An Empirical Analysis of the Relationship Between Information Technology and Firm Size1

Erik Brynjolfsson*

Thomas W. Malone*

Vijay Gurbaxani**

Ajit Kambil***

CCS-TR #123

Copyright 1993, 1994 MIT; Brynjolfsson, Malone, Gurbaxani and Kambil

All Rights Reserved

* Center for Coordination Science, MIT

** Graduate School of Management, UCI

*** New York University

Previous Drafts: September, 1989; September 1991

This Version: January, 1993

1 Previously Circulated under the title:

"Does Information Technology Lead to Smaller Firms?"




Many changes in the organization of work in the United States since 1975 have been attributed in the literature to the large technological improvements and the sharply higher levels of investment in computers and related technologies over the same period. Few empirical studies have attempted to empirically examine these relationships. The primary goal of this paper is to assess the hypothesis that the rapid growth of information technology (IT) is at least partially responsible for one important organizational change, the shift of economic activity to smaller firms. We examine this hypothesis using industry-level data on IT capital and four measures of firm size, including employees per firm. We find broad evidence that investment in IT is significantly associated with subsequent decreases in the average size of firms. We also find that the effects of IT on organizations are most pronounced after a lag of two to three years. Although the correlations we find do not necessarily imply causality, they are consistent with the theory that IT makes market-based coordination more attractive relative to internal coordination.

1. Research problem

Industrialized economies have recently entered a period of substantial organizational change. This transition has been likened to a "second industrial divide" (Piore & Sabel, 1984) or the "coming of the new organization" (Drucker, 1988) and has been widely discussed in the both the business and academic literatures (see (Huber, 1990); and the studies reviewed therein). Among the postulated aspects of the transition are decreases in firm size, a shift to externally provided services, and a shift from mass production to more flexible arrangements. The growing attention to these organizational changes has coincided with a rapid drop in the price of computing power (Gordon, 1987), significant increases in information technology (IT)[1] usage, and one infers, decreases in the costs of information processing in general.

Because both firms and markets have often been modeled as information processing entities (Arrow, 1973; Galbraith, 1977; Hayek, 1945), one might reasonably suspect that some of the recent organizational changes are related to the massive deployment of IT. Indeed, the theoretical links between the two have often been made in the literature (see (Crowston & Malone, 1988) for a review.) Unfortunately, empirical research on the relationship between IT and organizational structure has produced few if any reliable generalizations, in part because case studies have predominated. Indeed, when taken as a whole, the literature on IT impacts presents many contradictory results (Attewell & Rule, 1984) .

The principal aim of this paper is to empirically examine the impact of IT on one important characteristic of economic organization: firm size. To address this question, we have obtained economy-wide data on investments in IT in the U.S., enabling the use of econometric techniques to more broadly study its impact. To our knowledge, this is the only public domain data on IT investments in the US economy. It includes reasonably accurate hedonic price deflators which take into account quality improvements of over 20% per year in computing power. In order to understand the changes more fully, we examined four different measures of firm size: 1) the number of employees per establishment, 2) the number of employees per firm, 3) the sales per firm and 4) the value-added per firm.

The paper consists of five sections. Section 2 provides background on the recent trends in IT usage and average firm size, and discusses the theoretical relationship between these two variables. In section 3, we explain the methodology and data used in this study. Section 4 presents the results of the regressions and explores some explanations for the results. We conclude with some interpretations of the results and suggestions for further research in section 5.

2. Background and hypotheses

2.1 Trends in the key variables

In this section, we summarize previous evidence of two significant trends in the last 15 years: 1) The number of employees in the average business establishment has decreased substantially, and 2) the real stock of IT has grown enormously. The previous work does not examine whether these two trends are related, but the evidence for each independently is strong.

2.1.1 Firm size

According to several sources, firm size, as measured by the number of employees, has declined.[2] For instance, Piore (1986) cites data from County Business Patterns showing that the average establishment size has been decreasing since the 1970s, reversing an earlier trend towards ever-larger firms. The Bureau of Labor Statistics reports that from 1980 to 1986 firms of under 100 employees created six million new jobs while firms of over 1000 employees experienced a net loss of 1.5 million jobs. The BLS data also showed that intermediate size classes had employment gains inversely proportional to their size.

The phenomenon is not unique to the U.S. but is also being experienced by other major industrial countries. As shown in figure 1, Huppes (1987) reports that data from the UK, Germany, and the Netherlands also show a recent decrease in average firm size, despite increases until the early 1970s .

Our examination of data from County Business Patterns and from Compustat reveals the number of employees per firm has indeed declined in the manufacturing sector, but there is no strong trend in the service sector (figure 2). We also find that although firms tend to be smaller in the service sector, employment has grown more rapidly in services than in manufacturing (figure 3). In addition, we examined two other measures of firm size, sales per firm and value-added per firm, for manufacturing industries. We did not find any overall declines in these measures of firm size. (figure 4 and figure 5)

2.1.2 Information Technology

The data confirm the almost self-evident increase in power and ubiquity of IT. We find that investment in computers has increased steadily and dramatically since at least 1971. After taking into account quality improvements (which allow each dollar to buy more IT), there has been over a tenfold increase in IT investments between 1971 and 1990 (see figure 6). Each of the major business sectors shows the same accelerating trend toward increased use of IT (figure 7). A category of equipment that was largely insignificant two decades ago is now very important.

Driving much of this investment are exponential declines in the price/performance of computers and related technology (Gurbaxani & Mendelson, 1990). Furthermore, "Moore's Law", which posits a doubling of transistor density every 18 months is projected to hold into the next century; the one to two million transistors per chip in 1990 may be 50-100 million by the year 2000 (Grove, 1990)(figure 8). Whatever effects of IT we detect today, these numbers are of sufficient magnitude to augur even greater effects in the near future.

2.2 The Relationship between IT and firm size

In order to understand the impact of IT on firm size, it is useful to understand the economic rationale underlying the determination of firm size in the first place. As Friedman (1955, p. 233) notes:

"The appropriate size of firm to produce, say, copper, may be different for two different mines, and both can exist simultaneously because it is impossible to duplicate either one precisely. . . . The existing distribution [of firm sizes] reflects both 'mistakes' and intended differences designed to take advantage of the particular specialized resources under the control of different firms."

Ijiri and Simon (1977) provide more detailed models of the processes that give rise to a distribution of firms of different sizes in an industry. They note that the size distribution of firms in an industry is a result of the successes and failures of numerous individual firms over time. The success of each individual firm, in turn, is determined by a number of factors, including how well it takes advantage of its specialized resources and also how well it adapts to changing conditions in the industry as a whole.

In this paper, we are especially concerned with how firms adapt to one particular kind of change: the availability of increasingly powerful and inexpensive IT. The overall effects of this technology clearly depend on how individual firms use it and how this usage affects their subsequent performance. If each firm's use of IT leads to widely varying results, then there might be no overall patterns of change in the economy as a whole. On the other hand, if certain kinds of adaptation to IT are generally more successful than others, then we may be able to observe the resulting changes as overall trends in the economy. For instance, if the widespread use of IT increases the viability of smaller firms relative to larger ones, then this might be part of the reason for the decrease in average firm size that we noted above.

A central theoretical question, then, is why the increasing use of IT might change the relative viability of small and large firms. We summarize the arguments that have been proposed to answer this question in two basic categories: (a) labor substitution, and (b) outsourcing.

2.2.1. Labor substitution

Perhaps the simplest explanation that has been proposed for why firm size might be related to IT is that firms can sometimes use IT to produce the same output with fewer people. By substituting automated processing for human labor, the argument goes, these firms can increase productivity and reduce costs.

Somewhat surprisingly, however, previous studies have not provided broad support for the hypothesis that IT substitutes for labor or even increases the productivity of labor. For example, a growing literature on what has come to be known as the "IT productivity paradox" finds that IT does not yet appear to have led to significant increases in productivity (Brynjolfsson, 1993 in press) provides a review of the literature. See also Loveman, 1988; Roach, 1987a). Robert Solow summarized the results of this literature most succinctly when he said "We see computers everywhere except in the productivity statistics." (Solow, 1990)

Furthermore, in direct studies of the relationship between IT and employment, there is some evidence that IT may actually increase employment. For instance, Osterman (1986) found that IT investment ultimately resulted in a complementary increase in the number of clerks and managers employed after a lag of several years. Two recent papers by Morrison and Berndt (1991; 1990) also addressed this question using essentially the same IT data set we are using. By specifying a general production function, they found that IT was on balance a complement, not a substitute for labor, especially white collar labor. Specifically, they conclude: "... rather than being aggregate labor-saving, increases in [IT] tend to be labor-using" (Berndt & Morrison, 1991, p. 1).

Even though this previous work casts some doubt upon the labor substitution hypothesis as an explanation for decreasing firm size, our study will allow us to examine the hypothesis from another perspective. If labor substitution due to IT is the primary explanation for decreasing firm sizes, then we should expect to see a decrease in the number of employees per firm associated with IT use, but no decrease in the sales per firm. In fact, if this hypothesis is correct, we might even see an increase in the sales per firm associated with IT use.

2.2.2. Outsourcing

Another possible explanation for why IT might lead to smaller firms is that IT might allow firms to "outsource" more of their activities. That is, the use of IT might lead firms to "buy" rather than "make" more of the components and services needed to make their primary products. For instance, when a firm like Ford needs tires for the cars it produces, it has two choices about how to obtain these tires: It can make them internally, or it can buy them from a tire supplier. Which of these choices is preferable in a given situation depends, in part, on their respective costs.

We can divide these costs into two categories--production costs and coordination costs. Production costs refer to the costs of the physical production process itself--tasks like molding and cutting the rubber for tires. Coordination costs, on the other hand, refer to the costs of managing the dependencies between production tasks. For example, coordination tasks include making sure that the rights things and the right people are at the right places at the right times.

