Does information technology lead to smaller firms?
DiittftY
HD28 .iyi414
DOES INFORMATION TECHNOLOGY LEAD TO SMALLER FIRMS? Erik Brynjolfsson
Thomas W. Malone Gurbaxani Ajit Kambil
Vijay
November 1989 CiSR Sloan
WP
WP
No. 211
No. 3142-90 No. 106
CCS WP
Center for Information Systems Research Massachusetts
Institute of
Sloan School of
Technology
Management
77 Massachusetts Avenue Cambridge, Massachusetts, 02139
DOES INFORMATION TECHNOLOGY LEAD TO SMALLER FIRMS? Erik Brynjolfsson
Thomas W. Malone Gurbaxani Ajit Kambil
Vijay
November 1989 CISR
WP
No. 211
WP No. 3142-90 CCS WP No. 106
Sloan
© 1989
E. Brynjolfsson,
T.W. Malone, V. Gurbaxani, A. Kambil
Center for Information Systems Research Sloan School of Management Massachusetts Institute of Technology
J I.
-'
t
1
1-
-
.:J0
Does Information Technolog>' Lead
to
Smaller Firms?
Abstract
We
examine the relationship between information technology capital and firm using industry data for the entire U.S. economy. The results indicate that increased
size
stocks of information technology are associated with significant decreases in firm size as
measured by the number of employees. One explanation
for this observations
is
that
information technology enables reduced levels of vertical integration and our analysis of data for the U.S. manufacturing sector supports
this
explanation.
We
also find that the
on organizations are most pronounced after a lag of two years. While the correlations we find cannot, of course, prove causality, the evidence is consistent with the hypothesis that information technology reduces transaction costs and coordination costs, enabling a shift from hierarchies to markets as a means of coordinating economic activity. effects of information technology
The authors would like comments and insights,
numerous helpful and in improving the specification of the econometric model. We would also like to thank the Management in the 1990s project. Digital Equipment Corporation and the UCI Committee on Research for pro\iding funding for this research. to thank Professor Ernst
Berndt for
his
especially in identifying the relevant data sources
1.
2.
Introduction: the research problem
1
Background and hypotheses 2.1 Trends in the Icey variables 2.1.1 Firm size
2
2
3
Information technology 2.2 Theoretical studies of IT and firm 2.1.2
3.
5 size
5
Data and methodology 3.1
8
The data
8
3.1.1
Information technology capital and total capital
3.1.2
Firm
size:
employees per establishment and per firm
3.1.4 Data grouping and Methodology 3.2.1 The basic model 3.2.2 The lagged model
dummy
variables
13
14 15
17
Results 4.1
Regressions on establishment
4.2
Are the
results
an
Conclusion 5.1
Summan,'
5.2 Future research
NOTES
and
17
size
artifact of the
4.3 Information technology
5.
11
12
3.3 Predicted signs of the coefficients
4.
9 10
3.1.3 Vertical integration
3.2
8
CBP
data?
vertical integration
20 23 25 25
26 28
1.
Introduction: the research problem
Industrialized
economies have recently entered a period of substantial
organizational change.
This transition has been likened to a "second industrial divide"
(Piore and Sabel, 1984) or even the beginnings of a "post-industrial society" (Huber, 1984; and the studies reviewed therein).
Among
the postulated aspects of the transition
are decreases in firm size, a shift to externally provided services, less hierarchical
management processes specialization,
and increases
The growing drop
in
within firms, a shift from mass production to flexible
in
product and service variety.
attention to these organizational changes has coincided with a rapid
the price of computing
power (Gordon.
technology (IT)' usage, and one in
general.
infers,
1987), significant increases in information
decreases
in the price of
information processing
Unfortunately, empirical research on the relationship between IT and
organizational structure has produced few
if
any reliable generalizations.
taken as a whole, the literature on IT impacts presents
many
Indeed,
when
contradictory results
(Attewell and Rule, 1984).
We investments
have recently been able to obtain detailed, economy-wide data on in
IT
in the U.S..'
enabling the use of econometric techniques to more
1
broadly study
impact.
its
The
principal
aim of
this
paper
is
to
perform an empirical
examination of the impact of information technology on an important characteristic of
economic organization: firm Because firm
economy
size
is
as well as
size as
measured by the number of employees per
a key variable in
some of
many
of the descriptions of a post-industrial
the theories on the impact of information technology, our
results should also help illuminate the theoretical literature
firmer empirical base.
For
firm.
and put future work on
a
instance, our findings that increases in information technology
stock are associated not only with declines in employees per firm, but also with reduced vertical integration, support the theory that information technology reduces transaction
costs (Brynjolfsson,
Malone and Gurbaxani, 1988) and coordination
costs (Malone, Yates
and Benjamin, 1987; Malone and Smith, 1988).
The paper
consists of five
main
sections.
Section
II
provides background on the
recent trends in IT usage and average firm size, and presents a brief discussion of the
hypothesized relationship between these two variables.
methodology and data used regressions and explores
in this study.
In section
III,
we
explain the
Section IV presents the basic results of the
some explanations
for the results.
We
conclude with some
interpretations of the results and suggestions for further research in section V.
2.
2.1
Background and hypotheses
Trends
in
the key variables
The data
number
the
and
Whether or not
Firm
new
trends in the
last
10 years:
1)
firm size, as measured by
of employees in the average business establishment, has decreased
substantially,
2.1.1
reveal two
2) the real stock of information technology has
grown enormously.
the trends are related, the evidence for each independently
is
strong.
size
Recent trends
are widely reported.
in firm size
Piore (1986) cites data from
Count)' Business Patterns showing that the average establishment size has been
Using a
decreasing since the 1970s, reversing an earlier trend towards ever-larger firms. different data set. Birch (1979)
data revealed that the
all
new
12%
of
jobs while the
findings.
His analysis of
Dun &
Bradstreet
66%
27%
of establishments under 20 employees accounted for
46%
of establishments that had over 100 employees generated only
of
new jobs.
More
recently, the trend has apparently intensified.
Statistics reports that
new
had similar
from 1980
to 1986 firms of
The Bureau
under 100 employees created
jobs while firms of over 1000 employees experienced a net loss of
The phenomenon
is
industrial countries.
not unique to the U.S. but
As shown
in
graph
Netherlands also show a recent decrease early 1970s (Huppes. 1987).
1,
is
six million
1.5 million jobs.^
also being experienced by other
data from the
in
of Labor
average firm
UK, Germany, and size,
major
the
despite increases until the
Carlsson (1988) also found a similar pattern
in
the
manufacturing sectors of other Western industrial countries and hypothesized that facilitated
was
it
by computer-based technology.
Our examination a function of
of data reveals that the decrease in average establishment size
two complementar)' trends:
1) in sectors that
have many large firms, such as
manufacturing, overall emplojTnent has not kept up with the growth firms, resulting in significant declines in average firm size,
is
and
in
the
number
of
2) in sectors with typically
smaller firms, such as services, employment has grown rapidly, while average firm size
has remained small. (See graph
A
related trend, which
is
2).
consistent with the above observations,
externalization or "decoupling" of business functions.
a
45%
This
is
is
the increasing
manifested, for example,
in
increase from 1980 to 1986 in the use of outside contractors and consultants (Wall
Street Journal, 1988).
A
detailed study of the metal-working industry found that
amount
of value added to shipments declined in 88 of 106 sectors between 1972 and 1982 and that this could
article
be tied to increased use of information technology' (Carlsson. 1988).
on how "value-adding partnerships" are supplanting
companies, Johnston and Lawrence (1988) also argue that
In
an
vertically integrated
this
phenomenon
enabled by information technolog}'. Apparently, the wave of mergers and
is
partly
LBOs
in the
1980s have not reversed this trend as highly leveraged firms have spun-off divisions and
downsized even their core business (Scherer, 1988).
2.1.2 Information technology,'
Evidence of the increase
abundant even without looking
computer workstation 1988).
in
power and ubiquity of information technology
A
at the statistics.
for every five
employees
in
1988 study found that there was one surveyed companies (Nolan Norton,
After controlling for depreciation and quality improvements,
we
find that the
stock of IT has increased almost exponentially since 1970, from just over capital stock to nearly seven percent in 1985 (see
sectors shows the
equipment
that
graph
Each of
3).
same accelerating trend toward increased use of
was
largely insignificant
is
two decades ago
is
of overall
the major business
IT.
rapidly
1%
A
category of
becoming very
important.
