University of Southern California Dissertations and Theses

Trade, training, employment, and wages: evidence from the U.S. manufacturing industry

(USC Thesis Other)

Trade, training, employment, and wages: evidence from the U.S. manufacturing industry

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content TRADE, TRAINING, EMPLOYMENT, AND WAGES: EVIDENCE FROM THE U.S. MANUFACTURING INDUSTRY by Hao-Chung Li A Dissertation Presented to the FACULTY OF THE USC GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulllment of the Requirements for the Degree DOCTOR OF PHILOSOPHY (ECONOMICS) August 2010 Copyright 2010 Hao-Chung Li ii Dedication To My Parents, My Wife, and Elena, My Newborn Daughter iii Acknowledgments Finishing my Ph.D. study would never have been possible without the guidance of my committee members and the support from my friends and family. I would like to explicitly express my appreciation to my advisors, John Ham, RobertDekle,JohnStraussandGaryPainter. Theirguidance,advice,andquestions challenge me to become a better researcher. The care they have shown to me also set an excellent model of how to be a good advisor myself in the future. I would also like to give my utmost thanks to my parents. They have given me their full support to go after my own academic interest even though since my childhood they always hope I can follow my fathers footstep to become an electrical engineer. Finally, I would like to express my gratitude to my wife, Chia-I. Her positive and optimistic attitude lifts me through my bad days and brings me constant joy to my life. iv Tableof Contents Dedication ii Acknowledgments iii List of Tables vi Abstract viii Chapter 1 Introduction 1 Chapter 2 Import Competition and Company Training: Evidence from 7 the U.S. Microdata on Individuals 2.1 Introduction 7 2.2 Literature Review 14 2.2.1 Product Market Competition and Innovation 14 2.2.2 Imports, Outsourcing, and Labor Productivity 17 2.3 Theoretical Predictions 19 2.3.1 The Competition E¤ect of Imports on Training: 19 An Illustrative Model 2.3.1.1 The Framework 20 2.3.1.2 The Autarky Case 21 2.3.1.3 Foreign Entry and Competition 22 2.3.1.4 The Technology 24 2.3.1.5 The Training Decision 26 2.3.2 Skill Upgrading E¤ectunderlying Hypothesis 3 30 2.4 Data Description and Summary Statistics 31 2.4.1 The NLSY79: Microdata on Training, Work 32 History, and Other Personal Characteristics 2.4.1.1 Training Information in NLSY79 33 2.4.1.2 Other Personal Information in NLSY79 35 2.4.1.3 Construction of the Training Variable 35 2.4.2 Construction of Import Competition Measures 36 2.4.3 Other Industry Level Variables 38 2.4.3.1 Dening High and Low R&D Industries 39 2.4.3.2 Computing Industry-Level Nonroutine 40 Cognitive Task Requirement 2.4.4 Summary Statistics 41 v 2.5 Econometrics Framework 45 2.6 Results and Discussions 48 2.6.1 Basic Specication 48 2.6.2 The E¤ect of Import Penetration on Employer- 52 Paid Non-Company Training 2.6.3 Testing of Hypothesis 1: Do the E¤ects Vary by 53 the Income Levels of Importing Countries? 2.6.4 Testing of Hypothesis 2: Are the Training E¤ects 56 of Imports Di¤erent between High-Tech and Low-Tech Industries? 2.6.5 Testing of Hypothesis 3: Are There Di¤erences 62 between Imported Intermediate Goods and Imported Final Goods? 2.7 Robustness Check: Fixed E¤ects Estimation and IV 63 Estimation 2.8 Concluding Remarks 70 Chapter 3 Imports, Exports, and the Determination of Employment and 75 Wage in the U.S. Manufacturing Industries 1979-2001 3.1 Introduction 75 3.2 Literature Review 77 3.3 Empirical Framework 80 3.4 Data Description 84 3.5 Results and Discussions 86 3.6 Concluding Remarks 97 Chapter 4 Conclusion 100 References 103 Appendix A Model for the Competition E¤ect of Import Competition 107 on Training vi Listof Tables Table 1 Summary Statistics of Key Variables 42 Table 2 Correlation between Training and Individual and Industry-Level 43 Variables Table 3 Comparison of the E¤ects of Import Competitions on Company 44 Trainings with and without Industry Dummies Table 4 Baseline Result-E¤ect of Import Penetration on Company 49 Training Table 5 E¤ect of Import Penetration on Non-Company Training 54 Table 6 Testing Hypothesis 1-Comparison of the E¤ects on Training of 55 Imports from Countries of Di¤erent Income Levels Table 7 Testing Hypothesis 2-Comparison of the E¤ects of Imports on 58 Training in High-Tech and Low-Tech Industries (Comparing High RD and Low RD Industries) Table 8 Testing Hypothesis 2-Comparison of the E¤ects of Imports on 59 Training in High-Tech and Low-Tech Industries (Comparing Industries with Di¤erent Nonroutine Task Inputs Intensity) Table 9 Comparing the E¤ects of High Income Countries Imports and 61 Non-High Income Countries on Training for High-Tech and Low- Tech Industries (Comparing High RD and Low RD Industries) Table 10 Testing Hypothesis 3-Comparison of the E¤ects of Intermediate 63 and Final Goods Imports Table 11 Comparison of the Random E¤ects (RE) Model, Fixed E¤ects 66 (FE) Model, and the Instrumental Variable (IV) Model vii Table 12 Summary Statistics of Key Variables (Census 3-digit Industry 85 Level, Manufacturing Industries Only; Year 1979-2001) Table 13 Change in Industry Total Employment, Shipments, Domestic 87 Demand, Exports, and Imports: 1979-2001 Table 14 Change in Industry Production Worker Employment, Shipments, 90 Domestic Demand, Exports, and Imports: 1979-2001 Table 15 Change in Industry Non-Production Worker Employment, 91 Shipments, Domestic Demand, Exports, and Imports: 1979-2001 Table 16 Change in Industry Average Real Wage, Shipments, Domestic 93 Demand, Exports, and Imports: 1979-2001 Table 17 Change in Industry Production Worker Real Wage, Shipments, 95 Domestic Demand, Exports, and Imports: 1979-2001 Table 18 Change in Industry Non-Production Worker Real Wage, 96 Shipments, Domestic Demand, Exports, and Imports: 1979-2001 viii Abstract In this dissertation, I analyze the e¤ects of trade on the U.S. domestic labor market. I extend the current literature in two dimensions. First, I investigate the e¤ectofimportcompetitiononcompanytrainingwithinUnitedStatesmanufacturing industries. Second, I extend Freeman and Katzs (1991) and Kletzers (2002) studies on the employment and wage e¤ects of trade through the year 2001. My focus on the e¤ects of imports on company training is new to the literature, and it is also important as such training is an important factor in earnings and job security. Specically, I look at the e¤ect of imports on the incidence of company training for individuals in the National Longitudinal Survey of Youth. Overall, I nd that import competition has a negative e¤ect on company training. I also nd that imports from low- and middle- income countries have a more severe negative e¤ect on training than do those from high-income countries. However, I do not ndasignicantdi¤erencebetweenthee¤ectofimportsinhigh-technologyandlow- technologyindustries. Finally,Indthatthenalgoodsimportsinanindustryhavea morenegativee¤ectontrainingthantheintermediategoodsimportsintheindustry. Thus it is not surprising there is pressure to limit import competition, especially from low- or middle- income countries, since reduced training opportunities for U.S. workers can be perceived as reducing good jobs. My research on company training suggests that nonproduction workers bear the brunt of this negative e¤ect on training, while the e¤ect on production workers is ix insignicant. In my chapter on the employment and wage e¤ects of trade, I demon- strate that the results in my training study do not tell the full story. Typically, production workers might su¤er lower employment and wage levels when faced with import competition. On the other hand, rising demand for exports, through their e¤ectonmountingdomesticproductdemand,isassociatedwithincreasesinindustry employment and wage levels for both production and nonproduction workers. This suggeststhatwhenwediscussthee¤ectoftradeonemploymentandwages,weshould not overlook the positive e¤ect that arises from increasing foreign demand. The e¤ect of trade on the U.S. labor market is of great importance given the continuingriseintradeinboththemanufacturingandservicesectors. Mydissertation suggests that workers could potentially bear greater costs in the face of increased globalization. How to mitigate these potential negative e¤ects is a crucial policy question. 1 Chapter 1 Introduction During the last three decades, the United Stateseconomy has become increasingly open. This is evident in a steady increase of both exports and imports throughout the manufacturing industries. For example, between 1979 and 2001, the ratio of manufacturing exports to domestic manufacturing shipments almost doubled from 8.18%to15.6%,whiletheratioofmanufacturingimportstodomesticmanufacturing shipmentsalmosttripledfrom8.32%to24.3%. Duringthesameperiod,employment inU.S.manufacturingindustriesfellsteadilyfrom19.8millionin1979to15.6million in 2001, despite the fact that the population increased from 226 million in 1980 to 281 million in 2000. Concerns are rising about the widening gap between unskilled andhighly-skilledworkersintheUnitedStates,asunskilledworkersarefacingfalling real wages, while highly-skilled workers continue to enjoy substantial growth in their real wages. 1 These parallel developments have led some observers, both in the public and in academia, to link increased trade with changes in labor market conditions. Existing economicstudies(mostlyfromtheperiodbeforemid-1990s)focusonhowtradea¤ects wage structure and employment. In this dissertation, I will extend the literature in 1 For example, Autor, Katz, and Kearney (2006) nd that, between 1979 and 2005 and in terms of real wage, workers with less than 12 years of education su¤er a 16.5% decrease, those with 12 years of education su¤er a 4% decrease, while those with 13 to 15 years of education see a 5.4% increase, and those with at least 16 years of education enjoy a 22.1% increase. 2 twodimensions. First,Iwillinvestigateimportcompetitionse¤ectontheprobability that a worker received company training in the United States for the period between 1988 and 1996. Second, I will extend the literature on the e¤ect of trade on U.S. manufacturing wages and employment before 2001. Mystudyonthee¤ectsofimportsoncompanytrainingprovidesanewperspective forinvestigatingthee¤ectoftradeontheoveralllabormarket. Inthelaborliterature, companytraininghasbeenfoundtobeveryimportantforaworkerslifetimehuman capital accumulation. Many studies conclude that company training received from oneemployerisvaluablebothwiththeemployerthatprovidedthetrainingaswellas otherfutureemployers. Theseresultsimplythattheopportunitytoreceivecompany trainingmaybenettheworkerthroughouthiscareer. Inaddition,manystudiesnd thatworkerswhoreceivecompanytrainingwillbelesslikelytoleavetheirjobs. This implies that receiving company training may be an indicator of job security from a workers perspective. Inthisstudy,Iproposeandempiricallytesttwopotentialchannelsthroughwhich import competition may a¤ect rmsincentives to provide company training. The net e¤ect of these two channels suggests that the e¤ect of import competition is a priori ambiguous, and must be determined empirically. The rst channel is the competition e¤ect, which considers how foreign imports maya¤ectincentivestoprovidetrainingatdomesticrmsthatproducesimilarprod- ucts. To illustrate, I wrote a simple theoretical model which implies two testable hy- 3 potheses. The rst hypothesis argues that import competition from low- or middle- income countries will have a more negative e¤ect on training than that from high income countries. This occurs because these low- or middle- income countriesrela- tively low labor costs hurt the competitiveness of American products in the market, and could have a negative e¤ect on the survival of U.S. rms. Consequently, U.S. producers may be less willing to invest in training because they are afraid that they maybedrivenoutofthemarketandwillneverseereturnsontheircostlyinvestment. Mysecondhypothesisarguesthatworkersinhigh-techindustriesarelessa¤ected by foreign competitors. The idea is that domestic rms can escape low-cost foreign competitionandsurviveifthequalityoftheirproductsisfarsuperiortothatofforeign goods. Since,inhigh-techindustries,U.S.rmscandi¤erentiatethemselvesfromlow- costcompetitors, theymaystillhaveincentivestotraintheirworkers. Conversely, in low-tech industries, the goods are standardized, so U.S. rms cannot compete with low-cost foreign producers, either in cost or quality. Another channel of import competition that may a¤ect company training is the skill upgrading e¤ect, which considers the e¤ect of foreign intermediate imports on training. In the outsourcing literature, researchers argue that intermediate good imports benet the domestic industry because they allow domestic rms to focus on high-end jobs, which require more training. Some also argue that intermediate good imports encourage nal good producers to upgrade their product quality, since using these cheaper imports will increase the prots associated with quality upgrading. 4 Motivated by these arguments, I test the following third hypothesis: for an industry asawhole,theoverallnegativee¤ectofimportedintermediategoodsontrainingwill be less than that of imported nal goods. Empirically, using the National Longitudinal Survey of Youth, at the three-digit industry level I look at how imports a¤ect incidences of company training for indi- viduals. Overall, I nd that import competition has a negative e¤ect on company training, especially for non-production workers. I also nd that imports from low- and middle- income countries bring about a more severe negative e¤ect on training thandothosefromhigh-incomecountries. However, Idonotndasignicantdi¤er- ence between the e¤ect of imports in high-technology and low-technology industries. Finally, I nd that nal goods imports have a more negative e¤ect on training than intermediate good imports in a given industry. Thus it is not surprising that there ispressuretolimitimportcompetition, especiallyfromlow-ormiddle-incomecoun- tries, sincereducedtrainingopportunitiesforU.S. workerscaneasilybeperceivedas a reduction in "good jobs." Inmystudyoncompanytraining,Indthatmostofthenegativee¤ectontraining is borne by nonproduction workers. This is in contrast with the usual notion that trade,especiallyimports,hurtsproductionworkersmorethannonproductionworkers. One potential reason may be that most companies o¤er training to nonproduction workers. To see if production workers also su¤er, I return to the common theme of the literature and study the e¤ect of trade on industry employment and wages. 5 Toinvestigatethee¤ectsonemploymentandwages,Iusetheempiricalframework developedinFreemanandKatz(1991)andKletzer(2002). Thisframeworkattempts to model an industrys labor market and see how changes in a domestic industrys product demand, which potentially may uctuate due to changes in imports and exports, a¤ects the industrys employment level and average wage. My period of study starts from 1979 and ends at 2001. Thus, I am extending Freeman and Katzs work by 17 years, and Kletzers work by 7 years. I nd that increased imports are associated with decreased employment, while increased exports and domestic consumption are associated with increased employ- ment. The employment e¤ects on production and non-production workers are simi- lar. Conversely, the wage response to changes in exports and domestic consumption is qualitatively similar to typical employment responses. However, the response to changes in imports is more ambiguous. While I nd some evidence that there is an insignicant negative correlation between imports and wages for production workers, the same correlation is insignicantly positive for nonproduction workers. These negative correlations between imports and both employment and wages among production workers validates the common notion that production (or low skilled) workers su¤er in the face of import competition. Based on my ndings of diminished training and employment in the nonproduction workers, one can surmise that there is a cost to these workers as well. Moreover, my nding on employment is logically consistent with my nding of a negative e¤ect of imports on training. 6 Specically,themainrationaleforfewerhoursofcompanytrainingforworkerswhose employers face import competition is that domestic rms may want to quitand therefore will stop training altogether. The negative correlation between change in import penetration and change in employment suggests that U.S. manufactures are indeed cutting their labor force in the face of import competition. The structure of this dissertation is as follows: In the next chapter, I detail my study on the e¤ect of import competition on company training. I then present the theoretical model as well as my empirical work that tests several of my hypotheses. In the third chapter, I discuss the correlations between trade and both employment and wages. I discuss the rationale behind my empirical framework, and present the results. Finally, in the last chapter, I conclude my dissertation. 7 Chapter 2 Import Competition and Company Training: Evidence from the U.S. Microdata on Individuals 2.1 Introduction Importcompetitionsimpactonthelabormarkethasbecomeanareaofgreatconcern forthepublicandacademicsinrecentyears. Forexample,politiciansoftenarguethat foreign competitors take away American manufacturing jobs, pointing to the ood of Japanese cars and Chinese tires as one of the main causes of the challenges faced by many U.S. manufacturing industries and their workers over the past two decades. The impact of import competition on the wage structure and employment in U.S. manufacturing industries has been studied extensively using industry level data in thepasttwodecades. 2 InthischapterIextendtheliteraturebyexaminingthee¤ect of import penetration on the provision of company training by U.S. manufacturing 2 For example, Feenstra and Hanson (1996, 1999) study the e¤ects of foreign outsourcing on relative wage shares of production and non-production workers; Katz and Murphy (1992) look at several determinants that result in wage inequality, among them is import competition; Amiti and Wei (2006) extend Feenstra and Hansons study to service outsourcing. 8 rms using individual level data. This is an important topic as many theoretical and empiricalstudiespointtothesubstantiale¤ectsofcompanytrainingonlabormarket outcomes. Since the seminal paper by Becker (1962), labor economists often distinguish be- tween general and specic training. While general training raises a workers produc- tivitywithmanyemployers, specictrainingincreaseshis productivityonlywiththe employer providing the training. In the original Becker model, under the assump- tion of a competitive labor market, employers will not pay for any general training, as they cannot secure any of the future returns. Conversely, a worker who receives specic training will be more productive at one employer than elsewhere, so specic human capital creates a bilateral monopoly and the employer and worker share the costs of and returns to specic training. 3 However, recent theoretical papers argue that employers might also share the costs of general training if the labor market is imperfect, as labor market imperfections may prohibit job turnovers initiated by workers who receive general training, thereby allowing employers to receive returns from general training. Among several sources of labor market imperfection are mo- bility costs (Acemoglu and Pischke, 1999a, 1999b), asymmetric information between currentandpotentialemployers(AcemogluandPischke,1998),andcontractualprob- lems (LoewensteinandSpletzer, 1998). Recent empirical evidencealsolends support tothisnewstrandofliteratureanddemonstratesthatemployersdosometimesspon- 3 Hashimoto (1981) studies the division of costs and returns to specic training that maximize the expected value of an employer-worker match. 9 sor training that is deemed to be general (e.g. correspondence school). The skills acquired in those forms of training that are often regarded as specic (e.g. formal company-runtraining)areactuallytransferabletofuturepositionswithdi¤erentem- ployers. 4 The above theoretical and empirical studies provide us with the basis for under- standing the e¤ect of training on two important aspects of a workers labor market outcome: namely, his earnings and job turnover rate. First, as company training endows a worker with new skills, his earnings should increase. Moreover, since some proportion of employer-sponsored training is general, it would follow that this train- ing will be useful beyond his tenure with the current employer. In fact, empirically, both of these predictions have been conrmed. The returns from company training have been found to be at least as high as those fromformal schooling education, and trainingfrompreviousemployershasbeenfoundtoexertapositiveandpersistentin- uenceonwages. 56 Second,onewouldexpectthatcompanytraininggenerallyreduces 4 Loewenstein and Spletzer (1998, 1999). 5 Earlier studies usually study the wage e¤ect of training incidence. For instance, Blanchower and Lynchs (1994) study of wage growth of non-college graduates found that receiving company training raised wages by 12 percent. Few recent studies attempt to quantify the rate of return to training. For example, Bartel (1995) use the data from a company and estimated a rate of return of 35%. Frazis and Loewenstein (2005) use data from the 19792000 NLSY79 surveys and estimated a rate of return of about 40%, after considering various functional forms as well as corrections for direct costs, incidence of promotions, and etc. As a comparison, the estimated rates of return to schooling are usually around 10% (Card 1999). 6 Loewenstein and Spletzer (1998, 1999) use the data from the NLSY79, the same dataset I used for this chapter, and nd that the increase in wages associated with a training event is larger for future employers than for the employer providing the training. Booth and Bryan (2002) nd similar resultsusingtheUKhouseholdpaneldata. Oneexplanationforthislargerreturntofutureemployers is that the training provided has a large general component and has been at least partially paid for by the employer. 10 layo¤s and increases job tenure. Worker turnovers are costly not only because there are direct expenses associated with moving from one employer to another, but more importantlybecauserm-speciccapitalthatisacquiredthroughtrainingduringthe jobmatchwouldbelost. Empirically, thepredictionthatcompanytrainingcouldre- duceturnovershasalsobeensupported. Forexample,Lynch(1991)andLoewenstein and Spletzer (1997), both using the NLSY79 data in di¤erent time periods, nd that an increase in company training is associated with reduced job mobility. In this chapter, I propose and empirically test two potential channels through whichimportcompetitionmaya¤ectwhetherrmsprovidecompanytraining. Taken togetherthesetwochannelssuggestthatthee¤ectofimportcompetitiononthepro- visionofcompanytrainingisaprioriambiguous,andmustbedeterminedempirically. The rst channel is what I call a competition e¤ect,which refers to how foreign imports may a¤ect the incentives to provide training at domestic rms that produce similar products. As I will later show more formally in a simple model, this decision to invest in human capital may depend on the characteristics of foreign competitors, as well as the characteristics of the products domestic rms produce. In particular, my rst hypothesis on this competition e¤ect argues that import competition from low- or middle- income countries will have a more negative e¤ect on training than thatfromhighincomecountries, becausetheselow-ormiddle-incomecountriesrel- ativelylowlaborcostshurtthecompetitivenessofAmericanproductsinthemarket, and could have a more negative e¤ect on the continued existence of U.S. rms. As 11 their survival probability decreases, U.S. producers may be less willing to invest, as theymaynotbeabletorecoupthecostlyhumancapitalinvestmenttheyhavemade. My second hypothesis regarding the competition e¤ectargues that workers in more technologically advanced (High-Tech) industries may feel less impact from for- eign competitors. Here, I dene an industry to be High-Techif its products have multiple opportunities for improvement. The idea is that domestic rms can escape low-costforeigncompetitionandsurviveifthequalityoftheirproductsismuchmore superior to that of foreign goods. Therefore, domestic rms in High-Tech industries still have incentives to train their workers so as to realize these large improvements in product quality. 7 The second channel through which import competition might a¤ect company training is the skill upgrading e¤ect,which arises when foreign intermediate goods enter the domestic market. 8 Although these imported intermediate goods might put substantialcompetitionpressureondomesticintermediategoodproducers,andthere- fore may impose a potential negative competition e¤ect of imports on training, they, on the other hand, provide additional benets to the domestic industry. For exam- 7 There could be two reasons why providing company training will result in quality improvement. First, it is possible that company training allows workers to make use of the newest technology or production procedures, whose occurrence may be exogenously determined, resulting in a higher quality outcome. Second, it is also possible that company training helps workers to become innova- tors themselves and directly contribute to quality improvement. In this study, I do not attempt to distinguishbetweenthesetwopossibilities,butonlystressthepositivecorrelationbetweencompany training and quality improvement. 