Industry, Self-Interest, and Individual Preferences over Trade Policy. Sungmin Rho Stanford University

Transcription

1 Industry, Self-Interest, and Individual Preferences over Trade Policy Sungmin Rho Stanford University Michael Tomz Stanford University December 2012 Preliminary. Comments welcome! Abstract: Do voters have economically self-interested preferences about trade policy? Despite a proliferation of research on this question, no consensus has emerged. We argue that scholars can gain new insight by analyzing public attitudes toward protectionism for specific industries, instead of looking at sentiment toward free trade in general. Accordingly, we develop industryspecific measures of protectionism, incorporate them into original public opinion polls, and use the data to test several economic theories. We find surprisingly little evidence that the preferences of citizens fit the predictions of standard models, including Stolper-Samuelson, Ricardo-Viner, and new new models of trade with heterogeneous firms. These findings compel us to rethink the sources of public opinion about trade policy. Note: Prepared for the 2013 International Political Economy Colloquium, Goa, India. We thank Adam Berinsky, Bobby Gulotty, and Lisa Martin for helpful comments on an earlier version of this paper.

2 1. Introduction A large and growing literature examines how ordinary citizens think about trade policy. Scholars have investigated why some individuals favor free trade, whereas others support protectionism. The central controversy in this literature concerns the explanatory power of economic self-interest. To what extent do citizens have economically self-interested preferences about trade? Most studies on this topic have considered two economic theories. The first predicts that the costs and benefits of free trade for an individual depend on which factors of production they own. The second assigns greater importance to the individual s industry, rather than his or her factor endowments. These theories typically go by the names Stolper-Samuelson and Ricardo- Viner, in recognition of the economists who contributed to their development. Since these theories generate distinct hypotheses about who would gain or lose from free trade, researchers have asked how well each theory explains existing patterns of public opinion. Despite considerable effort, though, no scholarly consensus has emerged. Some analyses of polling data from the U.S. and other countries find support for both theories, others conclude in favor of one theory, and still others reject both (e.g., Scheve and Slaughter 2001; Beaulieu 2002; O Rourke and Sinnott 2002; Hays, Ehrlich, and Peinhardt 2005; Mayda and Rodrik 2005; Hainmueller and Hiscox 2006; Hiscox 2006). Thus, it is remains unclear whether and under what conditions we can connect the trade preferences of individuals with their economic interests. Moreover, while numerous studies have focused on Stolper-Samuelson and Ricardo-Viner, practically none have tested the explanatory power of other theories, including the new-new trade theory (Melitz 2003), which emphasizes that highly productive firms stand to gain the most from international trade, whereas less productive firms would wither without protection. We argue that scholars can gain new insight about all these theories by analyzing public attitudes toward protectionism for specific industries, instead of looking at sentiment toward free trade in general. Accordingly, we develop industry-specific measures of protectionism, incorporate them into original public opinion polls, and use the data to test the explanatory power not only of Stolper-Samuelson and Ricardo-Viner, but also of new-new trade (NNT) theories. We find surprisingly little evidence that these theories help explain public opinion about free trade. These findings compel us to rethink why some citizens favor free trade, whereas others oppose it. 2. Three Theories about Economic Self-Interest and Trade In this section, we briefly review three theories about effects of trade on individual income, and discuss the type of public opinion data scholars have used for empirical analysis. The Stolper-Samuelson Theory The Stolper-Samuelson theorem posits that an increase in the relative price of a product will increase the returns to the factor that is most intensively used to make that product, while decreasing the real earnings of other factors of production. The theorem rests on several assumptions, including perfect competition, constant returns to scale, and complete inter-industry mobility of the factors of production. When those assumptions do not hold, changes in relative 1

3 prices might affect people differently, even when they own the same factor of production. Despite these limitations, the theorem remains a touchstone for research about the political economy of protectionism. When a government protects an industry from foreign competition, its policies (such as tariffs and quotas) typically increase the relative price of the protected industry s products. For instance, if the government limited imports of capital-intensive products, this would raise the relative price of capital-intensive products and, according to Stolper-Samuelson, increase the returns to capital while depressing real wages of labor. Similarly, placing limits on imports of labor-intensive products would raise the relative price of labor-intensive products, which would raise the real wages of workers while lowering real rents for capitalists. 1 Thus, the theorem predicts that capitalists should favor limits on imports of capitalintensive products, whereas labor should oppose such a policy. Mussa (1974) provides an example of this prediction: The [Stolper-Samuelson] theory suggests that if automobile manufacturing is capital intensive, then General Motors will favor increases in the tariff on automobiles and the United Auto Workers will oppose such increases; and, if textile manufacturing is labor intensive, then textile workers will favor quotas on textile imports, and domestic textile manufacturers will oppose them. Most studies of individual attitudes toward trade policy have focused on a different factor of production: educated labor. Some workers are highly educated whereas others are not, and some products rely on educated labor whereas others do not. If educated labor is an important factor of production, then workers without much formal education should favor policies that increase the price of low-education products (items that are made with unskilled labor). At the same time, they should oppose policies that raise the price of education-intensive products, since such policies would impose pain by making the products they buy more expensive. Highly educated workers should have the opposite preferences: favoring policies that raise the price of high-education products, while opposing policies that make low-education products more expensive. The Ricardo-Viner Theory The Stolper-Samuelson theorem assumes that factors of production are completely mobile across industries, but the Ricardo-Viner theorem posits that at least one factor is immobile. The theorem suggests that an increase in the relative price of a product increases the returns to all immobile factors that are used to make that product, while decreasing the returns to all immobile factors that are used to make other products. 1 Jones and Scheinkman (1977) and Ethier (1982) together show that the results from the traditional 2 x 2 model of the Stolper-Samuelson theorem generally apply to higher dimensions. They prove (1) that some factors stand unambiguously to gain and to lose from the price chances caused by protection ; and (2) that the general pattern of gains and losses is related to the intensities with which factors are used in production of the goods (Deardorff 1994, p.19). However, the theorems say little about how to empirically identify winning and losing factors in a multi-dimensional model. For a succinct summary of higher dimensional models, see Deardorff (1994). 2

