight the observations causes of the trade-related layoffs (e.g.,imports,off- by the number of votes cast in the county in the 2000 shoring)was not collected during the previous election elections.14 cycle (1996-2000).However,I also later incorporate A few comments regarding the interpretation of the data from the previous elections using a more aggre- estimates and the limitations of my empirical approach gated measure of trade-related job losses. are in order.First,by including controls for the county's By using the first difference in the president's vote employment level in 2004,as well as controlling for share across the two elections as the dependent vari- changes in employment in the years between elections able,the model is essentially controlling for any unob- the value y in Equation(1)should be interpreted as an served time-invariant,county-level characteristics that estimate of the localized effect of the job dislocation are correlated with support for the president.I estimate resulting specifically from foreign competition,not as a linear model,where the main regression is an estimate of the total electoral impact of the job dislocation itself.Second.this analysis estimates the (ABush Vote)i.04-00=a+XiB+y(Trade Comp)i localized electoral effect of trade-related losses using the within-county variation.This specification allows +0(Unemp)i.04+02(AUnemp)i.04-03 us to estimate the effect of additional trade-related job +(△Unemp)i.04-0, losses across counties,but it does not capture nation- (1) wide shifts in support for the incumbent due to trade's where i denotes the county,and Xi is a vector of covari- ates of county-level social,economic,and demographic 13 In the Results section,I also test alterative measures of employ- ment shifts in the county based on Mass Layoff Statistics (MLS)data characteristics.Trade Comp is the percent of workers from the Bureau of Labor Statistics. hurt by trade-related competition as a share of the total 14 The results are similar if I weight by the votes cast in 2004 or by county workforce.To test whether there is an elec- the size of the population. 170
Costly Jobs February 2011 FIGURE 1. Media Reports on Offshoring or on Trade and Jobs in Major News Outlets (January 1996–March 2005) 0 10 20 30 40 50 60 70 80 90 100 Jan-96 Jun-96 Nov-96 Apr-97 Sep-97 Feb-98 Jul-98 Dec-98 May-99 Oct-99 Mar-00 Aug-00 Jan-01 Jun-01 Nov-01 Apr-02 Sep-02 Feb-03 Jul-03 Dec-03 May-04 Oct-04 Mar-05 Freqeuncy of menons in news reports Offshoring Trade and Jobs Note: This figure reports the monthly number of media reports that discuss offshoring and its variants or trade and jobs in five media outlets: New York Times, USA Today, Washington Post, Houston Chronicle, Los Angeles Times. dislocations claimed to be a result of trade openness. With this detailed data, I generate measures for each U.S. county of the proportion of its workforce whose employment was hurt by trade-related competition. I then estimate the electoral effect of those job dislocations on the change in the president’s vote share in each county between the two elections. I begin the analysis by examining the shift between the 2000 and 2004 elections because data on the exact causes of the trade-related layoffs (e.g., imports, offshoring) was not collected during the previous election cycle (1996–2000). However, I also later incorporate data from the previous elections using a more aggregated measure of trade-related job losses. By using the first difference in the president’s vote share across the two elections as the dependent variable, the model is essentially controlling for any unobserved time-invariant, county-level characteristics that are correlated with support for the president. I estimate a linear model, where the main regression is (Bush Vote)i,04−00 = α + Xiβ + γ(Trade Comp)i + θ1(Unemp)i,04 + θ2(Unemp)i,04−03 + θ3(Unemp)i,04−00, (1) where i denotes the county, and Xi is a vector of covariates of county-level social, economic, and demographic characteristics. Trade Comp is the percent of workers hurt by trade-related competition as a share of the total county workforce. To test whether there is an electoral consequence to the fact that job dislocations are caused specifically by international competition, the model also controls for the level of unemployment in the county, as well as the change in unemployment rate in the year preceding the elections and in the 4-year period between the elections.13 To ensure that the results are representative of the average voter, and because the precision of the county vote share decreases the smaller the number of votes, I weight the observations by the number of votes cast in the county in the 2000 elections.14 A few comments regarding the interpretation of the estimates and the limitations of my empirical approach are in order. First, by including controls for the county’s employment level in 2004, as well as controlling for changes in employment in the years between elections, the value γ in Equation (1) should be interpreted as an estimate of the localized effect of the job dislocation resulting specifically from foreign competition, not as an estimate of the total electoral impact of the job dislocation itself. Second, this analysis estimates the localized electoral effect of trade-related losses using the within-county variation. This specification allows us to estimate the effect of additional trade-related job losses across counties, but it does not capture nationwide shifts in support for the incumbent due to trade’s 13 In the Results section, I also test alternative measures of employment shifts in the county based on Mass Layoff Statistics (MLS) data from the Bureau of Labor Statistics. 14 The results are similar if I weight by the votes cast in 2004 or by the size of the population. 170
American Political Science Review Vol.105,No.1 impact on jobs.In the model,such a national time location that produces a specific product for a specific trend is captured in the constant term a in Equation company.Applications made to the TAA division are (1)];however,one cannot distinguish the specific na- then reviewed by investigative teams at the DOL and, tionwide effect of trade openness from other national if found to have merit,are granted a TAA certification trends that shifted the electorate's preferences.Given that provides the workers with eligibility to a range that the nationwide impact of trade-related job losses of benefits.16 If the petition is certified,then the cer- on support for the incumbent is probably not positive, tification covers all workers in the group regardless this analysis most likely provides a conservative esti- of whether their names appeared as claimants on the mate of the electoral effect of these job losses on the petition. nationwide vote TAA petitions include information about the name Third and finally,the analysis should not be seen as of the employer,the estimated number of affected estimating the fotal impact of trade openness on voting workers,the application and determination date,and It could be,for example,that a county that experienced the address of the workplace.17 In total,the dataset many trade-related job losses gained other jobs from I use for the 8-year period between 1996 and 2004 trade-related investments,perhaps leading to a net gain includes 22,287 applications representing 2,110,310 of jobs.Moreover,trade openness has an impact on the employees.18 Table 1 