We can further divide coordination costs into two subcategories--internal coordination costs and external coordination costs. When Ford produces its own tires, for instance, the internal coordination costs include the costs of managers and others who decide when, where, and how to produce the tires. When Ford buys tires from an outside supplier, the external coordination costs include (a) the supplier's costs for marketing, sales, and billing and (b) Ford's costs for finding suppliers, negotiating contracts, and paying bills.

In both cases, coordination costs include information intensive activities such as gathering information, communicating, and making decisions. Since IT is particularly useful in these kinds of information intensive activities, several previous theories suggest how IT might affect firm size by reducing these coordination costs. As the following subsections describe, the theories make different predictions, depending on which kinds of costs are affected most (see Gurbaxani and Whang (1991) for a summary). Reducing internal coordination costs more than external

If IT reduces the costs of internal coordination more than external coordination, then we would expect firms to make more things internally. This means we would expect firms to grow in size. For example, if IT greatly reduced the costs for managers to monitor and control what their subordinates were doing throughout a large organization, then this might lead firms to make more things internally where they could be more controlled more effectively and at less cost than if they were purchased from an outside supplier.

One kind of internal coordination cost simply involves moving information to the places where decisions are made and then informing others about the decisions. Another important kind of internal coordination cost arises because the interests of individual employees are often not the same as those of the firm as a whole. Research in agency theory has studied extensively how these conflicts of interest can be managed by approaches such as monitoring employees or providing them with performance-based pay (Eisenhardt, 1989; Jensen, 1983) . Since these approaches involve information-intensive activities, it seems plausible that IT might affect their costs. More generally, since many early applications of computers have focused on internal systems rather than interorganizational systems, we might expect that these systems would affect internal coordination costs more than external ones. Reducing external coordination costs more than internal

If IT reduces the costs of external coordination more than internal coordination, then we would expect to see firms buy more things externally. In this case, the average size of firms should decrease. For example, if it is easier and cheaper for a firm to find an external supplier for new parts than to make them internally, then the firm is more likely to buy the parts outside and less likely to need as much internal manufacturing capacity.

The factors that lead to high "transaction costs" for external coordination have been analyzed extensively by research in transaction cost theory (Williamson, 1975; 1985) . In general, the "opportunistic" behavior of firms negotiating contracts with each other often leads to costs (such as legal and accounting expenses) that would not be necessary if the same transactions were coordinated internally. For example, when a supplier invests in special machinery that is useful only for one customer, the supplier is vulnerable if that customer threatens to buy somewhere else. Similarly, when it is difficult for buyers to find out about alternative sources of supply, they are vulnerable to excessive charges from the supplier they customarily use.

While IT certainly cannot eliminate opportunistic human behavior, its functions (such as increasing the availability of information) can reduce the problems opportunism causes (Brynjolfsson, Malone & Gurbaxani, 1988; Malone, Yates & Benjamin, 1987) . More generally, by reducing the costs of many of the information searching and accounting activities that are needed for coordination with external suppliers, IT can make buying things externally more attractive to firms.

A related effect of IT is that it might reduce market coordination costs by changing the "specificity" of assets themselves. For instance, Klein, Crawford and Alchian (1978) , and Grossman and Hart (1986) have emphasized that when assets are specific to one another, market coordination will be inefficient and this may lead to common ownership of large, related sets of assets. However, if IT facilitates techniques like flexible manufacturing, it may decrease the specificity of assets, and thus transform internal production to production organized through smaller units coordinated by markets. Applying the Grossman-Hart paradigm, Brynjolfsson (1991b) formally modeled the role of asset flexibility, and also showed that under appropriate assumptions, a broader distribution of production knowledge and coordination knowledge would favor market coordination. Reducing coordination costs more than production costs

The theories reviewed so far allow us to predict either kind of change: if internal coordination costs decrease most, firms should grow; if external coordination costs decrease most, firms should shrink. Malone and colleagues argue that, in general, we should expect both kinds of coordination costs to decrease relative to production costs. They further argue that this would still favor buying rather than making (Malone, 1987; Malone & Smith, 1988; Malone, Yates & Benjamin, 1987) .

The basis of this argument is summarized in Figure 1. First, as noted above, the costs of finding suppliers, negotiating contracts, and paying bills often make external coordination more expensive than coordinating the same activities internally would be (Williamson, 1975; 1985) . However, when external suppliers pool the demands of multiple customers, they can often realize economies of scale or other production cost advantages that internal production could not achieve. Thus, in general, buying rather than making leads to higher coordination costs but lower production costs.

Coordination mechanism
Coordination Costs
Production Costs
External ("buying")
Internal ("making")

Figure 1. Relative costs of "buying" components and services externally vs. "making" them internally.

Now, if IT reduces both internal and external coordination costs, it will decrease the importance of the dimension on which buying has a disadvantage. Thus, it should increase the number of situations in which is more attractive than making.

If more outsourcing occurs because of this or any of the other effects described in this subsection, we should expect to see a decrease in the average number of employees per firm. Unlike the labor substitution hypothesis, however, the outsourcing hypothesis predicts that the activities required to produce a product will be divided among more separate firms. This means that, if the outsourcing hypothesis is correct, the average amount of "value added" and the average sales per firm should decrease.

There is some evidence suggesting that IT has led to a decline in firm size in certain industries. A detailed study of the metal-working industry found that vertical integrated firms were "decoupling" into smaller firms in 88 of 106 sectors between 1972 and 1982 and that this could be tied to increased use of IT (Carlsson, 1988). In an article on how "value-adding partnerships" are supplanting vertically integrated companies, Johnston and Lawrence (1988) also cite examples in which this phenomenon is partly enabled by IT. However, to date there has been no economy-wide study to determine whether these changes are part of a broader trend.

2.2.3. Summary

In summary, the theoretical literature suggests that IT will reduce the costs of coordination both within firms and between firms. We cannot know, a priori, however, which effect predominates and whether a resulting shift is of an economically significant magnitude. Furthermore, these theoretical arguments do not allow us to determine whether the decrease in average firm size noted above is related--either positively or negatively--to the increasing use of IT.

Fortunately, the question of how IT affects firm size is subject to empirical investigation, and as noted above, the different theories make different predictions about what changes we should see. In the remainder of this paper, we use econometric techniques to analyze the relationship in the U.S. economy as a whole between IT usage and firm size. We also interpret these results in light of the theories just described.

3. Data and methodology

Our approach was to use available U.S. data to directly examine the relationship between IT and average firm size. The data are divided into six sectors: durable goods manufacturing; non-durable goods manufacturing; transport and utilities; wholesale and retail trade; finance, insurance and real estate; and services. These six sectors represent substantially all manufacturing and services industries in the U.S.[3] A series of regressions were then run on this data to identify the direction and magnitude of the relationship between IT and firm size, while controlling for total capital stock, foreign trade, industry, and capital costs. Several regression models were examined and data from alternative sources was also used to help validate the results.

3.1 The data

There are at least three basic ways to measure the size of firms: number of employees, total sales, and total value added.

3.1.1 Two measures of employees 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. Average establishment size is derived by dividing the latter by the former, generating the variable SIZE1. [4]Consistent data on establishment size were available from 1976 to 1989, so we used those years as the primary period of our study.

Although the data from County Business Patterns is the most comprehensive census of establishments in the United States, another measure of firm size, SIZE2, can be created from an alternative source: Compustat. Compustat maintains data on every publicly-traded company, including number of employees and the SIC code of its principal line of business. Using the SIC code, we grouped all the companies into the same sectors of the economy used above and added up the number of employees and number of firms in each sector. This provided a second measure of firm size.

While the Compustat sample is presumably biased towards large firms (small firms are not usually publicly traded), it still includes over 2000 firms for each year. It also has a virtue: to the extent that "firms" differ from "establishments", the Compustat data measures firm size directly.

3.1.2 Sales per firm and value added per firm

We also explored the impact of IT on two other definitions of firm size: value of sales per establishment (SIZE3), and value added per establishment (SIZE4). These additional regressions serve as an extra robustness check on the employment-based measures of firm size. They also help provide insight into the mechanism by which IT affects firm size: if IT leads to a decrease in shipments and value-added per firm then substitution effects alone do not underlie the shift to smaller firms.

The data required for these two series is available from the Census of Manufacturers and the Annual Survey of Manufacturers for the manufacturing sectors of the economy. Unfortunately, no comparable data is available for other sectors of the economy. The data was analyzed for the period 1976 to 1989 at the two digit SIC code level of aggregation. This distinguishes 20 different industries within the manufacturing sector.

The value of industry sales is defined as the amount received on net receivable selling value, f.o.b plant, after discounts and allowances, and excluding freight charges and excise taxes. This quantity (which is sometimes called "shipments") is then divided by the number of establishments, from County Business Patterns, to derive the average sales per firm.

Value added by manufacture is defined as the value of the finished goods minus the value of the raw materials and other supplies. This number is derived by first converting the value of 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 costs of materials, supplies, containers, fuels, purchased electricity and contract work are subtracted from the value of production to obtain the value added of manufacture. Value added per establishment is then calculated by dividing total value added by the number of establishments.

3.1.3 IT capital and total capital

Data from the Bureau of Economic Analysis (BEA) was used to derive figures for IT investments, and total capital investments by industry for each year. The BEA data classifies all economic activity in the United States into 61 industries. For each industry, total annual investments are measured in 27 asset categories. We used category 14 -- "office, computing and accounting machinery" (OCAM) for our IT figures[5] and the sum of all other categories for our total capital figures. Over 90% of the OCAM category consisted of computers and associated peripherals, electronic calculating machines and accounting machines but it also included typewriters and "other office machines". Although imperfect, we considered the OCAM category to be the best proxy for the new electronics-based information processing technologies that may have an impact on organizational structure; the BEA data is the most authoritative government dataset on computing investments in the United States (see e.g. (Gorman, Silverstein, Gerald & Comins, 1985) (Berndt & Morrison, 1991) ).