8 Another reason that this skill upgradinge¤ect may take into e¤ect is that foreign imports, both nal and intermediate good imports, may bring new ideas into the market, and consequently encourage domestic rms to train their workers in order to adopt these new ideas. However, since it is di¢ cult to quantify these new ideas,I would not attempt to test this possibility in the study. 12 ple, Glass and Saggi (2001) have argued that imported intermediate goods allow the domestic nal good producers to obtain cheaper inputs and henceforth earn larger prots. Thiscreatesgreaterincentivesforrmstoupgradetheirproductquality,and one waythat this can occur is by providing more training. 9 Another oftencitedben- et of imported intermediate goods is that they allow U.S. industries to recongure jobs to those that are more complex and require more training. 10 The above discus- sion suggests my third hypothesis: for an industry as a whole, the overall negative e¤ect of imported intermediate goods on training will be less than that of imported nal goods. This occurs because while both nal and intermediate good imports have a competition e¤ecton domestic nal and intermediate good producers, re- spectively, intermediate good imports also have a benecial skill-upgrading e¤ect on the domestic industry. Toempiricallydetermine the net e¤ect of import competitionontraining, as well as testing the above three hypotheses, I rst use the National Longitudinal Survey of Youth 1979 Cohort (NLSY79), which provides the best data on training received byindividuals. Ithenmergethisdatasetwithseveralothertradeandmanufacturing datasets in order to see how changes in import penetration a¤ect the incidence of companytraining. Istartwithananalysisof theoverall e¤ect of import competition oncompanytrainingforindividualsindi¤erentthree-digitmanufacturingindustries. 9 In the Glass and Saggi (2001) original argument, innovation is the venue for quality upgrading. HereIborrowtheirideabutinsteadarguethatcompanytrainingisthevenueforqualityupgrading. 10 This is the main argument of the foreign outsourcing literature, e.g. as in Feenstra and Hanson (1996). 13 I nd overall that an increase in 1 percentage point of import penetration will result inadecreaseintheprobabilityofcompanytrainingby0.006. Giventhatinmysam- ple period, between 1988 and 1996, the average import penetration increases from 14% to 18%, this implies a reduction in the probability of training by 0.024. This is a very large e¤ect since each year the probability of receiving company training is approximately 0.098. I also nd that this e¤ect is concentrated primarily among non-production workers. Next, I test the three hypotheses described above. First, in support of the rst hypothesis, I nd that the e¤ect of imports on training depends on the productscountry of origin. Specically, while imports from middle and low income countries exert a signicant negative e¤ect on training, high income country imports have an insignicant positive e¤ect. Second, I do not nd any signicant di¤erence in the e¤ect of imports on training between High-Tech and Low-Tech in- dustries, so the empirical support of my second hypothesis is weak. 11 Third, I nd that, in accordance with my third hypothesis, while the e¤ect of foreign nal good imports on training is signicantly negative, the e¤ect of foreign intermediate good importsisalwaysinsignicant,implyingthatforeignintermediategoodimportshave at least some positive skill upgrading e¤ect that neutralize its negative competition e¤ect. The chapter is organized as follows. In the next section I discuss the related literature. In section 3 I present my illustrative model on the competition e¤ect 11 When I have an insignicant coe¢ cient (or a di¤erence in coe¢ cients) with a large condence interval, I treat it as 0. To be accurate, however, in these cases we do not really empirically learn that much about whether we should accept or reject the null hypotheses. 14 of imports on training, as well as an outline of my hypotheses to be empirically investigated. Sections 4 and 5 describe the data used and my econometric approach respectively. I present and interpret my empirical results in Section 6. In Section 7 I showthatmyresultsarerobusttoreasonablechangesintheempiricalspecications. In Section 8 I summarize my ndings and o¤er concluding remarks. 2.2 Literature Review Few papers, either empirically or theoretically, look into the relationship between trade and company training. Therefore, in this literature review I rst focus on pa- pers that motivate my theoretical predictions by discussing the relationship between product market competition and innovation. Given that company training, like re- search and development (R&D) activity, is a form of costly investment that aims to improve future protability, it is reasonable to look at the theoretical guidance that this strand of literature provides. Next, I discuss several papers that study the rela- tionship between import competition and labor productivity. I argue that my study can increase our understanding on the channels through which imports may a¤ect labor productivity. 2.2.1 Product Market Competition and Innovation Although the literature seldom directly looks into the relationship between product market competition and company training, the correlation between competition and 15 innovation has been extensively studied in the Industrial Organization literature and in the endogenous growth literature. In the earlier literature, although there exists some work that argues for a potential positive e¤ect of competition on innovation, themostprominentmodelsalwayspredictanegativecorrelationbetweencompetition and innovation. 12 This occurs because under more competition, the monopoly rents thatrewardnewinnovationswillbereduced, sormshavelessincentivetoinnovate. However, there is another factor to be considered. For example, Porter (1990) arguesthatproductmarketcompetitionisgoodforinnovationbecauseitforcesrms to innovate in order to acquire a substantial lead over their rivals so as to escape competition. Aghion et al. (2004) theoretically formalize this idea of an escape- competitione¤ect in the context of foreign entry. Specically, the authors study howpotentialtechnologicallyadvancedforeignentry, intheformofforeigndirectin- vestment,a¤ectsdomesticincumbentrmsincentivestoconductR&D.Theauthors especially emphasize the role of the di¤erence between an incumbents initial state of technological development and that of potential foreign entrants. In their model, they nd that the threat of technologically advanced entry encourages innovation by incumbent rms in sectors that are initially close to the foreign entrantstechnologi- cal level. This occurs because incumbent rms need to successfully innovate in order to compete with potential foreign entrants when they enter. On the other hand, the 12 For example, in the industrial organization literature, see the Hotelling model and the monop- olistic competition model by Dixit and Stiglitz (1977), while in the endogenous growth literature, see the product variety framework of Romer (1990) and the Schumpeterian model of Aghion and Howitt (1992). 16 threat of technologically advanced entry would discourage innovation by incumbent rms that are initially further behind the foreign entrantstechnological level. This result arises because incumbent rms know that even if they successfully innovate, they may not be able to compete with foreign entrants because of the initial large technologicalgap. Intheirlaterpaper, Aghionetal. (2009)ndthatthistheoretical argument is supported by the data. Specically, using UK manufacturing rm data, they rst calculate the distance between the labor productivity of the industry with theworldtechnologyfrontier(representedbythecorrespondingU.S.industryaverage labor productivity). They then show that labor productivity growth responds pos- itively to increased foreign entry for rms in industries that are more-than-median close to the foreign technology frontier, while labor productivity growth responds negatively to foreign entry for rms in industries that are more-than-median distant from the frontier. 13 My work contributes to this literature in two aspects. First, I look at another di- mensionthatmaya¤ectarmsdecisionwithregardtoqualityimprovement. Specif- ically, while earlier papers such as Aghion et al. (2009) focus on how di¤erences in technological levels between domestic and foreign rms a¤ect quality upgrading de- cisions, I consider, both theoretically and empirically, how di¤erences in production costs a¤ects these decisions. This dimension is important because currently in the 13 In their paper, they calculate the distance to technology frontier for each industry-year cell and obtain a distancedistribution of the whole sample. More-than-median closerefers to those industry-year cells whose distance to technology frontier is smaller than or equal to the median distance to technology frontier in the distribution. 17 U.S. the main concern about foreign competition is that lower income countries pro- vide goods of acceptable quality but at much lower prices. Second, instead of R&D activities, I focus on worker training, which is another mode of quality upgrading that may take place. This is an area that has not been studied before, and given its substantial e¤ects on a workers welfare, it is also an important issue. 2.2.2 Imports, Outsourcing, and Labor Productivity I am not aware of any paper discussing the relationship between imports and train- ing. However, thereareseveralpapersthatdiscusstherelationshipbetweenimports, outsourcing, and labor productivity. 14 For example, one paper that looks at im- port penetration and labor productivity is McDonald (1994). He nds that import penetration increases labor productivity, especially in those industries that have a high concentration ratio. His argument for this nding is that import competition serves as a slack-reducing device which forces the domestic rms, especially those whoinitiallyenjoylittledomesticcompetition,toworkmoree¢ ciently. 15 Therehave also been several papers that study the relationship between foreign outsourcing and productivity, both at the industry level and the rm level. 16 The main theoretical 14 In this chapter, the concepts of foreign outsourcingand imported intermediate goodsare exchangeable. Most of the time I use imported intermediate goods,though when referring to arguments stemming from the outsourcing literature, I use the words foreign outsourcing. 15 A theoretical model which formalizes this notion that product market competition enhances productive e¢ ciency is Hart (1983). 16 One of the earliest studies is Feenstra and Hanson (1999). For a recent comprehensive survey, please see Olsen (2006). 18 argument that foreign outsourcing will increase labor productivity is that outsourc- ing opportunities allow rms to relocate their tasks to those over which they have a comparative advantage. In addition, there are also theoretical models which argue that imported intermediate goods will bring about new ideas and technologies that benettheindustry. Impactssuchasthesewillallresultinhigherlaborproductivity. Empirically, however, theresultsaremixed. Somendpositivee¤ectsofoutsourcing on productivity, while others do not nd signicant e¤ects. In the U.S. case, Amiti and Wei (2006) use industry level data between 1992 and 2000 and nd a positive and signicant, but small, impact of material outsourcing on labor productivity. My work can contribute to this literature in two ways. To begin with, although there may be an increase in labor productivity, we do not really know how this is achieved. Generally speaking, there are three ways that this could happen. First, there could be an improvement in workersskills. Second, as recent papers such as Melitz (2003) emphasize, within an industry there may be a reallocation e¤ect, i.e. rms that are more e¢ cient thrive while less e¢ cient ones wither. Third, there may belayo¤s. Thethreechannelshavesignicantlydi¤erentimplicationsfromthepoint of view of the worker, and my study sheds light on the rst channel. My second contribution consists of using individual micro data, which gives us a more clear- cut understanding of the e¤ects on workers. Moreover, since our observation is at the individual workers level, we are also less concerned about the simultaneity or endogeneity, issues that sometimes plague industry level studies. 19 2.3 Theoretical Predictions As mentioned earlier in the introduction, I expect import competition to a¤ect do- mestic rmsincentives to train through two channels: the competition e¤ectand theskillupgradinge¤ect.Inthissection,Iwillrstdiscussthroughasimplemodel themechanismbehindthecompetitione¤ect,andlayoutmyhypothesesthatcharac- teristicsofforeigncompetitorsaswellastheproductsmayhaveane¤ectontraining incentives. Next, I will discuss the skill upgrading e¤ect, which focuses on how im- ported intermediate goods o¤er additional benets to the industry and may increase rmsinterest inimproving their workersskills. Inparticular, I argue that the e¤ect of imported intermediate goods on company training should be less negative to that of imported nal good. 2.3.1 The Competition E¤ect of Imports on Training: An Illustrative Model The model presented here will show how an increase in import competition may a¤ectadomesticproducersincentivestoprovidetraining. Themodelfollowsclosely that of Aghion et al. (2004), which considers how potential technologically advanced foreign entry a¤ects domestic incumbent rmsincentives to innovate. The main di¤erence between my model and theirs is that while they only allow for di¤erence in the technological levels of the foreign and domestic goods, assuming same costs in production,mymodelexplicitlyallowsforcostdi¤erencesintheproductionofforeign 20 anddomesticgoods. Thisdistinctionisimportant,asIwillshowhowlowcostforeign producers may discourage domestic producers from quality upgrading (here, in the form of training), even when the domestic producers already have a technologically superior product. 2.3.1.1 The Framework Time is discrete. In each period t, consider an industry with a nal good, y(t), which is produced competitively, that could use either one, but only one, of the intermediate inputs, x d (t) and x f (t). Here, x d (t) denotes the ow of domestically produced input used at date t, and it embodies quality A d (t); x f (t) denotes the ow offoreignproducedinputusedatdatet,anditembodiesqualityA f (t). Thedomestic intermediate good producer can improve its product quality A d (t) through company training. While nal good producers can always get access to domestic inputs, they can only use foreign inputs when foreign input producers enter the domestic market. If a nal good producer uses domestic intermediate inputs, her production function will take the following form: y(t) =A d (t) 1 x d (t) (1) On the other hand, if she uses foreign inputs, then her production function will be written as: y(t) =A f (t) 1 x f (t) (2) 21 To produce one unit of domestic intermediate good, x d , the domestic input pro- ducer uses one unit of labor, so the unit cost of x d is w d , the domestic wage level. Similarly, the unit cost of x f is w f , the foreign wage level. 2.3.1.2 The Autarky Case When there is no foreign entry into the domestic intermediate input market, the domestic input market is monopolized by an incumbent producer, so nal good pro- ducers can only produce based on equation (1). The domestic monopolists prot at time t, m (t), measured in units of nal goods, will equal to m (t) =p(t)x d (t)w d x d (t) where p(t) is the price of the intermediate input, and w d is the domestic wage level and also the unit cost of x d . Since the nal good is produced in a perfectly competitive market, the equilibrium price p(t) of one unit of x d is the value of its marginal product. p m (t) =@y(t)=@x d (t) =A d (t) 1 x d (t) 1 Therefore,afterchoosingtheoptimalquantityx d (t),themonopolistcouldgetthe monopoly prot, m (t), which will equal to m (t) =A d (t);where (1) 1+ 1 w 1 d (3) 22 2.3.1.3 Foreign Entry and Competition The monopolized intermediate input producer is subject to an entry threat from foreign producers. Let q denotes the probability that the foreign producer shows up. Therefore, a larger q implies that the domestic rm is more likely to face foreign competition. Oncetheforeignintermediateinputproducersenterthedomesticmarket,domes- tic nal goodproducers couldchoose betweenx d andx f forproduction, sonowboth productionfunctions(1)and(2)areavailable. Followingtheliterature,Iassumethat domesticandforeignrmscompeteinpricesforthemarket,thoughoneneedstotake into account of the quality di¤erence between domestic and foreign goods, indicated byA d (t) andA f (t). Fromequations(1)and(2),itisobviousthatwhentheproductioncostsofx d and x f arethesame, thenalgoodproducerwillchoosetheintermediateinputthatisof higher quality. Therefore, the domestic (foreign) producer will dominate the market whiletheforeign(domestic)producerwill bedrivenoutof themarketif A d (t)> (<) A f (t). Hence, the only determinant of the intermediate input provider is the relative quality between the domestic and foreign producers. However,whentheunitcostsofx d andx f aredi¤erent,thedomesticproducermay still lose the market even if her product is of better quality. To see this, rearranging equation (2) in the following way: 23 y(t) =A f (t) 1 x f (t) =A d (t) 1 A f (t) A d (t) 1 x f (t) (2a) =A d (t) 1 " A f (t) A d (t) 1 x f (t) # =A d (t) 1 b x f (t) whereb x f (t) = A f (t) A d (t) 1 x f (t): Comparing equations (1) and (2a), we see that, from the domestic nal good producersperspective, thequalityofoneunitofx d isequivalenttothatofb x f . Since the domestic and foreign intermediate good producers compete in prices, domestic nalgoodproducerswillchoosetheonethatischeaper. Thelowestpricethedomestic inputproducercouldchargeforoneunitofx d isw d ,herunitproductioncost,whilethe lowestpricetheforeigninputproducercouldchargeforaunitofb x f is A d (t) A f (t) 1 w f , thequalityadjustedcostoftheforeignrm. Itisthenevidentthatifw f issu¢ ciently low, then even if A d (t) > A f (t), i.e. the domestic product has better quality, the quality-adjustedcostoftheforeignproductcouldstillbelower,i.e. A d (t) A f (t) 1 w f < w d , and hence drive the domestic input producer out of the market. 17 Next, following the assumption made in Aghion et al. (2004), I assume that the potential foreign entrant (which will show up with probability q) can observe the e¤ectiveness of training before it actually enters. This implies that if the foreign 17 When rms compete in prices, and their products are the same, the producer with a lower unit cost will always win the entire market, and the price she is going to charge will be the unit cost (minus epsilon) of the second-lowest unit cost producer. 24 company nds that the domestic producer has lower (or equal) quality adjusted pro- duction costs, it will not enter at all, and the domestic input producer can maintain its monopoly position. Therefore, from the domestic intermediate input producers perspective, after the realization of e¤ectiveness of training, i.e. A d (t) is known, two cases could happen when a foreign competitor shows up. Case 1: A d (t) A f (t) 1 w f <w d . Under case 1, the domestic intermediate input producer has a larger quality- adjustedproductioncost,soshewillbedrivenoutofthemarketandearnzeroprot, i.e. c (t) = 0. Here, c (t) is the prot when a potential foreign entrant shows up. Case 2: A d (t) A f (t) 1 w f w d . Incase2, foreignentrywill not occurevenif apotential foreignentrant shows up because the foreign rm realizes that the domestic rm will win the market anyway. Consequently,thedomesticinputproducerwillstillholdthemonopolyinthemarket and earn monopoly prots as in (3), i.e. c (t) = m (t). 2.3.1.4 The Technology The quality of the input takes on discrete values. At the end of period t 1, the domestic inputs quality is A d (t 1) = k1 A 0 , and the quality of foreign inputs is A f (t 1) = j1 A 0 . Both j and k are integers, and A 0 is just a constant. For the 25 foreignintermediateinput,Iassumethatitsqualitywillalwaysimprovebyafactorof in each period, so A f (t) = j A 0 . Conversely, the domestic input producer need to incuratrainingcost of 1 2 cz 2 , wherec is aconstant, inordertoimproveherproducts quality by a factor of s in the next period with probability z. Hereafter, I will call z the training intensity the domestic producer chooses. The above implies that if the training is e¤ective, which occurs with probability z, the quality of the domestic input at period t would be A d (t) = k1+s A 0 ; however, if the training is ine¤ective, which occurs with probability 1z, the quality would remain to beA d (t) = k1 A 0 . The index s =f1; 2g is the number of steps that the domestic rm can improve once training is e¤ective, and its value is determined by the characteristics of the industry. In particular, I dene an industry to be High-Techwhen s = 2. This is intended to reect the situation where conditional on e¤ective training the quality improvement of the domestic product (2-steps) could be greater than that of the foreign product (1-step). On the other hand, when s = 1, the industry is viewed as Low-Tech. The quality improvement of the domestic product under e¤ective training is the same as that of the foreign product (both 1-step improvement). Making the above assumptions implies that the prot levels achieved can be in- dexed by parameters: k, j, s, w d , and w f . For instance, when training is e¤ec- tive but no foreign rms show up, the associated monopoly prot can be written as m = m (k 1 +s;w d ); when training is e¤ective and a foreign competitor shows up, the prot can be written as c = c (k 1 +s;j;w d ;w f ). 26 2.3.1.5 The Training Decision From the domestic input producers perspective, the decision of training intensity, z, will depend on three factors. First, the probability, q, that a foreign rm shows up. Second,herpotentialforeigncompetitorsproductqualityandunitcost. Finally,the extent of quality improvement it may attain when training is e¤ective. Consider a domestic intermediate input producer that chooses to train with in- tensityz. Inthis case, thedomesticinput producers prot, not includingthe cost of training,willbe c (k1+s;j;w d ;w f )withprobabilityzq,theprobabilitythattrain- ing is e¤ective and that a foreign rm shows up; her prot will be m (k 1 +s;w d ) with probabilityz(1q), corresponding to the case when training is e¤ective and no foreignrmsshowup;withprobability (1z)q,theprotwillbe c (k1;j;w d ;w f ), andthishappenswhentrainingisine¤ectiveandaforeigncompetitorshowsup; and nally, the domestic input producers prot will be m (k 1;w d ) with probability (1z)(1q), and this arises when training is ine¤ective and no foreign competitors show up. Therefore, the prot maximization problem that the domestic input producer attempts to solve, including the cost of training, is the following: max z E =zq c (k 1 +s;j;w d ;w f ) +z(1q) m (k 1 +s;w d ) + (1z)q c (k 1;j;w d ;w f ) + (1z)(1q) m (k 1;w d ) 1 2 cz 2 27 Fromtherst ordercondition, weknowthatthesolutiontotheabovemaximiza- tion problem is: cz =q c (k 1 +s;j;w d ;w f ) + (1q) m (k 1 +s;w d ) q c (k 1;j;w d ;w f ) (1q) m (k 1;w d ) So the optimal intensityz depends on the probabilityq that a foreign competitor will show up, and the prot levels under each scenario. Given that the main interest hereistoseehowincreasedimportcompetition,i.e. alargerq,willa¤ectthetraining intensity, z, one could di¤erentiate the above equation with respect to q to get dz dq = 1 c h c (k 1 +s;j;w d ;w f ) m (k 1 +s;w d ) (4) c (k 1;j;w d ;w f ) + m (k 1;w d ) i Itisnowapparentthathowanincreaseinimportcompetitionmaya¤ecttraining intensity depends crucially on the parameters k, j, s, w d , and w f . As I will show in detailintheAppendix,thee¤ectofimportcompetitiononcompanytrainingcouldbe either one of the three e¤ects: a discouragement (negative) e¤ect, an encouragement (positive)e¤ect,ornoe¤ect. However,inthischapterIexcludethelastcasebecause itcorrespondstotheuninterestingcasewherethedomesticrmissosuperiorthatit will always win the domestic market regardless of the outcome of company training orwhetheraforeigncompetitorshowsup. Onewouldonlybeinterestedinthee¤ect 28 of import competition when potential foreign entrants actually impose a threat to domestic rms. 18 Basedonequation(4)andtheanalysisintheAppendix,Iwillmakethefollowing two testable hypotheses, where the technical derivation that leads to them is also in the Appendix. My rst hypothesis is the following: H1: The negative e¤ect of import competition on company training will be more severe for imports from middle or low income countries than those from high income countries. The rationale for this rst hypothesis is straightforward. In face of rising import competition, the domestic intermediate rm needs to upgrade its products quality through training so that it at least has a chance to survive the competition. Hold- ingtheforeigncompetitorsqualitylevelxed,whenthepotentialforeigncompetitor comesfromahighincomecountry,thedi¤erenceinquality-adjustedproductioncosts ofthedomesticandforeignrmsisrelativelysmall,sothedomesticrmhasanincen- tivetoinvestmoreintraining,i.e. increasingtheprobabilitythattrainingise¤ective, because e¤ective training will allow its products quality-adjusted production cost to belowerthanthatoftheforeignersandsubsequentlytowinthecompetition. Onthe otherhand, whenthe potential foreigncompetitorcomes fromalowincome country, the di¤erence in quality-adjusted production costs of the domestic and foreign rms 18 Empirically, it is also di¢ cult to empirically test this no-e¤ect of import competition when domesticqualityissu¢ cientlysuperior. Thereasonbehindthisisthatinthiscase, theforeignrms will never enter, so we could not observe an increasein import competition anyway. 