provides the key summary overall state of the economy (e.g.,by spurring growth), statistics19 which in turn also affects voters'preferences.The find- As noted,the key dependent variable of interest in ings of this analysis should therefore be interpreted as the analysis is the relative change in Bush's vote share capturing the electoral impact of one aspect of trade, between the 2000 and 2004 elections.I therefore cal- albeit a politically salient one,namely,the loss of local culate all trade-related job dislocations that occurred in the 4 years between the two elections based on the determination dates provided by the TAA.Within this DATA AND MEASUREMENT period,I classify the applications along two dimensions. The first dimension distinguishes between applications The analysis covers all 3.111 U.S.counties on which that the DOL certified for TAA benefits and those complete voting data are available.The counties of that were denied.The second dimension is the cause Alaska are excluded because its electoral wards do not that brought about the job dislocation:(1)offshoring match the county boundaries.The voting data comes of production,(2)import competition,and(3)indirect from the Congressional Quarterly's Voting and Elec- foreign competition.20 I then calculate the percentage tions Collection.Data on trade-related job losses are based on applications submitted to the TAA program in the DOL.Because the measure of trade-related job 6 These benefits include a"trade readjustment allowance"for up to losses is central to all subsequent analyses,this section 52 weeks after workers'unemployment compensation benefits are describes the TAA data in some detail. exhausted (provided during the period in which a worker is partic- ipating in an approved full-time training program).a 2-year"wage insurance"that covers 50%of the difference between workers'old Trade Adjustment and Assistance Data and new salaries,support for the costs associated with retraining for another job or career,and tax credits for health care costs covering The TAA program was established as part of the Trade up to 65%of the workers'monthly health insurance premium.The exact benefits have changed somewhat over the years. Expansion Act in 1962,with the objective of providing Approximately 15%of the applications in the DOL's archive had reemployment services and benefits to workers whose incomplete address information that did not provide ZIP code data. employment was hurt by America's trading relations. For these slightly more than 3,000 cases,I used the companies'Web By providing legislatures with a means to compensate sites and the U.S.Post Office Web site to locate the addresses and match the exact ZIP codes of the affected businesses.The informa- constituents hurt by the signing of new free trade agree- tion was then manually recorded. ments,the TAA was seen as a useful political tool for 18 The TAA division reports a slightly different set of figures.The fig- helping pass legislation geared toward trade liberaliza- ures reported here are obtained after aggregating data from the dif- tion.To be eligible for TAA benefits.a worker mus ferent TAA programs,eliminating duplicate entries and terminated receive certification by the DOL confirming that cer- applications,as well as applications that had missing data or that took tain requirements are met:(1)the worker's employer place in the same calendar year but occurred after the election date Note that the figure treats TAA applications made by different divi- produces an article;and(2)the workers applying have sions of the same "mother company"as separate petitions.See the been totally or partially laid off as a result of(a)import online Appendix at http://www.journals.cambridge.org/psr2011003 competition that led to decline in sales or production for a more detailed discussion of consolidating the TAA petitions (b)a shift in production to another country with which into the final dataset. the United States has a trade agreement,or (c)due to 19 Following the signing of NAFTA in December 1993,the DOL established an additional program(NAFTA-TAA)that operated in loss of business as an upstream supplier or downstream parallel to the general TAA program in order to deal specifically producer for another company that is TAA certified.15 with job dislocations caused by NAFTA.With the passing of the Each application,which requires at least three pe- 2002 Trade Act,the two programs were consolidated and continued titioners,refers to a group of workers at a specific to operate under the banner of the TAA program.The data used in the analysis include all applications made to the TAA and the NAFTA-TAA programs during the period 1996-2004. 15 As part of the "Recovery Act."these eligibility criteria were 20 The third category includes the job dislocations resulting from a loss of business of an upstream supplier or a downstream producer slightly altered in May 2009. for a TAA-certified company. 171
American Political Science Review Vol. 105, No. 1 impact on jobs. In the model, such a national time trend is captured in the constant term [α in Equation (1)]; however, one cannot distinguish the specific nationwide effect of trade openness from other national trends that shifted the electorate’s preferences. Given that the nationwide impact of trade-related job losses on support for the incumbent is probably not positive, this analysis most likely provides a conservative estimate of the electoral effect of these job losses on the nationwide vote. Third and finally, the analysis should not be seen as estimating the total impact of trade openness on voting. It could be, for example, that a county that experienced many trade-related job losses gained other jobs from trade-related investments, perhaps leading to a net gain of jobs. Moreover, trade openness has an impact on the overall state of the economy (e.g., by spurring growth), which in turn also affects voters’ preferences. The findings of this analysis should therefore be interpreted as capturing the electoral impact of one aspect of trade, albeit a politically salient one, namely, the loss of local jobs. DATA AND MEASUREMENT The analysis covers all 3,111 U.S. counties on which complete voting data are available. The counties of Alaska are excluded because its electoral wards do not match the county boundaries. The voting data comes from the Congressional Quarterly’s Voting and Elections Collection. Data on trade-related job losses are based on applications submitted to the TAA program in the DOL. Because the measure of trade-related job losses is central to all subsequent analyses, this section describes the TAA data in some detail. Trade Adjustment and Assistance Data The TAA program was established as part of the Trade Expansion Act in 1962, with the objective of providing reemployment services and benefits to workers whose employment was hurt by America’s trading relations. By providing legislatures with a means to compensate constituents hurt by the signing of new free trade agreements, the TAA was seen as a useful political tool for helping pass legislation geared toward trade liberalization. To be eligible for TAA benefits, a worker must receive certification by the DOL confirming that certain requirements are met: (1) the worker’s employer produces an article; and (2) the workers applying have been totally or partially laid off as a result of (a) import competition that led to decline in sales or