Network of Practice, IT Knowledge Spillovers, and Productivity: Evidence

from Enterprise Software

Peng Huang

R.H. Smith School of Business

University of Maryland

College Park, MD 20742

huang@umd.edu

Marco Ceccagnoli, Chris Forman and D. J. Wu

Scheller College of Business

Georgia Institute of Technology

800 West Peachtree Street, NW

Atlanta, GA 30308

{marco.ceccagnoli;chris.forman;dj.wu}@scheller.gatech.edu

Abstract

Prior research on IT spillover often uses spillover pools with undefined transmission mechanisms and

ignores the direction of the spillovers. We invent an alternative measurement of IT spillover derived from

a different transmission pathway: knowledge transfer through interactions in an Internet-enabled network

of practices where IT professionals exchange ideas and help each other resolve technical problems they

encounter. Our method separates the effects of knowledge spillover from spillovers that are embedded in

intermediate inputs. We show that IT knowledge spillovers through these networks contribute positively

to firm productivity. The economic payoff of the spillovers is significant: by our estimate, one percent

increase in the inward knowledge spillovers translates to a $0.46 million increase in added value for an

average firm in our sample. Further, we find that IT spillovers complement a firm’s investment in IT

capital in the sense that the productivity of IT capital is increased with higher level of IT spillovers, but

such compelmentarity does not exist for either non-IT capital or non-IT labor. We discuss the

implications for research and practice.

1

1. Introduction

As knowledge becomes one of the central sources of competitive advantage, the identification, acquisition

and use of knowledge has generated sustained interest among researchers (Alcácer and Chung 2007). The

non-rival, public-goods characteristics of knowledge implies that when external knowledge is not

protected by intellectual property rights, it can be acquired through indirect means of spillovers. For

example, Romer (1986) argues that increasing economy-wide returns of innovation are likely to be a

result of spillovers, despite the fact that there is decreasing firm-specific returns to innovation. A large

body of literature has examined the role of knowledge spillovers in R&D (Eeckhout and Jovanovic 2002,

Griliches 1992, Nadiri 1993, O'Mahony and Vecchi 2009). In contrast, IT knowledge spillovers have

been examined only recently (Chang and Gurbaxani 2011, Cheng and Nault 2007, Han et al. 2011), with

good reasons. First, for a long time, the IT productivity paradox (Brynjolfsson 1993) has plagued IS

researchers due to the lack of robust and consistent measures of IT capital investment, and the fact that a

large fraction of IT investment is spent on intangible assets that are not recorded on the balance sheet

(Brynjolfsson et al. 2002). In addition, as Krugman (1991) pointed out, “knowledge flows… leave no

paper trail by which they may be measured and tracked.” The only exception to this is in the form of

patent citations, which has been used by researchers to explore R&D spillovers in inventive activities

(Jaffe and Trajtenberg 1999, Jaffe et al. 1993). In contrast, it is difficult to observe a similar linkage of IT

knowledge spillovers as that in patent citations. Therefore, studies on IT spillovers have to rely on

vaguely defined spillover pools, with little understanding of the mechanisms of spillovers that underlie

these pools. Such methodological treatment is further complicated by the well-known issue of

measurement errors in IT investments, which is likely to result in estimation bias in the magnitude of the

spillover effect (Tambe and Hitt 2012).

This paper aims to provide yet another counterexample to Krugman’s observation by identifying “paper

trails” of IT knowledge spillovers through fine-grained data obtained from an Internet-enabled network of

practices, where IT practitioners help each other solve technical problems and exchange knowledge and

2

ideas on the use of technologies. While prior research stresses the role of geographical proximity in

knowledge spillovers (Jaffe et al. 1993, Keller 2002a), spillovers are increasingly taking place over virtual

channels such as Internet. IT knowledge spillovers are especially susceptible to this path of transmission,

as many large platform technology companies have established Internet-enabled professional

communities of practice to facilitate knowledge exchange and self-support among the technology users.

IT knowledge spillovers are likely to enhance efficiency in the implementation, deployment, and use of

the latest information technology, and therefore shift total factor productivity. That is, given the same

amount of productive inputs such as labor, capital, and IT investment, firms that receive greater inward IT

knowledge spillovers tend to produce more production output than firms that do not. We test this

hypothesis by examining the knowledge spillovers that take place in the context of enterprise software, an

information technology that represents a large fraction of firms’ IT investment that is accompanied by

recent innovation in corporate IT (McAfee and Brynjolfsson 2008), and has significantly improved firm

financial and operational performances (Hitt et al. 2002, McAfee 2002). We measure knowledge

spillovers by tracking SAP Community Network forum activities of the members who are employees of

the Fortune 1000 firms that have installed SAP software, and use a production function framework to

examine the extent to which IT knowledge spillovers drive variations in total factor productivity.

There are several major results from our study. First, we find that the SAP community network has

attracted a large number of IT professionals in a relatively short time period and has experienced

exponential growth. For example, the total number of registered users has grown to around 268,000 in

about 7 years. In addition, consistent with prior anecdotal evidence, we find that the community has been

used as an effective means of knowledge acquisition and peer-support through “crowd sourcing” (Howe

2008). For example, there have been over 1.8 million discussion threads (Q&A conversations) posted in

the forums that focus on various technical and business process-related questions. On average, about a

quarter of all the questions raised by knowledge seekers are solved by the collective efforts of the

community members, and the average time it takes to get a correct solution is between 3-5 days.

3

Second, we find evidence that is consistent with interpreting these interactions as IT spillovers.

Particularly, results from fixed effect panel data estimates of a production function that incorporates the

role of IT spillovers suggest that, among firms that are SAP customers, those that receive higher levels of

knowledge spillovers have greater productivity than firms that do not. The economic payoff of the IT

knowledge spillovers is significant: by our estimate, a one percent increase in the inward knowledge

spillovers translates to a $0.46 million increase in added value for an average firm in our sample. We

conduct a series of falsification tests to explore the robustness of our findings. For example, we find the

identified effect is absent for the rest of the firms in the Fortune 1000 that do not use SAP software.

Further, IT spillovers complement a firm’s investment in IT capital in the sense that the productivity of IT

capital is increased with higher level of IT spillovers, but such compelmentarity does not exist for either

non-IT capital or non-IT labor. Finally, the identified effects of IT spillovers are also robust to

instrumental variables tests.

The reminder of the paper is organized as follow: Section 2 presents a brief overview of literature on

knowledge spillovers in information technology and its relationship to productivity. Section 3 introduces

the research context from which our measure of IT spillover is derived. The data and empirical methods

are introduced in Section 4. We present the results of data analyses in Section 5. In Section 6 we

summarize the findings and discuss their implications.

2. Knowledge Spillovers, Information Technology, and Productivity

The analysis of economic growth by Solow (1957) has sparked much interest in the search of factors that

underlie the productivity residual, which represents the part of output growth that is not explained by the

changes in factor inputs (Hulten 2001). Such factors include firm investment in R&D (Griliches 1979,

1986) and the use of information technologies (Brynjolfsson and Hitt 1995, Brynjolfsson and Hitt 2003).

A significant part of literature on endogenous growth has started to examine not only the effects of firm’s

own investment in these factors but also their social returns to the rest of the economy in the form of

4

spillovers (Romer 1986). Traditionally, empirical research that examined the relationship between

spillovers and productivity has focused primarily on the investments in R&D, and it has revealed positive

spillovers at the firm level (Griliches 1986, Jaffe 1986), the industry level (Goto and Suzuki 1989, Keller

2002b), and the country level (Coe and Helpman 1995, Madsen 2007).1 In some cases, the estimates of

the social return of R&D appear to be unusually high and exceed the internal return, especially when

R&D embedded in upstream industries where intermediate inputs are purchased are included in the

specification of production functions (O'Mahony and Vecchi 2009). In addition, R&D spillovers are

found to display high levels of technological proximity (Jaffe and Trajtenberg 1999, Jaffe 1986) and

geographical proximity (Agrawal et al. 2006, Alcácer and Chung 2007, Griffith et al. 2006, Jaffe et al.

1993).

There is a surge of empirical research in the spillovers of IT investment in recent years (Chang and

Gurbaxani 2011, Cheng and Nault 2007, Cheng and Nault 2011, Han et al. 2011, Tambe and Hitt 2012).

