Theory

The study of individual trade preferences within international political economy has focused largely on using the labor market effects of economic openness to explain variations in preferences. Individuals are expected to form trade policy preferences according to the real or anticipated effects of openness on their wages and/or job security. This body of research can be roughly subdivided into three categories, based on researchers’ assumptions about the relative importance of different barriers to labor market mobility.

The Stolper-Samuelson (SS)—or Heckscher-Ohlin—model of trade opinion argues that trade preferences vary with individual factor endowments and national comparative advantage in capital or labor costs, implying that national borders constitute the only relevant barriers to labor mobility. This perspective assumes high labor mobility between sectors.Footnote ¹¹ If instead many individuals have sector-specific skills, their interests may be influenced by sector comparative advantage.Footnote ¹² This critical insight has produced the second major strand of trade opinion research: the specific factors (Ricardo-Viner, RV) models suggesting that individual trade preferences vary with sector of employment.Footnote ¹³ Recent research extends Hiscox's logic to other labor market barriers. In particular, the increasing globalization of production has led scholars to focus on occupation.Footnote ¹⁴ In this model, workers are defined by tasks they perform in the global labor market, with barriers in transferring from task to task generating trade preference cleavages.

While focusing on occupation represents a useful advance toward understanding the increasingly complex distributional implications of globalization, the task-based model nonetheless omits any consideration of firms. These models (like SS and RV models) implicitly assume that individuals are participants in a single global marketplace for labor (albeit one with national, sectoral, or occupational barriers to mobility). However, according to foundational work by Coase, by Helpman, Itskhoki, and Redding, and by Williamson, transaction costs associated with switching employers constitute important barriers to labor mobility.Footnote ¹⁵

Further, recent trade economics research suggests that the growing complexity of the global economy has created an important new cleavage in trade preferences based on heterogeneity in firm productivity. The “new new trade theory” argues that only the most productive firms will participate in and benefit from exporting,Footnote ¹⁶ importing,Footnote ¹⁷ or foreign direct investment,Footnote ¹⁸ given the high fixed costs required by these cross-border economic activities. Firm productivity has been shown to precede, rather than result from, participation in exporting or foreign investment.Footnote ¹⁹ Less productive firms instead focus on domestic markets.Footnote ²⁰ Thus economic liberalization allows productive firms to increase their market share and reduce input costs but exposes less productive, domestically oriented firms to import competition, leading to declining margins and possible market exit.Footnote ²¹ Consistent with research in economics, recent research in political science shows that firm preferences and political activities regarding economic liberalization vary with productivity.Footnote ²²

Thus we extend the logic introduced by Hiscox to argue that firms constitute an important additional dimension of cleavages separating relative winners and losers from trade.Footnote ²³ Table 1 shows how different theories about individual trade preferences—SS, RV, occupation, and firm-based models—lead to different predictions about the direction of a change in trade preferences for various combinations of skill, industry, occupation, and firm characteristics in a country, like Japan, that is relatively well endowed with high-skilled labor.

TABLE 1.

Labor market cleavages due to openness

Notes: This table assumes that the preferences described are in a country, like Japan, that is relatively well endowed with high-skilled labor. A minus sign denotes decreasing demand for trade protectionism, and a plus sign, increasing demand for protectionism.

But why should employees’ preferences over trade openness converge with their employers’? First, employees of more (less) productive firms can experience wage gains (cuts) and increased (decreased) job security on liberalization. Even when the average effect of openness on employment is negative, employees of more productive firms will be relatively less affected. This is consistent with research in trade economics showing that the primary determinant of employment dislocation due to trade liberalization is employer productivity, not skill or sector.Footnote ²⁴ Further, extensive research in labor economics has established that workers displaced by mass layoffs (for example due to import competition) suffer short- and long-term earnings losses even after securing new employment, whether in the same sector or not.Footnote ²⁵ Given the persistent negative effects of layoffs on welfare, we expect employees to be particularly interested in keeping their current jobs.