It should be noted that the OCAM data include capital equipment only, and do not include spending on software development, maintainance or related services. Furthermore, some computing power is embedded in scientific instruments, robotic machinery and even appliances. These "computers" are also excluded from our data. Instead, we focus on a somewhat narrower definition of informaton technology that stresses office and coordination uses. While such uses account for the vast majority of computing power and are in any event likely to be highly correlated with other uses, our data is far from "ideal". Accordingly, we generalize our findings about IT to other manifestations of computing power only with great caution.

According to the BEA, the OCAM and total capital investment data were based on US National Income and Product Accounts (NIPA) annual investment expenditures by type of asset and allocated proportionately across industries by using data from the BEA's capital flow tables, which were available at five year intervals (1967, 1972, 1977, 1982 and 1987). The BEA made interpolations and extrapolations for other years, using control totals from numerous independent sources (e.g. the Annual Survey of Manufactures, payroll data, employment data, inventory data, etc.) and some technical modifications were also made, for instance to account for ownership changes and investments by non-profit institutions. The industry-level data originate from a variety of sources and the overall methodology is described by the Bureau of Economic Analysis (1987) and Gorman et al. (1985).[6]

Each asset category also has an associated quality-adjusted (hedonic) price deflator. The deflators were the same as those used in deriving constant dollar GNP estimates. Most were based on Producer Price Indexes published by the Bureau of Labor Statistics. Those for computers were based on industry data as described in Cartwright (1986). By dividing each investment by its associated deflator, nominal investments were converted into constant-dollar or "real" investments.[7] One advantage of this deflator is that allows comparison of different types of computers on the basis of "computing power", including processor speed, memory, storage capacity, display and various specific features. As a result, if some types of firms purchase one type of computers and other purchase another type, our results will be based on the difference in computing power delivered.

3.1.4 Foreign Trade and Interest Rates

A principal alternative explanation for the decline in firm size is that increased competition from foreign companies has forced American companies to "downsize" and become more efficient. To explore this hypothesis, and to help control for the impact of changes in foreign trade, we included it as an independent variable. Trade was defined as the sum of US merchandise exports and imports for each year, as reported by the Department of Commerce. The data exclude military transactions and were converted from nominal dollars into constant dollars using the GNP deflator.

Because high interest rates discourage business expansion and could also lead to a decline in firm size, the yearly rate of interest on AAA rated bonds was also included. This data was obtained from the Economic Report of the President (Bush, 1991) .

3.1.5 Data grouping and dummy variables

The data for the dependent and explanatory variables was categorized by the Standard Industrial Classification (SIC) codes. The 61 industries used by the BEA were identified by their SIC codes and were then grouped into the six sectors identified by CPB for the regressions on SIZE1 and SIZE2 and, as mentioned above, into 20 manufacturing industries for regressions on SIZE3 and SIZE4.

To control for effects specific to a particular sector that do not change over time, we included dummy variables for each of the industry sectors. The most natural interpretation of the interaction among the dependent and explanatory variables is in terms of percentage-effect, so in accordance with common practice, all the relevant variables were expressed in term of the natural logarithm. Another feature of this specification is that it can be interpreted as a multiplicative structural model in which the parameters are exponents. The estimating equation we use is then derived by taking the natural logarithm of both sides.

3.2 Methodology

The basic technique used for analyzing the data was a two stage least-squares regression estimate of the correlation between IT and various measures of firm size, while controlling for other explanatory variables. The data for all sectors over the time period were pooled and, as described below, corrections were made for potential heteroskedasticity, serial correlation and simultaneity.

Because IT capital and other capital are long-lived, a distributed lag structure for the investments is appropriate. This enables us to assess the impact of several years of IT and other capital investments and to determine the time-path of the effects. We allow for general patterns of depreciation (or more accurately, effectiveness) by examining a polynomial distributed lag for IT investments. By fitting a second-order polynomial distributed lag structure, we can include the current year and four previous years of data on investments in IT without introducing excessive collinearity or unduly reducing the degrees of freedom. A concave pattern (peaking in an arbitrary year), as well as various other patterns such as the conventional straightline or declining-balance depreciation, can be approximated by second order polynomials.[8]

By construction, only changes in firm size within a sector were examined by our models. However, part of the decline in average firm size was due to shifts among sectors. This effect was expressly excluded from our specification, because so many other factors may affect shifts among sectors, and because it does not appear to account for most of the decline in firm size.[9] Nonetheless, to the extent that the value chain for any one product involves production across several sectors, the outsourcing hypothesis discussed above would also be consistent with an intersectoral shift which is not captured in our specification. [10]

Finally, as discussed above, four different dependent variables were fitted, to reduce the sensitivity of the results to particular data-gathering biases and to further explore the mechanism(s) by which IT affects firm size.

3.2.1 The model

The model measures the relationship between levels of IT stock and the average size of firms for a given sector in a given year, while controlling for total capital, foreign trade, interest rates and industry specific effects.

The basic regression model was:

This specification constrains the lag weights to follow a second degree polynomial for the first five lagged values. By substituting into the initial equation, a coefficient on the each of the lagged values of IT can be computed and the resulting equation can then be directly estimated using ordinary least squares.[11]

The sum of coefficient on IT, S wi , will indicate the extent to which a percentage change in IT capital is correlated with changes in firm size, after controlling for the other variables in the equation. Predicted values of all the coefficients are given in section 3.3 below.

3.2.3 Data pooling and correction procedures

To increase the efficiency of the estimation, the time series and cross-sectional observations were pooled. This provided an adequate sample size but introduced potential problems like heteroskedasticity among sectors and serial correlation between successive years. Furthermore, by pooling the data, we implicitly restricted the effect of IT and the other explanatory variables to be the same across sectors and across time, and indeed, by estimating a least squares regression, assumed that the relationship is (log-)linear. If the true values differ across sectors or the relationship is non-linear, the resulting estimates will be a weighted average, and should be interpreted in this light. For instance, a zero coefficient could be the result of positive effects in one sector being cancelled by negative effects elsewhere, or a positive relationship over one range of values being cancelled by a negative relationship for other values. Finally, it is possible that the explanatory variables (IT and other capital) will be jointly determined with firm size, at least in the long run, and thus may be correlated with the error term leading to inconsistent estimates. Fortunately, we were able to correct for these potential problems by modifying the specification of the model as outlined in the remainder of this section.

Heteroskedasticity was a potential problem because the size of the six sectors varied significantly in the pooled data, suggesting a violation of the assumption of constant error variance. Although the exact relative error variances cannot be determined, it is common practice in the economics literature to assume that they vary inversely with the size of the sector. Accordingly, the weighted least squares correction technique was applied: each observation on a sector was weighted by the size of the sector, as proxied by the share of gross domestic product (GDP) produced in that sector that year. We first computed the mean of the weighting series (GDP) and then multiplied all the variables by the ratio of the weighting series to its mean (GDP/GDP) yielding a specification that corrects for heteroskedasticity and provides efficient parameter estimates (Pindyck & Rubinfeld, 1991) .

As might be expected in time series regressions, the initial regressions also revealed the potential presence of serial correlation, as manifested in generally low Durbin-Watson statistics. Accordingly, the Hildreth-Lu correction procedure was used in all regressions: generalized differencing was applied to all the variables and results were reported for the value of the estimated correlation coefficient, r, that resulted in the lowest sum-of-squared residuals (Hildreth & Lu, 1960). The corrected regressions did not exhibit significant serial correlation.

As mentioned above, there is the possibility that some of the independent variables (IT and other capital stock) are correlated with the error term. This could occur for a number of reasons. First, according to standard production theory, the optimal levels of the various factors of production: employment, IT capital, and other capital are each functions of the other two (at least in the long-run) as well as other exogenous variables which are observable to managers, but not to us. Secondly, theory suggests that managers minimizing coordination costs will choose both the firm size and IT investment with regard to levels of the other. Either of these complications would result in simultaneity: a shock that affected firm size might also affect investment in the same year. Thirdly, all of the variables are measured only with some error. It is possible that the measurement errors for the independent variables may be correlated with the error term from the regression. The appropriate correction for these possible biases is to use Instrument Variables (IV) estimation as outlined below, instead of Ordinary Least Squares (Berndt, 1991; Pindyck & Rubinfeld, 1991) .

To derive consistent parameter estimates, we ran the regressions in two stages, using lagged values of IT and TOT as well as price index for IT investments as instruments to minimize the simultaneity problem. The first stage consisted of estimating a reduced form equation for ITt, TOTt and the other independent (and presumably exogenous) variables using ordinary least squares by regressing them on a constant, a polynomial distributed lag of ITt-1, a polynomial distributed lag of TOTt-1, TRADEt, BONDt, the industry dummies and polynomial distributed lag of PRICEt (a price index for computers, to overidentify the equation). By using these variables, all of which are presumed to be predetermined, the fitted values of IT and TOT will be asymptotically independent of the current error term, by construction. The second stage regression simply replaced the original variables in equation (1) with these fitted variables, yielding consistent estimators, at the expense of somewhat higher standard errors.

It should be noted that, although we modified the regression to correct for potential sources of bias and to improve efficiency, the interpretation of the sign and magnitude of the cofficients is the same as for ordinary least squares (Pindyck & Rubinfeld, 1991) .[12]

3.3 Predicted signs of the coefficients

According to the hypothesis that IT leads to smaller firms by reducing the costs of inter-firm coordination, we should expect the sum of the coefficients on IT to be negative for the regressions on all four measures of firm size.. Conversely, a positive sign would indicate that IT facilitated internal coordination more than market coordination. If IT is a substitute for labor, we should also see negative coefficients for the regressions on SIZE1 and SIZE2, but SIZE3 and SIZE4 would not be affected. Thus, while all four measures give indications of the significance of IT's impact on firm size, the latter two can provide some insight into the mechanism of the effect.