29 is already too large. There is no hope to win against the potential entrant even after upgrading, so the e¤ect of increased import competition is to reduce the training intensity as the domestic rm knows that investing in training will not be protable. My second hypothesis considers how rms in High-Tech industries and Low-Tech industries may behave di¤erently in provision of training when import competition increases. Specically, the hypothesis I am going to test is: H2: Thenegativee¤ectofimportcompetitiononcompanytrainingwill beless severe for High-Tech industries than Low-Tech industries. Therationaleforthissecondhypothesisisalsointuitive. Inthemodel,thereason thatthereisanegativee¤ectofimportsontrainingisbecausethedi¤erenceinquality- adjusted costs of the domestic and foreign rms is too large, so that the domestic rm will lose to the potential entrant even after successful upgrading. However, in a High-Tech industry, with e¤ective training the domestic rm could improve its qualitybyamuchlargerextentnotethatthedomesticrmenjoysatwo-stepquality improvement while the foreign rm only improves by one stepso that it could now win the market through e¤ective training. Consequently, the domestic rm has the incentive to invest more in training in order to escape foreign competition. In the empirical sections of this chapter, I will test the above two hypotheses. As a preview, although the data empirically support the rst hypothesis, I nd little evidence in support of the second hypothesis. 30 2.3.2 Skill Upgrading E¤ectunderlying Hypothesis 3 Importsfromforeigncountriesconsistnotonlyofnalgoods,butalsoofintermediate goods. This distinction is crucial because, from the industry perspective, imported intermediate goods provide additional benets to the industry compared with those provided by imported nal goods. For example, it has been argued that imported intermediategoodsallowtheimportingrmswhousethemtohaveaccesstocheaper inputs, holding quality constant. With regard to these ideas, Glass and Saggi (2001) point out that within an industry, importing intermediate goods lowers the marginal cost of production and increases prots for the nal good producers within the in- dustry, which in turn creates greater incentives to these producers to upgrade their productquality. 19 Onewaythatqualityupgradingcanoccuristhroughmoreworker training. Inaddition, theliteratureonforeignoutsourcing, whichcanalsobethoughtofas importingintermediategoods,oftenarguesthatanoutsourcingopportunityenablesa rmtorelocateitsrelativelyine¢ cientproductionprocessestoexternalproviders,so thatthermcanfocusonandexpandinareaswhereithasacomparativeadvantage. As U.S. rms are usually thought to be more technologically advanced, this means that they can focus on tasks with higher complexity levels. Accordingly, they must provide more training to workers in order to meet this expansion of high-end jobs. 19 In this study, I only look at how the intermediate goods import in a 3-digit industry a¤ects the domestic producers within the same 3-digit industry. I argue that there will be a skill upgrading e¤ectbecausetherearedomesticnalgoodsproducerswhouseimportsinthesame3-digitindustry intermediate goods imports. 31 Attheindustrylevel,theabovediscussionsuggeststhatonewouldexpectamore negative e¤ect of nal good imports on training than that of intermediate good im- ports. Bothimportedintermediategoodsandimportednalgoodshavethepotential to impose a negative competition e¤ecton training conducted by domestic inter- mediate good and nal good producers, respectively. However, intermediate good importsbringadditionalbenetstothedomesticindustryastheyallowthedomestic nal good producers to obtain cheaper inputs and have greater incentives to provide training in order to upgrade product quality. Intermediate goods imports also allow theU.S.industrytorecongurejobstothosethataremorecomplexandrequiremore training. Therefore, the third hypothesis I will empirically test is the following: H3: Imports of nal goods will have a more negative e¤ect on training than that of intermediate goods imports. 2.4 Data Description and Summary Statistics In this section, I will rst discuss the data I use, the NLSY79. Second, I will dis- cuss the derivation of import competition measures as well as the data sources for these variables. Third, I will discuss other industry-level data used in my regression analysis. Finally, I will look at the summary statistics. 32 2.4.1 The NLSY79: Microdata on Training, Work History, and Other Personal Characteristics ThemicrodatasetI useinthischapteristheNational Longitudinal Surveyof Youth 1979 Cohort (NLSY79). The individuals in this survey were between 14 and 21 in 1979, and were followed annually until 1994. Since 1994, the survey has been conducted biannually. In this chapter, I restrict the sample to men working in the manufacturingindustriesbetween1988and1996. Thechoiceofthisparticularperiod of time is due to two main constraints. First, as I will discuss in detail later, there is a signicant change in the training questions before and after 1988. Therefore, the training information is not comparable across the two periods. Second, in 1997 there was a large change in the U.S. industry classication system, from the original Standard Industry Classication (SIC) system to the new North America Industry Classication System (NAICS). The correspondence between the two classication systems is in many cases not straightforward, so the e¤ects of import competition before and after 1997 are not comparable. 20 Therefore, I only consider the years before 1997. Finally, the decision to conne my study to manufacturing workers not only is because import competition mainly occurred in the manufacturing industries duringthatperiod,butalsobecausetheinformationonserviceimportsintheservice 20 There is another reason why only focusing on years before 1997 may not be a big limitation. Since the NLSY is a panel of data that follows the same cohort across years, the age distribution will shift to the right each year. It has been shown in many papers that most training occurs in the early years of a persons life cycle. Hence, even if I use data after 1997, it is not evident that I should merge the two time periods (before and after 1997) together since it is quite possible that the training behaviors for younger and older people will be di¤erent. 33 industries is scarce and usually of unsatisfactory quality. In total, there are 2147 individuals in the sample with a total of 7208 observations. This is an unbalanced paneldatasetduetothefactthatindividualsmovedinandoutofthemanufacturing sector across years. 2.4.1.1 Training Information in NLSY79 A variety of formal training questions were asked in the NLSY79 in all survey years, except in 1987. Individuals were asked to report on several vocational or technical programs in which they were enrolled since the previous survey. Until 1986, the maximum number of reported programs was two, and in 1988 it was increased to four. However,themainreasonthatIconcentrateontheperiodafter1988isbecause up until 1986, questions regarding the type of training program and the start and end dates were asked only if the training lasted over four weeks. Starting in 1988, these questions were asked for all programs, regardless of their length. Given the starkdi¤erencebetweenthescopesoftrainingquestionsbeforeandafter1988,Ionly focused on the latter period. TheNLSY79alsoreportsthetypesoftrainingprogramsinvolved. Fromthestart in 1988, the following categories have been reported: 1. Business college; 2. Nurses program; 3. Apprenticeship; 4. Vocational or technical institution; 5. Barber or beauty school; 6. Flight school; 7. Correspon- dence; 8. Aformalcompanytrainingrunbyemployerormilitarytraining(excluding 34 basic training); 9. Seminars or training programs at work run by someone other than employer; 10. Seminars or training programs outside of work; 11. Vocational rehabilitation center; and 12. other. Following Bartel and Sichermans (1998) denition of company training, I will dene categories 810 as company training. The inclusion of categories 8 and 9 is self-evident, but I also include seminars outside work as company training because severalpapersshowthatthiscategoryseemstobemoresimilartocategories8and9 than to the other categories. For example, in a series of paper using the NLSY79 as theirmaindatasource, LoewensteinandSpletzer(1997, 1998, 1999)showthatwhile employers pay for around 90% of training programs running at work and seminars outside work (specically, employers pay for about 82% of trainings in the form of seminars outside work), employers only pay for about 45% of other kinds of training programs. In addition, based on a question asked only in the 1993 NLSY79 survey regarding the generality of the training program, the content of training programs at work and in seminars outside work are more specic than those in other categories of training programs. Therefore, in this chapter I focus on these three categories of company training programs, as they should cater more to the rms actual needs and are the ones that rms would more likely to resort to in the face of a changing environment,suchasincreasingimportcompetition. However,sincerecenttheoretical workalsosuggeststhatcompaniesmaypayforotherformsoftrainingthataredeemed to be more general, in one section of the chapter I also consider the e¤ect of imports 35 onemployer-sponsoredtrainingincategories1-7and11-12. Theseothercategoriesof training are muchless important as they represent only 13%of all the employer-paid training. Myregressionresultshowsthatimportcompetitionhasnoe¤ectontraining in these categories, suggesting that for my problem at least, focusing on training in categories 8-10 is appropriate. 21 2.4.1.2 Other Personal Information in NLSY79 The NLSY79 contains records on personal characteristics such as work experience, tenure, education that are available each year. Although the NLSY79 records a maximum of ve jobs in each survey year, it designates only one of these jobs as a CPS job,which in most cases is the most recent or current job at the time of the interview. AseriesofimportantquestionsareaskedfortheCPSjobsonly,specically the3-digitCensusindustrycodesoftheCPSjobs. SinceIwanttolinkindustry-level information in the NLSY79 to trade data, my analysis in this chapter is restricted to CPS jobs. 2.4.1.3 Construction of the Training Variable I construct my training variable on a yearly basis. In particular, since I have infor- mationonthestartingandendingmonthandyearofeachtrainingprogram, Idene 21 LaterinthechapterwhenIpresenttheempiricalresultofthee¤ectofimportsonnon-company training,Iarguethatonepotentialreasonfortheno-e¤ectonnon-companytrainingcouldbebecause attendinganon-companytrainingprogrammaybeinitiatedbytheworker, insteadoftheemployer. The employer only pays for it because it is considered a benet provided to the workers. 36 a dummy variable for the incidence of training that occurred in the past year. 22 Moreover, since this chapter considers how import competition a¤ects employersin- centivestotraintheirworkers,Iwillonlyconsiderthosecompanytrainingprograms, i.e. training programs in categories 8-10, that are at least partly sponsored by the CPS employers, as I only know the industry a¢ liations of CPS employers. 23 2.4.2 Construction of Import Competition Measures The measure adopted for import competition is import penetration (impen), which hasbeenusedextensivelyintheempirical tradeliterature. Foranindustry j attime t, it is dened as impen jt = imports of industry j at t (shipment + import export) of industry j at t 100. (5) The denominator on the RHS of (5) is the value of domestic consumption of industryjatyeart,anditequalsdomesticproductionplusimportsandminusexports in the industryj and yeart. The import penetration measure captures the extent of competition in the domestic market an industry faces. For rms in industries with a highratioofimportstoconsumption,theirdomesticrevenuemaybemorevulnerable 22 Although the survey was conducted every year before 1994, some respondents might not have taken the survey in a particular year. In this case, it is crucial to have information on the starting and ending dates of training to identify yearly training incidences, since the original questions asked about training activities that occurred since last survey. 23 The NLSY79 asks questions about who pays for the training. Potential sponsors may be self/family, employer, or a variety of governmental programs such as Job Training Partnership Act (JTPA), Trade Adjustment Act (TAA), etc. 37 to foreign shocks, as a substantial part of their competition comes from abroad. I construct this import penetration measure from the U.S. import and export data provided by Robert Feenstra on his website. 24 The annual U.S. imports and exports data are originally available at the 4-digit SIC manufacturing industry level. Since the industry level information in NLSY79 is at the 3-digit Census industry code, I aggregate the above information to the 3-digit Census industry level. Inthisstudy,Imaketwoadditionaldistinctionsregardingtheimportcompetition information. First, I distinguish between imports from low, middle, and high income countries. Second, I distinguish between imports of nal goods and of intermediate goods. For the former distinction, lowincome countries are dened as countries with income no more than 20% of the US GDP per capita on average during 1985-1996; middleincomecountriesaredenedasthosewhoseGDPpercapitaliesbetween20% to70%oftheUSGDPpercapita;andhighincomecountriesaredenedascountries whose income are at least 70% of that of the US. For each of the countries, I have deneda separate import penetrationmeasure. Specically, I have a variable forlow income country (LIC) import penetration (LIC_impen), which is dened as LIC_impen jt = imports of industry j from LIC at t (shipment + import export) of industry j at t 100. 24 http://www.econ.ucdavis.edu/faculty/fzfeens/. 38 Twoothervariableshavebeendenedsimilarlyformiddleandhighincomecoun- triesimport penetration, respectively. My data for intermediate and nal goods imports come from Peter Schotts web- site. 25 It contains information on the share of imported intermediate goods within each 4-digit SIC industry, whereas intermediate goods are dened as those imports that include the word partsin their import product codes. Since there is a change inthetradecodesystemsin1989fromtheoldTSUSAsystemtothenewHSsystem, the classications of intermediate and nal goods before and since 1989 are not com- parable. Therefore, for the empirical exercise that involves this distinction, I have separate coe¢ cients forthe two time periods so as to reect the potential di¤erential e¤ects of imports under di¤erent classications. 2.4.3 Other Industry Level Variables I have included several other industry level variables to serve as additional controls. The rst variable is the proportion of IT capital to total equipment capital usage withintheindustry. Thisvariableisincludedtocapturethetechnologyadvancement coming from the utilization of IT capital, and it is calculated it based on the capital andmultifactorproductivityinformationprovidedbytheBureauofLaborStatistics. This variable is only available at the 2-digit SIC level. The second variable I include is the capital-labor ratio, and it is included because we would expect changes in 25 http://www.som.yale.edu/faculty/pks4/. 39 capital usage may also a¤ect rmsincentives to train workers. Following Bernard et al. (2006), I also include a variable that measures the wage share of non-production workers within an industry. This attempts to capture the change in the industrys relativedemandforskilledworkersduringtheperiod. Boththecapital-laborratioand non-production workerswage share are calculated from the NBER Manufacturing Database, andtheyare originallyavailable at the 4-digit SIClevel. Finally, I include a measure of the annual percentages of unionized workers in the industry. This informationcomesfromtheonlineUnionMembershipandCoverageDatabase,which is compiled from the Current Population Survey at the 3-digit Census level. 26 2.4.3.1 Dening High and Low R&D Industries In one of my extensions, I allow di¤erential e¤ects of imports on trainings for indus- trieswithhighR&Dintensity(HighR&D)andlowR&Dintensity(LowR&D).This distinction is made because I hope to see whether rms in High-Tech and Low-Tech industries respond di¤erently in their provision of training when facing increased im- port competition. The level of R&D intensity is my rst indicator of whether an industry is High-Techor Low-Tech. The classication of an industry to be High R&D or Low R&D is based on the OECD denition. The OECD denes the following manufacturing industries as HighR&Dindustries: O¢ ceMachinery,MotorVehicles,OtherTransport,Electronic Equipment, Instruments, Chemical and Pharmaceutical Products, and TV, Radio, 26 http://unionstats.gsu.edu. 40 andCommunicationsEquipment. TheremainingindustriesaredenedasLowR&D industries. 2.4.3.2 Computing Industry-Level Nonroutine Cognitive Task Require- ment In another extension, I attempt to see whether industries that use di¤erent inten- sities of nonroutine cognitivetask inputs behave di¤erently in the face of import competition. Nonroutine cognitive task inputs can be thought of as those tasks that professional personnel carry out, such as product development and design, strategic planning, and managerial duties. The intensity of nonroutine cognitive task inputs within an industry is my second indicator of whether the industry is more High- Techor Low-Tech.This information is from David Autors website, and is the samedatasetheandhiscoauthorsuseinAutor,Levy,andMurname(2003). 27 Autor and coauthors draw the information on detailed task contents from the U.S. Depart- mentofLaborsDictionaryofOccupationTitles(DOT)foreachofthehighlydetailed 12,000+ occupations. They then aggregate this information on occupations into the 3-digitlevelindustriesusingoccupationemploymentsharesasweights. Therefore,for eachindustry,theyhaveconstructedmeasuresthatreecttheimportanceofphysical demands, skill requirements, and favorable worker aptitudes at work. Autor,Levy,andMurname(2003)specicallychoosetwovariablestomeasurethe extentsofnonroutinecognitivetasksneededinimplementingajob. Therstvariable 27 http://econ-mit.edu/faculty/dautor/data. 41 captures the extent to which occupations involve Direction, Control, and Planning of activities (DCP), and this variable often takes on a high value for occupations that emphasize managerial and interpersonal tasks. The second variable, MATH, measuresthequantitativereasoningrequirementoftheoccupations,anditusuallyhas a high value in scientic and engineering occupations. I interact these two variables respectivelywithmyimportpenetrationmeasureinordertoseewhetherthetraining activities in those industries that involve many nonroutine cognitive tasks are less a¤ected by imports. 28 2.4.4 Summary Statistics Table 1 provides the summary statistics of the variables used in this chapter. As one canseefromthetable,onaverage9.8%oftheobservationsreceiveemployer-sponsored companytrainingduringthemostrecentyear. Itisalsoworthnotingthat,duringthe period of study, there is a substantial increase in total import penetration. Between 1988 and 1996, the total import penetration in the manufacturing sector increased from14%to18%. Thesharesofimportsfromcountriesofdi¤erentincomelevelsalso changeddramatically. Whilethesharesofimportscomingfromhighincomecountries fellfrom63%to52%,thesharesofimportsfromlowincomecountriesincreasedfrom 15% to 29%. 28 Autor provides information on selected years between 1960 and 1998, including years 1989 to 1991 that match the time period of my study. I would only show the results using the 1990 data, though the results from using the other two years are very similar. 42 Table 1: Summary Statistics of Key Variables Mean Standard Deviation Individual Variables Incidence of Company Training 0.098 0.297 Experience (weeks) 536.29 179.75 Tenure (weeks) 228.38 215.44 Grades Completed 12.55 2.40 White Dummy 0.571 0.495 Work in Large Firms (>1000 people) 0.450 0.498 Union Member 0.172 0.377 Total number of individuals 2147 Total number of observations 7208 Industry Characteristics 1985 1 1988 1992 1996 Import Penetration (%) 12.2 13.8 14.7 17.6 High Income Country Import Share (%) 63.4 59.9 56.5 51.7 Middle Income Country Import Share (%) 21.2 23.5 20.4 18.8 Low Income Country Import Share (%) 15.4 16.6 22.9 29.4 Intermediate Good Share (%) 2 15.4 18.2 15.1 12.9 IT Capital Ratio (%) 18.9 22.7 26.3 24.4 Capital Labor Ratio ($1000/worker) 5.5 4.9 6.0 7.5 Union Coverage (%) 27.7 24.9 22.0 20.0 Wage Share of Production Workers (%) 61.8 61.0 58.9 60.6 Note 1: I start from 1985 since I use lagged 3-yrs moving averages of industry level variables. 2: There is a reclassication of the intermediate goods in 1989; so values before and after 1989 are not comparable. 43 Table 2: Correlation between Training and Individual and Industry-Level Variables Training Grades completed 0.195 Production worker -0.170 White 0.089 Large rms 0.161 Hi-RD industry 0.129 Import penetration 0.061 Capital-labor ratio 0.091 Union rate -0.008 y Note: All correlations signicant at the 1% level except those with subscript " y " (insignicant). Table 2 provides information on the correlation between training incidence and several individual and industry-level variables. In accordance with most of the liter- ature on training, in general those people who are highly educated, non-production workers, whites, and who work in large rms and in High-Tech (high R&D) indus- triesaremorelikelytoreceivecompanytraining. Interestingly,thissimplecorrelation shows that trainings and import penetration is positively correlated. This seems to supporttheargumentsthatincreasedforeigncompetitionwillpushindustriesforward and upgrade their workersskills. However, this simple exercise that relies on cross-sectional comparisons of the e¤ect of imports has omitted an important fact: Each industry has its own special characteristics that may not be observed by econometricians, yet are correlated with import penetration. In this case, it is crucial to add industry xed e¤ects so as to control for these industry-specic time-invariant unobserved characteristics as well. Table 3 shows the result when I run two simple regressions that regress training 44 incidence on import penetration, year dummies, and with/without industry xed e¤ects. As we can see, when there are no industry dummies, the e¤ect of imports on training is signicantly positive, while after adding dummies the e¤ect of imports on training becomes signicantly negative. A Wald test shows that the industry dummies are jointly signicant, indicating the importance of adding industry xed e¤ectssothatthee¤ectofimportsoncompanytrainingscouldbecorrectlyidentied bythetimetrendsinimportpenetrationwithinindustries. Moreover,thispreliminary result of a negative correlation between import penetration and company training suggests that, in the context of previous theoretical discussions, in the United States the competition e¤ectof imports on training is negative, and it overshadows the positiveskill-upgradinge¤ectofimports,sotheoveralle¤ectofimportsoncompany training is negative. In fact, as I will demonstrate more later, this pattern will be preserved(andactuallystronger)evenafteraddingadditionalcontrolsonworkerand industry characteristics. Table 3: Comparison of the E¤ects of Import Competitions on Company Trainings with and without Industry Dummies Dependent variable: Incidence of company training Variable (1) (2) Coe¢ cient Coe¢ cient (Std. Errors) (Std. Errors) import penetration 0.0012 -0.0039 (0.0003)*** (0.0019)** N 7208 7208 industry dummies N Y time dummies Y Y Note 1: Signicance of the coe¢ cients***1%,**5%,*10%. 2. Value of import penetration is its 3-yrs lagged moving average. 