For example, recent studies have shown that IT investments made by a firm’s suppliers or customers

reduce transaction costs through information sharing and coordination, resulting in positive IT spillovers

within supply chains (Cheng and Nault 2007, Cheng and Nault 2011). Similar conclusions were drawn by

earlier works (e.g., Mun and Nadiri 2002). However, not all the firms benefit equally from IT spillovers;

Han et al. (2011) document that two characteristics of downstream industries – IT intensity and

competitiveness – influences the ability and motivation to capture and appropriate IT spillovers of the

firms in the industry. Moreover, Chang and Gurbaxani (2011) find that a large fraction of IT spillover

benefits is derived from trade relationships with IT services industry, and IT-related spillovers have

sustained contribution to productivity in the long run, in some cases persisting for over ten years in IT

intensive firms.

1 For comprehensive surveys, see Griliches (1992) and Nadiri (1993).

5

There are several reasons behind this recent interest in IT spillovers. First, a series of studies on

endogenous growth present evidence that support the hypothesis that information technology is behind

the productivity growth in the late 1990s (Jorgenson et al. 2008, Jorgenson and Stiroh 1999, Jorgenson et

al. 2000, Stiroh 2002b), and that it is the primary reason that explains the difference in the growth rates

between US and most European countries (Bloom et al. 2012, Severgnini 2010). Second, while the IT

productivity paradox has largely been resolved by using better data and improved methodology

(Brynjolfsson and Hitt 1996, Brynjolfsson and Hitt 2003), studies in this area have often detected an

unusually large output elasticity of IT (Brynjolfsson and Hitt 1995, Dewan and Min 1997), which tends to

be substantially higher than its input share. In a neoclassical economics framework where all capital input

must be paid at their marginal product, the output elasticity of IT must equal to its input factor share. This

contradiction leads to the hypothesis that at least some of the excess returns of IT are attributed to IT-

related spillovers (Stiroh 2002a).

Although significant progresses have been made in the research of IT spillovers, scholars in this field are

facing several challenges. One of such challenges is the inability to directly observe IT knowledge

spillovers with clearly defined transmission mechanism and the direction of knowledge flow. While

remedies for this issue exist in R&D spillovers,2 there hasn’t been any satisfactory solution to this

problem in the spillovers of IT. In the absence of such data, researchers rely heavily on the use of

aggregate spillover pools with undefined spillover transmission pathways. For example, Tambe and Hitt

(2012) reveal that the measurements of IT spillover pools are usually calculated by aggregating IT

capitals of other firms using either supply chain weights, industry weights, or trading weights.3 This

2 For example, researchers of R&D spillovers have started to examine the actual mechanisms of knowledge transfer:

some use patent citations to trace the source and destination of knowledge spillovers (Thompson 2006), while others

seek to employ variations in alliance and inventor mobility (Rosenkopf and Almeida 2003).

3 The pooled approach is not unique to IT spillovers. For example, many studies on R&D spillovers implicitly

assume that the technological distance between two firms is primarily determined by their industrial proximity, and

6

methodological treatment leads to several implications. First, it assumes that spillovers are nondirectional

and ignores asymmetry of spillovers in the sense that all firms in the industry/region draw equal benefits

from the contributions of all other firms. However, the endogenous growth literature maintains the view

that knowledge flows exclusively from frontier firms to follower firms, which promotes inequality and

results in free riding by followers (Eeckhout and Jovanovic 2002). More generally, this literature implies

that the sizes of relevant spillover pools differ across firms (Atallah 2005). Empirically, results from

Knott et al. (2009) support the notion that knowledge does have directionality and reject the hypothesis

that spillovers are pooled.

Second, the use of spillover pool has the limitation of being unable to separate rent spillover from

knowledge spillover (Griliches 1979, 1992), although the differences between the two have been

highlighted in prior literature. Rent spillover happens when factor inputs are purchased from other

industries at a price that does not fully reflect the improvements in the quality (Griliches 1992, Severgnini

2010). For example, an upstream supplier’s IT investment may improve the quality/variety of its

production offerings or timelineness /convenience of its services, which are used by a downstream firm as

intermediate input. If the upstream firm is unable to internalize the full return of its IT investment due to

competition among suppliers, externalities result in the downstream industry in the form of spillovers.

Knowledge spillovers, on the other hand, refers to IT-enabled innovations and practices that can be

transferred to other firms through interactions over time (Han et al. 2011). To the extent that the owner of

the knowledge cannot perfectly protect its invention through patent, and such knowledge is non-rival and

non-excludable, it can be learned and replicated by others. Such spillovers may happen via a number of

channels, such as learning, employee mobility, and leakage at trade conferences. As Griliches (1992)

correctly points out, rent spillover is more of a consequence of conventional measurement problem rather

than a true spillover effect.

therefore construct the spillover pool by calculating a weighted sum of R&D stocks of other firms within the same

industry (Ornaghi 2006).

7

In addition, IT spillover researchers encounter another challenge that is quite unique to IT value studies:

the lack of robust, consistently available measures of IT investment. Even in situations where data sets

that are available, the measurement errors in these IT data are rampant, in some cases constituting as high

as 30-40% of the total variance (Tambe and Hitt 2011). While this measurement error creates a downward

bias on the estimate of the productivity of own IT investment, it creates a different issue when the mis-

measured IT data is used to construct IT spillover pool, which often leads to significant overestimate of

the spillover effects. This estimation bias is caused by the fact that the spillover pool constructed in this

way (for example, industry weighted spillover pool) tends to be highly correlated with a firm’s own IT

investment, as the firms in the same industry often share common operating environment and

technological opportunities, therefore make similar IT investments. Tambe and Hitt (2012) provide a

formal analysis of this problem, and they present evidence that when such measurement errors are

corrected using instrumental variable method, the magnitude of the estimate of IT spillovers is

significantly reduced.

The aforementioned challenges call for a better understanding of the mechanisms through which IT

spillovers take place, as well as more accurate measurements of IT spillovers with carefully constructed

spillover functions. In consideration of these issues, this study invents an alternative measurement of IT

spillovers derived from directly observable linkage of knowledge flow. In contrast to prior literature, we

focus on a different spillover transmission pathway: knowledge transfer through interactions in virtual,

Internet-enabled network of practices where IT professionals exchange ideas and help each other resolve

technical problems they encounter. A unique characteristic of these online communities is that they have

the capability to generate spillovers that are not bounded by geographic spaces, as they permit long

distance transmission of technologies (Severgnini 2010). Recent research has theorized this community-

based model of knowledge creation and transfer as an evolutionary process of learning driven by criticism

(Lee and Cole 2003), which may expand beyond the boundary of firms (O'Mahony and Ferraro 2007).

Compared to a traditional closed model, such an model is said to result in faster and high quality solutions

8

to technical questions raised by members and greater variety of innovations (Füller et al. 2007). More

importantly, with the advent of Internet and online communities, there is a possibility to track the “paper

trail” of knowledge flows between IT professionals that participate in these communities. For example,

using information on the collaborative relationships between open-source software (OSS) developers,

Fershtman and Gandal construct a two-mode social network of OSS developers and projects, and link

project success with contributor spillovers and project spillovers (Fershtman and Gandal 2011).

We take this nascent line of research one step further to examine if the knowledge spillovers that take

place in these online networks of practice between the IT professionals lead to the variation in the

accumulation of IT knowledge capital of the firms that employ these professionals, and therefore

influence the total factor productivity of the firms. Although several prior studies have shown that IT

investment may generate externalities (Chang and Gurbaxani 2011, Cheng and Nault 2007, Cheng and

Nault 2011, Han et al. 2011), none of them have demonstrated that such externalities are due to pure IT

knowledge spillovers in the sense of Griliches (Griliches 1992). Compared to the spillover pool approach,

our method has the advantage of identifying knowledge flow with observable source and destination of

the spillovers in a way similar to that observed in patent citations for R&D spillovers. Therefore, our

study aims to add to the recent empirical evidence on IT spillovers by separating the effect of IT

knowledge spillovers on productivity from that of IT-related rent spillover, resulting in cleaner

identification. In addition, our work extends the current understanding of emerging online networks of

practice by presenting evidence of their business value. For example, although there is anecdotal evidence

that technology platform owner-sponsored online communities are often used for promoting innovation

and peer-support among platform adopters, and they bear similarity to open source communities in many

ways (Gorman and Fischer 2009, Von Hippel 1994), there has been a lack of formal studies that examine

the extent to which user firms of the underlying technology platform benefit from participation in those

online communities. Given that technology platform sponsors have invested heavily in building such

9

online user communities and they were embraced by millions of members in some cases, answer to this

question is critical in the understanding of the economic payoff to the investment in these technologies.