production, (b) a shift in production to another country with which the United States has a trade agreement, or (c) due to loss of business as an upstream supplier or downstream producer for another company that is TAA certified.15 Each application, which requires at least three petitioners, refers to a group of workers at a specific 15 As part of the “Recovery Act,” these eligibility criteria were slightly altered in May 2009. location that produces a specific product for a specific company. Applications made to the TAA division are then reviewed by investigative teams at the DOL and, if found to have merit, are granted a TAA certification that provides the workers with eligibility to a range of benefits.16 If the petition is certified, then the certification covers all workers in the group regardless of whether their names appeared as claimants on the petition. TAA petitions include information about the name of the employer, the estimated number of affected workers, the application and determination date, and the address of the workplace.17 In total, the dataset I use for the 8-year period between 1996 and 2004 includes 22,287 applications representing 2,110,310 employees.18 Table 1 provides the key summary statistics.19 As noted, the key dependent variable of interest in the analysis is the relative change in Bush’s vote share between the 2000 and 2004 elections. I therefore calculate all trade-related job dislocations that occurred in the 4 years between the two elections based on the determination dates provided by the TAA. Within this period, I classify the applications along two dimensions. The first dimension distinguishes between applications that the DOL certified for TAA benefits and those that were denied. The second dimension is the cause that brought about the job dislocation: (1) offshoring of production, (2) import competition, and (3) indirect foreign competition.20 I then calculate the percentage 16 These benefits include a “trade readjustment allowance” for up to 52 weeks after workers’ unemployment compensation benefits are exhausted (provided during the period in which a worker is participating in an approved full-time training program), a 2-year “wage insurance” that covers 50% of the difference between workers’ old and new salaries, support for the costs associated with retraining for another job or career, and tax credits for health care costs covering up to 65% of the workers’ monthly health insurance premium.The exact benefits have changed somewhat over the years. 17 Approximately 15% of the applications in the DOL’s archive had incomplete address information that did not provide ZIP code data. For these slightly more than 3,000 cases, I used the companies’ Web sites and the U.S. Post Office Web site to locate the addresses and match the exact ZIP codes of the affected businesses. The information was then manually recorded. 18 The TAA division reports a slightly different set of figures. The figures reported here are obtained after aggregating data from the different TAA programs, eliminating duplicate entries and terminated applications, as well as applications that had missing data or that took place in the same calendar year but occurred after the election date. Note that the figure treats TAA applications made by different divisions of the same “mother company” as separate petitions. See the online Appendix at http://www.journals.cambridge.org/psr2011003 for a more detailed discussion of consolidating the TAA petitions into the final dataset. 19 Following the signing of NAFTA in December 1993, the DOL established an additional program (NAFTA-TAA) that operated in parallel to the general TAA program in order to deal specifically with job dislocations caused by NAFTA. With the passing of the 2002 Trade Act, the two programs were consolidated and continued to operate under the banner of the TAA program. The data used in the analysis include all applications made to the TAA and the NAFTA-TAA programs during the period 1996–2004. 20 The third category includes the job dislocations resulting from a loss of business of an upstream supplier or a downstream producer for a TAA-certified company. 171
Costly Jobs February 2011 TABLE 1.Descriptive Statistics: the location of trade-related job losses was not evenly spread geographically.As the map in Figure 2a shows. TAA Applications 1996-2004,by the areas that suffered the highest share of trade- Administration Period related layoffs were the Northeast,the "Rust Belt," Time Period the South,and the Midwest.In contrast,employment in the Great Plains region (e.g.,Nebraska,North and Variable 1996-2000 2001-04 South Dakota,Oklahoma)was almost unaffected by trade-related competition.Notably,Figure 2 also high- Applications made 9,462 12,825 lights the fact that the geographic pattern of trade- Workers represented in 906,675 1.203,635 related layoffs differs from the pattern of the overall applications Applications approved 63.4% 60.4% change in unemployment in that period (Figure 2b),the Workers certified 692.999 909.873 concentration in absolute levels of unemployment in Workers denied 211,292 290,432 2004(Figure 2c),and the pattern of"generic"(i.e.,not Average workers per 95.8 93.9 only trade-related)layoffs in that same 4-year period.22 application These variables capture related,yet clearly different phenomena. Notes:NAFTA,North American Free Trade Agreement:TAA. The prevalence of trade-related layoffs varied not Trade Adjustment and Assistance.Data prior to 2002 includes ooth TAA and NAFTA-TAA programs combined. only across geographic units,but also across indus- tries.Overall,applications to the TAA were made by 340 different industries.23 Between 2001 and 2004,the industry with the highest number of affected work- ers was the "electronic components and accessories," of workers affected by each cause as a share of the total followed by "men's and boys'furnishing,work cloth- workforce in the county. ing,and allied garments"(with 115,218 workers and The DOL produces a report explaining each decision 65,119 workers,respectively).See Table A2 in the on- to deny workers TAA certification.A reading of those line Appendix(available at http://www.journals.cam- reports,as well as conversations with personnel at the bridge.org/psr2011003)for more details on the distri- TAA division,suggest that many of the cases denied bution of TAA applications across industries. TAA certification are nonetheless instances in which One potential concern in using the TAA data for this workers were hurt by foreign competition,even if not analysis is selection bias.For selection bias to affect the in a way that met the DOL's eligibility criteria for re- results of the analysis,one must consider two distinct ceiving government compensation.The TAA's reports possibilities.The first is that the "nonapplicants"(i.e., reveal two common reasons for denial.The first is when those workers whose employment was hurt by trade- the applicant is not able to prove that the job disloca- related competition but did not apply to the TAA)are tion was caused primarily by foreign competition.That dispersed geographically in a similar fashion to those is,workers whose employer was hurt by both foreign that did apply.Put differently,no location factors ac- competition and other reasons are often denied TAA count for,or correlate with,workers'decision to apply certification.The second reason is when the employer for TAA