4 When placed in the context of international trade, the theorem generates the following prediction: if capital is immobile, then import barriers will increase the real returns to the owners of capital used to make that product, while decreasing the real returns to capital that is used for other products. Capitalists should, therefore, want to protect their own industry, while opposing protective measures for other industries. If workers cannot easily switch from one industry to another, their preferences should exhibit similar patterns: they should seek to protect their own industry, while opposing tariffs and non-tariff barriers for other industries. 2 New-New Trade Theory Since the pioneering work of Melitz (2003), the scholarly debate on effects of trade on income has focused on trading behavior of individual firms and within-industry reallocations of resources. Building on the established empirical link between firm productivity and intraindustry reallocations, Melitz provides a general equilibrium model with heterogeneous firms and shows that trade policy liberalization induces the more productive firms to export, the less productive firms to produce only for the domestic market, and the least productive firms to exit the industry. His work highlights that trade liberalization might create winners and losers within each industry. Subsequent studies, including Melitz and Ottaviano (2008), provide additional insight about whether individuals would favor or oppose unilateral trade policies such as tariff protection. Melitz and Ottaviano argue that asymmetric trade liberalization results in welfare gains for the liberalizing state in the short-run, since increased competition in the importcompeting sectors enhances industry-wide productivity. This theory yields a prediction that an individual employed in less productive firms might be more likely to support restrictive policies on imports in his own industry, since his firm might be able to operate in a less-competitive environment as the result of import restriction, at least in the short run. This prediction is bolstered by empirical studies about the effects of antidumping duties on industry productivity. Konings and Vandenbussche (2008), for example, show that antidumping protection generated different effects across individual firms in EU, depending on the firm s initial productivity: firms with low initial productivity experienced an increase in productivity, while more productive firms experienced productivity losses during protection. Likewise, based on the U.S. plant-level data in the manufacturing sector, Pierce (2011) concludes that antidumping duties lower industry-wide productivity by helping less productive firms to survive. In summary, the NNT theory emphasizes that the economy is composed of heterogeneous firms, which vary in their level of productivity. The most highly productive firms benefit from free trade, whereas less productive firms do better with protectionism. One might, therefore, 2 If workers are mobile, then the effect of trade policy on workers is ambiguous. The level of real wages will depend on their consumption preferences, since the real wages of workers will fall with respect to purchases of that product, but increase with respect to purchases of other products. Those who have a relatively high demand for the product might experience welfare loss, thus opposing such a policy. Therefore, the level of support for placing limits on a product should be determined by industry of employment for the immobile factors, and by consumption preferences for the mobile factors. For a more detailed explanation, see Jones (1971). 3

5 hypothesize that an individual s desire to protect their own industry should be negatively correlated with the productivity of the firm where the individual works. 3 Appraisal of Existing Evidence The most commonly used surveys, such as International Social Survey Program (ISSP), the World Values Survey, and the National Election Survey (NES), ask only about trade in general, using abstract terms such as free trade, instead of measuring protectionist sentiments about specific industries. As noted earlier, this research has failed to produce a consensus about whether the trade policy preferences of ordinary citizens are consistent with economic selfinterest. In this paper, we aim to advance the debate by developing industry-specific measures of support for protectionism and using them to assess the explanatory power of the Stolper- Samuelson, Ricardo-Viner, and NNT frameworks. Recent work by other scholars illustrates the potential payoffs from using specific, rather than generic, measures of economic policy. For example, Lü, Scheve, and Slaughter (2012) carried out an innovative survey about trade policy preferences in China and the United States. Although their study did not name specific industries, it did ask respondents to consider protection for industries with varying wage levels. Their unique showed that inequity aversion influences trade policy. Likewise, Malhotra, Margalit and Mo (forthcoming) adopted an industry-specific approach when studying immigration. They sampled respondents in the high-technology industry and asked their opinions about a U.S. policy to increase the number of H-1B visas. Since most of H-1B visa holders are Indians working in high-technology industries, this question essentially measured people s attitudes toward immigration of high-skilled workers. The authors found that employees in the high-technology industry were not especially negative about immigration in general, but were strongly opposed to increasing the number of H-1B visas. The authors conclude that lack of support for the industry-based logic may have more to do with the data (i.e. the samples) used to test the competing hypotheses than with the actual merits of the argument (pp.32-33). In a similar spirit, we argue that scholars can gain new insights by studying protectionism for specific industries. 3. Measuring Preferences for Protection Sample characteristics We designed and fielded an original survey about U.S. public attitudes toward protectionism. The survey was fielded in two waves. The first, in August 2011, involved 496 respondents; the second, in November 2012, yielded an additional 649 respondents. Both waves were administered over the internet to a diverse sample of U.S. adults. We recruited respondents by inviting members of Amazon Mechanical Turk (MTurk), an online service, to take a 10 3 This hypothesis assumes that the labor market is at least partially sticky, such that workers cannot transition easily from failing firms to highly productive firms in the own industry or other industries. 4