3. Research Context

Our research setting is the online community network run by SAP AG, the largest enterprise software

vendor by revenue. As part of its platform strategy, SAP established its Internet-based community of

innovation since 2004, with SAP developer network (SDN) and business process expert (BPX) as its two

major modules. It serves as a resource repository and a platform for SAP users, developers, architects,

consultants and integrators to collaborate and exchange knowledge on the adoption, implementation and

customization of SAP solutions. SAP Community Network (SCN) hosts forums, expert blogs, a technical

library, article downloads, a code sharing gallery, e-learning catalogs, wikis and other facilities through

which its members contribute their knowledge. All these web 2.0 technologies support open

communication between active members of the community, which amount to over 268,000 registered

users from 224 different countries as of 2010.

We choose enterprise software as the background for measuring IT knowledge spillovers for several

reasons. First, wide adoption of enterprise software – such as enterprise resource management (ERP),

customer relationship management (CRM), and supply chain management (SCM) – marks the start of a

period of innovation in corporate IT, and it coincides with the productivity revival of the US in the mid-

1990s (McAfee and Brynjolfsson 2008). Research has shown that investment in enterprise software and

its implementation makes a significant portion of a firm’s overall IT spending (Brynjolfsson et al. 2002)

— in some cases accounting for as high as 75% of corporate IT investment (McAfee and Brynjolfsson

2008), and adoption of enterprise software is associated with significant improvement in firm financial

and operational performance (Hitt et al. 2002). The use of corporate IT such as enterprise software

multiplies the value of innovation in business processes, intensifies competition among rival firms, and

drives economic growth (McAfee and Brynjolfsson 2008). Second, the adoption of complex IT platforms

10

such as enterprise software often requires complementary, specialized knowledge to unlock their

productivity. Enterprise software products are highly business process-oriented and usually need to be

tailored to fit business practices, where idiosyncratic local needs usually drive innovations in work places

(Hitt et al. 2002, Von Hippel 2005). For example, implementation of the off-the-shelf enterprise software

modules usually requires users to customize the configuration of a series of system parameters, modify

existing business processes, choose specific add-on functions and features, and sometimes even devise

specific tools to meet heterogeneous user needs. In addition, to facilitate the interoperability of the

enterprise system with legacy IT infrastructure, or the seamless integration with the information systems

of its partners, suppliers and customers, the end users often have to create workarounds and solutions to

integrate various IS components. Accumulation of specialized knowledge is likely to result during this

process of adaption and customization, and such knowledge is particularly susceptible to spillovers as

most of it is not protected by intellectual property rights. Third, although IT knowledge spillovers may

include that of knowledge related to other technologies such as computer operating systems and personal

productivity software, we choose not to include them in this study as they may be driven by the need of

individual skill-building that is not necessarily linked to the accruement of knowledge capital of the firm

that the individual works for. Unlike many other open source software communities, the knowledge

learned from enterprise software communities is most likely to be applied to drive firm productivity,

instead of being applied to pursue personal interest of the individuals.

A unique feature of SAP community network is that its members’ knowledge contribution to the

community can be quantified. To reward active members, SAP’s online community adopts a contributor

recognition program (CRP), which awards points to community members for each technical article, code

sample, video, wiki contribution, forum post, and weblog authored. For example, in the case of forum

discussion participation, points may be awarded for posting solutions in reply to existing discussion

threads marked as questions, if the answer is deemed helpful by the person who asks the question. SAP

publicly recognizes its most active members. For example, on the “Top Contributors” page, the top 50

11

contributors are listed in recognition of their contribution. On each discussion forum page, the top three

contributors to that forum are listed, with their total reward points displayed. In addition, SAP identifies

and provides special status to exceptional and high-value members by granting them the tile of “SAP

Mentor”. SAP Mentors are offered unique opportunities for access to SAP senior management, early

access to information on products and programs and greater visibility in the on-line communities as well

as at SAP events such as the SAP Tech Ed conference.

The participation in the community network is completely voluntarily and anyone can register as a

member by providing basic personal information. One piece of such profile information is the company

that employs the individual. Using this piece of information, it is possible to aggregate individual level

knowledge contribution and exchange to the firms that employ these individuals, and to derive the

knowledge flow patterns among firms (the definition of such knowledge flow will be introduced in the

next section). Other identifying information includes the country that the user comes from, her

relationship to SAP, email address, phone number, expertise, and LinkedIn profile page, etc. Figure 1

presents a sample user profile.

[Insert Figure 1 Here]

To track knowledge flows between the members of SAP online community network, we focus on user

interactions through the most frequently used communication format: the discussion forums. The primary

purpose of the discussion forums is to provide an avenue for conversations between the community

members so that they help each other solve problems that they encounter during the implementation,

deployment and use of SAP software (Fahey et al. 2007). The forums are organized according to the

domains of the relevant knowledge or expertise, each of which usually corresponds to a particular SAP

software module, or the application of the software solutions in a particular industry. Examples of SAP

forums include ERP manufacturing, product life cycle management, CRM-interaction center, and SAP for

automotive solutions. Each discussion thread is initiated by a knowledge seeker, who posts a specific

12

technical question in a topic forum of her choice. Knowledge contributors, on the other hand, post

responses to the question and try to solve the problem. A discussion thread is comprised of a list of

messages, and each message (either a question or an answer attempt) contains the information about the

member who posts the message, the body of the message, and a time stamp. Once a correct answer (at the

discretion of the knowledge seeker) is received, the discussion thread is closed. We developed a web

scripting tool and obtained the complete history of SAP forum discussions from 2004 to 2010. The

dataset includes about 1.8 million discussion threads with over 8 million messages posted in 243 topic-

specific forums. Table 1 presents some summary statistics of the evolution of the SAP community

network, including numbers of registered members, topic forums and the discussion threads posted in

these forums. Overall, we find that the online community has experienced rapid growth since its

establishment, attracting close to 268,000 registered users in just 7 years, although the growth rate has

slowed recently. In addition, the discussion forums are heavily employed by the members of SCN, with

over 463,000 discussion threads generated in 2008 alone. Further, our data are consistent with earlier

anecdotal evidences that online communities of practice are an effective means of using the “wisdom of

the crowd” for peer support: on average, about a quarter of all the questions raised are solved by the

collective effort of the community members, and the average time it takes to get a correct solution is

between 3-5 days.

[Insert Table 1 Here]

4. Data and Methods

4.1. Estimation models

We adopt the production function approach and extend it by introducing our measurement of IT-related

spillovers. A typical production function relates firm output to factors of input of production (Hulten

2001). For example, a simple form of three-factor Cobb-Douglas production function has been widely

13

used in prior studies on IT productivity (Brynjolfsson and Hitt 1996, Dewan and Min 1997, Mittal and

Nault 2009):

(1)

Where Y is the quantity of production output, K is the stock of non-IT capital, L is the stock of labor, C is

the stock of IT-capital, and A denotes the total factor productivity (TFP). TFP is defined as the output

contribution that is not explained by the factor inputs and often interpreted as technological progress. In

this case, the output elasticity of IT-capital, , represents the percentage increase in output

due to a one percent increase in IT capital. To incorporate the role of knowledge spillovers, we consider

the following modification of (1), which also exploit longitudinal variation in a way similar to that is used

in Thornton and Thompson (2001):

(2)

Where S denotes the measure of inward IT-related knowledge spillovers. In this equation i and t index

firm and time period, respectively. The Cobb-Douglas production function can be employed to estimate

the factor productivities by implementing the following stochastic model:

(3)

OLS estimates of IT spillovers effects are likely to suffer from unobserved firm heterogeneities that are

correlated with inward spillovers. To address this issue, we introduce a set of firm- and year-fixed effects

to control for unobserved heterogeneities. This amounts to the modification of (3)

(4)

Where I and T represent firm fixed effect and year fixed effect, respectively.

14

4.2. Data

We conduct the empirical tests by constructing a dataset of firms that form the SAP install base and are

publicly traded. Our data come from a variety of sources. Particularly, we obtained the SAP installation

data, with a detailed list of product modules that are sold to and installed at all its clients in the United

States prior to the end of year 2004. In addition, we use the Harte Hanks Computer Intelligence (Ci)

Technology database to collect firm-level IT investment data. The Ci database records detailed

information about IT infrastructure for most of the Fortune 1000 firms, which include data on the quantity

of mainframes, peripheral, minicomputers, servers and PC systems, as well as other IT hardware stocks.

