Democratizing Innovation by Eric von Hippel - HTML preview

PLEASE NOTE: This is an HTML preview only and some elements such as links or page numbers may be incorrect.
Download the book in PDF, ePub, Kindle for a complete version.

49

Democratizing Innovation

290

6 Why Users Often Freely Reveal Their Innovations

to fully understand that information or put it to use. However, if

the possessor of the information does not profit from any such ex-

291

Products, services, and processes developed by users become

penditures made by its adopters, the information itself is still freely

more valuable to society if they are somehow diffused to others

revealed, according to our definition. This definition of free reveal-

that can also benefit from them. If user innovations are not dif-

ing is rather extreme in that revealing with some small constraints,

fused, multiple users with very similar needs will have to invest to

as is sometimes done, would achieve largely the same economic

(re)develop very similar innovations, which would be a poor use of

effect. Still, it is useful to discover that innovations are often freely

resources from the social welfare point of view. Empirical research

revealed even in terms of this stringent definition.

shows that new and modified products developed by users often do

diffuse widely---and they do this by an unexpected means: user-

Routine and intentional free revealing among profit-seeking firms

295

innovators themselves often voluntarily publicly reveal what they

was first described by Allen (1983). He noticed the phenomenon,

have developed for all to examine, imitate, or modify without any

which he called collective invention, in historical records from the

payment to the innovator.

nineteenth-century English iron industry. In that industry, ore was

processed into iron by means of large furnaces heated to very

292

In this chapter, I first review evidence that free revealing is frequent.

high temperatures. Two attributes of the furnaces used had been

Next, I discuss the case for free revealing from an innovators' per-

steadily improved during the period 1850--1875: chimney height

spective, and argue that it often can be the best practical route for

had been increased and the temperature of the combustion air

users to increase profit from their innovations. Finally, I discuss the

pumped into the furnace during operation had been raised. These

implications of free revealing for innovation theory.

two technical changes significantly and progressively improved the

energy efficiency of iron production---a very important matter for

293

Evidence of Free Revealing

producers. Allen noted the surprising fact that employees of com-

294

When my colleagues and I say that an innovator “freely reveals”

peting firms publicly revealed information on their furnace design

proprietary information, we mean that all intellectual property rights

improvements and related performance data in meetings of profes-

to that information are voluntarily given up by that innovator and all

sional societies and in published material.

parties are given equal access to it---the information becomes a

public good (Harhoff, Henkel, and von Hippel 2003). For example,

After Allen's initial observation, a number of other authors searched

296

placement of non-patented information in a publicly accessible site

for free revealing among profit-seeking firms and frequently found

such as a journal or public website would be free revealing as we

it. Nuvolari (2004) studied a topic and time similar to that studied by

define it. Free revealing as so defined does not mean that recipi-

Allen and found a similar pattern of free revealing in the case of im-

ents necessarily acquire and utilize the revealed information at no

provements made to steam engines used to pump out mines in the

cost to themselves. Recipients may, for example, have to pay for

1800s. At that time, mining activities were severely hampered by

a subscription to a journal or for a field trip to an innovation site

water that tended to flood into mines of any depth, and so an early

to acquire the information being freely revealed. Also, some may

and important application of steam engines was for the removal of

have to obtain complementary information or other assets in order

water from mines. Nuvolari explored the technical history of steam

SiSU

web.mit.edu/evhippel/www/

50

Democratizing Innovation

engines used to drain copper and tin mines in England's Cornwall

modifications they had developed. Reasons given for not reveal-

District. Here, patented steam engines developed by James Watt

ing the remainder had nothing to do with considerations of intel-

were widely deployed in the 1700s. After the expiration of the Watt

lectual property protection. Rather, users who did not share said

patent, an engineer named Richard Trevithick developed a new

they had no convenient users' group forum for doing so, and/or

type of high-pressure engine in 1812. Instead of patenting his in-

they thought their innovation was too specialized to be of interest

vention, he made his design available to all for use without charge.

to others.

The engine soon became the basic design used in Cornwall. Many

Innovating users of sports equipment also have been found to

mine engineers improved Trevithick's design further and published

298

freely reveal their new products and product modifications. Franke

what they had done in a monthly journal, Leans Engine Reporter.

and Shah (2003), in their study of four communities of serious

This journal had been founded by a group of mine managers with

sports enthusiasts described in chapter 2, found that innovating

the explicit intention of aiding the rapid diffusion of best practices

users uniformly agreed with the statement that they shared their in-

among these competing firms.

novation with their entire community free of charge---and strongly

297

Free revealing has also been documented in the case of more re-

disagreed with the statement that they sold their innovations (p <

cent industrial equipment innovations developed by users. Lim

0.001, t-test for dependent samples). Interestingly, two of the four

(2000) reports that IBM was first to develop a process to manu-

communities they studied engaged in activities involving signifi-

facture semiconductors that incorporated copper interconnections

cant competition among community members. Innovators in these

among circuit elements instead of the traditionally used aluminum

two communities reported high but significantly less willingness to

ones. After some delay, IBM revealed increasing amounts of pro-

share, as one might expect in view of the potentially higher level of

prietary process information to rival users and to equipment sup-

competitive loss free revealing would entail.

pliers. Widespread free revealing was also found in the case of

Contributors to the many open source software projects extant

299

automated clinical chemistry analyzers developed by the Techni-

(more than 83,000 were listed on SourceForge.net in 2004) also

con Corporation for use in medical diagnosis. After commercial

routinely make the new code they have written public. Well-known

introduction of the basic analyzer, many users developed major

open source software products include the Linux operating system

improvements to both the analyzer and to the clinical tests pro-

software and the Apache web server computer software. Some

cessed on that equipment. These users, generally medical per-

conditions are attached to open source code licensing to ensure

sonnel, freely revealed their improvements via publication, and at

that the code remains available to all as an information commons.

company-sponsored seminars (von Hippel and Finkelstein 1979).