Current investments business (Kriebel, 1988).
in
IT are estimated
at
50%
of
all
net capital purchases by
Meanwhile, over half of the labor force
tasks that primarily involve information processing (Porat, 1977).
we
detect today, these
in the
numbers are of
sufficient
magnitude
to
is
currently engaged in
Whatever
effects of
IT
augur even greater effects
near future.
2.2 Theoretical studies of
IT and firm
Theory suggests ways enable a reduction
size
that IT can lead to larger firms as well as
in firm size.
A
ways that IT can
seminal article by Leavitt and Whisler (1958)
5
emphasized the
One
organization.
centralization
effect of
IT on reducing the costs of information transmission within the
of the conclusions of this argument was that IT would lead to greater
and broader spans of hierarchical
control.
The
principal-agent literature
has also emphasized that the agency costs of hierarchy decrease
in
better information
and monitoring by the principal (Jensen, 1983)/ To the extent that the costs of centralization
and agency are reduced by
IT,
one might expect
to see
more use of
hierarchy and therefore larger firms.
This hypothesis contrasts with an analysis of IT's impacts based on transaction cost
economics. According to Williamson (1975, 1985). production
governance of a single hierarchy
in
is
brought under the
response to high transaction cost of organizing
through the market. These "market failures" are caused by asset and
bounded
rationality
site specificity,
and uncertainty, impacted information, and small numbers
bargaining. Information technology can be expected to mitigate these market failures and
thus reduce the need to rely on hierarchies (Malone, Yates and Benjamin, 1987; Brynjolfsson,
A
Malone and Gurbaxani,
line of analysis
technology
will
1988).
by Malone and Smith which considers that information
reduce the costs of coordination both inside and outside the organization
reached similar conclusions (Malone (1987) and Malone and Smith (1988)). They argue that because of the relatively larger coordination requirements of
markets as opposed to
hierarchies, reductions in coordination costs will be particularly beneficial for
market
Accordingly, these models predict a decrease in the proportion of
coordination.
hierarchical coordination following investments in IT.
A the
related effect of information technology
"minimum
efficient scale" of production.
is
that
it
might be expected to change
For instance, Klein, Crawford and Alchian
Grossman and Hart (1986) and Hart and Moore (1988) have emphasized
(1978),
strong technological complementarities between assets can
common
lead to
ownership of
large, related sets of assets.
techniques like flexible manufacturing,
may decrease
it
make markets However,
if
IT
that
inefficient
and
facilitates
the specificity of assets, and thus
transform hierarchical production to production organized through smaller units
coordinated by markets.
In summan,'. the theoretical literature suggests that IT will reduce the costs of
coordination both within firms and between firms. the balance will shift towards
know, a
priori
,
whether
There
is
some reason
more use of markets and smaller
this effect exists
and
is
to believe that
firms, but
we cannot
of an economically significant magnitude.
Furthermore, theorists have concentrated on the organizational forms which dominated the last centun-: markets and hierarchies.
It is
very possible that IT
organizational forms with different characteristics.
For instance,
if
will
enable
new
large firms used
information technology to set-up or simulate internal markets, average firm size might increase with information technology' usage despite a greater reliance on markets."
Fortunately, the question of
how information
technolog)' affects firms size
is
subject to empirical investigation. Previously, a case study of the evolution of airline
reservation systems (Malone, Yates and Benjamin, 1987). found that information
technology was associated with increased market coordination.
In this paper,
we
use
econometric techniques to generalize those results to the U.S. economy as a whole.
3.
Data and methodolog>
Our approach was
to use available U.S. data to directly test the relationship
between information technolog)' stock and average firm eight sectors: agriculture; mining; durable
manufacturing; transport and real estate;
and
the direction
services.
A
utilities;
are divided into
series of regressions
retail trade; finance,
were then run on
this
and magnitude of the relationship between IT and firm
tested and data
insurance and
data to
size,
identify'
while
and year. Several regression models were
from alternative sources was also used
The data
3.1.1
The data
goods manufacturing; non-durable goods
wholesale and
controlling for total capital stock, industr)'
3.1
size.
Information technology capital and total capital
to help validate the results.
Data from the Bureau of Economic Analysis (BEA) was used
to derive figures for
information technolog>' stock and investments, and total capital stock by industry for each year.
The
industries.
BEA
data classifies
The data
each industry,
total
economic
all
activity in the
gathering methodology
is
described
annual investments are measured
in
United States into 61
in
Gorman
For
et al. (1985).
27 asset categories.
We
use the
category "office, computing and accounting machinery" for our IT figures and the
all
categories for our total capital figures.
quality-adjusted (hedonic) price deflator.
Each
asset category also has
sum
of
an associated
By multiplying each investment by
its
associated deflator, nominal investments were converted into constant-dollar or "real"
To
investments.*
derive capital stocks from investment flows,
Winfrey table which assumes their expected
mean
a
Normal
distribution of
lifetime as described in
percentage of existing equipment
is
Gorman,
we used
a modified
equipment service et al. (1985).
eliminated based on
its
lives
around
Thus each
age while
new
year, a
stock
is
added
from that year's investments.
3.1.2
Firm
size:
employees per establishment and per firm
Data on average establishment
size
is
available
from County Business Patterns"
(CPB) annual summaries. The number of business establishments and number of employees
is
provided for each sector of the economy.
derived by dividing the latter by the former.^ size
were only available from 1977
to 1984.
Average establishment
size
is
Because consistent data on establishment
we used
those years as the priman,' period of
our study.
Although the data from County Business Patterns in the
census of establishments
sectors of the
number
of firms
While
this
Using the SIC code, we grouped
in
sample
is
it
still
size directly.
was
that the
the companies into the
number
of employees and
includes over 2000 firms for each year.
size.
It
also has
two
from "establishments", the Compustat data
Secondly, consistent Compustat data was available for a
CPB
data.
The
only change from the previous
Compustat data included too few firms
sector to provide a reliable statistic, so this sector
Compustat
its
presumably biased towards large firms (small firms are not
longer time period (1975-1985) than the specification
the
all
each sector. This provided a second measure of firm
First, to the extent that "firms" differ
measures firm
can be
Compustat maintains data on every
economy used above and added up
usually publicly traded),
virtues.
size
company, including number of employees and the SIC code of
principal line of business.
same
the most comprehensive
United States, another measure of firm
created from an alternative source: Compustat. publicly traded
is
in
the agricultural
was omitted from the regressions with
data.
3.1.3 Vertical integration
We
also explore the impact of IT on levels of vertical integration to help provide
10
insight into the
mechanism by which IT
Vertical integration
affects firm size.
degree to which multiple stages of production are combined under
and coordinated within a 1962).
Our
operational measure of
the total value shipped per firm.^ variation of the value
and
later
opposed
single firm as
added over
this
to a
concept
is
common
is
ownership
sequence of smaller firms (Gort,
the ratio of value added per firm to
This operationalization of vertical integration sales
measure
the
originally
is
a
proposed by Adelman (1951)
modified to the current version by Tucker and Wilder (1977).
The data required
to construct the vertical integration index
Census of Manufacturers and the Annual Survey of Manufacturers sectors of the
economy. Unfortunately, no comparable data
of the economy.
The data on
1985 (see Kambil. 19SS)
at the
vertical integration
two
digit
SIC code
is
was analyzed
is
available from the
for the manufacturing
available for other sectors for the period 1975 to
level of aggregation.
This
distinguishes 20 different industries within the manufacturing sector.
3.1.4
Data grouping and
The
dummy
variables
data for the dependent and explanatory variables was categorized by the
Standard Industrial Classification (SIC) codes. The 61 industries used by the identified by their
CPB
SIC codes and were then grouped
for the regressions
were
into the eight sectors identified by
on each of the measures of firm
size and, as
mentioned above,
into 20 manufacturing industries for the vertical integration regressions.
11
BEA
We
recognize that other trends
may
under study. To control for time-specific to IT,
we
included
a particular sector,
dummy
also have
been important during the period
effects like recessions or other trends unrelated
To
variables for each year.
we included dummy
control for effects specific to
variables for each of the industry sectors as well.
Because the most natural interpretation of the both the dependent and explanatory variables
in
is
terms of percentage-effect, the natural logarithm was taken of
all
the
relevant variables.
3.2
Methodology
The
basic technique used for analyzing the data
The panel
panel data.
CBP
was least-squares regression on
consisted of eight sectors observed from 1977 to 1984 for the
data and 1975 to 1985 for the other regressions on firm
integration regressions discussed in section
IV.D were on
size.
The
a panel of
vertical
20 manufacturing
industries for the period 1975 to 1985.
