Worker Training and Foreign Competition:
Evidence from the US Manufacturing Sector
Vasilios D. Kosteas
Cleveland State University
2121 Euclid Avenue, RT 1707
Cleveland, OH 44115-2214
b.kosteas@csuohio.edu
Tel: 216-687-4526, fax: 216-687-9206
Abstract
Relatively little attention has been paid to the effect of trade on investment in human capital through job
training. We examine the relationship between rising imports and the amount of time workers in the US
manufacturing sector spend in training events. Using industry level exchange rates as an instrument for
import shares, we find that rising imports lead workers to spend more time in training to upgrade their
skills, consistent with firms responding to foreign competition by increasing their technology adoption. In
contrast, rising imports lead to a lower probability of investing in training for the purpose of career
advancement.
JEL Codes: F16, J31
Key Words: training, imports, US manufacturing
1. Introduction
Past research has shown that training constitutes an important source of wage increases
(Veum 1999). Much of the research on training provision and participation has focused on
explaining why it is that firms pay for general training (see for example Acemoglu and Piscke
1998 and Aghion et al 1999, amongst others), while other studies work towards understanding
the determinants of enrollment in training. Relatively little attention has been paid to the
possible impact of foreign competition on training provision and enrollment. While the
substantial literature examining the labor market effects of trade has focused primarily on
employment, wages and wage inequality, rising foreign competition might result in greater
enrollment in training by inducing firms to require more training in order to remain competitive
and preserve market share. It has been shown that trade can spur innovative activity and
technology adoption (Bloom et al 2008, Gorodnichenko et al 2011) which in turn leads to greater
training provision (Zeytinoglu et. al. 2009, Xu et al 2011).
Clearly, enrollment in training will increase if the firm makes training a required part of
continued employment and advancement. Even when not compelled to do so, increased access
to training might induce more participation in training events if either the financial or time cost
of participating in employer sponsored training is lower than alternative training opportunities.
Independent of these training provision effects, the effect of greater product market competition
on participation in training depends on how it affects the expected returns to investing in human
capital acquisition. Workers who believe the greater competition increases the likelihood of
suffering a layoff and also believe that training will not help them retain their job or find a new
one may be less likely to enroll in training events. By contrast, workers who believe that training
will help them keep their jobs in the face of higher layoffs or find a new one if they are in fact
laid off are more likely to participate in training. Thus, the effect of foreign trade on worker’s
training decisions may not be uniform.
The present paper contributes to the literature by analyzing the effect of foreign product
market competition on the extensive (whether the individual participates in any training events)
and the intensive margin (weeks spent in training) of enrollment in training programs by US
workers in the manufacturing sector. This is the first paper to focus on the link between foreign
competition and participation in training by US workers. Controlling for training provision by
the firm, we find that workers spend more time in training that is associated with the adoption of
1 new technology or general upgrading of skills when faced with greater foreign competition. In
contrast, they are less likely to enroll in training for the purpose of career advancement with their
current employer.
There are relatively few studies examining the link between job training and product
market competition. We are aware of only one paper investigating the link between foreign
competition and training using U.S data. As part of his doctoral dissertation, Li (2010) examines
the effect of foreign competition on the incidence of company provided training using the 19881996 waves of the NLSY79 data, restricting his analysis to the sample of men only. The author
finds that rising imports result in a lower probability that a worker’s employer will offer
company sponsored training. A handful of papers investigate the link between product market
competition and training using Canadian data. Using the Workplace and Employee Survey
(WES), Zeytinoglu and Cooke (2009) find that workers in firms implementing information
technology and those facing competition from foreign firms are more likely to train. Xu and Lin
(2011) also employ the WES data and report a positive impact between international competition
and participation in employer provided training, in addition to a positive link between
technological innovation and training incidence. It should be noted that the foreign competition
variables used in these studies are developed from self-reported measures of the perceived
degree of foreign competition faced by the firm. Tat-kei and Ng (2011) find a positive impact on
training provision from competition in general.
This paper also ties into the literature examining the effects of foreign competition on
labor markets by analyzing worker level data. This is a relatively small, but growing literature as
older studies in this area have generally relied on industry-level data. Worker-level data has
been used to investigate the effect of globalization on wages (Kosteas 2008, Ebenstein et al
2011), offshoring on the skill-premium (Tempesti 2010), the sensitivity of wages to the
unemployment rate as foreign competition increases (Bertrand 2004) for US workers in the
manufacturing sector, and the effect of trade induced job displacement on wages in occupations
receiving these displaced workers (Kosteas and Park 2013). As with the present analysis, each
of these studies merges industry-level trade data with individual-level data using either NLSY
(Kosteas 2008, Kosteas and Park 2013), CPS data (Bertrand 2004, Ebenstein et al 2011 and
Tempesti 2010), or the Panel Study of Income and Dynamics (Kosteas and Park 2013).
2 2. Theoretical Background and Estimation Strategy
While theorists have tackled the question of how competition affects firm behaviors such
as innovation, the theoretical literature is relatively silent on how product market competition
affects workers’ decisions about whether (and how much) to invest in their human capital via job
training. However, the models dedicated to analyzing firm behavior do provide some important
insights for the question at hand. Ultimately, the issue is whether greater competition raises or
lowers the return to training. The answer rests on the tension between two mechanisms. First,
greater competition from foreign producers might increase the level of job displacement in the
worker’s industry (Addison et al 2000 and Kletzer 1998), which decreases the expected return to
training by lowering the expected amount of time spent working. At the same time the
probability an individual worker is laid off should decrease with training. In industries
characterized by low levels of layoffs and job displacement, this benefit from training may be
relatively small. However, in industries characterized by high levels of job loss, the job saving
aspect (from the worker’s perspective) of training may be relatively large. Furthermore, the
relative strength of these two effects is likely to vary across individuals. More productive
workers may be able to keep from being laid off by investing in training while less productive
workers may feel that training will do nothing to save their jobs. Conversely, more productive
workers may feel more secure in their employment, feeling less pressure to participate in
additional training in order to keep their jobs. Karabay and McLaren (2010) develop a model
where trade leads to lower wages but less wage volatility while offshoring raises wages and
increases income volatility for workers in richer countries while Krishna and Senses (2011) find
evidence that trade is associated with higher income risk. This increased risk is not due solely to
job loss and thus represents a broader measure of labor market uncertainty for workers in import
competing industries. This increased risk can lead to decreased investment for risk averse
individuals if the training has any significant cost to the individual.