Conceivably, any apparent effect of employer productivity on employee preferences for trade openness could simply be an artifact of the well-established process of matching between employees and employers on the basis of productivity: more (less) productive employees tend to work for more (less) productive employers.Footnote ²⁶ Thus, any apparent differences in preferences over trade openness between employees of productive firms and employees of unproductive firms may be a result of individual-level rather than firm-level variation in productivity. That is, employees of productive firms may favor trade openness more than employees of unproductive firms simply because productive firms have productive employees who could benefit more from trade openness, as predicted by Stolper-Samuelson.

However, a large body of research describes two types of transaction costs leading to labor frictions, suggesting an important role for the influence of firm-level productivity on employee preferences. First, employees develop firm-specific human capital. While generic forms of human capital transfer readily between firms, firm-specific human capital, much of which is proprietary and relies on patents or trademarks, is less productive for other employers.Footnote ²⁷ Second, firms experience substantial costs from labor turnover.Footnote ²⁸ One study estimates the average direct costs to businesses of labor turnover per employee at about 20 percent of employees’ salaries.Footnote ²⁹ Moreover, firms risk losing employee-specific knowledge in the separation process. Search and matching costs for qualified replacements constitute an additional layer of expenses.

The consequence of these frictions is an imperfectly competitive labor market: individuals of comparable productivity are employed by firms of varying productivity, creating variation in wages across individuals with the same intrinsic productivity. Thus, employees of highly productive firms may receive wages above their intrinsic marginal productivity due to firm rent/profit sharing.Footnote ³⁰

Researchers have identified two related mechanisms that may underlie this process. First, firms may share their profits with workers because it will be costly to replace them due to search and matching costs, causing wages to vary with firm revenues.Footnote ³¹ Second, what is perceived as a fair wage sufficient to induce worker efforts may vary with firm profitability since it is believed that more profitable firms should pay higher wages.Footnote ³² Moreover, productive firms are willing to pay higher wages because of their productivity advantage from firm-specific knowledge (for example, technical or management know-how), implying that the cost of replacing employees should be even higher for these firms.Footnote ³³

Using employer–employee matched data in the UK and Mexico, Faggio, Salvanes, and Van Reenen and Frías, Kaplan, and Verhoogen find that workers with similar characteristics receive different wages as a result of firm-specific wage premia.Footnote ³⁴ Specifically, using UK panel data, Faggio, Salvanes, and Van Reenen find that increases in wage inequality between firms can be primarily attributed to an increase in firm productivity due to technology diffusion, rather than high-skilled (low-skilled) workers selecting into productive (non-productive) firms. Similarly, using employer–employee and plant-level data from Mexico, Frías, Kaplan, and Verhoogen show that two-thirds of the higher level of wages in more productive plants can be attributed to plant-specific variation. Thus, even given the (implausible) assumption of perfect sorting of employees into employers by ability, higher wages paid by productive firms at every level of employee productivity are sufficient to generate varying employee preferences by employer productivity alone.

Consequently, combining the “new new trade theory” with the labor market frictions logic of Hiscox yields the following insight: individuals with equivalent levels of skill and task competitiveness employed in the same industry may differ widely in their trade attitudes.Footnote ³⁵ Instead, their preferences may vary systematically with employer productivity, representing an important new cleavage in trade preferences. While beyond the scope of this paper, this suggests the need for an updated generalized political-economy model of preferences over globalization, building on Iversen and Soskice, where individual interests (for example, factor, sector, occupation, or firm based) affect attitudes conditional on the extent and locus of labor market frictions.Footnote ³⁶

While we focus here on the rent-sharing mechanism to explain how firm-level productivity can shape individual trade preferences, recent research in American politics, as well as an extensive literature in organizational psychology, suggests that there are other channels through which firms can influence employee preferences and political behavior. First, studies in American politics show that firms often actively or indirectly mobilize their employees to engage in various political actions on behalf of their preferred policy outcomes.Footnote ³⁷ Moreover, employers are likely to convey their preferences on trade policy to their employees.Footnote ³⁸ Second, a large literature on organizational identification (OI) suggests that employees gradually adopt their organizations’ views as their own.Footnote ³⁹ Future research is required to assess the extent and conditions under which each of these mechanisms operates to bring about convergence in employee–employer trade preferences.

