show that much of overall wage inequality arises within sector occupations and for workers with similar observable characteristics. Cosar, Guner and Tybout  explore the combined effects of reductions in trade frictions, tariffs, and firing costs on firm dynamics, job turnover, and wage distributions. They find that integration with global product markets has increased both average income and job turnover in Colombia. Using data on trade-induced displacements, Kondo  shows that regions that face more foreign competition in the U.S. have higher job destruction rates, lower job creation rates, and, consequently, lower employment rates. Yotov  shows that trade-induced unemployment and trade-adjustment costs may result in an incumbent politician granting protection to an unorganized industry, even when there is political pressure by organized sectors of the labor market. This is consistent with the theoretical predictions of Grossman and Helpman  who argue that the government should protect organized industries but should subsidize imports in unorganized sectors.
3. Data and Variable Construction
3.1. Data Sources
To test the negotiation power hypothesis, which posits a negative relationship between import penetration and unionization rates, we obtain historical data on U.S. imports, exports, tariffs, foreign direct investment, and unionization by industry from several sources.
U.S. import and export data are from Peter Schott’s International Economics Resource Web site (http://www.som.yale.edu/faculty/pks4/sub_international.htm). The import database contains information on which commodity, indicated by the 10-digit Harmonized Tariff Schedule (HTS) Code, was imported from which country for every year from 1989 to 2008. The export database contains similar information that shows which commodities were exported to which country for the same 11 year time span. We obtain the eight-digit Harmonized Tariff Schedule (HTS) Code−level tariff data provided by John Romalis, which are available at his Web site http://www.johnromalis.com/. Tariff data cover the years between 1989 and 2001.
We retrieve the foreign direct investments by industry from the OECD’s Web site (http://stats.oecd.org/Index.aspx?DataSetCode=FDI_FLOW_INDUSTRY).
Census Industry Classification (CIC)−level unionization data between 1984 and 2006 are gathered from the Union Membership and Coverage Database (www.unionstats.com). This database, which is maintained by Barry Hirsch and David Macpherson, is compiled from the Current Population Survey. Hirsch and Macpherson  discuss the details on the construction of this unique and comprehensive dataset.
To merge the unionization data and the export-import databases, we use the concordance file that links 1989-2006 U.S. Harmonized System (HS) codes to U.S. SIC codes provided by Pierce and Schott  . They describe the concordances between the 10-digit HS categories used to classify products in U.S. international trade that cover the years 1989 to 2006. Because the data for import penetration and unionization are constructed using different industry definitions, we use a common denominator, two-digit SIC code level industry definition to merge the data.
3.2. Variable Descriptions
Our dependent variable, C3UNION, is the change in unionization rate in an industry within three years following the year import penetration is measured. To measure labor force unionization rate (UNION), we follow Connolly, Hirsch and Hirschey  and measure unionization in industry i in time t as the percentage of employed workers in a firm’s primary industry covered by unions in the collective bargaining with the employers, i.e.:
There may be several unions in a given industry; however, data do not allow us to identify unions by name. Therefore, the unionization measure basically captures the percentage of labor affiliated to unions. Because our industry definition is at the two-digit SIC code level, we are able to capture the spillover effects within an industry. For example, any import penetration shock occurring in the household furniture industry (SIC code 2510) may potentially impact the office furniture industry (SIC code 2520) because they have the same two-digit SIC code (25). Similarly, any import penetration in the knitting mills industry (SIC code 2253) may also affect the carpet and rugs industry (SIC code 2273).
We define the change in unionization rate as:
We choose a change period of three years to accommodate the possibility that a unionization decision (voluntarily or forced) may take time to respond to import penetration shock. We use one and two year horizons to investigate the robustness of this time difference.
We compute two import penetration measures. Our low-wage-country import penetration measure is:
where and Mit represent the value of imports from low-wage countries and all countries in industry i and year t, respectively, and Xit represents U.S. exports. This measure has the advantage of not only considering where imports originate and but also their amounts. We follow Bernard, Jensen & Schott  and classify a country as low wage in year t if its per capita GDP is less than 5% of U.S. per capita GDP. Their list of 52 countries, which is provided in the Appendix, contains well-known countries such as China and India as well as lesser known ones such as Lesoto and Togo.7 Similarly, our import penetration measure for countries not classified as low-wage, i.e., other-wage is:
Several factors may influence affect unionization rates. These factors include general market conditions affecting unemployment rates, regulations governing labor standards and Social Security, union recruitment strategies, unions’ prior success in bargaining, predominant social and political values  , and employers’ opposition to unionization  . To control for these factors, we include industry and time fixed effects in our econometric specification. We include industry fixed effects to capture differences in unionization rates across industries that may exist because of industry-specific regulations. We include year (time) fixed effects to account for variation in unionization that can be attributed to a general market condition in a particular year. Because the year fixed effects capture the effect of time trend, we do not include a separate time trend variable.