The magnitude of these coefficients will tell us how much a percentage change in IT will affect firm size, all else being equal. The sum of the coefficients can be interpreted as the percentage change in firm size associated with a 1% change IT capital. Because of lags due to learning and other adjustments to the technology, we expect the impact of the IT investments will be increasing for small lags and then returning close to zero in distant years, exhibiting a concave pattern.[13] The rise results as the technology is more fully exploited and the fraction of purchased IT stock that is "effective" increases. As the IT capital begins to deteriorate, the coefficient would fall.

It is sometimes argued that the rapidly increasing globalization of the economy increases competitive pressures on American firms and forces them to "downsize" (Kanter, 1989) ; Roach, 1987a #59. The underlying assumption is that previously they were inefficiently large or overstaffed. This effect would lead to a negative coefficient on TRADE, at least for SIZE1 and SIZE2. However, an opposite effect is at least equally plausible. Globalization leads to larger markets and provides an opportunity to further exploit scale economies (Caves & Bradburd, 1988) . This could lead to a larger average firm size, and a thus positive coefficient on TRADE.

The control variables have relatively straightforward interpretations. Growth in the output of the economy is indicative not only of cyclical business expansion, but also increased opportunity to exploit scale economies. This should lead to a positive coefficient on GNP. According to a transaction cost framework, industries which require large fixed capital investments will, ceteris paribus, have larger average firm size. The coefficient on TOT should thus be positive. The current costs of capital, as proxied by interest rates on AAA bonds, should be negatively correlated with growth of firms, leading to a negative coefficient on BOND. The sector dummy variables control for systematic variations in firm size between sectors. For instance, manufacturing firms are typically much larger firms than firms in the service sector, and this should be reflected here.

4. Results

The overall results indicate that IT is correlated with a decline in all measures of firm size and the TRADE is correlated with an increase in average firm size. We also found evidence suggesting the impact of IT is not immediate but rather that it is strongest after about 2 years.

4.1 Results for SIZE1

We found that the current level of IT stock is strongly correlated with a decline in average establishment size, SIZE1. The coefficients indicates that each 1% increase in IT investment for five years is associated with a 0.13% decrease in the number of employees per establishment, after controlling for the other variables, and is significant at the 99.9% level (see table A). A distinct concave pattern of effects is evident in the individual years' investments in IT. Declines in establishment size are correlated with IT investments for all lags, but the strongest impact is for investments that have been lagged for one to two years, with a coefficients of 0.032 and 0.031 respectively and t-statistics of 3.5 and 2.7.

Interestingly, TRADE was strongly and unambiguously correlated with an increase in average firm size, as measured by SIZE1. The coefficient was 0.32 and the t-statistic was 4.5

The other variables are consistent with our predictions. Total capital stock and GNP are positively correlated with establishment size. The coefficient on BOND was negative, but it was not significantly different than zero. Dummy variables for each sector show that manufacturing establishment tend to be larger than service firms. The unusually high R2 is encouraging but somewhat misleading because it is largely due to the dummy variables which "explain" the large intersectoral variations in firm size.[14]

The results of this regression suggest that increased levels of IT in a sector lead to a decrease in the average size of firms in that sector. The null hypothesis that IT does not affect firm size can be rejected at the 99.9% level of significance. Furthermore, the impact of IT on establishment size is of a sufficient magnitude to be economically important. Because of the large increase in IT investment over the period, this effect could account for the 20% decline in establishment size between 1977 and 1985, reversing its long term upward trend. The alternative hypothesis that increased globalization lead to smaller firms is not supported. It is in fact contradicted by the data.

While the data from County Business Patterns is the most comprehensive available, we sought to confirm our findings by running the same regressions on several different data sets to help guard against the possibility that our findings were somehow an artifact of the County Business Patterns data. These results are presented in the next sections.

4.2 Results for SIZE2

This impact of IT was further examined by constructing a different measure of firm size (SIZE2) from data available through Compustat as described in section 2. The Compustat data is less susceptible to the same data gathering biases as the CBP data used to construct SIZE1 and furthermore, allows us to detect changes in the size of multi-establishment firms. For instance, it is possible is that IT reduces the size of individual establishments but allows firms to grow by adding more establishments. The CBP data would not detect such growth.

The results using this new source for the dependent variable also show that increasing IT stock is associated with decreasing firm size (table A). Thus sum of the coefficients on IT was -0.14, almost identical to that in the SIZE1 regressions, and it was significant at the 99% level of confidence. Furthermore, TRADE again had a positive coefficient, and it was significant at the 95% level. Two differences from the previous regression were that 1) the fit did not seem to be quite as good, as indicated by the lower F-statistic, R2 and t-statistics and 2) the coefficients on the individual lags for IT investment did not follow the hypothesized concave pattern, although the once-lagged value was still the most significant.

The coefficients on the other variables are also qualitatively similar to those in the previous regressions with the notable exception that the trade sector tends to have larger firms, ceteris paribus, as measured by Compustat, presumably because chain stores involving multiple establishment are counted as only one firm.[15]

The combined evidence of the preceding analysis of CBP and Compustat data strongly suggest that increases in IT capital stock are associated with significant declines in the number of employees per firm. This result is consistent with both the hypothesis that IT leads to a reduction in the proportion of internal coordination as well as the hypothesis that IT substitutes for labor within firms, enabling firms to continue to provide roughly the same scope of activities, but with fewer employees. While we cannot rule out either of these explanations looking only at employment per firm, the next set of regressions are particularly supportive of the coordination cost hypothesis.

4.3 Results for SIZE3 and SIZE4

We next examined the relationship between IT and two alternative measures of firm size: gross shipments per firm (SIZE3) and value-added per firm (SIZE4). As discussed in section 3.1.3, data for these variables were available at the two-digit SIC code level of aggregation for manufacturing.

The results for both these alternative measures provide very strong additional evidence that IT has lead to a decline in the average size of firms (table B). They suggest that, ceteris paribus, doubling IT stock in an industry leads to a decrease in average sales per firm in the same industry of about 13% and a decrease of about 12% in average value added per firm over a period of five years. Both of these estimates are significantly different from zero at a 99.9% level of confidence and are remarkably consistent with the estimates from the SIZE1 and SIZE2 regressions on different data. The distributed lag specification for both SIZE3 and SIZE4 show the expected concave shape, peaking after two years. The coefficients on lags of from 1 to 3 years are all negative and significant at the 99% level.

Because these alternative measures of firm size are unlikely to be affected by shifts among factors of production, the combined evidence of these regressions with other studies of the effect of IT on labor intensity and productivity (Berndt & Morrison, 1991; Loveman, 1988; Osterman, 1986), suggest that the decline in employees per firm was not caused solely by the substitution of capital for labor.

As discussed earlier, the observed reduction in all measures of equilibrium size of firms is consistent with the hypothesis that IT is facilating the "decoupling" of existing vertically integrated firms and the supplanting of existing firms by value-adding networks of new, smaller firms.

4.4 Results on subsamples of the data

4.4.2 Different time periods

To assess whether the relationship between IT and firm size changed over time, we ran the model on subsamples disaggregated by year. There was sufficient data to examine the effect on each of the four measures of firm size in three overlapping samples corresponding respectively to the first 2/3 of the time period (i.e. 1976-1984), the second 2/3 (1981-1989), and a long difference of the first and last third of the period (excluding 1981-1984).

As shown in tables C and D, the net relationship between IT and all four measures of (SIZE1, SIZE2, SIZE3 and SIZE4) was consistently negative in all three time periods. The effect was statistically significant at at least the 99% level in 9 of the 12 cases examined. The coefficients on the sum of IT investments ranged from a low of .08 to a high of .39. For all four measures of firm size, the effect was strongest in the latter 2/3 years as compared to the first 2/3 of the sample. In all twelve cases, IT had the statistically most significant impact after a lag of 1-2 years and a concave pattern was apparent in 6 of the twelve regressions.

4.4.1 Manufacturing versus services

For the SIZE1 and SIZE2 regressions, we were also able to disaggregate the data into "manufacturing" and "services"[16]. "Manufacturing" consisted of two of the six sectors (Durable and Non-durable goods manufacturing) grouped together. The other four sectors (Transport and utilities; Trade; Finance, insurance and real estate; and, other Services) comprised "services". Although the number of observations were limited for each subsample, we were able to identify some potential differences and to further assess the robustness of our overall results.

In manufacturing industries, IT investments were associated with economically important declines in both SIZE1 and SIZE2 at the 99% level of significance when all five years of investments were considered. (table E). For service industries, IT was correlated with declines in SIZE1 at the 99% level of significance and at the 95% level for SIZE2. Once again, IT investments were most strongly correlated with declines in firms size when they were lagged by either 1 or two years.

Undoubtedly, the impact of IT differs across sectors, industries and even firms. In our sample, there was only a small difference between manufacturing and services. The coefficients were slightly smaller in the service industry regressions for both measures of firm size, but a Chow test indicated that the difference was not statistically significant.

International trade in both sectors was correlated with increases in both measures of firm size. This increase was statistically significant in all regressions except the manufacturing subsample for SIZE2.

On balance, the results of this set of regressions and those on the different time periods suggest that the correlations between IT and declines in various measures of firm size are not unique to any one sector or any one time period. Although some differences were found, they were relatively small so we are hesitant draw any inferences from them. Instead, we conclude only that the relationship between IT and firm size appears to be fairly widespread in the economy.

5. Conclusion

5.1 Summary

There is substantial evidence of a relationship between increased levels of IT usage and smaller firm size. The overall relationship is robust to a variety of specifications and at least four measures of firm size. However, our findings should not be interpreted to apply to all industries and all time periods. Our data were far from perfect and the relatively undeveloped state of the economic theory of the firm and of its relationship to IT in particular made the use of detailed functional specifications unrealistic. Our purpose was primarily to develop generalizable models that could be applied broadly across the economy. As we continue to develop more detailed and industry-specific models, and secure better data sets, we will be able to make sharper predictions. Nevertheless, our findings are consistent with the hypothesis that IT leads to a decline in the average size of firm, enabling us to empirically distinguish among some of the predictions of prior theoretical work.