45 2.5 Econometrics Framework In order to estimate the e¤ect of import competition on the likelihood of company training, I adopt a linear probability framework. In each year, an individual i in region r, industry j, and at time t could either receive company training (Y ijrt = 1), or not (Y ijrt = 0). The linear probability model determining Y ijrt is the following: Y ijrt =X it +impen j;t1 +Z jt1 + i + j + rt +" ijrt (6) Y ijrt = 1if received training; Y ijt = 0 if received no training In equation (6), the variable impen j;t1 is a three year moving average of the import penetration measure in years t 1, t 2, and t 3 for industry j in which the individual is working at time t. 29 Further,X it is a vector of current time-varying individualcharacteristics,includingmaritalstatus,race,yearsofeducation,residence in a standard metropolitan statistical area, years of experience and its square, union membership,andwhetherornottheindividualisemployedbyalargermwithmore than 1000 employees. Moreover, Z jt1 is a vector of time-varying industrial charac- teristics consisting of the three year moving averages of t 1, t 2, andt 3 of the capital-labor ratio, the ratio of IT capital to total capital usage, the wage share of non-production workers, and the industry union coverage, respectively. Finally, j is the industry dummy for industry j used to capture the unobserved industry-specic 29 I use a three years lagged moving average for impen jt and Z jt to obtain a smoother version of the variables. Nevertheless, I do not nd much di¤erence in my results using 1-year, 2-year, or 3-year lagged values of the import penetration variable. 46 time invariant e¤ects that may bias the results, while rt is a vector of region-time dummiesusedtocapturetheunobservedtime-variantmacroeconomicchangeswithin each region that may a¤ect training incidence. 30 In addition, I have added i to rep- resent individual unobserved characteristics, and use the random e¤ects estimation. In calculating standard errors, I allowed for heteroskedasticity as I am using linear probability models. There may be potential endogeneity problems under the above random e¤ects specication. First,endogeneitymaycomefromunobservedindividualcharacteristics that are correlated with import penetration. For example, individuals with certain characteristics (say, lack of motivation) may not be likely to be trained by their employers. At the same time, however, these workers may also intentionally choose industrieswithlessforeigncompetitioninanattempttosecuretheirjobs. Therefore, workersjob selection may introduce bias in my estimates. To address concerns of this kind, I also consider xed e¤ects specications. A second potential source of endogeneity is unobserved time-varying industry characteristics. For example, during the period studied there may be a restructuring withintheindustry(e.g. intensemergersandacquisitionsactivities)ororganizational changes within rms (e.g. attening of the organizational hierarchies). These sce- narios are mostlyunobservedbut may involve changing training practices, while also a¤ecting the competence of the industry. Hence the industrys import competition 30 In the NLSY79, there is no publicly available detailed information on the geographic area indi- viduals live. The only information available is the region (northeast, north central, south, and west) these respondents live in. 47 measure may be endogenous. In order to take account of this second source of bias, I have also considered instrumental variable specications for my study. The instru- mental variables I use are the three year moving averages over t 4,t 5, andt 6 of the industry level real exchange rates, industry level tari¤cost, and industry level freightcost. 31 Here,theindustryrealexchangerateisdenedastheimport-weighted average of the real exchange rates of importing countries by industry; 32 tari¤cost is calculated as the ratio between the duties collected and the customvalue of imports; andfreightcostisthesumoffreightandinsurancecostsdividedbythecustomsvalue of imports. 33 It should be noted that, after conducting bootstrap Hausman tests I could not reject random e¤ects models in either case. 34 Therefore, when I report my results I mainly discuss the random e¤ects models, given their e¢ ciency. I discuss the re- sults for xed e¤ects and instrumental variable specications later in the chapter as robustness checks. 31 I have considered smaller lagsspecically, the moving average of t-1, t-2, and t-3, as in the import penetration variablefor the instruments but reject the overidentication tests. 32 The industry real exchange rate I use is the industry import-weighted real exchange rate. The weights for a given industry each year are the shares of each foreign countrys imports of the total imports in that year, and I calculate them from the U.S. import and export data. Real exchange rates are constructed from the IMF International Financial Statistics Database. 33 The data for industry level tari¤and freight costs again both come from Peter Schotts website. Both variables are initially at the 4-digit SIC level, and I aggregate them to the 3-digit level. 34 These tests use the bootstrap to calculate the variance of the di¤erence in RE and FE (or IV) coe¢ cients. 48 2.6 Results and Discussions 2.6.1 Basic Specication Table 4 shows the baseline results. 35 In addition to running a regression on the men for the full sample, I also run separate regressions for production and non- production workers respectively. This distinction is made because one would expect that the nature of production and non-production jobs is quite di¤erent and would therefore expect that the determination of training for these two groups of workers be di¤erent. 36 In fact, I have found that non-production workers are more likely to receive company training (17%) than production workers (6%). As can be seen from the column 1 of the table, I nd a signicant negative e¤ect of import penetration on company training for the full sample. The results suggest that when import competition increases by 1 percentage point, the probability of getting training will decrease by 0.0059. Given that in my sample period, between 1988 and 1996, the average import penetration has increased from 14% to 18%, this implies a reduction in the probability of training for about 0.024, holding all other potential factors constant. This is a substantial e¤ect since the average incidence of company training 35 Ihavealsoconductedmyanalysisonwomen. Theresultindicatesnoe¤ectofimportcompetition on training for women (with p-value 0.652). One potential reason that this occurs may be that the number of women in the manufacturing industries is less (I have only about 3800 observations in that sample), so the estimated e¤ect is less precise. Another potential reason may be that the determinants of training may be di¤erent for women, given that women usually have higher job mobility (Royalty, 1996). 36 Bartel and Sicherman (1998) also distinguish between production and non-production workers when they consider the e¤ect of technological change on company training. 49 each year is only 0.098. As jobs that provide training are often viewed as those that provide better job security and earning prospects, this result seems to support those who argue that import penetration results in the export of more desirable jobs. Table 4: Baseline ResultE¤ect of Import Penetration on Company Training Dependent Variable: Incidence of Company Training Variable Coe¢ cient Coe¢ cient Coe¢ cient (Std. Errors) (Std. Errors) (Std. Errors) (1) (2) (3) All Workers Production Nonproduction Workers Workers Import Penetration -0.0059 -0.0019 -0.0103 (0.0019)*** (0.0021) (0.0036)*** IT Capital Ratio 0.0017 -0.0002 0.0040 (0.0016) (0.0017) (0.0029) Capital-Labor Ratio 0.0014 0.0005 0.0027 (0.0007)** (0.0008) (0.0012)** Non-Production Worker Wage Share 0.0091 0.0013 0.0205 (0.0051)* (0.0053) (0.0102)** Union Coverage -2.11e-05 -1.17e-05 -4.37e-05 (2.22e-05) (2.21e-05) (5.49e-05) Experience (in weeks) 0.0001 -3.64e-05 0.0004 (0.0001) (0.0001) (0.0002)** Experience Square 6.97e-08 1.69e-08 -3.65e-07 (8.23e-08) (8.73e-08) (2.00e-07)* 1-8 years of Schooling -0.0194 -0.0144 -0.0698 (0.0107)* (0.0129) (0.0270)*** 9-11 years of Schooling -0.0165 -0.0163 -0.0097 (0.0075)** (0.0084)** (0.0242) 13-15 years of Schooling 0.0184 0.0079 0.0331 (0.0102)* (0.0119) (0.0217) 16 years of Schooling 0.0785 0.0900 0.0672 (0.0163)*** (0.0342)** (0.0221)*** 17+ years of Schooling 0.1462 0.0798 0.1462 (0.0248)*** (0.0606) (0.0296)*** 50 Table4,Continued: BaselineResultE¤ectofImportPenetrationonCompanyTrain- ing Dependent Variable: Incidence of Company Training Variable Coe¢ cient Coe¢ cient Coe¢ cient (Std. Errors) (Std. Errors) (Std. Errors) (1) (2) (3) All Workers Production Nonproduction Workers Workers Production worker dummy -0.0351 (0.0094)*** White Dummy 0.0131 0.0089 0.0311 (0.0077)* (0.0083) (0.0185)* Married Dummy 0.0249 0.0260 0.0212 (0.0071)*** (0.0074)*** (0.0168) Large Firm (>1000) Dummy 0.0595 0.0397 0.0959 (0.0074)*** (0.0080)*** (0.0165)*** Union Member Dummy -0.0027 0.0001 0.0017 (0.0098) (0.0098) (0.0300) Year Dummies Y Y Y Industry Dummies Y Y Y SMSA Dummies Y Y Y Random E¤ects Y Y Y N 7208 4839 2369 Note: 1. See notes to table 3. 2. All industry-level variables are measured as their 3-years lagged moving average. 51 It is also useful to discuss this result in the context of the competition e¤ect and skill-upgrading e¤ectpreviously discussed. As argued earlier, while one would expect the skill-upgrading e¤ectof imports on training is positive, one cannot tell a priori thesignofthecompetitione¤ectofimports. However, thisbaselineresult suggeststhatthiscompetitione¤ectisnegative,anditissosubstantialinabsolute value that it reverses the positive skill-upgrading e¤ect. Another interesting point about the result is that this negative e¤ect is mainly borne by non-production workers. While I only nd insignicant negative e¤ects for production workers, I nd much more signicant negative e¤ects for non-production workers. For the non-production worker sample, a 1 percentage point increase in import penetration decreases the probability of getting training by 0.0103. This e¤ectislarge,althoughoneshouldkeepinmindthattheaverageprobabilityfornon- production workers to receive training in a given year is 0.17. One potential reason thatane¤ectonproductionworkersisnotapparentmaybethatproductionworkers aremuchlesslikelytobetrainedintherstplace,soaproductionworkersreception of training is less responsive to import competition. For the other control variables, as we can see from Column 1, in line with most papers on training incidence, I nd that training opportunity increases for men who are whites, married, higher educated, non-production workers, and those working in large rms. Two other industry level variables are also important in explaining the 52 incidence of training. 37 First, an increase in industrys non-production worker wage share will increase training, and second, an increase in the industry capital intensity will increase the probability that the workers receive training. Both results are not surprising, as an increase in both factors often signal that the industry is moving toward production procedures that are more sophisticated and capital intensive, so rms will need to upgrade their workersskills to adapt to new procedures. 2.6.2 The E¤ect of Import Penetration on Employer-Paid Non-Company Training Before formally test my hypotheses, here I will rst show the results of the e¤ect of import competition on non-company training that is paid for by employers. As described above, non-company training refers to training programs that are not run bycompanyandnotintheformofseminarsatoroutsidework. Thesenon-company training programs are also the ones that provide more general training. 38 37 I do not nd a signicant positive e¤ect of IT capital usage (as a measure of technological change)oncompanytraining, whichatrstsightmayseemtobeinconsistentwithwhatBarteland Sicherman(1998)havefoundonthee¤ectoftechnologicalchangeontraining. Onepotentialreason for the inconsistency may be the choice of variable to measure technological change. For example, if I use a variable indicating the growth in an industrys total factor productivity (TFP)Bartel and Sicherman use it as one of their measures of technological changein the regression, then I nd a signicant positive e¤ect of this variable on training, a nding in accordance with Bartel and Sichermans main result. Nevertheless, I intentionally exclude the use of this growth in TFP variable because it is a measure that captures the residual growth that cannot be explained by tangible inputs, so it may capture too much information, including the potential benecial e¤ects of imports. Hence, if I control for this growth in TFP variable, I could end up only measuring the detrimental e¤ects of imports on training, which is undesirable. In fact, when I include this variable in my regression, the coe¢ cient for import penetration becomes more negative (from -0.59 to -0.69), suggesting that the growth in TFP variable has captured part of the benecial e¤ect of import penetration. 38 Loewenstein and Spletzer (1999). 53 The results are in Table 5, and we can see that there is no signicant e¤ect of imports on non-company training. This could be due to two reasons. First, given their very low frequency (13% of all employer-provided training) in the rst place, it may be di¢ cult to nd any e¤ect at all. The second reason, and probably an even more important one, is that one cannot be sure if a workers taking of these non-company training programs is initiated by the rm or by the worker himself. It is possible that in some cases it is the worker that decides to take the training, and his employer is only paying for it because the employer provides this kind of benet. In cases like these, the training decision is made by the employee and it may di¤er substantiallyfromwhatthermmaywanttodohadthermistheonewhodecides. Forexample, inthecurrentsituationofrisingimportcompetition, theemployermay decide to send fewer workers to non-company training because he is concerned that he would never recoup the human capital investment, but the workers would want to seek for more training in order to prepare themselves for an potentially uncertain future. Inasituationlikethis, theemployerandhisemployeesmaymakecompletely oppositedecisions, resultinginanonnon-e¤ect of imports onnon-companytraining. 2.6.3 Testing of Hypothesis 1: Do the E¤ects Vary by the Income Levels of Importing Countries? Fromthissubsectiononwards, Iwill empiricallytestthehypothesesIproposeabove. Hypothesis 1 argues that the negative e¤ects of import competition will be more 54 Table 5: E¤ect of Import Penetration on Non-Company Training Dependent Variable: Incidence of Non-Company Training Variable Coe¢ cient Coe¢ cient Coe¢ cient (Std. Errors) (Std. Errors) (Std. Errors) (1) (2) (3) All Workers Production Nonproduction Workers Workers Import Penetration -0.0010 -0.0005 -0.0016 (0.0009) (0.0011) (0.0016) N 7208 4839 2369 Note: 1. All control variables are the same as in Table 4. 2. See notes to Table 3. severe for imports from low and middle income countries. To carry this out, low income countries are dened as countries with income no more than 20% of the US GDP per capita on average during 1985-1996; middle income countries are dened as those whose GDP per capita lies between 20% to 70% of the US GDP per capita. Relative to the US, these low and middle income countries enjoy a large labor cost advantage once they are able to produce acceptable quality goods (e.g. compete in the US market). Therefore, if my hypothesis about the importance of the foreign competitorscharacteristics is correct, we should expect a more negative e¤ect on training for the imports from these low and middle income countries. Table 6 shows the results of investigating the hypothesis. From column 1, we see that while the point estimates of the e¤ect on training of low and middle income country imports are bothabout -0.95, the point estimate of the e¤ect of highincome countryimports is insignicantly positive at 0.29, though I could not reject the null hypothesis that thesecoe¢ cientsarethesame(withap-valueof0.12). However, giventhesimilarity between the coe¢ cients on low and middle income country imports, onwards I have 55 pooled their data to achieve better estimation e¢ ciency. In the bottom panel where Inowonlydistinguishbetweenimportsfromhighandnon-highcountries, Indthat we can reject the equivalence of the e¤ects at the 5% level. As for the comparison betweenproductionandnon-productionworkers, fromthe secondandthirdcolumns of Table 6, again we see that the e¤ect of imports on training is mostly borne by non-production workers, while the e¤ect of imports on production workers is always insignicant, regardless of the income levels of importers. Table6: TestingHypothesis1-ComparisonoftheE¤ectsonTrainingofImportsfrom Countries of Di¤erent Income Levels Dependent Variable: Incidence of Company Training Variable Coe¢ cient Coe¢ cient Coe¢ cient (Std. Errors) (Std. Errors) (Std. Errors) (1) (2) (3) All Workers Production Nonproduction Workers Workers Panel A: Comparison between low,.middle, and high income country imports (a) Low income country -0.0094 -0.0036 -0.0166 import penetration (0.0025)*** (0.0026) (0.0050)*** (b) Middle income country -0.0096 -0.0051 -0.0179 import penetration (0.0039)*** (0.0044) (0.0074)*** (c) High income country 0.0028 0.0030 0.0078 import penetration (0.0046) (0.0048) (0.0092) Test (a)=(b)=(c) Accept Accept Accept (p-value=0.12) (p-value=0.52) (p-value=0.11) Panel B: Comparison between non-high and high income country imports (d) Non-high income country -0.0094 -0.0039 -0.0169 import penetration (0.0026)*** (0.0027) (0.0048)*** (e) High income country 0.0028 0.0026 0.0073 import penetration (0.0046) (0.0048) (0.0090) Test (d)=(e) Reject** Accept Reject** (p-value=0.04) (p-value=0.27) (p-value=0.03) N 7208 4839 2369 Note: See notes to Table 5. 56 2.6.4 Testing of Hypothesis 2: Are the Training E¤ects of Imports Di¤erent between High-Tech and Low-Tech Industries? Mysecondhypothesisarguesthatthee¤ectof importsontrainingwill belesssevere in High-Tech industries than in Low-Tech industries, where a High-Tech industry is denedasonewhoseproductshavemoreroomforqualityimprovement. Totestthis second hypothesis, I attempt to distinguish between High-Tech and Low-Tech indus- tries using two di¤erent denitions. The rst one I use is the OECDs denition of High-TechandLow-Techindustries,andtheclassicationofanindustryisbased on its R&D intensities. This denition is appropriate in the current context because onewouldassumethatinR&Dintensiveindustriesthereareconstantimprovements, allowingdomesticrmstowinthedomesticmarketthroughachievingbetterproduct quality. My second denition is based on Autor, Levy, and Murnames (2003) mea- sures of an industrys requirement of nonroutine cognitivetasks, i.e. those tasks thatemphasizeinteractive&managerialskills,and/oranalyticreasoningskills. Here, I would assume that industries that are Higher-Techinvolve more intensive non- routine cognitive tasks, and the rationale is that industries which emphasize these nonroutine cognitive tasks are more capable of di¤erentiating their products from those of their foreign competitors. For example, those industries that require a lot of analyticreasoningskillsmaybetheonesthathavecontinuousproductimprovements or innovations, and those industries that value interactive and managerial skills may 57 be the ones where improvement in business operations or provision of di¤erentiable products or services is possible. Totestmyhypothesis, Irstcomparethee¤ectofimportsontrainingincidences in high R&D industries and low R&D industries based on the OECD classication. Table 7 shows the results, and I nd little support for Hypothesis 2. From the rst column of table 7, we see that while there is a signicant negative e¤ect of -0.74 for Low-Tech industries, there is an insignicant negative e¤ect of -0.43 for High-Tech industries. I cannot reject the equality of the two coe¢ cients. A similar result holds for non-production workers. However, the result for production workers is slightly di¤erent. WhileIndasignicantnegativee¤ectofimportsontraininginlowR&D industries, the e¤ect is insignicantly positive in high R&D industries. Nevertheless, a Wald test shows that I still cannot reject the hypothesis that the coe¢ cients are di¤erent. Instead of comparing industries with di¤erent R&D intensities, another interest- ing exercise in order to see if U.S. industries are capable of escaping competition is tolookatwhetherindustriesthatinvolvedi¤erentintensitiesofnonroutinecognitive tasks may have di¤erent training behavior in face of increased import competition. Following Autor, Levy, and Murname (2003), I use two variables to measure non- routine cognitive task inputs. The rst variable measures an industrys emphasis on Direction, Control, and Planning of activities (DCP), while the second variable, MATH, measures the importance of quantitative reasoning. I have interacted these 58 Table 7: Testing Hypothesis 2-Comparison of the E¤ects of Imports on Training in High-Tech and Low-Tech Industries (Comparing High RD and Low RD Industries) Dependent Variable: Incidence of Company Training Variable Coe¢ cient Coe¢ cient Coe¢ cient (Std. Errors) (Std. Errors) (Std. Errors) (1) (2) (3) All Workers Production Nonproduction Workers Workers (a) High R&D industry -0.0043 -0.0020 -0.0087 *import penetration (0.0031) (0.0045) (0.0046)* (b) Low R&D industry -0.0074 -0.0041 -0.0130 *import penetration (0.0023)*** (0.0020)** (0.0059)** Test (a)=(b) Accept Accept Accept (p-value=0.40) (p-value=0.20) (p-value=0.56) N 7208 4839 2369 Note: See notes to Table 5. two variables respectively with my import penetration measure, and a signicant positive coe¢ cient on these interaction terms is be evidence in support of the second hypothesis. Table 8 shows the results. As we can see, although the coe¢ cients on the inter- action terms are positive in the full sample (Column 1) and non-production worker sample(Column3),theyareneversignicant. Therefore,similartowhatIndwhen comparing industries by their R&Dintensities, I do not nd signicant di¤erences in the e¤ects of imports on training for High-Tech and Low-Tech industries. FromtheresultsshowninTables7and8,weseethatempiricaltestsonHypothesis 2seemtoindicatethatthee¤ectsofimportsontraininginHigh-TechandLow-Tech are not too di¤erent. These results lead to the following question: Are the negative e¤ects of imports in both High-Tech and Low-Tech industries due to imports from 59 Table 8: Testing Hypothesis 2-Comparison of the E¤ects of Imports on Training in High-TechandLow-TechIndustries(ComparingIndustrieswithDi¤erentNonroutine Task Inputs Intensity) Dependent Variable: Incidence of Company Training Variable Coe¢ cient Coe¢ cient Coe¢ cient (Std. Errors) (Std. Errors) (Std. Errors) (1) (2) (3) All Workers Production Nonproduction Workers Workers Import penetration -0.0090 -0.0014 -0.0221 (0.0040)*** (0.0045) (0.0087)** Import penetration 0.0005 -0.0001 0.0018 *Math skill (0.0007) (0.0010) (0.0012) Import penetration -0.0073 0.0004 -0.0200 (0.0040)* (0.0048) (0.0086)** Import penetration 0.0002 -0.0004 0.0014 *DCP skill (0.0007) (0.0010) (0.0012) N 7208 4839 2369 Note: 1. See notes to Table 5. 2. DCP is the acronym for Direction, Control, and Planning. 60 non-highincomecountries,asHypothesis1hasemphasized? Toinvestigatethisissue, I allow the coe¢ cients on the e¤ect of high and non-high income country imports on training to be di¤erent for high R&D and low R&D industries. Table 9 shows the result. We see that the e¤ects of both high and non-high incomecountryimportsontrainingdonotseemtovarybasedonanindustrysR&D intensity. Therefore, although one might expect that the high innovation capacity of U.S.High-Techindustriescouldallowthemtoescapeforeigncompetition,empirically this does not hold. It is worthwhile to discuss this nding from the perspective of the model. In the model, the main reason that a domestic rm in a High-Tech industry might escape foreigncompetitionisbecause,throughe¤ectivetraining,theimprovementindomes- tic quality is much greater than that in the foreign competitors quality. However, it is not clear that in the real world this is true even in High-Tech industries. In the past two decades, newly industrialized countries such as the Asian Tigers play im- portant roles in High-Tech industries, including the information and communication technologyindustries. Thesenewlyindustrializedcountriesareusuallymiddleincome countries which boast high innovation capacity and mid-level labor costs. Hence, in theseHigh-Techindustriesinwhichthesecountrieshavestrength,U.S.manufactures maynotnecessarilyimprovefasterthantheirforeigncounterparts. 39 Inconsequence, U.S.manufacturesstillmaynotescapecompetition,eventhroughqualityupgrading. 39 Another reason that middle or low income countries may improve in quality faster is because it is easier to imitate than to innovate. 61 Table 9: Comparing the E¤ects of High Income Countries Imports and Non-High Income Countries on Training for High-Tech and Low-Tech Industries (Comparing High RD and Low RD Industries) Dependent Variable: Incidence of Company Training Variable Coe¢ cient Coe¢ cient Coe¢ cient (Std. Errors) (Std. Errors) (Std. Errors) (1) (2) (3) All Workers Production Nonproduction Workers Workers (a) High R&D* high income 0.0020 0.0034 0.0043 country import penetration (0.0058) (0.0064) (0.0103) (b) High R&D* non-high income -0.0079 -0.0005 -0.0146 country import penetration (0.0047)* (0.0058) (0.0066)** (c) Low R&D* high income 0.0045 0.0008 0.0192 country import penetration (0.0065) (0.0073) (0.0167) (b) Low R&D* non-high income -0.0107 -0.0051 -0.0212 country import penetration (0.0026)*** (0.0023)** (0.0065)*** Test (a)=(c) Accept Accept Accept (p-value=0.77) (p-value=0.64) (p-value=0.45) Test (b)=(d) Accept Accept Accept (p-value=0.59) (p-value=0.43) (p-value=0.47) N 7208 4839 2369 Note: See notes to Table 5. 