Ci database has been widely used by prior studies to investigate issues related to IT productivity

(Chwelos et al. 2010). The Ci data were then matched with Standard and Poor’s Compustat database to

obtain financial information of the publicly traded companies. Using similar method of prior research

(Brynjolfsson and Hitt 1996, Brynjolfsson and Hitt 2003, Chwelos et al. 2010, Dewan and Min 1997), we

use the financial data to construct measures of production output, non-IT capital stock and labor expenses.

4.2.1. Sample

As we are primarily interested in the knowledge spillovers restricted to those associated with a particular

platform technology – enterprise software by SAP, our sample is chosen as the set of firms among the

Fortune 1000 that had installed SAP product prior to the beginning of our sample period. The SAP

community network was established around the end of 2003 and beginning of 2004, so we choose 2004

as the starting year of our analyses. Although we have complete data of Ci Technology database from

2004 to 2009, the methods that are used to collect information on critical data items – such as numbers of

PC and Server owned by firms – has been changed dramatically by Harte Hanks from year 2009. To

maintain consistency of the measurement of IT capital over the sample years, we choose to exclude data

from 2009, resulting a five year sample period from 2004 to 2008. The sample is derived in several steps.

First, we retrieve the set of firms that had ever made into the Fortune 1000 list during 2004-2008, and

match them to Computstat data. We then match these firms with Ci Technology data and get the firms in

15

the intersection of the two databases. Finally, using the list of SAP clients, we obtain those firms that had

installed at least one SAP module prior to the end of year 2004. The final sample consists of 278 firms

with 1264 observations over a 5-year period.

4.2.2 Variables

IT Knowledge Spillovers

We derive the variable of inward IT-related spillovers, Sit, from forum conversations took place on the

SAP community network. Specifically, the rules of SAP reward program dictates that, for each question

that is posted in a topic forum, the knowledge seeker may use her discretion to judge the quality of

answers posted by knowledge contributors, and she can distribute 10 reward points to a user whose

answer is deemed correct (at most 1 answer can be evaluated as correct), 6 points if very helpful (at most

2 answers), and 2 points if helpful (no limit on number of helpful answers). We use a crawler program to

acquire the information on the user profiles for all the registered users of SAP Community Network, such

as their names, addresses, companies, profession, email addresses, countries, personal websites, LinkedIn

profiles, etc. Next, we select all the members that reside in the United States, and match them to

companies in our sample by examining their company affiliations and domains of their email addresses.

For each individual a who is an employee of firm i, we retrieve all the discussion threads that were

initiated by a in year t, and examine the history of the answers posted by other forum members. If a

received any correct, very helpful, or helpful answers in year t, the total number of reward points she gave

to the knowledge contributors are used as a proxy for inward IT spillovers to a. The reward points were

then aggregated across all the threads posted by a in year t to derive an individual level spillover variable,

Sait. The firm level spillover variable is defined as the sum of spillover measures of all the individuals who

are employees of the firm:

16

IT Capital

The measure of IT capital is derived from Computer Intelligence Technology database. The information

in the database covers major categories of IT hardware investments made by Fortune 1000 firms, such as

personal computing, systems and servers, networking, software, storage and managed services (Gu et al.

2008). We adopt the method used by Brynjofsson and Hitt (1995), Hitt and Brynjofsson (1996), and

Dewan and Min (1997) that define the IT capital stock as the sum of computer capital and three times of

IT labor. Inclusion of IT labor expense in the calculation of IT capital is justified by the fact that a large

fraction of IT labor expenses is dedicated to the development of computer software, which is a capital

good. The assumption that underlies this method is that the current IT labor spending is a good proxy of

the IT labor expenses in the recent past, and IS staff “stock” depreciates fully in three years (Brynjolfsson

and Hitt 1995).

The first component of this variable is the market value of total PCs and Servers currently owned by the

firm, converted to constant 2005 US dollars. To be specific, we collect market prices of PCs and Servers

in the United States from two report series produced by Gartner Dataquest Market Statistics database:

Gartner Worldwide Server Forecast and Gartner Worldwide PC Forecast from 2004 to 2008. These two

report series present detailed statistics on the number of shipments, prices, vendor revenues and other

related information about PC and Servers, which is broken down to the level of each geographic region

and market segment. 4 Our market prices of PC and Server are calculated as the average user price across

their respective market segments within the region of United States. These prices are then multiplied by

the quantities of PCs and Servers owned by the firms in our sample, which are retrieved from Harte

Hanks Ci Technology database, to derive the market value of the IT computer assets. Our approach of

calculating the computer market value is similar to that in Brynjofsson and Hitt (2003). Finally, we

4 Gartner Dataquest defines PC market segments as: desk-based, mobile, professional, and home. Server market

segments are defined by CPU types, which include x86, IA64, RISC, and other. The database covers regions of

Asia/Pacific, Eastern Europe, Latin America, Middle East & Africa, and West Europe. Several country level

statistics are also available, which include United States, Canada, and Japan.

17

deflate the market value by Bureau of Economic Analysis (BEA) Price Index for computers and

peripherals.

The second component of IT capital stock is IT-related labor expenses. Ci Technology database provides

the number of IT employees of the sample firms at the site level. We aggregate the site-level employee

numbers to the firm level to derive the total number of IT-related employees hired by the firm.5 IT labor

prices are obtained from Occupational Employment and Wage Estimates series of Bureau of Labor

Statistics (BLS) Occupational Employment Statistics (OES), and we use the mean annual wage of

computer and mathematical occupations as the average labor price for IT employees. As the wage

reported by OES series does not reflect benefits, we multiply the wage number by the ratio of total

compensation to salary, which is obtained from BLS Employer Costs for Employee Compensation

(ECEC) series. The IT labor expense is then deflated by BLS Employment Cost Index (ECI) for private

industry workers.

Production output

We follow prior literature (Brynjolfsson and Hitt 2003, Dewan and Min 1997) and use added value as the

measure of production output, which equals to deflated sale less deflated material. Compared to sales,

added value is said to be less noisy and more comparable across industry sectors (Dewan and Min 1997).

Annual sales numbers are retrieved from Compustat, and we deflate them using industry-specific (at two-

digit NAICS sector) price deflators from BEA Gross Output and Related Series by Industry. Materials are

calculated by subtracting undeflated labor and related expenses (Compustat data item XLR) from

undeflated total operating expenses (Compustat data item XOPR), and deflating by BLS Producer Price

Index (PPI) for intermediate materials, supplies, and components.

5 Ci database actually records a range of IT employees at each site. The ranges are defined as: 1-4, 5-9, 10-24, 25-

49, 50-99, 100-249, 250-499, and 500 or more. For each range, we take the middle value of the range as the number

of IT employees.

18

Non-IT capital

The calculation of total capital stock is similar to that in Brynjofsson and Hitt (2003) for ordinary capital.

Specifically, the gross book value of capital stock (Property, Plant and Equipment (Total-Gross),

Compustat data item PPEGT) is deflated by industry-specific capital investment deflator reported in BLS

1987-2010 Detailed Capital Measures.6 In order to apply the deflators, the average age of capital stock is

calculated as the ratio of total accumulated depreciation (Compustat data item DPACT) to current

depreciation (DP). We then subtract the deflated computer capital from deflated total capital to get the

value of non-IT capital.

Non-IT labor

Consistent with prior studies on IT productivities (Bloom and Van Reenen 2007, Bresnahan et al. 2002,

Brynjolfsson and Hitt 2003), total labor expense is either obtained directly from Compustat Labor and

Related Expenses (data item XLR), or calculated as the product of a firm’s reported number of employees

(Compustat data item EMP) and industry-average labor cost per employee, and deflated by BLS

Employment Cost Index (ECI) for private industry workers. Average labor cost per employee is obtained

from National Sector NAICS Industry-Specific estimates series of BLS Occupational Employment

Statistics (OES). To account for the fraction of benefits in total compensation, we multiply the wage

number by the ratio of total compensation to salary, which is obtained from BLS Employer Costs for

Employee Compensation (ECEC) series. Non-IT labor is defined as the difference between deflated total

labor expense and IT labor expense.