Because of these added protections, open source code does not

Mishina (1989) found free, or at least selective no-cost revealing

quite fit the definition of free revealing given earlier in this chapter.

in the lithographic equipment industry. He reported that innovating

(The licensing of open source software will be discussed in detail

equipment users would sometimes reveal what they had done to

in chapter 7.)

machine manufacturers. Morrison, Roberts, and I, in our study of

library IT search software (discussed in chapter 2 above), found

Henkel (2003) showed that free revealing is sometimes practiced

300

that innovating users freely revealed 56 percent of the software

by directly competing manufacturers. He studied manufacturers

SiSU

web.mit.edu/evhippel/www/

51

Democratizing Innovation

that were competitors and that had all built improvements and ex-

able from holding an innovation secret or licensing it, free revealing

tensions to a type of software known as embedded Linux. (Such

should be the preferred course of action for a profit-seeking firm or

software is “embedded in” and used to operate equipment ranging

individual.

from cameras to chemical plants.) He found that these manufac-

Others Often Know Something Close to “Your” Secret

turers freely revealed improvements to the common software plat-

304

form that they all shared and, with a lag, also revealed much of the

Innovators seeking to protect innovations they have developed as

305

equipment-specific code they had written.

their intellectual property must establish some kind of monopoly

control over the innovation-related information. In practice, this can

301

The Practical Case for Free Revealing

be done either by effectively hiding the information as a trade se-

302

The “private investment model” of innovation assumes that inno-

cret, or by getting effective legal protection by patents or copyrights.

vation will be supported by private investment if and as innovators

(Trademarks also fall under the heading of intellectual property, but

can make attractive profits from doing so. In this model, any free re-

we do not consider those here.) In addition, however, it must be

vealing or uncompensated “spillover” of proprietary knowledge de-

the case that others do not know substitute information that skirts

veloped by private investment will reduce the innovator's profits. It

these protections and that they are willing to reveal. If multiple in-

is therefore assumed that innovators will strive to avoid spillovers of

dividuals or firms have substitutable information, they are likely to

innovation-related information. From the perspective of this model,

vary with respect to the competitive circumstances they face. A

then, free revealing is a major surprise: it seems to make no sense

specific innovator's ability to protect “its” innovation as proprietary

that innovators would intentionally give away information for free

property will then be determined for all holders of such information

that they had invested money to develop.

by the decision of the one having the least to lose by free reveal-

ing. If one or more information holders expect no loss or even a

303

In this subsection I offer an explanation for the puzzle by pointing

gain from a decision to freely reveal, then the secret will proba-

out that free revealing is often the best practical option available

bly be revealed despite other innovators' best efforts to avoid this

to user innovators. Harhoff, Henkel, and von Hippel (2003) found

fate.

that it is in practice very difficult for most innovators to protect their

innovations from direct or approximate imitation. This means that

Commonly, firms and individuals have information that would be

306

the practical choice is typically not the one posited by the private

valuable to those seeking to imitate a particular innovation. This

investment model: should innovators voluntarily freely reveal their

is because innovators and imitators seldom need access to a spe-

innovations, or should they protect them? Instead, the real choice

cific version of an innovation. Indeed, engineers seldom even want

facing user innovators often is whether to voluntarily freely reveal

to see a solution exactly as their competitors have designed it:

or to arrive at the same end state, perhaps with a bit of a lag, via in-

specific circumstances differ even among close competitors, and

voluntary spillovers. The practical case for voluntary free revealing

solutions must in any case be adapted to each adopter's precise

is further strengthened because it can be accomplished at low cost,

circumstances. What an engineer does want to extract from the

and often yields private benefits to the innovators. When benefits

work of others is the principles and the general outline of a possi-

from free revealing exceed the benefits that are practically obtain-

ble improvement, rather than the easily redevelopable details. This

SiSU

web.mit.edu/evhippel/www/

52

Democratizing Innovation

information is likely to be available from many sources.

many other help-line participants also knew a solution, and could

have provided an answer if they themselves had not done so (table

307

For example, suppose you are a system developer at a bank and

6.1).

you are tasked with improving in-house software for checking cus-

Table 6.1 Even very specialized information is often widely known.

tomers' credit online. On the face of it, it might seem that you would

309

Tabulated here are answers to a question asked of help-line infor-

gain most by studying the details of the systems that competing

mation providers: “How many others do you think knew the answer

banks have developed to handle that same task. It is certainly true

to the question you answered?”

that competing banks may face market conditions very similar to

your bank, and they may well not want to reveal the valuable inno-

310

vations they have developed to a competitor. However, the situa-

Frequent providers (n = 21)

tion is still by no means bleak for an imitator. There are also many

Other providers (n = 67)

non-bank users of online credit checking systems in the world---

Many

38%

61%

probably millions. Some will have innovated and be willing to reveal

A few with good Apache knowledge

38%

18%

what they have done, and some of these will have the information

A few with specific problem experience

24%

21%

you need. The likelihood that the information you seek will be freely

revealed by some individual or firm is further enhanced by the fact

that your search for novel basic improvements may profitably ex-

Source: Lakhani and von Hippel 2003, table 10.

311

tend far beyond the specific application of online credit checking.

Even in the unlikely event that a secret is held by one individual,

312

Other fields will also have information on components of the solu-

that information holder will not find it easy to keep a secret for long.

tion you need. For example, many applications in addition to online

Mansfield (1985) studied 100 American firms and found that “infor-

credit checking use software components designed to determine

mation concerning development decisions is generally in the hands

whether persons seeking information are authorized to receive it.

of rivals within about 12 to 18 months, on the average, and in-

Any can potentially be a provider of information for this element of

formation concerning the detailed nature and operation of a new

your improved system.

product or process generally leaks out within about a year.” This

observation is supported by Allen's previously mentioned study of

308

A finding by Lakhani and von Hippel (2003) illustrates the possi-

free revealing in the nineteenth-century English iron industry. Allen

bility that many firms and individuals may have similar informa-

(1983, p. 17) notes that developers of improved blast furnace de-

tion. Lakhani and von Hippel studied Apache help-line websites.

signs were unlikely to be able to keep their valuable innovations

These sites enable users having problems with Apache software

secret because “in the case of blast furnaces and steelworks, the

to post questions, and others to respond with answers. The au-

construction would have been done by contractors who would know

thors asked those who provided answers how many other help-line

the design.” Also, “the designs themselves were often created by

participants they thought also knew a solution to specific and often

consulting engineers who shifted from firm to firm.”

obscure problems they had answered on the Apache online forum.