6 minute survey in exchange for $1.50. Our invitation did not indicate the topic of the survey, so respondents could not have selected into the study on the basis of their substantive interests. 4 A growing number of academics have used MTurk to conduct surveys and implement experiments. As a consequence, we now have detailed information about the demographic characteristics of the MTurk community. In general, members of MTurk are younger, more likely to be female, more highly educated, and more liberal than the national population (Berinsky, Huber, and Lenz 2012). Our data exhibited these same patterns. Although respondents ranged in age from 19 to 82, the mean was 34 years. Approximately 56% of respondents were female and 47% had college degrees. Finally, 52% classified themselves as liberal, 26% called themselves moderates, and 22% said they were conservative. Nevertheless, we regard the data from MTurk as a useful starting point for two reasons. First, our sample exhibits excellent variation on several demographic variables (including education) that are central to the theories we are testing. Second, as shown later in the paper, we successfully replicated several key findings in the literature, notwithstanding demographic differences between MTurk subscribers and the national population. Our replications confirm that samples from MTurk can be highly informative. Ultimately, though, we urge readers to treat the findings in this paper as preliminary. In future research we plan to assess the robustness of our findings by interviewing more representative samples of adults in the United States and other countries, including India. Preferences by industry In both waves, we asked whether respondents supported or opposed protection for specific types of industries. We placed these industry-specific questions early in survey to avoid any risk that other items would color the responses. We also took care not to frame the issue by asserting that protection would save jobs, raise prices, or affect respondents in other ways. Instead, our prologue simply stated that U.S. businesses and consumers buy many products that are made in foreign countries. The products from foreign countries are called imports. There is much debate about whether the U.S. government should use laws to limit imports by U.S. businesses and consumers. We then subdivided the economic landscape in several ways and included questions about each subdivision. First, we classified businesses based on the factors of production they employ. Consistent with other work, we treated labor as a heterogeneous input and distinguished several levels of education, or skill intensity, in the production process. Respondents read that Foreign countries have many types of businesses, which vary in their use of workers with college degrees. We mentioned three categories of foreign businesses those employing a low, medium, or high percentage of workers with college degrees and asked whether the U.S. government should or should not limit imports from each type of business. Second, we classified business according to their output. We gave participants in the 2011 survey a list of seven product categories: cars, cell phones, clothing, computers and software, fruits and vegetables, furniture, and medicines. As discussed in detail below, we selected these products not only because of their importance for the economy, but also because they require different levels of education to produce. For example, fruits and vegetables can be 4 Respondents could have quit the study after learning that it focused on economics, trade, and politics. In fact, though, less than a dozen people stopped before completing the study. 5

7 grown and processed by workers with little education, whereas computers, software, and pharmaceuticals require highly educated workforces. We listed the categories in random order and asked: in general, please say whether the U.S. government should or should not limit imports of each of these types of products. Participants in the 2012 survey considered the same seven goods, plus four types of services: business consulting, customer service, data entry, and typing services. They, too, indicated whether they wanted to protect those industries or not. Finally, we included a branched series of questions to find out whether respondents wanted protection for their own industry. The series began by asking whether the respondent was currently employed or had ever been employed. Those with current or previous jobs were instructed to describe, in no more than two words, the type of product or service offered by the place they worked most recently. Later, they were reminded: You said that the business where you work now [or where you worked most recently] sells X, where we replaced the variable X with the two-word description the respondent had supplied. Could U.S. businesses or consumers import that kind of product or service from foreign countries? Those who answered yes, implying that their industry produced tradable goods or services, were then asked, Do you think the U.S. government should or should not use laws to limit imports of X? where once again we replaced X with the respondent s pithy description. In summary, we measured each person s preference about protection for various industries, disaggregated in three ways: by inputs (level of education needed for production), outputs (the type of good or service produced), and employment (whether the respondent worked in that industry). Below, we use these measures to assess whether respondents have economically self-interested preferences over trade. Preferences in general For comparability with previous research, we also included two generic questions about trade policy. These items tapped sentiment about trade in general, without referring to particular products or industries. Our first generic question measured support for new limits on imports. Some people have suggested placing new limits on foreign imports in order to protect American jobs. Others say that such limits would raise consumer prices and hurt American exports. Do you favor or oppose placing new limits on imports, or haven't you thought much about this? This question, which Scheve and Slaughter (2001) analyzed in their seminal study of individual trade policy preferences, comes from the U.S. National Election Study. We created a variable called limit imports that took a value of 1 if the respondent favored new limits on imports, was 0 if they opposed new limits on imports, and was missing if respondents said had not thought much about the issue. The second question asked whether the U.S. should encourage or discourage international trade. As you may know, international trade has increased substantially in recent years. This increase is due to the lowering of trade barriers between countries, that is, tariffs or taxes that make it more difficult or more expensive to buy and sell things across international borders. Do you think the government should try to encourage international trade or to discourage international trade? Do you think the government should encourage/discourage international trade a lot, or only a little? Mansfield and Mutz (2009) used this item when comparing the explanatory power of self-interest versus sociotropic and cultural variables. Based on this question, we created an ordinal variable called discourage trade that was 1 if the government 6