Table 2 reports the summary statistics of the variables. The average firm in the sample has sales of $16.43

billion, added value of $5.42 billion, and 40,736 employees, consistent with the fact that our sample being

publicly-traded, Fortune 1,000 SAP adopters. In addition, the firms invest heavily in IT capital, which has

6 Retrieved from http://www.bls.gov/mfp/mprdload.htm

19

a mean level of $100.59 million and maximum of $1.22 billion. Table 3 provides the correlation matrix

among the key variables.

[Insert Table 2 and Table 3 Here]

Table 4 presents a breakdown of the sample firms by vertical industries, which is based on 2-digit NACIS

sectors. It is notable that firms in manufacturing industry account for the majority (66%) of the sample,

followed by utilities firms (8%).

[Insert Table 4 Here]

5. Results

Three-Factor Productivity Analysis

Although the primary objective of this work is to examine the role of IT knowledge spillovers on firm

productivity, considering a large body of literature has centered on the role of IT capital investment in

driving productivity growth (Brynjolfsson and Hitt 1995, 1996, Brynjolfsson and Hitt 2003, Dewan and

Min 1997), we present a set of results in comparison with prior studies on IT productivities using the

same theoretical framework. One of the reasons that we include this analysis is the lack of studies that

present the evidence of IT productivity using data of recent years, partly due to the lack of robust and

consistent data that measure IT investment. Although exceptions do exist (Chwelos et al. 2010, Tambe

and Hitt 2011), the ways of constructing IT capital variable in these studies are usually different from

earlier literature.7 In addition, economists have raised the concern that returns in IT investments may

decline in recent years, implying that the stock of IT-enabled innovations is being depleted (Stiroh 2008).

7 For example, Tambe and Hitt (Tambe and Hitt 2011) use IT personnel data derived from a job search website,

while Chwelos et al. (Chwelos et al. 2010) use hedonic regression to impute IT equipment price, which is different

from the method employed by Ci database in earlier years.

20

To make the results comparable to earlier works, the sample used in this exercise is chosen as the

complete set of Fortune 1000 firms, instead of the one that we will use in the spillover analyses which

only includes the SAP installed bases. This selection criteria results in 991 firms with 4286 observations

over years 2004-2008, for which we have complete production output and input data. In Column 1 of

Table 5 we present the baseline OLS regression of a 3 factor Cobb-Douglas production function, where

we use robust standard errors that are clustered by firms. To control for heterogeneity in IT productivity

across different industries, we create a set of industry dummies according to SIC divisions, and run a test

with these industry segments as controls. The results are presented in Column 2 of Table 5. In Column 3

we present the results of the OLS model with a set of year dummies that control for productivity shocks

over different time period, and in Column 4 the model includes both industry and year dummies. Results

from panel data models, including fixed-effects and random-effects models, are presented in Column 5

and Column 6. We notice that our estimate of output elasticity is similar to those in Brynjofsson and Hitt

(2003),8 but significantly lower than some of the other IT productivity studies (Dewan and Min 1997, Hitt

and Brynjolfsson 1996), probably due to the different ways of constructing IT capital, different estimation

models, or different sample periods. For example, Hitt and Brynjofsson (1996) use survey data to

construct the measure of computer capital, while Dewan and Min (1997) adopt translog and CES-translog

estimation models.

[Insert Table 5 Here]

Baseline Spillover Analyses

We next turn to the primary variable of interest and consider the role of IT knowledge spillovers in

driving the variations in total factor productivity. We add the variable of IT knowledge spillovers in

8 For comparison, Brynjofsson and Hitt report the output elasticity of IT capital ranging from 0.0085 (using one-year

differences) to 0.0456 (using seven-year differences) in the semi-reduced-form model (Brynjolfsson and Hitt 2003,

p. 800).

21

addition to the usual production factors into the regression and use panel data model as the starting point

of the analyses to control for unobserved firm heterogeneity. In Column 1 and 2 we report the result from

fixed-effects and random-effects models, respectively. As we expect, the coefficients of the spillover term

are significant in both models (p<0.05), indicating that firms with greater amount of inward IT spillovers

produce more production output, given the same amount of input of capital, labor and IT investment.

Particularly, results from the fixed effects model imply that one percentage increase in the amount of

inward spillovers is associated with 0.0086 percentage increase in the added value produced by a firm.

Considering the added value of an average firm in our sample is $5.421 billion, this translates into a $0.46

million increase in production output.

We notice that the estimated output elasticity of IT capital using the SAP installed base as sample is lower

(and sometimes insignificant) than the estimates in Table 5, where we use the complete set of Fortune

1000 firms as sample. Considering the SAP installed bases consist primarily of manufacturing firms, and

IT intensity9 in the SAP sample is considerably lower than that of the rest of the Fortune 1000 sample

(.033 vs. .062, p<0.01), this is consistent with the observation from prior studies (Dewan and Min 1997)

that output elasticity of IT capital is lower in manufacturing industry than in service industry, and it is

higher for IT intensive firms.

To explore the robustness of our findings, we consider several alternative explanations that may

contradict our interpretation. One of such explanations is that the longitudinal variation in the spillover

term merely reflects a positive time trend, which is correlated with the increasing number of firm IT

employees who participate in the SAP community network as it gains popularity. As a result, positive

spillover is driven by a passive learning effect by the increasing number of registered users, which is

unobserved in our model specification. If this is the case, when the cumulative number of registered users

who are the firm’s employees is added into the regression, the effect of our measure of IT spillovers

9 IT intensity is defined as the ratio of IT capital to total capital.

22

would go away. Our results from Column 3, in which we add this variable to the regression, indicate that

this is not the case, and the magnitude of the spillover effect is actually even higher when we control for

this variable. Another possible explanation is that the spillover effect is caused by some unobserved firm

characteristics that drive up the need for knowledge seeking. For example, changes in complexity in a

firm’s IT architecture as a result of an acquisition may prompt compatibility issues with SAP enterprise

software, leading to a greater need for seeking outside consultancy to resolve some technical issues,

which in turn causes a larger spillover. In Column 4 we present a model that explicitly includes the total

number of questions that are raised by a firm’s employees (recall that a question is usually the first

message that initiates a discussion thread) in a year as a control. The results indicate that our finding is

robust to this alternative explanation.

[Insert Table 6 Here]

Instrumental Variable Test

Although the use of fixed effects model in the production function estimation is helpful in teasing out

effect of time-invariant, unobserved firm heterogeneity, our estimates of the spillover effect may still

suffer from endogeneity issues as a result of changes in unobserved firm characteristics over time.

Particularly, firm productivity may be influenced by a variety of unobservable factors that are potentially

correlated with our measure of inward knowledge spillovers, such as changes in the complexity of a

firm’s IT architecture that drives up the need for support related to SAP products, variation in the maturity

of the installed SAP modules, how well these modules work with other existing IT solutions in the user

environment, or the degree to which the existing business processes need to be modified. While it is

impossible to control for all these unobserved factors in the estimation models, we seek to employ

instrumental variable method to increase our confidence in establishing a causal relationship between IT

spillovers and increased productivity.

23

The rationale behind our choice of instrumental variables is that these variables should exogenously shift

the likelihood of getting inward knowledge spillovers, and at the same time being uncorrelated with

unobserved error component mentioned above. To find such variables, we make use of the SAP

installation data and take advantage of the variations in the different product modules installed by

different SAP clients.10

The idea is that firms with different installed product modules may derive varying

degree of benefits from the use of SAP community forums, and the variation in the potential benefits in

turn shapes the incentive of firm employees to seek knowledge in the community. Particularly, we

observe that the online forums are usually organized in such a way that each forum is dedicated to a

special topic of interest, which often corresponds to a specific SAP product module or an industry

solution. In addition, SAP introduces the forums gradually over the years, with only 57 forums in 2004

and 243 forums in 2010 (see details in Table 1). To make use of this longitudinal variation, we create a

mapping table that associates each SAP product module with the forums that are most relevant to this

module.11

In this way, we can identify the forums that are most useful for a firm, by examining the

product modules that the firm installs. Once such firm-to-forum correspondence is established, some of

the forum characteristics that may shift knowledge spillovers can be used to construct our instrumental

variables.