The decline in firm size is greatest with a lag of one to two years following investments in IT, suggesting that the impacts of the new technology are not fully felt immediately. This finding may shed light on previous studies which found little or no impact of IT in the same year that the investments were made, and is consistent with earlier work on the effective time-path of capital depreciation (Pakes & Griliches, 1984).

The decline in two of our measures of firm size (SIZE1 and SIZE2) could be explained by the substitution of IT capital for labor, but the substitution hypothesis would not explain the decline in sales per firm (SIZE3) or value added per firm (SIZE4). In contrast, our findings on all four measures are completely consistent with a relative increase in the reliance on markets for coordination following IT investments. Such a shift is specifically predicted by Malone and Smith (1988) and Malone, Yates and Benjamin (1987) and also provides an empirical basis for evaluating the relative importance of the various forces identified in the other models of the relationship between IT and firm structure that are reviewed in section 2.2.

It is worth noting that the principal alternative explanation for the decline in firm size in this period, increased international competition, apparently does not explain the strong correlation we found with increased investments in IT. On the contrary, trade is strongly correlated with increases in firm size in virtually all of our regressions.

Although the null hypothesis that IT was not correlated with declines in firm size is strongly rejected by the data, other factors may also have played a role. Furthermore, we cannot rule out more complex relationships. For instance, if some third trend, like increased turbulence in the economy, is associated with both increased IT usage and with younger, smaller firms after a brief lag, it could produce the correlation we found in our regressions.

There is, however, strong reason to believe that most of the growth in IT investment has an exogenous technological basis. Increased investment in IT appears to be almost entirely explained by the rapid drop in its price (Gurbaxani & Mendelson, 1990; Chow, 1967) , and these price declines are directly attributable to improvements in the technology, especialy for chip production (Grove, 1990) . Although causality can never be proven by statistical regression, the data do support the hypothesis that improvements in technology have enabled a shift toward smaller firms.

5.2 Implications

This finding may be useful to corporate practitioners. For instance, one implication is that the current downsizing of firms, the popularity of outsourcing, and the rise of value-adding partnerships is not simply a management fad, but rather may have a technological and theoretical basis. Companies evaluating success strategies in an environment of increasingly inexpensive information technology will benefit from considering alternative forms of organizing which depend more heavily on market coordination. For instance, the comment by the former chairman of GM that their historically high levels of vertical integration were no longer a competitive advantage but had instead become a "semi-disadvantage" worth several hundred dollars per car is indicative of the way in less-integrated smaller firms have benefitted. Our findings also suggest that it may be no accident that IT is often the catalyst for "reenginering" projects that result in greater outsourcing and leaner internal staffing.

Government can help or hinder this process. Rather than seeking to support large failing firms, government may wish to direct more efforts toward understanding whether a coalition of smaller firms would be more successful in the long run. It is also important to recognize that initiatives like funding a fiber optic communications infrastucture and setting standards for EDI can accelerate the trend which we have identified and may require commensurate changes in industrial organization to fulfill their potential. However, since we find evidence that industries are already adapting to the new technological environment, we cannot conclude that any special taxes or subsidies based on firm size are required.

5.3 Future research

In this paper, we have focused on changes in firm size associated with IT investments within the same sector or industry. However, to the extent that IT enables the externalization of services, it may be contributing to the shift in employment from the manufacturing sector to the service sector. Because the service sector generally has smaller firms, such a shift would tend to amplify the trend we identified in this study. An examination of this possibility would be a fruitful direction for future research.

Secondly, in interpreting our results we have largely followed the standard approach of presuming a dichotomy between markets and firms, and therefore clear firm boundaries (Williamson, 1975) . Perhaps some of the most interesting effects of IT will be the enabling of new organizational forms such as "networks" (Antonelli, 1988; Piore, 1989), "ad-hocracies" (Toffler, 1982) or more complex forms. Future research should seek to identify and where possible quantify these new forms in order to establish whether, how, and why IT affects their implementation.

Thirdly, while the reduced form models analyzed in this paper are provocative, a better understanding of the theory of the firm and a more formal theory of the relationship between IT and firm structure are needed for more definitive hypothesis testing. The recent work in agency theory and the property rights approach to the theory of the firm are promising avenues to explore and should lend themselves to econometric analysis of specific functional forms.

Fourthly, our finding that IT is related to the decline in firm size may shed light on another aspect of the current restructuring of the Western economies: the recently-discovered benefits of "focus" by firms. The new conventional wisdom has been bolstered by Comment and Jarrell (1991), who concluded "Whatever the reasons, the dramatic shift in corporate policy to reduce corporate diversification and to increase corporate focus, which is empirically verified by our data for the 1980s, has been associated with substantial gains to shareholders in the focus-increasing firms." Similar findings are reported by Morck, Shleifer and Vishny (1990) [17]who found that diversification was worse for shareholders in the 1980s than in the 1970s, and Lichtenberg (1990) who confirmed that de-diversification became a profitable strategy between 1985 and 1989. Because diversification is typically modeled as a response to market failures, it would be interesting to assess whether IT has helped enable the emerging strategy of increased focus by increasing the relative efficiency of product and capital markets.

Finally, this paper has examined the relationship between IT stock and one key indicator of the restructuring of the American economy, firm size. By examining more refined measures, it will be possible to evaluate stronger conclusions about the size and form of these changes. While IT constituted less than 10% of all capital stock in most of the period we examined, net investment in IT amounts to up to 30% of net increases in capital stock in 1990. This trend, combined with the relationship between IT and firm size identified above, portend a potentially more radical restructuring of the American economy in the next decade.


Coordination MechanismCoordination CostsProduction Costs
External ("buying")HighLow
Internal ("making")LowHigh
Figure 1. Relative costs of "buying" components and services externally vs. "making" them internally.

No other figures available.

Regression Tables


Dependent Variable is SIZE1Dependent Variable is SIZE2
VariableCoefficientT StatisticCoefficientT Statistic
C -3.806 -4.537*** -2.911 -1.465**
ITNV(0) -0.02803 -0.792 -0.1375 -1.641*
ITINV(-1) -0.03217 -3.525*** -0.05542 -2.564***
ITINV(-2) -0.03144 -2.771*** -0.00063 -0.023
ITINV(-3) -0.02583 -1.469** 0.02687 0.645
ITINV(-4) -0.01535 -1.101 0.02708 0.82
IT SUM -0.1328 -5.916*** -0.1395 -2.625***
TTINV(0) 0.1522 2.278*** 0.2194 1.386*
TTINV(-1) 0.09994 4.270*** 0.07407 1.336*
TTINV(-2) 0.05854 3.134*** -0.02057 -0.465
TTINV(-3) 0.02809 1.013 -0.06446 -0.981
TTINV(-4) 0.00857 0.385 -0.05760 -1.093
TT SUM 0.3474 5.286*** 0.1509 0.969
GDP 0.2418 1.604* 0.2749 0.7699
BOND -0.0002447 -0.06345 0.003698 0.4049
TRADE 0.3211 4.501*** 0.2826 1.672**
DU 0.7530 46.69*** 0.4523 11.84***
ND 0.7259 18.88*** 0.3915 4.300***
TU 0.1952 4.650*** 0.8495 0.8545
TR -0.04841 -3.371** 0.4001 11.76***
FI -0.07374 -2.990*** -0.2277 -3.898***
R - Squared 0.9910 0.9608
Durbin-Watson Statistic 2.0360 1.868
F Statistic 655.6 145.0
Number of Observations 84 84 84 84

Dependent Variable SIZE3Dependent Variable is SIZE4
VariableCoefficientT StatisticCoefficientT Statistic
C-4.225 -9.019*** -4.728 -8.180***
ITNV(0) -0.01185 -0.490 -0.01611 -0.540
ITINV(-1) -0.04352 -4.865*** -0.03780 -3.425***
ITINV(-2) -0.05094 -3.504*** -0.04125 -2.300**
ITINV(-3) -0.03409 -2.668*** -0.02646 -1.678**
ITINV(-4) 0.0070 0.361 0.00657 0.275
IT SUM -0.1334 -9.193*** -0.1150 -6.426***
TTINV(0) 0.05650 1.501* 0.02419 0.521
TTINV(-1) 0.06273 4.224*** 0.03339 1.822**
TTINV(-2) 0.06131 3.165*** 0.03473 1.453*
TTINV(-3) 0.05224 3.001*** .02820 1.313*
TTINV(-4) 0.03553 1.281 0.01380 0.403
TT SUM 0.2683 5.096*** 0.1343 2.067**
GDP 0.1492 3.200*** 0.009204 0.1599
BOND -0.009570 -2.600*** -0.01318 -2.902***
TRADE 0.5544 8.310*** 0.6775 8.231***
D20 0.7248 7.934*** 0.7348 6.519***
D21 1.967 18.44*** 1.939 14.74***
D22 0.6481 8.454*** 0.5891 6.227***
D23 0.07540 1.127 0.08673 1.051
D24 -0.1478 -2.047** -0.1982 -2.224**
D25 0.2749 3.190*** 0.2580 2.426***
D26 0.8301 9.888*** 0.9123 8.807***
D27 -0.09009 -1.195** 0.1627 1.750**
D28 0.7599 7.587*** .9980 8.076***
D29 1.593 20.75*** 1.102 11.64***
D30 0.3695 4.866*** 0.4593 4.902***
D31 0.6330 3.158*** 0.4147 1.677**
D32 0.2838 3.903*** 0.3743 4.171***
D33 0.9182 10.81*** .9320 8.897***
D34 0.1600 1.963** 0.3311 3.292***
D35 0.2257 2.381** 0.4852 4.147***
D36 0.6441 7.013*** .9150 8.075***
D37 1.078 10.88*** 1.222 9.991***
D38 0.6423 8.336*** 0.8713 9.164***
R - Squared0.9956 0.9874
Durbin-Watson 1.464 1.342
F Statistic 2040 706.0
Number of Obs. 280 280 280 280

