A Multimethod Approach for Creating New

Business Models: The General Motors

OnStar Project
Vince Barabba
General Motors Corporation, Corporate Strategy and Knowledge Development, 400 Renaissance Center, P.O. Box 400,
Detroit, Michigan 48265

Chet Huber Fred Cooke

General Motors Corporation, OnStar Headquarters, 1400 Stephenson Highway, Troy, Michigan 48083

Nick Pudar
General Motors Corporation, Corporate Strategy and Knowledge Development, 400 Renaissance Center, P.O. Box 400,
Detroit, Michigan 48265

Jim Smith
General Motors Corporation, OnStar Headquarters, 1400 Stephenson Highway, Troy, Michigan 48083

Mark Paich
Decisio, 320 West Cheyenne Road, Colorado Springs, Colorado 80906
[email protected] [email protected] [email protected] [email protected]
[email protected] [email protected]

We developed a multimethod modeling approach to evaluate strategic alternatives for GMs

OnStar communications system. We used dynamic modeling to address some decisions GM
faced in 1997, such as the companys choice between incremental and aggressive marketing
strategies for OnStar. We used an integrated simulation model for analyzing the new telematics industry, consisting of six sectors: customer acquisition, customer choice, alliances,
customer service, nancial dynamics, and dealer behavior. The modeling effort had important
nancial, organizational, and societal results. The OnStar business now has two million subscribers, an 80 percent market share of the emerging telematics market, and has been valued
at between $4 and $10 billion. The OnStar project set the stage for a broader GM initiative in
service businesses that ultimately could yield billions in incremental earnings. Most important,
OnStar has saved many lives that otherwise would have been lost in vehicle accidents.
(Industries: communications. Transportation: automotive.)

eneral Motors (GM) assembled a project team

consisting of the authors of this paper to develop
its OnStar telematics business. Telematics is the provision of communications services to cars, including
crash notication, navigation, Internet access, and trafc information. We used a multimethod modeling approach to design the OnStar business model and to
analyze the fundamental strategic decisions GM faced
in 1997. We explain the modeling process and some

Interfaces, 2002 INFORMS

Vol. 32, No. 1, JanuaryFebruary 2002, pp. 2034

specics of the model that we used to analyze the strategic choices, and we present the nancial, organizational, and social impacts the project created.
OnStar is GMs two-way vehicle communication system that provides a variety of services that enhance
safety, security, entertainment, and productivity (Figure
1). The vehicle communicates with either an automated
system, called the virtual advisor, or with a human advisor through a cell-phone connection. Two-way
0092-2102/02/3201/0020$05.00
1526-551X electronic ISSN

BARABBA ET AL.
The General Motors OnStar Project

Figure 1: OnStar provides safety and security, Internet, and communications services.

communication enables safety and security services,

such as crash notication, in which the call center is
notied if the vehicle crashes or the airbag is deployed.
A built-in global positioning system (GPS) precisely locates the vehicle and, if necessary, the call center dispatches emergency services to the accident scene.
Two-way communication facilitates a variety of
other services that provide information and enhance
the users productivity. For example, users can obtain
real-time trafc information through the virtual advisor in most major cities. In addition, they can access
content, such as the Wall Street Journal, and have it read
to them. Many other services have either been implemented or are currently in development.
OnStar began in 1994 as a promising communications technology. A GM engineering group proposed
Project Beacon to test the application of advanced communications technology in GM vehicles. To test the
Interfaces
Vol. 32, No. 1, JanuaryFebruary 2002

concept, in 1996, GM made OnStar available as an option on some Cadillac models. The services at that time
were limited to safety and security and a few other
features, such as remote door unlocking. The OnStar
system required complicated installation by the dealer,
costing about $1,300. The high cost and hassle of installation limited customer acceptance, but market research showed that buyers found the system extremely
valuable and that the customer retention rate was very
high. Some senior GM executives believed that with
the appropriate strategy, OnStar could become an important product.