One of such variables is the number of forums that are most useful for a firm in a particular year. Since

the forums are introduced at different times, this variable is time-varying. In addition, the introduction of

the forums reflects SAP decisions that are exogenous to the characteristics of the firms that may drive

changes in the firm’s productivity. Particularly, we obtain the number of forums that are associated with

each firm-year, and create a binary variable high_forum which is set to 1 if the number is greater than the

sample median and to 0 otherwise. Two additional instrumental variables that we construct are related to

10

Some of the most frequently installed SAP product modules include: Financial accounting (FI), material

management (MM), business intelligence and data warehouse (BIW).

11 The mapping table is available upon request.

24

the potential benefits of participation in the forums. Particularly, we calculate the average number of

replies a question receives across all the discussion threads in a forum-year, and then derive a second IV

as the log of weighted average of “number of replies per question” at the firm-year level by referencing

the firm-to-forum matching table. In addition, we calculate the average of total number of useful answers

(which is equal to the sum of correct, helpful and very helpful answers) that a question receives across all

the discussion threads in a forum-year, and then construct a weighted average of “number of useful

answers per question” at the firm-year level by using our firm-to-forum mapping. We then create a binary

indicator high_useful as the third IV. These two variables should increase the expect payoff that shape the

incentive of participation, and therefore exogenously shift the likelihood of generating inward spillover.

We run an instrumental variable model with fixed effects using these 3 IVs, and present the results in

column 1 of Table 7. The results suggest that the positive spillover effect is robust to the endogeneity

issue.

Falsification Tests

To increase the confidence in our interpretation of the regression results, we conduct a series of

falsification tests. First, since the variable we use for IT spillover includes only the knowledge on a

narrowly defined technology platform – SAP enterprise software, such inflow of knowledge would only

be useful for those firms that install SAP software product. In other words, if the observed productivity

increase is indeed due to positive knowledge spillover, we should expect that no such effect exists for

firms that do not use SAP product, despite the fact that some of their employees also participate in forum

activities. We test the same fixed effects spillover model using the rest of the Fortune 1000 firm as sample

(which includes 713 firms and 3022 firm-year observations), and present the result in Column 2 of Table

7. Consistent with our expectation, we find that such knowledge spillover does not boost productivity for

companies that are not in the SAP install base. This result also confirms prior findings that spillovers have

directionality such that the sizes of relevant spillover pools differ across firms (Knott et al. 2009).

25

Second, a particular concern over our panel data estimation model is that serial correlation among the

spillover measure is partly driving the significant results. To rule out this interpretation, in Column 3 of

Table 7 we report a model in which we add the spillover variable in the next period in addition to its

current value. If our interpretation is correct, we expect that the future value of spillover should have no

effect on the productivity of the current year. This is indeed the result we find.

Third, to the extent that our spillover variable measures the knowledge on the implementation and use of

enterprise software, a major component of a firm’s IT investment, we expect that there is a synergy

between IT capital and IT spillovers, in the sense that the output elasticity of IT capital would be higher if

a firm receives greater inward spillovers, due to more effective use of its existing IT infrastructure. We

interact the IT capital variable with the spillover term, and enter it into the fixed effects model. The result

is presented in Column 4 of Table 7. The coefficient estimate of the interaction term is indeed positive,

consistent with the theory of complementarity between IT capital and IT knowledge spillover. For the

purpose of comparison, we also interact the spillover variable with non-IT capital and non-IT labor, and

enter them into the fixed effects model. The results from Column 5 of Table 7 suggest that no such

complementarity exist for non-IT capital or non-IT labor.

[Insert Table 7 Here]

6. Conclusions and Discussion

While most of the existing studies that assess the contribution of IT spillovers to firm productivity have

adopted the empirical strategy of constructing an aggregate pool of external IT investments and

embedding it along with other factor inputs in the production function, such identification method has

been shown to produce significant upward bias on the estimated effects of spillovers due to the

measurement errors in IT capital, and a high correlation between IT capital and the spillover pool (Tambe

and Hitt 2012). In many cases the estimated elasticity of spillover pool is even higher than that of a firm’s

own IT investment, making it difficult to justify these investments in the first place. In contrast, in this

26

study we invent a new method of measuring IT-related spillovers through directly observable linkage of

knowledge flows by using data on the knowledge exchanges and transfers that take place on Internet-

enabled network of practice. We obtain data on the inward knowledge spillovers at individual levels, and

use the company affiliation of these individuals to construct firm level spillover measures. Our analyses

indicate that greater level of IT-related knowledge spillovers is associated with higher productivity, and

our results are robust to a series of falsification tests and instrumental variable tests. To the best of our

knowledge, this is the first study that observes IT spillover with clearly defined transmission paths in a

way similar to that is observed in patent citations for R&D spillovers. Our measure of IT spillovers is less

likely to suffer from estimation bias caused by the pooled approach, since there is little correlation

between the spillover variable and IT capital. In addition, while prior research on knowledge spillovers

has highlighted the distinctions between rent spillover and knowledge spillover (Griliches 1979, 1992),

prior empirical studies of IT spillovers were unable to separate the two due to the way that the spillover

variable is constructed. In this sense, the estimated spillover effect in this study is a constituent of that

estimated by prior researchers, and we show that IT knowledge spillovers alone can drive increase in

productivity, even when the component of rent spillovers is removed.

Our findings reveal several theoretical implications. First, our data support the notion that spillovers are

asymmetric and different firms receive different spillover benefits (Knott et al. 2009, Singh 2007), even if

they are in the same industry or situated in similar positions in the supply chain. For example, a

downstream firm who uses SAP enterprise software systems is unlikely to receive IT knowledge spillover

benefits from an upstream supplier who implement Oracle enterprise software (however this does not

preclude the benefits of rent spillover). More generally, researchers should be more cautious in adopting

the pooled approach to spillover measurement and in choosing the correct spillover function. Second,

much of the R&D spillovers research has pointed to the phenomenon of localized spillovers (Alcácer and

Chung 2007, Jaffe et al. 1993, Keller 2002a), which often leads to the geographic concentration of the

firms in the same industry. Our study suggests that information technology has the capability to bridge the

27

distance created by geography, and that IT knowledge spillovers are increasingly taking place via virtual,

online channels. This implies that the choice of geographic location may play a lesser role in deriving

spillover benefits, at least for technology companies.

Our analyses also offer several useful managerial insights to practitioners. Many organizations face the

trade-off between encouraging their employees’ participation in Internet-enabled networks of practice or

open source software communities to reap learning benefits, and restricting excessive engagement to the

extent of lowering the productivity of its personnel. Our results suggest that the key to this trade-off is the

“fit” – the nature of these communities of practice should fit the set of skills that is most relevant to the

firm’s long term IT strategy. For platform technology owners, our analyses show that nurturing a thriving

Internet-enabled professional community and providing a set of facilities to encourage the knowledge

exchange and peer-support to its technology users can accelerate the spillover process, enhance the

learning experience, and reduce the cost of technical support, which translate to tangible gains to its

technology adopters. This could in turn accelerate the adoption and diffusion of its technology.

28

Figure 1: Sample User Profile

Table 1 Evolution of SAP Community Network

Year Cumulative

number of

registered

users

Number

of

active

forums

Number

of new

threads

initiated

in the

year

Average

number of

replies for

threads

initiated in

the year

Percentage

of questions

solved

Percentage

received

helpful

answers

Percentage

received

very

helpful

answers

Number of

days until

correct

answer

2004 19,289 57 16,296 4.678756477 0.106587713 0.072539 0.098075 13.37847222

2005 43,226 83 67,225 5.39369116 0.24422301 0.270816 0.29484 4.735419274

2006 80,981 141 176,422 5.159815038 0.241686198 0.292963 0.314088 3.359192766

2007 137,552 179 394,183 4.730564867 0.227132791 0.259766 0.286584 4.219277108

2008 198,975 209 463,740 4.625396825 0.252165725 0.255206 0.255556 4.512215939

2009 238,796 233 362,442 4.679898006 0.274699345 0.261455 0.256584 4.003047936

2010 267,997 243 274,296 4.421258701 0.273250193 0.240188 0.235741 3.334158854

29

Table 2 Summary Statistics

Variable Obs Mean Std. dev Min Max

Annual sales (million $) 1264 16432.97 32774.9 298.9129 364392.4

Added value (million $) 1264 5421.446 8745.522 118.1137 73242.29

Non-IT capital (million $) 1264 12306 29417.79 48.44337 321767.7

IT capital (million $) 1264 100.5927 147.0563 0.001968 1224.126

Non-IT labor expense (million $) 1264 2748.302 4372.957 28.75259 40586.13

No. of employees (thousands) 1261 40.73639 59.1636 0.658 428

IT spillover (reward points) 1264 10.51108 92.93818 0 2190

Table 3 Pearson Correlation Matrix of Selected Variables

Variable 1 2 3 4 5 6 7

1 Annual sales 1.0000

(-)