certification.If that is the case,estimates of is judged not to produce an"article,"a definition that can exclude some providers of services.21 In both cases, the electoral effect of trade-related layoffs will be over- stated.However,this concern is at least partly allevi- the workers included in the applications had good rea- ated by the fact that the analysis relies on the firm-level son to view themselves as hurt by trade openness,de- data collected as part of the DOL's investigation of an spite being ineligible for TAA certification.For this application,not on each affected worker petitioning reason,in the subsequent analysis,I measure trade- individually.In other words,a single application from related job dislocations using all applications submitted a plant that laid off workers is sufficient for the entire to the TAA,whether certified or denied.I then repeat number of affected workers in the plant to be recorded the analysis using separate measures for certified and in the TAA data,regardless of whether those workers denied cases. then collected the TAA benefits.24 Businesses from about half the U.S.counties (50.2%) applied to the TAA in the 4 years preceding the 2004 elections.In those counties from which TAA applica- 22"Generic"layoffs are calculated using data from the MLSprogram tions were submitted,an average of 2%of the work- at the Bureau of Labor Statistics(BLS).I return to discuss these data force was represented.In total,76%of the applicants and the generic layoffs measure in greater detail in the Robustness section. were certified to receive TAA compensation.The most 23 Industries defined at the three-digit Standard Industrial Classifi- common reason for certification was import competi- cation (SIC)leveL tion (43%),followed by offshoring (42%).Notably, 24 To make this distinction clearer,consider an example of a U.S company relocating overseas and laying off its 200 workers.Let us assume that only 20 workers apply to the DOL for TAA certifica- 21 That some of the companies providing services were ineligible for tion and that after receiving the DOL's certification,only 10 of the TAA compensation means that the coverage of the TAA data does workers actually collect the full TAA benefits,whereas the others not completely represent all instances of trade-related job losses. immediately find new jobs.When the DOL assesses the petitions However,this is not a major issue because trade-related service job from this plant,it produces a single report that either certifies or losses still account for only a small share of the overall jobs lost due denies all affected workers.The dataset I use thus includes the total to foreign competition(Blinder 2009,1). number of workers considered by the DOL as affected by the plant 172
Costly Jobs February 2011 TABLE 1. Descriptive Statistics: TAA Applications 1996–2004, by Administration Period Time Period Variable 1996–2000 2001–04 Applications made 9,462 12,825 Workers represented in 906,675 1,203,635 applications Applications approved 63.4% 60.4% Workers certified 692,999 909,873 Workers denied 211,292 290,432 Average workers per 95.8 93.9 application Notes: NAFTA, North American Free Trade Agreement; TAA, Trade Adjustment and Assistance. Data prior to 2002 includes both TAA and NAFTA-TAA programs combined. of workers affected by each cause as a share of the total workforce in the county. The DOL produces a report explaining each decision to deny workers TAA certification. A reading of those reports, as well as conversations with personnel at the TAA division, suggest that many of the cases denied TAA certification are nonetheless instances in which workers were hurt by foreign competition, even if not in a way that met the DOL’s eligibility criteria for receiving government compensation. The TAA’s reports reveal two common reasons for denial. The first is when the applicant is not able to prove that the job dislocation was caused primarily by foreign competition. That is, workers whose employer was hurt by both foreign competition and other reasons are often denied TAA certification. The second reason is when the employer is judged not to produce an “article,” a definition that can exclude some providers of services.21 In both cases, the workers included in the applications had good reason to view themselves as hurt by trade openness, despite being ineligible for TAA certification. For this reason, in the subsequent analysis, I measure traderelated job dislocations using all applications submitted to the TAA, whether certified or denied. I then repeat the analysis using separate measures for certified and denied cases. Businesses from about half the U.S. counties (50.2%) applied to the TAA in the 4 years preceding the 2004 elections. In those counties from which TAA applications were submitted, an average of 2% of the workforce was represented. In total, 76% of the applicants were certified to receive TAA compensation. The most common reason for certification was import competition (43%), followed by offshoring (42%). Notably, 21 That some of the companies providing services were ineligible for TAA compensation means that the coverage of the TAA data does not completely represent all instances of trade-related job losses. However, this is not a major issue because trade-related service job losses still account for only a small share of the overall jobs lost due to foreign competition (Blinder 2009, 1). the location of trade-related job losses was not evenly spread geographically. As the map in Figure 2a shows, the areas that suffered the highest share of traderelated layoffs were the Northeast, the “Rust Belt,” the South, and the Midwest. In contrast, employment in the Great Plains region (e.g., Nebraska, North and South Dakota, Oklahoma) was almost unaffected by trade-related competition. Notably, Figure 2 also highlights the fact that the geographic pattern of traderelated layoffs differs from the pattern of the overall change in unemployment in that period (Figure 2b), the concentration in absolute levels of unemployment in 2004 (Figure 2c), and the pattern of “generic” (i.e., not only trade-related) layoffs in that same 4-year period.22 These variables capture related, yet clearly different phenomena. The prevalence of trade-related layoffs varied not only across geographic units, but also across industries. Overall, applications to the TAA were made by 340 different industries.23 Between 2001 and 2004, the industry with the highest number of affected workers was the “electronic components and accessories,” followed by “men’s and boys’ furnishing, work clothing, and allied garments” (with 115,218 workers and 65,119 workers, respectively). See Table A2 in the online Appendix (available at http://www.journals.cambridge.org/psr2011003) for more details on the distribution of TAA applications across industries. One potential concern in using the TAA data for this analysis is selection bias. For selection bias to affect the results of the analysis, one must consider two distinct possibilities. The first is that the “nonapplicants” (i.e., those workers whose employment was hurt by traderelated competition but did not apply to the TAA) are dispersed geographically in a similar fashion to those that did apply. Put differently, no location factors account for, or correlate with, workers’ decision to apply for TAA certification. If that is the case, estimates of the electoral effect of trade-related layoffs will be overstated. However, this concern is at least partly alleviated by the fact that the analysis relies on the firm-level data collected as part of the DOL’s investigation of an application, not on each affected worker petitioning individually. In other words, a single application from a plant that laid off workers is sufficient for the entire number of affected workers in the plant to be recorded in the TAA data, regardless of whether those workers then collected the TAA benefits.24 22 “Generic” layoffs are calculated using data from the MLS program at the Bureau of Labor Statistics (BLS). I return to discuss these data and the generic layoffs measure in greater detail in the Robustness section. 