Professional Documents
Culture Documents
REFERENCES
Linked references are available on JSTOR for this article:
https://www.jstor.org/stable/2118134?seq=1&cid=pdf-
reference#references_tab_contents
You may need to log in to JSTOR to access the linked references.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms
is collaborating with JSTOR to digitize, preserve and extend access to The American Economic
Review
There have been few studies in the United its the reliability of the productivity analysis.
States on the impact of human-capital in- Nevertheless, Bartel finds evidence that re-
vestments, such as education and employer- turns to training investments increase produc-
provided training, on productivity. Some of tivity on the order of 16 percent. In a follow-up
the researchers who have been able to ex- study using longitudinal data on manufactur-
amine the linkage between training and pro- ing firms, Bartel (1992) found that lagged
ductivity (e.g., John Barron et al., 1987; John training investments rather than current train-
Bishop, 1994) have used a subjective mea- ing yield positive effects on productivity.
sure of productivity such as the answer to the Our paper, using unique data from the Na-
survey question: "on a scale of 1-4 how tional Center on the Educational Quality of the
has your productivity changed over the last Workforce (EQW) National Employers' Sur-
year?" The main problem with subjective vey, presents findings on the impact of human-
measures of productivity is that they are not capital investments on business productivity.
comparable across firms or even within firms The survey was designed to overcome some
over time. However, given these caveats, of the limitations of previous studies and col-
Bishop (1994) concludes that employer- lect more precise data on human-capital inputs
provided training raises this subjective pro- and establishment output. Part of the survey
ductivity measure by almost 16 percent. (see Lynch and Black [1995 ] for more details)
Ann Bartel (1989) has linked data from a was designed to replicate the questions asked
survey of human-resource management prac- in the Annual Survey of Manufacturers (ASM)
tices at the establishment level with firm- on the dollar value of sales, receipts or ship-
level data from Compustat on productivity ments, the book value of the capital stock,
and financial performance. However, this ap- and the cost of materials (including energy)
proach is also problematic because most of the used in production during calendar year 1993.
establishments in the survey used by Bartel Given the survey design, these variables can
(1989) were part of multiple-establishment then be augmented with worker characteristics
firms. Therefore there is a discrepancy in the (such as education and training), other estab-
unit of analysis of inputs and outcomes, since lishment characteristics (such as the age of the
the Compustat data refer to the firm as a whole, capital stock), and workplace practices (such
not to specific establishments. This problem, as TQM [total quality management], bench-
along with a low overall response rate in the marking, etc.) to estimate a much richer
survey used by Bartel ( 1989) (6 percent), lim- production function than has been possible
before. In addition, since we have collected
sales, materials, and capital data specifically
for the establishment we are not forced to
* Department of Economics, Harvard University, Cam-
bridge, MA 02138, and Fletcher School of Law and match establishment-level workplace prac-
Diplomacy, Tufts University, Medford, MA 02155, re- tices with enterprise-level sales information.
spectively. The work reported herein was supported under The EQW National Employers Survey (de-
the Education Research and Development program, agree-
signed by Lynch in collaboration with EQW Co-
ment number R17QOOO11-91, CFDA 84.117Q, as ad-
ministered by the Office of Educational Research and Directors Robert Zemsky and Peter Cappelli)
Improvement, U.S. Department of Education. The funding was administered by the U.S. Bureau of the
was administered through the National Center on the Ed- Census as a telephone interview in August and
ucational Quality of the Workforce (EQW), University of
September 1994 to a nationally representative
Pennsylvania. The findings and opinions expressed in this
sample drawn from the Census Standard Statis-
paper do not necessarily reflect the position or policies of
the U.S. Department of Labor, the U.S. Department of tical Establishment Listing (SSEL) data base
Education, or the Bureau of the Census. of private establishments with more than 20
263
employees. The survey oversampled establish- ployed at the establishment for less than one
ments in the manufacturing sector and those with year, whether or not grades or communication
over 100 employees. Public-sector employers, skills are considered important for recruitment
not-for-profit institutions, and corporate head- purposes, dummy variables for the use of
quarters were excluded from the sample. Of the TQM or benchmarking, a dummy variable
4,633 eligible employers contacted by the Cen- equal to 1 if the establishment exports any
sus Bureau, 1,275 refused to participate in the sales of its principal product, a dummy vari-
survey. This represents a 72-percent response able equal to 1 if there is an R&D center any-
rate, substantially higher than many similar em- where in the enterprise, controls for capacity
ployer surveys. Of the 3,358 establishments that utilization, and whether or not the establish-
participated in the survey, not all respondents ment is unionized. Since the values of sales,
completed all parts of the'survey by the inter- capital stock, and materials need to be deflated
view cutoff date of October 1, 1994. The final by the appropriate price deflators, we use
number of employers who completed all parts dummy variables for the two-digit standard in-
of the survey was 1,621 in the manufacturing dustrial classification of the establishment in
sector and 1,324 in nonmanufacturing. This rep- our cross-section estimation.