Fundamental Decisions in 1997

In 1997, GM faced fundamental strategic decisions
with respect to OnStar. First, GM had to decide
whether to view OnStar as a car feature or as a service

21

BARABBA ET AL.
The General Motors OnStar Project

business. The default and safe strategy was to market

OnStar as a car feature that would improve vehicle
safety and security. GM had decades of experience
with new car features and had well-developed models
for pricing and packaging vehicle options.
An alternate strategy was to view OnStar as a service
business that could contribute greatly to GMs prots
(Figure 2). The OnStar business would collect monthly
subscription fees in exchange for a portfolio of services.
The business would be responsible for acquiring, developing, and retaining customers and would provide
customer service. In contrast with the vehicle business
in which GM interacts with customers infrequently,
the service business would put customers in touch
with GM with every use of OnStar.
GM would have to choose between an evolutionary
and a revolutionary strategy. If it decided to create an
OnStar service business with the evolutionary option,
GM would take a cautious approach to the telematics
market. In 1997, the telematics category barely existed,
and no vehicle manufacturer had invested in it aggressively. GM faced almost complete uncertainty
about technological approaches, major competitors,

and what competitive and complementary technologies would emerge among the Internet, digital cellular
services, and hand-held devices, such as PDAs.
A reasonable approach to such pervasive uncertainty
would be to postpone any major investments until the
picture became clearer. GM could run a portfolio of
small technological experiments to develop and preserve its options until the direction of the market became clearer. Once the situation was better dened, GM
could be a fast follower that could prot from the mistakes of bleeding-edge competitors. In the Internet eld,
for example, the blood of failed rst movers, such as
WordPerfect, CPM, and VisiCalc, and various Internet
companies is splattered all over the battleeld.
Specically, if GM took an overly aggressive approach, it might make large investments in vehiclecommunications hardware and infrastructure and
then fail to recover the costs because of low customer
demand. A good example of the failure of an aggressive strategy in communications is Iridiums launch of
satellites that cost several billion dollars. The system
never attracted many subscribers, and Motorola was
forced to write off most of its investment. On its face,

Figure 2: In 1997, GM faced fundamental strategic decisions for OnStar.

22

Interfaces
Vol. 32, No. 1, JanuaryFebruary 2002

BARABBA ET AL.
The General Motors OnStar Project

the conservative strategy is sensible, and indeed, both

Ford and DCX have taken a cautious approach; neither
will launch a telematics service until 2003.
On the contrary, in some cases in the automobile industry, the go-slow approach has served rms badly.
Auto companies missed the opportunity to be major
players in the cell-phone business even though 60 to 80
percent of cellular minutes are consumed in cars and
even though some auto companies understood the potential for cell phones years before demand exploded.
The cell-phone analogy is well known at GM. A second
example of go-slow failure was Detroits sluggishness
in responding to consumer demand for small, energyefcient, high-quality vehicles in the 1970s.
GMs second option was to choose the revolutionary
strategy of preempting the market. In this strategy, GM
would move quickly to build a large installed base and
gain the cost advantages and network externalities that
come from being the earliest and biggest player. Network externalities result from a process through
which, as more people adopt a service, the service becomes more valuable to both existing and potential
customers. There is evidence that the get-big-fast strategy can be very successful. The literature on rstmover advantages, while mixed, provides examples in
which preemptive strikes have generated big payoffs
(Gurumunrthy et al. 1995). In addition, economic analysis demonstrates that properly managed network effects can create a long-lived competitive advantage
(Shapiro and Varian 1999).
GM faced a strategic choice: should it follow an evolutionary wait-and-see strategy or should it make a
revolutionary move to develop and preempt the telematics market? The choice was difcult because the
market did not exist and data were scarce. In addition,
GM had minimal experience in subscriber service businesses, so its senior management had limited intuition
about which direction would be better. The preemptive strategy would require a huge investment; failure
of the preemptive strategy would be costly. On the
other hand, its success would bring large returns.

Modeling an Industry That Did

Not Exist
GM formed a project team (this papers authors) to
consider alternative strategies for OnStar. GM makes
Interfaces
Vol. 32, No. 1, JanuaryFebruary 2002

important strategic decisions through the dialogue decision process (DDP). DDP consists of four stages to
reaching agreement on decisions: framing, alternatives, analysis, and connection (Figure 3). At each step,
the project team interacts with the decision board that
is responsible for actually making the decision and
committing resources.
Dynamic modeling can be a part of each stage. For
example, in the alternatives phase, analysts often use