Hypotheses

Following this logic, we argue that individuals’ trade opinion varies predictably with the international competitiveness of their employers. Because more productive firms are likely to benefit from increased trade openness and to engage in rent sharing with their employees, while less productive firms will likely suffer from increased import competition, we expect that employees of more productive firms engaged in globalized activities will be systematically more supportive of trade openness than employees of less productive firms producing for the domestic market.

Second, existing theories of trade opinion make no specific predictions about how the effects of their favored explanators (education, sector, and task routineness/offshorability) should vary by firm productivity. However, extending the logic introduced by HiscoxFootnote ⁴⁰ implies that if the distributional impact of trade openness on employment and wages is largely, or even primarily,Footnote ⁴¹ felt at the level of individual firms and determined by firm productivity,Footnote ⁴² then education, sector of employment, and occupational characteristics should have heterogeneous effects on individual trade preferences, conditional on employer productivity. Thus, we compare the empirical implications of this theory with naive extensions of alternative explanations.

First, a simple extension of the RV hypothesis would suggest that sector subsumes firm productivity as a trade preference cleavage, meaning that conditioning on sector-level competitiveness, we would not expect employees to differ systematically in their trade opinions. Instead, we hypothesize that the positive association between sector disadvantage and support for protection should obtain among the employees of less productive firms, but not among those of more productive firms.

Second, a simple extension of the occupation hypothesis would suggest that due to occupational sorting, we should not expect employees of less productive or more productive firms to differ systematically in their trade opinions once we condition on task routineness and/or offshorability. Instead, we expect that at higher levels of task offshorability, employees of globalized firms will hold particularly pro-trade opinions compared to employees of domestic firms. In addition, we expect that the positive interaction effect of routineness and offshorability on support for protection identified by Owen and Johnston will be more pronounced among employees of domestic firms than among employees of globalized firms.Footnote ⁴³

Third, a simple extension of the SS hypothesis would suggest that education— that is, individual skill or productivity—subsumes firm productivity, meaning that we would not expect systematic differences in trade opinion based on employer productivity, conditional on education. Instead, we expect that higher-skill employees of globalized firms should hold particularly pro-trade opinions, perhaps due to firm rent sharing.

Moreover, employees’ trade opinions will also likely vary by their relative position within an employing firm. We expect employees who are more likely to benefit (through rent sharing, or simply anticipated continued employment) from the success of their employers to be more favorable toward trade openness than those who are less likely to benefit. For example, managers should be more supportive of trade than ordinary employees (accounting for income and education), and permanent employees more favorable toward liberalization than temporary employees.

This hypothesis is well supported by existing theories. First, consistent with the logic of open economy politics models of individual preferences,Footnote ⁴⁴ intra-firm variations in trade opinion may arise from the anticipated distribution of benefits due to openness. Those employees best positioned to benefit from higher firm performance under openness are most likely to support liberalization. Research suggests that employees with higher status may benefit the most from rent sharing because they are likely to enjoy greater bargaining power.Footnote ⁴⁵ For example, managers often receive substantial stock-based compensation, tying their income more directly to firm performance than ordinary employees’, such that even unexpected increases in firm earnings are associated with increased executive compensation.Footnote ⁴⁶