Information providers generally were of the opinion that some or

Low Ability to Profit from Patenting

313

SiSU

web.mit.edu/evhippel/www/

53

Democratizing Innovation

314

Next, suppose that a single user-innovator is the only holder of a

for excluding imitators or for capturing royalties in most industries.

particular unit of innovation-related information, and that for some

(Fields generally cited as exceptions are pharmaceuticals, chem-

reason there are no easy substitutes. That user actually does have

icals, and chemical processes, where patents do enable markets

a real choice with respect to disposing of its intellectual property: it

for technical information (Arora et al. 2001).) Most respondents

can keep the innovation secret and profit from in-house use only, it

also say that the availability of patent protection does not induce

can license it, or it can choose to freely reveal the innovation. We

them to invest more in research and development than they would

have just seen that the practical likelihood of keeping a secret is

if patent protection did not exist. Taylor and Silberston (1973) re-

low, especially when there are multiple potential providers of very

ported that 24 of 32 firms said that only 5 percent or less of their

similar secrets. But if one legally protects an innovation by means

R&D expenditures were dependent on the availability of patent pro-

of a patent or a copyright, one need not keep an innovation secret in

tection. Levin et al. (1987) surveyed 650 R&D executives in 130

order to control it. Thus, a firm or an individual that freely reveals

different industries and found that all except respondents from the

is forgoing any chance to get a profit via licensing of intellectual

chemical and pharmaceutical industries judged patents to be “rela-

property for a fee. What, in practical terms, is the likelihood of

tively ineffective.” Similar findings have been reported by Mansfield

succeeding at this and so of forgoing profit by choosing to freely

(1968, 1985), by Cohen et al. (2000, 2002), by Arundel (2001), and

reveal?

by Sattler (2003).

315

In most subject matters, the relevant form of legal protection for in-

Despite recent governmental efforts to strengthen patent enforce-

317

tellectual property is the patent, generally the “utility” patent. (The

ment, a comparison of survey results indicates only a modest

notable exception is the software industry, where material to be

increase between 1983 and 1994 in large firms' evaluations of

licensed is often protected by copyright.) In the United States, util-

patents' effectiveness in protecting innovations or promoting

ity patents may be granted for inventions related to composition of

innovation investments. Of course, there are notable exceptions:

matter and/or a method and/or a use. They may not be granted for

some firms, including IBM and TI, report significant income from

ideas per se, mathematical formulas, laws of nature, and anything

the licensing of their patented technologies.

repugnant to morals and public policy. Within subject matters po-

tentially protectable by patent, protection will be granted only when

Obtaining a patent typically costs thousands of dollars, and it can

318

the intellectual property claimed meets additional criteria of useful-

take years (Harhoff, Henkel, and von Hippel 2003). This makes

ness, novelty, and non-obviousness to those skilled in the relevant

patents especially impractical for many individual user-innovators,

art. (The tests for whether these criteria have been met are based

and also for small and medium-size firms of limited means. As a

on judgement. When a low threshold is used, patents are easier to

stark example, it is hard to imagine that an individual user who has

get, and vice-versa (Hall and Harhoff 2004).)

developed an innovation in sports equipment would find it appeal-

ing to invest in a patent and in follow-on efforts to find a licensee

316

The real-world value of patent protection has been studied for more

and enforce payment. The few that do attempt this, as Shah (2000)

than 40 years. Various researchers have found that, with a few ex-

has shown, seldom gain any return from licensees as payment for

ceptions, innovators do not think that patents are very useful either

their time and expenditures.

SiSU

web.mit.edu/evhippel/www/

54

Democratizing Innovation

319

Copyright is a low-cost and immediate form of legal protection that

nificant money and time to ensure that their innovations are seen

applies to original writings and images ranging from software code

in a favorable light, and that information about them is effectively

to movies. Authors do not have to apply for copyright protection;

and widely diffused. Writers of computer code may work hard to

it “follows the author's pen across the page.” Licensing of copy-

eliminate all bugs and to document their code in a way that is very

righted works is common, and it is widely practiced by commercial

easy for potential adopters to understand before freely revealing it.

software firms. When one buys a copy of a non-custom software

Plant owners may repaint their plant, announce the availability of

product, one is typically buying only a license to use the software,

tours at a general industry meeting, and then provide a free lunch

not buying the intellectual property itself. However, copyright pro-

for their visitors.

tection is also limited in an important way. Only the specific original

writing itself is protected, not the underlying invention or ideas. As

Innovators' active efforts to diffuse information about their innova-

323

a consequence, copyright protections can be circumvented. For

tions suggest that there are positive, private rewards to be obtained

example, those who wish to imitate the function of a copyrighted

from free revealing. A number of authors have considered what

software program can do so by writing new software code to im-

these might be. Allen (1983) proposed that reputation gained for

plement that function.

a firm or for its managers might offset a reduction in profits for the

firm caused by free revealing. Raymond (1999) and Lerner and

320

Given the above, we may conclude that in practice little profit is

Tirole (2002) elaborated on this idea when explaining free reveal-

being sacrificed by many user-innovator firms or individuals that

ing by contributors to open source software development projects.

choose to forgo the possibility of legally protecting their innovations

Free revealing of high-quality code, they noted, can increase a pro-

in favor of free revealing.

grammer's reputation with his peers. This benefit can lead to other

benefits, such as an increase in the programmer's value on the job

321

Positive Incentives for Free Revealing

market. Allen has argued that free revealing might have effects

322

As was noted earlier, when we say that an innovator “freely reveals”

that actually increase a firm's profits if the revealed innovation is to

proprietary information we mean that all existing and potential in-

some degree specific to assets owned by the innovator (see also

tellectual property rights to that information are voluntarily given up

Hirschleifer 1971).

by that innovator and that all interested parties are given access to

it---the information becomes a public good. These conditions can

Free revealing may also increase an innovator's profit in other

324

often be met at a very low cost. For example, an innovator can

ways. When an innovating user freely reveals an innovation, the

simply post information about the innovation on a website without

direct result is to increase the diffusion of that innovation relative

publicity, so those potentially interested must discover it. Or a firm

to what it would be if the innovation were either licensed at a fee

that has developed a novel process machine can agree to give

or held secret. The innovating user may then benefit from the in-

a factory tour to any firm or individual that thinks to ask for one,

crease in diffusion via a number of effects. Among these are net-

without attempting to publicize the invention or the availability of

work effects. (The classic illustration of a network effect is that the

such tours in any way. However, it is clear that many innovators

value of each telephone goes up as more are sold, because the

go beyond basic, low-cost forms of free revealing. They spend sig-

value of a phone is strongly affected by the number of others who

SiSU

web.mit.edu/evhippel/www/

55

Democratizing Innovation

can be contacted in the network.) In addition, and very importantly,

ing users also report being motivated to freely reveal their code

an innovation that is freely revealed and adopted by others can

under a free or open source license by a number of additional fac-

become an informal standard that may preempt the development

tors. These include giving support to open code and “giving back”

and/or commercialization of other versions of the innovation. If, as

to those whose freely revealed code has been of value to them

Allen suggested, the innovation that is revealed is designed in a

(Lakhani and Wolf 2005).