We also include two additional variables to capture industry protectionism: 1) the foreign direct investments an industry attracts and 2) the average tariff rate for the products associated with an industry. If foreign firms face high entry, then we expect the foreign direct investments to be low in that industry. Similarly, if the average tariff rate for the products of the industry is high, we expect less competition due to foreign companies. Overall, these two factors may influence the level of unionization in an industry if they create protection for domestic firms that would likely affect foreign direct investments, e.g., taxes and transportation costs.
We hand match the OECD’s ISIC3-based industry codes to two-digit SIC codes and combine these data with the Harmonized Tariff Schedule (HTS) Code-level tariff data. The tariff data cover the years between 1989 and 2001. Thus, to prevent observations from dropping out of the sample, we assume 1) tariff rates in 2001 prevail in the years between 2001 and 2005 and 2) foreign direct investment is zero if no foreign direct investment is reported in the OECD database. These two assumptions, collectively or separately, do not drive our results. To match the eight-digit harmonized code level tariff data to the two-digit SIC codes, we use the concordance file that links 1989-2006 U.S. HS codes to U.S. SIC codes provided by Pierce and Schott  . Of the variables provided in the tariff database, we use the ad valorem portion of the Most Favored Nation (MFN) duty rate as our measure of tariff rate. Because multiple products are associated with two-digit SIC codes, we use the average of these tariff rates as our measure of industry tariff rate. In addition to these factors, we include the level of unionization (UNION) as an explanatory variable because a 1% change in unionization (C3UNION) may have a different economic implication if the level of unionization was, say, 40% rather than 4%. Excluding the level variable would bias the coefficients if there were a relation between UNION and C3UNION.
In Table 1, we summarize the descriptive statistics of the variables introduced in this section. Of special importance are the variables associated with the incidence of unionization and import penetration. The mean unionization rate (UNION) is 17.0%, while the mean three-year change in this rate (C3UNION) is −1.23%. The mean of import penetration by low wage countries (LWPEN) is −0.173%, while the same measure for all other countries (OWPEN) is 1.20%. In addition, both of the penetration variable means are associated with relatively very large standard deviations (7.62% for LWPEN and 31.4% for OWPEN), a consequence their small denominator (i.e., net imports may be close to zero) creating outliers. This phenomenon is also evinced by the 5th to 95th ranges for the sevariables. Our results, however, are not sensitive to 1) excluding top and bottom outliers, 2) using outlier robust regression methods, and 3) employing other import penetration measures that generate fewer outliers.
Table 1. Summary statistics of variables. This table summarizes the univariate statistics of the primary variables. UNIONit is the percentage of employees that were members of a union in industry i and year t. C3UNIONi,t to t+3 is the change in unionization rate in an industry within three years following the year import penetration is measured. FDIit is the foreign direct investment; TARIFFit is the average of estimated MFN ad valorem rate of products; LWPENit is the ratio of low-wage country imports to net imports (x100); and OWPENit is the ratio of other than low-wage county imports to net imports (x100). The industry and time subscripts are suppressed in the table headings.
4. Empirical Design and Results
If competition originates from low-wage countries, employers’ opposition to unionization and/or unions’ inadequate bargaining power to negotiate (negotiation power hypothesis), then we predict a negative relation between import penetration originating from low-wage countries and unionization rates. We use the following specification to test this hypothesis:
To obtain exogenous variation in unionization, we instrument the current level of unionization by its lagged values. For ease of reading, where appropriate we sometimes denote a variable lagged one year as Lag1(.) and if lagged two years is Lag2(.) instead of using subscripts such as t − 1 and t − 2, respectively.
We report the results of the estimation and the specification tests for the instrument in Table 2. The R2 of first-stage regression is 0.40, and the coefficient of the instrument is statistically significant. The under- identification test also shows that the excluded instrument, Lag1(UNION), is relevant, i.e., correlated with the endogenous regressor, UNION. The weak identification tests show that our instrument is not a weak instrument, suggesting that Lag1(UNION) is not a poor predictor of UNION. Overall, the two statistical tests indicate that Lag1(UNION) carries desirable properties as an instrument.