62 2.6.5 Testing of Hypothesis 3: Are There Di¤erences be- tweenImportedIntermediateGoodsandImportedFi- nal Goods? Hypothesis 3 claims that imports of nal goods will have a more negative e¤ect on trainingthanthatof intermediateimports. Thisoccursbecauseimportedintermedi- ategoodsmaybenetthedomesticnal goodproducerbyprovidingcheaperinputs. This in turn may increase the incentive for quality upgrading, and also may allow U.S. workers to relocate to higher-end jobs, which require more training. Table 10 examinesthishypothesis. Noteherethatsincetheclassicationsofintermediateand nal goods are not comparable before and after 1989, I allow di¤erent coe¢ cients in the regression. Moreover, since I am using lagged 3 years moving average, I remove the observations in years 1990 and 1991. AswecanseefromColumn1ofthetable,inbothperiodsIndasignicantneg- ativee¤ectofimportednalgoodsontrainingwhilethecoe¢ cientsfortheimported intermediate goods are insignicant. However, due to the large standard errors as- sociate with imported intermediate goods, while I reject the null hypothesis that the e¤ects of intermediate goods and nal goods are the same for the full sample in the earlier period, I cannot do so for the later period. From Column 3 of the table, we see that the above pattern regarding the e¤ects of imported intermediate goods and nalgoodsagaincomefromthee¤ectsonnon-productionworkersinbothperiods. If I focus onthis particulargroup, I canreject the equivalence of the e¤ects of interme- 63 diate and nal goods at a 6% level in the latter period, while for the former period I can only reject the null at 11% level. From these results, it appears that there exists some, though not very strong, empirical support for Hypothesis 3. Table 10: Testing Hypothesis 3-Comparison of the E¤ects of Intermediate and Final Goods Imports Dependent Variable: Incidence of Company Training Variable Coe¢ cient Coe¢ cient Coe¢ cient (Std. Errors) (Std. Errors) (Std. Errors) (1) (2) (3) All Workers Production Nonproduction Workers Workers (a) Old nal goods -0.0051 -0.0023 -0.0099 import penetration(88-89) (0.0026)* (0.0028) (0.0058)* (b) Old intermediate goods 0.0033 0.0003 0.0064 import penetration(88-89) (0.0050) (0.0052) (0.0100) (c) New nal goods -0.0050 -0.0010 -0.0114 import penetration(92-96) (0.0023)** (0.0023) (0.0050)** (d) New intermediate goods -0.0005 -0.0055 0.0073 import penetration(92-96) (0.0046) (0.0053) (0.0085) Test (a)=(b) Reject* Accept Accept (p-value=0.09) (p-value=0.63) (p-value=0.11) Test (c)=(d) Accept Accept Reject* (p-value=0.36) (p-value=0.39) (p-value=0.06) N 5332 3579 1753 Note: See notes to Table 5. 2.7 Robustness Check: Fixed E¤ects Estimation and IV Estimation In this section, I o¤er two robustness checks for my results to address the potential endogeneity of import penetration. First, endogeneity may arise from unobserved 64 individual characteristics. Individuals with certain characteristics may choose their industries intentionallyandthus mayresult inbiasedestimates. Toaddress this rst concern, I use xed e¤ects (FE) models. Second, endogeneity may come from unob- servedtime-varyingindustrycharacteristics. Forexample,unobservedfactorssuchas industry restructuring, or organizational structural changes, may be correlated with both training and import penetration, which would result in inconsistent estimates. To address this second concern, I separately estimate my model using a random ef- fects instrumental variables (IV) model. The instrumental variables I use are the three year moving averages over t-4, t-5, and t-6 of the industry level real exchange rates, industry level tari¤costs, and industry level freight costs. 40 Table 11 shows the results for the full sample using a random e¤ect (RE) speci- cation, a FE specication, and an IV specication, respectively. 41 Panel A in Table 11 shows the baseline results for all three specications. We see that when using FE model, the import penetration coe¢ cient seem to be slightly lower (-0.0047) than that under the RE model (-0.0059), but the coe¢ cient under the IV specication is 40 I have considered shorter lagsspecically, the moving average of t-1, t-2, and t-3 (as in the case of the import penetration variable)for the instruments but reject the overidentication tests with these IVs. 41 For IV estimations, I conduct overidentication tests to see if the instruments are valid, i.e. uncorrelated with the errors. I also check whether the instruments are weak.Following Baum, Scha¤er,andStillmans(2007)suggestion,IusetheKleibergenPaap rk WaldStatistic,whichisthe robust analog of the CraggDonald statistic, as my test statistics and compare them to the critical values compiled by Stock and Yogo (2005) to check for weak instruments. However, since the Stock and Yogo (2005) test for weak instruments are based on i.i.d. errors, the weak instrument tests I conduct should be treated with caution because of the presence of heteroskedasticity in the data. Note also that since Stock and Yogo (2005) do not provide critical values when there are more than three endogenous variables, I could not conduct this weak instrument test when testing Hypothesis 3, as in that case I have four endogenous variables. 65 larger(-0.0116). Nevertheless,thequalitativemessagesunderthethreespecications are all the same, i.e. there is a signicant negative e¤ect of import competition on company training. To formally test whether my concerns about endogeneity are sta- tistically important, I conduct Hausman tests to see whether the RE model is valid. Itshouldbenoted,however,thatinthiscaseIcannotdirectlyusethestandardform of Hausman test, because inthe original formulation, one must assume an e¢ cient model as the null. Since I need to use the Whites standard errors to correct for the heteroskedasticity problem associated with a linear probability model, my RE model is not e¢ cient. I instead rely on a bootstrapping method to conduct my Hausman tests. 42 The test results suggest that the coe¢ cients under the FE model and IV model are both not statistically di¤erent from that under the RE model, indicating that using the RE specication is valid. PanelBshowstheresultswhenIconsiderRE,FE,andIVestimationswhendistin- guishingbetweenimportsfromhighincomecountriesandnon-highincomecountries. Looking across the columns, we see that the qualitative pattern carries over di¤er- ent specications. Specically, I nd a signicant negative e¤ect of non-high income country imports on training, but nd an insignicant positive e¤ect of high income country imports. In addition, in neither the FE nor the IV case can I reject the null hypothesis that the RE estimation is appropriate. 42 For the bootstrap Hausman tests, I sample by individuals (not observations) and perform 300 bootstrap replications. 66 Table11: ComparisonoftheRandomE¤ects(RE)Model,FixedE¤ects(FE)Model, and the Instrumental Variable (IV) Model RE Model FE Model IV Model Panel A: Baseline Result (a) Import Penetration -0.0059 -0.0047 -0.0116 (0.0019)*** (0.0026)* (0.0048)** Hausman Test Accept Accept (p-value=0.21) (p-value=0.24) Overidentication Test Accept (p-value=0.41) Weak Instrument No Panel B: Non-high income country imports vs. high income country imports (b) Non-high income country -0.0094 -0.0087 -0.0166 import penetration (0.0026)*** (0.0034)* (0.0066)** (c) High income countries 0.0028 0.0048 0.0191 import penetration (0.0046) (0.0059) (0.0257) Test (b)=(c) Reject** Reject* Accept (p-value=0.04) (p-value=0.07) (p-value=0.25) Hausman Test Accept Accept (p-value=0.43) (p-value=0.45) Overidentication Test Accept (p-value=0.28) Weak Instrument No Panel C: High-tech industry vs. low-tech industry (Measure: R&D) (d) Low R&D Industry -0.0074 -0.0069 -0.0139 *import penetration (0.0023)** (0.0030)** (0.0063)** (e) High R&D Industry -0.0043 -0.0026 -0.0087 *import penetration (0.0031) (0.0040) (0.0056) Test (b)=(c) Accept Accept Accept (p-value=0.40) (p-value=0.38) (p-value=0.50) Hausman Test Accept Accept (p-value=0.46) (p-value=0.50) Overidentication Test Accept (p-value=0.82) Weak Instrument No 67 Table 11, Continued: Comparison of the Random E¤ects (RE) Model, Fixed E¤ects (FE) Model, and the Instrumental Variable (IV) Model RE Model FE Model IV Model Panel D: High-tech industry vs. low-tech industry (Measure: Math Skill) (f) Import penetration -0.0090 -0.0090 -0.0141 (0.0040)** (0.0054) (0.0084)* (g) Import penetration 0.0005 0.0004 0.0008 *Math skill (0.0007) (0.0010) (0.0011) Hausman Test Accept Accept (p-value=0.41) (p-value=0.67) Overidentication Test Accept (p-value=0.19) Weak Instrument No Panel E: High-tech industry vs. low-tech industry (Measure: DCP Skill) (h) Import penetration -0.0073 -0.0058 -0.0084 (0.0040)* (0.0054) (0.0087) (i) Import penetration 0.0002 0.0002 -0.0003 *DCP skill (0.0007) (0.0009) (0.0011) Hausman Test Accept Accept (p-value=0.44) (p-value=0.42) Overidentication Test Accept (p-value=0.44) Weak Instrument No Panel F: E¤ects of high and non-high income country imports on high- and low-tech industries (j) High R&D*High income 0.0020 0.0053 country import penetration (0.0058) (0.0073) (k) High R&D *Non-high income -0.0079 -0.0073 country import penetration (0.0047)* (0.0060) (l) Low R&D* High income 0.0045 0.0030 country import penetration (0.0065) (0.0083) (m) Low R&D* Non-high income -0.0107 -0.0096 country import penetration (0.0026)*** (0.0036)*** Test(j)=(l) Accept Accept (p-value=0.77) (p-value=0.83) Test(k)=(m) Accept Accept (p-value=0.59) (p-value=0.73) Hausman Test Accept (p-value=0.59) 68 Table 11, Continued: Comparison of the Random E¤ects (RE) Model, Fixed E¤ects (FE) Model, and the Instrumental Variable (IV) Model RE Model FE Model IV Model Panel G: Comparison of the E¤ects of Intermediate and Final Goods Imports (n) Old nal goods -0.0051 -0.0047 -0.0097 import penetration(88-89) (0.0026)** (0.0037)** (0.0060) (o) Old intermediate goods 0.0033 0.0096 0.0136 import penetration(88-89) (0.0050) (0.0075) (0.0187) (p) New nal goods -0.0050 -0.0043 -0.0120 import penetration(92-96) (0.0023)** (0.0032) (0.0048)** (q) New intermediate goods -0.0005 0.0038 0.0241 import penetration(92-96) (0.0046) (0.0069) (0.0210) Test (n)=(o) Reject* Reject* Accept (p-value=0.09) (p-value=0.05) (p-value=0.20) Test (p)=(q) Accept Accept Reject* (p-value=0.36) (p-value=0.26) (p-value=0.09) Hausman Test Accept Accept (p-value=0.40) (p-value=0.16) Overidentication Test Accept (p-value=0.35) Note: See notes to Table 5. 69 Panels C, D, and E shows the results when I allow for the e¤ects of imports on training to be di¤erent for High-Tech and Low-Tech industries. As we can see across thecolumnsinPanelsCthroughE,theresultsunderdi¤erentspecicationsareagain very similar to what I have reported above, i.e. I could not nd the e¤ect of imports ontrainingtobesignicantlydi¤erentforHigh-TechandLow-Techindustries. Again, the Hausman tests indicate that using RE is consistent with the data. In Panel F, I interact income status of the import with the High-Tech status of the industry. We can see that the result using the xed e¤ects models reinforce the random e¤ects model result which suggests that the negative e¤ects of imports in High-Tech and Low-Tech industries are both due to imports from non-high income countries, and that this negative e¤ect is similar for High-Tech and Low-Tech industries. I cannot reject the equality of the RE and FE coe¢ cients. For this particular case, I do not use IV estimation because the instrumental variables are too weak to identify all of the coe¢ cients. 43 Finally, Panel G shows the result for the comparison between intermediate and nal goods imports using RE, FE, and IV specications. Recall that since the clas- sications of intermediate and nal goods are not comparable before and after 1989, I allow di¤erent coe¢ cients for di¤erent periods in the regression. Looking across 43 Since Stock and Yogo (2005) do not provide the critical values to determine whether the instru- mentsareweaknotforfourendogenousvariables, Icouldnotformallytestwhethertheinstruments are weak. However, when I use the IV estimation for this particular case, the point estimate of the coe¢ cient for the high-income country import in the low R&D industry is 0.15 and with a standard error of 0.099. Both the point estimates and the standard errors blow up by 20 times (relative to the RE coe¢ cient). This causes one to seriously doubt the strength of the instrumental variables. 70 thecolumns, weseethatthepointestimatesunderdi¤erentspecicationsallsuggest thatthenegativee¤ectofnalgoodimportsismoreseverethanthatofintermediate good imports. However, as the standard errors under the FE model and IV model are much larger, most coe¢ cients now become insignicant in these specications. Nevertheless, the hypothesis tests under all three specications lendsome support to the argument that the e¤ects of intermediate good imports and nal good imports aredi¤erent. Specically,fortheREandFEmodels,Irejectthenullhypothesisthat the e¤ects of intermediate good imports and nal good imports are the same for the earlierperiod(between1988and1989),butIcannotrejectthenullhypothesisforthe later period (between 1992 and 1996). On the other hand, for the IV model, while I reject the null hypothesis in the later period, I cannot do so in the earlier period. The Hausman test results again suggest the validity of the RE specication. Finally, with Table 11 we can see from the coe¢ cients, and especially from the Hausman test results, that my empirical ndings are robust to di¤erent estimation methods. Therefore, although a priori one may have some concerns about poten- tial endogeneity of the import penetration variable due to individual job choice or unobserved industry trends, empirically this is less of a concern. 2.8 Concluding Remarks Thischapterstudieshowimportcompetitiona¤ectsworkersopportunitiestoreceive company training. This is an important issue, as company training has been found 71 tobeaveryimportantsourceofaworkerslifetimehumancapitalaccumulation,and thus will have important implications for ones earnings and job security. My baseline estimation shows that an increase in 1 percentage point of import penetration will result in a 0.006 decrease in the probability of company training. Given that in my sample period, between 1988 and 1996, the average import pene- tration rose from 14% to 18%, this implies a reduction in the probability of training forabout0.024,holdingallotherfactorsconstant. Thisisalargee¤ectgiventhatthe average probability of receiving company training each year is approximately 0.098. I also nd that this e¤ect is concentrated primarily among non-production workers. In order to explain this nding, I rst derive a theoretical model and arrive at two testable hypotheses. Hypothesis 1 argues that domestic producers have less incentivetotraintheirworkerswhentheyfaceimportcompetitionfromlowormiddle income countries. This occurs because these countrieslow-cost products may drive thedomesticrmsproductsoutofthemarketinthefuture,preventingthedomestic rms from recouping their training investment. I nd empirical support for this hypothesis. Specically, while the e¤ect of low or middle income country imports on training is signicantly negative, that of high income country is insignicantly positive. Hypothesis 2 argues that, when import competition increases, rms in High-Tech industrieswillbemorelikelytotraintheirworkersthanrmsinLow-Techindustries. This prediction arises because, through e¤ective training, while rms in High-Tech 72 industries can achieve substantial quality improvement and win the market, rms in Low-Tech industries have little room for quality improvement so they will eventually losethemarkettolow-costforeignproducers. Empirically,however,thedatadoesnot seemtosupportthishypothesis. Idonot ndasignicant di¤erenceinthee¤ects of importsontrainingbetweenHigh-TechandLow-Techindustries,andIalsondthat the negative e¤ect of middle and low income country imports on training is similar in both High-Tech and Low-Tech industries. In addition to the above two hypotheses, I also propose the following Hypothesis 3: Imports of intermediate goods will have a less negative e¤ect on training than that of nal goods imports. This hypothesis follows from arguments made in the foreign outsourcing literature regarding incentives to product upgrading when rms can access intermediate goods imports. I nd empirical support for this hypothesis. Whilethee¤ectofforeignnalgoodimportsontrainingissignicantlynegative,the e¤ect of foreign intermediate good imports is always insignicant. Finally, for the baseline estimation as well as all of the other extensions, I show that my results are robusttoreasonablechangesintheempiricalspecicationsthatareaimedtoaddress potential endogeneity problems. The results in this chapter have implications for the literature. In particular, as argued earlier in the chapter, there have been some industry-level studies that nd evidence of improving labor productivity in face of import competitions or foreign outsourcing. The results of this chapter suggest that this improved productivity is 73 not from increased training. Hence, as discussed earlier, it could come from the exit and entry of rms within the industry (a potentially good scenario), a decrease in the employment level (a bad scenario), or both. The identication of the exact channel is crucial as each has di¤erent implications for U.S. workers, and therefore is an important venue for future study. There are two important caveats regarding this chapter that need to be pointed out. First,althoughinthischapterIshowthatintheU.S.,importcompetitioncauses rms to train their workers less, it may still be a stretch to conclude that U.S. rms alsoslowdowntheirinnovativeactivitiestodevelopnewproducts. Itispossiblethat U.S. rmsstill maintainahighlevel of R&Dactivitiesbutreducespendingonlabor, either through using more capital intensive production techniques or shutting down their domestic production and moving to a foreign site. Therefore, the results in thischapterarenotnecessarilyatoddswithargumentsthatforeigncompetitionwill induce innovation. The important message of this chapter is, however, that foreign competition will impose a cost borne by domestic labor. It is also important to note that this study is conducted only for a specic time period, and it is possible that the e¤ect of imports on training could be di¤erent for di¤erent time periods. In this chapter, I have not conducted my study for years beyond1996becausetheindustryreclassicationin1997doesnotallowmetomerge the industry data before and after 1997. Although in principle one could conduct a similarstudyusingtheNLSY79foryearsafter1997andcompareitsresultswiththis 74 one, one important concern would be that the age distribution of the sample will be di¤erent, making the comparison more tenuous. Specically, since the NLSY79 is a panel data that follows the same cohort across years, while in my study the age of individuals varies between 23 and 39, a study starting in 1998 and ending in 2006, the last year of data currently available, will have a sample of individuals whose age is between 33 and 49. Insummary,itisnotsurprisingthaterceimportcompetition,especiallyfromlow income countries and middle income countries, is an important policy and political issue. Through the reduction in training opportunities for U.S. workers, they may su¤er lower lifetime earnings and face more a unstable future. The results in this chapter seem to validate some of the concerns about exporting good jobs. 75 Chapter 3 Imports, Exports, and the Determination of Employment and Wages in the U.S. Manufacturing Industries 1979-2001 3.1 Introduction In the past three decades, the United States has become an increasingly open econ- omy. For example, between 1979 and 2001, the ratio of manufacturing exports to domestic manufacturing shipments almost doubled from 8.18% to 15.6%, while the ratio of manufacturing imports to domestic manufacturing shipments almost tripled from 8.32% to 24.3%. During the same period, employment in U.S. manufacturing industries fell steadily from 19.8 million in 1979 to 15.6 million in 2001, despite the fact that the population increased from 226 million in 1980 to 281 million in 2000. Furthermore, there have been rising concerns about the widening gap between un- skilled and highly-skilled workers in the United States. In the manufacturing sector, thewageshareofproductionworkers,whoareusuallydeemedasless-skilledworkers, had decreased from 64.7% in 1979 to 57.7% in 2001. 76 These parallel developments have many researchers to link increasing trade with changes in employment and wage patterns. For example, using industry level data, Freeman and Katz (1991, hereafter FK) analyze the impact of imports and exports onemploymentandwagesfortheperiodbetween1958and1984. Also, Kletzerlooks at the e¤ect of imports and exports on employment for the period between 1979 and 1994. Inthischapter,usingtheanalyticframeworkofFKaswellasKletzer,Ianalyze the impact of imports and exports on industry employment and average wages for theperiodbetween1979and2001. Thus,IamextendingFKsworkby17years,and Kletzers work on employment by 7 years. Theory predicts that exports will raise employment while imports will decrease employment. Thisoccursbecauserisingexportscanbeviewedasincreasingdemand for domestic products while rising imports may substitute domestic production. On the other hand, the e¤ect of exports and imports on wages is less obvious. Straight- forward intuition may lead us to expect that the wage response should move in the same direction as the employment response, since higher (lower) demand for labor will often result in higher (lower) wages. However, if we consider that the quality of workers are heterogeneous, then employment and wage changes may not move in the same direction. For example, it is likely that an increase in exports will cause employers to hire new workers whose labor productivity is less than that of older workers. This e¤ect will mitigate and may even reverse the e¤ect of exports and industry average wages. Similarly, an increase in imports may be associated with 77 decreasing or increasing wages. Hence, in both cases the e¤ect of trade on wages is an empirical matter. This chapter is organized as follows. In the next section, I will briey discuss previous literature on trade and employment/wages. Section 3 and 4 discuss the em- pirical framework and the data, respectively. In section 5, I will present my results. Specically,similartopreviousndings,Indthatmoreimportsareassociatedwith decreasedemployment,whilemoreexportsareassociatedwithincreasedemployment. The employment e¤ects on production and non-production workers are similar, and the employment e¤ects of high income country imports and non-high income coun- try imports are also similar. On the other hand, although in general the correlation betweenimports(exports)andwagesapproachesthecorrelationbetweenimports(ex- ports) and employment, there is some evidence that most of the negative correlation between imports and wages comes from non-high income imports. The correlation between high income countries and wages is actually insignicantly positive. Finally, in Section 6, I summarize my ndings and o¤er concluding remarks. 3.2 Literature Review There is a very large literature on the e¤ect of international trade on wages and/or employment. Traditionally,thefocushasmostlybeenonhowtradea¤ectsworkersof di¤erent skill levels. This particular lens may be motivated by the dual observations of the declining real and relative wages of lesser skilled workers since the 1980s, and 78 therisingimportcompetitionfromlowandmiddle-incomecountriesduringthesame period. Probably the most prominent theoretical basis for international trade to have an impactontherisinginequalityinthedomesticlabormarketistheStolper-Samuelson Theorem (1941): in a two-factor, two-good model, trade liberalization changes the relative price of goods, leading to an increase in the real return to the factor used intensively in the production of that good, and a decrease in the real return to the other factor that is used less intensively. Since the U.S. is often perceived as a skill abundant country, trade especially with low income countries is viewed as a potentialsourceforthedi¢ cultsituationfacedbylow-skilledworkerssincethe1980s. Empirically, the importance of the link between trade and wage inequality has not been settled, and di¤erent studies usually come to di¤erent conclusions. 44 For example, based on changes in the relative prices of skilled labor-intensive and un- skilled labor-intensive goods, which is assumed to occur because of foreign compe- tition, Lawrence and Slaughter (1993) argue that trade did not contribute to rising inequality in the 1980s. 45 On the other hand, Borjas, Freeman, and Katz (1992) conclude that trade plays a much more important role in rising wage inequality. In particular, based on their calculated factor contents of traded goods, they estimate 44 For a comprehensive literature review, please refer to Slaughter (2000) and Kletzer (2002). 