way that is especially appropriate to conditions unique to the inno-

vator, this can result in creating a permanent source of advantage

By freely revealing information about an innovative product or pro-

327

for that innovator.

cess, a user makes it possible for manufacturers to learn about that

innovation. Manufacturers may then improve upon it and/or offer it

325

Being first to reveal a certain type of innovation increases a user

at a price lower than users' in-house production costs (Harhoff et al.

firm's chances of having its innovation widely adopted, other things

2003). When the improved version is offered for sale to the general

being equal. This may induce innovators to race to reveal first.

market, the original user-innovator (and other users) can buy it and

Firms engaged in a patent race may disclose information voluntarily

gain from in-house use of the improvements. For example, con-

if the profits from success do not go only to the winner of the race. If

sider that manufacturers often convert user-developed innovations

being second quickly is preferable to being first relatively late, there

(“home-builts”) into a much more robust and reliable form when

will be an incentive for voluntary revealing in order to accelerate the

preparing them for sale on the commercial market. Also, manufac-

race (de Fraja 1993).

turers offer related services, such as field maintenance and repair

programs, that innovating users must otherwise provide for them-

326

Incentives to freely reveal have been most deeply explored in the

selves.

specific case of open source software projects. Students of the

open source software development process report that innovating

A variation of this argument applies to the free revealing among

328

users have a number of motives for freely revealing their code to

competing manufacturers documented by Henkel (2003). Com-

open source project managers and open source code users in gen-

peting developers of embedded Linux systems were creating soft-

eral. If they freely reveal, others can debug and improve upon

ware that was specifically designed to run the hardware products

the modules they have contributed, to everyone's benefit. They

of their specific clients. Each manufacturer could freely reveal this

are also motivated to have their improvement incorporated into the

equipment-specific code without fear of direct competitive reper-

standard version of the open source software that is generally dis-

cussions: it was applicable mainly to specific products made by a

tributed by the volunteer open source user organization, because

manufacturer's client, and it was less valuable to others. At the

it will then be updated and maintained without further effort on the

same time, all would jointly benefit from free revealing of improve-

innovator's part. This volunteer organization is the functional equiv-

ments to the underlying embedded Linux code base, upon which

alent of a manufacturer with respect to inducing manufacturer im-

they all build their proprietary products. After all, the competitive

provements, because a user-developed improvement will be as-

advantages of all their products depended on this code base's be-

sured of inclusion in new “official” software releases only if it is

ing equal to or better than the proprietary software code used by

approved and adopted by the coordinating user group. Innovat-

other manufacturers of similar products. Additionally, Linux soft-

SiSU

web.mit.edu/evhippel/www/

56

Democratizing Innovation

ware was a complement to hardware that many of the manufactur-

mathematics. She notes that “scholars in diverse disciplines are

ers in Henkel's sample also sold. Improved Linux software would

adopting open-access practices at a surprisingly high rate and are

likely increase sales of their complementary hardware products.

being rewarded for it, as reflected in [citations].”

(Complement suppliers' incentives to innovate have been modeled

Implications for Theory

by Harhoff (1996).)

332

We have seen that in practice free revealing may often be the best

333

329

Free Revealing and Reuse

practical course of action for innovators. How can we tie these

330

Of course, free revealing is of value only if others (re)use what has

observations back to theory, and perhaps improve theory as a re-

been revealed. It can be difficult to track what visitors to an infor-

sult? At present there are two major models that characterize how

mation commons take away and reuse, and there is as yet very

innovation gets rewarded. The private investment model is based

little empirical information on this important matter. Valuable forms

on the assumption that innovation will be supported by private in-

of reuse range from the gaining of general ideas of development

vestors expecting to make a profit. To encourage private invest-

paths to pursue or avoid to the adoption of specific designs. For

ment in innovation, society grants innovators some limited rights

example, those who download software code from an open source

to the innovations they generate via patents, copyrights, and trade

project repository can use it to learn about approaches to solving a

secrecy laws. These rights are intended to assist innovators in get-

particular software problem and/or they may reuse portions of the

ting private returns from their innovation-related investments. At

downloaded code by inserting it directly into a software program of

the same time, the monopoly control that society grants to innova-

their own. Von Krogh et al. (2004) studied the latter type of code

tors and the private profits they reap create a loss to society relative

reuse in open source software and found it very extensive. Indeed,

to the free and unfettered use by all of the knowledge that the in-

they report that most of the lines of software code in the projects

novators have created. Society elects to suffer this social loss in

they studied were taken from the commons of other open source

order to increase innovators' incentives to invest in the creation of

software projects and software libraries and reused.

new knowledge (Arrow 1962; Dam 1995).

331

In the case of academic publications, we see evidence that free

The second major model for inducing innovation is termed the col-

334

revealing does increase reuse---a matter of great importance to

lective action model. It applies to the provision of public goods,

academics. A citation is an indicator that information contained in

where a public good is defined by its non-excludability and non-

an article has been reused: the article has been read by the cit-

rivalry: if any user consumes it, it cannot be feasibly withheld from

ing author and found useful enough to draw to readers' attention.

other users, and all consume it on the same terms (Olson 1967).

