Information Technology, Workplace Organization and the ...tbres/research/itwold8-21f.pdf · workplace organization, and the demand for skilled labor. In both a short-run factor demand

Information Technology, Workplace Organization andthe Demand for Skilled Labor: Firm-level Evidence

Draft: Comments Welcome

Timothy F. BresnahanDepartment of Economics

Stanford, [email protected]

Erik BrynjolfssonMIT Sloan School of Management

Cambridge, MAAnd

Stanford Graduate School of BusinessStanford, CA

[email protected]

Lorin M. HittUniversity of Pennsylvania, Wharton School

Philadelphia, [email protected]

First Draft: January, 1998This Draft: August, 1998

We thank Gary Burtless and participants in the Brookings/MIT Human CapitalConference for valuable comments. This research has been generously supported by theMIT Center for Coordination Science, the MIT Industrial Performance Center, theNational Science Foundation (Grant IIS-9733877) and the Stanford Computer IndustryProject under grants from the Alfred P. Sloan Foundation and NationsBanc MontgomerySecurities. Incon Research, the Center for Survey Research, Computer IntelligenceInfocorp and Informationweek provided or helped to collect essential data.

Information Technology, Workplace Organization and the Demandfor Skilled Labor: Firm-level Evidence

Abstract

Recently, the relative demand for skilled labor has increased dramatically. We

investigate one of the causes, skill-biased technical change. Advances in information

technology (IT) are one of the most powerful forces bearing on the economy. Employers

who use information technology (IT) often make complementary changes in their

organizations and in the services they offer. These co-inventions by IT users change the

mix of skills that employers demand. We examine the specific hypothesis that it is a

cluster of complementary changes involving IT, workplace organization and services that

is the key skill-biased technical change.

Our evidence is based on new firm-level data linking several indicators of IT use,

workplace organization, and the demand for skilled labor. In both a short-run factor

demand framework and a production function framework, we find evidence for

complementarity. IT use is complementary to a new workplace organization which

includes broader job responsibilities for line workers, more decentralized decision-

making, and more self-managing teams. In turn, both IT and that new organization are

complements with worker skill, measured in a variety of ways. Further, the managers in

our survey believe that IT increases skill requirements and autonomy among workers in

their firms. Taken together, the results highlight the roles of both IT and IT-enabled

organizational change as important components of skill-biased technical change.

1 Introduction

1.1 The growing demand for skilled laborThe distribution of wages and earnings has been spreading out in the United

States, leading to increases in income inequality. The wages and earnings of the well off

have been rising much more rapidly than those of low and middle-income workers.1 An

impressive body of empirical studies shows that the proximate cause has been a shift in

the demand for labor.2 Employers’ demand has shifted from low- and middle-wage

occupations and skills toward highly rewarded jobs and tasks, those requiring exceptional

talent, training, autonomy, or management ability. The total effect has been large; as the

gap between wages at the 75th percentile of the distribution and the 25th has increased by

nearly 50 percentage points.3 The total effect has also been widespread, shifting relative

wages in the top, middle, and bottom of the income distribution.4 The shift in labor

demand appears to be approximately a quarter of a century old.5

A shift of this magnitude should be explained carefully. Clearly, parts of the shift

in labor demand are explained by such broader economic patterns as globalization and

sectoral shifts in employment. Yet these forces appear too small to explain the breadth

and depth of the shift.6 A substantial proportion of the labor demand shift must be

explained as skill-biased technical change. In this context, skill-biased technical change

means technical progress that, on net, led employers to increase demand for more highly

1 There are a number of reviews of the growing empirical literature. See Autor, Katz and

Krueger (1997), Bresnahan (1997), and the symposium in the Spring, 1997 issue of the Journal ofEconomic Perspectives, especially Johnson (1997).

2 See Autor, Katz and Krueger (1997) for a summary of this evidence, a bibliography, and aninteresting supply-demand framework. See also Katz and Murphy (1992) and Bound and Johnson (1992)on the supply vs. demand distinction.

3 Murphy and Welch (1993) show percentiles of the wage distribution of prime-age males: theinterquartile range has increased from about 1.75 to 1 to about 2.25 to 1. More extreme percentiles havemoved even farther from the median, so that the entire distribution of wages is widening over time.

4 In countries with very different labor market institutions and employment relationships thanthose found in the US, the same labor demand shifts have affected relative wages less and relativeemployment more. See Gottschalk and Smeeding (1997) for a review of cross-national comparisons.

5 The shift in labor demand precedes the shift in relative wages because of supply shifts. Thebaby boom's graduation from college, for example, was a supply event that transitorily depressed the wagepremium for higher education in the 1970s. See Karoly (1996) and Juhn, Murphy and Pierce (1993).

IT, Organization and Labor Demand Draft: Please do not quote Page 2

skilled workers more rapidly.7 The size, breadth, and timing of the labor demand shift

have led many observers to link it to a largest and most widespread technical change of

the current era, information technology.8 In this paper, we examine the firm-level

evidence for a specific theory of how information technology (IT), together with the

organizational change it encourages, could be skill-biased technical change.

Employers adopt IT-based production processes to improve product quality or

increase efficiencies. In either case, effective use of IT involves changes to organization.

Examination of the form of the organizational changes suggests a theory of why IT-based

technical change is skill-biased (Bresnahan (1997)). Improved service quality is one of

the potential advantages of IT-based production (Barras (1990)). First in service-

producing sectors like finance, then in the service parts of goods-producing industries,

IT-using firms have found ways to take advantage of the new production processes.

They have also found it very difficult to make improvements using only computers and

telecommunications gear. The new production process involves global changes to the

organization.9 These often involve replacing low-skill human work with automation, but

passing on to humans a variety of tasks related to the higher level of service. A similar

pattern holds for attempts to achieve IT-based efficiencies in production. Only with

organizational change, typically of a kind that involves complementarity with high skill

as well as substitutability for low skill human work, do employers get the benefit they

seek from IT.

These observations lead us to an analysis based on a cluster of complementarities

that we see as at the heart of recent changes in labor demand. Intensive use of IT, higher

6 See Krugman and Lawrence (1993) for this view of globalization and trade. While such forces

as changing conditions of domestic and international competition and immigration are obviously relevant,they are too small to provide the entire explanation of changing relative wages.

7 The core idea is shifts to the production side of the economy which shift the demand for moreskilled vs. less skilled workers. This could be capital-skill complementarity, as in Griliches (1969), orcomplementarity between skill and high technology capital, as in Berndt, Morrison, and Rosenblum (1992),or measured as a relationship between technology proxies and skill mix as in Berman, Bound and Griliches(1994).

8 See for example Autor, Katz and Krueger (1997) and the references therein.9 Among the first to predict this effect were Leavitt, and Whisler (1958). There are by now many

papers which illustrate the role of IT-enabled organizational change at varying levels of detail, notablyMilgrom and Roberts (1990), Brynjolfsson, Renshaw and van Alstyne (1997), Bresnahan and Greenstein(1997), and Brynjolfsson and Hitt (1997). It has also been an important theme in the management literature(Davenport and Short (1990) and Hammer, (1990)).


service levels for customers, and organizational change all go together, and together call

for higher-skilled labor. These form a mutually reinforcing cluster of inventions for

employers. Critically, the organizational changes associated with IT-based service

improvements and IT-based efficiencies are skill using. This cluster of complements is,

together, a skill-biased technical change. The "technical" side of this cluster is the

ongoing declines in IT prices and ongoing improvements in IT performance. It is tightly

linked to labor demand through its organizational side.

1.2 Aggregate data suggest IT is behind the labor demand shiftsBroad-based studies of the labor market are consistent with the hypothesis of an

IT-based demand shift.

An important body of studies looks at wage determination at the level of

individual workers or jobs. A wide variety of studies have examined individual worker

wage equations.10 Based on large data sets, such as the Current Population Survey

(CPS), these studies predict wages with both observables – education and experience –

and unobservables – the residual in the wage equation. The observables are interpreted as

proxies for skills. Changes in their coefficients over time are interpreted as changes in the

prices of those skills. When the distribution of the residual spreads out, it is interpreted as

an increase in the price of an unobserved skill. The important results from these studies

are that the relative demand for more highly educated workers is rising (probably related

to general cognitive skill), that the relative demand for more experienced workers is

rising (likely specific knowledge or managerial/people skills) and that the relative

demand for "residual" highly skilled workers (skills not captured by education or

experience) is rising as well.

A related literature classifies occupations by skills and examines the wages and

employment changes for work thus classified.11 It, too, finds considerable support for the

view that relative demand is shifting toward cognitive skills and toward such valuable

worker attributes as autonomy, independence, ability to give and receive direction, etc.

10 Surveyed in Autor, Katz and Krueger (1997) and in Gottschalk (1997). See also Juhn, Murphy

and Pierce (1993), Bound and Johnson (1992) and Murphy and Welch (1993).11 See Howell and Wolff (1991).


The difficulty with both these bodies of empirical inquiry is the same. While they

can reveal the effects of changes in labor demand, they do not examine the demanding

unit. They look at the demanded unit – the worker or the job. Accordingly, their ability

to examine alternative stories of demand is quite limited.

Another body of studies gets closer to the demanding unit by looking at

industries. Here there is also a very clear finding. The IT-intensive industries, have seen

the demand shift earlier (Wolff (1996)) and to a larger extent (Autor, Katz and Krueger

(1997)) than other industries. This finding appears to be robust to how computer-

intensity is measured, as looking at investments in all IT capital and looking at the

fraction of workers in an industry that use a computer or terminal yield comparable

results. The finding is particularly strong in services.

An industry may be simply too wide a unit to capture the relevant flows of

causation. Industries contain variety across firms in IT-use strategies. Industries contain

variety in service and product strategies. Industries contain variety in organizations of

work. Our study, focussing on the firm level behavior, is pointed more closely at the

relevant unit of demand.

2 A Production Function Framework for White-collarBureaucracies

In this section, we take up the implementation of computer business systems for

labor demand in white-collar bureaucracies.12 There are two stages. First, we summarize

our overall view of the complementarities. This revolves around the graphical

presentation in Figure 1. We then discuss the specifics of the relationship between

organizational changed based on IT and changing labor demand patterns.

Computer business systems change white-collar work. They change it by

organizing, routinizing and regularizing tasks that people- and paper-based systems did

more intuitively but more haphazardly. In the service sectors, and in the white-collar

12 Many of the same implications are present in blue-collar contexts (see for instance the case

study of “Macromed” in Brynjolfsson, Renshaw and van Alstyne, 1997), but for brevity we do notemphasize them. A very substantial portion of work in modern economies is in purely white-collarindustries, such as service industries, or in the white-collar functions of goods-producing industries. Wefocus particular attention on the parts of bureaucracies that are associated with transactions. These are


activities of the goods sectors, they also change work by changing the nature of the firm’s

output. Computer-based production leads higher levels of service or even whole new

(service) products. The labor-demand impact comes at the firm level, as computer

business systems form the modern production process for many service industries (and

for the service functions of other industries.) As computers have grown cheaper, and

especially as computer networking has improved, computer-based production has spread

more and more widely through white-collar work.

2.1.1 IT, Organizational Innovations and Improved Products and Services areAll ComplementsAn emerging view of the way IT comes to be used in companies is represented in

the bottom half of Figure 1. This view, which has considerable support in earlier work,13

emphasizes complementarity among three distinct kinds of technical change. The three

complements are:

• Technical change embodied in IT Capital• Organizational Change• New Products, Services or Quality

The theory we will test is that the cluster of three complements at the bottom of

Figure 1 represents, together, a form of skill-biased technical change. Implementing all

three of these forms of technical change together transforms companies and leads to

substantial changes in labor demand by type. We do not make any assumption that one

of these three uniquely causes the other, only that they tend systematically to go together.

Since all three involve substantial amounts of invention, the complementarity arises in

the long run. Taken together, the three forms of technical change are the cluster of driver

technologies for labor demand we study.

associated not only with a great deal of the IT-related technical change of the past, but also, as electroniccommerce grows more important, with the IT-related technical progress of the future.

13 See for example Applegate, Cash and Mills (1988), Attewell and Rule (1984), Barras (1990),Crowston and Malone (1988), Davenport and Short (1990), Levy and Murnane (1996), Malone andRockart (1991), Milgrom and Roberts (1990). Scott Morton (1991), Steiner and Teixera (1991), Zuboff(1988).


Figure 1 Causal Flows in Our Framework

(E)(E’) (E”)

Information Technology Drivers (IT)• Falling Costs of IT Capital• Rising Capabilities of IT Capital e.g.,

new networked platforms• Legacy Applications, Data and

Systems

Organization's Strategic Goals• New Products and Services• New Quality Levels• New Capabilities• New Efficiencies

(C)

Workplace Organization (WO)• Authority and Hierarchy• Task content of operational work• Cognitive content of managerial /

professional work

Demand for Human Capital (HK)• Education Mix• Occupation Mix• Skills Mix• Training Mix• Recruitment and Retention• Wages Offered

(C")(C')


The next step moves us toward specificity by having a simple framework for the

inputs and outputs of white-collar work. We show our framework in a descriptive way in

Figure 2 and in a more analytical way in Figure 3.

Figure 2-- The Production Function Framework

Detailed Inputs Detailed ProductionProcess

DetailedOutputs

Labor of several types Tasks for each input Services

Job responsibilities &computer programs Transactions

Capital goods ofseveral types

=>

Organization of inputs,hierarchy, etc.

=>

Control

The view we take of the production function is one that has a detailed presentation

of inputs, processes, and outputs. Labor and capital of several different types are the

inputs. This gives us the opportunity to represent our fundamental long-run driver,

declines in the price of IT capital goods. It also gives us the opportunity to represent our

fundamental dependent variable, demand for labor of different skill levels. The detailed

production process and detailed outputs play a different role. They permit us to state a

specific theory about the mechanisms by which changes in the price of IT affect the

demand for labor in the intermediate run. They also offer some further dependent

variables tied to the mechanisms of the theory.

The production of detailed outputs is particularly important here, and illustrates

why we take a detailed view of the production process. A white collar bureaucracy's

outputs do not only consist of what the bureaucracy literally does – the services it makes

for customers or the transactions it processes with customers or suppliers. Above and

beyond the simple completion of its task, a bureaucracy produces managerial control of

the task.

This may be illustrated by simple examples. Consider a department with

responsibility for accounts receivable. What does it produce? It produces, in a literal

sense, transactions in which the firm collects money and a record of those collections. If

we think literally about the "service quality" associated with those, it suggests collecting


most of the money owed the firm, not annoying customer/debtors too much, and so on.

A broader view includes the firms' control of its receivables and of the collection process

in the output. Can the firm identify problem accounts for special attention? In allocating

new trade credit associated with new sales, can the firm discriminate between rapid

payers and slow ones? Can it do this unobtrusively and quickly?

Both the narrow and the broad view of output suggest a changed role with

computerization. Better record keeping clearly makes it possible to collect each bill once

and only once. Combined with some changes in policies and procedures, it also makes it

possible for salespeople to know the credit status of an account while in the process of

making a sale.

More generally, these examples reveal two things about the IT-based production

process from a very operational perspective. First, as IT has grown cheaper and more

powerful, the number and kind of white-collar processes it can automate in this way has

grown. More and more interactions in the business world are undertaken with this kind

of computer support. That means more and more substitution out of certain kinds of

human effort. Computers have been especially successful in routine, repetitive tasks,

which are typically those that have not been highly rewarded historically. Second,

computerizing tasks makes it much easier to accumulate data, both intentionally and as a

by-product of other tasks.14 A firm can decide to retain a systematic record of all its

interactions with a customer, or all of a given employee's interactions with customers, for

example. This is a much lower-cost decision given that the interactions are mediated

through IT. As more and more interactions are so mediated, the complexity of record

keeping and record analysis can grow. This offers opportunities for the new uses of

human intelligence.

14 The introduction of bar code scanners to supermarkets saved labor by cashiers. However, this

benefit was vastly outweighed by the benefit of analyzing scanner-derived product data to improveinventory control, supply-chain management, marketing and promotion programs, and pricing strategies.


Figure 3 Short Run Production Function Notation

DetailedInputs

DetailedProductionProcess

Detailed Outputs

L1 . . . . LM

Indexed by T, S XK1 . . . . KM

=> =>

Q

Moving to Figure 3, we show the way we will represent the detailed outputs and

inputs. X measures the quantity of various services or transactions, and Q represents the

service level or degree of control over those transactions that the bureaucracy offers. T

and S index the technologies used in production and the organizational structure of

production, respectively. These are the things usually left in the black box.

The link to factor demand is via a short run production function:

0 = F (L1 . . . . LM , K1 . . . . KM, X; Q, T, S)

That is, the detailed labor and capital requirements are determined by the level of

output, output qualities, and the technology and organizational structure used in the firm.

This is a short-run production function because it conditions on the organizational

variables fixed by the firm in the intermediate run.

Our analysis will focus on the short run and the intermediate run. The short run

has Q, T, and S fixed, and we look at the relationship between the capital (IT) and labor

factors of production. In the intermediate run, Q, T, and S can be improved by the firm,

perhaps because of the complementarities associated with (C'), (C"), (C) and so on. The

relationship between the two runs and Figure 1 is as follows. In the short run, Q, T, and

S are fixed, and only flow of causation (E") is available to change labor demand. In the

intermediate run, organizational structure and service are free, so that flows of causation

(E) and (E') also come into play. We believe that both the short and intermediate run

causal flows can be seen in our data. We are less certain that they can be distinguished.