8 should encourage trade a lot, 2 if the government should encourage trade a little, 3 if the government should discourage trade a little, and 4 if the government should discourage trade a lot. 4. An Analysis of General Attitudes toward Trade To validate our sample and set the stage for subsequent analysis, we investigated how closely we could reproduce the major findings in the literature. Many authors have found a positive relationship between education and support for free trade (e.g., Scheve and Slaughter 2001; Mayda and Rodrik 2005). To measure education, we asked each respondent for the highest level of education they attained and sorted their answers into four categories: no college, some college, college degree, or advanced degree. For some analyses we created a unit variable, education, which was 0 if the respondent had not gone to college, 1/3 if they had some college education, 2/3 if they had earned a college degree, and 1 if they had attained an advanced degree. For other analyses we included dummy variables for some college, college degree, and advanced degree, while treating no college as the omitted reference category. Our survey also asked about other demographic variables that might be correlated with education. Based on the responses, we constructed variables for gender (female or not), age, and household income. Although our raw data contained age in years and income in dollars, we divided age by 10 and income by 100,000 to make the statistical tables easier to read. Thus, in our analysis, age is measured in decades and income is measured in hundreds of thousands of U.S. dollars. Previous research suggested that union membership, employment status, and party affiliation might also be important predictors of attitudes toward trade. Thus, we created dummy variables for whether the respondent belonged to a union and whether the respondent was unemployed. Our measure of party identification ranged from 0 if the respondent called themselves a strong Democrat to 1 if the respondent called themselves a strong Republican. Following Mansfield and Mutz (2009), we also constructed three cultural variables. The first, isolationism, summarized the respondent s level of agreement or disagreement with each of the following five statements. The U.S. needs to play an active role in solving conflicts around the world. The U.S. government should just try to take care of the well-being of Americans and not get involved with other nations. It is essential for the United States to work with other nations to solve problems, such as overpopulation, hunger, and pollution. It will be best for the future of the country if we stay out of world affairs. The United States has the responsibility to play the role of world policeman, that is, to fight violations of international law and aggression wherever they occur. We coded each item such that larger values represented more isolationist preferences, standardized each item by subtracting its mean and dividing by its standard deviation, and measured isolationism as the average of the five. In a similar way, we measured nationalism by scaling responses to three items from Mansfield and Mutz (2009): In the United States, our people are not perfect, but our culture is 7

9 superior to others ; I would rather be a citizen of America than of any other country in the world ; and The world would be a better place if people from other countries were more like Americans. Finally, we measured stereotypes of three racial groups whites, blacks, and Hispanics by asking to what extent members of each group were hard working versus lazy, efficient versus wasteful, and trustworthy versus untrustworthy. The variable ethnocentrism summarized how much more favorably respondents perceived their own racial group than other racial groups. We calculated this variable for white, black, and Hispanic respondents. We could not calculate it for other respondents, because we did not know how they perceived their own racial group. The results, shown in Table 1, are quite consistent with previous research. Columns (1) and (2) summarize the estimates from a logistic regression in which the explanandum was limit imports, a binary dependent variable. Columns (3) and (4), on the other hand, present an ordered logit model in which the attitude of interest, discourage trade, had four levels. All four columns confirm that protectionism declines with education. When we measured education as a continuous variable, ranging from 0 to 1, the estimated coefficient was negative and several times larger than its standard error. When we replaced the continuous measure of education with a series of dummy variables, we again found negative effects, which were most pronounced among respondents with college and advanced degrees. Several other variables in the model performed as expected. Consistent with previous studies, support for protectionism was higher among females than among males, and higher among isolationist respondents than among people who thought the United States should play an active role in world affairs. Protectionist sentiment also increased with age and was stronger among members of labor unions. The remaining variables in Table 1 had insignificant or unstable effects. For instance, income exerted no effect on the desire to limit imports (columns 1 and 2), but did suppress the tendency to discourage trade (columns 3 and 4). Neither nationalism nor party identification was statistically significant at conventional levels. Finally, when we included ethnocentrism in the model (not shown), the effect was statistically indistinguishable from zero. As noted earlier, though, ethnocentrism was only observed for white, black, and Hispanic respondents. Including ethnocentrism reduced the sample size without increasing our explanatory power or noticeably altering the estimates for other key variables. For this reason, we omitted ethnocentrism from the models in Table 1. 8