Recent empirical studies are finding that articles to which readers

The collective action model assumes that innovators are required

have open access---articles available for free download from an au-

to relinquish control of knowledge or other assets they have devel-

thor's website, for example---are cited significantly more often than

oped to a project and so make them a public good. This require-

are equivalent articles that are available only from libraries or from

ment enables collective action projects to avoid the social loss as-

publishers' fee-based websites. Antelman (2004) finds an increase

sociated with the restricted access to knowledge of the private in-

in citations ranging from 45 percent in philosophy to 91 percent in

vestment model. At the same time, it creates problems with respect

SiSU

web.mit.edu/evhippel/www/

57

Democratizing Innovation

to recruiting and motivating potential contributors. Since contribu-

projects may engage in no active recruiting beyond simply posting

tions to a collective action project are a public good, users of that

their intended goals and access address on a general public web-

good have the option of waiting for others to contribute and then

site customarily used for this purpose (for examples, see the Fresh-

free riding on what they have done (Olson 1967).

meat.net website). Also, projects have shown by example that they

can be successful even if large groups---perhaps thousands---of

335

The literature on collective action deals with the problem of recruit-

contributors are involved. Finally, open source software projects

ing contributors to a task in a number of ways. Oliver and Marwell

seem to expend no effort to discourage free riding. Anyone is free

(1988) and Taylor and Singleton (1993) predict that the description

to download code or seek help from project websites, and no ap-

of a project's goals and the nature of recruiting efforts should mat-

parent form of moral pressure is applied to make a compensating

ter a great deal. Other researchers argue that the creation and de-

contribution (e.g., “If you benefit from this code, please also con-

ployment of selective incentives for contributors is essential to the

tribute . . .”).

success of collective action projects. For example, projects may

grant special credentials to especially productive project members

What can explain these deviations from expected practice? What,

337

(Friedman and McAdam 1992; Oliver 1980). The importance of

in other words, can explain free revealing of privately funded

selective incentives suggests that small groups will be most suc-

innovations and enthusiastic participation in projects to produce a

cessful at executing collective action projects. In small groups, se-

public good? From the theoretical perspective, Georg von Krogh

lective incentives can be carefully tailored for each group member

and I think the answer involves revisiting and easing some of the

and individual contributions can be more effectively monitored (Ol-

basic assumptions and constraints conventionally applied to the

son 1967; Ostrom 1998).

private investment and collective action models of innovation.

Both, in an effort to offer “clean” and simple models for research,

336

Interestingly, successful open source software projects do not ap-

have excluded from consideration a very rich and fertile middle

pear to follow any of the guidelines for successful collective action

ground where incentives for private investment and collective

projects just described. With respect to project recruitment, goal

action can coexist, and where a “private-collective” innovation

statements provided by successful open source software projects

model can flourish. More specifically, a private-collective model

vary from technical and narrow to ideological and broad, and from

precise to vague and emergent (for examples, see goal statements

recruit contributors. He simply expressed his private motivation in a message he

posted by projects hosted on Sourceforge.net).8

Further, such

posted on July 3, 1991, to the USENET newsgroup comp.os.minix (Wayner

2000): Hello netlanders, Due to a project I'm working on (in minix), I'm interested

8As a specific example of a project with an emergent goal, consider the

in the posix standard definition. [Posix is a standard for UNIX designers. A

beginnings of the Linux open source software project. In 1991, Linus Torvalds, a

software using POSIX is compatible with other UNIX-based software.] Could

student in Finland, wanted a Unix operating system that could be run on his PC,

somebody please point me to a (preferably) machine-readable format of the

which was equipped with a 386 processor. Minix was the only software available

latest posix-rules? Ftp-sites would be nice. In response, Torvalds got several

at that time but it was commercial, closed source, and it traded at US$150.

return messages with Posix rules and people expressing a general interest in

Torvalds found this too expensive, and started development of a

the project. By the early 1992, several skilled programmers contributed to Linux

Posix-compatible operating system, later known as Linux. Torvalds did not

and the number of users increased by the day. Today, Linux is the largest open

immediately publicize a very broad and ambitious goal, nor did he attempt to

source development project extant in terms of number of developers.

SiSU

web.mit.edu/evhippel/www/

58

Democratizing Innovation

of innovation occupies the middle ground between the private

7 Innovation Communities

341

investment model and the collective action model by:

It is now clear that users often innovate, and that they often freely

342

338

• Eliminating the assumption in private investment models that free

reveal their innovations.

But what about informal cooperation

revealing of innovations developed with private funds will rep-

among users? What about organized cooperation in development

resent a loss of private profit for the innovator and so will not

of innovations and other matters? The answer is that both flourish

be engaged in voluntarily. Instead the private-collective model

among user-innovators. Informal user-to-user cooperation, such

proposes that under common conditions free revealing of propri-

as assisting others to innovate, is common. Organized coopera-

etary innovations may increase rather than decrease innovators'

tion in which users interact within communities, is also common.

private profit.

Innovation communities are often stocked with useful tools and

339

• Eliminating the assumption in collective action models that

infrastructure that increase the speed and effectiveness with which

a free rider obtains benefits from the completed public good

users can develop and test and diffuse their innovations.

that are equal to those a contributor obtains.

Instead, the

private-collective model proposes that contributors to a public

In this chapter, I first show that user innovation is a widely dis-

343

good can inherently obtain greater private benefits than free

tributed process and so can be usefully drawn together by inno-

riders. These provide incentives for participation in collective

vation communities. I next explore the valuable functions such

action projects that need not be managed by project personnel

communities can provide. I illustrate with a discussion of free and

(von Hippel and von Krogh 2003).

open source software projects, a very successful form of innova-

tion community in the field of software development. Finally, I point

340

In summation: Innovations developed at private cost are often

out that innovation communities are by no means restricted to the

revealed freely, and this behavior makes economic sense for

development of information products such as software, and illus-

participants under commonly encountered conditions. A private-

trate with the case of a user innovation community specializing in

collective model of innovation incentives can explain why and

the development of techniques and equipment used in the sport of

when knowledge created by private funding may be offered freely

kitesurfing.

to all. When the conditions are met, society appears to have the

best of both worlds---new knowledge is created by private funding

User Innovation Is Widely Distributed

344

and then freely revealed to all.

When users' needs are heterogeneous and when the informa-

345

tion drawn on by innovators is sticky, it is likely that product-

development activities will be widely distributed among users,

rather than produced by just a few prolific user-innovators.

It

should also be the case that different users will tend to develop

different innovations.

As was shown in chapter 5, individual

users and user firms tend to develop innovations that serve their

particular needs, and that fall within their individual “low-cost

SiSU

web.mit.edu/evhippel/www/

59

Democratizing Innovation

The analogy to distributed user innovation is, of course, that each

347

user has a different set of innovation-related needs and other as-

sets in place which makes a particular type of innovation low-cost

(“shallow”) to that user. The assets of some user will then gener-

ally be found to be a just-right fit to many innovation development

problems. (Note that this argument does not mean that all innova-

tions will be cheaply done by users, or even done by users at all. In

essence, users will find it cheaper to innovate when manufacturers'

economies of scale with respect to product development are more

than offset by the greater scope of innovation assets held by the

collectivity of individual users.)