Because the sectors were of different
sizes,
each observation on a sector was weighted by the
employment presence of correlation
in that sector that year.^
serial correlation so the
was used
The
initial
leading to possible heteroskedasticity, size of the sector, as
proxied by total
regressions revealed the potential
generalized differencing correction for serial
in all regressions.'^
12
3.2.1
The
basic
Our integration.
technology
model
basic
A
model allows
number
may be
for lags in the
impact of IT on firm
the investment.
their full
new
technologies because in effect
impact on firm
is
to include
both
only
it
some
Lags are commonly observed
takes time to learn
how
to best use
in the
and apply
of the most recent years' investments will have
size.
The technique used at first
vertical
and Gordon, 1988; Curley and
particularly susceptible to lags (Baily
Thus,
and
of authors have suggested that the effects of information
Pyburn, 1982; Nolan Norton, 1988; Loveman. 1988). assimilation of
size
to
model the
ITSTOCK
to
possibility that
IT
may be
only partially effective
measure the basic impact of
IT,
and the
t^A•o
most
recent years of IT investments (ITINV). to measure the extent to which the effects of current investments are not immediately felt."
ITINV and ITINV(-])
ITSTOCK, "effective"
will
will
will
learning
is
indeed taking place,
have the opposite sign but be smaller
thus reducing the effect of
IT stock
If
be captured
in
ITSTOCK
in
for recent years.
the coefficient on
ITSTOCK
magnitude than All of the impact of
while the learning
be measured by the relative magnitude of the coefficient on ITINV.''
A
measure of
total capital stock in the sector,
TOTSTOCK.
control for the fact that firm size will in general be larger larger.
Finally, effects specific to
each
industry'
13
when
was included
to
total capital stock
and year were controlled
is
for by including
two
appropriate
sets of
The
basic
model
dummy
is
variables.
specified as:''
log(SIZE) = Bo + 6,*log(ITSTOCK) +B2*]og(ITINV) + 33*log(ITINV(-l))
+ 64*log(TOTSTOCK) + S B,*INDUSTRY + 1 Bj*YEAR.
While there
is
likely to
be some collinearity
among
the variables in this model,
regression theory assumes that the independent variables are coUinear and the least
squares formulation
is
expressly designed to separate out the effects of each of the
However, while even severe
variables.
required for accurate regression,
it
collinearity
will in
does not violate any of the conditions
general lead to lower significance levels of the
coefficients.
3.2.2
The lagged model
To and firm years.
better observe the overall time path of the relationship
size, a series
between IT stock
of separate regressions was run with IT lagged from zero to four
Although none of these regressions individually
will
learning effect, taken together, they would be expected to effectiveness of IT over time.
14
be able to capture any
show any changes
in
the
The
five regressions
used to observe the time path of IT were:
log(SIZE) = Bo + Bi*log( ITSTOCK[-N]
+
)
B.*log(
TOTSTOCK
)
+ 1 B,*INDUSTRY + 2 Bj*YEAR. where
N
e {0,...,4}, representing lags
from
to 4 years.
3.3 Predicted signs of the coefficients
If
information technology leads to smaller firms, then
negative sign on the coefficient of
how much
ITSTOCK
in
both models.
we would Its
a percentage change in IT stock will reduce firm size,
expect to see a
magnitude all
will tell us
else being equal.
In
the basic model, the coefficient should be interpreted as the impact of one unit of
"effective"
IT capital and should thus be larger than the coefficient
in
any one of the
lagged regressions.
The
learning curve effect would lead to a positive sign on
model, opposite the sign on of the coefficient on
ITSTOCK.
ITSTOCK.
Its
Including
magnitude should be
ITINV
will
ITINV less
in
the basic
than the magnitude
thus cancel a fraction of the most
recent years' investments in IT because they have not yet reached their
full
effect
on firm
size.
In the lagged model, the impact of the
15
IT stocks should be increasing for small
lags
and then returning close
rise results as the
stock that
would
is
technology
to zero in distant years, exhibiting a
is
more
"effective" increases.
As
fully
concave pattern. The
exploited and the fraction of purchased IT
the IT capital began to depreciate, the coefficient
fall.
Total capital stock
is
included primarily as a control variable.
According to a
transaction cost framework, industries which require large fixed capital investments ceteris paribus
,
have larger average firm
The
size.
coefficient
on
TOTSTOCK
will,
should
thus be positive.
The sectors.
sector
dummy
variables control for systematic variations in firm size
For instance, manufacturing firms are
much
typically
between
larger firms than firms in
the service sector.
The time dummies capture year or external shocks.
Because firms
be expected to be lower years like 1984.
If
there
in
is
to year
lay-off
economy-wide changes such
workers
in recessions,
the recession years of 1980
any overall trend
in
variables already in the model, this should also
16
average firm size would
and 1982 and higher
firm size that
show up
in
is
as recessions
in
recovery
not captured by the
the time
dummies.
4.
4.1
Results
Regressions on establishment
As predicted by
ITSTOCK
is
negative
magnitude
is
large
size
the hypothesis that IT leads to smaller firms, the coefficient on
in all regressions. It
enough
to
have the opposite sign of the
is
significant at the 0.01 level
ITSTOCK
coefficients
variables are also as predicted.
correlated with establishment
size.
Dummy
space limitations, were also as expected.
all
and are smaller
in
magnitude, as
is
positively
1).
Total capital stock
variables, though not
shown here because of
For instance, manufacturing establishments
were larger than service establishments and firm 1982, after controlling for
its
be economically important. The coefficients on ITINV
predicted by the learning curve hypothesis (see table
The other
and
the other variables.
but perhaps somewhat misleading because the intersectoral variations in firm size.'^
17
size
was lowest
in
the recession year of
The unusually high
dummy
R*^
is
encouraging
variables "explain" the large
Table
1:
Basic Model
Dependent Variable: Establishment
VARIABLE
Size from
CBP
COEFFICIEN1
ITSTOCK ITINV ITINV(-l)
TOTSTOCK R' Durbin-Watson significant at the 0.1 level (one-tailed test)
significant at the 0.05 level (one-tailed test) significant at the 0.01 level (one-tailed test)
Table
2:
Model with
ITSTOCK
Lagged by Zero
Dependent Variable: Establishment
N =
2
(t-statistic)
(1.470)
(1.282)
TOTSTOCK (t-statistic)
0.8156*** (4.011)
(3.863)
R' Durbin-Watson
(1.802)
0.7484***
0.5811***
0.999
0.999
0.999
0.999
0.999
1.770
1.753
1.822
1.865
1.862
significant at the 0.1 level (one-tailed test) significant at the 0.05 level (one-tailed test)
***
significant at the 0.01 level (one-tailed test)
technology
The
-0.0232 (0.555)
0.6309***
(2.852)
(2.842)
**
The
-0.0447
(1.028)
*
4
3
-0.0824**
-0.0653*
-0.0732
CBP
Size from
I
ITSTOCK(-N)
Four Years
to
0.6922***
(3.000)
results of the regressions suggest that increased levels of information
in a sector lead to a
null hypothesis that
decrease
in
the average size of firms in that sector.
IT does not affect firm
significance in the basic specification
and
size
can be rejected
at the 0.05 level
at the 0.01 level of
when ITSTOCK
is
lagged by
two years.
While the data from County Business Patterns available,
we sought
is
the most comprehensive
to confirm our findings by running the
same regressions on a
different data set to help guard against the possibility that our findings
artifact of the
County Business Patterns
data.
are presented in the next section.
19
The
were somehow an
results of the confirmatory regressions
4.2
Are the
results
an
artifact of the
CBP
data?
County Business Patterns, the source of our dependent every establishment
in
the country.
for better data gathering
in the
It
may be
and small firms
sample. Another possibility
is
that
that information technology simply allows
were previously omitted are now included
that information technology reduces the size of
individual establishments but enables firms to
CBP
variable, seeks to record
grow by adding more establishments. The
data would not detect such growth.
This alternative was examined by constructing a different measure of firm size
from data available through Compustat as described
same data gathering
less susceptible to the
in
section
biases as the
CBP
The Compustat data
II.
is
data and furthermore,
allows us to detect changes in the size of multi-establishment firms.
The table
4.
results using this
They show,
decreasing firm
ITSTOCK
is
size.
in
new source
dependent variable are
even stronger fashion, that increasing IT stock
In the basic
significant
for the
model and four out of
beyond the
0.01 level.
five of the
The lagged
years.