It is worth noting at this stage that this framework is designed to look at the

individual firm. Further organizational changes associated with changes in the boundary


of the firm will not be captured in firm data, nor will they enter our framework. We

conjecture that we are, as a result, leaving out one of the important flows of causation

from IT-based organizational change to labor demand. We will miss, both theoretically

and empirically, the creation of new specialty firms employing highly paid workers, for

example.

3 Implications of the Theory for Labor Demand

3.1 Limited substitutionComputer business systems involve the regularization and routinization of white-

collar tasks. Simple, repetitive tasks are far more amenable to regularization and

routinization than more complex and idiosyncratic ones. The result has been the

systematic substitution of computer decisionmaking for human decisionmaking in

clerical (and similar bureaucratic) work. Decisions that were once reached by humans in

a paper- and people-based system are now reached by software. Advances in artificial

intelligence notwithstanding, the scope of this substitution has been limited. Simple

decisions, closely related to individual transactions or other operational actions, have

been most amenable to computerization. More complex and cognitively demanding

work, such as that of managers and professionals, has proved remarkably difficult to

automate with computers. Similarly, tasks that require judgment, creativity and frequent

exceptions are much more difficult to computerize than well-defined and repetitive

tasks.15 Computer automation of white-collar work has been correspondingly limited in

its scope, affecting demand for clerks and expediters far more than for managers and

professionals.

This limited substitution is related to all three of the causal arrows (E), (E'), (E")

in Figure 1. First, but probably least important empirically, is the narrow flow of

causation from the use of IT capital alone to substitution out of low-cognitive skill labor;

this is (E"). In the short run, holding Q and S constant, it is sometimes possible to

literally replace people with IT, holding other aspects of the organizational system

15 Traditionally one of the most important steps in software development has been the creation of

an extremely detailed system specification, outlining precisely how the computer system should respond inall possible contingencies.


constant. A telephone-switching computer replaces a telephone operator in a variety of

tasks, an automatic teller machine replaces a bank teller, and documents once organized

and filed by clerks are instead handled entirely by machine. Yet these examples – and

even they are imperfect, involving considerable change in Q, and S as well as T – are

rare.

Far more important is substitution of computer for human decisionmaking in the

intermediate run, with corresponding changes in T, Q, and S. This reflects that fact that

job design and the allocation of tasks to workers combine tasks which use similar

information sets, or which it is efficient to monitor together, which have similar skill

requirements, or which should be performed geographically close together. (Holmstrom

and Milgrom (1991), Jensen and Meckling (1992), Williamson (1979)). Yet, computers

have very different strengths and weaknesses than humans. As a rule, computerization

significantly changes at least one, and more often all, of these factors influencing job

design. As a result, successful computerization also calls for a rethinking of how work is

organized and the types of workers required.

This can perhaps be most simply seen in a homey example from Accounts

Receivable (AR). Almost everyone has heard of how early applications lacked the

commonsense of human clerks, for example sending bills for $0.00. A computer needs to

be told not to send a bill for $0.00, whereas (many) clerks could be relied upon to do this

without explicit instruction. After a while, the software system in an AR application

improves and the system stops making these silly mistakes. Or the company adopts an

organizational change, such as a human clerk who can override the system when it is

obviously doing something silly. Improvements in T – better software systems – made

the computer a better substitute for a clerk. Improvements in S – such as using the fact

that the clerk's actions are now well recorded to permit autonomy, or using the computer's

memory to permit any of a wide variety of clerks to deal with a particular account in a

teamwork fashion – make the clerk plus the computer an improved production process.

These improved rules draw on the very distinct capabilities of IT and humans. They can

only effect labor demand with some invention by the firm. The invention is as much of

improved work organization as it is of improved software.


This tendency for improvements in T and S to offer opportunities to substitute out

of L into K is quite general. There are some common-sense rules in any bureaucracy.

Where common sense can be reduced to rules, or where the business can get by with a

little less common sense, or where common sense can be reorganized into a human-

override, IT systems can be made to substitute for simple human decisionmaking. It is

especially true where S changes.

There are, however, strict limits to this substitution, for all its power.

3.2 Increased demand for cognitive skillsThe IT-related cluster of innovations is not merely a substitute for low cognitive

skill labor. It is also a complement for high cognitive skill labor. These two tendencies

together make the skill-biased technical change. In this subsection, we look at the more

complex causal relationship between the IT-related cluster and high cognitive skill

workers such as managers and professionals.

3.2.1 Direct Causal FlowsThere is an important causal flow directly from IT capital to use of very skilled

technical workers of an (E") form. Computer departments themselves employ highly

skilled people. So, to, does the computer services industry. (This industry is an

extension of company computer departments. It offers custom programming, consulting,

and systems integration to user companies.) Finally, there is a related effect on demand

in the computer hardware, software, and networking industries themselves. In the US,

though not to any very large extent in other countries, this represents a substantial

demand for cognitive skill. Aggregating all these flows will not, however, amount to a

very large fraction of employment.16 We need to look elsewhere for a substantial

complementarity between the IT-based cluster and high cognitive skill.

Another direct causal flow of the (E") variety has been widely discussed but is

likely to be of limited importance. This is the direct use of personal computers (PCs) by

individual workers in organizations. By far the largest single use of PCs is for word

16 Bresnahan (1997) has a more quantitative version of the discussion in this paragraph.


processing.17 Anecdotes about excessive font-futzing and lost data notwithstanding,

typists and secretaries have clearly become much more productive as a result. However,

this is an implausible source of wage dispersion. Meanwhile, managers and professionals

who do their own typing may also become somewhat more productive with PCs, but this

is also an unlikely explanation for their higher relative wages. Similar arguments apply

to the other dominant PC applications, spreadsheets and graphics. However, a minority

of more specialized applications, such as PC-based computer aided design and project

management, are more promising candidates for skill complementarity, as are the

growing organizational uses of PCs which can affect the wages of even those managers

who do not literally use computers.18

3.2.2 Causal Flows Through Organizational ChangeMany effective uses of computers make use of IT's comparative advantage in

doing simple tasks quickly, and in remembering a large volume of simple information

and recalling it rapidly. Many times, these features of computers are most effective when

combined with human intelligence. For example, a typical "analytical" application

consists almost entirely of complex data storage and recall. Rapid, detailed memory by

computers is, it turns out, often a complement, not a substitute, for human judgement in

decisionmaking. In our accounts-receivable example, the computer may quickly report a

summary of a particular customer's payment history. That is an "analytical" application.

Human judgement might have the final say in whether a particular account's value to the

firm warrants extending trade credit in particular circumstances.

A related kind of effective use of the computer's detail and completeness turns the

"analytical" lens on workers and departments within the firm. Detailed reports on the

success – measured by anything the computer system knows systematically – of

individual workers or departments are a kind of management tool that has grown cheaper

with the falling costs of computerization. In Figure 1, this kind of increased demand for

17 The statements about the relative importance of various PC applications are based on data from

Computer Intelligence Infocorp.18 See Bresnahan (1997) for more detail on the likely role of personal computers in driving labor

demand.


highly skilled labor shows up as an (E) flow, that is, one that involves changes both in

computerization and in organization.

This complementarity between machine and human capabilities leads to a wide

variety of organizational structures, but some specific ones are worth mentioning here.

One is the information-enabled decentralized worker. In the most extreme form, this

involves use of a worker who is supported by analytical applications that let her know

everything she needs to know to accomplish her task. She is in turn monitored by a

computer analytical application that permits her superiors to monitor her performance

quickly and efficiently. Less extreme examples vary in a number of ways. They might

have the human unit that is supported and monitored be a team rather than an individual

worker. They might use only one of the support/monitoring flows of information.

Note that this tendency toward computer-based reorganization calls for several

very different changes in worker capabilities. The first is cognitive. Individual workers

or teams supported by analytical applications means must be able to interpret the

information that is presented. Often, this involves bridging from the quantitative data

provided by computers to complex decisions involving many factors. In a wide variety

of applications, computers have made information vastly cheaper and more abundant.

This increases the demand for humans who can process information processing in ways

that machines cannot. Indeed, it creates an information-overload bottleneck. Simon

(1973) noted this not long after the first widespread use of computers in business that

"The scarce resource is not information, it is the processing capacity to attend to

information. Attention is the chief bottleneck in organizational activity, and the

bottleneck becomes narrower and narrower as we move to the tops of organizations."

Thus, in addition to a general increase demand for human cognitive skills,

computerization can lead to attempts to bypass the managerial bottleneck via increased

lateral communication and coordination.

A second set of human capabilities associated with this kind of organizational

change has to do with autonomy more than with any specific cognitive skill. Monitoring

technologies in general are useful for moving responsibility and authority. The

monitoring technology associated with computer-based work appears to be systematically


related to incentives, at least long run incentives, which are more closely tied to

measurable performance. Not all workers come with equal tolerance for or capability in

these kinds of incentives. The change in the mode of supervision calls for changed talents

A third set of human capabilities associated with this kind of organizational

change involves the skills associated with dealing with customers and suppliers,

influencing teammates and colleagues and inspiring and coaching subordinates. More

generally, they involve providing the people skills which computers lack.

Additional complementarities of the (E) form, i.e., involving the three-way

complementarity with organization, IT, and product change, arise from inventive

dynamics of computer-based organizations. As we emphasized above, effective use of

computer systems involves a great deal of invention by the using firms. It is a complex

activity to discover and implement ways to gain from computers’ capabilities. As

computers have become more powerful, more flexible and more user-friendly, the largest

and growing fraction of this inventive activity occurs outside of the programmers’ job

function. It is now the duty of those who use computers, directly and indirectly, in large

ways and in small ways. In particular, this puts pressure on managers' and professionals'

cognitive capabilities. Adding more complexity, the most computer-intensive businesses

typically use computers for improved customer service and as the basis for new and

improved services. Invention of the new products those processes will deliver and of the

human side of the delivery mechanism are very difficult cognitive tasks. They call for

managers who can think of ways to take advantage of the new production processes

offered by computing. This calls for new cognitive skills, having a deep understanding of

one’s own organization and one’s customers’ needs.19 This raises the demand for high

levels of cognitive skills in managers and professionals.

Some of these new demands on highly skilled workers can call for a great deal of

human capability. It can be quite difficult to think about one's customers, for example,

both through the lens of knowing them and the more analytical lens of seeing their

19 Bartel and Lichtenberg (1987) suggest that high levels of cognitive skills may be particularly

important in creating and adapting to change, notably in implementing new technology. The managerialside of computer-based production processes is an excellent example of this story. The constantimprovements in computer technology mean that organizations that use IT intensely have this demand forhigh levels of cognitive skill on a standing basis, not just for a single transition.


behavior systematically in databases. Once the firm knows new and interesting things

about its customers, further, it must invent new ways to interact with them.

The last point is not a trivial one. As more and more of business is mediated

through computers, there are more and more opportunities for managers and

professionals to undertake two activities. The first is research-like, combining IT-based

quantitative information about customers or employees with deep knowledge of the

business. The second is designing organizations, products, and so on, taking advantage

of the new research findings. This, too, combines hard quantitative skills with deep

business knowledge. To the extent that the new organization or product will be enabled

by computer, the firm is specifying a new computer program. Even those managerial and

professional workers who never touch computers have their work transformed in this

way, calling for more and more complex bodies of skill and knowledge. Thoughtful

designers of business information systems recognize the information overload bottleneck:

"While we've seen a great revolution in bandwidth and computing power, unfortunately

the input/output capability of human beings hasn't changed much over the last decade".20

3.3 The net impact of IT and IT-enabled organizational changeTo summarize the implications of our theory for labor demand, we think first that

in computerized work environments, high levels of human information processing skills

should be getting more valuable. In part, this reflects the potential of present-day

computers to substitute more effectively for routine information processing tasks than for

complex or ill-structured cognitive tasks. It also reflects a potential complementarity

between some of the strengths of computers -- storing, communicating and executing

instructions on a flood of data -- and many types of human judgement and decision-

making. As Simon (1971) has noted "What information consumes is rather obvious: it

consumes the attention of its recipients."

One implications of this theory is that computerized firms will seek to address the

increased relative demand for high cognitive skill labor by hiring more skilled and

educated workers and by investing more heavily in training for existing workers.

Another implication is that the information overload bottleneck is likely to be especially


severe higher up in the hierarchy. Human capital should be especially at a premium here.

But firms will also seek non-hierarchical forms of organizing to bypass the bottleneck

(e.g. "self-managing teams"). As a result, computerized firms should make greater use of

self-managing teams and decentralized decision-making. Furthermore, in these firms,

employees with "team" skills, and "autonomy" skills should become more valuable.

Information technology impacts the optimal organization of work, and impacts the

demand for skills both directly and through changes in workplace organization.

4 Data DescriptionThe data set used for our analyses is a cross sectional survey of organizational

practices and labor force characteristics conducted in 1995 and 1996 matched to a panel

detailing IT capital levels and mix over the 1987-1994 time period. For some analyses,

we will also use data on other production function inputs and outputs, including non-IT

capital, total employment, sales and value-added. Data cover approximately four

hundred large US firms. Here we briefly describe each data source and then our

measures of key variables, with supplementary detail in Appendix A.21

4.1 Data sources: workplace organization and labor characteristicsWe undertook a new survey based on questions from prior surveys on workplace

organization and human resource practices (Huselid, 1994; Ichniowski, Shaw and

Prennushi, 1994; Osterman, 1994). Questions address aspects of workplace organization

including allocation of various types of decision-making authority, job responsibilities of

line workers, the use of self-managing teams, and promotion criteria. Labor force

questions include the levels and types of investments in training, percentages of the

workforce with various levels of education, hiring criteria, skill levels and the degree of

computerization of the work. We also included some additional questions about the

perceived effect of computers on the workplace.

Administered to senior human resource managers or their designees, the survey

asked questions about organizational practices at the most typical establishment. The

20 Robert Walker, CIO of Hewlett Packard, quoted in Mendelson and Pillai (1998).21 A more detailed description of our data set, and its relationship to earlier studies, can be found

in Brynjolfsson and Hitt (1997).


approach of Osterman (1994) was followed in focusing on a single class of employee

termed “production employees” (which corresponds to Osterman’s “core employee”). A

production employee was defined as “non-managerial, non-supervisory personnel

directly involved in producing a firm’s product or delivering its service”.

One potential difficulty of our sampling approach is that the practices reported by

the respondent may not be representative of the work practices across the entire firm. To

address this issue, the revised instrument contained questions about how representative

production workers were in terms of total employment and the uniformity of work

practices for this category of workers. Overall, for the average firm in the second survey

subsample, production workers account for about two thirds of total employment and

organizational practices are found to be fairly uniform: 65% of respondents said that all

production workers have the same work practices. Furthermore, 82% reported that at

least 80% of workers had the same work practices.

In addition to the detailed questions about work organization for production

workers, we also asked several questions that that relate to the overall structure of the

firm. In particular, we obtained the percentages of the labor force in various broad

occupational categories (managers, professionals, clerical, skilled blue collar and

unskilled blue collar). We also measured the extent of computerization by two questions:

the percentage of employees that use a general-purpose computer, and the percentage of

employees that use electronic mail.

These data were collected in three waves. A total of 416 firms provided at least

some data for the study, including 93 from the first survey, conducted in Summer, 1995,

138 from a survey administered in Fall, 1995 and 199 from a survey in Summer, 1996.

The survey instrument is reproduced in Appendix A, with links to the labels for all the

variables. Descriptive statistics are in Table 1.

4.2 Data sources: information technologyThe measures of IT use were derived from the Computer Intelligence Infocorp

installation database that details IT hardware spending by site for companies in the

Fortune 1000 (approximately 25,000 sites are aggregated to form the measures for the set

of companies that represent the total population in any given year). This database is


compiled from telephone surveys that detail the ownership for specific pieces of IT

equipment and related products.22 Most sites are updated at least annually with more

frequent sampling for larger sites. The year-end state of the database from 1987 to 1994

was used for the IT measures. These data include variables capturing the total capital

stock of IT (central processors, PCs, and peripherals) as well as measures of computing

capacity of central processors in millions of instructions per second (MIPS) and the

number of PCs. The IT data do not include all types of information processing or

communication equipment and are likely to miss a portion of computer equipment which

is purchased by individuals or departments without the knowledge of information

systems personnel.23 Descriptive statistics can be found in Table 2 and more detailed

discussion in Appendix Section 10.2.

4.3 Data sources for other inputs and outputCompustat data were used to provide additional firm information, such as

employment levels, not covered by other sources. Measures were created for output,

capital, labor, and value-added. These let us estimate a production function relationship

between value-added and the various inputs following the procedures in Hall (1990) and

Brynjolfsson and Hitt (1997). Approximately 55% of the sample are from the

manufacturing, mining, or construction sectors and 45% are in services. Table 3 provides

sample means for these variables and Appendix Section 10.3 more discussion.

4.4 Measures of key variablesIn our empirical work, we have used a simplified and streamlined version of

Figure 1.