Available data support these expectations. In chapter 2 we saw ev-

348

idence that users tended to develop very different innovations. To

test whether commercially important innovations are developed by

just a few users or by many, I turn to studies documenting the func-

tional sources of important innovations later commercialized. As is

evident in table 7.1, most of the important innovations attributed to

users in these studies were done by different users. In other words,

user innovation does tend to be widely distributed in a world char-

acterized by users with heterogeneous needs and heterogeneous

stocks of sticky information.

Table 7.1 User innovation is widely distributed, with few users

349

developing more than one major innovation. NA: data not avail-

able.

Number of users developing this number of major innova-

350

tions

351

1

2

3

6

NA

Sample (n)

Scientific Instrumentsa

28

0

1

0

1

32

Scientific Instrumentsb

20

1

0

1

0

28

Process equipmentc

19

1

0

0

8

29

SiSU

web.mit.edu/evhippel/www/

60

Democratizing Innovation

innovation niches.” For example, a mountain biker who specializes

1

2

3

6

NA

Sample (n)

in jumping from high platforms and who is also an orthopedic

Sports equipmentd

7

0

0

0

0

7

surgeon will tend to develop innovations that draw on both of

these types of information: he might create a seat suspension

that reduces shock to bikers' spines upon landing from a jump.

a. Source: von Hippel 1988, appendix: GC, TEM, NMR Innova-

352

Another mountain biker specializing in the same activity but with a

tions.

different background---say aeronautical engineering---is likely to

b. Source: Riggs and von Hippel, Esca and AES.

draw on this different information to come up with a different inno-

c. Source: von Hippel 1988, appendix: Semiconductor and pultru-

vation. From the perspective of Fleming (2001), who has studied

sion process equipment innovations.

innovations as consisting of novel combinations of pre-existing

d. Source: Shah 2000, appendix A: skateboarding, snowboarding,

elements, such innovators are using their membership in two

and windsurfing innovations.

distinct communities to combine previously disparate elements.

Innovation Communities

353

Baldwin and Clark (2003) and Henkel (2004a) explore this type of

situation in theoretical terms.

User-innovators may be generally willing to freely reveal their in-

354

formation. However, as we have seen, they may be widely dis-

tributed and each may have only one or a few innovations to offer.

346

The underlying logic echoes that offered by Eric Raymond regard-

The practical value of the “freely revealed innovation commons”

ing “Linus's Law” in software debugging. In software, discovering

these users collectively offer will be increased if their information is

and repairing subtle code errors or bugs can be very costly (Brooks

somehow made conveniently accessible. This is one of the impor-

1979). However, Raymond argued, the same task can be greatly

tant functions of “innovation communities.”

reduced in cost and also made faster and more effective when it is

I define “innovation communities” as meaning nodes consisting

opened up to a large community of software users that each may

355

of individuals or firms interconnected by information transfer links

have the information needed to identify and fix some bugs. Un-

which may involve face-to-face, electronic, or other communica-

der these conditions, Raymond says, “given a large enough beta

tion. These can, but need not, exist within the boundaries of a

tester and co-developer base, almost every problem will be char-

membership group. They often do, but need not, incorporate the

acterized quickly and the fix obvious to someone. Or, less formally,

qualities of communities for participants, where “communities” is

`given enough eyeballs, all bugs are shallow.”' He explains: “More

defined as meaning“networks of interpersonal ties that provide so-

users find more bugs because adding more users adds more ways

ciability, support, information, a sense of belonging, and social

of stressing the program. . . . Each [user] approaches the task

identity” (Wellman et al. 2002, p. 4).9

of bug characterization with a slightly different perceptual set and

analytical toolkit, a different angle on the problem. So adding more

Innovation communities can have users and/or manufacturers as

356

beta-testers . . . increases the probability that someone's toolkit

9When they do not incorporate these qualities, they would be more properly

will be matched to the problem in such a way that the bug is shallow

referred to as networks---but communities is the term commonly used, and I

to that person.” (1999, pp. 41--44)

follow that practice here.

SiSU

web.mit.edu/evhippel/www/

61

Democratizing Innovation

members and contributors. They can flourish when at least some

used to create and maintain such software (the community-based

innovate and voluntarily reveal their innovations, and when others

development process).

find the information revealed to be of interest. In previous chapters,

Open Source Software

360

we saw that these conditions do commonly exist with respect to

user-developed innovations: users innovate in many fields, users

In the early days of computer programming, commercial “pack-

361

often freely reveal, and the information revealed is often used by

aged” software was a rarity---if you wanted a particular program for

manufacturers to create commercial products---a clear indication

a particular purpose, you typically wrote the code yourself or hired

many users, too, find this information of interest.

someone to write it for you. Much of the software of the 1960s and

the 1970s was developed in academic and corporate laboratories

357

Innovation communities are often specialized, serving as collection

by scientists and engineers. These individuals found it a normal

points and repositories for information related to narrow categories

part of their research culture to freely give and exchange software

of innovations. They may consist only of information repositories

they had written, to modify and build on one another's software,

or directories in the form of physical or virtual publications. For ex-

and to freely share their modifications. This communal behavior

ample, userinnovation.mit.edu is a specialized website where re-

became a central feature of “hacker culture.” (In communities of

searchers can post articles on their findings and ideas related to

open source programmers, “hacker” is a positive term that is ap-

innovation by users. Contributors and non-contributors can freely

plied to talented and dedicated programmers.10 )

access and browse the site as a convenient way to find such infor-

mation.

In 1969, the Defense Advanced Research Projects Agency, a part

362

of the US Department of Defense, established the ARPANET, the

358

Innovation communities also can offer additional important func-

first transcontinental high-speed computer network. This network

tions to participants. Chat rooms and email lists with public post-

eventually grew to link hundreds of universities, defense contrac-

ings can be provided so that contributors can exchange ideas and

tors, and research laboratories. Later succeeded by the Internet,

provide mutual assistance.

Tools to help users develop, eval-

it also allowed hackers to exchange software code and other in-

uate, and integrate their work can also be provided to commu-

formation widely, easily, and cheaply---and also enabled them to

nity members---and such tools are often developed by community

spread hacker norms of behavior.

members themselves.