Second, higher-status employees may identify more closely with employers’ goals. This logic draws on the extensive OI literature.Footnote ⁴⁷ OI can be seen as a mechanism of persuasion, whereby employees are gradually influenced to adopt the organization's goals (for example, its desired trade policies) as their own.Footnote ⁴⁸ Intuitively, correlates of higher employee status are reliable predictors of OI. A meta-analysis of ninety-six quantitative studies, incorporating nearly 21,000 respondents, found job type and scope of responsibilities, job satisfaction, and anticipated retention to be highly and reliably correlated with OI.Footnote ⁴⁹

Third, higher-status employees may be better informed about employers’ trade policy positions and thus more likely to be influenced by them. Employers often provide explicit trade policy cues to employees.Footnote ⁵⁰ For example, a leaked memo from Nike management encouraged employees to support TPP ratification.Footnote ⁵¹ But relative exposure may depend on employees’ positions within firms. Higher-status employees, like managers and full-time workers, likely will know more about employer trade preferences, even when these have not been explicitly announced.

Data

Evidence

We evaluate our hypotheses through a series of ordinary least squares (OLS) regressions using JGSS data.Footnote ⁶⁷ Consistent with expectations, we find that employees of globally competitive firms view trade openness more favorably. They are also more likely to support trade openness within a given sector, skill level, or level of task routineness and offshorability. Finally, the relationship of firm characteristics with employee trade preferences is moderated by relative status. Employees of globalized firms who benefit more directly from firm success (for example, managers) are particularly supportive of free trade.

Table 2 tests H1: that employees of internationally competitive firms are more likely to favor trade openness than employees of less competitive firms. Across all models, globalized employer is highly significant in the expected direction and substantively comparable to other prominent variables identified in the literature, such as skill level or sector competitiveness (Figure 1). Moreover, except for specifications including disadvantaged industry, we employ industry fixed effects (based on twenty-one industry classifications) throughout to identify within- rather than between-industry effects.Footnote ⁶⁸

FIGURE 1.

Hypothesis 1: substantive effects comparison

TABLE 2.

Firm characteristics and support for protection (OLS)

Notes: Industry-clustered standard errors in parentheses. * p < .10; ** p < .05; *** p < .01.

Model 1 shows results from our base OLS model. Consistent with H1, employees of firms engaged in globalized economic activity are significantly less likely to support import restrictions. Model 2 includes a variable indicating sector competitiveness, disadvantaged industry, and Model 3 adds a sociotropic variable, localism.

Model 4 shows that our results are robust to the inclusion of Liberal Democratic Party (LDP) support as a measure of political ideology and support for redistributive policy. Model 5 demonstrates that employer competitiveness predicts trade attitudes, even controlling for occupational (task) attributes. Model 6 provides a fuller model with variables measuring non-self-interested influences on trade opinions (sociotropic trade, cosmopolitanism index, redistribution, and job insecurity).

To facilitate interpretation of Table 2 results, Figure 1 presents comparisons of substantive effects. The y-axis shows predicted values for protection from OLS estimations of the model specification shown in models 1 and 2 of Table 2. Analogous plots showing predicted probabilities based on ordered logistic regressions are substantively identical (see the online supplement).Footnote ⁶⁹ Consistent with the results in Table 2, the observed effect of globalized employer on protection (the difference in the predicted values of protection for globalized employer and domestic employer, shown in Figure 1a) is broadly comparable in magnitude to that of college graduate (Figure 1b) and disadvantaged industry (Figure 1c).

Table 3 tests H2: The association between an individual's economic attributes (such as skill level, sector, or occupation) and trade opinion will be moderated by employer competitiveness. Model 1 tests the expectation that the frequently observed relationship between trade opinion and sector comparative advantage (workers employed in comparatively disadvantaged sectors are more likely to support protection) is conditioned by employer type. Consistent with the expectations set out in H2, Globalized employer × disadvantaged industry is negative and significant. Employees of globalized firms in disadvantaged sectors are significantly less likely to support protectionism than employees of domestic firms in similarly disadvantaged sectors. Model 2 tests the expectation that the frequently observed effect of skill (as measured by education) on trade preferences is conditioned by employer type. Consistent with H2, globalized employer × college graduate is negative and significant, suggesting that college-educated employees of globalized firms are significantly less supportive of protectionism than college-educated employees of domestic firms.