23 Industries defined at the three-digit Standard Industrial Classifi- cation (SIC) level. 24 To make this distinction clearer, consider an example of a U.S. company relocating overseas and laying off its 200 workers. Let us assume that only 20 workers apply to the DOL for TAA certification and that after receiving the DOL’s certification, only 10 of the workers actually collect the full TAA benefits, whereas the others immediately find new jobs. When the DOL assesses the petitions from this plant, it produces a single report that either certifies or denies all affected workers. The dataset I use thus includes the total number of workers considered by the DOL as affected by the plant 172
American Political Science Review Vol.105,No.1 FIGURE 2.Geographic Distribution of Job Dislocations and Unemployment(November 2000-November 2004) (a)Trade-related Layoffs(2000-04) (b)Change in Unemployment (2000-04) 0% 41指92 01%1% 0.1%-0.% 1.1%3% %例 3%-5% 21%.33w 51%-28% 33%-71% (d)'Generic'Layoffs(2000-04) (c)Unemployment Rate(2004) % 0%+43% 通1%。修 44%-57% 1.1%-% 58%.74% 7.5%10% 1% 156. 1% Note:Each map represents a different measure of the employment situation by county(clockwise):(a)the share of county workforce applying for Trade Adjustment and Assistance (TAA)compensation for trade-related job dislocations between 2000 and 2004,(b)the unemployment change in the county(2004 level minus 2000 level),(c)the county's absolute unemployment rate in 2004,and(d)the share of county workforce laid off for any reason as part of a mass layoff(>50 workers)between 2000 and 2004. The second possibility is that some characteristics of represent a lower bound of the true electoral effect of the counties from which applications were submitted trade-related job losses.In conclusion then.selection also account for the propensity of workers to apply bias is unlikely to be the explanation for the findings for TAA compensation.Given this possibility,we must presented in the later analyses. again consider two scenarios.The first is that these county characteristics are also correlated with decreas- ing support for the president.Although theoretically Other Controls possible,for this to be the case.any determinant of the decision to apply for TAA compensation must be- The benchmark specification in the analyses controls as revealed by the results presented here-dispersed for three types of county-level factors that may account geographically in a manner that correlates with op- for variation in voting outcomes:unemployment,in- posite partisan shifts between the 1996-2000 elections come,and demographic characteristics. (Democrats losing votes)and the 2000-04 elections Unemployment measures are included in the analysis (Republicans losing votes).This possibility,in itself, in order to control for the "generic"electoral effect seems quite improbable.The second option is that the associated with absolute level and change in level of same county characteristics that explain workers'deci- employment in the county.Employment data are ob- sions to apply for TAA compensation are orthogonal tained from the Local Area Unemployment Statistics to voting preferences.In this case,the results would be (LAUS)program of the Bureau of Labor Statistics biased against finding that trade-related layoffs have a (BLS).As noted,estimations include three unemploy- larger electoral effect;this would be because counties ment measures:unemployment rate in the year of the from which no TAA applications were made did in election,the change in unemployment rate in the year fact experience trade-related job losses that are not preceding the election,and the change in unemploy- captured in the data.My estimates,in this case,would ment between the two election years.For robustness. I also use measures of general layoffs (i.e.,not only trade related)at the county level using data from the closure(in this case,200 workers),notjust the 20 workers that applied Mass Layoff Statistics (MLS)program at the BLS. nor just the 10 workers that ultimately used the TAA benefits As Figure 2 indicated,the correlation between the 173
American Political Science Review Vol. 105, No. 1 FIGURE 2. Geographic Distribution of Job Dislocations and Unemployment (November 2000–November 2004) Note: Each map represents a different measure of the employment situation by county (clockwise): (a) the share of county workforce applying for Trade Adjustment and Assistance (TAA) compensation for trade-related job dislocations between 2000 and 2004, (b) the unemployment change in the county (2004 level minus 2000 level), (c) the county’s absolute unemployment rate in 2004, and (d) the share of county workforce laid off for any reason as part of a mass layoff (≥ 50 workers) between 2000 and 2004. The second possibility is that some characteristics of the counties from which applications were submitted also account for the propensity of workers to apply for TAA compensation. Given this possibility, we must again consider two scenarios. The first is that these county characteristics are also correlated with decreasing support for the president. Although theoretically possible, for this to be the case, any determinant of the decision to apply for TAA compensation must be— as revealed by the results presented here—dispersed geographically in a manner that correlates with opposite partisan shifts between the 1996–2000 elections (Democrats losing votes) and the 2000–04 elections (Republicans losing votes). This possibility, in itself, seems quite improbable. The second option is that the same county characteristics that explain workers’ decisions to apply for TAA compensation are orthogonal to voting preferences. In this case, the results would be biased against finding that trade-related layoffs have a larger electoral effect; this would be because counties from which no TAA applications were made did in fact experience trade-related job losses that are not captured in the data. My estimates, in this case, would closure (in this case, 200 workers), not just the 20 workers that applied nor just the 10 workers that ultimately used the TAA benefits. represent a lower bound of the true electoral effect of trade-related job losses. In conclusion then, selection bias is unlikely to be the explanation for the findings presented in the later analyses. Other Controls The benchmark specification in the analyses controls for three types of county-level factors that may account for variation in voting outcomes: unemployment, income, and demographic characteristics. Unemployment measures are included in the analysis in order to control for the “generic” electoral effect associated with absolute level and change in level of employment in the county. Employment data are obtained from the Local Area Unemployment Statistics (LAUS) program of the Bureau of Labor Statistics (BLS). As noted, estimations include three unemployment measures: unemployment rate in the year of the election, the change in unemployment rate in the year preceding the election, and the change in unemployment between the two election years. For robustness, I also use measures of general layoffs (i.e., not only trade related) at the county level using data from the Mass Layoff Statistics (MLS) program at the BLS. As Figure 2 indicated, the correlation between the 173