TABLE C1 Dependent Variable is by Time Period for SIZE1
1976-84 1976-80 &1985-89 1981-1989
VariableCoefficientT StatisticCoefficientT StatisticCoefficientT Statistic
C -4.577 -2.940*** -2.076 -1.767** -5.359 -3.399***
ITNV(0) -0.01275 -0.301 -0.08407 -2.274*** 0.01570 1.124
ITINV(-1) -0.02970 -2.528*** -0.03780 -3.930*** 0.00561 0.107
ITINV(-2) -0.03690 -2.866*** -0.00626 -0.469 -0.08581 -4.450***
ITINV(-3) -0.03435 -1.708** 0.01055 0.537 -0.1172 -2.990***
ITINV(-4) -0.02205 -1.376* 0.01264 0.822 -0.08862 -2.556***
IT SUM -0.1357 -4.649*** -0.1049 -3.760*** -0.1289 -0.982
TTINV(0) 0.03831 0.580 0.2140 2.949*** -0.1310 -0.543
TTINV(-1) 0.02940 0.939 0.1166 4.647*** 0.05011 0.564
TTINV(-2) 0.02111 0.629 0.04656 2.244** 0.1537 3.875***
TTINV(-3) 0.01345 0.351 0.00374 0.122 0.1799 2.429***
TTINV(-4) 0.00641 0.229 -0.01178 -0.479 0.1287 2.013**
TT SUM 0.1086 0.913 0.3692 5.221*** 0.3815 1.720**
GDP 0.4683 2.517*** 0.1564 0.7557 0.3392 1.657*
BOND 0.006395 1.014 0.01153 1.579* -0.003052 -0.2688
TRADE 0.3986 3.368*** 0.1058 0.9844 0.4240 2.524***
DU 0.8305 29.90*** 0.07424 38.64*** 0.7270 33.97***
ND 0.8190 17.27*** 0.7151 13.47*** 0.7269 9.151***
TU 0.3401 4.942*** 0.1881 3.362*** 0.1761 1.454*
TR -0.03902 -2.168** 0.04409 -2.544*** -0.06197 -2.353**
FI 0.01492 0.3924 -0.1001 -3.341*** -0.1015 -2.022**
R - Squared 0.9884 0.9924 .9920
Durbin-Watson Statistic 2.719 2.190 1.756
F Statistic 293.5 512.6 425.8
Number of Observations 54 54 60 60 54 54

TABLE C2 Dependent Variable is by Time Period for SIZE2
1976-84 1976-80 &1985-89 1981-1989
VariableCoefficientT StatisticCoefficientT StatisticCoefficientT Statistic
C -4.812 -1.257 -0.4331 -0.1520 -6.304 -1.950**
ITNV(0) -0.07456 -0.716 -0.2049 -2.287** -0.2570 -0.898
ITINV(-1) -0.05113 -1.770** -0.05842 -2.506*** -0.1310 -1.213
ITINV(-2) -0.03196 -1.010 0.03533 1.093 -0.04225 -1.069
ITINV(-3) -0.01705 -0.345 0.07632 1.602* 0.00916 0.114
ITINV(-4) -0.00639 -0.162 0.06454 1.732** 0.02325 0.327
IT SUM -0.1080 -2.523*** -0.08716 -1.288 -0.3979 -1.478*
TTINV(0) 0.07521 0.463 0.2913 1.656* 0.3842 0.777
TTINV(-1) 0.05761 0.748 0.07673 1.261 0.1829 1.004
TTINV(-2) 0.04128 0.500 -0.05969 -1.187 0.0438 0.538
TTINV(-3) 0.02624 0.279 -0.1179 -1.583* -0.03306 -0.218
TTINV(-4) 0.01248 0.181 -0.09805 -1.646* -0.04766 -0.364
TT SUM 0.2128 0.727 0.09235 0.539 0.5302 1.166*
GDP 0.3562 0.7790 0.04243 0.08455 0.3221 0.7677
BOND -0.001350 -0.08717 0.005526 0.3123 -0.01511 -0.6490
TRADE 0.4582 1.574* 0.1629 0.6255 0.6176 1.792**
DU 0.5448 7.981*** 0.4079 8.760*** 0.4283 9.762***
ND 0.4383 3.761*** 0.3187 2.478*** 0.2902 1.7824**
TU 0.04644 0.2746 0.05392 0.4129 -0.1149 -0.4627
TR 0.4098 9.264*** 0.3987 9.491*** 0.3692 6.841***
FI -0.1095 -1.172** -0.2957 -4.070*** -0.2983 -2.897***
R - Squared 0.9470 0.9641 0.9773
Durbin-Watson Statistic 3.120 2.209 1.466
F Statistic 61.26 105.7 146.7
Number of Observations 54 54 60 60 54 54

TABLE D1 Dependent Variable is SIZE3
1976-84 1976-80 &1985-89 1981-1989
VariableCoefficientT StatisticCoefficientT StatisticCoefficientT Statistic
C -4.299 -7.684*** -3.085 -4.194*** -6.252 -7.748***
ITNV(0) -0.02430 -0.599 -0.06862 -1.830** -0.08125 -1.923**
ITINV(-1) -0.04105 -3.614*** -0.02819 -5.013*** -0.08359 -4.685***
ITINV(-2) -0.04258 -3.087*** -0.00328 -0.137 -0.07083 -4.411***
ITINV(-3) -0.02890 -2.029** 0.00612 0.272 -0.04297 -3.047***
IT SUM -0.1368 -3.614*** -0.09396 -5.013*** -0.2786 -4.685***
TTINV(0) -0.00078 -0.017 .1650 4.449*** 0.1790 2.606***
TTINV(-1) 0.03863 2.338** 0.07562 4.891*** 0.1256 4.190***
TTINV(-2) 0.05190 2.184** 0.01828 0.792 0.07801 3.390***
TTINV(-3) 0.03903 1.815** -0.00692 -0.344 0.03612 1.823**
TT SUM 0.1287 2.338** .2520 4.891*** 0.4188 4.190***
GDP 0.4025 5.784*** 0.09170 1.913** 0.1901 3.138***
BOND -0.008448 -1.173** -0.007236 -0.8827 -0.01619 -2.334**
TRADE 0.5880 6.795*** 0.3937 3.733*** 0.8201 7.517***
D20 0.6692 6.215*** 0.7392 8.029*** 0.6765 5.161***
D21 1.933 18.08*** 1.922 16.50*** 2.079 14.65***
D22 0.6768 7.862*** 0.6457 8.254*** 0.5949 5.887***
D23 0.0527 0.6783 0.09028 1.305* -0.04364 -0.4707
D24 -0.1080 -1.278 -0.1171 -1.536* -0.2519 -2.512***
D25 0.2928 2.925** 0.2468 2.767*** 0.3167 3.163***
D26 0.8291 8.553*** 0.8213 9.734*** 0.8251 7.164***
D27 -0.1553 -1.720** -0.1119 -1.473* -0.02285 -0.2260
D28 0.7358 6.319*** -0.7827 7.738*** 0.6626 4.423***
D29 1.747 17.97*** 1.578 21.23*** 1.552 15.53***
D30 0.3986 4.587*** 0.3614 4.684*** 0.3255 3.201***
D31 .6630 3.342*** 0.5936 2.768*** 0.7323 2.681***
D32 0.2889 3.112*** 0.2555 3.490*** 0.4091 4.483***
D33 0.8941 8.701*** 0.9642 11.34*** 0.8538 7.697***
D34 0.1124 1.158* 0.1726 2.082** 0.1345 1.229
D35 0.1589 1.297* .2200 2.338** 0.3103 2.328**
D36 0.6232 5.518*** 0.6294 6.980*** 0.6589 5.118***
D37 1.005 8.532*** 1.087 10.90*** 1.059 7.537***
D38 0.6117 6.529*** 0.6278 8.165*** 0.7218 7.167***
R - Squared 0.9961 0.9968 0.9965
Durbin-Watson Statistic 1.756 1.352 1.917
F Statistic 1524 2108 1692
Number of Observations 180 180 200 200 180 180

TABLE D2 Dependent Variable is SIZE4
1976-84 1976-80 &1985-89 1981-1989
VariableCoefficientT StatisticCoefficientT StatisticCoefficientT Statistic
C -4.870 -6.465*** -3.865 -4.287*** -6.502 -6.954***
ITNV(0) -0.01716 -0.314 -0.0836 -1.820** -0.06880 -1.405*
ITINV(-1) -0.03598 -2.353** -0.02411 -3.499*** -0.07438 -3.598***
ITINV(-2) -0.03940 -2.121** 0.00968 0.331 -0.06477 -3.482***
ITINV(-3) -0.02740 -1.429* 0.01772 0.643 -0.03998 -2.447***
IT SUM -0.1199 -2.353** -0.08036 -3.499*** -0.2479 -3.598***
TTINV(0) -0.00953 -0.155 0.1529 3.364*** .1160 1.458*
TTINV(-1) 0.03017 1.356* 0.05852 3.088*** 0.05400 1.554*
TTINV(-2) 0.04499 1.406* 0.00153 0.054 0.01396 0.523
TTINV(-3) 0.03493 1.207 -0.01797 -0.728 -0.00404 -0.176
TT SUM 0.1005 1.356* .1950 3.088*** 0.1799 1.554*
GDP 0.2745 2.930*** -0.02960 -0.5041 -0.01629 -0.2322
BOND -0.01502 -1.550* -0.01377 -1.371* -0.01710 -2.127**
TRADE 0.6683 5.735*** 0.5194 4.019*** 0.9636 7.623***
D20 0.5183 3.575*** 0.6532 5.790*** 0.8798 5.7934***
D21 1.909 13.26*** 1.883 13.19*** 2.117 12.87***
D22 0.5673 4.894*** 0.5457 5.692*** 0.5916 5.053***
D23 0.06183 0.5905 0.09413 1.110 -0.008930 -0.0831
D24 -0.2233 -1.962** -0.1991 -2.131** -0.2415 -2.079**
D25 0.2785 2.066** 0.2282 2.088** 0.3072 2.648**
D26 0.7818 5.990*** 0.8257 7.986*** 1.050 7.868***
D27 0.01255 0.1032 0.07947 0.8532 0.3509 2.996***
D28 0.7963 5.079*** .9130 7.367*** 1.102 6.350***
D29 1.047 8.001*** 1.086 11.93*** 1.165 10.06***
D30 0.4047 3.459*** 0.3912 4.137*** .5240 4.447***
D31 0.5413 2.026** 0.4355 1.657* 0.3723 1.176
D32 0.3161 2.530*** 0.3073 3.426*** 0.5428 5.133***
D33 0.7983 5.770*** 0.8958 8.601*** 0.9745 7.583***
D34 .1890 1.446* 0.2664 2.622*** 0.4319 3.405***
D35 0.2989 1.812** 0.3842 3.332*** 0.7384 4.782***
D36 0.7785 5.119*** 0.7951 7.196*** 1.107 7.426***
D37 1.000 6.308*** 1.120 9.171*** 1.401 8.609***
D38 0.7615 6.036*** 0.7895 8.378*** 1.063 9.116***
R - Squared 0.9868 0.9911 0.9912
Durbin-Watson Statistic 1.561 1.088 1.875
F Statistic 411.5 742.2 665.7
Number of Observations 180 180 200 200 180 180