The communal hacker culture was very strongly present among

363

359

All the community functionality just mentioned and more is visible

10hacker n. [originally, someone who makes furniture with an axe] 1. A person

in communities that develop free and open source software pro-

who enjoys exploring the details of programmable systems and how to stretch

grams. The emergence of this particular type of innovation com-

their capabilities, as opposed to most users, who prefer to learn only the

munity has also done a great deal to bring the general phenomenon

minimum necessary. 2. One who programs enthusiastically (even obsessively)

to academic and public notice, and so I will describe them in some

or who enjoys programming rather than just theorizing about programming. 3. A

person capable of appreciating hack value . 4. A person who is good at

detail. I first discuss the history and nature of free and open source

programming quickly. . . . 8. [deprecated] A malicious meddler who tries to

software itself (the product). Next I outline key characteristics of

discover sensitive information by poking around. Hence password hacker,

the free and open source software development projects typically

network hacker. The correct term for this sense is cracker (Raymond 1996).

SiSU

web.mit.edu/evhippel/www/

62

Democratizing Innovation

a group of programmers---software hackers---housed at MIT's

oneering idea was to use the existing mechanism of copyright law

Artificial Intelligence Laboratory in the 1960s and the 1970s (Levy

to this end. Software authors interested in preserving the status of

1984). In the 1980s this group received a major jolt when MIT

their software as “free” software could use their own copyright to

licensed some of the code created by its hacker employees

grant licenses on terms that would guarantee a number of rights to

to a commercial firm.

This firm, in accordance with normal

all future users. They could do this by simply affixing a standard

commercial practice, then promptly restricted access to the

license to their software that conveyed these rights. The basic li-

“source code”11 of that software, and so prevented non-company

cense developed by Stallman to implement this seminal idea was

personnel---including the MIT hackers who had been instrumental

the General Public License or GPL (sometimes referred to as copy-

in developing it---from continuing to use it as a platform for further

left, in a play on the word “copyright”). Basic rights transferred to

learning and development.

those possessing a copy of free software include the right to use it

at no cost, the right to study its source code, the right to modify it,

364

Richard Stallman, a brilliant programmer in MIT's Artificial Intelli-

and the right to distribute modified or unmodified versions to oth-

gence Laboratory, was especially distressed by the loss of access

ers at no cost. Licenses conveying similar rights were developed

to communally developed source code. He also was offended by

by others, and a number of such licenses are currently used in the

a general trend in the software world toward development of pro-

open source field. Free and open source software licenses do not

prietary software packages and the release of software in forms

grant users the full rights associated with free revealing as that term

that could not be studied or modified by others. Stallman viewed

was defined earlier. Those who obtain the software under a license

these practices as morally wrong impingements on the rights of

such as the GPL are restricted from certain practices. For exam-

software users to freely learn and create. In 1985, in response, he

ple, they cannot incorporate GPL software into proprietary software

founded the Free Software Foundation and set about to develop

that they then sell.12 Indeed, contributors of code to open source

and diffuse a legal mechanism that could preserve free access for

software projects are very concerned with enforcing such restric-

all to the software developed by software hackers. Stallman's pi-

tions in order to ensure that their code remains accessible to all

11Source code is a sequence of instructions to be executed by a computer to

(O'Mahony 2003).

accomplish a program's purpose. Programmers write computer software in the

form of source code, and also document that source code with brief written

The idea of free software did not immediately become mainstream,

365

explanations of the purpose and design of each section of their program. To

and industry was especially suspicious of it. In 1998, Bruce Perens

convert a program into a form that can actually operate a computer, source code

and Eric Raymond agreed that a significant part of the problem

is translated into machine code using a software tool called a compiler. The

compiling process removes program documentation and creates a binary

resided in Stallman's term “free” software, which might understand-

version of the program---a sequence of computer instructions consisting only of

ably have an ominous ring to the ears of businesspeople. Ac-

strings of ones and zeros. Binary code is very difficult for programmers to read

cordingly, they, along with other prominent hackers, founded the

and interpret. Therefore, programmers or firms that wish to prevent others from

open source software movement (Perens 1999). Open source soft-

understanding and modifying their code will release only binary versions of the

ware uses the licensing practices pioneered by the free software

software. In contrast, programmers or firms that wish to enable others to

understand and update and modify their software will provide them with its

source code. (Moerke 2000, Simon 1996).

12See www.gnu.org/licenses/licenses.html#GPL

SiSU

web.mit.edu/evhippel/www/

63

Democratizing Innovation

movement. It differs from that movement primarily on philosoph-

Conner and Prahalad 1996). In sharp contrast, all are offered

ical grounds, preferring to emphasize the practical benefits of its

free access to the source code of open source software if that

licensing practices over issues regarding the moral importance of

code is distributed by its authors. In early hacker days, this free-

granting users the freedoms offered by both free and open source

dom to learn and use and modify software was exercised by in-

software. The term “open source” is now generally used by both

formal sharing and co-development of code---often by the physi-

practitioners and scholars to refer to free or open source software,

cal sharing and exchange of computer tapes and disks on which

and that is the term I use in this book.

the code was recorded. In current Internet days, rapid technolog-

ical advances in computer hardware and software and network-

366

Open source software has emerged as a major cultural and

ing technologies have made it much easier to create and sustain

economic phenomenon.

The number of open source software

a communal development style on ever-larger scales. Also, imple-

projects has been growing rapidly. In mid 2004, a single major

menting new projects is becoming progressively easier as effective

infrastructure provider and repository for open source software

project design becomes better understood, and as prepackaged

projects, Sourceforge.net,13 hosted 83,000 projects and had more

infrastructural support for such projects becomes available on the

than 870,000 registered users. A significant amount of software

Web.

developed by commercial firms is also being released under open

source licenses.