Costly Jobs February 2011 different unemployment variables is not particularly Iyengar 2000;Pew Research Center 2010).Summary high (ranges from 0.16 to 0.42).Therefore,multi- statistics for all these control variables are provided in collinearity between the unemployment covariates is Appendix Table A1. not a serious concern in this case.25 Income controls are included in the model because previous research finds a strong association between RESULTS economic standing and vote choice (e.g.,Gelman. Park,and Shor 2008;McCarty,Poole,and Rosenthal A central result of the analyses is a consistent and sta- 2006).Measures include the counties'per capita in- tistically significant association between the share of come in the year of the election and the percent change county workers hurt by trade-related job losses and a in per capita income from the previous elections. decline in the level of support for the president between These data are obtained from the Regional Economic the 2000 and 2004 presidential elections.This result Information System at the U.S.Bureau of Economic holds while including a range of controls for the em- Analysis(BEA).26 ployment shifts in the county,demonstrating the strong Demographic controls include measures of the racial electoral impact of trade-related job losses above and composition,the age distribution,and percent of home beyond the electoral effect of changes in the unem- ownership in the counties'population based on the ployment rate.The first set of analyses is presented in U.S.Census.27 Using data from the Religious Congre- Table 2.The dependent variable in the analysis is the gations Membership Study (RCMS),the model also change in Bush's vote share between the two elections controls for the breakdown of religious denominations The measure of trade-related job losses is based on in the county.The analysis also controls for the share the share of all county workers that applied to receive of small employers,small-to-medium enterprises,and TAA certifications in the 4-year period between the large corporations among the county's business estab- elections.As the results indicate,decline in support lishments using figures from the BLS's Business Pat- for President Bush was greater the higher the share of terns Data.28 These controls serve as a proxy for the adversely affected workers in the county ease with which the government could preemptively To examine the sensitivity of the findings,Table 2 assist local businesses to avoid layoffs.29 They also help shows the results from estimating a number of differ- reduce the possibility that the electoral effect associ- ent specifications.In the first column,I estimate Equa- ated with trade-related job losses is instead capturing a tion (1)without controls.The bivariate relationship response to the impact of government policies geared between trade-related layoffs and change in support toward different types of businesses.30 Finally,to exam- for Bush is negative yet statistically insignificant.In ine the potential impact of the media coverage of local column(2),adding to the model controls for level and job losses,some specifications control for the counties' change in the county's income,the coefficient of trade- designated market areas (DMAs).commonly referred related layoffs is more than doubled in size,remains to as"media markets."31 Accounting for the media cov- negative,and is statistically significant.The results also erage is important because despite the spread of cable show that Bush gained votes in counties that experi- and Internet services,Americans still receive a major enced higher rates of growth between the two elec- share of their news from local broadcasts(Gilliam and tions.In column(3),when adding controls for shifts in the counties'employment situation,the magnitude of 25 See Table A3 in the online Appendix for the correlations between the estimated effect of trade-related lavoffs decreases the unemployment variables. and the estimate is marginally significant in statistical 26 In 2001.the BEA redefined the boundaries of counties in Virginia, terms.As expected,the results show that an increase and thus its data do not apply to the exact same geographic units in in a county's unemployment rate is negatively asso- 2000 and 2004.I therefore exclude these counties from analyses that rely on the BEA income data,in which case the total sample is ciated with support for the incumbent.In column(4), reduced to 3,054 counties.See the online Appendix for detail on the which also includes the demographic controls,the point excluded counties. estimate of trade-related job losses is slightly larger 27 http://www.census.gov/support/DataDownload.htm (accessed and the estimate is more precise (as indicated by the October 24.2010). smaller standard error).In column (5),I add state fixed 28 Large corporations are businesses with 500 employees or more. Small businesses have less than 5 employees. effects because these may capture unobserved time- 29 It is presumably easier to provide assistance to a few very large invariant effects at the state level.Inclusion of the fixed local employers than to many small "mom and pop"shops.To deal effects substantially increases the variation explained with the possibility that the government preemptively assisted cer- by the model (from 33%to 64%).and the coefficient tain industries,I also test a model that includes as controls the share of trade-related job losses decreases somewhat.yet re- of county workers employed in each industry.These two estimations mains significant in statistical terms (p=.02).Finally. help address the potential limitation that the prevalence of trade- related job losses may not be entirely exogenous to the electoral to examine whether a longer-term trend of decline in process. counties'support for the Republicans accounts for the 0 During theelection campaign,Bush repeatedly touted his policies observed effects,column (6)also includes a control as pro"small business"and criticized the Democrats'tax policy as a for the voting trend in the preceding election cycle threat to small business owners.See,for example,Elizabeth Olson, (between 1996 and 2000).Controlling for this trend "Courting the Small-business Owner,"New York Times,September 23.2004. only marginally decreases the estimated effect of trade- 31 DMAs are constructed by Nielsen Media Research(Broadcasting related job losses,which remains negative and highly and Cable Yearbook 2003). significant. 174