^^
20
size,
and
associated with
lagged regressions,
regressions
concave pattern of negative correlation between IT and firm
is
in table 3
show
the
same
again peaking after two
Table
3:
Basic Model
Dependent
variable:
Firm Size from Compustat
VARIABLE
COEFFICIENT
ITSTOCK
-0.3253***
ITINV
0.0494*
(1.552)
ITINV(-l)
0.0962**
(2.578)
TOTSTOCK
0.8180***
(3.982)
R2 Durbin-Watson
0.999
Table
4:
Model with ITSTOCK Lagged by Zero Size
N =
(t-statistic)
(4.743)
1.0849***
4-
3
2
(4.019)
(5.192)
1.0104***
(3.455)
0.999
0.999
0.999
0.999
0.999
1.995
2.203
2.119
2.031
1.659
*
significant at the 0.1 level (one-tailed test)
**
significant at trie 0.05 level (one-tailed test)
***
significant at the 0.01 le\el (one-tailed test)
The combined evidence
of the preceding analysis of
21
(2.596)
0.6222***
0.8113***
(4.833)
(5.542)
(5.464)
R' Durbin-Watson
Four Years
-0.1429***-0.1698***-0.1934***-0.1745***-0.1302**
(4.140)
TOTSTOCK
to
from Compustat
1
ITSTOCK(-N)
(4.824)
2.075
Dependent Variable: Firm
(t-statistic)
T-ST ATI STIC
CBP
0.6004***
(3.060)
and Compustat data
clearly indicates that increases in IT capital stock are associated with significant declines
in the
number
of employees per firm.
While
this result
is
consistent with the hypothesis
that IT leads to a reduction in the proportion of hierarchical coordination, IT might
affect firms in other
For instance, activities,
it
is
ways that could also lead to the relationship observed
possible that firms continued to provide roughly the
in the data.
same scope of
but IT enabled them to do so with fewer employees. This could result
significantly increased the productivity of
each worker directly or
some way
in
if
IT
facilitated
the substitution of capital for labor.
While we cannot
rule out either of these explanations, previous studies
provided broad support for either proposition.
A
have not
growing literature on what has come to
be known as the "IT productivity paradox" finds that IT may not lead to significant increases in productivity (Loveman, 1988; Roach, 1989)'".
Nor
that IT leads to the substitution of capital for labor, at least
(Osterman, 1986).'^ While
it
is
beyond the scope of
IT productivity paradox, the positive coefficient on
this
per firm suggest that IT fact, total capital
is
among
paper to
TOTSTOCK
regressions as well as the results of additional regressions
there clear evidence
is
clerks
directly
in
and managers
examine the
the previous
we ran on
capital investments
not leading to a significant substitution of capital for labor.
spending per firm actually decreased
in industries
In
with greater IT
stock'^
Accordingly,
we
focus our attention on the less widely examined possibility that IT
22
facilitates a relative increase in
market coordination and a commensurate reduction
the scope of each individual firm's contribution to the value chain. series of smaller firms coordinated through
managed by
larger, integrated firms.
decline in firm size
we observed
markets undertake the
phenomenon could
This
This means that a activities previously
not only explain the
above, but also leads to the further prediction of
reduced levels of vertical integration following increases
4.3
in
in
ITSTOCK.'^
Information technology' and vertical integration
The added
results of the regressions of information technology
shown below, support
to sales, as
reduces firm size
ITSTOCK,
is
the hypothesis that one
it
feasible to divest functions that
Once
expensive to contract through the market.
when
a
100%
increase in IT
is
integration, with a significance in excess of
variables
is
ratio of value
mechanism by which IT
by reducing the degree of vertical integration.
firms apparently find
a 2 year lag,
on the
Following increases
were previously too
again, the impact of IT
associated with a
is
strongest after
3.1% decrease
in vertical
99%. The interpretation of the other
consistent with previous results.
23
in
Basic Model
Dependent Variable:
VARIABLE ITSTOCK
Vertical Integration Index
COEFFICIENT
T-STATISTIC
5.
Conclusion
5.1
Summary
Using data from different sources, we have been able to relationship
size.
The
between increased
decline in firm size
levels of information technology
is
immediately.
impact of IT
We
in
This finding
the
same year
may shed
light
on
usage and smaller firm
new technology
earlier studies
that the investments
are not fully
which found
little
is
at least partly
explained by a
on market coordination following IT investments,
accordance with earlier theory.
Not only does
or no
were made.^*'
also find evidence that the decline in firm size
relative increase in reliance
document a
greatest with a lag of two years following investments in
information technology', suggesting that the impacts of the felt
clearly
in
vertical integration decline following
IT
investments, but the Compustat data, which allowed for multi-establishment firms showed a greater decline in firms size following IT investments than did the data
based on single
establishments.
It is
worth noting that alternative explanations for the decline
25
in
firm size in this
period, such as increased international competition, would not directly explain the strong correlation
we found
However, we
with increased stocks of information technology.
cannot rule out more complex relationships.
For instance,
if
some
third trend in the
economy
is
associated with both increased IT usage and with smaller firm size after a
brief lag,
it
could produce the correlation
5.2
we found
in
our regressions.
Future research
In this paper,
we have focused on changes
investments within the same sector or industry. the externalization of services,
manufacturing sector to the smaller firms, such a
examination of
shift
in
may be
However,
in
employment from
identified in this studv.
we have
largely
presumed
a
dichotomy
between markets and hierarchies. Perhaps some of the most interesting will
An
a fruitful direction for future research.
interpreting our results
information technology
the
Because the service sector generally has
would exacerbate the trend we would be
to the extent that IT enables
contributing to the shift
ser\'ice sector.
this possibility
Secondly,
it
with IT
in firm size associated
effects of
be the enabling of new organizational forms such as
"networks" (Antonelli, 1988; Piore. 1988), "ad-hocracies" (Toffler, 1982) or more complex forms.
forms
Future research should seek to identify and where possible quantify these new in
order to establish whether, how, and why IT affects their implementation.
26
This paper has examined the relationship between information technology stock
and a key indicator of the restructuring of the American economy, firm constituted less than
investment trend,
in
1%
of
all
IT amounts to
capital stock in the period
50%
we examined,
size.
While IT
current net
of net increases capital stock (Kriebel, 1988).
combined with the relationship between IT and firm
American economy may see even more
size,
suggests that the
radical restructuring in the next decade.
27
This
Notes 1.
We
will
generally use the terms "information technology" and "IT" as shorth-
d for the
Bureau of Economic Analysis (BEA) category "Office, Computing and Accounting Machinery", which is comprised primarily of computers. Other authors use slightly different definitions but the basic trends are similar regardless of exact specification.
This data was compiled by Lou Gorman at the BEA and for the first time includes reasonably accurate hedonic price deflators for the computer industry. Because information 2.
technology has experienced such large annual performance improvements, old data which did not take quality changes into account could be severely misleading.
was reported in a chart in the Wall Street Journal on April 27, 1988, p. 29, which also showed that intermediate size classes showed employment gains inversely proportional to their size. The data from County Business Patterns cited by Piore and the data compiled by David Birch refer to establishments, several of which are sometimes owned by a single corporation. 3.
The data from
4.
A less
the
Bureau of Labor
Statistics
noted feature of agency analysis is that this result may be reversed when improved is provided to the agent as well as, or instead of, to the principal (Brynjolfsson,
information 1989).
manufacturing firms. Eccles and White (1988) found that the use of transfer prices between divisions in the same firm was a ver\' common alternative to pure "price" or "authority" mechanisms. 5.
In a field study of 13
The deflator for information technology was more IT each year than it did the year before.
6.6.
7.
It
should be noted that
strictly
actually an "inflator".
A
dollar bought
speaking, establishments are not equivalent to firms,
CBP. A study by Carlsson (1988) found that the correlation between changes in the number of establishments and the number of firms in a sample of manufacturing industries was over 91%. although the vast majority of firms consist of a single establishment as defined
8.
The value
of industry shipments
is
in
defined as the amount received on net receivable
and allowances, and excluding freight charges and Value added by manufacture is derived by first converting the value c shipments to the value of production by adding in the ending inventory in finished goods and work in process, and subtracting the beginning inventory. The cost of materials, supplies, containers, fuels, purchased electricity, and contract work is subtracted from the value ol production to obtain the value of manufacture. The ratio of value added by manufacture to shipments will increase monotonically as the degree of vertical integration increases. Changes in reporting invemories to the LIFO method beginning in 1982, make value added selling value, f.o.b. plant, after discounts
*"
excise taxes.
28
rrsi..