In 1997, the telematics category barely

existed.
models to suggest hybrid strategies that combine the
original alternatives. In the analysis phase, they use
dynamic models to calculate ranges for important
business variables, such as cash ow and market share.
The OnStar case was difcult to model. In the vehicle
business, GM has decades of experience and plentiful
historical data. Modelers can build on a wealth of previous analyses and examples of best practice. The
OnStar business was very different in that the telematics market did not exist and no one had experience
or historical data.
To cope with the inherent uncertainty, we needed a
modeling process that would allow integration of various methods and data sources. Our method had to be
exible enough to absorb a wide variety of inputs
based on judgment, historical analogies, market research, and other sources. Our chosen modeling approach integrated concepts and techniques from several management sciences approaches:
(1) System dynamics was useful in two important
ways. First, the stock-ow structure used in system dynamics provided a good foundation for the physics of
customer migration from one state to another. For example, the model included structures that tracked the
ow of customers from unawareness, to adoption, to
churn (loss of customers), and to possible resubscription. Second, the feedback perspective used in system
dynamics was very helpful in modeling the network
externalities that are crucial to the telematics market.
(2) Conjoint analysis was used to calibrate the consumer choice relationships that govern consumer
adoption of OnStar.
(3) Dynamic optimization was used to assess the

23

BARABBA ET AL.
The General Motors OnStar Project

Figure 3: Dynamic modeling is a part of the GM strategic decision process.

correct magnitudes for price and for spending on

marketing.
(4) Diffusion models and the literature on previous
studies of product diffusion helped us to understand
the impact of market spending, word of mouth, price,
and product innovation on OnStar adoption.
(5) Concepts from lifetime customer-value analysis
helped us to analyze the impact of churn and customer
recapture.
(6) The real-options approach was useful for decomposing the decision to expand OnStar into a set of
smaller, less costly steps. As it implemented each step,
GM had the option of taking the next step if the outcome of previous decisions was favorable.

The Integrated Simulation Model

A simulation model was our core tool in the OnStar
strategy project (Figure 4). It had six key sectors: customer acquisition, customer choice, alliances, customer

24

service (which we discuss in more detail), nances, and

dealer behavior. The nancial sector calculated nancial metrics, such as revenue, cash ow, and prot. The
dealer-behavior sector dealt with issues of how dealers
made customers aware of OnStar and how much sales
effort they expended. The sectors interact over time to
generate time series for such important business variables as market share and cash ow.

Customer Acquisition and Retention

The customer acquisition and retention sector of the
simulation model describes the inows and outows
of OnStar subscribers. The model builds on concepts
from the literature on lifetime customer value by explicitly representing the causal mechanisms responsible for subscriber acquisition and churn. Since the accumulated number of subscribers produces monthly
revenue directly through the monthly subscription fee,
we could use the model to evaluate the nancial
Interfaces
Vol. 32, No. 1, JanuaryFebruary 2002

BARABBA ET AL.
The General Motors OnStar Project

Figure 4: The simulation model of the telematics industry included multiple, interacting sectors.

impact of decisions that would affect GMs acquisition

and retention of subscribers.
The detailed model structure begins with the simple
relationship that determines the number of new
subscribers:
NS TR * V,

(1)

where NS is the number of new subscribers added during a time period, TR is the take rate (a fraction between zero and one), and V is the number of new vehicles on which OnStar is available. V can include both
GM and non-GM vehicles, depending on GMs policy
toward alliances with other vehicle manufacturers.
The modeling of the take rate (TR) begins with ideas
developed in the product diffusion literature. The take
rate is the product of customer awareness and customer choice:
Interfaces
Vol. 32, No. 1, JanuaryFebruary 2002

TR A*C,

(2)

where A is the fraction of new car buyers who have

top-of-mind awareness of OnStar, and C is the fraction
of new-car buyers who choose to subscribe to OnStar
conditional on awareness. The product-diffusion literature suggests that the two factors that drive awareness are the coefcients of internal and external inuence. The coefcient of internal inuence usually
represents the effect of word-of-mouth communication
on sales. We believed, and the data have subsequently
conrmed, that word-of-mouth would be an important
factor in generating awareness for OnStar.
Several published metastudies of estimated diffusion models were helpful in calibrating the word-ofmouth effect (Mahajan et al. 1995, Sultan et al. 1995).
We had no direct evidence about the magnitude of the
effect in the telematics market, but we used published