45 FromtheStolper-SamuelsonTheorem,relativewageoflow-skilledworkersmaydecreasebecause the relative price of low-skilled labor-intensive goods decreases (potentially due to low-cost foreign competition). Lawrence and Slaughter (1993) nd no clear evidence that in the 1980s the relative price of low-skilled labor-intensive goods fell, so they interpreted it as evidence against the trade explanation for wage inequality. 79 that trade increasedthe U.S. e¤ective labor supply of low-skilledworkers much more than that of high-skilled workers. As a result, they estimate that a 15 to 25 percent increase in college graduate/high school graduate wage premiums over the period of 1980-1985 can be attributed to large net imports; although the importance of trade lessened in the later half of the 1980s. Other studies have conclusions which fall in between. For example, similar to Lawrence and Slaughter (1993) but with a much more sophisticated technique, Baldwin and Cain (2000) investigate the relationship between product-price changes and factor-price changes and conclude that increased import competition in industries that intensively used less-educated labor seems to haveplayedanimportantroleinbringingabouttheincreaseinwageinequalityduring the 1980s and early 1990s. Although a substantial portion of the papers focus on the e¤ect of trade on in- equality,therearealsosomepapersthatlookatthelevele¤ectoftradeonthelabor market. In this strand of study, the two papers most closely related to mine are FK andKletzer(2002). 46 AsIdiscussmorefullyinthenextsection,bothpapersusethe same framework and consider the potential labor market response to trade-induced changes in product demand. FK study the period between 1958 and 1984 and look at how trade a¤ects wage and employment determination within industry. Kletzer (2002) extends the FK study for the period between 1979 and 1994, but only looks 46 Twootherpapersthatlookatthelevele¤ectoftradeareRevenga(1992)andBertrand(2003). Revenga nds that changes in import prices have a sizable e¤ect on employment and a smaller yet signicant e¤ect on wages for the period of 1977-1987. Bertrend (2003) nds that increased import penetration causes lower wages for the period of 1975-1992. 80 at the correlation between trade and industry employment levels. In general, both studies nd that increasing exports are associated with a rise in employment, while increasing imports are associated with a fall in employment. FK also nd similar industry wage responses to trade. In my study, I extend the period of study to 2001. Similar to FK, I look at both the employment and wage responses to trade. 3.3 Empirical Framework Inthischapter,IusetheempiricalsetupfromFKandKletzer(2002)toinvestigatethe relationshipbetweenimportsandexportsontheonehand,andindustryemployment and wages on the other. As noted above, I extend these studies by including the period between 1995 and 2001. This empirical setup attempts to model an industry labor market that responds to(potentiallytrade-induced)changesindomesticproductdemand. Onthelaborde- mandside,usingrstdi¤erences,itisassumedthatthedemandforlaborinindustry i in year t (N it ) can be written as: d lnN it =kd lnW it +d lnZ it ; (7) whereW it istheindustrywage,Z it istheshiftinthedemandcurveduetoshiftsin productdemand,k isthewageelasticityoflabordemand,anddistherstdi¤erence operator. 81 Similarly, we can write the determination of labor supply within the industry in the following form: d lnN it =ed lnW it +d lnH it ; (8) where e is the wage elasticity of labor supply and H it is a vector of factors that shift labor supply. From Equations (7) and (8), when the labor market clears, they arrive at the following reduced form equations: d lnN it = (ed lnZ it kd lnH it )=(k +e) =a 1 d lnZ it +a 2 d lnH it ; (9) d lnW it = (d lnZ it +d lnH it )=(k +e) =b 1 d lnZ it +b 2 d lnH it : (10) Equations (9) and (10) relate changes in wages and employment to exogenous changes in labor demand and supply. Following both FK and Kletzer (2002), I use measures of industry shipments (S it ) as measures of exogenous shifts in product de- mand (Z it ). I further decompose S it into domestic market demand, exports, and imports. However, as in Kletzer (2002), I do not have information on exogenous shifts in the labor supply and thus neglect the last term in (9) and (10). 47 47 FK have information on the percentage of immigrants within each industry, and they use this informationtoindicateexogenouschangeinlaborsupply. However, thisinformationisnotavailable after1984. Nevertheless, whenIreplicatedFKresults, Idonotndthattheinclusionoftheselabor supply factors to have a substantial impact on the estimation of the labor demand factors for their sample period. 82 Following FK, dene D it as domestic consumption for industry i in yeart, where D it =shipments(S it )-exports(X it ) + imports(M it ). Then, straightforward rearranging yields S it =D it +X it M it ; which implies dS it =dD it +dX it dM it : Next dene import penetration (impen it ) as the ratio of imports in domestic consumption, M it =D it . Further, note that dS it = S it d lnS it , dD it = D it d lnD it , dX it =X it d lnX it . Then FK obtain 48 d lnS it =w 1it d lnD it w 2it dimpen it +w 3it d lnX it : (11) In (11),w 1it = (S it X it )=S it , w 2it =D it =S it , and w 3it =X it =S it . The purpose of theweightingistoadjusttherelevantchangesforthedi¤erenceinabsolutemagnitude of shipments generated by domestic demand and trade. Substituting (11) into (9) and (10) and dropping the labor supply term lnH it in each equation, we obtain: d lnN it =a 1 w 1it d lnD it a 1 w 2it dimpen it +a 1 w 3it d lnX it +u it ; (9) d lnW it =b 1 w 1it d lnD it b 1 w 2it dimpen it +b 1 w 3it d lnX it +u it : (10) 48 Note that FKs formulation treats imports asymmetrically from domestic consumption and exports; I follow their approach for comparability with their work. 83 Finally, note that the FK formulation implies that the labor market responds similarly to (weighted) changes in shipments due to trade-related factors as it does to those due to domestic factors. In practice, this need not be true. For example, if changes in imports and exports are due to temporary exchange rate volatility, then theire¤ectonwagesandemploymentmaybedi¤erentfromchangesindomesticcon- sumption. Inaddition,whenrmsfacemoreimportcompetitionoraremoreinvolved in an export market, their production technology may change as well. Therefore, in an actual estimation I use the following setup: d lnN it =a 1 w 1it d lnD it +a 2 w 2it dimpen it +a 3 w 3it d lnX it +a 4 dTFP it +a 5 dunion it +a 6 ln(capital=shipment) it + i + t +u it ; (12) d lnW it =b 1 w 1it d lnD it +b 2 w 2it dimpen it +b 3 w 3it d lnX it +b 4 dTFP it +b 5 dunion it +b 6 ln(capital=shipment) it + i + t +u it : (13) Here, I rst allow the e¤ects of imports, exports and domestic demand to be dif- ferent. Ialsoconsiderdi¤erentfactorsthatmaya¤ectemploymentandwages. These factors include the changes in TFP, union coverage rate, and the capital intensity of the industry. I include industry xed e¤ects, i , so my identication comes from withinindustrychangesovertime. Timedummies, t , arealsoincludedtocontrolfor macro shocks that a¤ect all industries. 84 3.4 Data Description I estimate the employment-shipments and wage-shipments equations for the period between 1979 and 2001 for manufacturing industries. I focus on manufacturing in- dustriesasmostofthetradeoccursinthissector. Theinformationonindustrywages andemploymentcomesfromtheNBERManufacturingDatabase,whichalsocontains informationonindustrydomesticshipmentsandcapitalintensity,denedastheratio of industry real capital and industry shipment. For the trade variables, the informa- tion comes from the U.S. import and export data on Robert Feenstras website. 49 Finally, the information on union membership comes from the Union Membership and Coverage Database. 50 Following Kletzer (2002), all variables are aggregated to the 3-digit Census level; 51 FK use 4 digit-level data, but I cannot follow them since the union information is only available at the 3 digit-level after 1983. Table 12 presents the summary statistics of the main variables. As is clear in the table,duringthesampleperiodthereisanaveragedeclineof1.35percentinmanufac- turing employment. The drop in production worker employment is more pronounced at 1.54 percent each year, while the drop in nonproduction worker employment is 49 http://www.econ.ucdavis.edu/faculty/fzfeens/. 50 http://unionstats.gsu.edu. 51 Kletzer (2002) constructs her employment series using the Current Population Survey (CPS). TherearetwopotentialdrawbacksusingtheCPS.First,withCPS,itisdi¢ culttoconstructawage seriesattheindustrylevel. Thatisprobablywhyshedoesnotstudythewageresponse. Second,and probably even more important, it is di¢ cult to construct a reliable yearly employment series at this detailed level of industry classication. The issue becomes especially severe since in the regression one needs to rst di¤erence the employment series, which will increase the noise contained in the dependent variable. 85 about 0.85percent. Onthe otherhand, there appears tobesome increase inaverage real wages during the period. On average there is an increase of 0.5 percent each year, and nonproduction workers seem to enjoy a higher growth in real wages. The patternsofemploymentchangeandwagechangeareconsistentwiththeusualnotion that nonproduction workers are relatively well-o¤during the period. Table 12: Summary Statistics of Key Variables (Census 3-digit Industry Level, Man- ufacturing Industries Only; Year 1979-2001) Mean Standard Deviation Dependent Variables (In Annual Change) ln(total employment) -0.0135 0.0605 ln(production worker employment) -0.0154 0.0650 ln(nonproduction worker employment) -0.0086 0.0713 ln(average real wage) 0.0051 0.0272 ln(production worker real wage) 0.0027 0.0289 ln(nonproduction worker real wage) 0.0068 0.0447 Explanatory Variables ln(shipment) 0.0368 0.0842 Unweighted ln(domestic consumption) 0.0408 0.0867 Unweighted ln(exports) 0.0681 0.2205 Unweighted ln(imports) 0.0830 0.3039 Unweighted ln(HIC imports) 0.0715 0.3007 Unweighted ln(nonHIC imports) 0.1107 0.3799 Weighted ln(domestic consumption) 0.0367 0.0763 Weighted ln(exports) 0.0071 0.0249 Weighted import penetration 0.0069 0.0221 Weighted HIC import penetration 0.0050 0.0168 Weighted nonHIC import penetration 0.0019 0.0131 Change in union coverage -0.0402 0.2482 Change in TFP level 0.0019 0.0445 Number of Industries 73 Number of Observations 1679 Note 1: Here only consider the unweighted means and standard deviations. 2: HIC is abbreviation for "high-income country." 86 Table 12 also shows the trend in other industry variables. During the period 1979-2001, shipments, domestic consumption, exports and imports all rose substan- tially. In particular, the annual increase in exports (6.29%) and imports (8.30%) all outpaced that of domestic consumption (4.08%), indicating the rising importance of trade during the period. Finally, we see that the annual increase in non-high-income country imports (11.07%) is also greater than that of high income country imports (7.15%). 3.5 Results and Discussions Table13showstheresultsfromtheindustrytotalemploymentregressionforthesam- ple period of 1979-2001. In the rst column, I look only at the correlation between changes in shipment and employment. In the second column, I decompose changes inshipmentintothreecomponentschangeindomesticdemand,export,andimport penetrationand allow for di¤erent employment responses to each individual com- ponent. Finally, in column 3, I further allow employment to respond di¤erently to imports from high-income countries and non-high-income countries. FromtherstcolumnofTable13,oneseesthata10percentincreaseinshipments isassociatedwitha5.16percentincreaseinemployment, allelsethesame. However, when we break down di¤erent components of shipments in the second column of Table 13, we see that the responses of employment to these di¤erent components of shipments are somewhat di¤erent. While a 10 percent increase in domestic demand 87 Table 13: Change in Industry Total Employment, Shipments, Domestic Demand, Exports, and Imports: 1979-2001 Dependent Variable: Change in Industry Total Employment (1) (2) (3) Coe¢ cient Coe¢ cient Coe¢ cient (Std. Errors) (Std. Errors) (Std. Errors) (a) Change in ln(shipments) 0.516 (0.070)*** (b) Weighted change in 0.509 0.509 ln(Domestic Demand) (0.075)*** (0.075)*** (c) Weighted change in 0.477 0.477 ln(Exports) (0.069)*** (0.069)*** (d) Weighted change in -0.698 Import Penetration (0.085)*** (e) Weighted change in -0.663 HIC import penetration (0.098)*** (f) Weighted change in -0.727 nonHIC import penetration (0.121)*** (g) Change in TFP level -0.121 -0.131 -0.131 (0.134) (0.131) (0.131) (h) ln(Capital/Shipment) -0.024 -0.023 -0.024 (0.012)* (0.012)* (0.012)* (i) Change in union 0.009 0.009 0.009 coverage rate (0.006)* (0.006) (0.006) Test (b)=(c)=-(d) Reject (p=0.001)*** Test (e)=(f) Accept (p=0.656) Industry Dummies Y Y Y Time Dummies Y Y Y Industry Number 73 73 73 Number of Observations 1679 1679 1679 Note: Signicance of the coe¢ cients***1%,**5%,*10%. 88 and exports implies approximately a 5 percent increase in employment (for domestic consumption, it is 5.09 percent; for exports, it is 4.77 percent), a 10 percent increase in import penetration, holding domestic consumption constant, is associated with a 6.98 percent decline in employment. 5253 At the bottom of the table, we see that the absolute value of the responses to domestic demand, exports, and imports are statistically signicant at the 1% level. Finally, from the third column in Table 13, I ndthatthenegativee¤ectsonemploymentaresimilarforimportsfromhighincome countriesandthosefromnon-highincomecountries. AWaldtestshowsthatIcannot reject the null hypothesis that the e¤ects are the same with a p-value of 0.656. Here, high income countries are dened as countries with GDP per capita that is at least 70% of that of the U.S. during the sampling period, while the remaining countries are categorized as non-high income countries. It is also interesting to discuss how well changes in shipments can help to explain the patterns of change in employment. Using the mean values of the annual changes in weighted domestic demand, exports, and import penetration from Table 12, and the point estimates from the second column of Table 13, we see that the changes in these di¤erent components of shipments imply a 1.725 percent increase in annual 52 When domestic consumption is held constant, a 10 percent increase in import penetration is equivalent to a 10 percent increase in imports. This occurs because import penetration is dened as imports/domesticconsumption,wheredomesticconsumptionisdenedasshipment+import-export. Infact, theassumptionthatdomesticconsumptionholdingconstantimpliesthatimportssubstitute for domestic shipment one for one, assuming foreign demand (exports) remains the same. 53 As argued in the previous footnote, a 10 percent increase in import penetration is equivalent to a 10 percent increase in imports, holding domestic consumption constant. This implies that it is stillsensibletocomparethecoe¢ cientsfor(changesin)domesticconsumption, exports, andimport penetration, albeit the asymmetric treatment of forms for imports in the regression equations. 89 employment. 54 This is in contrast with the actual gure of a 1.34 percent decrease in annual employment. Therefore, although domestic demand, exports, and imports areeachsignicantlyassociatedwithemploymentchanges,theycannotaloneexplain the actual industry employment change. The remaining variables in the specication indicate that more capital-intensive industries have smaller employment growth. There is also indication that union cov- erage increases are positively correlated with employment growth. However, the cor- relationbetweengrowthintotal factorproductivityandemployment is insignicant. I also run separate employment regressions for production and non-production workersinTables14and15,respectively. Imakethisdistinctionbetweenproduction andnon-productionworkers because one wouldexpect that the nature of production andnon-productionjobsisquitedi¤erent,andthereforewecanexpectthatrmsmay adjustemploymentdi¤erentlyforthesetwogroupsofworkers,whenfacingchangesin theindustrysproductdemand. FromthesecondcolumninTable14,weseethatthe e¤ectofdomesticdemand, exportsandimportsonproductionworkeremploymentis similarinmagnitude, whilefromthecorrespondingcolumninTable15thereissome evidence that non-production worker employment is more responsive to changes in import penetration. Again, the employment e¤ects of high and non-high income country imports are similar within both production and non-production workers. 54 During the sampling period, as we can see from Table 1, the mean annual increase in weighted domestic consumption is 3.67 percent, in weighted export is 0.71 percent, in weighted import pen- etration is also 0.69 percent; multiply these gures with the estimated coe¢ cients from the second column of Table 2, we see that the changes in these di¤erent components of shipments imply a 3.67*0.509+0.71*0.477-0.69*0.698=1.725 increase in annual employment. 90 Table14: ChangeinIndustryProductionWorkerEmployment,Shipments,Domestic Demand, Exports, and Imports: 1979-2001 Dependent Variable: Change in Industry Production Worker Employment (1) (2) (3) Coe¢ cient Coe¢ cient Coe¢ cient (Std. Errors) (Std. Errors) (Std. Errors) (a) Change in ln(shipments) 0.512 (0.070)*** (b) Weighted change in 0.508 0.508 ln(Domestic Demand) (0.076)*** (0.076)*** (c) Weighted change in 0.506 0.506 ln(Exports) (0.077)*** (0.077)*** (d) Weighted change in -0.610 Import Penetration (0.066)*** (e) Weighted change in -0.623 HIC import penetration (0.115)*** (f) Weighted change in -0.598 nonHIC import penetration (0.081)*** Test (b)=(c)=-(d) Accept (p=0.253) Test (e)=(f) Accept (p=0.864) Industry Number 73 73 73 Number of Observations 1679 1679 1679 Note 1: See notes in Table 13. 2: All covariates are the same as in Table 13. 91 Table 15: Change in Industry Non-Production Worker Employment, Shipments, Do- mestic Demand, Exports, and Imports: 1979-2001 Dependent Variable: Change in Industry Nonproduction Worker Employment (1) (2) (3) Coe¢ cient Coe¢ cient Coe¢ cient (Std. Errors) (Std. Errors) (Std. Errors) (a) Change in ln(shipments) 0.471 (0.089)*** (b) Weighted change in 0.446 0.446 ln(Domestic Demand) (0.088)*** (0.088)*** (c) Weighted change in 0.474 0.474 ln(Exports) (0.098)*** (0.098)*** (d) Weighted change in -0.918 Import Penetration (0.264)*** (e) Weighted change in -0.774 HIC import penetration (0.196)*** (f) Weighted change in -1.033 nonHIC import penetration (0.357)*** Test (b)=(c)=-(d) Reject (p=0.089)* Test (e)=(f) Accept (p=0.346) Industry Number 73 73 73 Number of Observations 1679 1679 1679 Note: See notes in Table 14. 92 Tables 16to18showtheresults of the estimationonwages forthe sample period of 1979-2001 for production workers and non-production workers, respectively. From theestimatespresentedinthesecondcolumnofTable16,weseethat,similartowhat Ifoundpreviouslyintheemploymentregressions,risingshipments,domesticdemand and exports are all associated with rising wages, while rising import penetration are associated with a decrease in wages. However, if we look at the results for the production workers and non-production workers in the second columns of Tables 17 and 18, we see that their wage response to imports may be somewhat di¤erent. Althoughbothareimpreciselyestimated,Indaninsignicantpositivewageresponse torisingimportpenetrationfornon-productionworkers,butaninsignicantnegative wage response for production workers. Moving to the columns 3 of Tables 16, 17, and 18, we can compare the e¤ects of high income country and non-high income country imports on wages. As we can see, whilethee¤ectofnon-highincomecountryimportsonwagesisalwaysnegative, as expected, the wage e¤ect of high income country imports is always insignicant (but positive). One potential reason that more imports may drive up the average wage might be the composition bias. In particular, the literature commonly nds that when there is a contraction within the industry, the workers who leave rst are usuallyless-skilled. 55 Consequently,whenincreasedimportsdiminishtheemployment 55 Using repeated cross-sections of the Current Population Survey (CPS), Weinberg (2001) nds that there is no systematic relationship between industry wages and low frequency changes in in- dustry labor demand. However, Devereux (2005) shows that a positive correlation between changes industry wages and changes in industry labor demand exist once individual heterogeneity is con- trolled for. This is interpreted as evidence suggests that growing industries attract less skilled 93 Table 16: Change in Industry Average Real Wage, Shipments, Domestic Demand, Exports, and Imports: 1979-2001 Dependent Variable: Change in Industry Average Wage (1) (2) (3) Coe¢ cient Coe¢ cient Coe¢ cient (Std. Errors) (Std. Errors) (Std. Errors) (a) Change in ln(shipments) 0.067 (0.014)*** (b) Weighted change in 0.065 0.065 ln(Domestic Demand) (0.015)*** (0.015)*** (c) Weighted change in 0.089 0.088 ln(Exports) (0.034)** (0.033)** (d) Weighted change in -0.069 Import Penetration (0.040)* (e) Weighted change in 0.059 HIC import penetration (0.055) (f) Weighted change in -0.173 nonHIC import penetration (0.055)*** (g) Change in TFP level 0.045 0.043 0.043 (0.020)** (0.021)** (0.021)** (h) ln(Capital/Shipment) 0.002 0.005 0.0004 (0.005) (0.005) (0.005) (i) Change in union -0.002 -0.002 -0.002 coverage rate (0.002) (0.002) (0.002) Test (b)=(c)=-(d) Accept (p=0.742) Test (e)=(f) Reject (p=0.007)*** Industry Number 73 73 73 Number of Observations 1679 1679 1679 Note: See notes in Table 14. 94 level, as I have conrmed from the employment equations, the workers who stay within the industry are more highly-skilled, and therefore we can observe higher averagewages. Anotherpotentialreasonthatmoreimportsmaydriveuptheaverage wage could be that rms may change (or improve) their production techniques when facingmoreimportcompetition. 56 Consequently,ifmoreimportsinducetechnological improvementsthatraiselaborproductivity,thentheaveragewageswillalsoincrease. To disentangle the two potential reasons, it may be worthwhile to investigate the wage issue using a micro panel dataset. Following the same individuals across time allows us to control for the composition bias that may exist at the industry level. The remaining variables in the specication indicate that industries experiencing higher growth in total factor productivity are also experiencing higher wage growth. However,thecapitalintensityoftheindustriesandchangesinunioncoveragearenot correlated with wage growth. Takingtheresultsforproductionandnon-productionworkersonemploymentand wages together, we see that it is di¢ cult to conclude whether the production or the non-production workers are better o¤in the face of trade. For example, comparing the second columns in Tables 14 and 15, we nd that the employment responses to workers in a manner that biases down the estimated relationship between industry employment and wages in repeated cross-sectional data. Similarly, in their study of cyclical upgrading, McLaughlin and Bils (2001) also arrive at a similar conclusion that the average quality of workers increases as industries decline and decreases as industries grow. 56 For example, McDonald (1994) nd some evidence that import competition force e¢ cient pro- duction; similarly, Lawrence (2000) nd positive correlation between import competition and a rise in TFP. In terms of industry case studies, both Schmitz (2005) and Dunne, Klimek, and Schmitz (2008) nd similar evidence for the U.S. iron ore and cement industry in the 1980s, respectively. A theoretical framework that link trade and innovation is in Acemoglu (2002). 95 Table 17: Change in Industry Production Worker Real Wage, Shipments, Domestic Demand, Exports, and Imports: 1979-2001 Dependent Variable: Change in Industry Production Worker Real Wage (1) (2) (3) Coe¢ cient Coe¢ cient Coe¢ cient (Std. Errors) (Std. Errors) (Std. Errors) (a) Change in ln(shipments) 0.068 (0.017)*** (b) Weighted change in 0.069 0.068 ln(Domestic Demand) (0.020)*** (0.020)*** (c) Weighted change in 0.057 0.056 ln(Exports) (0.036) (0.036) (d) Weighted change in -0.060 Import Penetration (0.047) (e) Weighted change in 0.050 HIC import penetration (0.071) (f) Weighted change in -0.149 nonHIC import penetration (0.066)** Test (b)=(c)=-(d) Accept (p=0.962) Test (e)=(f) Reject (p=0.007)*** Industry Number 73 73 73 Number of Observations 1679 1679 1679 Note: See notes in Table 14. 96 Table 18: Change in Industry Non-Production Worker Real Wage, Shipments, Do- mestic Demand, Exports, and Imports: 1979-2001 Dependent Variable: Change in Industry Nonproduction Worker Real Wage (1) (2) (3) Coe¢ cient Coe¢ cient Coe¢ cient (Std. Errors) (Std. Errors) (Std. Errors) (a) Change in ln(shipments) 0.060 (0.023)*** (b) Weighted change in 0.063 0.062 ln(Domestic Demand) (0.024)*** (0.023)*** (c) Weighted change in 0.088 0.086 ln(Exports) (0.055) (0.054) (d) Weighted change in 0.057 Import Penetration (0.111) (e) Weighted change in 0.206 HIC import penetration (0.127) (f) Weighted change in -0.064 nonHIC import penetration (0.146) Test (b)=(c)=-(d) Accept (p=0.251) Test (e)=(f) Reject (p=0.079)* Industry Number 73 73 73 Number of Observations 1679 1679 1679 Note: See notes in Table 14. 