38
7
Date Due
AUG,
29
]934
Lib-26-67
MIT LIBRARIES DUPl
3
TOAD DD7D15bb
T
HD28 .iyi414
DOES INFORMATION TECHNOLOGY LEAD TO SMALLER FIRMS? Erik Brynjolfsson
Thomas W. Malone Gurbaxani Ajit Kambil
Vijay
November 1989 CiSR Sloan
WP
WP
No. 211
No. 3142-90 No. 106
CCS WP
Center for Information Systems Research Massachusetts
Institute of
Sloan School of
Technology
Management
77 Massachusetts Avenue Cambridge, Massachusetts, 02139
DOES INFORMATION TECHNOLOGY LEAD TO SMALLER FIRMS? Erik Brynjolfsson
Thomas W. Malone Gurbaxani Ajit Kambil
Vijay
November 1989 CISR
WP
No. 211
WP No. 3142-90 CCS WP No. 106
Sloan
© 1989
E. Brynjolfsson,
T.W. Malone, V. Gurbaxani, A. Kambil
Center for Information Systems Research Sloan School of Management Massachusetts Institute of Technology
J I.
-'
t
1
1-
-
.:J0
Does Information Technolog>' Lead
to
Smaller Firms?
Abstract
We
examine the relationship between information technology capital and firm using industry data for the entire U.S. economy. The results indicate that increased
size
stocks of information technology are associated with significant decreases in firm size as
measured by the number of employees. One explanation
for this observations
is
that
information technology enables reduced levels of vertical integration and our analysis of data for the U.S. manufacturing sector supports
this
explanation.
We
also find that the
on organizations are most pronounced after a lag of two years. While the correlations we find cannot, of course, prove causality, the evidence is consistent with the hypothesis that information technology reduces transaction costs and coordination costs, enabling a shift from hierarchies to markets as a means of coordinating economic activity. effects of information technology
The authors would like comments and insights,
numerous helpful and in improving the specification of the econometric model. We would also like to thank the Management in the 1990s project. Digital Equipment Corporation and the UCI Committee on Research for pro\iding funding for this research. to thank Professor Ernst
Berndt for
his
especially in identifying the relevant data sources
1.
2.
Introduction: the research problem
1
Background and hypotheses 2.1 Trends in the Icey variables 2.1.1 Firm size
2
2
3
Information technology 2.2 Theoretical studies of IT and firm 2.1.2
3.
5 size
5
Data and methodology 3.1
8
The data
8
3.1.1
Information technology capital and total capital
3.1.2
Firm
size:
employees per establishment and per firm
3.1.4 Data grouping and Methodology 3.2.1 The basic model 3.2.2 The lagged model
dummy
variables
13
14 15
17
Results 4.1
Regressions on establishment
4.2
Are the
results
an
Conclusion 5.1
Summan,'
5.2 Future research
NOTES
and
17
size
artifact of the
4.3 Information technology
5.
11
12
3.3 Predicted signs of the coefficients
4.
9 10
3.1.3 Vertical integration
3.2
8
CBP
data?
vertical integration
20 23 25 25
26 28
1.
Introduction: the research problem
Industrialized
economies have recently entered a period of substantial
organizational change.
This transition has been likened to a "second industrial divide"
(Piore and Sabel, 1984) or even the beginnings of a "post-industrial society" (Huber, 1984; and the studies reviewed therein).
Among
the postulated aspects of the transition
are decreases in firm size, a shift to externally provided services, less hierarchical
management processes specialization,
and increases
The growing drop
in
within firms, a shift from mass production to flexible
in
product and service variety.
attention to these organizational changes has coincided with a rapid
the price of computing
power (Gordon.
technology (IT)' usage, and one in
general.
infers,
1987), significant increases in information
decreases
in the price of
information processing
Unfortunately, empirical research on the relationship between IT and
organizational structure has produced few
if
any reliable generalizations.
taken as a whole, the literature on IT impacts presents
many
Indeed,
when
contradictory results
(Attewell and Rule, 1984).
We investments
have recently been able to obtain detailed, economy-wide data on in
IT
in the U.S..'
enabling the use of econometric techniques to more
1
broadly study
impact.
its
The
principal
aim of
this
paper
is
to
perform an empirical
examination of the impact of information technology on an important characteristic of
economic organization: firm Because firm
economy
size
is
as well as
size as
measured by the number of employees per
a key variable in
some of
many
of the descriptions of a post-industrial
the theories on the impact of information technology, our
results should also help illuminate the theoretical literature
firmer empirical base.
For
firm.
and put future work on
a
instance, our findings that increases in information technology
stock are associated not only with declines in employees per firm, but also with reduced vertical integration, support the theory that information technology reduces transaction
costs (Brynjolfsson,
Malone and Gurbaxani, 1988) and coordination
costs (Malone, Yates
and Benjamin, 1987; Malone and Smith, 1988).
The paper
consists of five
main
sections.
Section
II
provides background on the
recent trends in IT usage and average firm size, and presents a brief discussion of the
hypothesized relationship between these two variables.
methodology and data used regressions and explores
in this study.
In section
III,
we
explain the
Section IV presents the basic results of the
some explanations
for the results.
We
conclude with some
interpretations of the results and suggestions for further research in section V.
2.
2.1
Background and hypotheses
Trends
in
the key variables
The data
number
the
and
Whether or not
Firm
new
trends in the
last
10 years:
1)
firm size, as measured by
of employees in the average business establishment, has decreased
substantially,
2.1.1
reveal two
2) the real stock of information technology has
grown enormously.
the trends are related, the evidence for each independently
is
strong.
size
Recent trends
are widely reported.
in firm size
Piore (1986) cites data from
Count)' Business Patterns showing that the average establishment size has been
Using a
decreasing since the 1970s, reversing an earlier trend towards ever-larger firms. different data set. Birch (1979)
data revealed that the
all
new
12%
of
jobs while the
findings.
His analysis of
Dun &
Bradstreet
66%
27%
of establishments under 20 employees accounted for
46%
of establishments that had over 100 employees generated only
of
new jobs.
More
recently, the trend has apparently intensified.
Statistics reports that
new
had similar
from 1980
to 1986 firms of
The Bureau
under 100 employees created
jobs while firms of over 1000 employees experienced a net loss of
The phenomenon
is
industrial countries.
not unique to the U.S. but
As shown
in
graph
Netherlands also show a recent decrease early 1970s (Huppes. 1987).
1,
is
six million
1.5 million jobs.^
also being experienced by other
data from the
in
of Labor
average firm
UK, Germany, and size,
major
the
despite increases until the
Carlsson (1988) also found a similar pattern
in
the
manufacturing sectors of other Western industrial countries and hypothesized that facilitated
was
it
by computer-based technology.
Our examination a function of
of data reveals that the decrease in average establishment size
two complementar)' trends:
1) in sectors that
have many large firms, such as
manufacturing, overall emplojTnent has not kept up with the growth firms, resulting in significant declines in average firm size,
is
and
in
the
number
of
2) in sectors with typically
smaller firms, such as services, employment has grown rapidly, while average firm size
has remained small. (See graph
A
related trend, which
is
2).
consistent with the above observations,
externalization or "decoupling" of business functions.
a
45%
This
is
is
the increasing
manifested, for example,
in
increase from 1980 to 1986 in the use of outside contractors and consultants (Wall
Street Journal, 1988).
A
detailed study of the metal-working industry found that
amount
of value added to shipments declined in 88 of 106 sectors between 1972 and 1982 and that this could
article
be tied to increased use of information technology' (Carlsson. 1988).
on how "value-adding partnerships" are supplanting
companies, Johnston and Lawrence (1988) also argue that
In
an
vertically integrated
this
phenomenon
enabled by information technolog}'. Apparently, the wave of mergers and
is
partly
LBOs
in the
1980s have not reversed this trend as highly leveraged firms have spun-off divisions and
downsized even their core business (Scherer, 1988).
2.1.2 Information technology,'
Evidence of the increase
abundant even without looking
computer workstation 1988).
in
power and ubiquity of information technology
A
at the statistics.
for every five
employees
in
1988 study found that there was one surveyed companies (Nolan Norton,
After controlling for depreciation and quality improvements,
we
find that the
stock of IT has increased almost exponentially since 1970, from just over capital stock to nearly seven percent in 1985 (see
sectors shows the
equipment
that
graph
Each of
3).
same accelerating trend toward increased use of
was
largely insignificant
is
two decades ago
is
of overall
the major business
IT.
rapidly
1%
A
category of
becoming very
important.