10 Table 1: Multivariate analysis of general protectionist sentiment Favor new limits Should discourage Characteristics of on imports international trade U.S. respondents (1) (2) (3) (4) Education (0.29) (0.20) Some college (0.25) (0.18) College degree (0.27) (0.19) Advanced degree (0.32) (0.22) Female (0.16) (0.16) (0.12) (0.12) Age (0.07) (0.07) (0.05) (0.05) Income (0.20) (0.21) (0.15) (0.15) Union member (0.35) (0.35) (0.22) (0.22) Unemployed (0.21) (0.21) (0.14) (0.15) Party ID (0.26) (0.26) (0.19) (0.19) Isolationism (0.11) (0.11) (0.08) (0.08) Nationalism (0.10) (0.10) (0.07) (0.07) Sample size Estimator Logit Logit Ologit Ologit Note: Columns (1) and (2) present parameter estimates and standard errors from logistic regressions in which the dependent variable 1 if the respondent wants new limits on imports and 0 otherwise. Columns (3) and (4) give parameter estimated and standard errors from ordered logistic regressions in which the dependent takes four values, ranging from 1 if the respondent strongly wants to encourage international trade, and 4 if the respondent strongly wants to discourage international trade. To save space, we do not list the intercept in models (1) and (2) and the three ancillary cutpoints in models (3) and (4). 9

11 5. Using Industry-Specific Measures to Assess Stolper-Samuelson In the previous section we analyzed public sentiment about trade in general. Consistent with earlier work, we found a negative correlation between the education of respondents and their overall desire to discourage trade. Previous studies have interpreted this correlation as evidence that U.S. citizens have economically self-interested preferences that are broadly consistent with Stolper-Samuelson. In this section we use new data to assess the accuracy of that interpretation. Do Citizens Want Protection for Their Factor of Production? The Stolper-Samuelson framework says that an increase in the relative price of a product should benefit the factor that is intensively used to make the product. If protectionism increases the relative price of the protected product, then people who are driven by economic self-interest should favor protection for industries that use their own factor intensively, while opposing protection for industries that use other factors intensely. To the extent that educated labor is a key factor of production, U.S. adults who never attended or completed college should be most enthusiastic about limiting imports from industries that rely on low-educated workers. In contrast, respondents with college or advanced degrees should prefer to restrict imports in sectors that employ a high proportion of college graduates. 5 To test these predictions, we divided our respondents into four educational categories: no college (13% of the sample), some college (40%), college degree (32%), and advanced degree (14%). Within each subgroup we noted how many respondents wanted to restrict imports from foreign companies that employ a low share of workers with college degrees, and compared the number who wanted to restrict imports from firms with medium or high shares of collegeeducated workers. Table 2 presents the results. Consider the first column, which summarizes the preferences of respondents who never attended college. Within that group, 64.1% wanted to limit imports made by low-educated foreign workers, whereas only 42.5% wanted to restrict imports made by highly educated foreign workers. The difference is nearly 22 percentage points, and the low/high gradient (the ratio of protectionist sentiment for low-education versus high-education industries) is 1.5. This pattern seems consistent with Stolper-Samuelson: low-educated workers appear to be promoting their economic self-interest by disproportionately blocking imports from foreign firms that use their own factor of production. 5 As in previous research that assesses the fit between Stolper-Samuelson and public opinion, we assume that foreign and domestic firms use educated and uneducated labor in similar ways to produce similar products. 10

12 Table 2: Percent of U.S. respondents who want to limit imports from foreign industries that utilize low, medium, or high levels of education Education level in Education of U.S. respondent the foreign industry No college Some college College degree Advanced degree Low Medium High Low High (10.7 to 32.5) (15.2 to 27.8) (11.0 to 25.0) (6.2 to 26.8) Low/High (1.2 to 1.9) (1.4 to 1.8) (1.3 to 1.8) (1.2 to 2.1) Observations Note: Each column gives the percentage of respondents who wanted to limit imports of products from foreign companies that employ a low, medium, or high percentage of workers with college degrees. The row labeled low-high gives the difference between low and high, and the row labeled low/high gives the ratio of low to high. 95% confidence intervals appear in parentheses. However, the remaining columns of Table 2 contradict the hypothesis that Americans are protecting their own educational group. It turns out that all types of U.S. citizens, including those with college and advanced degrees, want considerably more protection for low-education industries than for high-education ones. In all four columns, the difference exceeds 16 percentage points and the low/high gradient is at least 1.5. Moreover, the differences and gradients do not vary significantly by the education of the respondent. It would be difficult to reconcile this pattern with the Stolper-Samuelson model, as traditionally understood. Multivariate analyses reinforce this conclusion. Table 3 presents three logistic regressions, which analyze the desire to limit imports from three types of foreign firms: those employing a low, medium, or high percentage of workers with college degrees. Even after controlling for the same factors as in Table 1, the coefficient on education is negative and highly significant in all three models. Thus, other factors equal, educated Americans prefer free entry not only for products made by low-educated foreigners, but also for products made by highly educated foreigners. 11