25

BARABBA ET AL.
The General Motors OnStar Project

studies to construct a range of estimates that we

thought were reasonable. The ranges were the basis for
extensive sensitivity analysis on word-of-mouth and
other uncertain parameters.
The coefcient of external inuence represents the
effect of the rms marketing effort on product adoption. In several product-diffusion studies, researchers
have modeled the external effect as an increasing function of spending on marketing. We used a similar approach and used the metastudies to calculate reasonable bounds for the magnitude of the effect. We also

We believed word-of-mouth would

be an important factor in generating
awareness.
consulted with internal GM marketing experts and
outside advertising experts to estimate how much
OnStar would have to spend to reach different levels
of awareness. Consistent with previous studies, the relationship between marketing spending and awareness exhibited diminishing returns.
The choice variable is the conditional probability
that a new-car buyer will subscribe given that OnStar
is available on the vehicle. The probability of the
buyers choosing OnStar depends on the utility of
OnStar relative to the utility of alternative uses of the
money and the utility of any available competing systems. We modeled the probability of choice with the
logit-choice function that is commonly used in marketing studies (Meyer and Johnson 1995). We derived
the utilities we used to calibrate the logit choice model
from a consumer research study.
The choice function included the effects of network
externalities. Writers on the economics of new-product
diffusion make a strong case for the importance of network externalities or positive-feedback effects (Shapiro
and Varian 1999). Positive-feedback effects are the increases in the value of the service for all existing users
as additional users adopt the service. We were aware
of these effects and asked an outside consulting rm
to search for examples relevant to the OnStar situation.
Its research turned up several examples that were useful primarily for identifying potentially important
positive-feedback processes. The examples were not

26

useful, however, for projecting the number of OnStar

subscribers or estimating revenue and cost ows.
We believed that three positive-feedback processes
could be important in the telematics industry. Each
process could create positive feedback, but their implementation required managerial attention and effort.
The rst process concerned creation of new OnStar applications. In 1997, OnStar was limited to the core
safety and security features. Although these features
were extremely valuable, the market research showed
they would not be enough to drive widespread penetration. From the beginning, we recognized that GM
could never nd and build internally the myriad applications that OnStar would need. Alliances with important new-economy and old-economy players
would be crucial.
We considered the factors that would make GM an
attractive partner to prospective content partners and
the factors that would make it economic for GM to
invest in partnerships. Examples of potential content
partners included Fidelity Investments and several
vehicle-insurance companies. For both old- and neweconomy rms, partnerships become much more attractive as the installed base of subscribers grows. The
classic example of how applications partnerships can
create positive feedback is Microsoft Windows. Applications developers wrote applications for Windows
that made Windows more valuable. Increased value
increased the number of customers buying Windows,
which, in turn, made developing further applications
more attractive.
We hypothesized that a similar process could occur
with OnStar. The economic dynamics of a recently announced OnStar alliance for providing real-time personalized trafc information demonstrates this process. Market research revealed that consumers want
personalized trafc information and that providing it
could be the telematics killer application. Trafc information requires both GM and its partner to make major up-front investments. Providing the trafc information feature is economically unattractive with a
small installed base because the average cost per subscriber would be much too high. The economics become very attractive as the installed base reaches several million. The alliance mechanism forms a positive
feedback because the availability of the trafc service
Interfaces
Vol. 32, No. 1, JanuaryFebruary 2002

BARABBA ET AL.
The General Motors OnStar Project

makes OnStar more attractive to car buyers. In turn,

additional subscribers make OnStar alliances more lucrative for all involved.
Quantifying the magnitude of the positive-feedback
processes was a challenge. We were tempted to conclude that we could not accurately quantify these
mechanisms and to leave them out of the formal modeling. We decided to try to quantify the alliance feedback, because historical examples suggested that they
could be critical to OnStars development, and because
the strength of the positive feedback is affected by
many other variables, such as pricing and spending on
marketing. Omitting the mechanisms from the model
would greatly distort the effects of alternative pricing
and marketing-investment policies.
To quantify these effects, we considered a long list
of potential services and partners. We used GM managers judgment and nancial data to estimate how the