Today, an open source software development project is typically

370

initiated by an individual or a small group seeking a solution to an

367

Open Source Software Development Projects

individual's or a firm's need. Raymond (1999, p. 32) suggests that

368

Software can be termed “open source” independent of how or by

“every good work of software starts by scratching a developer's

whom it has been developed: the term denotes only the type of

personal itch” and that “too often software developers spend their

license under which it is made available. However, the fact that

days grinding away for pay at programs they neither need nor love.

open source software is freely accessible to all has created some

But not in the (open source) world. . . .” A project's initiators also

typical open source software development practices that differ

generally become the project's “owners” or “maintainers” who take

greatly from commercial software development models---and

on responsibility for project management.14 Early on, this individual

that look very much like the “hacker culture” behaviors described

or group generally develops a first, rough version of the code that

above.

outlines the functionality envisioned. The source code for this initial

version is then made freely available to all via downloading from an

369

Because commercial software vendors typically wish to sell the

code they develop, they sharply restrict access to the source code

14“The owner(s) [or `maintainers'] of an open source software project are those

of their software products to firm employees and contractors. The

who have the exclusive right, recognized by the community at large, to

redistribute modified versions. . . . According to standard open source licenses,

consequence of this restriction is that only insiders have the infor-

all parties are equal in the evolutionary game. But in practice there is a very

mation required to modify and improve that proprietary code fur-

well-recognized distinction between òfficial' patches [changes to the software],

ther (Meyer and Lopez 1995; Young, Smith, and Grimm 1996;

approved and integrated into the evolving software by the publicly recognized

maintainers, and `rogue' patches by third parties. Rogue patches are unusual

13‹http://www.sourceforge.net›

and generally not trusted.” (Raymond 1999, p. 89)

SiSU

web.mit.edu/evhippel/www/

64

Democratizing Innovation

Internet website established by the project. The project founders

“try it and see” testing much more practical, because much less is

also set up infrastructure for the project that those interested in

at risk if a new contribution inadvertently breaks the code. Toolk-

using or further developing the code can use to seek help, provide

its used in open source projects have been evolved through prac-

information or provide new open source code for others to discuss

tice and are steadily being improved by user-innovators. Individual

and test. In the case of projects that are successful in attracting

projects can now start up using standard infrastructure sets offered

interest, others do download and use and “play with” the code--

by sites such as Sourceforge.net.

-and some of these do go on to create new and modified code.

Two brief case histories will help to further convey the flavor of open

Most then post what they have done on the project website for use

372

source software development.

and critique by any who are interested. New and modified code

that is deemed to be of sufficient quality and of general interest by

Apache Web Server Software

373

the project maintainers is then added to the authorized version of

Apache web server software is used on web server computers that

the code. In many projects the privilege of adding to the authorized

374

host web pages and provide appropriate content as requested by

code is restricted to only a few trusted developers. These few then

Internet browsers. Such 7 computers are a key element of the

serve as gatekeepers for code written by contributors who do not

Internet-based World Wide Web infrastructure.

have such access (von Krogh and Spaeth 2002).

The web server software that evolved into Apache was developed

375

371

Critical tools and infrastructure available to open source software

by University of Illinois undergraduate Rob McCool for, and while

project participants includes email lists for specialized purposes

working at, the National Center for Supercomputing Applications

that are open to all. Thus, there is a list where code users can report

(NCSA). The source code as developed and periodically modified

software failures (“bugs”) that they encounter during field use of the

by McCool was posted on the web so that users at other sites could

software. There is also a list where those developing the code can

download it, use it, modify it, and develop it further. When McCool

share ideas about what would be good next steps for the project,

departed NCSA in mid 1994, a small group of webmasters who

good features to add, etc. All of these lists are open to all and

had adopted his web server software for their own sites decided

are also publicly archived, so anyone can go back and learn what

to take on the task of continued development. A core group of

opinions were and are on a particular topic. Also, programmers

eight users gathered all documentation and bug fixes and issued

contributing to open source software projects tend to have essen-

a consolidated patch. This “patchy” web server software evolved

tial tools, such as specific software languages, in common. These

over time into Apache. Extensive user feedback and modification

are generally not specific to a single project, but are available on

yielded Apache 1.0, released on December 1, 1995.

the web. Basic toolkits held in common by all contributors tends

to greatly ease interactions. Also, open source software projects

In 4 years, after many modifications and improvements contributed

376

have version-control software that allows contributors to insert new

by many users, Apache became the most popular web server soft-

code contributions into the existing project code base and test them

ware on the Internet, garnering many industry awards for excel-

to see if the new code causes malfunctions in existing code. If so,

lence. Despite strong competition from commercial software de-

the tool allows easy reversion to the status quo ante. This makes

velopers such as Microsoft and Netscape, it is currently used by

SiSU

web.mit.edu/evhippel/www/

65

Democratizing Innovation

over 60 percent of the world's millions of websites. Modification

bulletin board set up for that purpose. Many responded by volun-

and updating of Apache by users and others continues, with the

teering information on bugs they had found and perhaps fixed, and

release of new versions being coordinated by a central group of 22

by offering improvements they had developed for their own use.

volunteers.

The quality of these suggestions was often high because “contri-

butions are received not from a random sample, but from people

377

Fetchmail---An Internet Email Utility Program

who are interested enough to use the software, learn about how

378

Fetchmail is an Internet email utility program that “fetches” email

it works, attempt to find solutions to the problems they encounter,

from central servers to a local computer. The open source project

and actually produce an apparently reasonable fix. Anyone who

to develop, maintain, and improve this program was led by Eric

passes all these filters is highly likely to have something useful to

Raymond (1999).

contribute.” (ibid., p. 42)

379

Raymond first began to puzzle about the email delivery problem

Eventually, Raymond arrived at an innovative design that he

381

in 1993 because he was personally dissatisfied with then-existing

knew worked well because he and his beta list of co-developers

solutions. “What I wanted,” Raymond recalled (1999, p. 31), “was

had used it, tested it and improved it every day. Popclient (now

for my mail to be delivered on snark, my home system, so that I

renamed fetchmail) became standard software used by millions

would be notified when it arrived and could handle it using all my

users. Raymond continues to lead the group of volunteers that

local tools.” Raymond decided to try and develop a better solution.

maintain and improve the software as new user needs and

He began by searching databases in the open source world for

conditions dictate.

an existing, well-coded utility that he could use as a development

base. He knew it would be efficient to build on others' related work

Development of Physical Products by Innovation Communi-

382

if possible, and in the world of open source software (then gener-

ties

ally called free software) this practice is understood and valued.

User innovation communities are by no means restricted to the

383

Raymond explored several candidate open source programs, and

development of information products like software. They also are

settled on one in small-scale use called “popclient.” He developed

active in the development of physical products, and in very similar

a number of improvements to the program and proposed them to

ways. Just as in the case of communities devoted to information

the then maintainer of popclient. It turned out that this individual