Foreign competition may also affect training incentives through its impact on wage
inequality. Falvey et al (2010) develop a model showing that the increase in the skill premium
that results from trade liberalization induces more workers to become skilled. An important
insight from their analysis is that anticipated liberalization episodes will induce skill upgrading
prior to liberalization. Thus, focusing on contemporaneous effects of rising imports will
understate the effect on skill acquisition. While they model skill acquisition as formal schooling,
3 these insights extend to post-schooling human capital investments. In another recent theoretical
model (Sly 2010), workers increase their skill in response to falling consumer prices generated as
a result of lower trade costs.
While the above discussion focuses on workers’ decisions regarding voluntary training,
trade might also affect enrollment in training through firms’ decisions whether to require certain
training events. For example, firms responding to increased foreign competition by
implementing new technology are likely to require their workers learn to operate new equipment
and properly execute new processes. There is a growing theoretical and empirical literature
examining the link between competition and innovation. Blundell et al (1999) and Nickel (1996)
find that greater product market competition leads to greater innovation. More recent work by
Aghion et al (2005) finds evidence of an inverted U-shape relationship between innovation and
competition and develops a theoretical model to explain this empirical observation, while
Aghion et al (2009) shows that firms closer to the technological frontier will respond to greater
competition by increasing innovation efforts and raising productivity while the industry laggards
will decline in productivity. Focusing specifically on competition due to foreign trade, Long et
al (2011) find that trade liberalization increases aggregate R&D when trade costs are low and
decreases it when they are high.
2.1 A basic model of investment in training
So far, the discussion has focused on training in general. However trade might exhibit
different pressures on various types of training. This section develops a simple model to analyze
how an increase in foreign competition affects the optimal level of investment in training. The
analysis distinguishes between three types of training: training that is specific to the current job,
training that is specific to a new job and general training which applies to both. In all cases,
foreign competition refers specifically to the degree of foreign competition in the worker’s
current industry of employment. To keep the model simple, I assume that both foreign
competition and investment in training affect only the probability the worker suffers a layoff
from her current job and the probability that she will find a new job in the event that she does
lose her current job. Let i denote the amount of time invested in training and c denote the
marginal cost of training. We denote the probability the worker will lose her current job as p,
which is a function of foreign competition (m). In the cases of general training and training
4 specific to the current job, p is also a function of training. Finally, let f denote the probability of
finding a new job in the event the worker has suffered a layoff. In the case where training is
specific to the current job, f = φ, which is a constant. Otherwise, f is a function of foreign
competition. The worker chooses the level of investment to maximize his expected payoff in a
two period model.
All functions are twice differentiable. Regarding the probability of suffering a layoff, we
assume that imports increase the probability of suffering a layoff at a decreasing rate (pm > 0,
pmm < 0). In the cases of general training and training that is specific to the current job, we
assume investment in training lowers the probability of suffering a layoff at a decreasing rate (pi
< 0, pii > 0), while greater foreign competition decreases the effect of training on lowering the
probability of a layoff (pim > 0). In the case where training is specific to a new job, p is only a
function of foreign competition, allowing us to ignore the cross-partial derivatives as well as pi
and pii.
The exposition below begins with the most general case, where training affects both the
probability of suffering a layoff and the probability of finding a new job in the event of
experiencing a job loss. The other two scenarios are special cases of the general training
scenario.
Scenario 1: general training
In the basic scenario, training lowers the probability the worker will suffer a layoff and raises the
probability the worker will find a new job if she is indeed laid off. The worker maximizes her
expected payoff
1
(1)
,
,
.
Where w0 is the wage on the current job in period one, w1 is the wage on the current job in
period 2, β <1 is the discount factor and γ <1 reflects the wage decrease associated with the new
job. Taking the derivative with respect to i yields the following optimization condition
Taking the derivative with respect to i yields the following,
(2)
1
,
p
, f i βγ
0.
Taking the derivative of the left hand side of equation (2) with respect to m while evaluated at i*
gives the impact of increasing foreign competition on the optimal level of investment in training
specific to the current job. Applying the implicit function theorem,
5 (3)
.
,
.
,
,
? 0.
Since each term inside the brackets in the denominator is negative, the entire denominator is
negative. The numerator shows the competing tensions of rising imports on training incentives.
General training increases expected wages by both lowering the probability of losing the current
job and increasing the probability of finding a new job in the event the worker suffers a layoff.
Rising imports lower the returns to training by decreasing training’s negative effect on the
probability of suffering a layoff. This effect is represented by the first term in the numerator
which is negative since γf(i) < 1. The second term is positive since both pm(m,i) and fi(i) are
positive. Given these opposing effects, we are unable to sign the effect of imports on the optimal
level of investment.
Proposition 1: Rising foreign competition has an ambiguous effect on investment in training that
develops general human capital due to competing effects.
Scenario 2: Training is specific to a new job
In this scenario, training only affects the probability the worker will find a new job in the event
she has suffered a layoff. Therefore, the probability of suffering a layoff is now only a function
of m. Thus, equation (3) simplifies to the following:
(4)
0.
The numerator consists of two terms, the positive impact of rising foreign competition on the
probability of suffering a layoff times investment’s positive effect on the probability of finding a
new job. Taken together, these derivatives imply that increasing foreign competition raises the
return to training which is specific to a new job by increasing the likelihood the worker will be in
a new job in the second period.
Proposition 2: foreign competition increases the payoff to training specific to a new job by
raising the likelihood the worker will be employed in a new job in the second period. Thus,
workers respond to an increase in foreign competition by increasing their investment in training
specific to a new job.
6 Scenario 3: training is specific to the current job.
In this scenario, training only affects the probability the worker will suffer a layoff. Thus, f(i)
becomes a constant, denoted . Equation (3) now simplifies to the following:
,
(5)
,
0.
The numerator reflects how rising imports lower’s the ability to reduce the likelihood of
suffering layoff by investing more in training specific to the current job.
Proposition 3: foreign competition lowers the return to investment in training specific to the
current job by raising the probability of losing the current job. Thus, workers respond to an
increase in foreign competition by decreasing their investment in training specific to the current
job.