TABLE 3.

Conditional effects of firm productivity (OLS)

Notes: Industry-clustered standard errors in parentheses. * p < .10; ** p < .05; *** p < .01.

Models 3 to 6 present results from specifications based on research linking occupational characteristics to trade opinion.Footnote ⁷⁰ Using data from Autor and Dorn, we include measures for exposure to offshoring (job offshorability) and job task competitiveness (task routineness).Footnote ⁷¹ In Model 3, task offshorability is not significantly associated with support for protectionism. Nonetheless, consistent with H2, globalized employer × task offshorability is negative and significant, indicating that the observed effect of employer productivity increases with task offshorability. Following Owen and Johnston, model 4 interacts task offshorability with task routineness.Footnote ⁷² Contrary to their findings and our expectations, while employer type remains negatively and significantly associated with support for protection, task offshorability × task routineness, while positive, is not significant. Finally, models 5 and 6 show results from specifications identical to that in model 4, but stratified by globalized employer: model 5 shows results for employees of domestic firms only, and model 6 for employees of globalized firms only. Again, consistent with results from model 4, but contrary to expectations, the interaction of task offshorability and task routineness is neither significantly nor substantially associated with trade opinion regardless of respondents’ employer type.

To facilitate interpretation of the results shown in Table 3, Figure 2 presents a series of plots of substantive effects based on OLS estimates with all other covariates held at their median values. The y-axis shows predicted values for protection from OLS estimations of model specifications shown in Table 3 (see the online supplement for substantively identical plots based on ordered logistic estimations). Figure 2a shows that the effect of sector is conditioned by globalized firm employment. While employees of domestically oriented companies in comparatively disadvantaged industries are more likely to support trade restrictions compared to those in comparatively advantaged industries, within both comparative-advantage and comparative-disadvantage sector categories, employees of globalized firms are less likely to support protectionism than employees of domestically oriented firms. Moreover, we do not observe any significant difference in support for protectionism between employees of globalized firms working in comparatively advantaged and disadvantaged industries, suggesting that the effect of employment sector on trade opinion may be concentrated in employees of domestic firms. Figure 2b suggests that the effect of skill level on trade opinion is conditioned by employer type: college graduates employed by globalized firms are less likely to support protection than college graduates employed by domestic firms, while nongraduates employed by globalized firms are also less likely to support protection than nongraduates employed by domestic firms.Footnote ⁷³

FIGURE 2.

Hypothesis 2: effects of economic attributes conditioned by employer type

Figure 2c and Figure 2d depict how globalized firm employment may moderate a pair of findings described by Owen and Johnston. First, for a given level of routineness in their job tasks, individuals will be more likely to support trade protection as task offshorability increases; second, for a given level of task offshorability, predicted values of protection will increase with task routineness.Footnote ⁷⁴ They measure this relationship by interacting task routineness with task offshorability. Thus, Figure 2c and Figure 2d show substantive effects on observed levels of protection from this interaction: task routineness × task offshorability, stratified by globalized employer (models 5 and 6 of Table 3). For all levels of routineness (offshorability), support for free trade is higher among globalized firm employees than among domestic firm employees. As offshorability increases with routineness held at the sample mean, employees of domestic firms hold increasingly protectionist views. However, equivalent increases in offshorability have null (or even negative) observed effects on globalized firm employees’ support for protection. This is consistent with H2: The association between an individual's economic attributes (for example, skill level, sector, or occupation) and trade opinion will be moderated by employer competitiveness. However, contrary to expectations, increases in routineness (offshorability held at its mean) have null effects on protectionist views for employees of both domestic and globalized firms, although employees of the latter are consistently less supportive of protectionism at every level of task routineness.