resents a 64-percent overall "completed" sur- As shown in Table 1, human capital is an
vey response rate. important determinant of establishment pro-
Using these data we are able to estimate ductivity. The average educational level of the
the impact of human-capital investments on establishment has a positive and significant ef-
productivity. We assume a standard Cobb- fect in both the manufacturing and nonmanu-
Douglas production function where facturing sectors. The estimated coefficient in
the unrestricted Cobb-Douglas model implies
(1) log(Yi) = constant + a'Xi that a 10-percent increase in average education
in the establishment (approximately one more
+ b log Ki + c log Mi year of school) will lead to an 8.5-percent in-
crease in productivity in manufacturing and a
+ d log(LH)i + e log(LQ)i. 12.7-percent increase in nonmanufacturing.
In results reported in Lynch and Black
Output Yi is proxied by the dollar value of (1995) we also estimated a model with two
sales, receipts, or shipments for calendar year additional restrictions. The first restriction
1993 for establishment i, denoted by Si; Ki is tests (in the spirit of Dale Jorgenson and Zvi
the 1993 book value of the capital stock; Mi Griliches [1967]) whether or not the coef-
is the total cost of goods and services used in ficients on labor quality (average education)
the production of 1993 sales, including en- and labor hours are equal. If this is the case,
ergy; LHi are total labor hours for 1993, andthen we can consider the labor input to be
LQi is a measure of labor quality, which we adjusted for quality, where the adjustment
proxy with the average educational level of the factor is proportional to the average educa-
establishment, denoted by ED. The vector Xi tional level of the establishment. In addition
includes whether or not the establishment is we test to see whether the restriction of con-
part of a multi-establishment enterprise, the stant returns to scale is accepted by our data.2
age distribution of the capital stock, total num- Both of these restrictions are accepted. In the
ber of workers trained in 1990 and 1993, ' the
percentage of formal training that occurs off
the job, dummy variables for three specific
2Most studies on the determinants of productivity of U.S.
types of training activities, the percentage of establishments using cross-section data find constant re-
the full-time workforce that has been em- turns to scale. One difference between our results and
those of other studies is that the coefficient on materials
is much lower than what has been found elsewhere. We
believe that this is because respondents had more difficulty
' We have also included training in the form of the answering this question in a telephone interview than the
proportion of workers trained, but the reported results do questions on sales, capital, or labor. Therefore, there is
not change. probably significant measurement error for materials.
TABLE I-DETERMINANTS OF LOG SALES restricted model (see Lynch and Black, 1995)
(COBB-DOUGLAS PRODUCTION FUNCTION)
the implied coefficient on education would
suggest that for a 10-percent increase in edu-
Dependent variable = log(sales)
cation, productivity would rise 4.9 percent in
Independent variable Manufacturing Nonmanufacturing
manufacturing and 5.9 percent in nonmanu-
Constant 0.341 -1.252 facturing. These values combined with those
(0.317) (-0.659)
presented in Table 1 set the range of educa-
Log capital 0.25* 0.36*
(11.304) (9.957)
tion's impact on productivity. Although the
Log materials 0.26* 0.06*
range is somewhat large, the impact of edu-
(11.812) (2.958) cation on establishment productivity is sub-
Log hours 0.47* 0.628* stantial, especially in nonmanufacturing. Since
(12.45) (10.948)
the nonmanufacturing sector is expanding much
Multiple 0.13* -0.05
more rapidly than manufacturing in the United
establishments (2.257) (-0.382)
States, this result may shed some light on the
Percentage of -0.003 0.005
equipment <1 (-1.331) (1.249) widening college-high-school wage gap that
year old occurred during the 1980's.
Percentage of 0.003* -0.0003 Training has a more complex impact on the
equipment 1-4 (2.153) (-0.155)
years old
productivity of establishments. Unfortunately,
we are only able to measure the number of
log (average 0.86* 1.29t
education) (2.028) (1.793) workers involved in training at two points in
log (trained in 1993) -0.12 0.08 time: 1990 and 1993. This means that we do
(-1.294) (0.39) not have a measure of the accumulated stock
log (trained in 1990) 0.09 -0.11 of training for all workers; instead we are only
(0.994) (-0.515)
able to capture training that occurred during
Percentage of formal 0.002* -0.001
training outside (2.104) (-0.461)
two particular years. Thus our estimates of the
working hours impact of training are likely to underestimate
Computer training 0.05 0.26t the true returns. We find that the number of
(0.714) (1.895)
workers trained, especially in 1993, has no ap-
Teamwork training 0.07 -0.20
parent impact on productivity. In manufactur-
(0.983) (-1.617)
ing, the number of workers trained at the two
Supervisor training -0.03 0.20
(-0.336) (1.392) points in time is not statistically significant,
Grades a priority in -0.08 0.27t although the point estimates suggest that cur-
recruitment (- 1.226) (1.655) rent training lowers productivity, while past
Communication a 0.09 0.14 training raises current productivity. This is
priority in (1.52) (0.962)
recruitment
very similar to what we see happening with
Percent workers -0.003 -0.008*
the age of the capital stock in manufacturing.