TABLE E1 Dependent Variable is SIZE1
T StatisticServices
T Statistic
C -0.4570 -0.2341 -3.949 -3.127***
ITNV(0) -0.03034 -0.477 0.04049 1.006
ITINV(-1)-0.04035 -3.232*** -0.01018 -0.888
ITINV(-2)-0.04231 -1.699** -0.03956 -3.022***
ITINV(-3) -0.03625 -0.952 -0.04766 -2.415***
ITINV(-4) -0.02214 -0.743 -0.03447 -2.211**
IT SUM -0.1713 -6.128*** -0.09137 -2.929***
TTINV(0) 0.07461 0.785 -0.00709 -0.086
TTINV(-1) -0.01486 -0.227 0.0218 0.681
TTINV(-2) -0.6706 -0.836 0.03695 1.839**
TTINV(-3) -0.08198 -0.993 0.03837 1.303*
TTINV(-4) -0.05963 -1.048 0.02605 1.082
TT SUM -0.1489 -0.526 0.116O 1.283
GDP 0.08427 0.05065 0.393O 1.596*
BOND 0.007045 0.8698 -0.001389 -0.3336
TRADE 0.4897 4.024*** 0.3543 4.444***
DU 0.1182 2.841***
TU0.3215 4.911***
TR-0.03952 -2.437***
FI-0.006689 -0.1942
R - Squared 0.9979 0.9864
Durbin-Watson Statistic 2.712 2.46
F Statistic 1138 328.4
Number of Observations 28 28 56 56

TABLE E2 Dependent Variable is SIZE2
T StatisticServices
T Statistic
C 2.576 1.129* -7.258 -2.303**
ITNV(0) 0.00792 0.100 0.00624 0.062
ITINV(-1) -0.02795 -1.917** -0.0253 -0.885
ITINV(-2) -0.04667 -1.604* -0.04170 -1.277
ITINV(-3) -0.04826 -1.086 -0.04295 -0.873
ITINV(-4) -0.03270 -0.939 -0.02905 -0.747
IT SUM -0.1476 -4.520*** -0.1327 -1.706**
TTINV(0) 0.1454 1.310* -0.1499 -0.726
TTINV(-1) -0.00489 -0.064 -0.1059 -1.328*
TTINV(-2) -0.09458 -1.009 -0.06896 -1.376*
TTINV(-3) -0.1236 -1.282 -0.03898 -0.531
TTINV(-4) -0.09214 -1.387* -0.01599 -0.266
TT SUM -0.1698 -0.513 -0.3798 -1.683**
GDP 0.03315 0.1706 1.230 2.002**
BOND -0.000003392 -0.000359 0.003728 0.3590
TRADE 0.001297 0.009125 0.40832.053*
DU 0.1486 3.059***
FI-0.03163 -0.3682
R - Squared0.99460.9674
Durbin-Watson Statistic2.3642.726
F Statistic438.6133.6
Number of Observations 28 28 56 56

Appendix: Regressions on each of the measures of firm size with IT redefined as the sum of "Office, computing and accounting machinery" and "Communications Equipment". The new measure of IT is labeled "IT2" in Tables F & G, below.


Dependent Variable is SIZE1Dependent Variable is SIZE2
VariableCoefficientT StatisticCoefficientT Statistic
C -1.897 -2.663*** -1.698 -1.149
IT2(0) -0.05923 -1.782** -0.1427 -2.069**
IT2(-1) -0.03394 -3.215*** -0.05938 -2.712***
IT2(-2) -0.01536 -1.911** -0.00345 -0.207
IT2(-3) -0.00352 -0.263 0.02510 0.903
IT2(-4) 0.0016 0.146 0.02624 1.155*
IT2 SUM -0.1104 -4.343*** -0.1541 -2.923***
TTINV(0) 0.2542 3.865*** 0.2682 1.966**
TTINV(-1) 0.1221 4.871*** 0.08340 1.603*
TTINV(-2) 0.03075 1.586* -0.03585 -0.892
TTINV(-3) -0.02009 -0.748 -0.08951 -1.606*
TTINV(-4) -0.03034 -1.422* -0.07756 -1.753**
TT SUM 0.3567 4.755*** 0.1487 0.956
GDP -0.07264 -0.4772 -0.04505 -0.1426
BOND 0.004186 0.9809 0.005064 0.5721
TRADE 0.3282 3.751*** 0.4335 2.388**
DU 0.7275 42.48*** 0.4305 12.12***
ND 0.6503 15.51*** 0.3042 3.499***
TU 0.2721 5.153*** 0.1774 1.619*
TR -0.008021 0.4957 0.4653 13.86***
FI -0.1013 -3.958*** -0.2310 -4.351***
R - Squared 0.9884 0.9613
Durbin-Watson Statistic 1.995 1.940
F Statistic 508.7 146.9
Number of Observations 84 84

Dependent Variable SIZE3Dependent Variable is SIZE4
VariableCoefficientT StatisticCoefficientT Statistic
C-3.486 -8.466*** -4.060 -8.255***
IT2(0) -0.09613 -3.290*** -0.09912 -2.840***
IT2(-1) -0.04768 -5.424*** -0.05127 -4.883***
IT2(-2) -0.01059 -0.645 -0.01112 -0.567
IT2(-3) 0.01515 1.150 0.02131 1.355*
IT2(-4) 0.02953 1.723** 0.04604 2.262**
IT2 SUM -0.1097 -8.343*** -0.09416 -5.996***
TTINV(0) 0.1362 2.882*** 0.1165 2.063**
TTINV(-1) 0.06406 4.172*** 0.05019 2.737***
TTINV(-2) 0.01508 0.703 0.00103 0.040
TTINV(-3) -0.01070 -0.593 -0.03096 -1.438*
TTINV(-4) -0.01326 -0.507 -0.04581 -1.468*
TT SUM 0.1914 3.598*** 0.09094 1.431*
GDP 0.09238 1.971** -0.04250 -0.7593
BOND 0.003231 0.8953 -0.001731 -0.4017
TRADE 0.4814 7.871*** 0.5994 8.204***
D20 0.8471 9.131*** 0.8157 7.363***
D21 1.983 18.17*** 1.954 14.99***
D22 0.7608 9.707*** 0.6759 7.221***
D23 0.09531 1.395* 0.09979 1.223
D24 -0.02070 -0.2817 -0.09591 -1.093
D25 0.2692 3.047*** 0.2524 2.393***
D26 0.9367 10.90*** 0.9817 9.567***
D27 -0.03158 -0.4075 -0.1974 2.133**
D28 0.9711 9.570*** 1.152 9.511***
D29 1.686 21.54*** 1.166 12.48***
D30 0.4497 5.768*** 0.5112 5.491***
D31 0.5410 2.639*** 0.3472 1.418*
D32 0.3257 4.342*** 0.4000 4.464***
D33 1.017 11.81*** 0.9962 9.683***
D34 0.2685 3.245*** 0.4074 4.124***
D35 0.3239 3.320*** 0.5473 4.697***
D36 0.7869 8.282*** 1.008 8.890***
D37 1.212 12.02*** 1.310 10.88***
D38 0.7429 9.258*** 0.9418 9.828***
R - Squared0.99540.9876
Durbin-Watson 1.4821.446
F Statistic 1942717.6
Number of Obs. 280 280


Attewell, P. and Rule, J. Computing and Organizations: What We Know and What We Don't Know. Communications of the ACM, Vol. 27, Dec. (1984), pp. 1184-1192.

Antonelli, C. (Ed.),New Information Technology and Industrial Change: The Italian Case, A New Industrial Organization Approach, Dordrecht: Kluwer Academic Publishers, 1988.

Arrow, K.J. Information and Economic Behavior. Svanback & Nymans, Stockholm, 1973.

Berndt, E. The Practice of Econometrics: classic and contemporary. Addison-Wesley, Reading, MA, 1991.

Berndt, E.R. and Morrison, C.J. High-Tech Capital, Economic and Labor Composition in U.S. Manufacturing Industries: an Exploratory Analysis . National Bureau of Economic Research Manuscript , (April 24, 1991).

Brynjolfsson, E. An Incomplete Contracts Theory of Information, Technology, and Organization . MIT Sloan School of Management CCS TR #126, (December, 1991b).

Brynjolfsson, E. Information Technology and the 'New Managerial Work' . MIT Working Paper, (1990a).

Brynjolfsson, E. The Productivity Paradox of Information Technology: Review and Assessment. Communications of the ACM, (1993, in press), .

Brynjolfsson, E., Malone, T.W. and Gurbaxani, V. The Impact of Information Technology on Markets and Hierarchies . MIT Sloan School of Management Working paper #2113-88, (1988).