4.4.1 Human CapitalOur richest set of measures falls in the upper, human capital demand, box. Here

we proxy the choice of more highly skilled workers with

• Responding manager’s assessment of worker skill levels.

22 A question might be of the form “Do you still have the VAX-11/780 that your reported last year

at this location and if so, have you changed its configuration?”23 Another potential source of error in this regard is the outsourcing of computer facilities. To the

extent that the computers are located at the company site, they will still be properly counted by CII’scensus, but equipment at the outsourcer's other locations will be missed.


• Education mix of workplace, and• Occupation mix of workplace

While we observe skill demand, we also have indicators of its change over time,• Firms’ hiring and training policy for skills.

Our interpretation is that these are alternative measures of the same broad concept

-- a tendency at the firm level to use, or to attempt to use, more highly skilled workers.

We measure human capital in several direct ways: via the manager's assessment

of workforce skill (SKILL), the educational composition of the firm's workforce

(%SCED and %COLL) or the percentage of employees who are professionals (%PF),

managers (%MG), clerical (%CL), etc. Descriptive statistics on these variables may be

found in Table 1 and more discussion in Appendix Section 10.1. We also construct a

broader measure of information work based on the percentage of all employees who are

managerial, professional or clerical (%IW) -- precise definitions of all constructed

variables are in Table 5.

While we observe those human capital measures only once in cross section, we

also have a measure of the firm's policies toward human capital investment. HKI is

defined as a mean zero, standard deviation one variable that is created from the percent of

workers receiving training, the extent to which the firm screens for education in hiring

and the firms' use of cross-training (definition, Table 5). Like the human capital levels

variables, we observe this only at the end of our sample period.

4.4.2 Workplace OrganizationOn the left of Figure 1 lies the concept of workplace organization. Workplace

organization is notoriously difficult to measure and the changes in organizational

structure that are complementary to IT capital might be particularly so. Gathering new

data on workplace organization and linking these variables to IT capital at the firm level

is one of the advances that permits this study, and the earlier ones by Brynjolfsson and

Hitt.

In the broad and general sense, it is well established that organizational change

and information technology are complements.24 However, only a subset of the relevant

24 See, for example, the references listed in notes 8 and 12 above.


organizational changes is captured in our survey and used in our empirical work.25 We

focus on one particular type of organizational change that has been found to be important

in earlier work by Brynjolfsson and Hitt (1997). Our variable WO is defined as a mean

zero, standard deviation one variable reflecting the several distinct measures of

workplace organization greater use of teams, greater delegation of a variety of decision

rights to line workers, and activities that support the use of teams such as team building.

(Definition, Table 5.) 26 These variables have a clear interpretation as measures of

decentralization. Our concept of workplace organization is narrow and specific.

Since our data on organizational characteristics are based on a snapshot at the end

of the sample period, we do not know whether each firm had the same organizational

characteristics throughout the sample period. The reasonable interpretation based on

aggregate data, however, is that many of the firms have only recently adopted these

practices.27 In a measurement sense, most of WO is ∆WO.

4.4.3 Information TechnologyAt the right of Figure 1 is the concept of IT drivers. Our IT measures are stocks

of computer hardware at the firm level and also stocks of specific types of hardware (e.g.

PCs).28 Empirically, we view them largely as alternative measures of the same concept.

A distinction among the types of computers between those suitable for organizational

computing (mainframe and minicomputers) and those for personal computing (PCs)

25 For instance, we do not have any data on changes in outsourcing which might plausibly change

the skill mix of remaining employees. Nor do we attempt to catalog and measure all the types of internalreorganization that might occur. Omitting these other forms of organizational change probably leads us tounderstate the role of organizational factors.

26 This choice of measures reflects the first principal component of a factor analysis of responsesto the organizational practices survey. Interestingly, Osterman (1994) reports that a similar set of team-oriented practices loaded on the first principal component in his survey of 694 establishments.

27 The practices we use to define WO are very similar to the ones the "new work practices" that ,according to Ichniowski et al (1996), "have become increasingly common among U.S. businesses in recentyears". In fact, Osterman (1994) asked about a similar cluster of work practices in a survey of 694managers and found that 49.1% of the establishments in his sample reported introducing "teams" in the fiveyears prior to his survey year of 1992. He also reports that 38% introduced job rotation practices, 71%TQM programs and 67.9% problem solving groups in the years between 1987 and 1992, each of which alsoreflects increased decision-making by line workers.

28 It is more difficult to observe the portions of IT expenditure that are distributed throughout thefirm (personal computers, small web servers, etc.). It is more difficult to observe software investments than


sharpens this somewhat. In the period of our data, however, PCs could be an indicator

either of a tendency to personal computing or of new-architecture organizational

computing. Thus the sharper interpretation is ambiguous and most of our interpretation is

the broad one that some firms are more IT intensive than others.

4.4.4 Strategic GoalsThe most difficult observable in terms of the cluster of driver technologies is the

change in product and service quality and the invention of new products and services, the

box at the bottom of Figure 1. Success, which is related to one interpretation of our

production function results, may be an observable indicator of these. Our interpretation

is that firms that enjoy measured productivity growth have achieved one or more of the

subgoals at the bottom of the Figure.

4.5 Samples.Data for some firms could not be matched to either CII or Compustat data and had

to be dropped from the analysis. Some firms provided only partial data or were missing

data from Compustat, reducing the sample size for many of the analyses. We used the

largest available sample in each different kind of analysis. Thus there are several

different subsamples used in the analysis due to missing data or restrictions to a single

year. The major samples are outlined in Table 4.

4.6 The nature of the variation in our dataOur data is a mixture of a panel (the CII and Compustat sourced data) and a cross

section of organizational and human capital variables observed at the end of the sample

period. In our data, the correlations are largely driven by cross-sectional differences

between firms; we have only limited evidence regarding the effects of changes over time

in the variables of interest.29

These measures are related to the short-run production framework introduced

above in distinct ways. First, labor clearly can be treated as a quasi-fixed factor of

hardware investments, and it is particularly difficult to observe applications software that would reveal thepurposes of business computer systems.

29 While it would have been ideal to have panel data on organizational characteristics as well,these data were unavailable historically and extensive retrospective surveying was deemed too unreliable.


production (Hamermesh (1993)) and this applies with particular force to skilled labor (Oi

(1962)). Of course, the entire labor input to the firm may be varied by use of temporary

workers, outsourcing arrangements, specialized suppliers, and the like, but our measures

are of the skill composition of employees. Accordingly, we treat our various measures of

skill input as quasi fixed. In contrast, we treat HKI, policies toward changing the skill

mix, as highly variable in the relevant run.

Workplace organization in general is also quite fixed in the short run. This has

been particularly well documented for the kinds of organizational changes that are

complementary to IT investment.30 Information technology is a highly variable factor of

production, if by IT we mean the stock of hardware and purchased software. There are

large adjustment costs associated with IT, but those are the costs of adjusting the

complementary WO, not the IT itself (cf. Bresnahan & Greenstein (1997), Brynjolfsson

and Hitt (1997), and especially Ito (1996) on this point).31 Indeed, one of the barriers that

prevents competitors from quickly imitating the most successful computer users is the

subtlety and complexity of the organizational co-inventions needed to harness this

revolution in computer power.32 Like econometricians, rival managers may be able to

observe the broad outlines of a successful strategy without knowing the myriad details

necessary to implement it effectively.

Our approach to the factor demand system respects this pattern of relative

variability and fixity. We treat the HK and WO variables as the most quasi-fixed in cross

section, and the IT as the most variable. (This even though we believe that the largest

changes in time series to this cluster of complements is falling prices and rising

performance for IT.)

30 See, for example, Bresnahan and Greenstein (1997), Brynjolfsson, Renshaw and van Alstyne

(1997), David (1990) or Orlikowski (1992).31 If we were to expand the definition of IT to include applications software and the training and

reorganization needed to use it effectively, it would become more fixed in the short run. Our observables,however, correspond to the adjustment cost relationships described in the text. There might well be adifferent way to divide the cluster of complements, focussing on applications of IT rather than IT hardware,which would yield a more detailed and nuanced view of the pattern of adjustment costs than we canprovide here. That, however, would call for more detailed data on the software side of IT.

32 Indirect evidence for this fact is the high gross marginal products found for IT when standardproduction functions are estimated using firm level data (Brynjolfsson and Hitt, 1995, 1996, Lichtenberg,1995).


Since all three members of the cluster are complements, and since our purposes

are to understand how the cluster of the three of them functions as a driver of changes in

labor demand, the limited observability is not a major problem for us. What we will do is

use available measures, examine their tendency to move together, and examine their

tendency to be correlated with labor demand changes.

5 Basic Results

5.1 IntroductionThe theory laid out in the sections above could be tested in any of a variety of

ways in firm data. What we do in this paper is first examine all of the correlations that

might be indicators of the associated complementarities and then build successively more

structured models, addressing alternative explanations for our results along the way.

5.1.1 Our econometric strategy starts with simple models and then addsstructureOur hypothesis that IT, human capital, and decentralized organizational structure

are all complements has a number of testable implications for the firm-level data.

First, under plausible assumptions, we would expect levels of each of them to

covary in cross-sectional / time series data.33 We should expect firms that for some

reason have higher levels of IT than their competitors should, on average, also find it

more profitable to hire more educated workers, to train existing workers more intensively

and to organize production using self-managing teams and delegated decision-making.

Similarly, firms with higher levels of human capital, all else being equal, would be

expected to benefit more from investing in IT and organizational innovations.

Second, these simple correlations can be modeled as short-run demand equations

for IT and HKI, the most variable of the cluster of complements in the theory. There is

also a dynamic version of this point, given the substantial adjustment costs and invention

33 Holmstrom and Milgrom (1994) offer a theory of complementarity and a series of observations

about the conditions under which co-movements of different organizational practices and factors ofproduction are indicators of complementarity. Athey and Stern (1998) are rightly critical of interpretingco-movements in a cross section of firms as indicating complementarity when there is a possibility ofomitted common factors.


costs in our data. We will examine how these change with alternative definitions of the

variables, and with alternative treatments of the dynamics.

Third, output should be higher in firms that successfully match IT, organizational

structure and human capital investments than in those that cannot make such matches.

Specifically, estimates of a production function should find a that the interaction terms

among IT, human capital and decentralized organization are positive and significant,

indicating complementarity, if there is variation in firms' success in matching.

Finally, the firms' managers, to the extent they speak the relevant language,

should themselves report that such a pattern reflects their conscious efforts to match these

practices with each other. The managers with the strongest beliefs that IT increases the

demand for skilled labor and decentralized organization, should be most likely adopt

these practices jointly.

Of course, there are a number of statistical and econometric issues that will need

to be addressed for each of these analyses. In particular, the issue of causality, which is

always problematic in econometric work, is particularly murky in this case. In addition,

alternative explanations are possible for many of the specific statistical relationships that

we uncover. However, by examining the data from a variety of perspectives and in some

detail, we hope to distinguish which theory provides the most parsimonious explanation

for the overall pattern of results.

5.2 Correlations among IT, labor characteristics and workplaceorganization

As discussed in the data section, we have numerous different measures of firms’

IT capital, their employees’ human capital and overall workplace organization. The first

way we examine the data is simply to consider correlations among these measures. We

begin by examining Spearman rank-order correlations among pairs of the relevant

variables while controlling for industry, firm size (employment), and the composition of

the principal production workers in each firm. The last control is really a production-

process one; we control for it because of a concern that firms whose production workers

are professionals or clerical workers may be very different than those whose production

workers are blue collar. In Table 6 we report the correlations among our composite


measures of IT, HK, and WO. It is worthwhile, however, to examine these correlations

in somewhat greater detail.

5.2.1 IT use is correlated with employees' human capital in firm level dataWe find that most measures of IT are significantly correlated with most measures

of employees’ human capital (HK) in our sample of firms (Table 7). The correlation is

stronger when the measures of IT are taken from the same organizational survey as the

human capital data (first three columns) than when we use the IT measures from the

separate CII survey. This probably reflects a better match of the unit of observation.34

The finding of a correlation between IT and HK in firm-level data extends findings in

previous work, which used more aggregate data.

We also find that most measures of IT are correlated with greater investments in

human capital, such as training and screening new employees on the basis of their

education (Table 7, middle rows). IT can predict greater investments in human capital

(HKI) even when we control for current levels of human capital (by including SKILL and

EDUC as partial covariates in the bottom rows of the table). This is consistent with the

hypothesis that IT-intensive firms are transiting toward a new cluster of complements,

relative to less IT-intensive firms.

Classification of employment by job title is also correlated with IT. Firms which

employ more managers and especially professionals are more likely to have high levels

of IT relative to their industry competitors while those with more blue collar workers tend

to have less IT, and vice versa (Table 8). The employment job titles can be interpreted as

proxies for employee skills, as is common in earlier work, or as characteristics of

workplace organization.

5.2.2 IT use is correlated with specific characteristics of workplace organizationIn Table 9 we examine the correlations of various measures of IT with our

summary measure of decentralization, WO, with the measures that underlie it.35 All of

34 The respondents for the organizational practices survey were explicitly asked to consider a

representative site for both the HK and IT questions. In contrast, the CI data were for the company as awhole.

35 See Brynjolfsson and Hitt (1997) for more detail on the empirical relationship between IT andworkplace organization.


the components of WO are individually correlated with IT measures. Interestingly, the

strong correlations with self-managing teams and team building exercises suggest that

"people" skills, not just decision-making skills, are important complements of IT.

In these tables, one would conclude that there are significant relationships

between IT and human capital and between IT and workplace organization whichever

measure of IT was used. There are two further points worth noting. First, PCs are not the

strongest correlate for most measures of human capital. Instead, central computer use

appears to be a slightly better predictor, which is consistent with an important role for

organizational computing and, by inference, organizational impacts. Second, email use is

a particularly strong indicator of teams and related practices, as well as of high levels of

human capital. This could suggest an important role for communication, as opposed to

automation, in driving skill upgrading. Like the other correlations, it may also simply

reflect some residual unobserved heterogeneity in the sample.

5.2.3 Human capital is correlated with workplace organizationIn turn, the same decentralized workplace organization that is correlated with IT

is also is highly correlated with employee human capital and with firms' attempts to

increase human capital via training and employment screening (Table 10). Our finding

that certain workplace systems are correlated with higher levels of human capital extends

and amplifies a result reported by Osterman (1994) who observes "As the skill levels

required by an enterprise's technology increase, so does the use of the various work

organization innovations."36

Some, but not all of the correlation between IT and human capital that we found

can be “explained” through their common covariance with differences in organizational

structure. When we partial out workplace organization, the correlation between IT and

most measures of human capital drops somewhat (Table 11). While the three move

together, there is also substantial separate movement and co-movement of the pairs. This

is consistent with the view that all three are complements. However, it is also consistent

with the view that all three are driven by the same underlying causes or coincidences.


6 Estimates of More Structured ModelsThere are several possible explanations for the correlations we find among IT,

human capital and organizational decentralization in our sample of firms. We seek to test

one of these, the multi-way complementarity theory that IT is most effective in the

presence of skilled workers and decentralized organization when used for service

improvements. There are two distinct subtasks to the test. We will verify, in several

different frameworks (of which the correlations just seen are the first) that the

implications of the complementarity theory are observed. We will also examine a wide

range of types of evidence in order to rebut, to the extent feasible, alternative theories of

the correlations.

6.1 Alternative theoriesThere are two important classes of alternative theories. The first class posits

unobserved heterogeneity in the cross section of firms. If there are shocks to the value of

all the complements not observed by us (which there certainly are) and if those shocks

are correlated in the cross section of firms for reasons unrelated to complementarities,

then that provides an alternative explanation of the correlations among the hypothesized

"complements." It may merely be a coincidence that a common cause of all of them

occurs only in a subset of the sample firms. If the same shocks are correlated with

changes in productivity, then this class of alternative theories can also provide an

explanation of why a measured production function might appear consistent with

"complementarities" that aren't there (cf. Athey and Stern (1998)). Several specific

variants exist.

What could cause such a shock to the value of all these factors and organizations?

A plausible story is one in which some firms have skill or luck which makes their output

more valuable. To complete the story, we need to explain why the firms spend the

resulting revenue on these particular factors. A classical story follows from the

expansion path of inputs as the scale of output rises. That story seems very weak in the

36 The workplace innovations tracked by Osterman were teams, job rotation, TQM and quality

circles. His measure of skills was a dummy variable for whether "the production process requires a highlevel of skill."


present instance -- it posits larger firms systematically more computer-intensive, more

decentralized, and more using of skilled workers -- and we rule it out a priori.

A perhaps more sensible story in the present instance is related to managerial

rents and free cash flow. Managers might take pleasure from these particular inputs (it

might be nice to work with smarter and more capable people, etc.) and spend free cash on

them.

A related story is that we are missing some important inputs -- applications

software comes to mind -- and that those omitted inputs raise the value of HK, WO, and

IT. This theory may be true, and that may be a problem for causality in our estimates, but

this theory is simply a more elaborate version of our own story, so we are happy to accept

it.

Another variant posits that firms are simply in different economic circumstances.

More human capital, more computers, and more decentralization might simply be more

useful in some kinds of industries or some kinds of firms within industries. The variety

in their circumstances separately causes some to use more, some to use less, of the

"complements."