13 Table 3: Multivariate analysis of desire to limit imports from foreign industries that utilize low, medium, or high levels of education Education level in the Characteristics of U.S. respondents Low foreign industry Medium High Education (0.22) (0.22) (0.23) Female (0.12) (0.13) (0.13) Age (0.05) (0.06) (0.06) Income (0.16) (0.16) (0.17) Union member (0.26) (0.25) (0.25) Unemployed (0.15) (0.16) (0.16) Party ID (0.21) (0.21) (0.21) Isolationism (0.09) (0.09) (0.09) Nationalism (0.08) (0.08) (0.08) Sample size Estimator Logit Logit Logit Note: The table presents parameter estimates and standard errors from three logistic regressions. Each regression is based on 1,145 observations. Although Tables 2 and 3 contradict factor-based theories, they also reaffirm the conventional wisdom that protectionism declines with education. Previous researchers had documented this correlation by analyzing generic measures of protection. We show that the same pattern holds after disaggregating foreign products by their skill intensity. In each row of Table 2, for example, support for protection is highest among respondents who never went to college, lower among people with some college experience, even lower among people who graduated from college, and lowest among respondents with advanced degrees. Thus, previous researchers correctly identified a negative correlation between education and protectionism. However, this correlation does not seem inspired by a desire to maximize returns to one s own factor of production. 12

14 Do Citizens Have Different Preferences for Different Industries? For further insight we examined whether respondents expressed different policy preferences for different industries. In factor-based theories such as Stolper-Samuelson, respondents should not want free trade or protectionism across the board. Instead, recognizing that different products require different skills, each respondent should favor protection for products that rely intensively on their own factor of production, while opposing protection for products that utilize other factors. Do the data support this prediction? Recall that respondents were asked whether the U.S. should limit imports from three types of foreign businesses: those employing a low, medium, or high share of workers with college degrees. A majority, 57%, gave identical answers to all three questions (24% felt the government should block all three types of imports, and 33% wanted all three to enter the U.S. without restrictions). Only a minority, 43%, said the U.S. should limit imports from some educational categories but not from others. Thus, a majority of respondents failed to discriminate based on the skill intensity of foreign products. Next, we examined the 43% who expressed differentiated policy preferences. Did college graduates oppose imports from high-education firms but not low-education firms, and did people without college degrees voice the opposite preferences? Here, too, theory and data diverged. Only 55% of respondents with differentiated preferences sought protection for products in their own educational category while opposing protection for products at the opposite end of the educational spectrum. Putting these facts together, x 0.55 = 76% did not exhibit the kinds of nuanced preferences one would expect under Stolper-Samuelson. For additional leverage we analyzed what respondents said about various output categories. Within the Stolper-Samuelson model, an individual s trade policy preferences should be homogenous for all products with similar factor inputs. If, for example, respondents want free trade for some educationally intensive products, they should prefer free trade for other educationally intensive products, as well. We tested this prediction by analyzing attitudes toward seven products cars, cell phones, clothing, computers, fruits and vegetables, furniture, and medicine that vary in their reliance on college-educated workers. To quantify the variation, we compiled data from the U.S. Current Population Survey (CPS), which is administered every month to a sample of approximately 150,000 households. We pooled all monthly surveys from January 2006 through June 2011 and retained the 600,000 respondents who worked in agriculture, mining, or manufacturing. The share of college-educated workers was lowest in fruits and vegetables, clothing, and furniture. Reliance on college-educated workers was higher in automotives and cell phones, and highest in computers and medicines (left half of Table 4). Together, these seven categories employed around 130,000 of the 600,000 respondents, or roughly 22 percent of all employees in agriculture, mining, and manufacturing. 13