Bureau of Economic Analysis, U.S.D.o.C. Fixed Reproducible Tangible Wealth in the United States, 1925-85. U.S. Government Printing Office, Washington, D.C., 1987.

Bush, G. (Ed.),Economic Report of the President, Washington: United States Government Printing Office, 1991.

Carlsson, B. The Evolution of Manufacturing Technology and its Impact on Industrial Structure: An International Study . The Industrial Institute for Economic and Social Research, Stockholm, Sweden, (1988).

Cartwright, D.W. Improved Deflation of Purchases and Computers. Survey of Current Business, Vol. 66, March (1986), pp. 7-9.

Caves, R. and Bradburd, R. The Empirical Determinants of Vertical Integration. J Econ Behavior and Organization, Vol. 9, (1988), pp. 265-279.

Chow, G.C. Technological Change and the Demand for Computers. American Economic Review, Vol. 57, December (1967), pp. 1117-1130.

Comment, R. and Jarrell, G.A. Corporate Focus and Stock Returns . Simon Graduate School, University of Rochester Working Paper #MR 91-01, (May, 1991).

Crowston, K. and Malone, T.W. Information Technology and Work Organization. in Handbook of Human-Computer Interaction, North Holland, 1988.

Drucker, P.F. The Coming of the New Organization. Harvard Business Review, January-February (1988), pp. 45-53.

Eisenhardt, K. Agency Theory: An Assessment and Review. Academy of Management Review, (1989), pg. 57-74.

Galbraith, J. Organizational Design. Addison-Wesley, Reading, MA, 1977.

Gordon, R.J. The Postwar Evolution of Computer Prices . National Bureau of Economic Research, Cambridge, MA Working Paper #2227, (1987).

Gorman, J.A., Silverstein, M., Gerald, J.C., et al. Fixed Private Capital in the United States. Survey of Current Business, July (1985).

Grossman, S. and Hart, O. The Costs and Benefits of Ownership: A Theory of Vertical and Lateral Integration. Journal of Political Economy, No. 4, (1986).

Grove, A.S. The Future of the Computer Industry. California Management Review, Vol. 33, No. 1 (1990), pp. 148-160.

Gurbaxani, V. and Mendelson, H. An Integrative Model of Information Systems Spending Growth. Information Systems Research, Vol. 1, No.1, March (1990), pp. 23-46.

Gurbaxani, V. and Whang, S. The impact of information systems on organizations and markets. Communications of the ACM, Vol. 34, No. 1 (1991), pp. 59-73.

Hayek, F.A. The Use of Knowledge in Society. American Economic Review, Vol. 35, No. 4 (1945) .

Hildreth, G. and Lu, J.Y. Demand Relations with Autocorrelated Disturbances . Michigan State University Agricultural Expedition Station Technical Bulletin #276, (November, 1960).

Holmstrom, B.R. and Tirole, J. The Theory of the Firm. in Handbook of Industrial Organization, R. Schmalansee & R. Willig. Elsevier Science Pub. Co., Amsterdam, 1989.

Huber, G.P. A theory of the effects of advanced information technologies on organizational design, intelligence and decision making. Academy of Mangement Review, Vol. 15, No. 1 (1990), pp. 47-71.

Huppes, T. The Western EdgeWork and Management in the Information Age. Kluwer Academic Publishers, Dordrecht, the Netherlands, 1987.

Jensen, M.C. Organization Theory and Methodology. The Accounting Review, April (1983), pp. 319-339. (Sections I, II & VII).

Johnston, R. and Lawrence, P. Beyond Vertical Integration--the Rise of the Value-Adding Partnership. Harvard Business Review, July-August (1988), pp. 94-101.

Kanter, R.M. The New Managerial Work. Harvard Business Review, Nov-Dec (1989), pg. 85-92.

Klein, B., Crawford, R. and Alchian, A. Vertical Integration, Appropriable Rents and the Competitive Contracting Process. Journal of Law and Economics, Vol. 21, October (1978), 297-326.

Leavitt, H.J. and Whisler, T.L. Management in the 1980s. Harvard Business Review, November-December (1958).

Lichtenberg, F.R. Industrial De-Diversification and its Consequences for Productivity . Columbia University , (January, 1990).

Loveman, G.W. An Assessment of the Productivity Impact on Information Technologies . MIT Management in the 1990s Working Paper #88-054, (July, 1988).

Malone, T.W. and Smith Modelling the Performance of Organizational Structures. Operations Research, 36, May-June (1988), pp. 421-436.

Malone, T.W. Modelling Coordination in Organizations and Markets. Management Science, Vol. 33, (1987), pp. 1317-1332.

Malone, T.W., Yates, J. and Benjamin, R.I. Electronic Markets and Electronic Hierarchies. Communications of the ACM, Vol. 30, No. 6 (1987), pp. 484-497.

Morck, R., Shleifer, A. and Vishny, R. Do Managerial Objectives Drive Bad Acquisitions? Journal of Finance, Vol. 45, (1990), pp. 31-48.

Morrison, C.J. and Berndt, E.R. Assessing the Productivity of Information Technology Equipment in the U.S. Manufacturing Industries . National Bureau of Economic Research Working Paper #3582, (January, 1990).

Nakamura, A. and Nakamura, M. On the Performance of Tests by Wu and By Hausman for Detecting the Ordinary Least Squares Bias Problem. Journal of Econometrics, Vol. 29, (1985), pp. 213-227.

Osterman, P. The Impact of Computers on the Employment of Clerks and Managers. Industrial and Labor Relations Review, Vol. 39, (1986), pp. 175-186.

Pakes, A. and Griliches, Z. Estimating Distributed Lags in Short Panels with an Application to the Specification of Depreciation Patterns and Capital Stock Constructs. Review of Economic Studies, Vol. LI (2), No.165 (1984), pp. 243-262.

Pindyck, R.S. and Rubinfeld, D.L. Economic Models and Economic Forecasts. McGraw-Hill, New York, 1991.

Piore, M. and Sabel, C. The Second Industrial Divide. Basic Books, New York, 1984.

Piore, M. Corporate Reform in American Manufacturing and the Challenge to Economic Theory . MIT, Sloan School of Management Management in the 1990s Program , (1989).

Piore, M. The Changing Role of Small Business in the U.S. Economy . Institute of Labour Studies of the International Labour Organization New Industrial Organization Project, (December, 1986).

Roach, S.S. America's Technology Dilemma: A Profile of the Information Economy . Morgan Stanley Special Economic Study, (April, 1987a).

Solow, R. (1990). The Productivity Paradox, Personal Communication, March.

Toffler, A. The Third Wave. Pan Books, Ltd., London, 1982.

Wannell, T. Trends in the Distribution of Employment by Employer Size: Recent Canadian Evidence . Analytic Studies Branch, Statistics Canada, (1990).

Williamson, O. Markets and Hierarchy: Analysis and Antitrust Implications. Free Press, New York, 1975.

Williamson, O. The Economic Institutions of Capitalism. Free Press, New York, 1985.


[1] We will use the term "IT" as shorthand for the Bureau of Economic Analysis (BEA) category "Office, Computing and Accounting Machinery", which is comprised primarily of computers. Other authors have used slightly different definitions but the basic trends appear to be similar regardless of the exact definition used.

[2] Other interpretations of the term "firm size" -- for instance, in terms of sales, assets or market capitalization -- are less common in the empirical literature. As we show below, a decline in employees per firm does not necessarily imply a decline in these other characteristics.

[3] While this is a broader set of data than has been used before, it would arguably have been preferable to examine data on 100% of the US economy. We were unable to find any comparable data for the government or the "underground" economy and only relatively incomplete data for resource industries like agriculture. In any event, the theory of, and even the definition of, "the firm" does not readily translate to these sectors.

[4] Although often used synonymously, establishments are not, strictly speaking, equivalent to firms. Nonetheless, the vast majority of firms consist of a single establishment. 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 97%.

[5] We we also tested a model which included the communications equipment category with substantially identical results. See appendix A.

[6] See also Berndt and Morrison (1991) for additional detail and interpretation. We thank John Musgrave for useful conversations on the data gathering methodology.

[7] The deflator for IT was actually an "inflator"; a dollar bought more IT each year than it did the year before.

[8] We thank an anonymous referee for this suggestion, which simplifies the interpretation of our previous, more complex, specification. Two notes are in order. 1) This approach assumes that lag structure is constant over time. 2) A second degree polynomial allows for only one inflection point and so it cannot duplicate an S-shaped pattern of changes in effectiveness. If the true pattern were S-shaped, then the PDL might be either convex or concave, depending on which curve of the "S" is more closely fitted.

[9] For instance, a recent study of Canadian data concluded that "the increasing importance of service sector employment played a role in the growth of small firm jobs, but was generally less important than shifts in the size distribution within the major industrial sectors." (Wannell, 1990)

[10] Presumably, not all of the "new" firms created when a vertically integrated firm decouples will be classified in the same industry as their "parent". Consider for instance the outsourcing of design work from a manufacturer. However, the remaining manufacturing firm would be smaller than before, so the outsourcing would be reflected in average firm size within the sector as well.

[12] In any event, the OLS coefficients were very similar. On the recommendation of a referee, we report the IV estimates because statistical tests to choose between IV and OLS estimates do not always exhibit desirable statistical power properties and hence IV estimates, which are always consistent, are preferred to possibly biased OLS estimates (Nakamura & Nakamura, 1985)

[13] Formally, this is equivalent to the prediction that c1 will be positive and c2 will be negative in equation (2).

[14] An F-test rejected the restriction that the coefficients on IT were equal to zero at the 99.9% level, strongly supporting its explanatory power in the regression.

[15] A review of the residuals from both regressions revealed one possible outlier in the Transport and Utilities sector. Because this outlier could be due to an error in data coding, we re-ran the regression without this data point. The revised results, (table C), do not significantly affect findings discussed in the text except to further strengthen the negative relationship between IT and firm size.

[17] As cited in Lichtenberg, (1990).