Finally, there might be a fad: some managers might have decided that all three of

the complements are useful, with no particular foundation, while other managers decide

to keep their cool. These common unobserved cause or heterogeneity stories are a

general problem in the investigation of complementarities.

A second class of alternative theories posits a distinct direction of causation, but

one that is unrelated to production complementarities. It is possible that firms with many

skilled workers were most aggressive in investing in the technology because they

perceived IT as being a good substitute for their skilled workers. In other words, perhaps

the managers’ goal was to automate and replace skilled workers with ever-cheaper

computers. The organizational change could be the mechanism by which this

substitution took place.


A related story is that rather than computerization causing an increased demand

for skills, the causation is in the reverse direction and unrelated to productivity.37

Workers in high-rent jobs get the computers as toys. Note that it is not sufficient to

simply reverse the causation here. The lack of productivity is critical. Reverse causation

associated with productivity is just another complementarity at the firm level.

6.2 Estimates of the determinants of firm-level demand for ITSince the most variable of the factors in the cluster of complements is computer

capital, we first estimate a series of firm-level demand equations for IT. We cannot

estimate a classical system of factor demand equations, since we do not observe prices

for the factors varying across firms. However, we can estimate short-run factor demand

equations, with the relatively fixed factors as predictors of the more easily variable IT.

6.2.1 Human capital and decentralized organization shift out the IT demandcurveIn our baseline demand specification, we assume that (log) IT capital stock in a

firm will be a function of (log) firm output, primary production worker occupation,

industry and year.38 In addition, we treat WO and HK as quasi-fixed, including them as

regressors as well. If the skill levels of a firm’s workforce and changes workplace

organization are complementary with computerization, then WO or HK or both should be

predictive of firm-level demand for IT.

In Table 12, we present estimates of several variants of this IT demand equation.

We begin with a measure of worker skills, measured as a five-point scale and

standardized to mean 0 and variance 1. Firms with a one standard deviation higher level

of skill have about a .095 standard deviation increase in IT (Table 12, column 1) or

13.5% higher demand. When a variable is added to this equation for workforce

organization (column 3) we find that it is highly significant and reduces, but does not

eliminate, the coefficient on skill, similar to what we found in the raw correlations.

Moreover, the skill effect and the organizational effect are roughly the same order of

37 DiNardo and Pischke (1997) make this argument -- and provide convincing evidence for it at the

level of the individual worker and the PC.38 Under the assumption that all firms in the same industry face the same prices each year, this

specification implicitly accounts for changes in the price of IT and the prices of other inputs and outputs.


magnitude (standardized skill coefficient .0712, standardized organizational coefficient

.0799).

Similar results are found when we add two additional measures of human capital:

the percentage of college educated workers and the percentage with some college

education (the omitted category being workers with high school education or less). Firms

with a one standard deviation higher proportion of college educated workers have an

additional 14% higher demand for IT above the effects of (column 2). This suggests that

human capital is a multidimensional concept and is only imperfectly measured by any

single variable. As before, including a measure of workforce organization decreases but

does not remove the human capital effect, regardless of how it is measured (columns 3,

4).

Interestingly, lower levels of education appear to have a much smaller influence

on IT demand than higher; shifting workers from the “high school or less” category to the

“some college” category appears to have little (or even a slightly negative) effect on the

demand for IT. One interpretation is that not only is education a complement for IT, but

high levels of education are disproportionately important – a “twist” in the IT-education

complementarity.

Probing further into the nature of the organizational effects, we look at how the

composition of the workforce (percentage of professionals, clerical workers, managers,

skilled blue collar and unskilled blue-collar workers). We find that overall workforce

composition explains a substantial fraction of the demand for IT, although most of the

explanatory power comes from the percent professionals variable (Table 13). These

effects are above and beyond the existing effects of skills and education (which remain

significant). As before, adding WO lowers the coefficients on each HK measure

(including professionals) but the sum of the two effects exceeds the HK effect alone.

Finally, it is important to note that, while WO and HK are both good predictors of

all types of IT investments, they are relatively weak predictors of the demand for other

types of physical capital, highlighting the special relationship among IT, WO and HK.

Table 14 has the same specification as Table 12, except that the dependent variable is

switched from IT capital to non-IT capital. The coefficients on the human capital


variables shrink, to a quarter to a third of their levels in the IT capital regression. There is

far weaker skill-capital complementarity, it seems, than skill-computer complementarity.

Furthermore, the coefficient on WO completely disappears, indicating that the

organizational changes are linked to computerization but not to capital intensity.

In summary, several different measures of human capital and workplace

organization predict IT demand, even when entered simultaneously into a regression.

This reflects not only the fact that human capital and organization are multidimensional

concepts, but also the fact that each of our variables measures the underlying concepts

with error. The effect of a one-standard-deviation increase in WO plus HK (however

measured) exceeds that of HK entered alone. This is consistent with the view that both

are complements for IT.

Of course, it is also consistent with the view that there are common unobserved

causes or reverse causation at work. However, the inclusion of the sector and production

process (production worker composition) variables undercuts the theory that it is simply

variety in circumstances behind the correlations. The ability of HK and WO to predict IT

demand survives a substantial amount of conditioning out measures of the circumstances.

6.2.2 Dynamic Complementarities? Predictors of IT growthThe results in Table 12 and Table 13 indicate that various measures of human

capital and work organization shift outward the overall demand curve for IT. For any

given set of prices and business conditions, firms with more human capital, more

decentralized decision-making and more professionals tend to use more IT. In Table 15

we examine a version of this specification that is, at least in part, more dynamic. We add

interaction terms between a time trend and two human capital and organizational

structure measures. We can undertake this partially dynamic analysis as we have IT data

for each year of our sample period, but our measures of human capital and workplace

organization are only available for the end of the period.

To interpret this table, recall that firms' work organization has been changing over

the course of the sample period. As we said above, WO, measured at the end of sample

period, reflects the extent to which firms have made such changes. The human capital

and skills stocks of most of our firms have been relatively stable, by contrast. Table 15


shows that the interaction between WO and time can explain increases in IT demand,

while that between SKILL and time is insignificant and possibly negative. Even when

we exclude WOxTIME, the human capital measure does not pick up the effect.

The most direct interpretation is complementarities. The interactions with time

pick up changes in IT over time. WO is changing in many of our sample firms, and

where it is changing, IT is growing. HK is more stable, and so the level of HK at the end

of the period does not predict changes in IT. In one story, WO is a new organizational

invention39 (or IT-enabled WO is a new organizational invention). Those firms that have

moved forward in WO have also moved forward on the other complements, by this

interpretation.40 A closely related theory is that the complementarity between IT and

human capital is long-standing, as is the complementarity between IT and organizations

in general, but the complementarity with the particular type of workplace organization

that we measure is relatively new and less well understood. This would suggest that the

IT-HK complementarity is closer to equilibrium while we are observing firms actively

investing in both WO (narrow definition) and complementary IT.

The fact that the increase in IT over time is higher in firms with the decentralized

workplace organization, but is not higher in more human-capital intensive firms, is robust

to alternative definitions of WO or HK (as we discovered in results not shown).41

Furthermore, it is robust when we include tighter industry controls (1, 2 or 3 digit SIC

codes) or even separate firm dummies (fixed effects) in results not shown.42 Examination

of the correlations between the regressors in Table 15 and differences over time in IT

reveals that this is a slow-developing effect. The correlations are small for the

39 This is essentially the view that Ichniowski et al. (1996) put forth in their summary of the

literature on innovations in workplace organization.40 Accordingly, it is not a criticism to note a weaker correlation between WO and IT in the early

part of the sample (since our WO data are only contemporaneous with IT at the end of the sample period).Similarly, it is not an econometric criticism that WO(1995) is a measure of WO(t) with an error thatdeclines over time, rather, this econometric remark is closely linked to the economics of our interpretation.

41 For example other measures of WO such as share of professional workers, use of teams orextensive delegation of decision-making and other measures of HK such as share of college graduates orworkforce skill levels.

42 In the fixed effects regressions, WO and HK are included only as interactions with time, not asdirect variables, since the firm dummies pick up time-invariant differences between firms.


(notoriously noisy) one-year first differences in IT; it is only when we look at four-year

or longer first differences that the effect is strong.

Some differences emerge when we use different measures of IT for the dependent

variable (Table 16). The first of the three columns is reproduced from Table 15 for

comparison, and bears something of a resemblance to the MIPS column. There are

considerable differences between the MIPS and TOTPC columns. The coefficients of HK

and WO variables in the levels are larger in every case for the MIPS column. The

changes over time vary in the same direction.43 Neither WO nor SKILL has much effect

on the change over time in the demand for PCs. In contrast, higher WO firms are

investing in more and more MIPS over time. Our interpretation of this is as confirming

something about the technological history of our sample period (1987-1994). MIPS here

measures large, powerful computers, including both the mainframes that traditionally

supported organizational computing and the (mostly minicomputer and workstation)

servers which began to replace them in the "competitive crash." Investing in changed

WO and investing in organizational computing are the complements here. It was widely

rumored that PCs were the key to the new kind of decentralized computing. The

organizational significance of PCs may well have increased after our sample period and

not during it, as reported for example by Bresnahan and Greenstein (1997). These more

recent trends are that PCs have increasingly become networked and with the rise of the

Internet, browser-based organizational applications, and “downsized” organizational

computing.

We view these more dynamical results as consistent with the complementarity

story. One of the complements to IT is increasing over time in a subset of our sample,

and it is in those firms that IT is increasing particularly rapidly.

6.3 Both IT and workplace organization affect human capital investmentAnother more dynamic specification is possible if we treat not the level of human

capital but policies related to its rate of change as the variable factor. We estimate a

variety of equations for human capital investment (HKI) using our firm-level data in

Table 17. Our hypothesis suggests that both IT and workplace organization should affect


the demand for human capital investments such as training and screening new workers by

education. We are concerned, however, that the highly variable stock of IT may be

contemporaneously correlated with a free-cash-flow error in the HKI equation. We use

lagged values (four years) to reduce the possibility of spurious correlation that can arise if

"lucky" firms with excess free cash flow in one year decide to invest more in both

training (and hiring educated workers) and computers or "unlucky" firms cut back on

both simultaneously.

We find that the lagged44 value of (log) IT per worker predicts human capital

investment (column 1,Table 17) in our firm level data. This result is consistent with

earlier work with more aggregate data. We can directly interpret the coefficients: a one

standard deviation increase in lagged computer capital is associated with a increase in

human capital investments of about .23 standard deviations. When we add sector and

industry controls, they increase the R2 significantly, but they reduce the coefficient on

computerization only slightly. This suggests that, in these data, the within-industry

effects of computers dominate the cross-industry effects, highlighting the benefits of firm

level data.

If the correlation between IT and human capital investments in part reflects the

multiway complementarities that we hypothesize, then workplace organization should

also predict HKI. With or without sector controls, the organizational variable is highly

significant and predicts a substantial increase in HKI. In each specification, the sum of

the standardized coefficients for IT and WO exceeds the coefficient on IT alone. This is

consistent with the view that each of IT and WO is a complement for human capital, and

that IT and WO are not such perfect complements as to perfectly covary.

The first two columns of Table 18 present a similar set of regressions but

controlling for existing levels of workforce skills. While the coefficients on IT and WO

decline slightly they are qualitatively similar. Firms with high levels of IT and WO use

high HKI strategies, whether or not they already have a great deal of HK. Finally, we can

see (in the right of Table 18) that the coefficients are little changed when current levels of

43 We get much the same story when we interact SKILL and WO with time dummies, not trends.44 Fourth lags were chosen to be sufficiently long to examine long term effects, but short enough

to minimize data loss in our sample.


IT are instrumented with past levels of IT. It seems that the HK and IT investments are

not a reaction to some short run free cash-flow shock.45 When we change the measure of

IT to a question about the computer-intensity of work tasks (COMP), the coefficient is

significantly higher. This might simply reflect the fact that both COMP and HKI were

drawn from the same survey, improving the match of the unit of observation. More

plausibly, it might mean that broad measures of IT (like ITCAP, MIPS or TOTPC)

include both systems tightly connected to the work of the organization and other systems

that affect few jobs.

We can also repeat the analysis using different measures of computerization:

MIPS and PCs (Table 19). The MIPS results are quite similar to those using ITCAP.

However, PCs are a weaker predicator of HKI. In the case where WO is included, it does

most of the work leaving a small and insignificant coefficient for PCs. The technological

story is the same as above. It is organizational, not personal, computing that predicts

HKI in this time period, just as that was the form of computing most strongly predicted

by the HK variables above. In Table 20 we consider the effect of managers’ opinions

about the effects of IT on work. Firms where managers perceive that IT increases skill

requirements and worker freedom are especially likely to invest in increasing human

capital. Managers' opinions regarding IT’s effects on monitoring and routinization have

no detectable effect on human capital investments. The coefficients on WO and IT per

worker decline slightly when we add the managerial variables. If one takes the existing

level of worker skills as given in the short run, then the managers’ reported intentions

suggests that both IT levels and workplace organization determine the demand for new

human capital investments. This suggests that one can model the demand for new human

capital investments as a function of existing levels of human capital, IT, workplace

organization and exogenous factors like industry.

In summary, workplace organization and computerization predict policies

favoring human capital upgrading as the multi-way complementarity theory predicts. A

series of finer distinctions also are consistent with our more specific complementarities

theory. It appears to be organizational computing that matters the most, and it is

45 In fact, if current levels of computerization are used as a regressor, but not instrumented with


specifically the computerization of work tasks that is related to human capital upgrading.

It is in firms where managers articulate our narrow WO theory that human capital

upgrading occurs. As we condition out more and more effects, the complementarity

relationships do not disappear.

6.3.1 Relationship to Industry-Level StudiesAlthough our firm-level data are not directly comparable the industry data used by

Autor, Katz and Krueger (1997), we are able to parallel some of their basic analyses

(their table 8). We follow their specification of the regressors as closely as possible.

Their dependent variable is changes in HK over time, measured as changes in the

college-graduate wage bill share in industries. Our closest analog is HKI, human capital

investments by firms, and it forms the dependent variable in Table 21. Following AKK,

we consider several variants of the basic model, which also control for changes in other

types of capital, changes in (log) output, changes in IT per worker and separate

regressions for manufacturing and services (Table 21). In general, our results are

consistent with theirs. We find that the coefficient for IT per worker is little changed

across these specifications.46 We also find that most of the controls are not significant and

that the change in the current ratio of computers to labor (Table 21, column 4) has much

less explanatory power than the lagged effect of computers.

The next table (Table 22) adds our organizational decentralization variable (WO)

and industry controls to these same specifications. The results show a by-now-familiar

pattern. WO is highly significant and relatively stable across specifications. While the

coefficient on computerization falls, the sum of the WO and computerization coefficients

rises. These specifications, which were suggested by industry-level work, are also

consistent with the multiway complementarity theory.

The results in this subsection show the importance of not only IT but also

workplace organization in affecting the firm-level demand for human capital. They

extend the industry-level work in that they identify substantial linkages between IT and

past levels, the coefficient is substantially lower (not shown).

46 Although the coefficient on computers is higher in manufacturing than services, the difference isnot statistically significant (t=.8).


HK at the firm level.47 The firm-level evidence also permits telling a more detailed story,

with a significant causal flow through workplace organization.

6.4 How do various combinations of IT, human capital and workplaceorganization contribute to output?

If all managers fully understood the potential complementarities among IT,

workplace organization and human capital, if their decisions always reflected the profit-

maximizing strategies for their firms, and if there were no adjustment costs or mistakes in

implementing these strategies, then the demand equations IT and for HK would fully

reflect the optimal relationships among these factors. Furthermore, under these

assumptions, we would not expect to observe any firms that were not implementing the

optimal combinations of these practices. Of course, one of the reasons that we think

studying these relationships is interesting is that we believe that they are not fully

understood. With the benefit of hindsight, it is clear that many firms make significant

mistakes in their IT investments, organizational strategies or human capital policies.48

Furthermore, a large number of firms actively experiment with these policy levers in a

conscious effort to understand the how to best use the every cheaper processing, storage

and communications power delivered by IT. As with any important general-purpose

technology, a tremendous amount of organizational and strategic co-invention must

accompany technological advances in computing a most managers are fully aware of this

need.

Under these conditions, different combinations of inputs will lead to differences

in product and service quality in our sample of firms. Under the assumption that higher

quality levels will be manifested in increased revenues, a production function can help to

identify complementarities. Indeed, the production function approach will be most

effective to the extent that firms are not all employing a well-matched combination of

inputs. Thus, this approach provides a valuable counterpart to the correlation results and

47 It would be difficult to make a quantitative linkage between the industry and firm level results,

so we do not attempt it. IT-enabled movements of firm boundaries probably make industry (or supplychain) changes in HK as large as those within individual firms. Our sample covers only large firms, not thesmaller ones. And our dependent variable measures HK policies, not HK outcomes.

48 See, for instance, Kemerer and Sosa (1991) for a catalog of disastrous IT projects, some runninginto the tens of millions of dollars. Many of the failures are linked to organizational and strategic errors.


demand equations, which both provide the strongest results when firms are successfully

optimizing.