97 imports and exports are qualitatively (or even quantitatively) similar for both types of workers. When we compare the second columns in Tables 17 and 18, we nd that the wage responses to exports are similar for the two types of workers, while the responses to import penetration might be somewhat di¤erent. However, given the large standard errors associated with these estimates, it is still too early to conclude that the production workers are worse o¤in terms of wages. 3.6 Concluding Remarks Inthischapter,usinganempiricalframeworkdevelopedinFreemanandKatz(1991) and Kletzer (2002), I investigate how changes in imports and exports, along with changes in domestic consumption, are associated with changes in industry employ- ment and wages for the sample period of 1979-2001. My work extends the earlier work on employment by 7 years, while extending the study on wages by 17 years. Myresultsshowthatincreasedimportsareassociatedwithdecreasedemployment, whileincreasedexports anddomesticconsumptionareassociatedwithincreasedem- ployment. The employment e¤ects on production and non-production workers are similar, and the employment e¤ects of high income country imports and low income countryimportsarealsosimilar. Conversely,thewageresponsetochangesinimports, exports and domestic consumption is in general qualitatively similar to employment responses. However, there is some evidence that most of the negative correlation be- tweenimportsandwagescomesfromnon-highincomecountryimports. Idonotnd 98 a negative but only an insignicant positive correlation between non-high income country imports and industry average wage; this may be because of composition biasthat is, the average quality of workers is counter-cyclical: quality increases as industriesdeclineanddecreasesasindustriesgrow. Anotherpotentialreasoncouldbe that rms adjust their production techniques when facing more import competition. Itisworthwhiletorelatetheresultsinthischaptertomyworkonimportcompe- tition and training in the previous chapter. In particular, there are two points that I would like to make. First, my nding on employment is consistent with my rationale for nding a negative e¤ect of imports on training. Specically, the main argument inthepreviouschapterontrainingisthatwhenfacingimportcompetition, domestic rms may want to quitand therefore stop training workers. In this chapter, I nd a negative correlation between change in import penetration and change in employ- ment, suggesting that U.S. manufactures are indeed cutting their labor force in the face of import competition. Second, in my work on training I nd that most of the negative e¤ect of imports on training is on non-production workers, while the e¤ect on production workers is never signicant. My results in this chapter show that this nding on company training does not necessarily imply that nonproduction workers are relatively worse o¤. When facing more import competition, production workers also su¤er due to lower employment and wages. Finally, there are two important caveats regarding the work in this chapter that merit pointing out. First, in accordance with previous work, I use measures of in- 99 dustry actual shipments, decomposed into domestic market demand, exports and imports as measures of exogenous shifts in product demand. However, there may be concerns that these measures on di¤erent components of shipments are not necessar- ily exogenous and may also depend on the domestic labor market conditions which are correlated with the industry labor market. Hence, my empirical framework may su¤er from a simultaneity bias. 57 To deal with this issue, in the future it would be worthwhile to nd instruments that cause exogenous shifts in product demand in order to more convincingly identify the e¤ect. Second, I nd a potential positive correlation between industry average wages and high income country imports, and as argued earlier, this may be due to a compositional bias or a change in production technique. Inthefuture,itmaybeworthwhiletodisentanglethetwoe¤ectsbyusing micro labor panel datasets, such as the Panel Survey of Income Dynamics (PSID) or the National Longitudinal Survey (NLS), to investigate the issue. In particular, once we can follow the same individuals across time, we should be less concerned about the potential compositional bias that plagues industry level data. 57 To tackle the simultaneity issue, FK show that when the supply curve of industry output is at, so that the industry prices depend only on production costs (assuming there is only labor cost), thenasimpleOLSestimationwillnotleadtoalargebias. Thatis, theyarguethatthesimultaneity problem is not severe when running regressions such as the ones in this chapter. 100 Chapter 4 Conclusion In this dissertation, I analyze the e¤ects of trade on the domestic labor market. I extend the current literature in two dimensions. First, I investigate the e¤ect of import competition on company training within United States (U.S.) manufacturing industries. Thisnewfocusonthee¤ectsofimportsoncompanytrainingisimportant as such training is an essential factor in earnings and job security. Second, I extend FreemanandKatzs(1991)andKletzers(2002)studiesontheemploymentandwage e¤ects of trade through the year 2001. This extension is signicant, as most studies on this issue focus on the years before the mid-1990s. Inmyanalysisofthee¤ectofimportcompetitiononcompanytraining,Indthat importcompetitionhasanegativee¤ectontraining. Itisimportanttonotethatmost ofthesenegativee¤ectscomefromlow-andmiddle-incomecountryimports. Infact, I do not nd a signicant di¤erence between the e¤ect of imports in high-technology and low-technology industries. I also nd that nal goods imports in an industry haveamuchstrongernegativee¤ectontrainingthandointermediategoodsimports. My ndings seem to validate some concerns about import competition, especially from low- or middle- income countries, since reduced training opportunities for U.S. workers may have a negative e¤ect on their lifetime labor market outcomes. 101 It is clear from my research that nonproduction workers bear the brunt of this negativee¤ectoncompanytraining,whilethee¤ectonproductionworkersisinsignif- icant. This result stands in contrast with the popular notion that imports mainly adversely a¤ect production workers. In my chapter on the employment and wage e¤ects of trade, I demonstrate that the results in my training study do not tell the full story. Typically, production workers might su¤er lower employment and wage levels when faced with import competition. On the other hand, rising demand for exports, potentially through their e¤ect on mounting domestic product demand, is associatedwithincreasesinindustryemploymentandwagelevelsforbothproduction and nonproduction workers. This suggests that when we discuss the e¤ect of trade onemploymentandwages,weshouldnotoverlookthepositivee¤ectthatarisesfrom escalating foreign demand. The ndings in this dissertation suggest several interesting directions that merit future exploration. For instance, I nd that imports tend to have a negative e¤ect on company training. At rst sight, this stands in contrast with several papers ndings that import competition increases labor productivity, as one might expect thatcompanytrainingiscentraltolaborproductivity. However,asIargueinchapter 2, higher productivity can be achieved through di¤erent routes, for example the exit and entry of rms, a decrease in employment level, and so on. Identifying the exact channelthatbringsaboutincreasedlaborproductivityiscrucialaseachhasdi¤erent implications for workers, and hence each is an important element for future study. 102 Another very interesting nding in this dissertation is the insignicant positive correlation among nonproduction workers between industry average wages and im- ports. At rst glance, this is also counterintuitive, result as one would expect rising imports tosuppress wages throughnegative e¤ects ondomestic industryproduct de- mand. However,itisfeasiblethattheaveragequalityofworkcouldincreasewhenan industrydeclines,ifthosewhoarerstlaido¤arethosewithlimitedskills. Thiscom- position bias would support my counterintuitive ndings. Conversely, it is possible that rms may change their practices or production techniques in the face of import competition, whichwould lead to higher wage. I suggest using a labor panel dataset, as this allows us tocontrol forindividual unobservedcharacteristics, andwill helpus clarify whether a scenario based on composition bias is likely. To see if a situation basedonchangingworkpracticesorproductiontechniquesismorerelevant,onemay want to follow examples such as Schitmz (2005) and Dunne, Klimek, and Schitmz (2008) and conduct detailed industry studies. Thee¤ectoftradeontheU.S.labormarketisanintriguingeldforresearch. This isevenmoretrueinrecentyears,giventhecontinuingriseintradeinthemanufactur- ing and service sectors. My dissertation suggests that workers could potentially bear greater costs in the face of increased globalization. How to mitigate these potential negative e¤ects is a crucial policy question. 103 References Acemoglu,Daron(2002).TechnicalChange,InequalityandLaborMarket.Journal of Economic Literature, vol. 22: pp. 7-72. Acemoglu, D. and J.S. Pischke (1998). Why do rms train: theory and evidence. The Quarterly Journal of Economics 113(3): 79119. Acemoglu,D.andJ.S.Pischke(1999a).BeyondBecker: Traininginimperfectlabor markets.Economic Journal 109: F112F142. Acemoglu, D. and J.S. Pischke (1999b). The structure of wages and investment in general training.Journal of Political Economy 107(3), 539572. Aghion,Philippe,andP.Howitt(1992)Amodelofgrowththroughcreativedestruc- tion.Econometrica, 60: 323-51. Aghion,Philippe,RBlundell,RGri¢ th,PHowitt,andS.Prantl(2004).Entryand productivitygrowth: evidencefrommicro-levelpaneldata.Journal of the European Economic Association, Papers and Proceedings, 2: 265-276. Aghion,Philippe,RBlundell,RGri¢ th,PHowitt,andS.Prantl.(2009)Thee¤ects of entry on incumbent innovation and productivity.The Review of Economics and Statistics, (February) 91(1): 20-32. Amiti,M.andS.Wei(2006).Serviceo¤shoringandproductivity: evidencefromthe United States?NBER Working Paper, No. 11926. Autor, David, Levy, Frank, and R. Murnane (2003). The skill content of recent technological change: an empirical exploration, Quarterly Journal of Economics 118(4): 1279-1334. Aw, B.Y., B. Roberts, andT. Winston(2007). Export market participation, invest- mentsinr&dandworkertraining,andtheevolutionofrmproductivity.TheWorld Economy 83-104. Baldwin, Robert, and Glen Cain (2000). Shifts in relative U.S. wages: The role of trade, technology, andfactorendowments.The Review of Economics and Statistics, Vol. 82, No. 4: pp. 580-595. Bartel, A. (1995). Training, wage growth and job performance: evidence from a company database.Journal of Labor Economics, July 1995. Bartel, A. and N. Sicherman (1998). "Technological change and the skill acquisition of young workers." Journal of Labor Economics, October 1998. Baum, C.F., M.E. Scha¤er, and S. Stillman (2007). Enhanced routines for Instru- mental Variables/ GMM estimation and testing.Stata Journal 7:465-506. Becker, G. S. (1962). Investment in human capital: Atheoretical analysis.Journal of Political Economy 70: 949. Supplement (October). 104 Bernard, A., Jansen, J.B., and P. Schott (2006). Survival of the best t: low wage competition and the (uneven) growth of U.S. manufacturing plants." Journal of In- ternational Economics 68: 219-237. Bertrand, Marianne (2004). From the invisible handshake to the invisible hand? How import competition changes the employment relationship.Journal of Labor Economics, vol. 22: pp. 723-765. Blanchower, D. and L. Lynch (1994). "Training at work: A comparison of U.S. andBritishyouths."PrivateSectorandSkill Formation: International Comparisons, edited by Lisa Lynch, National Bureau of Economic Research, Chicago: Univ. of Chicago Press, 233-260. Blinder, A. (2009). "How many U.S. jobs might be o¤shorable,World Economics, April-June, 2009, 10(2): 41-78. Booth,AlisonL.&M.L.Bryan(2002).Whopaysforgeneraltraining? newevidence for British men and women.IZA Discussion Papers 486. Borjas, George, Richard Freeman, and Lawrence Katz (1992). On the labor market e¤ects of immigration and trade.In Immigration and the Workforce: Economic ConsequencesfortheUnitedStatesandSourceAreas,G.J.BorjasandR.B.Freeman, eds. Chicago: University of Chicago Press, pp. 213-244. Card, D. (1999). The causal e¤ect of education on earnings.In Orley Ashenfelter and David Card, editors, Handbook of Labor Economics Volume 3A. Amsterdam: Elsevier, 1999. Devereux, Paul (2005). Do employers provide insurance against low frequency shocks? Industry employment and industry wages.Journal of Labor Economics, vol. 23: pp. 313-340. Dixit,Avinash,andJ.Stiglitz(1977),Monopolisticcompetitionandoptimumprod- uct diversity.American Economic Review, 67: 297-308. Dunne, Timothy, Shawn Klimek, and James Schmitz (2008). Does foreign compe- tition spur productivity? Evidence from post WWII U.S. cement manufacturing. Working Paper, Federal Reserve Bank Minneapolis. Feenstra, R. and G. Hanson (1996). Foreign Investment, Outsourcing and Relative Wages,in Robert C. Feenstra, Gene M. Grossman and Douglas A. Irwin, eds., The PoliticalEconomyofTradePolicy: PapersinHonorofJagdishBhagwati,MITPress, 1996, 89-127. Feenstra, R.C. and G. H. Hanson (1999). The impact of outsourcing and high- technology capital on wages: estimates for the United States, 1979-1990.Quarterly Journal of Economics, 114 (3). Finegold,D.IsEducationtheAnswer: theSkillsoftheUSWorkforceinaChanging Global Economy,in OToole, J. and Lawler, E. (eds), Work in America II, SHRM 2006. 105 Frazis, H. and M. A. Loewenstein (2005). Reexamining the returns to training: Functionalform,magnitude,andinterpretation.Journal of HumanResources 40(2), 453476. Freeman, Richard, and Lawrence Katz (1991). Industrial wage and employment determination in an open economy.In Immigration, Trade, and the Labor Market, J.W. Abowd and R.B. Freeman, eds. Chicago: University of Chicago Press, pp. 235- 260. Glass, Amy Jocelyn, and K. Saggi (2001). Innovation and wage e¤ects of interna- tional outsourcing.European Economic Review 45(1) 67-86. Hart, O. (1983). The market mechanism as an incentive scheme.Bell Journal of Economics 14 (autumn): 366-82. Hashimoto, M. (1981). Firm-specic investment as a shared investment.American Economic Review, 475482. Katz, L. and K. Murphy (1992). Changes in the structure of wages, 1963-1987: supply and demand factors.Quarterly Journal of Economics, CVII, 35-78. Kletzer,Lori(2002).Imports, Exports, andJobs: WhatDoesTradeMeanforEmploy- ment and Job Loss? Kalamazoo, Michigan. W.E. Upjohn Institute for Employment Research. Lawrence, Robert, and Matthew Slaughter (1993). International trade and Ameri- can wages in the 1980s: Giant sucking sound or small hiccup?Brooking Papers on Economic Activity: Microeconomics 1993: pp. 161-210. Lawrence,Robert(2000).Doesakickinthepantsgetyougoingordoesitjusthurt? TheimpactofinternationalcompetitionontechnologicalchangeinU.S.manufactur- ing.In The Impact of International Trade on Wages, Robert Feenstra, ed. Chicago: University of Chicago Press, pp. 197-224. Loewstein, Mark, and J. Spletzer (1997). Delayed formal on-the-job training.In- dustrial and Labor Relations Review, October 1997, 82-99. Loewstein, Mark, and James Spletzer (1998), Dividing the costs and returns to general training.Journal of Labor Economics 16: 142-71. Loewstein, Mark, and J. Spletzer (1999). General and specic training: evidence and implications.Journal of Human Resources 34(4): 710-33. Lynch, L. (1991). The role of o¤-the-job vs. on-the-job training for the mobility of women workers.American Economic Review Papers and Proceedings, pp. 151156. McDonald, J.M. (1994) Does import competition force e¢ cient production?The Review of Economics and Statistics, 76(4) (November): 721-727. McLaughlin,Kenneth,andMarkBils(2001).Interindustrymobilityandthecyclical upgrading of labor.Journal of Labor Economics, vol. 19: pp. 94-135. Melitz,M.(2003).Theimpactoftradeonintra-industryreallocationsandaggregate industry productivity.Econometrica, 2003. 106 Olsen, K.B. (2006) Productivity impacts of o¤shoring and outsourcing: A review. OECD Working Paper. Porter, M. (1990). The Competitive Advantage of Nations. New York. Free Press. Revenga, Ana (1992). Exporting jobs? The impact of import competition on em- ploymentandwagesinU.S.manufacturing.Quarterly Journal of Economics 107(1): 255-284. Romer, Paul (1990). Endogenous technological change.Journal of Political Econ- omy, 98 (1990): 71-102. Royalty,A.B.(1996).Thee¤ectsofjobturnoveronthetrainingofmenandwomen. Industrial and Labor Relations Review (April) 506-21. Schmitz, James (2005). What determines productivity? Lessons from the dramatic recovery of the U.S. and Canadian iron ore industries following their early 1980s crisis, Journal of Political Economy, vol.113: pp582-625. Slaughter, Matthew (2000). What are the results of product-price studies and what canwelearnfromtheirdi¤erences?InThe Impact of International Trade on Wages, Robert Feenstra, ed. Chicago: University of Chicago Press, pp. 129-169. Stock, J. H., and M. Yogo (2005). Testing for weak instruments in Linear IV re- gression.In Identication and Inference for Econometric Models: Essays in Honor of Thomas Rothenberg, ed. D.W. Andrews and J. H. Stock, 80108. Cambridge Uni- versity Press. Stolper, Wolfgang, and Paul Samuelson (1941). Protection and real wages.Review of Economic Studies 9: pp. 58-73. Weinberg,Bruce(2001).Longtermcontractswithindustryspecichumancapital. Journal of Labor Economics, vol. 19: pp. 231-264. 107 Appendix: A Model for the Competition E¤ect of Import Competition on Training In this appendix, I will discuss in detail the model in Chapter 2 that describes the e¤ectofincreasedimportcompetitionontrainingfordi¤erentparametersk,j,s,w d , andw f . Recallthatkandjarethedomesticandforeignqualitylevels,respectively,at theendof periodt;s is thesteps of qualityimprovement thatthedomesticproducer canachieveconditionalone¤ectivetraining; w d andw f arethedomesticandforeign wage levels, respectively; and they are also the unit production cost of domestic and foreignintermediateinputs,respectively. Fortheconvenienceofillustration,equation (4) in the main text of Chapter 2 is replicated below as equation (A1). dz dq = 1 c h c (k 1 +s;j;w d ;w f ) m (k 1 +s;w d ) (A1) c (k 1;j;w d ;w f ) + m (k 1;w d ) i In the above equation, it is obvious that the second and the fourth terms on the RHS of (A1) only depend on the parameters associated with the domestic industry, while the rst and third terms also depend on the parameters describing the foreign competitors. Sincethefocusisonhowthecharacteristicsofforeigncompetitorsmay a¤ectthedomesticrmsincentivetotrainworkers,Iwillfocusontherstandthird terms on the RHS of equation (A1). 108 I will start with the rst term, c (k 1 +s;j;w d ;w f ), which corresponds to the prot level the domestic rm could obtain when training is e¤ective and a foreign competitor shows up. From the main text, we know that this prot level can take on one of the two values. In particular, when A d (t) A f (t) 1 w f = k1+s A 0 j A 0 1 w f = k1+sj 1 w f <w d , c (k 1 +s;j;w d ;w f ) = 0. This occurs when the domestic rm has a higher quality-adjusted production cost and hence been driven out of the market. On the other hand, when A d (t) A f (t) 1 w f = k1+sj 1 w f w d , c (k 1 + s;j;w d ;w f ) = m (k 1 +s;w d ). This occurs when the domestic rm has a lower (quality-adjusted) production cost, so it will remain the monopoly. The possible values for the third term, c (k 1;j;w d ;w f ), are also similar. Specically, when A d (t) A f (t) 1 w f = k1 A 0 j A 0 1 w f = k1j 1 w f < w d , c (k 1;j;w d ;w f ) = 0; on the other hand, when A d (t) A f (t) 1 w f = k1j 1 w f w d , c (k 1;j;w d ;w f ) = m (k 1;w d ). The above discussion about the potential values of c (k 1 +s;j;w d ;w f ) and c (k 1;j;w d ;w f ) impliesthatbasedondi¤erentvaluesofk,j,s,w d , andw f , there are three possible cases for the e¤ect of increased import competition on training: Case 1: k1j 1 w f < k1+sj 1 w f <w d Inthiscase,regardlessofthee¤ectivenessoftraining,thedomesticrmwillalways haveahigherquality-adjustedproductioncost, soitwillalwayslosethemarketonce 109 apotentialentrantshowsup. Consequently,fromequation(A1),itstrainingdecision when there is an increase in import competition, i.e. a rise in q, the probability that a potential foreign competitor shows up, will be: dz dq = 1 c h c (k 1 +s;j;w d ;w f ) m (k 1 +s;w d ) c (k 1;j;w d ;w f ) + m (k 1;w d ) i = 1 c [0 m (k 1 +s;w d ) 0 + m (k 1;w d )] = 1 c [ m (k 1 +s;w d ) + m (k 1;w d )]< 0 The last inequality comes from the fact that the domestic rms monopoly prot will be greater when its quality is better (note that s = 1or 2 so it is greater than zero). Theintuitionforthisisstraightforward: Onceaforeigncompanyshowsup,the domestic rm investment in human capital will be in vain. Therefore, the domestic rmwilldecreaseitstraininginvestmentbasedonanincreaseintheprobabilitythat a foreign company will show up. Case 2: k1j 1 w f <w d < k1+sj 1 w f In this case, when the training is e¤ective, the domestic rm will have lower quality-adjusted production costs, so it will win the market and earn monopoly rent. However, if the training is ine¤ective, the domestic rm will have higher quality- adjusted production costs, so it will lose the market once a potential entrant shows 110 up. Therefore, in terms of equation (A1), the training intensity the rm will choose when there is more import competition is: dz dq = 1 c h c (k 1 +s;j;w d ;w f ) m (k 1 +s;w d ) c (k 1;j;w d ;w f ) + m (k 1;w d ) i = 1 c [ m (k 1 +s;w d ) m (k 1 +s;w d ) 0 + m (k 1;w d )] = 1 c m (k 1;w d )< 0 Under this second case, the domestic rm will want to increase its training in- vestment when there is rising import competition. This result occurs because the domestic rm could win the market if its training is e¤ective, so it has the incentive to provide more training to increase its chance to survive the competition once the foreign competition shows up. Case 3: w d < k1j 1 w f < k1+sj 1 w f In this case, regardless of the e¤ectiveness of training, the domestic rms pro- duction cost is always lower than that of the foreign rm, so the domestic rm will always win the market and earn monopoly rent. Consequently, equation (A1) will have the following form: 111 dz dq = 1 c h c (k 1 +s;j;w d ;w f ) m (k 1 +s;w d ) c (k 1;j;w d ;w f ) + m (k 1;w d ) i = 1 c h c (k 1 +s;w d ) m (k 1 +s;w d ) m (k 1;w d ) + m (k 1;w d ) i = 0 In this case, when the probability of a potential foreign competitor arises, the domesticrmwill notchangeitstrainingdecision. Thisoccursbecausethedomestic rm is so superior that it does not require quality upgrading to defeat the potential entrants. Therefore, its training decision will not be a¤ected by the arrival of poten- tial competitors. However, in this chapter I ignore this third case for two reasons. First,onewouldonlybeinterestedinthee¤ectofimportcompetitionwhenpotential foreign entrants could actually impose a threat to domestic rms. From a domestic rms perspective, it would not be concerned about foreign rms unless they present direct competition. Second, empirically it is di¢ cult to test this no-e¤ect of import competition when domestic quality is su¢ ciently superior. The reason is that in this case, foreign rms will never enter, so we cannot observe an increasein import competition. Therefore, onwards I will only consider cases 1 and 2. 112 NowIwilldiscussthecomparativestaticsofthemodel. Comparingtheconditions inwhichcase1and2wouldoccur, wecanseethatthemaindi¤erenceisthatincase 1, e¤ective training, or equivalently, quality upgrading, cannot save the domestic rmfrompotentialentrants,whileincase2e¤ectivetrainingcan. Mathematically,it simplymeansthatwearecomparingthevaluesofw d and k1+sj 1 w f . Whenthe former is larger, we are in case 1 where increased import competition will discourage training,whilewewillbeincase2whereincreasedimportcompetitionwillencourage training if the latter has a larger value. Holding the domestic wage level, w d , and end of periodt 1 quality level,k 1, xed, it is then obvious that, holding all other parameters xed, we can perform the following comparative statics: When the foreign wage level w f is lower, we are more likely to be in case 1 where increased import competition will discourage training. This is the basis of Hypothesis 1 which argues that the negative e¤ect of import competition on company training will be more severe for imports from middle or low income countries than those from high income countries. When the number of steps of quality improvement under e¤ective training is larger, i.e. a larger s, we are more likely to be in case 2 where increased im- port competition encourages training. This observation constitute the basis of Hypothesis 2 which argues that the negative e¤ect of import competition on company training will be less severe for High-Tech industries where there are 113 many opportunities for improvement. Note that I dene an industry to be High-Techwhen s is larger. Finally, one can see that when the foreign quality level, j, is higher, case 1 is more likely to occur. In this chapter, since I do not have measures for quality levels for domestic or foreign products, I have not empirically tested this.