15 Table 4: Actual and Perceived Percentage of Workers with College Degrees Actual Percentage Perceived Percentage Product % N Low Medium High Fruits and vegetables , Clothing , Furniture , Cars , Cell phones , Computers and software , Medicines , Note: Actual percentages are based on data from the U.S. Current Population Survey. Perceived percentages are based on 1,145 observations per product. We asked respondents about the educational intensity of each industry. Below, we wrote, we have listed several kinds of products that are made in the United States. For each product, indicate whether U.S. businesses that make the product employ a low, medium, or high percentage of workers with college degrees. The right half of Table 4 reveals that respondents had fairly accurate perceptions. Most put fruits and vegetables, clothing, and furniture in the loweducation category; a majority put cars and cell phones in the medium-education category; and most regarded computers and medicines as high-education industries. Next, we clustered products according to each respondent s own perceptions about the nature of production process. In our study, 994 respondents assigned at least two products to the low-education category, but half wanted protection for some of those low-education products while advocating free trade for the other low-education products. People were more internally consistent about products they assigned to other educational clusters. Nevertheless, around onethird of respondents wanted protection for some but not all medium-education products, and nearly 40% wanted protection for some but not all high-education products. To sum up, when we asked about protection for various types of industries, many respondents failed to differentiate at all; others differentiated in favor of the wrong factor of production; and still others differentiated far too much. To quantify these deviations from theory, we tallied the number of respondents who displayed any of the following patterns: (1) when asked about foreign businesses that employed low, medium, or high levels of education, the respondent preferred a uniform trade policy for all three categories; (2) the respondent expressed differentiated policy preferences, but did not protect their own factor while leaving the opposite unprotected; or (3) having classified seven products according to their educational requirements, the respondent wanted limits for some but not all products in a single educational bin. Based on these three criteria, 96% of respondents did not fit the Stolper-Samuelson model. Despite the striking nature of these findings, it may be premature to dismiss factor-based accounts of individual preferences about trade policy. Perhaps the opinions of individuals reflect many considerations, in addition to the distributional consequences of trade. Or perhaps ordinary citizens (and the politicians and interest groups who help shape their opinions) think about many factors of production, not just education. At a minimum, though, the data suggest that 14

16 preferences about trade policy are not driven primarily by concerns about how free trade would affect the economic welfare of one s own educational group. 6. Using Industry-Specific Measures to Assess Ricardo-Viner Next, we tested the hypothesis that Americans disproportionately seek protection for their own industry. Table 5 summarizes the protectionist attitudes of the 558 respondents who worked in the tradeables sector, and were therefore asked whether the government should limit imports of whatever their firm made. Nearly 42 percent wanted protection for their own industry, but respondents were at least as willing to block other goods and services. Table 5: Percent Support for Protecting Specific Industries Industry Percent Customer service 66.0 Fruits and vegetables 57.5 Data entry 57.4 Typing services 56.2 Business consulting 53.1 Medicines 52.0 Own industry 41.9 Furniture 41.6 Clothing 40.7 Computers and software 38.9 Cars 38.5 Cell phones 34.1 Note: The table pertains to the 558 respondents who worked in a tradable industry. Four of the categories in the table customer service, data entry, typing, and business consulting appeared only in the second wave of the study; the statistics for those industries are based on 324 respondents, rather than 558. In fact, six sectors customer service, fruits and vegetables, data entry, typing services, business consulting, and medicines attracted substantially more protectionist sentiment than respondent s own industry. The differences were quite large (11-24 points) and almost certainly did not arise by chance (p <.0001). Moreover, although own industry ranked seventh in Table 5, the gaps between own industry and the next four sectors furniture, clothing, computers and software, and cars were neither substantively nor statistically significant. Only one sector, cell phones, seemed to merit substantially less protectionism the respondent s own industry. Overall, we found no evidence that people systematically privileged their own industry over others. We also checked whether anyone wanted protection for their own industry, while seeking free trade for all industries. Only 2% of respondents exhibited this pattern, and they 15

17 counterbalanced by a slightly larger group, comprising 3% of the sample, who wanted to protect all industries except their own! Next, we tested whether the desire to protect one s own industry was conditional on job mobility. Respondents were asked: Suppose you lost your job and could not get another job at a company that sells X. Would it be easy or hard for you to get a job at a business that does not sell X, but pays as well as the job you currently have? As in other parts of the survey, we replaced X with the respondent s two-word description of their own industry. Among respondents in the tradeables sector, 11% said that getting a job in another industry would be very easy, 30% said somewhat easy, 20% answered neither easy nor hard, 30% selected somewhat hard, and 9% said very hard. Table 6 summarizes the preferences of people who said it would be somewhat or very hard to get a job in another industry. Even within this fairly immobile subgroup, we found no clear tendency to protect one s own industry. Protectionism was significantly more popular in customer service, fruits and vegetables, data entry, typing services, and medicines. (Protectionist sentiment was also higher for business consulting, the difference was not significant at the.05 level because we did not inquire about consulting the first wave of the study.) We found no significant difference between own industry and the next four groups, ranging from clothing to computers and software. Own industry was favored only in comparison to cell phones. Table 6: Percent Support for Protecting Specific Industries, Among People Who said it would be Hard to Get Jobs in Other Industries Industry Percent Customer service 71.9 Fruits and vegetables 64.5 Data entry 64.0 Typing services 62.6 Medicines 59.9 Business consulting 57.6 Own industry 48.8 Clothing 47.5 Furniture 44.7 Cars 44.2 Computers and software 42.9 Cell phones 40.1 Note: The table pertains to the 217 respondents who worked in a tradable industry and, and indicated that it would be somewhat hard or very hard to get a job in another industry. Four of the industries customer service, data entry, typing, and business consulting appeared only in the second wave of the study, so the statistics for those industries are based on 139 respondents, rather than