More generally, suppose that managers are attempting to increase firm output

quantity and quality by choosing the optimal combinations IT, WO and HK. Suppose

further that they are imperfect in these attempts, but do better than mere chance. Then we

would expect to find evidence of complementarities both in the demand equations, which

reflect the non-random covariance of IT, WO and HK, and in the production function

equations, which reflect the increased value of output generated by the firms that get the

combinations “right” relative to those who don't. In this way, we can link differences in

firms' choices of IT, WO and HK to their degree of success in achieving their

organizational goals.

Accordingly, we estimate a series of production functions. Following common

practice, we focus on a loglinear specification and its variants. This functional form has

the advantage of its simplicity and the straightforward interpretation; the output elasticity

of any input can be directly read from its coefficient. It is the interaction effects between

different factors of production that measure complementarities and substitutabilities in

this framework, where in the demand curves we just reported it was the levels of the

coefficients of the quasi-fixed factors. Accordingly, we build a specification starting

from a very simple specification (consistent with Cobb-Douglas production) and adding

richer and richer sets of interactions.

The baseline production function includes controls for industry and year and three

types of productive inputs: labor, IT capital, and non-IT capital (Table 23, column 1). In

each of the rest of the columns of the table, we add a single measure of HK and the

interactions with IT. Except for college education, the levels of HK have positive

coefficients , significant for professionals and for percent information workers (defined as

the percentage of clerical, managerial and professional workers). The interaction terms

between IT and HK are positive and significant in all specifications except for

professionals.49 And they are not small. The measured output elasticity of IT is

49 Note that IT is centered for the interaction terms, while all the human capital and workforce

organization terms are standardized to mean zero, variance one. As a result, the coefficients on the crossterms can be interpreted as the increase in IT elasticity for a one standard deviation increase in the practice.


substantially higher in firms with more skilled workers and the new work organization.

Similar results also hold when the regression includes WO and its interaction with IT

(Table 24). When WO is entered by itself, the direct effects and the interaction with IT

are positive and significant. Most of the other coefficients are similar to those in the

previous table.

Similar results are also found when we examine the interactions between WO and

various measures of HK (Table 25). As before, all the direct terms and interactions are

positive. The interaction between WO and skills and WO and professionals are both

significant again supporting complementarity between these factors. The other two HK

measures do not tell as clear a story of complementarities. Nonetheless, a given increase

in HK is associated with more output in high WO firms, and vice-versa. It is intriguing

that the complementarities are especially strong with professional workers: that is a

category with the greatest increase demand in the nationwide aggregates.

Finally, when we consider three way interactions ITxHKxWO as well as all two-

way interactions simultaneously, we find that most of the interaction terms are positive

(Table 26) again suggesting broad complementarity between these practices. However,

given that the variables are fairly collinear, these effects are fairly imprecisely measured.

In most cases, the coefficients on WOxHK, ITxWO and ITxHK are reduced in the

presence of other interactions, suggesting that coefficient on any given pair of practices

may partially reflect the effects of all three aspects of the complementary system. This

also suggests that we have pushed the data right to the limit, as the number of firms who

are in categories like "high on WO, low on IT, high on HK" is necessarily relatively

small (especially given what we know from the complementarities evidence of the

demand equations.)

The interaction specifications permit us to undertake some simple predicted-

productivity calculations that are illuminating. Consider a firm that might be two

standard deviations away from the industry and year mean on any or all of HK, WO, IT

axes. When we look at the predicted values from Table 26 we see that a firm that is high

on all three axes has very high predicted productivity - about 10% above a firm that is at

the mean on all three. If we change to the "mixed" cases (high-low-high, etc.) we find


that the predicted productivity falls to worse than the mean in almost all cases. The very

interesting case of low-low-low, is one that is exceptional. It has about the same

predicted productivity as the mean (higher in the specific estimates in Table 26, though

not necessarily in all specifications) and a good bit higher than many of the "mixed"

cases. This is consistent with the complementarities notion. There is a perfectly

workable cluster of low-low-low, old-style firms. They have internal consistency in their

mix of complements, what Milgrom and Roberts (1990) call a "coherent combination" of

practices. Further, this result argues against heterogeneity arguments as an explanation

for the results. While some unobserved firm level shock (free cash flow is the obvious

one) could yield positive effects on human capital, IT and productivity at the same time,

it is difficult to explain why firms with both low IT and low HK have higher productivity

than those with one but not the other.50

Similar results follow from specifications with fewer interaction terms. Low-low

tends to be a relatively high productivity outcome when we treat only two potential

complements at a time (Table 27). Those firms in the off-diagonal cells (low-high or

high-low) of each table are usually less productive than those which match WO, HK and

IT. It should not be surprising that a majority of firms avoid these quadrants. The table

shows sample splits for only one measure of HK, namely SKILL. Qualitatively similar

results obtain for other measures, such as %COLL, and for related specifications with the

output elasticities varying between high and low IT firms. All these results show that we

are not simply projecting the positive effect in the positive quadrant into the negative

quadrant.51 However, a few attempts at finer subdivisions of the sample typically

increased the noise-to-signal ratio beyond levels that merit reporting.

50 One can, of course, construct a more intricate theory to explain the results. In this case, there

could be correlated shocks to the productivities of the factors at the firm level that are confined to certainranges of use of the factors in different ways in different firms. In empirical science, there is always analternative explanation of this form; this one is distinctly pre-copernican in structure.

51 This economic argument is one that Athey and Stern (1998) work to make more structural byusing a switching regimes model of firm choice between distinct clusters.


6.5 What managers believe about IT, human capital and organizationaldecentralization in these firms

Managerial opinions about the sources of firm decisionmaking and the effects of

different decisions offer a very different kind of evidence. This typically comes in the

form of anecdotes, case studies, and interviews.52 We can go farther in the present

instance, as our survey asks managers' opinions on the effects of information technology

on work. Both the average response and the correlations of these managerial opinions

with actions reinforce the evidence from the earlier sections.

In our survey, we had managers rate the extent to which computers have changed

skill levels, autonomy, monitoring and routinization of work for production workers. We

found that the responding managers at our sample firms overwhelmingly believed that IT

increased the need for skilled workers, rather than substituting for them (Table 28, first

row). Similarly, a clear majority reported that they thought that IT tended to increase

workers' autonomy. Notably, many of the same managers who reported that IT increased

worker autonomy also reported that it improved management’s ability to monitor

workers.53 Fewer reported that computers routinize work.

We go on to examine the correlation of these managerial responses with various

measures in the rest of Table 28. Of the four potential effects of computers, the increase

in skill requirements is clearly the one that is most strongly associated with the IT-WO-

HK cluster. Firms where more employees used computers were significantly more likely

to report that IT increased the skill requirements of their workers. Similarly, firms with

highly skilled and educated workers in their workforces were significantly more likely to

report that IT further increased the skill requirements of their workers. Finally,

decentralized work organization is also correlated with this perception (Table 28). This

pattern strongly suggests that the correlation among IT, skills and decentralization is due

52 In earlier work, we have conducted case studies and interviews inside many of the firms in our

present sample, which is an important part of the background to our interpretation of the econometricevidence.

53 In fact, a principal components analysis suggests that there are two dimensions characterizingthe effects of computers on workers: skills and worker autonomy represent one component whilemonitoring and routinization represent the other. There is no contradiction here, as one purpose of amonitoring technology may be to permit decentralization of decision-making.


to a conscious decision in some firms to match these practices with one another and that

this effort is still in the front of managers' minds in these firms.

While it is the change skill column that stands out in Table 28, some other

relationships are also worth noting. Decentralization (WO) is predicted not only by IT

changing skill requirements but also by IT increasing autonomy. IT changing monitoring

and IT changing skill requirements both predict high levels of human capital (measured

by (SKILL). The professionals variable, a strong measure above, is little predicted by

any of these variables, though it is negatively correlated with the view that computers

cause routinization. Perhaps the most surprising thing in this table is the plethora or zeros

in the computer capital rows. With the exception of changing skills, and to a far lesser

extent, changing autonomy, managers' opinions of computers' effects do not predict

computerization. While managerial opinion on the effects varies, IT investment is largest

in the firms where managers agree with the conclusions of this paper.

In summary, managers generally perceive IT as increasing skill requirements and

worker freedom, and this is especially true in firms that are skill-intensive, IT-intensive

and using decentralized workplace organization. Both the direct assessment of the

typical manager in our sample and the correlations between their assessments and firm

actions support the multiway complementarity theory.

7 Assessment and Implications

7.1 Assessment of the evidenceThe correlations, the factor demand equations, the production function estimates

and the managers' stated intentions are all consistent with the hypothesis that IT, HK and

WO are mutually complementary. How effective are they in rebutting alternative

theories of the correlations?

The problems with evidence of this form are well understood. If, as we said

above, there are common unobserved causes of all the complements, such as free cash

flow or fads, that could cause the correlations. Non-productivity reverse causation

interpretations are also a potential problem. The classical solutions to all these are also

well known -- find a set of clearly exogenous variables which separately shifts each

input's firm-specific supply curve, or use panel data to remove time-persistent common


causes. We are able to do only a limited version of the latter approach, and virtually none

of the former. There is still a surprising amount of information in the tables that we have

seen so far for the relevant hypotheses.

We can quickly discard the "fad" story on the production function evidence,

though the correlations and the short-run demand functions are consistent with it. The

cluster of inputs that go together are clearly related to increases in output or price

(perhaps because of service quality.) Thus they cannot be merely an inefficient fad.54

We have also accumulated a series of pieces of evidence that the "free cash flow"

heterogeneity theory does not explain these results. Perhaps the most compelling of these

is that WO and mainframes are very odd ways to consume a managerial rent (though HK

is clearly a likely one.) There are two problems with WO as an indicator of free cash

flow. First, it is decentralization, not centralization to managers, of power. Second, most

of WO is ∆WO, and organizational changes like that are notoriously difficult and

unpleasant. There are similar problems with our indicators of IT as a way to consume a

managerial rent. First, mainframes and servers make terrible toys. (Any manager who

thinks these are fun to buy will also probably think the change-management consultants

who help with ∆WO are a fun experience and is a strong candidate for institutional care.)

Second, we are not seeing a general tendency to invest, as non-IT capital is excluded

from the cluster of complements. Accordingly, any heterogeneity theory needs to

indicate why these specific kinds of capital are favored. Finally, there is the apparent

efficiency of low-low-low combinations of the three factors, a strong argument against

heterogeneity.

In fact, the efficiency of both low-low-low and high-high-high argues against any

heterogeneity theory, not just the free cash flow one. Similarly, the failure of non-IT

capital to appear as a false "complement" argues against any heterogeneity theory with a

shock to demand as its driver. Further, since the results survive controls for industry, any

heterogeneity theory has to explain why there has been a shock to the marginal value

product of these particular factors of production that hits firms heterogeneously within

industries. That is evidence against demand-side shock theories in general.


We also have considerable evidence against a nonproductivity alternative

direction of causation argument at the firm level. This stands in contrast to the

compelling evidence for a nonproductivity reverse causation argument at the individual

worker level. Individual workers who use computers do command higher wages

(Krueger (1993)). Yet it appears that both the higher wages and the computers may be

given to workers who are already particularly productive, rather than (or in addition to)

the computer making the worker more productive (DiNardo and Pischke (1997)). The

firm-level equivalent would be that firms with particularly high levels of HK got, for

some reason not related to production complementarities, WO and IT. The analogy to the

individual-worker theory is obviously false; WO and mainframes make poor "rewards"

for the reasons we just argued. Further, our partially dynamic IT factor demand

equations suggest that it is not so much that high HK firms are moving towards the other

two complements. Instead, firms that have adopted the new WO are the ones who are

most clearly moving toward the new system. Finally, as noted above, at the firm level we

have more direct evidence on the productivity of using the factors together.55

Last but not least, we need not treat the firms and their managers entirely as black

boxes. We had the opportunity to speak with many of the managers in our sample firms

(and in other firms) and we have some quantitative data on their stated managerial

perceptions and goals. While there is no unanimity of opinion, the majority perceives IT,

WO and HK to be complementary. Those firms with managers that have the strongest

beliefs on this point are furthest along in adopting the new system. This evidence

buttresses our causal interpretation of the econometric models. Any alternative theory

would face the additional hurdle of contradicting managers' stated perceptions and goals.

In summary, our empirical evidence provides very considerable support for the

multiway complementarity theory, and considerable evidence against the alternatives.

54 We exclude the fancier theory of a fad whose incidence is somehow confined to successful

firms. That is, of course, possible, if peculiar.55 Of course, there could be an unobserved factor causing both the positive outcomes and the

reverse causation. But that would be another variant of the heterogeneity story that we have alreadyrebutted.


7.2 ImplicationsStrong complementarity among human capital, new forms of workplace

organization, and IT is a finding with important implications for the whole economy. The

three complementary factors play different roles.

(IT) The price of IT has been falling rapidly and continues to fall. This shifts

firms in the direction of using all three complements. This is the route by which IT and

the workplace organization changes it enables become skill-biased technical change.

(HK) Supply curves for both more and less skilled workers are rising. The IT-

driven skill-biased technical change has driven changes in relative wages, hours, and

earnings.

(WO) Finally, the third complement, workplace organization, is characterized by

very substantial adjustment costs, as is the typical purpose of using the three

complements, changes in output quality. Firms must co-invent the specific organizational

changes and specific new product qualities that work in their particular circumstances, a

nontrivial challenge.

Computer and other information technology prices have been falling steadily,

tending to raise demand not only for IT Capital but also for the invention of new

organizational forms and for new services or new levels of quality. The demand for all

three is by no means automatic, however, because of those substantial adjustment costs.

At any given moment, we expect to see variety in a cross section of firms in the degree to

which they have adopted the three complements. Leading firms will have high levels of

all three complements; trailing firms will have low levels of all three complements.

Some firms will be further along on some dimensions than on others due to mistakes or

random factors that influence their adoption. Rising wages of highly skilled labor also

slow the diffusion of the cluster of complements (and presumably offer a powerful

incentive for inventing ways to use IT and organizational forms that economize on skill.)

Neither of these counter-effects, however, reverses the trend to the cluster of

complements. They simply slow it, in large measure by limiting the number of firms

using the cluster of complements.

We therefore believe that the current cross-firm variety in adoption of the three-

way cluster of innovations embodies a forecast of the future. An easy prediction is that


IT prices will continue to fall and IT performance will continue to improve (notably in

the data networking area). As a result, more and more firms will find the cluster of

complements worth adopting. In parallel, firms that are now experimenting or unsure

will increasingly overcome the costs and delays associated with the organizational

change. Over time, therefore, more firms will come to look like those using the cluster of

complements: IT-intensive, changed workplace organization, and high human capital.

The aggregate implication is that the recent tendency toward skill-biased technical

change will continue.

8 ConclusionIn this paper, we combine and extend two distinct research streams. One,

developed primarily by labor economists, links IT to the increased demand for skills.

Another, developed primarily by information systems researchers, links IT to changes in

organization and output quality. We hypothesize that the increased demand for skilled

labor is related to a particular cluster of technological change involving not only

increased use of IT, but also changes in workplace organization and changes in product

and service quality. IT is most effective when combined with organizational change and

that the most common combinations of IT-enabled organizational change increase the

relative demand for skilled workers.

We find that the empirical evidence is broadly consistent with our hypothesis. In

particular, our analysis of firm-level data suggests that IT use is correlated with increases

in the demand for various indicators of human capital and workforce skills. IT use is also

correlated with a pattern of work organization involving more decentralized decision-

making and greater use of teams. The combination of IT and work organization is a

much better predictor of the demand for human capital than is IT alone. Conversely, both

workplace organization and human capital are found to be very good predictors of the

firm-level demand for IT. Finally, increases in firms’ IT capital stock are associated with

the greatest increases in output in firms which also have high levels of human capital or

decentralized work organization, or both. All of these empirical facts are consistent with

the widespread perception among the managers in our sample (especially those in IT-


intensive and human capital-intensive firms) that IT increases the skill demands of work

and the autonomy of workers.

While it is difficult, both theoretically and empirically, to sort out the causality

among the cluster of factors which we identify, it is clear that the combination of

computerization, workplace organization and increased demand for skilled workers are

closely related. We conclude that the specific mechanisms of skill-biased technical

change can be better understood by more closely studying the nature of IT-enabled

organizational change in firms.


9 Bibliography

Applegate, L., Cash, J. and Mills, D. Q. (1988), “Information Technology andTomorrow’s Manager,” Harvard Business Review, November-December, pp. 128-136.

Athey, S. and S. Stern, (1998) "An Empirical Framework for Testing Theories aboutComplementarities in Organizational Design". NBER working paper 6600.

Attewell, P. and J. Rule (1984) “Computing and Organizations: What We Know andWhat We Don't Know.” Communications of the ACM 27(December): 1184-1192.

Autor, David, Lawrence F. Katz and Alan B. Krueger. (1997) “Computing Inequality:Have Computers Changed the Labor Market?” revised, March.

Barras, Richard. (1990) “Interactive Innovation in Financial and Business Services: TheVanguard of the Service Revolution,” Research Policy No. 19.

Bartel, A. and Lichtenberg, (1987) “The Comparative Advantage of Educated Workers inImplementing New Technology”, Review of Economics and Statistics, 64 1-11.