2.2 Empirical strategy
We investigate the effect of foreign competition on investment in human capital in terms
of total training and two specific categories of training: 1) training associated with promotion or
job advancement opportunity, and 2) training due either to the implementation of new
technology or as part of an ongoing program to upgrade employee skills (which we will refer to
as new skills from here onward). Examining the effect on total training enrollment provides
some insight as to whether foreign competition has any overall impact on workers’ human
capital investments. I isolate the first category because it has the greatest connection to training
that is specific to the current job. While there may well be a degree of generality to some of the
training captured in this category, it does provide the best fit to scenario 3 described in the theory
section. The second category was chosen for two primary reasons. First, these events likely
reflect the acquisition of skills with a high degree of generality, conforming to scenario 1 set
forth in the theory section. Second, both of these variables reflect firms’ decisions to acquire and
implement new technologies and production methods, providing a link to the existing body of
research which shows that greater foreign competition may lead firms to increase their
adoption/creation of new technology. While the data set does provide a category that makes a
good fit with scenario 2 in the theory section, training specific to a new job, it is observed only
for a very small fraction of the sample and frequently associated with individuals not employed
at the time of training.
7 The analysis looks at both the extensive and the intensive margins of enrollment in
training episodes for each of the three “types” of training. In this study we define the extensive
margin as whether the individual enrolled in any training (in that particular category) during the
period in question.1 Thus, the extensive margin is measured through indicator variables which
take a value of one if the individual enrolled in any training events in that category and zero
otherwise. In theory, we have multiple options for measuring the intensive margin of training.
We could look at the number of training episodes or some measure of the total time spent in
training (weeks, days, hours). While the number of training episodes is likely measured with less
error than time spent in training, it is also the least informative since a training episode could
range from a couple of hours to several weeks in duration. We opt for the most detailed
description of training time available in the dataset, the number of weeks spent in training. Of
course, this measure is subject to a potentially significant degree of measurement error since we
rely on the respondents’ reports on training episodes. The respondents may not accurately
remember the time spent in a training episode (in particular if it was eighteen to twenty months
ago). Perhaps more importantly, individuals may use different conventions for rounding of time
spent in training. The vast majority of individuals who report enrolling in a training event report
either zero or one week spent in training. In reality, the training event likely lasted some fraction
of a week. Some individuals who spent three days in training may report the event lasting for
zero weeks while others report it lasting for one week. Individuals who were enrolled in training
episodes lasting six days and eight days may both report their training episodes as lasting for one
week. In reality, the latter individual spent 33 percent more days in training. The potential for
significant measurement error in the dependent variables may lead to a substantial increase in our
estimated standard errors, making it less likely our estimates will uncover a statistically
significant relationship between imports and training.
To analyze the extensive margin of training, each model is fitted using probit and IV
probit estimation. For the intensive margin of training variables, we fit each model using a linear
fixed effects estimator and two-stage least squares estimator. The empirical specification is as
follows:
(6)
1
As we explain in the data description, in the dataset employed for this study, individuals are surveyed biennially
and asked to provide information relating to the years since the previous interview.
8 Where the subscript i denotes the individual, j the industry, and t the year. T is the training
outcome variable, m is a measure of foreign competition, X is a vector of individual and
workplace characteristics (including education, a measure of ability, age, tenure with the current
employer, demographics, whether the employer provides training opportunities, and firm size), Z
is a vector of industry characteristics (productivity growth, skill intensity, and the capital to labor
ratio) , T are year fixed effects, α are the individual fixed effects and ε is an iid error term.
The degree of foreign competition is captured via the industry import penetration share.
This variable is constructed by taking the value of all imports in the industry and dividing by the
value of domestic shipments plus the value of imports minus the value of exports. For industry j
in year t:
(7)
.
By construction, the import share variable takes values between zero and one. The denominator
is interpreted as the total value of all goods placed into the domestic market (which is why we
must subtract the value of exports). Thus, the import share is the share of goods in the domestic
market coming from foreign producers. This is a commonly used measure of import competition
(see Bernard et al 2006 and Mion and Zhu 2013).
While measurement error in our training variables might bias our results against finding a
statistically significant link between foreign competition and enrollment in training events by
inflating the standard errors for the coefficient estimates, other factors may bias the coefficient
estimates themselves. By focusing on participation in training events and controlling for whether
the employer makes training opportunities available, we mitigate some of the potential
endogeneity between foreign competition and training provision. The biggest remaining
challenge to identifying a causal link between foreign competition in the product market and
individual training decisions comes from the sorting of individuals into industries and firms with
different rates of training according to unobservable worker characteristics. For example,
individuals with higher levels of (unobservable) ability may also have higher returns to training.
As such, they are more likely to seek out careers with greater training opportunities. If more
skill-intensive industries also require more training, then high ability workers are likely to sort
into these industries. Given the comparative advantage the US has in skill-intensive industries,
these industries are likely to be characterized by relatively lower import penetration rates. Thus,
there will be a negative correlation between foreign competition and training driven by this
9 sorting of workers across industries. If the causal effect of greater foreign competition is to
increase training, then a failure to account for the sorting will lead to a downward bias in
estimates obtained via simple least squares estimation.
One simple solution to this problem is to control for industry fixed effects and rely on
within industry variation in foreign competition and training to identify the model. This
approach assumes worker sorting is driven by time-invariant industry characteristics. However,
this assumption may be flawed as industry-level rates of training may vary for reasons unrelated
to international trade. According to product cycle theories, younger industries are characterized
by higher rates of technical change. Thus, younger industries are also likely to exhibit higher
rates of training as workers need to adjust to constantly evolving production methods and
products.2 As the industry matures, the rate of change in products and processes slows down,
and along with it training intensity. In this case, industry fixed effects will not sufficiently
address the sorting problem.
2.2.1 Instrumental variables
To address the potential endogeneity of the imports variable, I estimate the models via
two-stage least squares, using industry-level exchange rates to instrument for the imports
variables. This approach has been employed in other studies (see, for example Mion and Zhu
2013). The industry-level exchange rates are calculated by the New York Federal Reserve
Banks and are available on the Bank’s website. Exchange rate movements should be driven
primarily by policy and macroeconomic factors. Thus, while it is reasonable to expect industrylevel exchange rates will exhibit a fairly high correlation with import shares, it is unlikely they
are correlated with unobserved, individual-level characteristics.