Current investments business (Kriebel, 1988).
in
IT are estimated
at
50%
of
all
net capital purchases by
Meanwhile, over half of the labor force
tasks that primarily involve information processing (Porat, 1977).
we
detect today, these
in the
numbers are of
sufficient
magnitude
to
is
currently engaged in
Whatever
effects of
IT
augur even greater effects
near future.
2.2 Theoretical studies of
IT and firm
Theory suggests ways enable a reduction
size
that IT can lead to larger firms as well as
in firm size.
A
ways that IT can
seminal article by Leavitt and Whisler (1958)
5
emphasized the
One
organization.
centralization
effect of
IT on reducing the costs of information transmission within the
of the conclusions of this argument was that IT would lead to greater
and broader spans of hierarchical
control.
The
principal-agent literature
has also emphasized that the agency costs of hierarchy decrease
in
better information
and monitoring by the principal (Jensen, 1983)/ To the extent that the costs of centralization
and agency are reduced by
IT,
one might expect
to see
more use of
hierarchy and therefore larger firms.
This hypothesis contrasts with an analysis of IT's impacts based on transaction cost
economics. According to Williamson (1975, 1985). production
governance of a single hierarchy
in
is
brought under the
response to high transaction cost of organizing
through the market. These "market failures" are caused by asset and
bounded
rationality
site specificity,
and uncertainty, impacted information, and small numbers
bargaining. Information technology can be expected to mitigate these market failures and
thus reduce the need to rely on hierarchies (Malone, Yates and Benjamin, 1987; Brynjolfsson,
A
Malone and Gurbaxani,
line of analysis
technology
will
1988).
by Malone and Smith which considers that information
reduce the costs of coordination both inside and outside the organization
reached similar conclusions (Malone (1987) and Malone and Smith (1988)). They argue that because of the relatively larger coordination requirements of
markets as opposed to
hierarchies, reductions in coordination costs will be particularly beneficial for
market
Accordingly, these models predict a decrease in the proportion of
coordination.
hierarchical coordination following investments in IT.
A the
related effect of information technology
"minimum
efficient scale" of production.
is
that
it
might be expected to change
For instance, Klein, Crawford and Alchian
Grossman and Hart (1986) and Hart and Moore (1988) have emphasized
(1978),
strong technological complementarities between assets can
common
lead to
ownership of
large, related sets of assets.
techniques like flexible manufacturing,
may decrease
it
make markets However,
if
IT
that
inefficient
and
facilitates
the specificity of assets, and thus
transform hierarchical production to production organized through smaller units
coordinated by markets.
In summan,'. the theoretical literature suggests that IT will reduce the costs of
coordination both within firms and between firms. the balance will shift towards
know, a
priori
,
whether
There
is
some reason
more use of markets and smaller
this effect exists
and
is
to believe that
firms, but
we cannot
of an economically significant magnitude.
Furthermore, theorists have concentrated on the organizational forms which dominated the last centun-: markets and hierarchies.
It is
very possible that IT
organizational forms with different characteristics.
For instance,
if
will
enable
new
large firms used
information technology to set-up or simulate internal markets, average firm size might increase with information technology' usage despite a greater reliance on markets."
Fortunately, the question of
how information
technolog)' affects firms size
is
subject to empirical investigation. Previously, a case study of the evolution of airline
reservation systems (Malone, Yates and Benjamin, 1987). found that information
technology was associated with increased market coordination.
In this paper,
we
use
econometric techniques to generalize those results to the U.S. economy as a whole.
3.
Data and methodolog>
Our approach was
to use available U.S. data to directly test the relationship
between information technolog)' stock and average firm eight sectors: agriculture; mining; durable
manufacturing; transport and real estate;
and
the direction
services.
A
utilities;
are divided into
series of regressions
retail trade; finance,
were then run on
this
and magnitude of the relationship between IT and firm
tested and data
insurance and
data to
size,
identify'
while
and year. Several regression models were
from alternative sources was also used
The data
3.1.1
The data
goods manufacturing; non-durable goods
wholesale and
controlling for total capital stock, industr)'
3.1
size.
Information technology capital and total capital
to help validate the results.
Data from the Bureau of Economic Analysis (BEA) was used
to derive figures for
information technolog>' stock and investments, and total capital stock by industry for each year.
The
industries.
BEA
data classifies
The data
each industry,
total
economic
all
activity in the
gathering methodology
is
described
annual investments are measured
in
United States into 61
in
Gorman
For
et al. (1985).
27 asset categories.
We
use the
category "office, computing and accounting machinery" for our IT figures and the
all
categories for our total capital figures.
quality-adjusted (hedonic) price deflator.
Each
asset category also has
sum
of
an associated
By multiplying each investment by
its
associated deflator, nominal investments were converted into constant-dollar or "real"
To
investments.*
derive capital stocks from investment flows,
Winfrey table which assumes their expected
mean
a
Normal
distribution of
lifetime as described in
percentage of existing equipment
is
Gorman,
we used
a modified
equipment service et al. (1985).
eliminated based on
its
lives
around
Thus each
age while
new
year, a
stock
is
added
from that year's investments.
3.1.2
Firm
size:
employees per establishment and per firm
Data on average establishment
size
is
available
from County Business Patterns"
(CPB) annual summaries. The number of business establishments and number of employees
is
provided for each sector of the economy.
derived by dividing the latter by the former.^ size
were only available from 1977
to 1984.
Average establishment
size
is
Because consistent data on establishment
we used
those years as the priman,' period of
our study.
Although the data from County Business Patterns in the
census of establishments
sectors of the
number
of firms
While
this
Using the SIC code, we grouped
in
sample
is
it
still
size directly.
was
that the
the companies into the
number
of employees and
includes over 2000 firms for each year.
size.
It
also has
two
from "establishments", the Compustat data
Secondly, consistent Compustat data was available for a
CPB
data.
The
only change from the previous
Compustat data included too few firms
sector to provide a reliable statistic, so this sector
Compustat
its
presumably biased towards large firms (small firms are not
longer time period (1975-1985) than the specification
the
all
each sector. This provided a second measure of firm
First, to the extent that "firms" differ
measures firm
can be
Compustat maintains data on every
economy used above and added up
usually publicly traded),
virtues.
size
company, including number of employees and the SIC code of
principal line of business.
same
the most comprehensive
United States, another measure of firm
created from an alternative source: Compustat. publicly traded
is
in
the agricultural
was omitted from the regressions with
data.
3.1.3 Vertical integration
We
also explore the impact of IT on levels of vertical integration to help provide
10
insight into the
mechanism by which IT
Vertical integration
affects firm size.
degree to which multiple stages of production are combined under
and coordinated within a 1962).
Our
operational measure of
the total value shipped per firm.^ variation of the value
and
later
opposed
single firm as
added over
this
to a
concept
is
common
is
ownership
sequence of smaller firms (Gort,
the ratio of value added per firm to
This operationalization of vertical integration sales
measure
the
originally
is
a
proposed by Adelman (1951)
modified to the current version by Tucker and Wilder (1977).
The data required
to construct the vertical integration index
Census of Manufacturers and the Annual Survey of Manufacturers sectors of the
economy. Unfortunately, no comparable data
of the economy.
The data on
1985 (see Kambil. 19SS)
at the
vertical integration
two
digit
SIC code
is
was analyzed
is
available from the
for the manufacturing
available for other sectors for the period 1975 to
level of aggregation.
This
distinguishes 20 different industries within the manufacturing sector.
3.1.4
Data grouping and
The
dummy
variables
data for the dependent and explanatory variables was categorized by the
Standard Industrial Classification (SIC) codes. The 61 industries used by the identified by their
CPB
SIC codes and were then grouped
for the regressions
were
into the eight sectors identified by
on each of the measures of firm
size and, as
mentioned above,
into 20 manufacturing industries for the vertical integration regressions.
11
BEA
We
recognize that other trends
may
under study. To control for time-specific to IT,
we
included
a particular sector,
dummy
also have
been important during the period
effects like recessions or other trends unrelated
To
variables for each year.
we included dummy
control for effects specific to
variables for each of the industry sectors as well.
Because the most natural interpretation of the both the dependent and explanatory variables
in
is
terms of percentage-effect, the natural logarithm was taken of
all
the
relevant variables.
3.2
Methodology
The
basic technique used for analyzing the data
The panel
panel data.
CBP
was least-squares regression on
consisted of eight sectors observed from 1977 to 1984 for the
data and 1975 to 1985 for the other regressions on firm
integration regressions discussed in section
IV.D were on
size.
The
a panel of
vertical
20 manufacturing
industries for the period 1975 to 1985.