Costly Jobs February 2011 different unemployment variables is not particularly high (ranges from 0.16 to 0.42). Therefore, multicollinearity between the unemployment covariates is not a serious concern in this case.25 Income controls are included in the model because previous research finds a strong association between economic standing and vote choice (e.g., Gelman, Park, and Shor 2008; McCarty, Poole, and Rosenthal 2006). Measures include the counties’ per capita income in the year of the election and the percent change in per capita income from the previous elections. These data are obtained from the Regional Economic Information System at the U.S. Bureau of Economic Analysis (BEA).26 Demographic controls include measures of the racial composition, the age distribution, and percent of home ownership in the counties’ population based on the U.S. Census.27 Using data from the Religious Congregations Membership Study (RCMS), the model also controls for the breakdown of religious denominations in the county. The analysis also controls for the share of small employers, small-to-medium enterprises, and large corporations among the county’s business establishments using figures from the BLS’s Business Patterns Data.28 These controls serve as a proxy for the ease with which the government could preemptively assist local businesses to avoid layoffs.29 They also help reduce the possibility that the electoral effect associated with trade-related job losses is instead capturing a response to the impact of government policies geared toward different types of businesses.30 Finally, to examine the potential impact of the media coverage of local job losses, some specifications control for the counties’ designated market areas (DMAs), commonly referred to as “media markets.”31 Accounting for the media coverage is important because despite the spread of cable and Internet services, Americans still receive a major share of their news from local broadcasts (Gilliam and 25 See Table A3 in the online Appendix for the correlations between the unemployment variables. 26 In 2001, the BEA redefined the boundaries of counties in Virginia, and thus its data do not apply to the exact same geographic units in 2000 and 2004. I therefore exclude these counties from analyses that rely on the BEA income data, in which case the total sample is reduced to 3,054 counties. See the online Appendix for detail on the excluded counties. 27 http://www.census.gov/support/DataDownload.htm (accessed October 24, 2010). 28 Large corporations are businesses with 500 employees or more. Small businesses have less than 5 employees. 29 It is presumably easier to provide assistance to a few very large local employers than to many small “mom and pop” shops. To deal with the possibility that the government preemptively assisted certain industries, I also test a model that includes as controls the share of county workers employed in each industry. These two estimations help address the potential limitation that the prevalence of traderelated job losses may not be entirely exogenous to the electoral process. 30 During the election campaign, Bush repeatedly touted his policies as pro “small business” and criticized the Democrats’ tax policy as a threat to small business owners. See, for example, Elizabeth Olson, “Courting the Small-business Owner,” New York Times, September 23, 2004. 31 DMAs are constructed by Nielsen Media Research (Broadcasting and Cable Yearbook 2003). Iyengar 2000; Pew Research Center 2010). Summary statistics for all these control variables are provided in Appendix Table A1. RESULTS A central result of the analyses is a consistent and statistically significant association between the share of county workers hurt by trade-related job losses and a decline in the level of support for the president between the 2000 and 2004 presidential elections. This result holds while including a range of controls for the employment shifts in the county, demonstrating the strong electoral impact of trade-related job losses above and beyond the electoral effect of changes in the unemployment rate. The first set of analyses is presented in Table 2. The dependent variable in the analysis is the change in Bush’s vote share between the two elections. The measure of trade-related job losses is based on the share of all county workers that applied to receive TAA certifications in the 4-year period between the elections. As the results indicate, decline in support for President Bush was greater the higher the share of adversely affected workers in the county. To examine the sensitivity of the findings, Table 2 shows the results from estimating a number of different specifications. In the first column, I estimate Equation (1) without controls. The bivariate relationship between trade-related layoffs and change in support for Bush is negative yet statistically insignificant. In column (2), adding to the model controls for level and change in the county’s income, the coefficient of traderelated layoffs is more than doubled in size, remains negative, and is statistically significant. The results also show that Bush gained votes in counties that experienced higher rates of growth between the two elections. In column (3), when adding controls for shifts in the counties’ employment situation, the magnitude of the estimated effect of trade-related layoffs decreases and the estimate is marginally significant in statistical terms. As expected, the results show that an increase in a county’s unemployment rate is negatively associated with support for the incumbent. In column (4), which also includes the demographic controls, the point estimate of trade-related job losses is slightly larger and the estimate is more precise (as indicated by the smaller standard error). In column (5), I add state fixed effects because these may capture unobserved timeinvariant effects at the state level. Inclusion of the fixed effects substantially increases the variation explained by the model (from 33% to 64%), and the coefficient of trade-related job losses decreases somewhat, yet remains significant in statistical terms (p = .02). Finally, to examine whether a longer-term trend of decline in counties’ support for the Republicans accounts for the observed effects, column (6) also includes a control for the voting trend in the preceding election cycle (between 1996 and 2000). Controlling for this trend only marginally decreases the estimated effect of traderelated job losses, which remains negative and highly significant. 174
American Political Science Review Vol.105,No.1 TABLE 2.Effect of Trade-related Job Losses on Change in Republican Presidential Vote Share (2004 Level-2000 Level) (1) (2) (3) (4) (5) (6) %Workforce applied to TAA -0.066 -0.163* -0.138t -0.142* -0.096* -0.091* (0.093) (0.080) (0.079) (0.062) (0.036) (0.033) Income per capita(A%2000-04) 0.053* 0.030 0.070* 0.055* 0.050* (0.025) (0.028) (0.030) (0.018) (0.017) Unemployment rate(2004) -0.077 -0.213 0.120 0.104 (0.164) (0.150) (0.120) (0.105) Unemployment(A%2003-04) -0.570 -0.372 -0.233 -0.191 (0.399) (0.352) (0.239) (0.241) Unemployment(A%2000-04) -0.407 -0.160 -0.337* -0.326* (0.311) (0.268) (0.137) (0.134) Labor force size(log) 0.073 0.342* 0.482* (0.140) (0.091) (0.104) Republican vote share(A%1996-2000) 0.148* (0.042) Constant 2.701* 30.432* 34.117* 187.069* 129.090* 108.674* (0.302) (9.505) (10.801) (31.908) (16.984) (18.699) Controls Income X X 十 Demographics X 十 十 State fixed effects X X 9 0.001 0.084 0.116 0.332 0.637 0.647 N 3,111 3.054 3,054 3.054 3,054 3.054 Notes:TAA,Trade Adjustment and Assistance.Standard errors are corrected for clustering at the state level.An observation in the model is a county.The dependent variable is a change in support of President Bush between 2004 and 2000. fsignificant at 10%;*significant at 5%;**significant