Table 4 presents tests of H3: The effect of employer competitiveness is conditioned by the extent to which employees identify their interests with their employers’. Results from model 1 show that globalized firm employees who feel insecure in their employment are significantly more protectionist. In model 2, the coefficient for the interaction term between temporary worker and globalized employer is positive, as predicted by H3, but not statistically significant. In contrast to models 1 and 2, model 3 examines the trade opinion of employees expected to identify more strongly with their employers’ interests: managers. Consistent with H3, globalized employer × management is negative and significant, suggesting that managers at globalized firms are distinctively less supportive of import restrictions.Footnote ⁷⁵ While non-management employees of globalized firms remain significantly less supportive of protection than employees of domestic firms, managers at globalized firms are particularly distinctive in their support for free trade.Footnote ⁷⁶

TABLE 4.

Employee status and firm productivity (OLS)

Notes: Industry-clustered standard errors in parentheses. * < .10; ** < . 05; *** < . 01.

Predicted values based on OLS estimations shown in Table 4 are depicted in Figure 3 and are consistent with H3.Footnote ⁷⁷ The observed effect of globalized firm employment is conditioned by employees’ relative position within firms. In globalized firms, relative “winners”—those who are more likely to benefit from employer success (for example, permanent employees and managers)—are distinctively less supportive of import restrictions, while relative “losers”—those less likely to share in employers’ success—express preferences similar to those of employees of domestically oriented firms.

FIGURE 3.

Hypothesis 3: intra-firm variation

For both theoretical and empirical reasons, it is unlikely that managers’ strong support for trade is an artifact of higher relative labor market mobility. Theoretically, highly skilled or higher-status employees are more likely to develop firm-specific skills, reducing inter-firm mobility.Footnote ⁷⁸ Empirically, in the JGSS data, college graduates and nongraduates have virtually identical job tenures for any given age cohort, while managers have notably longer job tenures than nonmanagerial employees.

Robustness Checks

Our choice to focus our analysis on Japan is motivated by data availability, rather than theoretical necessity or substantive focus. Still, generalizations from our findings should be viewed as provisional. In 2007, the average job tenure in Japan (10.88 years) was slightly above the OECD average (9.59 years), which raises the possibility that Japan is a “likely” case for our argument since employees may have stronger propensities to adopt employers’ trade preferences given an expectation of longer job tenure.Footnote ⁷⁹ Moreover, the Japanese labor market may be distinctive in ways that go beyond job tenure.

However, what we learn from Japanese data may extend to other middle- and high-income economies, since employee preferences in these countries are likely to be influenced by employer interests as long as (1) most workers are not self-employed, (2) transaction costs associated with switching employers create some barriers to labor mobility, and (3) firms vary systematically in their interests over openness (according to relative productivity). Admittedly, our study cannot show direct evidence for that extension, and future research is required to further evaluate the hypotheses in other political and economic settings.

A similar concern arises from the temporal coincidence of the JGSS 2008 survey with the global financial crisis. However, while Japan's economy was undoubtedly affected by the crisis, these effects were most pronounced among larger, export-oriented firms.Footnote ⁸⁰ Consequently, the idiosyncratic labor market effects of the global financial crisis are likely to bias against our findings by rendering employment less secure for employees of globalized firms, making them less likely to identify their interests with their employers’ compared to employees of domestically oriented firms.Footnote ⁸¹

We also conduct a series of robustness checks to address concerns that employees of globalized firms are distinctive in some way that is both unaccounted for by our theory and systematically related to trade opinion (Table A.13 in the the online supplement). While these tests cannot be fully dispositive, they increase our confidence in the robustness of our core results. First, we include specifications designed to control for the effect of firm size (apart from involvement in globalized economic activities). Moreover, we control for geographic clustering of globalized firms (for example, in the greater Tokyo area), which may affect local economic interests,Footnote ⁸² by including prefecture-level measures of total factor productivity (TFP), labor productivity, and TFP measured at the industry-prefecture level. These data are drawn from the Regional-Level Japan Industry Productivity database.Footnote ⁸³ To match JGSS data, we use 2008 data on productivity for forty-two prefectures and twenty-three industries (ten manufacturing and thirteen nonmanufacturing industries).