Because the sectors were of different
sizes,
each observation on a sector was weighted by the
employment presence of correlation
in that sector that year.^
serial correlation so the
was used
The
initial
leading to possible heteroskedasticity, size of the sector, as
proxied by total
regressions revealed the potential
generalized differencing correction for serial
in all regressions.'^
12
3.2.1
The
basic
Our integration.
technology
model
basic
A
model allows
number
may be
for lags in the
impact of IT on firm
the investment.
their full
new
technologies because in effect
impact on firm
is
to include
both
only
it
some
Lags are commonly observed
takes time to learn
how
to best use
in the
and apply
of the most recent years' investments will have
size.
The technique used at first
vertical
and Gordon, 1988; Curley and
particularly susceptible to lags (Baily
Thus,
and
of authors have suggested that the effects of information
Pyburn, 1982; Nolan Norton, 1988; Loveman. 1988). assimilation of
size
to
model the
ITSTOCK
to
possibility that
IT
may be
only partially effective
measure the basic impact of
IT,
and the
t^A•o
most
recent years of IT investments (ITINV). to measure the extent to which the effects of current investments are not immediately felt."
ITINV and ITINV(-])
ITSTOCK, "effective"
will
will
will
learning
is
indeed taking place,
have the opposite sign but be smaller
thus reducing the effect of
IT stock
If
be captured
in
ITSTOCK
in
for recent years.
the coefficient on
ITSTOCK
magnitude than All of the impact of
while the learning
be measured by the relative magnitude of the coefficient on ITINV.''
A
measure of
total capital stock in the sector,
TOTSTOCK.
control for the fact that firm size will in general be larger larger.
Finally, effects specific to
each
industry'
13
when
was included
to
total capital stock
and year were controlled
is
for by including
two
appropriate
sets of
The
basic
model
dummy
is
variables.
specified as:''
log(SIZE) = Bo + 6,*log(ITSTOCK) +B2*]og(ITINV) + 33*log(ITINV(-l))
+ 64*log(TOTSTOCK) + S B,*INDUSTRY + 1 Bj*YEAR.
While there
is
likely to
be some collinearity
among
the variables in this model,
regression theory assumes that the independent variables are coUinear and the least
squares formulation
is
expressly designed to separate out the effects of each of the
However, while even severe
variables.
required for accurate regression,
it
collinearity
will in
does not violate any of the conditions
general lead to lower significance levels of the
coefficients.
3.2.2
The lagged model
To and firm years.
better observe the overall time path of the relationship
size, a series
between IT stock
of separate regressions was run with IT lagged from zero to four
Although none of these regressions individually
will
learning effect, taken together, they would be expected to effectiveness of IT over time.
14
be able to capture any
show any changes
in
the
The
five regressions
used to observe the time path of IT were:
log(SIZE) = Bo + Bi*log( ITSTOCK[-N]
+
)
B.*log(
TOTSTOCK
)
+ 1 B,*INDUSTRY + 2 Bj*YEAR. where
N
e {0,...,4}, representing lags
from
to 4 years.
3.3 Predicted signs of the coefficients
If
information technology leads to smaller firms, then
negative sign on the coefficient of
how much
ITSTOCK
in
both models.
we would Its
a percentage change in IT stock will reduce firm size,
expect to see a
magnitude all
will tell us
else being equal.
In
the basic model, the coefficient should be interpreted as the impact of one unit of
"effective"
IT capital and should thus be larger than the coefficient
in
any one of the
lagged regressions.
The
learning curve effect would lead to a positive sign on
model, opposite the sign on of the coefficient on
ITSTOCK.
ITSTOCK.
Its
Including
magnitude should be
ITINV
will
ITINV less
in
the basic
than the magnitude
thus cancel a fraction of the most
recent years' investments in IT because they have not yet reached their
full
effect
on firm
size.
In the lagged model, the impact of the
15
IT stocks should be increasing for small
lags
and then returning close
rise results as the
stock that
would
is
technology
to zero in distant years, exhibiting a
is
more
"effective" increases.
As
fully
concave pattern. The
exploited and the fraction of purchased IT
the IT capital began to depreciate, the coefficient
fall.
Total capital stock
is
included primarily as a control variable.
According to a
transaction cost framework, industries which require large fixed capital investments ceteris paribus
,
have larger average firm
The
size.
coefficient
on
TOTSTOCK
will,
should
thus be positive.
The sectors.
sector
dummy
variables control for systematic variations in firm size
For instance, manufacturing firms are
much
typically
between
larger firms than firms in
the service sector.
The time dummies capture year or external shocks.
Because firms
be expected to be lower years like 1984.
If
there
in
is
to year
lay-off
economy-wide changes such
workers
in recessions,
the recession years of 1980
any overall trend
in
variables already in the model, this should also
16
average firm size would
and 1982 and higher
firm size that
show up
in
is
as recessions
in
recovery
not captured by the
the time
dummies.
4.
4.1
Results
Regressions on establishment
As predicted by
ITSTOCK
is
negative
magnitude
is
large
size
the hypothesis that IT leads to smaller firms, the coefficient on
in all regressions. It
enough
to
have the opposite sign of the
is
significant at the 0.01 level
ITSTOCK
coefficients
variables are also as predicted.
correlated with establishment
size.
Dummy
space limitations, were also as expected.
all
and are smaller
in
magnitude, as
is
positively
1).
Total capital stock
variables, though not
shown here because of
For instance, manufacturing establishments
were larger than service establishments and firm 1982, after controlling for
its
be economically important. The coefficients on ITINV
predicted by the learning curve hypothesis (see table
The other
and
the other variables.
but perhaps somewhat misleading because the intersectoral variations in firm size.'^
17
size
was lowest
in
the recession year of
The unusually high
dummy
R*^
is
encouraging
variables "explain" the large
Table
1:
Basic Model
Dependent Variable: Establishment
VARIABLE
Size from
CBP
COEFFICIEN1
ITSTOCK ITINV ITINV(-l)
TOTSTOCK R' Durbin-Watson significant at the 0.1 level (one-tailed test)
significant at the 0.05 level (one-tailed test) significant at the 0.01 level (one-tailed test)
Table
2:
Model with
ITSTOCK
Lagged by Zero
Dependent Variable: Establishment
N =
2
(t-statistic)
(1.470)
(1.282)
TOTSTOCK (t-statistic)
0.8156*** (4.011)
(3.863)
R' Durbin-Watson
(1.802)
0.7484***
0.5811***
0.999
0.999
0.999
0.999
0.999
1.770
1.753
1.822
1.865
1.862
significant at the 0.1 level (one-tailed test) significant at the 0.05 level (one-tailed test)
***
significant at the 0.01 level (one-tailed test)
technology
The
-0.0232 (0.555)
0.6309***
(2.852)
(2.842)
**
The
-0.0447
(1.028)
*
4
3
-0.0824**
-0.0653*
-0.0732
CBP
Size from
I
ITSTOCK(-N)
Four Years
to
0.6922***
(3.000)
results of the regressions suggest that increased levels of information
in a sector lead to a
null hypothesis that
decrease
in
the average size of firms in that sector.
IT does not affect firm
significance in the basic specification
and
size
can be rejected
at the 0.05 level
at the 0.01 level of
when ITSTOCK
is
lagged by
two years.
While the data from County Business Patterns available,
we sought
is
the most comprehensive
to confirm our findings by running the
same regressions on a
different data set to help guard against the possibility that our findings
artifact of the
County Business Patterns
data.
are presented in the next section.
19
The
were somehow an
results of the confirmatory regressions
4.2
Are the
results
an
artifact of the
CBP
data?
County Business Patterns, the source of our dependent every establishment
in
the country.
for better data gathering
in the
It
may be
and small firms
sample. Another possibility
is
that
that information technology simply allows
were previously omitted are now included
that information technology reduces the size of
individual establishments but enables firms to
CBP
variable, seeks to record
grow by adding more establishments. The
data would not detect such growth.
This alternative was examined by constructing a different measure of firm size
from data available through Compustat as described
same data gathering
less susceptible to the
in
section
biases as the
CBP
The Compustat data
II.
is
data and furthermore,
allows us to detect changes in the size of multi-establishment firms.
The table
4.
results using this
They show,
decreasing firm
ITSTOCK
is
size.
in
new source
dependent variable are
even stronger fashion, that increasing IT stock
In the basic
significant
for the
model and four out of
beyond the
0.01 level.
five of the
The lagged
years.