employed < 1 (-1.61) (-2.692) Capital equipment less than one year old low-
year
ers productivity (although the effect is not sta-
Unionized -0.05 0.35*
tistically significant), while capital stock aged
(-0.793) (2.494)
1-4 years raises current productivity. This
TQM -0.02 -0.01
(-0.347) (-0.121)
suggests some sort of adjustment story, where
Benchmark 0.03 0.08
initially there are adjusted costs associated
(0.539) (0.621) with the introduction of new skills and capital
R&D center -0.01 -0.05 into the workplace which are then followed by
(-0.200) (-0.334)
positive improvements in productivity.
N: 821 525 When we include other dimensions of train-
Adjusted R2: 0.8387 0.6512
ing we see more compelling evidence about
the impact of training on productivity. For
Notes: The estimated equation also includes two-digit SIC industry controls,
capacity utilization, unionization, age of the establishment, and a dummy vari- manufacturing, the greater the proportion of
able for whether the primary product is exported. Numbers in parentheses are
time spent in formal off-the-job training, the
t-test results.
t Statistically significant at the 10-percent level. higher the productivity. There are two pos-
* Statistically significant at the 5-percent level. sible reasons. First, training workers outside
TABLE 2-RELATIVE RANKING OF FACTORS grades are for most employers. Yet the results
IN MAKING HIRING DECISIONS presented in Table 1, at least for nonman-
ufacturing, suggest that those employers who
Applicant characteristic Rank
do focus on grades experience significantly
Applicant's attitude 4.6 higher productivity than their competitors.
Applicant's communication skills 4.2
Other interesting results in Table 1 include
Previous work experience 4.0
Recommendations from current employees 3.4
the negative impact that high labor turnover
Previous employer recommendation 3.4 has on productivity, especially for nonmanu-
Industry-based credentials 3.2 facturing. Although it is not news that turnover
Years of completed schooling 2.9 is high in nonmanufacturing, this table sug-
Scores on tests administered as part of the
gests that establishments pay quite a high
interview 2.5
Academic performance (grades) 2.5 price for this turnover in terms of lower sales.
Reputation of applicant's school 2.4 In nonmanufacturing, unionization has a pos-
Teacher recommendations 2.1 itive and rather large effect on establishment
productivity.
Note: Rank is based on a scale of 1-5, where 1 = not
important or not considered, 5 = very important.
At the same time, the use of TQM or bench-
Source: Lynch and Zemsky, 1995. These data come from marking does not seem to have any impact on
the answer to the following question: "When you consider current productivity of establishments. One
hiring a new non-supervisory or production worker (front- possible explanation of why these variables do
line worker), how important are the following in your de-
not enter significantly is that we have not con-
cision to hire?"
trolled for the timing of the introduction of
these practices. If an employer has only just
introduced these practices we would expect to
working hours lowers the output loss associ- see a delay in their impact on productivity, just
ated with on-the-job training. Second, those as in the introduction of new physical capital.
employers that train their workers off the job In addition, crude measures of the incidence
may be investing in more advanced and time- of TQM or benchmarking do not capture
intensive skills development. Unfortunately, how these programs are actually being im-
our survey does not allow us to explore this plemented. Perhaps what is most important
issue in more detail. is not the introduction of TQM, but rather
For nonmanufacturing, the content of the how it is introduced, when it is introduced,
training programs provided by employers seems and how it has been implemented. This issue
to have an important impact on productivity. In has been examined in much more detail in
particular, computer-skills development has a Casey Ichniowski et al. (1993) where, using
significant and positive impact on establishment more detailed industry information obtained in
productivity, even controlling for industry. Thispersonal interviews, they find the effect on
finding is consistent with evidence presented in productivity of high-performance work sys-
Alan Krueger (1993) of higher wage premiums tems depends on how they were actually im-
associated with workers who use computers on plemented. Therefore, future data-collection
the job. This result suggests that it is not so much efforts should focus on obtaining more infor-
whether you train workers, but rather what you mation on the content, timing, and implemen-
train the workers in that affects establishment tation of programs such as employer-provided
productivity. training and total quality management.
Recruitment strategies of establishments While the results in Table 1 highlight the
also play an important role in their productiv- importance of human capital, especially edu-
ity. In nonmanufacturing, those establishments cation and certain types of employer-provided
that cite grades as an important priority in hir- training, for establishments' productivity, there
ing also have higher productivity. Table 2 are several limitations with these findings. First
shows the ranking in importance by employers and foremost is the problem of endogeneity. The
of different applicant characteristics for re- presence of unobserved establishment character-
cruitment and selection. What is striking about istics that are time-invariant is likely to bias our
this table is how low priority an applicant's estimated coefficients. Although we have a