The New York FRB calculates both exporter and importer weighted exchange rates for
the manufacturing sector at the 3-digit NAICS level (nineteen different industries). I make use
of both the importer and exporter weighted exchange rates, including the first, second, and third
lags of each exchange rate as instruments (for a total of six instruments). The instruments are
calculated at a significantly higher level of aggregation than the imports variables (4-digit SIC),
limiting the degree to which the lagged industry exchange rates can explain variation in import
2
Vernon (1966) first developed the life cycle hypothesis. A recent paper (Xiang 2005) finds evidence that new
goods have substantially higher skilled-labor intensity, lending support to the notion that younger
industries/products are characterized by higher rates of R&D and technical change.
10 shares. Nonetheless, the exchange rate variables contain sufficient explanatory power to serve as
strong instruments.
3. Data Description
This study uses the 1991-2004 waves of the National Longitudinal Surveys of Youth
1979 cohort (NLSY79). The survey was conducted annually beginning in 1979 through 1994
and in even numbered years thereafter. The original sample contains 12,686 individuals who
were fourteen to twenty-two years old in 1979. This full sample includes oversamples of
minorities and the poor in addition to a supplemental sample representative of individuals
enrolled in the four branches of the military. The military sample was dropped after the 1984
survey while the oversampling of poor whites was dropped in 1991. The present study uses data
only for individuals in the nationally representative cross-section, which originally contained
6,111 individuals. Actual sample sizes are smaller due to attrition and missing information for
the key variables employed in the analysis. Furthermore, the study focuses on individuals
employed in the manufacturing sector, for which we have import values to construct our
measures of foreign competition.
The international trade data come from Peter K. Schott’s website. They were developed
for use in the study by Bernard, Jensen and Schott (2006) and are described in detail in Schott
(2010). The data are aggregated up from the product level to the 4-digit standard industry
classification (SIC) level and cover the period from 1972-2005. However, there was a
significant change in the way the value of trade was measured beginning in 1989, creating a
significant discontinuity in the data. The sample period is further restricted due to the fact that
one of the key training variables is only available for the 1991-2004 survey years. For all years,
contemporaneous trade measures are merged with the individual-level data. The trade dataset
contains information in the value of imports, exports and domestic production for each industry.
Additionally, the dataset provides separate information on the value of imports coming from
China and the value of imports from low-wage countries.3 This information is available for
roughly 400 4-digit industries in each year. These industries represent over ninety percent of all
manufacturing output.
3
A country is categorized as being low wage if its per-capita GDP was less than five percent of the US’s per capita
GDP in the year 2000. Hence, the set of countries considered low-wage does not change over the sample period.
11 Until the 2000 survey, the NLSY coded the industry of occupation according to the 1980
census classification. Beginning with the 2002 survey, the NLSY switched to the 2000 Census
industry classifications. In order to provide a consistent industry classification across all years of
data analyzed in this study, it was necessary to create a bridge between the two Census
classification systems. While it was relatively straightforward to incorporate most of the Census
2000 industry categories into the 1980 classification, some industries did not combine into a
single 1980 category. In order not to lose these observations, five industry pairs (using the 1980
classification) were merged, leaving us with seventy-four industries. Merging the industry trade
data into the NLSY required the use of a bridge between the SIC classification system and the
industry classifications used in the NLSY. The documentation for the industry and occupation
classifications used in the NLSY provides a bridge between 3-digit SIC industries and the 1980
Census industry classification. Accordingly, all industry-level data was aggregated into the
seventy-four final industry classes.
The NLSY contains information on up to four training events and also asks individuals
who report on four training episodes whether they participated in additional training beyond the
four events detailed. In these cases, the number of training events variable is given a value of
five. Thus, the number of training episodes variable takes values from zero to five. This topcoding of the training events variable does not raise any significant issues for the estimation
since there are only two observations where the respondent reported attending more than four
training events in our estimation sample. In fact, only three percent of the sample reports more
than one training event, mostly representing cases where the individual participated in two
training events. The training variables are constructed from follow up questions inquiring into
the nature and duration of the training events attended. We create the total number of weeks
spent in training variable by adding up the number of weeks spent in all reported training events
for the individual-year observation.4
The variables measuring training for specific reasons make use of a follow up question
which asks respondents to list the reason for each training event. Survey participants were asked
to select from the following list of responses: 1) training was associated with promotion or job
advancement opportunity, 2) new methods or processes introduced – additional training required
to continue same job, 3) the training was part of a regular program to maintain and upgrade
4
The respondents were asked to report the number of weeks spent in each training episode.
12 employee skills, 4) the training was necessary when I began a job, 5) the training was necessary
for a license or a certificate, 6) the training was associated with looking for a new job, 7) other.
The categories are mutually exclusive. We label training events under category (1) as training
related to career advancement opportunities (largely reflecting training specific to the current
employment relationship), while training episodes in categories (2) and (3) are labeled as those
related to obtaining new skills. Analogous to the total weeks of training variable, both of these
reason specific training variables are constructed by adding the total weeks spent in training for
that specific reason. The information for categories (4) and (5) is excluded from the analysis
since these training episodes generally reflect a new job and we do not want to pick up training
effects associated job turnover and foreign competition. While category (6) would appear to
provide a nice fit with scenario 2 in the theory section, it is only observed for a small fraction of
training episodes and frequently associated with unemployed workers looking to move into a
new occupation.5 With the exception of the new skills categories (2 and 3), these training events
are not restricted to employer sponsored or on-the-job training (as should be clear from reason
for training option six). This distinction is important since this paper focuses on the individual’s
training participation activity, not the firm’s decision of whether to offer training to its
employees. Since the total weeks spent in training variable includes all of these categories, it
captures a fairly comprehensive measure of training, regardless of whether the training events
were company sponsored. Focusing on this more inclusive measure of training enrollment is
important since individuals employed in an industry facing substantial foreign competition may
enroll in training events to prepare for employment in another industry or occupation in the event
of a job displacement. These pressures would not be captured by variable focusing exclusively
on on-the-job or company sponsored training.
The remaining explanatory variables are a combination of individual characteristics,
employer information and industry-level characteristics. Both the individual characteristics and
employer information come from the NLSY while the industry characteristics come from the
National Bureau of Economics Research’s productivity database. The individual characteristics
included in the empirical models are: highest grade completed, the armed forces qualify test
(AFQT) score percentile, age, log of tenure with the current employer, the number of children in
5
Since this category is included in the overall training variables, a robustness check is performed excluding workers
who have six months of tenure or less with the current employer.