2 Added value 0.8616 1.0000

(0.0000) (-)

3 Non-IT capital 0.8310 0.7693 1.0000

(0.0000) (0.0000) (-)

4 IT capital 0.3331 0.4824 0.3216 1.0000

(0.0000) (0.0000) (0.0000) (-)

5 Non-IT labor expense 0.5496 0.8108 0.3675 0.5725 1.0000

(0.0000) (0.0000) (0.0001) (0.0000) (-)

6 No. of employees 0.5248 0.7576 0.3730 0.5480 0.9334 1.0000

(0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (-)

7 IT spillover 0.0173 0.0451 -0.0006 0.0003 0.0600 0.0357 1.0000

(0.5387) (0.1089) (0.9834) (0.9908) (0.0329) (0.2055) (-)

30

Table 4 Industry Segments of the Sample

2-digit NAICS Description Freq. %

11 Agriculture, Forestry, Fishing and Hunting 5 0.4

21 Mining 25 1.98

22 Utilities 104 8.23

23 Construction 8 0.63

31-33 Manufacturing 839 66.38

42 Wholesale Trade 51 4.03

44-45 Retail Trade 55 4.35

48-49 Transportation and Warehousing 17 1.34

51 Information 72 5.7

52 Finance and Insurance 12 0.95

53 Real Estate and Rental and Leasing 10 0.79

54 Professional, Scientific, and Technical Services 31 2.45

56 Administrative and Support and Waste Management and Remediation Services 10 0.79

62 Health Care and Social Assistance 9 0.71

72 Accommodation and Food Services 16 1.27

99 Unclassified 5 0.4

Total 1,264 100

Table 5 Three-factor Productivity

(1) (2) (3) (4) (5) (6)

Variables OLS OLS OLS OLS Fixed effects Random

effects

Non-IT capital 0.236*** 0.288*** 0.238*** 0.290*** 0.0612*** 0.181***

(0.0171) (0.0234) (0.0171) (0.0234) (0.0214) (0.0184)

IT capital 0.0349*** 0.0133 0.0376*** 0.0165* 0.0113*** 0.0163***

(0.00752) (0.00844) (0.00765) (0.00862) (0.00335) (0.00320)

Non-IT Labor 0.641*** 0.630*** 0.640*** 0.627*** 0.792*** 0.688***

(0.0333) (0.0389) (0.0333) (0.0389) (0.0346) (0.0300)

Constant 1.182*** 1.039*** 1.189*** 1.040*** 1.582*** 0

(0.133) (0.231) (0.135) (0.228) (0.183) (0)

Year Dummies No No Yes Yes Yes Yes

Industry Dummies No Yes No Yes -- --

Observations 4,286 4,286 4,286 4,286 4,286 4,286

No. of firms 991 991 991 991 991 991

R-squared 0.854 0.883 0.856 0.885 0.520

Robust standard errors (clustered by firm) in parentheses except for Column 6.

*** p<0.01, ** p<0.05, * p<0.1

Industry dummies are defined as SIC division.

31

Table 6. Baseline Spillover Models

(1) (2) (3) (4)

Variables Fixed effects Random Effects Fixed effects Fixed effects

Non-IT capital 0.107** 0.239*** 0.107** 0.107** (0.0435) (0.0145) (0.0435) (0.0435) IT capital 0.00150 0.00347 0.00134 0.00166 (0.00578) (0.00494) (0.00579) (0.00580) Non-IT labor 0.730*** 0.680*** 0.731*** 0.732***

(0.0593) (0.0169) (0.0592) (0.0591)

IT spillover 0.00857** 0.0105** 0.0107*** 0.0134**

(0.00388) (0.00452) (0.00408) (0.00560)

Log(registered users) -0.0180 -0.00891

(0.0133) (0.0164)

Log(questions) -0.00927

(0.0102)

Constant 1.748*** 0 1.705*** 1.692***

(0.449) (0) (0.454) (0.454) Year dummies Yes Yes Yes Yes

Observations 1,264 1,264 1,264 1,264

Number of firms 278 278 278 278

R-squared 0.575 0.576 0.577

Robust standard errors (clustered by firm) in parentheses except for Column 2. *** p<0.01, ** p<0.05, * p<0.1

Table 7. IV and Falsification Tests

(1) (2) (3) (4) (5)

Variables Fixed effect IV Fixed effects Fixed effects Fixed effects Fixed effects

Non-IT capital .1173*** 0.0570** 0.0784 0.108** 0.106**

(.0415) (0.0231) (0.0478) (0.0436) (0.0439)

IT capital .0093 0.0148*** 0.000624 0.000584 -0.000430

(.0074) (0.00411) (0.00577) (0.00587) (0.00583)

Non-IT labor .7149*** 0.805*** 0.751*** 0.729*** 0.731***

(.0550) (0.0400) (0.0619) (0.0594) (0.0597)

IT spillover .0793* 0.00745 0.00773** -0.00428 0.0407**

(.0467) (0.00835) (0.00351) (0.00734) (0.0203)

IT spillover (t+1) 0.00261

(0.00321)

IT capital X spillover 0.00347* 0.00678***

(0.00183) (0.00232)

Non-IT capital X spillover 0.00263

(0.00347)

Non-IT labor X spillover -0.0108**

(0.00446)

Constant 1.415*** 1.838*** 1.748*** 1.749***

(0.199) (0.464) (0.450) (0.452)

Year Dummies Yes Yes Yes Yes Yes

Observations 1088 3,022 1,230 1,264 1,264

R-squared 0.445 0.509 0.579 0.576 0.578

Number of firms 232 713 271 278 278

Robust standard errors (clustered by firm) in parentheses. *** p<0.01, ** p<0.05, * p<0.1

32

References

Agrawal, A., I. Cockburn, J. McHale. 2006. Gone but not forgotten: knowledge flows, labor mobility, and

enduring social relationships. Journal of Economic Geography. 6(5) 571-591.

Alcácer, J., W. Chung. 2007. Location strategies and knowledge spillovers. Management Science. 53(5)

760-776.

Atallah, G. 2005. R&D cooperation with asymmetric spillovers. Canadian Journal of Economics. 38(3)

919-936.

Bloom, N., R. Sadun, J. Van Reenen. 2012. Americans Do IT Better: US Multinationals and the

Productivity Miracle. The American Economic Review. 102(1) 167-201.

Bloom, N., J. Van Reenen. 2007. Measuring and Explaining Management Practices Across Firms and

Countries. Quarterly Journal of Economics. 122(4) 1351-1408.

Bresnahan, T.F., E. Brynjolfsson, L.M. Hitt. 2002. Information Technology, Workplace Organization,

And The Demand For Skilled Labor: Firm-Level Evidence. Quarterly Journal of Economics.

117(1) 339-376.

Brynjolfsson, E. 1993. The productivity paradox of information technology. Communications of the

ACM. 36(12) 66-77.

Brynjolfsson, E., L. Hitt. 1995. Information Technology as a Factor of Production: The Role Of

Differences among Firms. Economics of Innovation and New Technology. 3(3&4) 183 - 200.

Brynjolfsson, E., L. Hitt. 1996. Paradox Lost? Firm-level Evidence on the Returns to Information

Systems Spending. Management Science. 42(4) 541-558.

Brynjolfsson, E., L.M. Hitt. 2003. Computing Productivity: Firm-Level Evidence. Review of Economics

& Statistics. 85(4) 793-808.

Brynjolfsson, E., L.M. Hitt, S. Yang. 2002. Intangible Assets: Computers and Organizational Capital.

Brookings Papers on Economic Activity. 2002(1) 137-181.

Chang, Y.B., V. Gurbaxani. 2011. The Impact of IT-Related Spillovers on Long-Run Productivity: An

Empirical Analysis. Information Systems Research. Forthcoming.

Cheng, Z., B.R. Nault. 2007. Industry level supplier-driven IT spillovers. Management Science. 53(8)

1199-1216.

Cheng, Z., B.R. Nault. 2011. Relative Industry Concentration and Customer-Driven IT Spillovers.

Information Systems Research. 23(2) 340-355.