Berman, E. J. Bound and Z. Griliches (1994) “Changes in the Demand for Skilled Laborwithin U.S. Manufacturing Industries,” Quarterly Journal of Economics, May, 109,367-98.

Berndt, E. R., C. J. Morrison and L. S. Rosenblum (1992). "High-Tech Capital,Economic Performance and Labor Composition in U.S. Manufacturing Industries: AnExploratory Analysis." MIT Working Paper 3414EFA.

Bound, John and Johnson, George. (1992) “Changes in the Structure of Wages in the1980s An Evaluation of Alternative Explanations,” American Economic Review82,3: 371-392.

Bresnahan, T. (1997) “Computerization and Wage Dispersion: An AnalyticReinterpretation.” Mimeo. Stanford University.

Bresnahan, T. and S. Greenstein (1997) “Technical progress and co-invention incomputing and in the uses of computers,” Brookings Papers on Economic Activity,Micro, pp. 1-83.

Brynjolfsson, E. and L. Hitt (1995). “Information Technology as a Factor ofProduction: The Role of Differences Among Firms.” Economics of Innovationand New Technology. 3 (4): 183-200.

Brynjolfsson, E. and L. Hitt (1996). “Paradox Lost? Firm-level Evidence on the Returnsto Information Systems Spending” Management Science (April).

Brynjolfsson, E. and L. M. Hitt (1997). "Information Technology and OrganizationalDesign: Evidence from Firm-level.” MIT Sloan School Working Paper.

Brynjolfsson, E. and Mendelson, H. (1993) “Information Systems and the Organizationof Modern Enterprise.” Journal of Organizational Computing. December.


Brynjolfsson, E., A. Renshaw and M. V. Alstyne. (1997) “The Matrix of Change,” SloanManagement Review, Winter.

Brynjolfsson, E. and S. Yang (1997) “The Intangible Costs and Benefits of ComputerInvestments: Evidence from Financial Markets”, Proceedings of the InternationalConference on Information Systems, December.

Crowston, Kevin and Malone, Thomas W. (1988), “Information Technology and WorkOrganization,” Chapter 49 in: M. Helander, ed., Handbook of Human-ComputerInteractions. Elsevier Science, Amsterdam, pp. 1051-1070.

Davenport, Thomas H., and James E. Short. (1990) “The New Industrial Engineering:Information Technology and Business Process Redesign.” Sloan ManagementReview 31.4: 11-27.

David, P. A. (1990). “The Dynamo and the Computer: A Historical Perspective on theModern Productivity Paradox.” American Economic Review Papers and Proceedings80(2 (May)): 355-361.

DiNardo, John and Jörn-Steffen Pischke. (1997) “The Returns to Computer UseRevisited: Have Pencils Changed the Wage Structure Too?” Quarterly Journal ofEconomics 1 1 2 (February).

Gottschalk, Peter. (1997) “Inequality, Income Growth, and Mobility: The Basic Facts,”Journal of Economic Perspectives 11, 2 (Spring): 21-40.

Gottschalk, Peter, and Timothy M. Smeeding (1997) "Cross-National Comparisons ofEarnings and Income Inequality" Journal of Economic Literature, (June), Vol.XXXV, No. 2.

Griliches, Z. (1969). “Capital-Skill Complementarity.” The Review of Economics andStatistics 51(4): 465-468.

Hall, B. H. (1990). The Manufacturing Sector Master File: 1959-1987, Documentation.NBER Working Paper 3366.

Hamermesh, Daniel S., Labor Demand Princeton, NJ: Princeton University Press, 1993.

Hammer, M. (1990). “Reengineering Work: Don't Automate, Obliterate.” HarvardBusiness Review (July-August): 104-112.

Holmstrom, B. and P. Milgrom (1991). “Multitask Principal-Agent Analyses: IncentiveContracts, Asset Ownership, and Job Design.” Journal of Law, Economics andOrganization 7 (Spring): 25-52.

Holmstrom, B. and P. Milgrom (1994). “The Firm as an Incentive System.” AmericanEconomic Review 84(4, September): 972-991.

Howell, C., and E. Wolff, (1991) "Trends in the Growth and Distribution of Skills in theU.S. Workplace," Industrial and Labor Relations Review, April.


Huselid, M. (1994). “The Impact of Human Resource Management Practices onTurnover, Productivity, and Corporate Financial Performance.” Academy ofManagement Journal.

Ichniowski, C. and T. A. Kochan (1995). “What Have We Learned From WorkplaceInnovations?” MIT Sloan School Working Paper.

Ichniowski, C. Thomas A. Kochan, David Levine, Craig Olson and George Strauss,(1996) "What works at work: Overview and Assessment", Industrial Relations, Vol.35, No. 3 (July). P. 299-333.

Ichniowski, C., K. Shaw and G. Prennushi, "The Effects of Human ResourceManagement Practices on Productivity." American Economic Review, June, 1997.

Ito, Harumi, Essays on Investment Adjustment Costs, Ph.D. Dissertation, StanfordUniversity, September 1996.

Jensen, M. and W. Meckling, (1992) “Knowledge, Control and OrganizationalStructure”, in Contract Economics, L. Werin and H. Wijkander, Editors., BasilBlackwell: Cambridge, MA. p. 251-274.

Johnson, George. (1997) “Changes in Earnings Inequality: The Role of Demand Shifts,”Journal of Economic Perspectives, 11, 2 (Spring): 41-54.

Juhn, Chinhui, Kevin M. Murphy and Brooks Pierce. (1993) “Wage Inequality and theRise in Returns to Skill,” Journal of Political Economy 101 June: 410-42.

Karoly, Lynn A. (1996) "Anatomy of the U.S. Income Distribution: Two Decades ofChange," Oxford Review of Economic Policy 12, 1: 76-95.

Kemerer, C. F. and G. L. Sosa, (1991) “Systems Development Risks in StrategicInformation Systems.” Information and Software Technology 33(3): 212-223.

Krueger, Alan B. (1993) “How Computers Have Changed the Wage Structure? Evidencefrom Micro Data,” Quarterly Journal of Economics (February): 33-60.

Krugman, Paul, and Robert Z. Lawrence. (1993) “Trade, Jobs, and Wages,” NBER WPNo. 4478, September.

Katz, Lawrence F. and Kevin M. Murphy. (1992) “Changes in Relative Wages, 1963-1987: Supply and Demand Factors,” Quarterly Journal of Economics 107 (February):35-78.

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

Levy, Frank and Richard Murnane. (1996) “With What Skills Are Computers aComplement?” American Economic Review, 86 (May): 258-262.

Lichtenberg, F. R. (1995). “The Output Contributions of Computer Equipment andPersonal: A Firm-Level Analysis.” Economics of Innovation and New Technology 3:201-217.


Malone, T. and Rockart, J. (1991), “Computers, Networks and the Corporation,”Scientific American, Vol. 265(3): 128-136.

Mendelson, Haim and Ravi Pillai, "Clockspeed and Informational Response: Evidencefrom the Information Technology Industry", Manuscript, 1998.

Milgrom, P. and J. Roberts (1990). “The Economics of Modern Manufacturing:Technology, Strategy, and Organization.” American Economic Review 80(3): 511-528.

Murphy, Kevin M. and Welch, Finns. (1993) "Lessons from Epirical Labor Economis:1972-1992. Inequality and Relative Wages," American Economic Review, 83, 2.

Oi, Walter. 1962. "Labor as a Quasi-Fixed Factor," Journal of Political Economy 70(6):538-555

Orlikowski, W. J. (1992). Learning from Notes: Organizational Issues in GroupwareImplementation. in Conference on Computer Supported Cooperative Work. J. Turnerand R. Kraut. Toronto, Association for Computing Machinery: 362-369.

Osterman, P. (1994). “How Common is Workplace Transformation and Who Adopts It?”Industrial and Labor Relations Review 47(2): 173-188.

Scott Morton, Michael S. (Ed.) (1991), The Corporation of the 1990s: InformationTechnology and Organizational Transformation, Oxford University Press.

Simon, Herbert A., "Applying Information Technology to Organization Design." PublicAdministration Review, (1973), p. 268-78.

Simon, Herbert A., "Designing Organizations for an Information-Rich World," inComputers, Communications, and the Public Interest, Martin Greenberger, ed.Baltimore: The Johns Hopkins Univ. Press, 1971, pp. 40-41.

Steiner, T. D. and Teixeira, D. B. (1991), Technology in Banking: Creating Value andDestroying Profits, Dow Jones Irwin, Homewood, IL.

Wolff, Edward N. (1996) "The Growth of Information Workers in the U.S. Economy,1950-1990: The Role of Technological Change, Computerization, and StructuralChange," The C.V. Starr Center for Applied Economics, New York University.

Williamson, Oliver E. 1979. Transaction-Cost Economics: The Governance ofContractual Relations. Journal of Law and Economics 22 (October):233-261.

Zuboff, Shoshana (1988) In the Age of the Smart Machine: The Future of Work andPower, New York: Basic Books.


10 Appendix: Variables and Data Construction

This appendix is divided into three sections. Section 10.1 contains the survey questionsand variable definitions for items obtained from the organizational practices survey.Section 10.2 contains the data construction procedures for variables obtained fromComputer Intelligence Infocorp (CII). Section 10.3 contains the data constructionprocedures for variables derived from Compustat. The formulas for all the compositevariables are in Table 5. Table 4 describes the subsamples used for various analyses.

10.1 Organizational practices survey

The organizational practices survey was designed to be administered over the phone andtook, on average, 30 minutes for the respondent to complete. The questions relevant tothis analysis are shown below.

10.1.1 Data Validation Questions

(PWPCT). To the nearest 10%, what percentage of your firm’s employees are productionworkers?

_____%

(PWSAME). In terms of the whole firm, please estimate to the nearest 20% thepercentage of all production employees covered by the same human resource practices asthose at your most typical establishment? Would you say 20%, 40%, 60%, 80% of100%?

______%

I.B. Characteristics of Production Workers

(PRBL, PRCL, PRPF). Can you name three titles of employees you consider to beproduction workers at that establishment:_____________________________________________________________________

Variables are coded by determining the fraction (out of 3) of jobs listed that are bluecollar clerical or professional.


Now I’d like you to rate the working conditions for production workers at your mosttypical establishment. First, would you rate...

VeryHigh

ModeratelyHigh Medium

ModeratelyLow

Very Low

a. (ICDA) the averagelevel of individual controland decision making thatworkers have in conductingtheir own work?

o o o o o

b. (SKILL) How wouldyou rate the average levelof skill required to performproduction work?

o o o o o

c. (BROAD) How wouldyou rate the amount ofdiversity associated withthe production work itself?

o o o o o

d. (COMP) How wouldyou rate the level ofcomputerization of thework tasks?

o o o o o

Next I want to know about the arrangement between workers and managers in theconduct of the work.

ExclusivelyWorkers

MostlyWorkers Equally

MostlyManagers

ExclusivelyManagers

a. (XPACE) Who sets thepace of work?

o o o o o

b. (XSCHED) Whoschedules productionwork?

o o o o o

c. (XDIST) Whodistributes this workamong the workers?

o o o o o

d. (XMETH) Who decideshow the tasks should beaccomplished?*

o o o o o

e. (XPROB) Who dealswith difficult situations inproduction?

o o o o o

f. (XCUST) Who deals o o o o o

* The variables METH and PACE were the original variables from survey one, coded 1-3XMETH and XPACE are from survey 2 and 3 and were recoded onto the 3 point scale (1=1, 2=1.5, 3=2,4=2.5, 5=3).


with customers in routinesituations?g. (XCOMPL) Who dealswith customers overproblems or complaints?

o o o o o

Next I would like you to tell me about how computerization has affected productionworkers?

GreatlyIncreased

SlightlyIncreased

SlightlyDecreased

GreatlyDecreased

NoEffect

a. (CSKILL)Would yousay that computerizationhas greatly increased,slightly increased, slightlydecreased, greatlydecreased or had no effecton the skill requirements ofproduction jobs?

o o o o o

b. (CROUTI) Would yousay that computerizationhas greatly increased,slightly increased, slightlydecreased, greatlydecreased or had no effecton the repetitiveness ofwork?

o o o o o

c. (CMONIT1) Using thesame categories, hascomputerization affectedthe need to monitor work?

o o o o o

d. (CMONIT2) Hascomputerization affectedthe ability to monitorwork?

o o o o o

e. (CFREED) Hascomputerization affectedworker autonomy?

o o o o o

The next set of questions is about the extent to which workers are provided with decisionmaking and training opportunities.

(SMTEA). Does your firm use “self-managing teams”?____ (yes, no) [If no, skip to next question]


Would you say your firm uses self-managing teams very heavily, heavily, moderately orslightly?

___ Very heavily___ Heavily___ Moderately___ Slightly

(QUALC). Does your firm use “employee involvement groups”?____ (yes, no) [If no, skip to next question]

Would you say your firm uses employee involvement groups very heavily, heavily,moderately or slightly?


(TEAMB). Does your firm use team-building or group cohesion techniques?____ (yes, no) [If no, skip to next question]

Would you say your firm uses these techniques very heavily, heavily, moderately orslightly?


(XTRAI). Does your firm cross-train workers?____ (yes, no) [If no, skip to next question]

Would you say your firm cross-trains very heavily, heavily, moderately or slightly?___ Very heavily___ Heavily___ Moderately___ Slightly

Now I would like to ask about the promotion of workers. How important are thefollowing factors when promoting production workers?

Extremely Very Somewhat Not too Not at alla. (PROTE) Would yousay teamwork is extremelyimportant, very important,somewhat important, nottoo important or not at all

o o o o o


important?b. (PROSK) Howimportant is skillacquisition?

o o o o o

How important are the following criteria when conducting pre-employment screens fornew production workers?

Extremely Very Somewhat Not too Not at alla. (SCNED) Would yousay educationalbackground is extremelyimportant, very important,somewhat important, nottoo important or not at allimportant?

o o o o o

Please tell me about what percentage of your production workforce has a high schooleducation or less, what percentage has some college or technical school, and whatpercentage has a college degree?

_____% High school or less (%HSED)_____% Some college or technical school (%SCED)_____% College grad (%COLL)

(TRAIN). What percentage of production workers received any work-related trainingoff-the-job during the last 12 months? (“Off-the-job” training includes classroomtraining, or courses or seminars apart from regular work activities.)

_____ %

I. C. Firm-Wide Characteristics

Next I want to ask you to estimate how various jobs are distributed across the entire firm,and about any large scale changes that might have happened recently. Some of thesequestions are quite detailed. If you are not sure of an answer please feel free to give youbest guess.

Approximately what percentage of jobs in your firm are in each of the followingcategories? I will read the categories first and then ask you the percentage of each?

Clerical ______% (%CL)Unskilled blue-collar workers ______% (%SK)Skilled blue-collar workers ______% (%US)Managers and supervisors ______% (%MG)Non-managerial professionals ______% (%PF)Total 100%


(%GP). What fraction of your employees use a general purpose computer in the courseof their regular work?

_____%

(%EMAIL). What fraction of your employees use electronic mail (e-mail) in the courseof their regular work?

_____%

10.2 Computer Intelligence variables

IT Capital (ITCAP). We take total purchase value of computer equipment as reportedby Computer Intelligence Infocorp. (CII) and deflate it using an extrapolation ofGordon’s (1990) deflator for computers (price change -19.3% per year). The totalpurchase value represents the current market value of mainframes, minicomputers andperipherals as well as personal computers during the 1991-1994 portion of our sampleperiod. Prior to 1991, the purchase value only represented the value of mainframes,minicomputers and peripherals, but excluded personal computers.

Central Processing Power (MIPS). This variable is taken straight from the CII databaseand represents the total processing power of central processors, measured in millions ofinstructions per second (PCs are not included in this calculation).

Personal Computers (TOTPC). This is also taken straight from the CII database andrepresents the total number of personal computers in use at the firm.

10.3 Compustat-based variables

Output (OUTPUT). Total Sales as reported on Compustat [Item #12, Sales (Net)]deflated by 2-digit industry level deflators from Gross Output and Related Series byIndustry from the BEA for 1988-1992, and estimated for 1993-1994 using five-yearaverage inflation rate by industry. When an industry deflator is not available, the sectorlevel producer price index for intermediate materials, supplies and components is used(Council of Economic Advisors, 1995).

Ordinary Capital (NITCAP). This figure was computed from total book value ofcapital (equipment, structures and all other capital) following the method in (Hall, 1990).Gross book value of capital stock [Compustat Item #7 - Property, Plant and Equipment(Total - Gross)] was deflated by the GDP implicit price deflator for fixed investment.The deflator was applied at the calculated average age of the capital stock, based on thethree year average of the ratio of total accumulated depreciation [calculated from


Compustat item #8 - Property, Plant & Equipment (Total - Net)] to current depreciation[Compustat item #14 - Depreciation and Amortization]. The calculation of average agediffers slightly from the method in Hall (1990) who made a further adjustment for currentdepreciation. The constant dollar value of IT capital (as calculated above) was subtractedfrom this result. Thus, the sum of ordinary capital and IT capital equals total capitalstock.