13 the household and indicator variables for whether the individual is married, in a white-collar
occupation, female, black or Hispanic.6 The employer characteristics are the log number of
employees at the respondent’s location of employment and an indicator variable for whether the
individual’s employment is covered by a union contract. The industry-level characteristics are
the log of total factor productivity (TFP), the non-production worker wage share (referred to as
skill intensity) and the log of capital per worker (referred to as capital intensity) where capital is
measured in dollar value. Each specification also includes a vector of indicator variables for year
and industry.
Before turning to the empirical results, a quick look at the annual statistics for each of the
trade measures helps to put the findings into context. Figure 1 shows the upward trend in the
aggregate import share for the US manufacturing sector over the sample period. The import
share nearly doubles between 1989 and 2005, increasing from 13.37 to 25.81 percent. We will
use these changes in the import shares to calculate the estimated impact of rising foreign
competition on training enrollment.
Table 1 presents summary statistics for the training variables, imports and industry
characteristics variables as well as the individual and employment characteristics included in the
empirical models. Workers report having attended at least one training episode in 14.09 percent
of the observations. They spent a total of 0.627 weeks in training on average, with 0.128 weeks
enrolled in training due to a promotion/career advancement, 0.266 weeks due to the
implementation of new technologies or upgrading their skills. In nearly sixty-one percent of
observations workers report that the firm makes company sponsored training available to its
employees. The average import share over the sample period is 17.76 percent. Roughly thirtyone percent of worker-year observations are for white-collar workers. The average educational
attainment is just over thirteen years of schooling and the average AFQT score percentile is
49.36.7 Consistent with previous reports, the sample shows that manufacturing is a maledominated sector. There are also relatively few minorities employed in the manufacturing
sector, only ten percent of the sample is black and 5.5 percent is Hispanic.
6
White collar occupations include the following broad categories: executive, administrative and managerial,
professional specialty occupations, technicians and related support and sales occupations. These occupations
correspond to codes 1-299 in the 1980 occupation classification.
7
The average educational attainment and AFQT score percentile are comparable to those for the full sample
including individuals employed in all sectors.
14 4. Results and Discussion
4.1 Extensive margin of training
Table 2 presents the results for the extensive margin of training estimates. Standard
errors are reported in parentheses with marginal effects in brackets. For the sake of comparison,
each model is fitted via both probit and IV probit estimation. Both the probit and IV probit
estimates indicate that rising imports do not show a significant correlation with the probability a
worker will attend any training. Individuals working in industries with a higher skill intensity
and capital to labor ratio are more likely to enroll in some form of training event, while
individuals working for a firm that offers training are 11 percentage points more likely to attend
training. Given the limitations of the data, we are unable to determine how much of this
correlation is due to an increase in voluntary enrollment in training due to greater access to
training opportunities and how much of it is due to the fact that firms offering training often also
make that training a required part of the job. More educated and more able (as measured by the
AFQT score percentile) individuals are more likely to participate in a training event, as are those
working for larger firms. The remaining variables fail to show a statistically significant
correlation with the probability an individual will attend training.
Next, we turn to the models investigating enrollment in training due to a promotion or
career advancement opportunity. The probit model estimates indicate a ten percentage point
increase in the import share is associated with a 3.2 percentage point decrease in the likelihood
of engaging in this type of training. The IV probit model shows an even stronger effect; a ten
percentage point increase in the import share results in an 11.2 percentage point decline in the
probability a worker will pursue training for the purpose of career advancement. The substantial
increase in the coefficient and marginal effect estimates for the IV probit over the probit model
support the assertion in the methodology section that unobserved worker characteristics are
likely to bias the estimates against finding statistically significant effects of foreign competition
on training. Other than the variable indicating whether the firm offers training opportunities, the
remaining fail to exhibit a significant correlation with the likelihood of attending training events
for career advancement. Evidence regarding the instruments’ validity is provided by table 3,
which presents estimates and test statistics from the first stage of the IV probit model. Both the
partial R-squared and the F-test indicate we are working with strong instruments while the
Hansen J-statistic indicates the instruments meet the excludability condition. In fact, these first
15 stage tests support the validity of the instruments for all three independent variables. These
findings are consistent with the predictions of the theoretical model. By lowering the returns to
investment in training specific to the current employment relationship, greater foreign
competition leads to less investment in this type of training.
Both the probit and iv probit models fail to find a significant relationship between import
shares and the likelihood a worker will engage in training associated with the implementation of
new technology or the upgrading of skills. However, workers employed in more skill and
capital-intensive industries or in larger firms and firms that make training opportunities available
are more likely to engage in training associated with the acquisition of new skills. The same
holds true for workers with more education and higher ability; a one standard deviation increase
in the AFQT score percentile results in a 28.87 percentage point increase in the likelihood the
individual will attend a training event with the purpose of upgrading or developing new skills.
4.2 Intensive margin of training
Estimates for the effect of rising imports on the intensive margin of training are presented
in table 4. These models are estimated over the sample of workers who report attending any
training events (for the total weeks of training model) and any training events for the purpose of
upgrading or acquiring new skills (for the weeks of training for new skills models). Given the
small sample size, we do not estimate a model for the intensive margin of training for the
purposes of career advancement with the current employer. For both dependent variables, we
estimate the model via linear fixed effects and instrumental variables.
We have observations for 995 individuals who attended any training since the previous
interview (approximately two years prior). Both the FE and IV estimates show that increasing
foreign competition increases the total number of weeks a worker will spend in training events,
however only the latter is statistically significant (and only at the ten percent level). Thus, there
is some weak evidence that workers respond to foreign competition by increasing the total
amount of time spent in training. In general, the model fails to uncover many statistically
significant relationships at all. This is unsurprising in the FE model since we generally only
have an average of 1.9 observations per individual for this sample.