The AMIC standards could be a

double-edged sword for OnStar.
number of subscribers would inuence the economics
of the different services. We used a combination of
market research and judgment to estimate how consumers would value additional services. We used sensitivity analysis to determine how the system would
react to different values of uncertain parameters.
The second positive-feedback process concerned the
dynamics of third-party sales of OnStar. Third-party
sales would be installations of OnStar through stereo
stores; electronics retailers, such as Circuit City and
Best Buy; and discount retailers, such as WalMart. Previous research reported in the marketing literature isolated a positive-feedback process in which products
with high sales and large market shares receive more
display space and attention at retailers, which, in turn,
further increases market share and sales (Reibstein and
Farris 1995).
Third-party distribution is the tool for reaching the
200 million existing vehicles and the 70 percent of new
car buyers who buy competitors cars. The viability of
third-party distribution depends on the cost and ease
of retrotting vehicles with OnStar hardware. We
already knew that replicating the existing dealerInterfaces
Vol. 32, No. 1, JanuaryFebruary 2002

installation procedure would be too costly and would

generate few sales. Also it would be time-consuming
and expensive to modify OnStar to interface with multiple car electronics systems. GM acted to reduce the
cost of third-party installation by sponsoring the Automotive Multimedia Interface Consortium (AMIC).
The AMIC is in the process of setting standards for
connecting to vehicle electrical systems. The standards
will enable any manufacturer of telematics systems to
connect to the vehicle electronics systems of any automobile without compromising the integrity and
safety of any of the systems.
The AMIC standards could be a double-edged
sword for OnStar. On one side, they enable businesses
to connect OnStar to competitors cars. On the other
side, they allow businesses to connect future competitors telematics systems with GM cars. The AMIC
standards increase the value of building a viable largescale telematics system because the consumer value
initially created for GM cars can be extended to other
platforms.
The third positive-feedback process concerns including other vehicle manufacturers in the OnStar alliance.
We believe that our project is one of the rst published applications to analyze the strategic implications of network effects in a real-life situation. Although the importance of network effects is clear, ex
post, from many historical case studies, few newproduct models actually incorporate them. Gupta et al.
(1999, p. 327) wrote the following:
Network effects have attracted signicant attention from
economists in recent years. However, marketing scientists
have been slow to respond to the growing importance of this
phenomena in new product adoption. For instance, most new
product models in the marketing science literature assume
that new products are autonomous and that the adoption of
new products is not affected by the presence or absence of
complementary products. These assumptions are being called
into question in almost every durable product market in the
Network Economy, where rms rarely act alone to create new
products.

Customer Choice
To calibrate the customer-choice sector, GM commissioned a conjoint study to estimate how consumers
would respond to different subscription fees, initial

27

BARABBA ET AL.
The General Motors OnStar Project

costs, and combinations of features (Reibstein and

Farris 1995). The sample for the conjoint study was 621
new-car buyers. In the conjoint study, researchers estimated the utility of 13 potential attributes of the
OnStar system, including route guidance, remote vehicle diagnostics, trafc information, initial price, and
monthly subscription fees.
The study showed that consumers could be divided
into six market segments. The segments had different
utilities for the service attributes and prices. In the customer acquisition sector, we calculated take rates separately for each market segment.
We used the market study to calibrate the consumerchoice decision. For example, we tested the impact of
different attribute combinations and prices on longterm OnStar penetration and protability. We also experimented with alternative price trajectories, such as
skim and penetration pricing. Skim pricing involved
installing OnStar on a few expensive GM models, such
as Cadillac, and charging a premium price for OnStar.
Penetration pricing involved installing OnStar on all
GM models and charging a low price that would maximize the take rate.

Vehicle Manufacturer Alliances

The option of offering OnStar to other vehicle manufacturers emerged early in the project. Clearly, enrolling other manufacturers could be benecial. First, increasing the vehicles in the alliance would create a
large OnStar installed base and strengthen positivefeedback processes. Second, GM could collect licensing
fees for the use of OnStar. The disadvantage of making
OnStar generally available would be that GM would
lose a competitive weapon for selling vehicles.
We evaluated the option of offering OnStar to other
manufacturers by including an additional positivefeedback process in the model. Manufacturers had four
options in the telematics market: do nothing, start their
own services, join the OnStar coalition, or join another
coalition. The probability that a manufacturer (other than
GM) would choose one of the options was given by
P f(S, V, T, C, F, M),

(3)

where S is the number of subscribers for a specic service (OnStar, Ford, and so forth), V is the number of