Because our model is exactly identified (number of instruments is equal to number of instrumented variables), we cannot test the assumption that the instruments are not correlated with the error term in the equation of interest. If the adjustment costs are high, it is plausible to expect lagged unionization to predict current unionization; however, it is less clear whether lagged unionization would predict change in unionization in the coming three years (the error term contains unaccounted variation in unionization changes with the future three years). The time gap between the instrument (t − 1) and the error terms (t + 1 to t + 3) provide some assurance that the dis-
Table 2. We use the instrumental variables method to estimate the parameters of Equation (5). All variables are defined in Table 1. In the first four (last two) columns, we calculate import penetration using imports and exports of all low wage countries (China and India only). UWETS (under identification) refers to the Kleibergen-Paaprk LM F-statistic and WWETS (weak identification) refers to the Cragg-Donaldwald F-statistic. P-values are reported in brackets. In under the identification and weak identification tests, (a) and (b) refer to statistical significance at the 0.01 and 0.05 levels, respectively.
tant historical levels of unionization may not necessarily cause future unionization level changes. When we apply lagged values of unionization as an instrument to the current unionization level to obtain exogenous variation in unionization, we find that the economic impact of import penetration on the future unionization rate is −1.88 and statistically significant.8
In the third and fourth column of Table 2, we find that the impact of import penetration originating from other-wage countries (OWPEN) on unionization rate change is one-third that of low-wage countries, suggesting that imports from low-wage countries have distinct effects. In the last two columns of Table 2, we replicate the main specification using penetrations measures based on imports originating from India and China only. The coefficients of LWPEN are noticeably more negative than those reported in the first two columns, confirming the observation that these two countries are dominant importers in most of the industries.
In Table 3, we extend our analysis by using lagged values of unionization, FDI, and TARIFF as instruments for their current value. As suggested before, these variables may be correlated with the error term in the presence of omitted variables. For example, if industries with higher tariffs enjoy protection from politicians who
Table 3. Change in unionization and import penetration from low-wage countries. The first three columns of this table report the estimates of instrumental variables estimation in which Lag1(UNION), Lag1(FDI), and Lag1(TARIFF) are used as instruments for UNION, FDI, and TARIFF. The last column reports the estimates of second stage of instrumental variables regression in which unionization (UNIONCOV) and change in unionization (C3UNIONCOV) are defined by the ratio of number of covered employees (rather than union members) to total number of employees. All other variables are defined in Table 1. UWETS (under identification) refers to the Kleibergen-Paaprk LM F-statistic and WWETS (weak identification) refers to the Cragg-Donald Wald F-statistic. P-values are reported in brackets. In under the identification and weak identification tests, (a) and (b) refer to statistical significance at the 0.01 and 0.05 levels, respectively.
create barriers to foreign competitors, then the error term is likely to be correlated with the tariff rate if we cannot measure political protection. The statistical properties of both Lag1(FDI) and Lag1(TARIFF) as instruments are similar to that of Lag1(UNION). The results in Table 3 indicate that the coefficient of the low-wage penetration (−1.99) is very close to what we report in Table 2 (−1.88).
In a right-to-work state, the law gives employees the right to decide whether to become labor union members. According to U.S. federal labor laws, unions are required to negotiate the same wage settlement for all employees in the bargaining unit regardless of their membership status. Eren  finds that members enjoyed, on average, a wage premium of 9% over comparable covered nonmembers. Using number of members, rather than number of covered employees, to calculate unionization rate in right-to-work states would underestimate the total amount of covered members in collective bargaining. This suggests that if industries in the right-to-work states were attracting more import penetration from low-wage countries, then reduction in a unionization measure that ignores “covered but not-member” labor would be incorrectly attributed to switches from being “member” to “covered but not-member” as the loss of labor unions’ negotiation power.
We replicate our main results with a unionization measure that also incorporates covered nonmembers and obtain similar results. In other words, we define unionization as follows:
In the last column of Table 3, we report the results of instrumental variables regression’s second stage, which uses lagged values of UNIONCOV, FDI, and TARIFF as instruments for these variables. Our results indicate that this alternate definition of unionization is also negatively related to import penetration.
In the analysis presented thus far, we assumed import penetration is exogenous. In Table 4 we study the implications of modeling import penetration as an endogenous variable. We use tariff and lagged import penetration as instruments of import penetration. Lagged import penetration is likely to be related to the current year’s penetration because importing countries may find it costly to change their production mix and strategies in the short run. Similarly, tariff rates of the previous year are likely to be related to import penetration because countries are not likely to respond to tariff changes quickly.
We use the Arellano and Bond  approach to measure the relationships. Specifically, we take the first difference in both dependent and independent variables to eliminate the fixed effects across panels. Next, we use lagged levels of the variables as instruments for the first differenced variables. In our case, because UNION and FDI are considered endogenous, lags of the second order are potentially valid instruments. In other words, we use LWPENit−2, UNIONit−2 and FDIit−2 as instruments for ΔLWPENit, ΔUNIONit, and ΔFDIit, where Δ is the difference operator.9 The results in Table 4 indicate that the sign of change in import penetration is negative but statistically insignificant (p = 0.16). This result could simply be due to imperfect instruments, because the null hypothesis that the instruments are weak cannot be rejected by the Cragg-Donald Wald F-statistic (0.12).