By contrast, weeks spent in training related to new skills models shows a stronger
relationship with foreign competition. The IV estimates indicate a ten percentage point increase
16 in the import share is associated with nearly one additional week spent in training for this
purpose. As one would expect, workers in more skill and capital intensive industries are more
likely to enroll in training events with the purpose of obtaining new skills; however industry
productivity growth and training for new skills are negatively related. This could reflect either
the initial decline in productivity experienced by firms implementing new technologies or
production processes (Sakellaris 2004) or firms attempting to offset lower productivity growth
through technology adoption. Counter to our intuition, we also observe a negative impact of
education on training due to technology adoption. Overall, the results for the intensive margin of
training indicate that workers who elect to enroll in training in order to upgrade or learn new
skills are likely to spend more time in these training episodes. The theoretical model developed
in this paper yields an ambiguous relationship between general training and foreign competition
due to competing forces. These results indicate that the benefits of this type of training carrying
over to other employment outweigh the lower return from this type of training with the current
employer (since there is a greater probability the worker will lose his current job and end up
unemployed).
4.2 Robustness check: excluding workers with less than six months tenure
To mitigate the prospect our results are being driven by workers who are in the early
phases of new jobs (a time when workers are more likely to spend significant time in training,
regardless of competitive pressures), we restrict the sample to individuals reporting more than six
months (twenty-six weeks) tenure with the current employer. This leads to a loss of
approximately 700 observations in the probability of any training sample (roughly ten percent of
the sample). Table 5 presents IV estimates for the extensive and intensive margins of both total
and new skills related training as well as estimates for the extensive margin of training related to
job advancement (for the sake of brevity, only the estimates for the import share variable are
presented). The results using this further restricted sample are highly similar to those presented
in tables 2 and 4. The one notable difference in the results is that the estimated effect of imports
on total weeks spent in training is now statistically significant at the one percent level. A ten
percentage point increase in the import share results in an additional 1.2 weeks spent on training.
5. Conclusion
17 This paper estimates the impact of rising imports on worker training. It is the first paper
to focus on foreign competition’s impact on worker enrollment in training events (as opposed to
firms’ decisions about whether to provide training to its workers) for US workers in the
manufacturing sector. The results find a robust, positive relationship between rising imports total
weeks spent in training as well as weeks spent in training related to the adoption of new
technology or skill upgrading. Consistent with the theoretical model presented in the paper, we
also find a negative relationship between rising imports and the probability a worker will enroll
in training related to career advancement.
These findings have important implications for the literature estimating the labor market
effects of globalization. Failure to account for post-schooling training may lead to an incorrect
assessment of the downward wage pressures faced by more educated workers employed in the
manufacturing sector. The evidence presented here adds another piece to the puzzle in our
understanding of the links between international trade and labor market outcomes for US
workers.
18 References
Acemoglu, Daron and Jorn-Steffen Pischke (1998). “Why do firms train? Theory and evidence,”
Quarterly Journal of Economics, Vol. x: 79-119.
Addison, John T, DA Fox and Christopher J. Ruhm (2000). “Technology, trade sensitivity, and
labor displacement,” Southern Economic Journal, Vol. 66(3): 682-699.
Aghion, Phillipe, Nick Bloom, Richard Blundell, Rachel Griffith and Peter Howitt (2005).
“Competition and innovation: an inverted U relationship,” The Quarterly Journal of Economics,
Vol. :701-728.
Aghion, Phillipe, Richard Blundell, Rachel Griffith, Peter Howitt, and Susanne Prantl (2009).
“The effects of entry on incumbent innovation and productivity,” The Review of Economics and
Statistics, Vol. 91(1): 20-32.
Altonji, Joseph G. and James R. Spletzer (1991). “Worker characteristics, job characteristics and
the receipt of on-the-job training,” Industrial and Labor Relations Review, Vol. 45(1): 58-79.
Barron, John M., Mark C. Berger and Dan A. Black (1997). “How well do we measure
training?” Journal of Labor Economics, Vol. 15(3): 507-528.
Bernard, Andrew B, J. Bradford Jensen and Peter K. Schott (2006). “Survival of the best fit:
exposure to low-wage countries and the (uneven) growth of us manufacturing plants.” Journal of
International Economics, Vol. 68: 219-237.
Bloom, Nicholas, Mirco Draca and Marco Van Reenen (2008). “Trade induced technical
change? The impact of Chinese imports on IT and innovation,” mimeo.
19 Blundell, Richard, Rachel Griffith and John Van Reenen (1999). “Market share, market value
and innovation in a panel of British manufacturing firms,” Review of Economic Studies, Vol. 66:
529-554.
Booth, Alison L., Francesconi, Marco and Zoega, Gylfi (2003), “Unions, work-related training
and wages: evidence for British men”, Industrial and Labor Relations Review, Vol.(57(1): 68-91.
Cunat, Vicete and Maria Guadalupe (2009). “Globalization and the provision of incentives inside
the firm: The effect of foreign competition,” Journal of Labor Economics, Vol. 27(2): 179-212.
Ebenstein, Avraham, Ann Harrison and Shannon Philips (2011). “Estimating the impact of trade
and offshoring on American workers using the Current Population Surveys,” mimeo.
Falvey, Rod, David Greenaway and Joana Silva (2010). “Trade liberalization and human capital
adjustment,” Journal of International Economics, Vol. 81: 230-239.
Gashi, Ardiana N., Geoff Pugh and Nick Adnett (2010). “Technological change and employerprovided training: evidence from UK workplaces,” International Journal of Manpower, Vol.
31(4): 426-448.
Guadalupe, Maria and Julie Wulf (2010). “The Flattening Firm and Product Market Competition:
The Effect of Trade Liberalization on Corporate Hierarchies,” American Economic Journals:
Applied Economics 2: 105-127.
Gorodnichenko, Yuriy, Jan Svejnar and Katherine Terrell (2010). “Globalization and innovation
in emerging markets,” American Economic Journals: Macroeconomics 2: 194-226.
Karabay, Bilgehan and John McLaren (2010). “Trade, offshoring, and the invisible handshake,”
Journal of International Economics, Vol. 82: 26-34.
20 Kletzer, Lori G. (1998). “Job displacement,” Journal of Economic Perspectives, Vol. 12(1): 115136.
Kosteas, Vasilios D. (2008). “Manufacturing wages and imports: Evidence from the NLSY,”
Economica, Vol. 75: 259-279.
Krishna, Parvin, and Mine Zeynep Senses (2009). “International trade and labor income risk in
the United States,” NBER working paper #14992.
Lai Tat-kei and Travis Ng (2011). “The impact of product market competition on training
provision: evidence from Canada,” mimeo.
Li, Hao-Chung (2010). “Trade, training, employment, and wages: evidence from the U.S.
manufacturing industry,” mimeo.