Abstract (if available)

Abstract In this dissertation, I analyze the effects of trade on the U.S. domestic labor market. I extend the current literature in two dimensions. First, I investigate the effect of import competition on company training within United States manufacturing industries. Second, I extend Freeman and Katz's (1991) and Kletzer's (2002) studies on the employment and wage effects of trade through the year 2001.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Beyond spatial mismatch: immigrant employment in urban America

PDF

Vertical integration and two-sided market pricing: evidence from the video game industry

PDF

Essays on labor and development economics

PDF

Essays on family and labor economics

PDF

Growth, trade and structural change in low income industrializing economies

PDF

Inter-generational tranfsers and bargaining power within the family in South Korea

PDF

Essays in development economics

PDF

The EITC, labor supply, and child development

PDF

The impact of trade liberlaization on firm performance in developing countries -- new evidence from Pakistani manufacturing sector

PDF

Essays on development and health economics: social media and education policy

PDF

The economic and political impacts of U.S. federal carbon emissions trading policy across households, sectors and states

PDF

Essays on the empirics of risk and time preferences in Indonesia

PDF

The impact of 9/11 on the judicial treatment of Middle Eastern asylum applicants in the U.S. courts

PDF

The political economy of implementation: intellectual property rights protection across the world

PDF

The determinants and measurement of human capital

PDF

Let us fake out a frontier: dissent and the settler colonial imaginary in U.S. literature after 1945

Asset Metadata

Creator Li, Hao-Chung (author)

Core Title Trade, training, employment, and wages: evidence from the U.S. manufacturing industry

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Economics

Publication Date 05/25/2010

Defense Date 05/03/2010

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag company training,employment,import competition,OAI-PMH Harvest,trade,wage

Place Name USA (countries)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Dekle, Robert (committee chair), Ham, John C. (committee chair), Painter, Gary D. (committee member), Strauss, John (committee member)

Creator Email haochunl@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-m3099

Unique identifier UC1441037

Identifier etd-Li-3767 (filename),usctheses-m40 (legacy collection record id),usctheses-c127-332417 (legacy record id),usctheses-m3099 (legacy record id)

Legacy Identifier etd-Li-3767.pdf

Dmrecord 332417

Document Type Dissertation

Rights Li, Hao-Chung

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Repository Name Libraries, University of Southern California

Repository Location Los Angeles, California

Repository Email cisadmin@lib.usc.edu