Chwelos, P., R. Ramirez, K.L. Kraemer, N.P. Melville. 2010. Does Technological Progress Alter the

Nature of Information Technology as a Production Input? New Evidence and New Results.

Information Systems Research. 21(2) 392-408.

Coe, D.T., E. Helpman. 1995. International r&d spillovers. European Economic Review. 39(5) 859-887.

Dewan, S., C.-K. Min. 1997. The Substitution of Information Technology for Other Factors of

Production: A Firm Level Analysis. Management Science. 43(12) 1660-1675.

Eeckhout, J., B. Jovanovic. 2002. Knowledge spillovers and inequality. The American Economic Review.

92(5) 1290-1307.

Fahey, R., A.C. Vasconcelos, D. Ellis. 2007. The impact of rewards within communities of practice: a

study of the SAP online global community. Knowledge Management Research & Practice. 5

186-198.

Fershtman, C., N. Gandal. 2011. Direct and indirect knowledge spillovers: the “social network” of

open‐source projects. The RAND Journal of Economics. 42(1) 70-91.

Füller, J., G. Jawecki, H. Mühlbacher. 2007. Innovation creation by online basketball communities.

Journal of Business Research. 60(1) 60-71.

Gorman, A., G. Fischer. 2009. Toward an Analytic Framework for Understanding and Fostering Peer-

support Communities in Using and Evolving Software Products. Proceedings of the fourth

international conference on communities and technologies, University Park, PA, USA.

33

Goto, A., K. Suzuki. 1989. R & D capital, rate of return on R & D investment and spillover of R & D in

Japanese manufacturing industries. The Review of Economics and Statistics. 71(4) 555-564.

Griffith, R., R. Harrison, J. Van Reenen. 2006. How special is the special relationship? Using the impact

of US R&D spillovers on UK firms as a test of technology sourcing. The American Economic

Review. 96(5) 1859-1875.

Griliches, Z. 1979. Issues in assessing the contribution of research and development to productivity

growth. Bell Journal of Economics. 10(1) 92-116.

Griliches, Z. 1986. Productivity, R and D, and Basic Research at the Firm Level in the 1970's. The

American Economic Review. 76(1) 141-154.

Griliches, Z. 1992. The Search for R&D Spillovers. Scandinavian Journal of Economics. 94(Supplement)

S29-S47.

Gu, B., L. Xue, G.N. Ray. 2008. IT Governance and IT Investment Performance: An Empirical Analysis.

Proceedings of the 29th International Conference on Information Systems, Paris, France.

Han, K., Y.B. Chang, J. Hahn. 2011. Information Technology Spillover and Productivity: The Role of

Information Technology Intensity and Competition. Journal of Management Information

Systems. 28(1) 115-146.

Hitt, L.M., E. Brynjolfsson. 1996. Productivity, Business Profitability, and Consumer Surplus: Three

Different Measures of Information Technology Value. MIS Quarterly. 20(2) 121-142.

Hitt, L.M., D.J. Wu, X. Zhou. 2002. Investment in Enterprise Resource Planning: Business Impact and

Productivity Measures. Journal of Management Information Systems. 19(1) 71-98.

Howe, J. 2008. Crowdsourcing: Why the power of the crowd is driving the future of business. Crown

Business, New York.

Hulten, C.R. 2001. Total factor productivity: a short biography. C.R. Hulten, E.R. Dean, M.J. Harper, eds.

New Developments in Productivity Analysis. University of Chicago Press, 1-54.

Jaffe, A., M. Trajtenberg. 1999. International Knowledge Flows: Evidence From Patent Citations.

Economics of Innovation and New Technology. 8(1-2) 105-136.

Jaffe, A.B. 1986. Technological Opportunity and Spillovers of R & D: Evidence from Firms' Patents,

Profits, and Market Value. The American Economic Review. 76(5) 984-1001.

Jaffe, A.B., M. Trajtenberg, R. Henderson. 1993. Geographic Localization of Knowledge Spillovers as

Evidenced by Patent Citations. The Quarterly Journal of Economics. 108(3) 577-598.

Jorgenson, D.W., M.S. Ho, K.J. Stiroh. 2008. A Retrospective Look at the U.S. Productivity Growth

Resurgence. Journal of Economic Perspectives. 22(1) 3-24.

Jorgenson, D.W., K.J. Stiroh. 1999. Information Technology and Growth. The American Economic

Review. 89(2) 109-115.

Jorgenson, D.W., K.J. Stiroh, R.J. Gordon, D.E. Sichel. 2000. Raising the Speed Limit: U.S. Economic

Growth in the Information Age. Brookings Papers on Economic Activity. 2000(1) 125-235.

Keller, W. 2002a. Geographic Localization of International Technology Diffusion. The American

Economic Review. 92(1) 120-142.

Keller, W. 2002b. Trade and the Transmission of Technology. Journal of Economic Growth. 7(1) 5-24.

Knott, A.M., H.E. Posen, B. Wu. 2009. Spillover asymmetry and why it matters. Management Science.

55(3) 373-388.

Krugman, P.R. 1991. Geography and trade. the MIT Press, Cambridge, MA.

Lee, G.K., R.E. Cole. 2003. From a Firm-Based to a Community-Based Model of Knowledge Creation:

The Case of the Linux Kernel Development. Organization Science. 14(6) 633-649.

Madsen, J.B. 2007. Technology spillover through trade and TFP convergence: 135 years of evidence for

the OECD countries. Journal of International Economics. 72(2) 464-480.

McAfee, A. 2002. The Impact of Enterprise Information Technology Adoption on Operational

Performance: An Empirical Investigation. Production & Operations Management. 11(1) 33-53.

McAfee, A., E. Brynjolfsson. 2008. Investing in the IT That Makes a Competitive Difference. Harvard

Business Review. 86(7/8) 98-107.

34

Mittal, N., B.R. Nault. 2009. Research Note--Investments in Information Technology: Indirect Effects

and Information Technology Intensity. Information Systems Research. 20(1) 140-154.

Mun, S.B., M.I. Nadiri. 2002. Information technology externalities: empirical evidence from 42 US

industries. NBER working paper.

Nadiri, M.I. 1993. Innovations and technological spillovers. NBER Working Paper, No. 4423.

O'Mahony, M., M. Vecchi. 2009. R&D, knowledge spillovers and company productivity performance.

Research Policy. 38(1) 35-44.

O'Mahony, S., F. Ferraro. 2007. The emergence of governance in an open source community. Academy of

Management Journal. 50(5) 1079-1106.

Romer, P.M. 1986. Increasing returns and long-run growth. The Journal of Political Economy. 94(5)

1002-1037.

Rosenkopf, L., P. Almeida. 2003. Overcoming Local Search Through Alliances and Mobility.

Management Science. 49(6) 751-766.

Severgnini, B. 2010. Is ICT a Jack-in-the-Box? A Counterfactual Approach for Identifying Productivity

Spillovers. Working Paper.

Singh, J. 2007. Asymmetry of knowledge spillovers between MNCs and host country firms. Journal of

International Business Studies. 38(5) 764-786.

Solow, R.M. 1957. Technical change and the aggregate production function. The Review of Economics

and Statistics. 39(3) 312-320.

Stiroh, K. 2008. Information Technology and Productivity: Old Answers and New Questions. CESifo

Economic Studies. 54(3) 358-385.

Stiroh, K.J. 2002a. Are ICT spillovers driving the New Economy? Review of Income and Wealth. 48(1)

33-57.

Stiroh, K.J. 2002b. Information Technology and the U.S. Productivity Revival: What Do the Industry

Data Say? American Economic Review. 92(5) 1559-1576.

Tambe, P., L.M. Hitt. 2011. The Productivity of Information Technology Investments: New Evidence

from IT Labor Data. Information Systems Research. Forthcoming.

Tambe, P., L.M. Hitt. 2012. Measuring spillovers from information technology investments. New York

University Working Paper.

Thompson, P. 2006. Patent citations and the geography of knowledge spillovers: evidence from inventor-

and examiner-added citations. The Review of Economics and Statistics. 88(2) 383-388.

Thornton, R.A., P. Thompson. 2001. Learning from experience and learning from others: An exploration

of learning and spillovers in wartime shipbuilding. American Economic Review. 91(5) 1350-1368.

Von Hippel, E. 1994. The Sources of Innovation. Oxford University Press, New York.

Von Hippel, E. 2005. Democratizing Innovation. The MIT Press, Cambridge, MA.