Competition created by international trade forces firms and labor unions to adjust to new environments. Bernard, Jensen and Schott  provide compelling evidence that firms exposed to imports from low-wage countries have lower growth rates and lower survival probabilities. These firms change their business focus to include more capital- and skill-intensive processes. In this paper, we present evidence that complements their findings by analyzing the effects of low-wage country imports on unionization rates.
We find that import penetration originating from low-wage countries is negatively related to unionization rates. This finding suggests that firms become reluctant to agree with unions that ask for sticky labor costs and multi-year contracts as a result of declining company profit margins. It also suggests that employees may find it less beneficial to join a union by predicting unions’ loss of negotiation power in the new era. Our findings indicate that increasing the low-wage country import penetration by 1% reduces unionization rates by 1.88% within three years following import penetration after controlling for several factors that have been argued in the literature to affect unionization rates. Using the last year of our sample (2008), as a reference point this percentage translates into a loss of roughly 640,000 union members at the end of the 3-year period. If we consider China and India as the only low-wages countries the 1% increase in imports increases the reduction to −3.71%.
More broadly, our analysis raises the question about the net effect of import penetration coming from low-
Table 4. Changes in unionization and endogenous import penetration. This table reports the first and second stage values of instrumental variables estimation in which lag unionization is used as an instrument for current unionization. All variables are defined in Table 1. UWETS (under identification) refers to the Kleibergen-PaaprkLM F-statistic and WWETS (weak identification) refers to the Cragg-Donald Wald F-statistic. P-values are reported in brackets. In under the identification and weak identification tests, (a) and (b) refer to statistical significance at the 0.01 and 0.05 levels, respectively.
wage countries. Many heterogeneous firm models of international trade that predict output and employment are directed toward higher-productivity firms, leading to an increase in average industry productivity. Our evidence shows that at least part of this increase in productivity is associated with a reduction in the union participation rate. Whether this trade-off is advantageous to society has yet to be decided. Some arguments have been made to support the assertion that union membership has benefits for its members, including better health care, sounder pensions, and safer work environment. One of the goals of government is to find an appropriate socially acceptable balance between various trade restrictions promulgated by international trade agreements and the organizational structure of the domestic labor force. To perform this task fairly and successfully, the government must understand not only the economic but also the social trade-offs involved.
Cite this paper
AyferGurun,G. GeoffreyBooth, (2016) Trade Liberalization, Import Penetration and Unionization: The U.S. Experience. Theoretical Economics Letters,06,75-86. doi: 10.4236/tel.2016.61010
Appendix. Low-Wage Countries
Source:  .
1The dates associated with the trade liberalization initiatives refer to when the initial agreement was completed. It often took many years to modify the agreements until eventually accepted by the governments involved. As of the date of this paper the Trans-Pacific Partnership has passed the initial agreement stage but has not yet received the final approval of the U.S. government.
2Their contributions, as illustrated in Smith  and Ricardo  , are considered by many economists and historians to be part of the foundation of modern economics.
3A recent exception appears to be the South Korea - U.S. trade agreement. Schneider  points out that “[t] he United Autoworkers union and key auto-state legislators in both parties also have endorsed the final agreement.” He also reports that a spokesperson for the Obama White House in reference this agreement stated that “[i]t has been a long time since a union supported a trade agreement…”
4Unions have not always been hostile to the concept of free trade as evinced by their support of the Reciprocal Trade Agreements Act (1934) and the Trade Expansion Act (1962). Simmons  suggests that the early support of free trade initiatives by labor was because of their belief that reducing tariffs was tantamount to decreasing the cost of living through the lowering of prices.
5Bailey and Bosworth  report that average manufacturing output as a percentage of total GDP in real terms was about 12% between 1960 and 2012 (see their Figure 1). However, during the same time, manufacturing employment as a share of total U.S. work force declined from 33% to 10% between 1950 and 2013. More importantly, the number of persons engaged in production (measured as full-time equivalent employees plus the self-employed) shows a sharp decline from 1980 to 2010, a period that spans the four recent trade agreements as well as our research sample.
6In this particular case, Robinson  maintains that unions effectively blocked closure of some of General Motor’s existing plants because the United Auto Workers negotiated a contract with the auto producers that enforces a fixed wage “irrespective of the performed work.”
7Bernard, Jensen and Schott  note that the list “…represents the world’s most labor-abundant cohort of countries and therefore the set of countries most likely to have an effect on U.S. manufacturing plants”.
8In unreported results, we estimate specification (1) using unionization rate changes for the one and two years instead of three years. We find that the magnitude of the import penetration coefficient is insignificant for the first year. For the second year, however, the effect is statistically significant, and its magnitude is 37% of the magnitude reported in Table 2. This suggests that it takes more than two years for employees, firms, and unions to adjust to a new environment that includes cheaper foreign imports.
9Recall that we also denote lagged variables using the Lag operator so that, e.g., Lag2(LWPEN) is equivalent to LWPENit−2.