While our empirical results are robust to controlling for observable factors that have been found to predict both individual trade preferences and place of employment, there may be unobservable factors leading individuals to select into employment at productive firms. For example, globalized employer may be capturing unobservable employee competitiveness or specialization not explained by education-, sector-, or task-based measures of competitiveness. While our research design does not allow us to fully address this concern, we adopt the sensitivity test introduced by Altonji, Elder, and Taber using selection on observables to calculate potential bias from unobservables.Footnote ⁸⁴ By calculating how much stronger selection on unobservables has to be relative to selection on observables to fully account for the pro-trade effect of globalized firm employment, sensitivity tests let us evaluate the seriousness of threats to inference arising from unobservable confounders—conditional on the assumption that relationships between observable variables are informative about unobservables.

This measure is calculated as $\widehat{{\beta ^{\rm R}}}-\widehat{{\beta ^{\rm F}}}$ using estimated coefficients for the main explanatory variable from a model with restricted ($\widehat{{\beta ^{\rm R}}}$) and full sets ($\widehat{{\beta ^{\rm F}}}$) of covariates, respectively. Higher values indicate that selection on unobservables must be much stronger relative to observables to account for the entire effect. Intuitively, the smaller $\widehat{{\beta ^{\rm R}}}-\widehat{{\beta ^{\rm F}}}$, the less the estimated effect is affected by selection on observables, and the stronger selection on unobservables must be relative to observables to account for the entire effect. Moreover, as $\widehat{{\beta ^F}}$ increases, the observed effect becomes less likely to be explained by selection on unobservables alone.

Table 5 presents a series of sensitivity analyses in which $\widehat{{\beta ^{\rm F}}}$ from three separate full models is compared to $\widehat{{\beta ^{\rm R}}}$ from a bivariate model.Footnote ⁸⁵ Our first full model replicates model 1 of Table 2. We also consider a second full model including nonmaterial controls (replicating model 6 of Table 2) and a third including task-related variables. We find it is unlikely that the observed effect of globalized employer can be entirely attributed to unobserved confounders. The Altonji ratios range from approximately 2.5 to 5.8, so the effect of unobservables would have to be 2.5 to 5.8 times the effect of observables in our model to explain the observed effect of employer competitiveness on trade opinion.

TABLE 5.

JGSS sensitivity analyses

However, Chaudoin, Hays, and Hicks suggest this approach may artificially inflate calculated sensitivity ratios by decreasing $\widehat{{\beta ^{\rm R}}}-\widehat{{\beta ^{\rm F}}}$ by including covariates orthogonal to selection in the sensitivity analysis.Footnote ⁸⁶ Thus, we also present theoretically informed sensitivity ratios (implied Altonji ratio) in Table 5, calculated using Chaudoin, Hays, and Hicks’s POET software. According to Chaudoin, Hays, and Hicks, we should have confidence that the observed effect of employer competitiveness on trade opinion is robust if the implied Altonji ratio is greater than 1.

The negative implied Altonji ratio of –1.53 shown for the base model specification in Table 5 indicates that the effect of unobservables must be proportionally greater in the opposite direction to account for the observed effect.Footnote ⁸⁷ This is encouraging because unobserved factors (for example, latent productivity not captured by education) sorting employees into globalized firms seem unlikely to make them less likely to support trade. All together, while sensitivity tests and other diagnostics alone cannot rule out alternative explanations, results from multiple sensitivity analyses increase our confidence that our results are not entirely driven by the presence of unobserved confounders.