28

vehicles that offer a service, T is the take rate for a

service, C is the estimated investment cost of setting
up a telematics service, F is the fee charged by the coalition for using the service, and M is the manufacturer
that sponsors the coalition. In the model, we assumed
that the probability of choosing an existing service increased with increases of S, V, T, and C. A large base
of subscribers (S) and a high take rate (T) demonstrate
that the service is successful and will be more successful in the future. Other manufacturers would prefer to
enroll in a successful coalition. In addition, a high take
rate shows that consumers want the service and that
manufacturers without a service are at a competitive
disadvantage. If the sum of the Vs across coalitions is
large, most vehicles offer telematics services and the
holdouts are under pressure to join one of the coalitions. A high cost of establishing a service (C) makes it

The analysis showed that the costfocused strategy would cause the
effort to fail.
disadvantageous to create a new service and more advantageous, especially for small manufacturers, to join
a coalition. High fees (F) for participating in a coalition
reduce the probability that an outside manufacturer
will join. Finally, some manufacturers will be very hesitant to partner with other manufacturers, such as Ford
with GM, so that the identity of a coalition sponsor (M)
inuences the probability of an alliance.
Original equipment manufacturers (OEMs) benet
greatly by joining an OnStar coalition. First, replicating
the GM system would be very costly for auto manufacturers with much lower volumes than GM, especially if GM were able to exploit positive feedback and
add high-value services. Second, assuming that consumers nd telematics services attractive (market research supports the conclusion), if several competing
OEMs were to join the coalition, the holdout competitor could lose precious market share in the vehicle
business. Finally, if OnStar were to build a credible
third-party distribution system with the AMIC standard, OnStar would have access to the competitors
cars even if they didnt join the coalition. OEMs could
conclude that their best interests lie in joining the coalition and cutting the best deal possible, instead of
Interfaces
Vol. 32, No. 1, JanuaryFebruary 2002

BARABBA ET AL.
The General Motors OnStar Project

letting GM capture their customers through the

aftermarket.
For each major vehicle OEM, the team considered
the costs and benets of partnering with OnStar from
the perspective of that competitor. Our reasoning process was similar to that of estimating competitor payoffs in a game-theory analysis. During each time period, the model calculated the probability that a
manufacturer would choose one of the four options;
once a manufacturer chose to join a coalition or to start
its own service, it could not change its decision.
The alliance decision structure creates another
positive-feedback process. Additional partners increase the value of the system through multiple mechanisms, such as word-of-mouth and additional applications. In turn, a more valuable system attracts new
subscribers and additional partners. We did not believe that these processes would be so strong as to create a single system for the whole vehicle industry. We
acknowledged that some competitive automakers
would be so averse to a GM-sponsored system that
they would never join an alliance.

limiting the time spent per call, and always running at

close to full utilization. The model-based analysis
showed that the cost-focused strategy would cause the
entire OnStar effort to fail. OnStar depends on a staff
of intelligent, well-trained service personnel to provide
excellent service during difcult situations, such as car
wrecks and serious illness. It takes time to recruit and
train people who are up to the task. Consequently,
OnStar has adopted a policy of purposely overstafng
the call center in order to build customer-service capacity in advance of expected demand. The overstaffing policy gives service employees opportunities to
learn systems and scenarios before they have full
customer-service responsibilities. This policy is the
only one that is consistent with the strategy of building
a large installed base, and we estimate that it will pay
for itself several times over in terms of lower churn
rates and positive word-of-mouth.

Customer Service
The customer-service sector represented demands for
customer service and the acquisition and retention of
service capacity. Poor customer service could restrain
the long-run growth of OnStar. Rapid subscriber
growth increases the demand on the call centers.
OnStar must be able to match customer-service capacity to demand or, beyond a point, the quality of its
customer service will deteriorate. Common sense suggests and the literature on customer service conrms
that customers poor experiences with service reduce
the attractiveness of the service, reduce the rms acquisition of new subscribers, increase churn, and generate negative word-of-mouth.
A rm can minimize the negative effects of inadequate customer-service capacity by choosing the right
customer-service policy. Often, rms run their call centers with a cost mentality. Their objective is to minimize the cost of the call center by paying low wages,
Executive summaries of Edelman award papers are presented here. The complete article was
published in the INFORMS journal Interfaces [2002, 32:1, 20-34]. Full text is available by
subscription at http://www.extenza-eps.com/extenza/contentviewing/viewJournal.do?journalId=5
Interfaces
Vol. 32, No. 1, JanuaryFebruary 2002

29