Long, Ngo Van, Raff Horst and Frank Stahler (2011). “Innovation and trade with heterogeneous
firms,” Journal of International Economics, Vol. 84: 149-159.
Mion, Giordano and Linke Zhu (2013). “Import competition from and offshoring to China: A
curse or blessing for firms?” in Journal of International Economics, Vol. 89: 202-215.
Nickell, Steven (1996). “Competition and corporate performance,” Journal of Political
Economy, Vol. 109: 724-746.
Sakellaris, Plutarchos (2004). “Patterns of plant asjustment,” Journal of Monetary Economics,
Vol. 51(2): 225-250.
Schott, Peter K. (2010). “U.S. Manufacturing Exports and Imports by SIC or NAICS Category
and Partner Country, 1972 to 2005,” mimeo.
21 Sly, Nicholas (2010). “Skill acquisition, incentive contracts and jobs: labor market adjustment to
trade.” Mimeo
Tempesti, Tommaso (2010). “Offshoring and the skill-premium: Evidence from individual
workers’ data,” mimeo.
Vernon, Raymond. “International Investment and International Trade in the Product Cycle.”
Quarterly Journal of Economics, 1966, 80(2): 190–207.
Veum, Jonathan R. (1999). “Training, wages, and the human capital model,” Southern Economic
Journal, Vol. 65(3): 526-538.
Xiang, Chong (2005). “New Goods and the relative demand for skilled labor,” Review of
Economics and Statistics, Vol. 87(2): 285–298.
Xu, Kuan and Zhengxi Lin (2011). “Participation in workplace employer-sponsored training in
Canada: Role of firm characteristics and worker attributes,” Contemporary Economic Policy,
Vol. 29(3): 416-430.
Zeytinoglu, Isik U. and Gordon B. Cooke (2009) “On-the-job training in Canada: Associations
with information technology, innovation and competition,” Journal of Industrial Relations, Vol.
51(1): 95-112.
22 Table 1: summary statistics
Mean
Std. Dev.
Attend training dummy
.1408882
.3479395
New skills dummy
.0953292
.2936973
Advancement dummy
.0225881
.1486003
Total weeks spent in training
.6274885
4.11595
Weeks training: job advancement
.1284456
1.584484
Weeks training: new skills
.2662711
1.664614
import share
.1776078
.1335408
productivity growth
1.126529
1.159721
skill intensity
.4045997
.1567817
log capital to labor ratio
4.257133
.8265997
firm offers training
.6083461
.4881667
white collar occupation
.3110643
.4629734
education
13.02584
2.251024
AFQT test score
47.36619
28.87236
log of tenure in years
1.297577
1.347068
log number of employees
5.064188
2.099539
age
34.78752
4.763609
married
.6428025
.4792196
number of children
1.265314
1.196097
female
.3539433
.4782376
black
.101072
.3014529
hispanic
.0549387
.2278825
Observations
5,224
Table provides summary statistics for the key variables.
23 Table 2: Probit and IV probit estimates of foreign competition on training
Attend Training
import share
productivity growth
skill intensity
log capital to labor ratio
firm offers training
white collar occupation
education
AFQT test score
Job Advancement
Probit
IV Probit
Probit
IV Probit
Probit
IV Probit
-0.0120
0.101
-0.483+
-1.601*
-0.303
0.271
(0.156)
(0.435)
(0.272)
(0.659)
(0.204)
(0.529)
[-0.003]
[0.023]
[-0.032]
[-0.112]
[-0.051]
[0.046]
-0.0225
-0.0258
0.0241
0.0556+
-0.0195
-0.0361
(0.0177)
(0.0211)
(0.0267)
(0.0312)
(0.0209)
(0.0253)
[-0.005]
[-0.006]
[0.002]
[0.004]
[-0.003]
[ -0.006]
0.488**
0.487**
0.229
0.217
0.561**
0.567**
(0.122)
(0.122)
(0.235)
(0.232)
(0.158)
(0.158)
[0.110]
[0.110]
[0.015]
[0.015]
[0.095]
[0.097]
0.0924**
0.0959**
0.0397
0.00385
0.0803*
0.0978**
(0.0237)
(0.0267)
(0.0447)
(0.0464)
(0.0314)
(0.0346)
[0.021]
[0.022]
[0.003]
[0.000]
[0.014]
[0.017]
0.486**
0.484**
0.336**
0.351**
0.499**
0.488**
(0.0456)
(0.0460)
(0.0940)
(0.0935)
(0.0639)
(0.0643)
[0.110]
[0.110]
[0.022]
[0.025]
[0.085]
[0.083]
0.0664
0.0662
-0.0147
-0.0122
0.102
0.101
(0.0486)
(0.0486)
(0.0925)
(0.0913)
(0.0627)
(0.0627)
[0.015]
[0.015]
[-0.001]
[-0.001]
[0.017]
[0.017]
0.0475**
0.0476**
0.0390+
0.0388+
0.0428**
0.0427**
(0.0113)
(0.0113)
(0.0210)
(0.0207)
(0.0145)
(0.0145)
[0.011]
[0.011]
[0.003]
[0.003]
[0.007]
[0.007]
0.00449**
0.00448**
-0.00091
-0.00079
0.0052**
0.0051**
(0.0009)
(0.0009)
(0.0017)
(0.0017)
(0.0012)
(0.0012)
[0.001]
[0.001]
[-0.000]
[-0.000]
[0.001]
[0.001]
24 New Skills
log of tenure in years
log number of employees
age
married
number of children
female
black
Hispanic
Observations
-0.0136
-0.0134
0.0130
0.00988
0.103**
0.104**
(0.0155)
(0.0155)
(0.0279)
(0.0277)
(0.0214)
(0.0215)
[-0.003]
[-0.003]
[0.001]
[0.001]
[0.018]
[0.018]
0.0434**
0.0426**
0.0270
0.0355+
0.0559**
0.0515**
(0.00933)
(0.00970)
(0.0183)
(0.0185)
(0.0120)
(0.0127)
[0.010]
[0.010]
[0.002]
[0.002]
[0.010]
[0.009]
-0.0208*
-0.0212*
-0.00019
0.00358
-0.00808
-0.0101
(0.00852)
(0.00863)
(0.0153)
(0.0154)
(0.0111)
(0.0113)
[-0.005]
[-0.005]
[-0.000]
[0.000]
[ -0.001]
[ -0.002]
-0.00483
-0.00535
0.0133
0.0167
0.0200
0.0181
(0.0443)
(0.0444)
(0.0840)
(0.0830)
(0.0600)
(0.0600)
[-0.001]
[-0.001]
[0.001]
[0.001]
[0.003]
[0.003]
0.0234
0.0231
-0.00804
-0.00619
0.0127
0.0114
(0.0177)
(0.0177)
(0.0348)
(0.0346)
(0.0232)
(0.0232)
[0.005]
[0.005]
[-0.001]
[-0.000]
[0.002]
[0.002]
0.0140
0.0126
0.0193
0.0303
-0.0327
-0.0395
(0.0391)
(0.0392)
(0.0726)
(0.0724)
(0.0518)
(0.0515)
[0.003]
[0.003]
[0.001]
[0.002]
[-0.006]
[-0.007]
0.00253
0.00256
-0.0810
-0.0863
-0.140
-0.138
(0.0712)
(0.0712)
(0.131)
(0.130)
(0.101)
(0.101)
[0.001]
[0.001]
[-0.005]
[-0.006]
[-0.024]
[-0.024]
-0.109
-0.111
-0.0622
-0.0401
-0.169
-0.182
(0.0913)
(0.0919)
(0.172)
(0.172)
(0.122)
(0.122)
[-0.025]
[-0.025]
[-0.004]
[-0.003]
[-0.029]
[ -0.031]