18 Interestingly, both tables show especially high enthusiasm for protecting service industries. We are not aware of other studies that compare individual preferences toward trade in goods versus services, but this would be an interesting area for future research. In summary, respondents did not systematically privilege their own industry over other common industries. This finding was evident not only in the sample as a whole, but also for the subset who would struggle to find jobs in other industries. Our data undercut the idea Americans are pursuing their narrow economic interest by protecting their own jobs at the expense of other industries. 7. Using Industry-Specific Measures to Assess NNT Finally, we tested the hypothesis that the trade preferences of individuals depend on the productivity of their own firm. If highly productive firms perform well under free trade, whereas less productive firms need protection, then preferences for protection should be most evident among workers at relatively unproductive firms. Our initial survey in August 2011 did not measure productivity, but the subsequent wave in November 2012 included several relevant questions. First, we asked respondents about the efficiency of their firm. We elicited perceptions of absolute efficiency by explaining: Efficiency means making things quickly, without sacrificing quality or wasting materials. How would you rate the place where you work on a scale from 0 to 10, where 0 is extremely inefficient and 10 is extremely efficient? We then inquired about perceptions of relative efficiency, by asking whether the place where they worked was more or less efficient than other businesses that made similar products. Second, we asked employees at private, for-profit companies to comment on their firm s profitability. Some businesses make money, some businesses lose money, and some businesses break even, we explained. Which phrase best describes the business where you work now or worked most recently? The choices were losing money, breaking even, or making money. We also asked about relative profitability. When it comes to making money, is your business doing better or worse than other businesses that make similar products? As Figure 1 shows, our sample contains good variation on both efficiency and profits. The distributions in Figure 1 apply only to respondents in the tradeables sector those who indicated that U.S. businesses and consumers could potentially import the kinds of goods and services their firm makes. We focus on that subset, because they were the only people for whom we could test how perceptions of efficiency and profitability affected their desire to limit imports in their own industrial category. 17

19 0 0 Percent Percent Percent Percent Figure 1: Perceptions about the Efficiency and Profitability of One s Own Firm Absolute Efficiency Relative Efficiency Absolute efficiency Relative efficiency Absolute Profits Relative Profits Profits Profits Note: The histograms pertain to the 324 respondents who participated in wave 2 of the survey and said their current or most recent job was in the tradeables sector. Relative efficiency was measured in five levels, ranging from much less efficient to much more efficient than other firms that make similar products. Relative profitability was measured in five levels, ranging from doing a lot worse to doing a lot better than other businesses that make similar products. We found no systematic relationship between these measures and the respondent s desire to protect their own industry. The top left panel of Figure 2 splits the sample into five groups, based on whether the respondent thought their firm was more or less efficient than other businesses that make similar products. For each group, a dot indicates the percentage of respondents who favored protection for their own industry, and the line represents the 95% confidence interval around that percentage. The figure shows that respondents from relatively efficient firms favored protection at least as much people from relatively inefficient firms. 18

20 Figure 2: Desire to Protect Own Industry, By Efficiency or Profitability of Own Firm By Relative Efficiency By Absolute Efficiency Much less efficient Scores 0-4 Somewhat less efficient Scores 5-6 Neither more nor less Score of 7 Somewhat more efficient Score of 8 Much more efficient Scores Protect own industry(%) Protect own industry(%) By Relative Profits By Absolute Profits Doing much worse Losing money Doing somewhat worse About the same Breaking even Doing somewhat better Doing much better Making money Protect own industry(%) Protect own industry(%) Note: The graphs pertain to the 324 respondents who participated in wave 2 of the survey and said their current or most recent job was in the tradeables sector. The bottom left panel of Figure 2 slices the sample according to perceptions of relative profitability. Once again, the pattern runs contrary to expectation. The preference for protection was at least as strong when the respondent s firm was outperforming its peers, than when the firm was underperforming its peers. Together, these plots undermine the hypothesis that employees at relatively efficient or profitable firms want less protection than employees at other firms. The right side of Figure 2 presents similar analyses, based on absolute rather than relative measures. In the top right corner, we split the sample into quintiles based on the absolute efficiency of the respondent s firm. The desire to protect one s own industry was highest in the third and fourth efficiency quintiles (scores of 7 and 8), and lower but roughly similar in the other three efficiency quintiles. Overall, we find no systematic relationship between absolute efficiency and protectionist tendencies. Finally, the bottom right corner distinguishes respondents whose firms were losing money, breaking even, or making money. Here, the pattern runs exactly contrary to theory: respondents who work at profitable firms are actually more protectionist than respondents from firms that are not making money. Our survey included three additional measures of relevance to NNT theory. We asked respondents how many employees their company had at all its locations, because firms with more employees are more likely to participate in international trade (Bernard and Jensen 1997, 1999, 2004). In addition, respondents indicated whether their firm sold products or services in foreign countries; and whether it had employees in one location, in many locations but only in 19