at 1%. Robustness Tests tion between the trade-related job losses that occurred between 2000 and 2004 and the electoral change in the Before proceeding to the substantive interpretation of earlier elections is weak.and the parameter estimate is the relationship between trade-related job losses and the decrease in support for the president,I subject the a good deal smaller than its standard error.33 Column(4)examines the robustness of the findings results to a number of robustness tests by using a differ- when using local media markets (DMAs)instead of ent coding of the dependent variable or a different unit states as the control unit.The presumption is that expo- of analysis as control,estimating a placebo treatment, sure to common information via media reporting may or employing an alternative measure of shifts in the account for similar shifts in voting behavior within a labor market. media market.The model estimated in column (4)is Table 3 indicates that replicating the analysis us- the same as the benchmark specification but includes ing the two-party vote share instead of the all-party vote share as the dependent variable does not produce fixed effects for media markets instead of state fixed effects(and also clusters the standard errors by DMAs substantially different results [column(1)].This indi- instead of by states).The results show that the addi- cates that the findings are unrelated to the relatively tion of more than two hundred controls (the DMA successful run of the third-party candidate Ralph fixed effects)slightly decreases the magnitude of the Nader in the 2000 elections.The results are also robust estimated effect associated with trade-related layoffs to controlling for the vote share in 2000.instead of (from -0.091 to-0.085).although the standard error using the change in vote share between the 2004 and remains small and the coefficient is highly significant 2000 elections [column(2)]. (p=.02).The fact that the overall fit of the model in- As another test of the identification strategy,column creases suggests that the shared media content to which (3)presents the equivalent of a placebo treatment, voters in different counties are exposed may have an examining the effect of trade-related job dislocations influence on their voting behavior. between 2000 and 2004 on changes in the Republican A final robustness test deals with the possibility that vote share in the previous election cycle (1996-2000). The expectation is that job dislocations that happened the electoral effect associated with trade-related job losses is due to the fact that the various measures after the elections took place will have no effect on their outcome.32 Indeed,as column(3)indicates,the correla- used to capture the "generic"(i.e.,the full,not only 33 Note that although the differences in magnitude are large,one 32 This is expected to be the case unless the layoffs are anticipated can only marginally reject the possibility that the coefficients across by voters in advance. elections are the same. 175
American Political Science Review Vol. 105, No. 1 TABLE 2. Effect of Trade-related Job Losses on Change in Republican Presidential Vote Share (2004 Level – 2000 Level) (1) (2) (3) (4) (5) (6) % Workforce applied to TAA −0.066 −0.163∗ −0.138† −0.142∗ −0.096∗ −0.091∗∗ (0.093) (0.080) (0.079) (0.062) (0.036) (0.033) Income per capita (% 2000–04) 0.053∗ 0.030 0.070∗ 0.055∗∗ 0.050∗∗ (0.025) (0.028) (0.030) (0.018) (0.017) Unemployment rate (2004) −0.077 −0.213 0.120 0.104 (0.164) (0.150) (0.120) (0.105) Unemployment (% 2003–04) −0.570 −0.372 −0.233 −0.191 (0.399) (0.352) (0.239) (0.241) Unemployment (% 2000–04) −0.407 −0.160 −0.337∗ −0.326∗ (0.311) (0.268) (0.137) (0.134) Labor force size (log) 0.073 0.342∗∗ 0.482∗∗ (0.140) (0.091) (0.104) Republican vote share (% 1996–2000) 0.148∗∗ (0.042) Constant 2.701∗∗ 30.432∗∗ 34.117∗∗ 187.069∗∗ 129.090∗∗ 108.674∗∗ (0.302) (9.505) (10.801) (31.908) (16.984) (18.699) Controls Income — X X X X X Demographics — — — X X X State fixed effects — — — — X X R2 0.001 0.084 0.116 0.332 0.637 0.647 N 3,111 3,054 3,054 3,054 3,054 3,054 Notes: TAA, Trade Adjustment and Assistance. Standard errors are corrected for clustering at the state level. An observation in the model is a county. The dependent variable is a change in support of President Bush between 2004 and 2000. †significant at 10%; ∗significant at 5%; ∗∗significant at 1%. Robustness Tests Before proceeding to the substantive interpretation of the relationship between trade-related job losses and the decrease in support for the president, I subject the results to a number of robustness tests by using a different coding of the dependent variable or a different unit of analysis as control, estimating a placebo treatment, or employing an alternative measure of shifts in the labor market. Table 3 indicates that replicating the analysis using the two-party vote share instead of the all-party vote share as the dependent variable does not produce substantially different results [column (1)]. This indicates that the findings are unrelated to the relatively successful run of the third-party candidate Ralph Nader in the 2000 elections. The results are also robust to controlling for the vote share in 2000, instead of using the change in vote share between the 2004 and 2000 elections [column (2)]. As another test of the identification strategy, column (3) presents the equivalent of a placebo treatment, examining the effect of trade-related job dislocations between 2000 and 2004 on changes in the Republican vote share in the previous election cycle (1996–2000). The expectation is that job dislocations that happened after the elections took place will have no effect on their outcome.32 Indeed, as column (3) indicates, the correla- 32 This is expected to be the case unless the layoffs are anticipated by voters in advance. tion between the trade-related job losses that occurred between 2000 and 2004 and the electoral change in the earlier elections is weak, and the parameter estimate is a good deal smaller than its standard error.33 Column (4) examines the robustness of the findings when using local media markets (DMAs) instead of states as the control unit. The presumption is that exposure to common information via media reporting may account for similar shifts in voting behavior within a media market. The model estimated in column (4) is the same as the benchmark specification but includes fixed effects for media markets instead of state fixed effects (and also clusters the standard errors by DMAs instead of by states). The results show that the addition of more than two hundred controls (the DMA fixed effects) slightly decreases the magnitude of the estimated effect associated with trade-related layoffs (from −0.091 to −0.085), although the standard error remains small and the coefficient is highly significant (p = .02). The fact that the overall fit of the model increases suggests that the shared media content to which voters in different counties are exposed may have an influence on their voting behavior. A final robustness test deals with the possibility that the electoral effect associated with trade-related job losses is due to the fact that the various measures used to capture the “generic” (i.e., the full, not only 33 Note that although the differences in magnitude are large, one can only marginally reject the possibility that the coefficients across elections are the same. 175