7362
7362
5416
5416
5421
5421
25 Coefficient estimates, with standard errors in parentheses and marginal effects in brackets.
+, *, ** denote significance at the 10%, 5% and 1% levels, respectively.
All models contain year indicators.
26 Table 3: First stage estimates and statistics for IV probit models
Attend
Job
New Skills
Training
Advancement
Training
Lagged exchange rate
-0.007**
-0.0088**
-0.0088**
for exports (1st lag)
(0.0012)
(0.0013)
(0.0013)
Lagged exchange rate
-0.0039**
-0.0078**
-0.0078**
for imports (1st lag)
(0.0006)
(0.00076)
(0.00076)
Lagged exchange rate
0.0084**
0.0089**
0.0089**
for exports (2nd lag)
(0.0015)
(0.0016)
(0.0016)
Lagged exchange rate
0.99973
0.0051**
0.0051**
for imports (2nd lag)
(0.00096)
(0.0012)
(0.0012)
Lagged exchange rate
-0.0046**
-0.0043**
-0.0043**
for exports (3rd lag)
(0.0011)
(0.0012)
(0.0012)
Lagged exchange rate
-0.0056**
-0.0065**
-0.0065**
for imports (3rd lag)
(0.00069)
(0.00081)
(0.00081)
Partial R-squared for instruments
0.1245
0.1404
0.1443
F-test for instruments
160.9
145.1
152.3
Hansen J-statistic
3.03
2.13
3.51
(0.70)
(0.84)
(0.62)
First stage statistics:
First stage estimates and test statistics for the IV probit models presented in table 2.
+, *, *** denote significance at the 10%, 5% and 1% levels, respectively.
All models contain the full set of covariates, including year effects.
27 Table 4: Imports and the intensive margin of training
Weeks spent in training episodes
import share
productivity growth
skill intensity
log capital to labor ratio
firm offers training
white collar occupation
education
AFQT test score
log of tenure in years
log number of employees
age
Total
FE
IV
FE
IV
6.295
8.686+
4.736+
9.083*
(4.357)
(5.125)
(2.648)
(4.034)
-0.0938
-0.296
-0.0804
-0.537**
(0.345)
(0.301)
(0.201)
(0.202)
0.811
0.732
-2.175
1.082
(3.341)
(1.396)
(2.645)
(1.251)
-0.501
0.488
-0.556
0.500*
(0.501)
(0.561)
(0.403)
(0.232)
0.0318
-0.351
0.880
-0.516
(1.880)
(1.386)
(1.701)
(0.733)
0.159
-1.033
0.840
0.213
(0.912)
(0.856)
(0.753)
(0.559)
-0.288
-0.372
-0.426
-0.372**
(1.320)
(0.265)
(0.694)
(0.127)
.
-0.000862
.
0.0157
.
(0.0246)
.
(0.0112)
-1.238+
-0.895+
-0.669
-0.337
(0.657)
(0.508)
(0.434)
(0.222)
0.271
0.0197
0.0483
0.118
(0.245)
(0.172)
(0.150)
(0.0979)
0.717
0.227
-1.170
-0.0380
28 New Skills
married
number of children
female
black
Hispanic
Observations
R-squared
(0.845)
(0.257)
(1.014)
(0.105)
1.263
-0.251
-1.492
-0.622
(1.512)
(0.775)
(1.108)
(0.618)
-0.270
-0.611+
-0.474
-0.432*
(0.636)
(0.331)
(0.602)
(0.199)
.
-0.181
.
-0.655
.
(0.786)
.
(0.433)
.
-0.302
.
0.0994
.
(1.717)
.
(0.715)
.
-2.666**
.
-0.660
.
(0.912)
.
(0.558)
993
993
503
503
0.0824
0.0659
0.1387
0.0420
Standard errors are in parentheses.
+, *, *** denote significance at the 10%, 5% and 1% levels, respectively.
All models contain the full set of covariates, including year effects.
29 Table 5: Robustness check using only observations with 6 months plus tenure
Attend Training
import share
Extensive
Intensive
Extensive
0.06
12.16**
(0.458)
(4.345)
[0.014]
Observations
Job Advancement
6,630
919
Intensive
Extensive
Intensive
-1.588*
0.386
8.847*
(0.695)
(0.54)
(4.022)
[-0.116]
[0.07]
4,920
4,927
Standard errors are in parentheses.
+, *, *** denote significance at the 10%, 5% and 1% levels, respectively.
All models contain the full set of covariates, including year effects.
30 New Skills
491
+, *, *** denote significance at the 10%, 5% and 1% levels, respectively.
All models contain the full set of covariates, including industry and year effects.
Figure 1: Aggregate Import Share
28
26
24
22
20
18
16
14
12
10
1989
1991
1993
1995
1997
1999
2001
2003
Figure 1 shows the aggregate import share for the sample period from 1989-2005.
31 2005