Labor Expense (LABOR). Labor expense was either taken directly from Compustat(Item #42 - Labor and related expenses) or calculated as a sector average labor cost peremployee multiplied by total employees (Compustat Item #29 - Employees) when laborexpense was not available, and deflated by the price index for Total Compensation(Council of Economic Advisors, 1995). The average labor expense per employee wastaken from BLS data on hourly cost of workers (including benefits) for 10 sectors of theeconomy. For firms which had labor expense directly reported on Compustat which didnot include benefits (identified by Compustat Item - Labor Expense Footnote), weadjusted the labor figure by multiplying reported labor expense by the totalcompensation/wages ratio for each sector as reported by BLS.

Employees (EMPLOY). Number of employees was taken directly from Compustat(Item #29 - Employees). No adjustments were made to this figure.

Materials (MATL). Only used in computations. Materials was calculated bysubtracting undeflated labor expenses (calculated above) from total expense and deflatingby the industry level output deflator. Total expense was computed as the differencebetween Operating Income Before Depreciation (Compustat Item #13), and Sales (Net)(Compustat Item #12).

Value-Added (VA). Computed from deflated Output (as calculated above) less deflatedMaterials.

Sector Dummy Variables. The industry controls used in this most analyses in this papercorrespond to an intermediate level between 1-digit and 2-digit SIC codes. Based on thereported primary SIC code on Compustat we construct the following variables:

Mining /Construction (MI) – SIC 11xx - 20xxProcess Manufacturing (PR) – SIC 26xx, 28xx and 29xxOther Non-Durable Manufacturing (MN) – SIC 20xx – 23xx and SIC 27xxHigh Technology Manufacturing (HI) – SIC 36xx – 38xx and 3571 (computers)Other Durable Manufacturing (MD) – SIC24xx-25xx, 30xx-35xx (except 3571) and 39xxTransportation (TP) – SIC40xx-47xxUtilities (UT) – SIC48xx-49xxTrade (TR) – SIC50xx-59xxFinance (FI) – SIC 60xx-69xxOther Services (SR) – SIC70xx-79xx


11 Tables and Figures

Table 1: Organizational Practice Survey Variables

Range Variable N Mean Std. Dev.

Team-Based Work OrganizationUse of Self-Managing Teams 1-5 SMTEA 345 2.11 1.13Use of Employee Involvement Groups 1-5 QUALC 345 2.85 1.21Use of Team Building Activities 1-5 TEAMB 345 2.95 1.17Promote for Teamwork 1-5 PROTE 345 3.59 0.95Breadth of Jobs 1-5 BROAD 345 3.25 0.99

Individual Decision AuthorityWho Decides Pace of Work (3=workers) 1-3 PACE 345 1.33 0.37

Who Decides Method of Work (same) 1-3 METH 345 1.39 0.38

Human Capital (Levels)Skill Level of Work 1-5 SKILL 345 3.60 0.86Education Level 1-5 EDUC 345 2.48 0.66Workers w/ High School or Less Ed. 0-100% %HSED 263 59.3% 27.8%Workers with Some College Education 0-85% %SCED 263 23.3% 17.5%Workers Completed College 0-100% %COLL 263 17.4% 21.0%

Human Capital InvestmentPre-Employment Screen for Education 1-5 SCNED 345 3.31 0.89Training (% workers involved) 0-100% TRAIN 345 48.0% 36.1%Cross-train Workers 1-5 XTRAI 345 3.16 0.98

Production Worker CompositionBlue Collar (fraction of jobs listed) 0-100% PRBL 345 61.9% 46.2%Clerical (fraction of jobs listed) 0-100% PRCL 345 31.4% 43.4%Professional (fraction of jobs listed) 0-100% PRPF 345 4.6% 17.5%

Workforce CompositionUnskilled Blue Collar (%) 0-95% %US 337 18.4% 21.4%Skilled Blue Collar (%) 0-85% %SK 337 24.7% 21.1%Clerical (%) 0-80% %CL 337 19.4% 17.6%Professionals (%) 0-90% %PF 337 20.7% 16.8%Managers (%) 0-50% %MG 337 16.8% 8.5%Source: Authors' Survey. More on data definitions in section 10.1


Table 2: IT Variables

Variable N Mean Std. Dev.

CII SurveyLog(IT Capital) LITCAP 333 3.07 1.66Total MIPS (Millions of instructions/sec.) MIPS 333 2,624 8,737Total PCs TOTPC 333 4,560 10,997

Organizational SurveyDegree of Computerization of Work COMP 343 3.28 1.11% Workers using General Purp. Computers %GP 290 53.0% 33.8%% Workers using E-mail %EMAIL 290 31.0% 32.2%Source: CII. More on data definitions in section 10.1

Table 3: Factors of Production

Variable N Mean$MM

log(Mean) Std. Dev.

1994 Cross Sectionlog(Output) LOUTPUT 311 2,251.38 7.72 1.00log(Value Added) LVA 311 890.25 6.79 1.02log(Labor Expense) LLABOR 311 477.42 6.17 1.09log(Non-IT Capital) LNITCAP 311 1,693.58 7.43 1.44log(Employment) LEMPLOY 311 12.84 2.55 1.09log(IT Capital) LITCAP 311 24.66 3.21 1.49

All Years Pooledlog(Output) LOUTPUT 2466 2,431.59 7.80 0.99log(Value Added) LVA 2466 946.53 6.85 1.02log(Labor Expense) LLABOR 2466 530.33 6.27 1.08log(Non-IT Capital) LNITCAP 2466 1,771.53 7.48 1.39log(Employment) LEMPLOY 2466 14.41 2.67 1.09log(IT Capital) LITCAP 2466 11.86 2.47 1.42Source: Compustat. More on data definitions in section 10.3


Table 4: Samples

There are several different subsamples used in the analysis due to missing survey data orrestrictions to a single year. The major samples are outlined in this table.Sample Where Used DescriptionHKA Regressions –1994 data only

Table 17, Table 18,Table 19, Table 20,Table 21, Table 22

250 complete observations of HKA,SKILL, 4th lagged IT, WO and IT Effects(CONTROL, UPGRADE). Restricted to1994 cross section only.

Production Factors,WO, HKI and Skill(Sample 1)

Table 12, Table 14,Table 23, Table 24,Table 25, Table 26,Table 27

2225 complete observations of VA,EMPLOY, ITCAP, NITCAP, LABOR,WO, HKA, PR_PROF, PR_CLER; 1994-5cross section of organizational practicesdata matched to 1987-1994

Sample 1 withEducation Controls(Sample 2)


1413 complete observations of variables insample 1 plus %COLL and %SCED;1994-5 cross section of organizationalpractices data matched to 1987-1994

Sample 1 withWorkforceComposition(Sample 3)


1854 complete observations of variables insample 1 plus %PF, %CL, %MG, %US,%SK and %IW; 1994-5 cross section oforganizational practices data matched to1987-1994

Completeorganizationalvariables(Sample 4)

Table 13 1331 complete observations; all variablesthat appear in samples 1-3; 1994-5 crosssection of organizational practices datamatched to 1987-1994

OrganizationalPractices

Table 1 345 observations with complete data onWO and HKA

Education Table 1 263 observations with complete data on%COED, %SCED

WorkforceComposition

Table 1 337 observations with complete data onworkforce composition (%MG, %PF,%SK, %US, %CL)

CI IT data Table 2 333 observations with complete data onITCAP, TOTPC, MIPS

Production Data Table 3 2466 observations (311 in 1994) withcomplete data on VA, ITCAP, LABOR,NITCAP;

Survey-based ITData

Table 2 343 observations of COMP; 293 withcomplete data on COMP, %GP, %EMAIL


Table 5: Definitions of Composite Variables

Description Variable FormulaWork Organization WO SMTEA+QUALC+TEAMB+PROTE+

PACE+METHHuman Capital Investment HKI SCNED+TRAIN+XTRAIComposite DecisionAuthority

DA7 PACE+XSCHED+XDIST+METH+XPROB+XCUST+XCOMPL

Information WorkerPercent*

%IW %CL+%PF+%MG

All composites are formed by summing standardized values (except as noted by *). Thesum is then restandardized to mean 0, unit variance.

Table 6: Correlations between Composite Measures of IT, Human Capitaland Organization

Measure ITCAP WO SKILL %COLL %PF

Computer Capital (ITCAP) 1Work Organization (WO) .18*** 1Worker Skill (SKILL) .12* .28*** 1Percent College (%COLL) .05 .34*** .17*** 1Percent Professional (%PF) .30*** .21*** .06 .21*** 1

Spearman partial rank order correlations controlling for industry (9 sector dummyvariables), employment (EMPLOY) and production worker composition (PRBL, PRCL).N=214-313, due to non-response and some measures limited to second and third wavesurveys. Key: * - p<.1, ** - p<.05, *** - p<.01; test is against the null hypothesis thatthe correlation is zero.


Table 7: Correlations between IT and Human Capital

Measure(scale in parenthesis) %GP

%EMAIL COMP ITCAP MIPS TOTPC

Skills/Education (N=371)Skill Levels (SKILL) .20*** .29*** .36*** .09 .15*** .13**Education (EDUC) .15** .26*** .24*** .13 .08 .01

Education Distribution (N=237)High School Education (%HSED) -.28*** -.37*** -.32*** -.11* -.18*** -.15**College Graduate (%COLL) .27*** .36*** .28*** .07 .14 .05

Skill Acquisition (N=370)Training (TRAIN) .17*** .14** .20*** .14** .15*** .14**Screen for Education (SCNED) .11* .15*** .28*** .16*** .18*** .21***

Skill Acquisition (control forSKILL and EDUC)Training (TRAIN) .14** .10* .19*** .15*** .14*** .14***Screen for Education (SCNED) .07 .08 .15** .16*** .15*** .19***Spearman partial rank order correlations controlling for industry (9 sector dummyvariables), employment (EMPLOY) and production worker composition (PRBL, PRCL).Sample size varies due to non-response.Key: * - p<.1, ** - p<.05, *** - p<.01; test is against the null hypothesis that thecorrelation is zero.


Table 8: Correlations between IT and Workforce Composition



Workforce Composition (N=303)Clerical (%CL) -.02 .04 -.05 -.08 -.01 -.03Unskilled Blue Collar (%US) -.22*** -.25*** -.17*** -.08 -.13** -.09Skilled Blue Collar (%SK) -.05 -.05 .05 .05 .09 .02Managers (%MG) .16** .13** .11* .17** .15** .10Professionals (%PF) .16** .27*** .18*** .29*** .39*** .28***Spearman partial rank order correlations controlling for industry (9 sector dummyvariables), employment (EMPLOY) and production worker composition (PRBL, PRCL).Key: * - p<.1, ** - p<.05, *** - p<.01; test is against the null hypothesis that thecorrelation is zero.


Table 9: Correlations between IT and Work Organization



Our Main MeasureDecentralization (WO) .25*** .30*** .23*** .17*** .23*** .15***

Subcomponents of WOSelf-Managing Teams (SMTEA) .17*** .27*** .18*** .17*** .22*** .20***Employee Inv. Grps. (QUALC) .19*** .19*** .08 .07 .08 .08Team Building (TEAMB) .24*** .26*** .13** .15*** .19*** .18***Promote for Teamwork (PROTE) .12** .14** .15** .02 .10* .00Pace of Work (PACE) .10* .10* .14** .04 .06 .02Method of Work (METH) .21*** .24*** .14*** .16*** .20*** .15***

Other Organizational Structure MeasuresBroad Jobs (BROAD) .20*** .17*** .24*** .07 .12** .10*Scheduling of Work (XSCHED) .20*** .20*** .22*** .11* .20*** .15**Distrib. Work to Wrkrs. (XDIST) .11* .19*** .20*** .11* .14** .12**Production Problems (XPROB) -.01 .04 .03 .07 .07 .05Customer Interaction (XCUST) .13** .10* .03 .01 .07 .03Handling Complaints (XCOMPL) .17*** .20*** .13** .04 .10 .08Composite of 7 measures (DA7) .21*** .29*** .23*** .12* .14** .16***Individual Control (ICDA) .15** .24*** .22*** .11* .15** .15**

Spearman partial rank order correlations controlling for industry (9 sector dummyvariables), employment (EMPLOY) and production worker composition (PRBL, PRCL).N=240-372, due to non-response and some measures limited to second and third wavesurveys.Key: * - p<.1, ** - p<.05, *** - p<.01; test is against the null hypothesis that thecorrelation is zero.


Table 10: Correlations between Human Capital and Teams/Decentralization

Measure(scale in parenthesis)

SKILL EDUC %COLL SCNED TRAIN

Our Main MeasureDecentralization (WO) .33*** .16*** .30*** .24*** .20***

Subcomponents of WOSelf-Managing Teams (SMTEA) .29*** .14** .25*** .22*** .16***Employee Inv. Grps. (QUALC) .18*** -.00 .10 .09 .14***Team Building (TEAMB) .20*** .02 .14** .16*** .22***Promote for Teamwork (PROTE) .12** .03 .08 .08 .06Pace of Work (PACE) .12** .20*** .27*** .17*** .03Method of Work (METH) .24*** .20*** .38*** .19*** .10*

. .Other Measures of Organizational Structure

Broad Jobs (BROAD) .20*** .16*** .14** .18*** .07Scheduling of Work (XSCHED) .17*** .05 .08 .15** .16**Distrib. Work to Wrkrs. (XDIST) .18*** .11* .10 .15** .12*Production Problems (XPROB) .14** -.04 -.01 .06 .09Customer Interaction (XCUST) .03 -.00 -.03 .02 .21***Handling Complaints (XCOMPL) .10 .10* .02 .08 .27***Composite (DA7) .20*** .07 .21*** .13** .22***Individual Control (ICDA) .47*** .09 .20*** .26*** .11*

Spearman partial rank order correlations controlling for industry (9 sector dummyvariables), employment (EMPLOY) and production worker composition (PRBL, PRCL).N=240-372, due to non-response and some measures limited to second and third wavesurveys.Key: * - p<.1, ** - p<.05, *** - p<.01; test is against the null hypothesis that thecorrelation is zero.


Table 11: Correlations between IT and Human Capital with OrganizationalControls

%GP%

EMAIL COMP ITCAP MIPS TOTPCSkills/Education (partialling WO)Skill Levels (SKILL) .12* .21*** .27*** .04 .09 .09Education (EDUC) .13** .24*** .19*** -.04 .01 -.03High School Education (%HSED) -.22*** -.34*** -.27*** -.08 -.13** -.16**College Graduate (%COLL) .19*** .30*** .20*** .02 .08 .04

Same Analysis without controlsSkill Levels (SKILL) .20*** .29*** .36*** .09 .15*** .13**Education (EDUC) .15** .26*** .24*** .13 .08 .01High School Education (%HSED) -.28*** -.37*** -.32*** -.11* -.18*** -.15**College Graduate (%COLL) .27*** .36*** .28*** .07 .14 .05Spearman partial rank order correlations controlling for industry (9 sector dummyvariables), employment (EMPLOY) and production worker composition (PRBL, PRCL).First four rows also include WO as a covariate.Key: * - p<.1, ** - p<.05, *** - p<.01; test is against the null hypothesis that thecorrelation is zero.


Table 12: IT demand as a function of human capital and organization

Dependent Variable log(ITCAP) log(ITCAP) log(ITCAP) log(ITCAP)Specification

Variable

Base Base +Education

Base+WO

Base+Education+

WO(1) (2) (3) (4)

Worker Skill(SKILL)

.0952***(.0156)

.0837***(.0224)

.0712***(.0162)

.0566**(.0231)

College Education(%COLL)

.0977***(.0240)

.0786***(.0242)

Some College(%SCED)

-.00672(.0206)

-.0138(.0206)

Decentralization(WO)

.0799***(.0162)

.0935***(.0220)

log(Value-Added)log(VA)

.724***(.0153)

.657***(.0205)

.706***(.0156)

.644***(.0207)

Controls SectorYear

Composition(PRCL, PRPF)

SectorYear


SectorYear


SectorYear


R2 56.4% 51.6% 56.8% 52.3%N 2225 1413 2225 1413Key: * - p<.1, ** - p<.05, *** - p<.01All variables standardized to mean 0, unit variance.