^^
20
size,
and
associated with
lagged regressions,
regressions
concave pattern of negative correlation between IT and firm
is
in table 3
show
the
same
again peaking after two
Table
3:
Basic Model
Dependent
variable:
Firm Size from Compustat
VARIABLE
COEFFICIENT
ITSTOCK
-0.3253***
ITINV
0.0494*
(1.552)
ITINV(-l)
0.0962**
(2.578)
TOTSTOCK
0.8180***
(3.982)
R2 Durbin-Watson
0.999
Table
4:
Model with ITSTOCK Lagged by Zero Size
N =
(t-statistic)
(4.743)
1.0849***
4-
3
2
(4.019)
(5.192)
1.0104***
(3.455)
0.999
0.999
0.999
0.999
0.999
1.995
2.203
2.119
2.031
1.659
*
significant at the 0.1 level (one-tailed test)
**
significant at trie 0.05 level (one-tailed test)
***
significant at the 0.01 le\el (one-tailed test)
The combined evidence
of the preceding analysis of
21
(2.596)
0.6222***
0.8113***
(4.833)
(5.542)
(5.464)
R' Durbin-Watson
Four Years
-0.1429***-0.1698***-0.1934***-0.1745***-0.1302**
(4.140)
TOTSTOCK
to
from Compustat
1
ITSTOCK(-N)
(4.824)
2.075
Dependent Variable: Firm
(t-statistic)
T-ST ATI STIC
CBP
0.6004***
(3.060)
and Compustat data
clearly indicates that increases in IT capital stock are associated with significant declines
in the
number
of employees per firm.
While
this result
is
consistent with the hypothesis
that IT leads to a reduction in the proportion of hierarchical coordination, IT might
affect firms in other
For instance, activities,
it
is
ways that could also lead to the relationship observed
possible that firms continued to provide roughly the
in the data.
same scope of
but IT enabled them to do so with fewer employees. This could result
significantly increased the productivity of
each worker directly or
some way
in
if
IT
facilitated
the substitution of capital for labor.
While we cannot
rule out either of these explanations, previous studies
provided broad support for either proposition.
A
have not
growing literature on what has come to
be known as the "IT productivity paradox" finds that IT may not lead to significant increases in productivity (Loveman, 1988; Roach, 1989)'".
Nor
that IT leads to the substitution of capital for labor, at least
(Osterman, 1986).'^ While
it
is
beyond the scope of
IT productivity paradox, the positive coefficient on
this
per firm suggest that IT fact, total capital
is
among
paper to
TOTSTOCK
regressions as well as the results of additional regressions
there clear evidence
is
clerks
directly
in
and managers
examine the
the previous
we ran on
capital investments
not leading to a significant substitution of capital for labor.
spending per firm actually decreased
in industries
In
with greater IT
stock'^
Accordingly,
we
focus our attention on the less widely examined possibility that IT
22
facilitates a relative increase in
market coordination and a commensurate reduction
the scope of each individual firm's contribution to the value chain. series of smaller firms coordinated through
managed by
larger, integrated firms.
decline in firm size
we observed
markets undertake the
phenomenon could
This
This means that a activities previously
not only explain the
above, but also leads to the further prediction of
reduced levels of vertical integration following increases
4.3
in
in
ITSTOCK.'^
Information technology' and vertical integration
The added
results of the regressions of information technology
shown below, support
to sales, as
reduces firm size
ITSTOCK,
is
the hypothesis that one
it
feasible to divest functions that
Once
expensive to contract through the market.
when
a
100%
increase in IT
is
integration, with a significance in excess of
variables
is
ratio of value
mechanism by which IT
by reducing the degree of vertical integration.
firms apparently find
a 2 year lag,
on the
Following increases
were previously too
again, the impact of IT
associated with a
is
strongest after
3.1% decrease
in vertical
99%. The interpretation of the other
consistent with previous results.
23
in
Basic Model
Dependent Variable:
VARIABLE ITSTOCK
Vertical Integration Index
COEFFICIENT
T-STATISTIC
5.
Conclusion
5.1
Summary
Using data from different sources, we have been able to relationship
size.
The
between increased
decline in firm size
levels of information technology
is
immediately.
impact of IT
We
in
This finding
the
same year
may shed
light
on
usage and smaller firm
new technology
earlier studies
that the investments
are not fully
which found
little
is
at least partly
explained by a
on market coordination following IT investments,
accordance with earlier theory.
Not only does
or no
were made.^*'
also find evidence that the decline in firm size
relative increase in reliance
document a
greatest with a lag of two years following investments in
information technology', suggesting that the impacts of the felt
clearly
in
vertical integration decline following
IT
investments, but the Compustat data, which allowed for multi-establishment firms showed a greater decline in firms size following IT investments than did the data
based on single
establishments.
It is
worth noting that alternative explanations for the decline
25
in
firm size in this
period, such as increased international competition, would not directly explain the strong correlation
we found
However, we
with increased stocks of information technology.
cannot rule out more complex relationships.
For instance,
if
some
third trend in the
economy
is
associated with both increased IT usage and with smaller firm size after a
brief lag,
it
could produce the correlation
5.2
we found
in
our regressions.
Future research
In this paper,
we have focused on changes
investments within the same sector or industry. the externalization of services,
manufacturing sector to the smaller firms, such a
examination of
shift
in
may be
However,
in
employment from
identified in this studv.
we have
largely
presumed
a
dichotomy
between markets and hierarchies. Perhaps some of the most interesting will
An
a fruitful direction for future research.
interpreting our results
information technology
the
Because the service sector generally has
would exacerbate the trend we would be
to the extent that IT enables
contributing to the shift
ser\'ice sector.
this possibility
Secondly,
it
with IT
in firm size associated
effects of
be the enabling of new organizational forms such as
"networks" (Antonelli, 1988; Piore. 1988), "ad-hocracies" (Toffler, 1982) or more complex forms.
forms
Future research should seek to identify and where possible quantify these new in
order to establish whether, how, and why IT affects their implementation.
26
This paper has examined the relationship between information technology stock
and a key indicator of the restructuring of the American economy, firm constituted less than
investment trend,
in
1%
of
all
IT amounts to
capital stock in the period
50%
we examined,
size.
While IT
current net
of net increases capital stock (Kriebel, 1988).
combined with the relationship between IT and firm
American economy may see even more
size,
suggests that the
radical restructuring in the next decade.
27
This
Notes 1.
We
will
generally use the terms "information technology" and "IT" as shorth-
d for the
Bureau of Economic Analysis (BEA) category "Office, Computing and Accounting Machinery", which is comprised primarily of computers. Other authors use slightly different definitions but the basic trends are similar regardless of exact specification.
This data was compiled by Lou Gorman at the BEA and for the first time includes reasonably accurate hedonic price deflators for the computer industry. Because information 2.
technology has experienced such large annual performance improvements, old data which did not take quality changes into account could be severely misleading.
was reported in a chart in the Wall Street Journal on April 27, 1988, p. 29, which also showed that intermediate size classes showed employment gains inversely proportional to their size. The data from County Business Patterns cited by Piore and the data compiled by David Birch refer to establishments, several of which are sometimes owned by a single corporation. 3.
The data from
4.
A less
the
Bureau of Labor
Statistics
noted feature of agency analysis is that this result may be reversed when improved is provided to the agent as well as, or instead of, to the principal (Brynjolfsson,
information 1989).
manufacturing firms. Eccles and White (1988) found that the use of transfer prices between divisions in the same firm was a ver\' common alternative to pure "price" or "authority" mechanisms. 5.
In a field study of 13
The deflator for information technology was more IT each year than it did the year before.
6.6.
7.
It
should be noted that
strictly
actually an "inflator".
A
dollar bought
speaking, establishments are not equivalent to firms,
CBP. A study by Carlsson (1988) found that the correlation between changes in the number of establishments and the number of firms in a sample of manufacturing industries was over 91%. although the vast majority of firms consist of a single establishment as defined
8.
The value
of industry shipments
is
in
defined as the amount received on net receivable
and allowances, and excluding freight charges and Value added by manufacture is derived by first converting the value c shipments to the value of production by adding in the ending inventory in finished goods and work in process, and subtracting the beginning inventory. The cost of materials, supplies, containers, fuels, purchased electricity, and contract work is subtracted from the value ol production to obtain the value of manufacture. The ratio of value added by manufacture to shipments will increase monotonically as the degree of vertical integration increases. Changes in reporting invemories to the LIFO method beginning in 1982, make value added selling value, f.o.b. plant, after discounts
*"
excise taxes.
28
rrsi..
38
7
Date Due
AUG,
29
]934
Lib-26-67
MIT LIBRARIES DUPl
3
TOAD DD7D15bb
T