Table 13: IT demand as a function of human capital and organization -controlling for workforce composition

Dependent Variable log(ITCAP) log(ITCAP) log(ITCAP) log(ITCAP)Specification

Variable


Base+WO

Base+Education+

WOWorker Skill(SKILL)

.0598***(.0186)

.0772***(.0240)

.0428**(.0191)

.0539**(.0247)


.0506***(.0276)

.0393(.0276)

Some College(%SCED)

-.00342(.0218)

-.0111(.0217)


.0673***(.0185)

.0894***(.0238)


.659***(.0175)

.611***(.0218)

.645***(.0179)

.599***(.0219)

Professionals(%PF)

.142***(0205)

.111***(.0289)

.133***(.0206)

.0957***(.0291)

Clerical(%CLER)

.0121(.0231)

.0246(.0291)

.0143(.0229)

.0319(.0290)

Managers(%MGR)

.0417**(.0185)

-.0202(.0239)

.0432**(.0184)

-.0155(.0238)

Unskilled Blue Collar(%USBC)

-.0223(.0227)

-.00629(.0300)

-.0174(.0227)

-.00511(.0299)

Controls SectorYear

Composition(PRBL, PRPF)

SectorYear


SectorYear


SectorYear




Table 14: Non-IT Capital demand, same specification as Table 12

Dependent Variable log(NITCAP) log(NITCAP) log(NITCAP) log(NITCAP)Specification

Variable


Base+WO

Base+Education+

WOWorker Skill(SKILL)

.0200(.0127)

.0344*(.0194)

.0166(.0133)

.0352*(.0202)


-.0512**(.0208)

-.0506**(.0211)

Some College(%SCED)

.0113(.0179)

.0115(.0180)


.0115(.0132)

-.00267(.0193)


.684***(.0124)

.656***(.0179)

.682***(.0128)

.656***(.0181)

Controls SectorYear


SectorYear


SectorYear


SectorYear




Table 15: Changes in IT demand as a function of human capital andorganization - controlling for workforce composition

Dependent Variable log(ITCAP) log(ITCAP) log(ITCAP)Specification

Variable

Base+WO

Base+Education+

WO

Base+Ed.+WO w/o

WOxTimeWorker Skill x Time(SKILL x YEAR)

.000255(.00759)

-.0194(.0220)

-.00557(.0203)

College Ed. x Time(%COLLxYEAR)

.00656(.0210)

.0109(.0207)

WO x Time(WO x YEAR)

.0183***(.00766)

.0358(.0219)

Worker Skill(SKILL)

.0438***(.0191)

.0560***(.0247)

.0543***(.0247)


.0388(.0276)

.0392(.0277)


.0668***(.0186)

.0876***(.0238)

.0890***(.0238)


.647***(.0186)

.600***(.0219)

.599***(.0219)

Professionals(%PF)

.133***(.0206)

.0967***(.0291)

.0958***(.0291)

Controls SectorYear

Composition(PRBL, PRPF%MG, %CL,

%SK)%SCED

SectorYear


%SK)%SCED

SectorYear


%SK)%SCED

R2 53.7% 51.0% 50.9%N 1854 1331 1331Key: * - p<.1, ** - p<.05, *** - p<.01All variables standardized to mean 0, unit variance except interaction terms whichrepresent a standardized variable multiplied by a mean 0 time measure (units are years).


Table 16: Changes in IT Demand, Varying the IT Measure

Dependent Variable log(ITCAP) log(MIPS) log(TOTPC)Specification

Variable

Base+Education+

WO

Base+Education+

WO

Base+Education+

WO

Worker Skill x Time(SKILL x YEAR)

-.0193(.0219)

.0191(.0212)

.00812(.0214)

College Ed. x Time(%COLL x YEAR)

.00656(.0210)

-.00339(.0203)

.0167(.0204)

WO x Time(WO x YEAR)

.0358(.0219)

.0810***(.0218)

-.00147(.0214)

Worker Skill(SKILL)

.0560***(.0247)

.0414*(.0239)

.0382(.0241)


.0388(.0276)

.0629**(.0267)

-.0119(.0269)


.0876***(.0238)

.0945***(.0231)

.0869***(.0233)


.600***(.0219)

.456***(.0231)

.561***(.0214)

Professionals(%PF)

.0967***(.0291)

.139***(.0282)

.0667**(.0284)

Controls SectorYear


%SK)%SCED

SectorYear


%SK)%SCED

SectorYear


%SK)%SCED

R2 51.0% 54.1% 53.3%N 1331 1331 1331Key: * - p<.1, ** - p<.05, *** - p<.01All variables standardized to mean 0, unit variance except interaction terms whichrepresent a standardized variable multiplied by a mean 0 time measure (units are years).


Table 17: Human Capital Investment as a function of computers andorganization adding separate variables for organization and industry.

Dependent Variable HKI HKI HKI HKI HKI HKI Specification

Variable

Base Base +WO

Base+Sector

Base +Sector+

WO

Base+2-digit

Industry

Base+Industry+

WOComputerizationlog(ITCAP/EMPLOY)-4

.228***(.0618)

.174***(.0542)

.212***(.0686)

.163***(.0612)

.172**(.0732)

.150*(.0651)

Work Organization(WO)

.481***(.0542)

.449***(.0558)

.449***(.0603)

Industry Controls None None Sector Sector 2-digitSIC

2-digitSIC

R2 5.2% 28.1% 15.6% 33.7% 28.7% 43.9%N 250 250 250 250 250 250Key: * - p<.1, ** - p<.05, *** - p<.01All variables standardized to mean 0, unit variance.

Table 18: Adding Skill and Contemporary IT (Instrumented)

Dependent Variable HKI HKI HKI HKI HKI HKI Specification

Variable

Base +Skill +Sector

Base +Skill+WO+Sec

IVIT/Ew/o WO

IV:IT/Ew/ WO

IV:COMPw/o WO

IV:COMPw/ WO

Computerizationlog(ITCAP/EMPLOY)-4

.180***(.0673)

.154***(.0614)

Computerizationlog(ITCAP/EMPLOY)

.204***(.0801)

.184**(.0735)

Computerization(COMP)

1.12**(.547)

.994*(.522)


.419***(.0589)

.409***(.0594)

.314***(.0982)

Skills(SKILL)

.237***(.0629)

.0948(.0607)

.232***(.0632)

.0930(.0609)

-.178(.229)

-.240(.200)

Industry Controls Sector Sector Sector Sector Sector SectorN 250 250 250 250 250 250

Key: * - p<.1, ** - p<.05, *** - p<.01All variables standardized to mean 0, unit variance.IV: Computerization instrumented with 4th lagged log(ITCAP/EMPLOY); all othervariables considered exogenous.


Table 19: Human Capital Investment as a function of computers andorganization, using alternative measures of computers

Dependent Variable HKI HKI HKI HKISpecification

Variable

MIPS:Base+Sector

MIPS:Base+Sector

+ WO

PCs:Base+Sector

PCs:Base+Sector

+WOComputerizationlog(TOTPC/employ)-4

.156**(.0729)

.0964(.0644)

Computerizationlog(MIPS/employ)-4

.258***(.0710)

.187***(.0639)


.442***(.0558)

.454***(.0565)

Industry Controls Sector Sector Sector Sector



Table 20: Human Capital Investment as a function of computers andorganization - different managerial expectations about computers and work

Dependent Variable HKI

VariableAdding

IT EffectsComputerizationlog(ITCAP/EMPLOY)-4

.156***(.0608)


.428***(.0559)

Monitoring +Routinization(CONTROL)

-.0421(.0534)

Skill + Freedom(UPGRADE)

.147***(.0538)

Industry Controls SectorR2 35.9%N 250Key: * - p<.1, ** - p<.05, *** - p<.01All variables standardized to mean 0, unit variance.


Table 21: Changes in Demand for HK as a function of IT and othervariables, paralleling Autor, Katz and Krueger (AKK), 1997, table 8.

Dep. Variable HKI HKI HKI HKI HKI HKISpecification

Variable

AKK T.8.Col. 2

AKK T.8.Col. 3

AKK T.8.Col. 4

AKK T.8.Col. 5

Manufact.Only

ServicesOnly


.228***(.0618)

.228***(.0618)

.221***(.0621)

.255***(.0636)

.294***(.0105)

.185**(.0805)

∆4log(ITCAP/EMPLOY)

.109*(.0637)

∆4log(CAPITAL/EMPLOY)

-.0286(.0618)

-.0359(.0638)

.119(.0972)

-.175**(.0876)

∆4log(CAPITAL/OUTPUT)

-.250(.184)

∆4logOUTPUT -.0322(.0723)

Other Controls None None None None None NoneR2 5.2% 5.3% 5.9% 6.4% 7.6% 10.4%N 250 250 250 250 136 107Key: * - p<.1, ** - p<.05, *** - p<.01All variables standardized to mean 0, unit variance.Note: Manufacturing and services do not add to the full sample because some firms arein agriculture, mining or construction.


Table 22: Human Capital Investment as a function of computers andorganization with additional control variables

Dep. Variable HKI HKI HKI HKI HKI HKISpecification

Variable

AKK T.8.Col. 2

AKK T.8.Col. 3

AKK T.8.Col. 4

AKK T.8.Col. 5

Manufact.Only

ServicesOnly


.163***(.0612)

.164***(.0612)

.163***(.0614)

.174***(.0534)

.165*(.0917)

.143(.0907)


.449***(.0558)

.452***(.0558)

.446***(.0558)

.449***(.0151)

.445***(.0712)

.424***(.0955)

∆4log(ITCAP/EMPLOY)

.0373(.0541)

∆4log(CAPITAL/EMPLOY)

-.0608(.0539)

-.0630(.0562)

.0406(.0819)

-.154*(.0826)

∆4log(CAPITAL/OUTPUT)

-.250(.162)

∆4logOUTPUT -.0223(.0671)

Industry Dummies Sector Sector Sector Sector Sector SectorR2 33.6% 34.0% 34.4% 34.1% 38.9% 25.8%N 250 250 250 250 136 107Key: * - p<.1, ** - p<.05, *** - p<.01All variables standardized to mean 0, unit variance.


Table 23: Production Functions with IT-HK and IT-WO interactions

Dependent Variable log(VA) log(VA) log(VA) log(VA) log(VA)Specification

Variable

Baseline Baseline+Skills

Baseline+College

Baseline+Prof’l.

Baseline+Info

WorkersIT Stocklog(ITCAP)

.0383***(.00698)

.0372***(.00702)

.0408***(.00967)

.0347***(.00751)

.0333***(.00746)

Capital Stocklog(NITCAP)

.155***(.00757)

.157***(.00755)

.152***(.00994)

.152***(.00830)

.148***(.00796)

Labor Inputlog(LABOR)

.754***(.00939)

.753***(.00934)

.748***(.0119)

.754***(.0103)

.753***(.00997)

Worker Skill(SKILL)

.00244(.00664)


-.00518(.0104)

Professionals(%PF)

.0223***(.00774)

InformationWorkers (%IW)

.0246***(.00892)

Work Organization(WO)IT x Skilllog(ITCAP)xSKILL

.0301***(.00657)

IT x Collegelog(ITCAP)x%COLL

.0637***(.0103)

IT x Professionalslog(ITCAP) x %PF

.00521(.00659)

IT x InformationWorkerslog(ITCAP) x %IW

.0192***(.00671)

IT x Work Org.log(ITCAP) x WOControls Sector

YearSectorYear

SectorYear

SectorYear

SectorYear

N 2225 2225 1413 1854 1854R2 92.7% 92.8% 90.9% 91.7% 91.7%Key: * - p<.1, **- p<.05, *** - p<.01


Table 24: Production Functions with IT-HK and IT-WO interactions

Dependent Variable log(VA) log(VA) log(VA) log(VA) log(VA)Specification

Variable

Baselinew/WO

Baseline+Skills

Baseline+College

Baseline+Prof’l.

Baseline+Info


.0359***(.00703)

.0352***(.00705)

.0390***(.00977)

.0264***(.00769)

.0302***(.00748)


.154***(.00754)

.156***(.00754)

.152***(.00994)

.148***(.00796)

.146***(.00789)


.749***(.00941)

.749***(.00939)

.747***(.0120)

.747***(.00997)

.747***(.00993)

Worker Skill(SKILL)

-.00426(.00664)


-.0104(.0108)

Professionals(%PF)

.0176**(.00783)

InformationWorkers (%IW)

.0254***(.00679)


.0218**(.00672)

.0231***(.00702)

.0165*(.00947)

.0231***(.00731)

.0267***(.00716)

IT x Skilllog(ITCAP)xSKILL

.0245***(.00722)

IT x Collegelog(ITCAP)x%COLL

.0668***(.0107)

IT x Professionalslog(ITCAP) x %PF

.00288(.00660)

IT x InformationWorkerslog(ITCAP) x %IW

.0192***(.00671)

IT x Work Org.log(ITCAP) x WO

.0218***(.00597)

.0123*(.00657)

-.00762(.00927)

.0264***(.00684)

.0317***(.00691)

Controls SectorYear

SectorYear

SectorYear

SectorYear

SectorYear

N 2225 2225 1413 1854 1854R2 92.8% 92.8% 91.0% 91.7% 91.8%Key: * - p<.1, **- p<.05, *** - p<.01


Table 25: Production Functions with WO-HK and Interactions

Dependent Variable log(VA) log(VA) log(VA) log(VA)Specification

Variable

Skill College Professional InformationWorkers

IT Stocklog(ITCAP)

.0345***(.00706)

.0348***(.00988)

.0261***(.00770)

.0289***(.00753)


.155***(.00754)

.149***(.0101)

.158***(.00656)

.148***(.00794)


.752***(.00936)

.756***(.0121)

.748***(.00999)

.750***(.00995)

Worker Skill(SKILL)

.000414(.00712)


.0130(.0106)

Professionals(%PF)

.0130(.00811)

Information Workers(%IW)

.0223***(.00881)


.0216***(.00703)

.0137(.00958)

.0246***(.00739)

.0265***(.00761)

Work Orgn. x Skill(WO x SKILL)

.0238***(.00633)

Work Orgn.x College(WO x %COLL)

.00939(.00818)

Work Orgn. x Prof’l.(WO x %PF)

.0156**(.00658)

Work Orgn. x InfoWorker (WO x %IW)

.0114(.00696)

Controls Sector***Year

Sector***Year

Sector***Year

Sector***Year

N 2225 1413 1854 1854R2 92.8% 90.7% 91.7% 91.7%Key: * - p<.1, **- p<.05, *** - p<.01


Table 26: Production Functions with IT-HK-WO Interactions

Dependent Variable log(VA) log(VA) log(VA) log(VA)Specification

Variable

HK asSkill

HK asCollege

HK asProfessional

HK asInformation


.0335***(.00723)

.0406***(.0101)

.0272***(.00771)

.0310***(.00767)


.156***(.00753)

.152***(.00997)

.149***(.00800)

.146***(.00790)


.750***(.00939)

.746***(.0121)

.745***(.0101)

.747***(.00790)

Human CapitalHK

-.00140(.00720)

-.00952(.0111)

.0154*(.00836)

.0254***(.00911)

DecentralizationWO

.0216***(.00720)

.0174*(.00959)

.0255***(.00749)

.0268***(.00716)

WO x HK .0153**(.00674)

-.0000117(.00959)

.0131*(.00744)

.00662(.00715)

WO x log(ITCAP) .00934(.00670)

-.00758(.00939)

.0234***(.00713)

.0319***(.00695)

HK x log(ITCAP) .0216***(.00735)

.0704***(.0119)

-.000296(.00688)

.0253***(.00731)

HK x WO xlog(ITCAP)

.00351(.00646)

-.00730(.0118)

-.00542(.00749)

.00422(.00737)

Controls Sector***Year

Sector***Year

Sector***Year

Sector***Year

N 2225 1413 1854 1844R2 92.8% 91.0% 91.8% 91.8%Key: * - p<.1, **- p<.05, *** - p<.01


Table 27: Productivity with Matches and Mismatches on Complements

Table 27A: IT x HK (Using ITCAP and SKILL, respectively) IT \HK

Low High

High -.0293(.0221)N=552

.0422(.0199)N=762

Low 0(N/A)N=561

-.0321(.0174)N=358

Standard Error in parenthesisPearson chi-square Test for Association: χ2=82.4 (p<. 001)Table unbalanced due to ties and discrete scales (SKILL)

Table 27B: IT x WO (Using ITCAP and WO, respectively) IT \WO

Low High

High .00356(.0179)N=448

.0664(.0197)N=664

Low 0(N/A)N=671

.00153(.0199)N=442

Standard Error in parenthesisPearson chi-square Test for Association: χ2=89.0 (p<. 001)Table unbalanced due to ties and discrete scales (SKILL)

Table 27C: HK x WO (Using SKILL and WO, respectively) HK \WO

Low High

High -.0481(.0208)N=492

.0372(.0165)N=822

Low 0(N/A)N=627

-.0522(.0178)N=284

Standard Error in parenthesisPearson chi-square Test for Association: χ2=211 (p<. 001)Table unbalanced due to ties and discrete scales (SKILL)


Table 28: Correlations between manager's perception of IT effects with ITuse, Human Capital and Decentralization

Measure CHANGESKILL

CHANGEAUTONOMY

CHANGEMONITOR

ROUTINIZEWORK

(CSKILLΡ) (CFREEDΡ (CMONIT1Ρ+CMONIT2Ρ)/2

(CROUTΡ)

Mean 4.17 3.58 3.59 2.96

Computer Capital (ITCAP) .09 -.02 -.06 -.08% Use Computer (%GP) .25*** .07 -.04 .00% Use e-mail (%E-mail) .20*** .12** -.04 -.10*

Work Organization (WO) .19*** .13*** .03 .05

Worker Skill (SKILL) .27*** .20*** .12** .04Percent College (%COLL) .19*** .03 -.07 -.12*Human Capital Investment (HKI) .22*** .10* -.01 -.01

Professionals (%PF) .03 -.04 -.04 -.12**

Ρ Measured on a scale of 1-5; 5 is the highest level.Spearman partial rank order correlations controlling for industry (9 sector dummyvariables), employment (EMPLOY) and production worker composition (PRBL, PRCL).N=238-295.Key: * - p<.1, ** - p<. 05, *** - p<.01; test is against the null hypothesis that thecorrelation is zero.

Documents

Information Technology, Workplace Organization and the ...tbres/research/itwold8-21f.pdf · workplace organization, and the demand for skilled labor. In both a short-run factor demand