RESEARCH ARTICLE

THE ECOSYSTEM OF SOFTWARE PLATFORM: A STUDY

OF ASYMMETRIC CROSS-SIDE NETWORK EFFECTS
1
AND PLATFORM GOVERNANCE
Peijian Song
School of Business, Nanjing University, Nanjing CHINA {songpeijian@nju.edu.cn}

Ling Xue
J. Mack Robinson College of Business, Georgia State University, Atlanta, GA 30303 U.S.A. {lxue5@gsu.edu}

Arun Rai
J. Mack Robinson College of Business, Georgia State University, Atlanta, GA 30303 U.S.A. {arun.rai@eci.gsu.edu}

Cheng Zhang
School of Management, Fudan University, Shanghai CHINA {zhangche@fudan.edu.cn}

In the context of software platforms, we examine how cross-side network effects (CNEs) on different platform
sides (app-side and user-side) are temporally asymmetric, and how these CNEs are influenced by the platform’s
governance policies. Informed by a perspective of value creation and capture, we theorize how the app-side
and the user-side react to each other with distinct value creation/capture processes, and how these processes
are influenced by the platform’s governance policies on app review and platform updates. We use a time-series
analysis to empirically investigate the platform ecosystem of a leading web browser. Our findings suggest that
while the growth in platform usage results in long-term growth in both the number and variety of apps, the
growth in the number of apps and the variety of apps only leads to short-term growth in platform usage. We
also find that long app review time weakens the long-term CNE of the user-side on the app-side, but not the
short-term CNE of the app-side on the user-side. Moreover, we find that frequent platform updates weaken
the CNEs of both the user-side and the app-side on each other. These findings generate important implications
regarding how a software platform may better govern its ecosystem with different participants.

Keywords: Software platform, two-sided markets, network effects, platform governance, value creation and
capture

Introduction1 devices, and software platforms (Evans 2011; Sriram et al.

Extant literature has categorized four types of platform busi-
nesses: exchanges, advertising-supported media, transaction
1
Arvind Malhotra was the accepting senior editor for this paper. Zhengrui
Jiang served as the associate editor.
The appendices for this paper are located in the “Online Supplements”
section of the MIS Quarterly’s website (http://www.misq.org).
2015). Software platforms, such as operating systems and
web browsers, are extensible codebases of software systems
that provide core functionalities for the applications that run
on them (Baldwin and Woodard 2009). They are emerging as
dominant models for software-based services and innovation
(Evans et al. 2006; Tiwana et al. 2010). Based on their tech-
nological architectures, software platforms operate as two-
sided markets to attract both end users and third-party devel-
opers that supply complementary apps (Boudreau 2012;

DOI: 10.25300/MISQ/2018/13737 MIS Quarterly Vol. 42 No. 1, pp. 121-142/March 2018 121
Song et al./Asymmetric Cross-Side Network Effects & Platform Governance
Ceccagnoli et al. 2012). Similar to other platforms, software
platforms are characterized by cross-side network effects
(CNEs).2
In the context of software platforms, CNE refers to the
phenomenon that the size and growth of one side, either the
user-side or app-side, reinforce the size and growth of the
other side (Anderson et al. 2014; Boudreau 2012). CNEs
differ from the traditional notion of network effects that focus
primarily on the mutual influences among network partici-
pants on a single side of a market and have strategic implica-
tions for the governance of a platform ecosystem with both
users and developers. First, CNEs influence a platform’s
strategies to build the installed base on either side. The
installed base is typically assessed using measures such as the
numbers of apps and users, the variety of apps, or the usage
behavior of users. For instance, a platform may subsidize one
side (e.g., free usage for users) and make the ecosystem
appealing to the other side (e.g., attract more third-party app
developers) (Hagiu 2006; Parker and Van Alstyne 2005; Weyl
2010). Second, CNEs can help incumbents by allowing them
to leverage their market advantages on one side by building
dominant positions on the other side and creating substantial
entry barriers. A software platform that starts with minor
leads on both sides is likely to win the entire market over
time, even if it is of inferior quality (Zhu and Iansiti 2012).
Therefore, understanding how the two sides mutually influ-
ence each other through CNEs is critical to the software
platform establishing competitive advantages by leveraging
CNEs.
In the extant empirical literature on CNEs, the focus is
primarily on investigating their presence and magnitude
(Sriram et al. 2015). The asymmetry among CNEs on dif-
ferent sides of the platform has received less empirical
attention. Some recent studies in other non-software platform
contexts have shown that CNEs on different sides can be
asymmetric in both signs and magnitude. For example,
Wilbur (2008) shows that, in television networks, increases in
viewers make networks more attractive to advertisers, but
increases in advertisements make networks less attractive to
viewers.3 In a setting of customer-to-customer (C2C) ex-
change platform, Chu and Manchanda (2016) show that the
2
Different labels have been used in the two-sided platforms literature to refer
to cross-side network effects—for example, indirect network effects, cross-
network externalities, cross-network effects, or cross-market network effects
(e.g., Chu and Manchanda 2016; Sriram et al. 2015; Zhu and Iansiti 2012).
3
It is worth noting that, in the television setting, the viewer-side is driven by
the demand for programs rather than advertisements. In contrast, in the
setting of software platforms, the user-side is driven by the demand for third-
party apps.
122 MIS Quarterly Vol. 42 No. 1/March 2018
growth of the seller-side has a larger CNE on the growth of
the buyer-side than vice versa. We examine the asymmetry
among CNEs but differ from previous studies in two respects.
First, we focus on the setting of software platforms, in which
platform participants (i.e., users and app developers) may
have distinct objectives, behaviors, and capabilities from
those in other platform contexts. Second, and more impor-
tantly, we consider CNE asymmetry in the temporal dimen-
sion. Specifically, we address the following research ques-
tion: How do the influences of the user-side on the app-side
(which we refer to as user-to-app CNEs) and the influences of
the app-side on the user-side (which we refer to as app-to-
user CNEs) in software platforms differ in terms of their long-
term and short-term effects? The examination of temporal
asymmetry is mainly motivated by the consideration that
CNEs evolve over time (see Clements and Ohashi 2005).
From a complex ecosystem perspective (Tanriverdi et al.
2010), uncovering the temporal dynamics of mutual adapta-
tion between two sides should help the platform owner adjust
the platform’s strategies on both sides and better manage
feedback loops.
In addition to CNEs, strategic actions and decisions about
platforms are examined in the extant empirical literature.
Prior studies, however, have focused primarily on issues of
pricing (e.g., Kaiser and Wright 2006; Liu 2010) and platform
competition (e.g., Dubé et al. 2010; Zhu and Iansiti 2012).
Less effort has been devoted to revealing how the governance
actions of a platform affect CNEs. Since platform governance
influences the ecosystem in which the two sides interact with
each other, its impact on CNEs is nontrivial and therefore
deserves research attention. We bridge this gap in the litera-
ture by addressing a second research question: How does
software platform governance affect the mutual influences
between the app-side and the user-side? In particular, we
differentiate from the existing literature on platforms by
focusing on two non-pricing governance policies of a soft-
ware platform: the duration of app reviews and the frequency
of platform updates.
Drawing on the theoretical lens of value creation and value
capture (Bowman and Ambrosini 2000; Lepak et al. 2007),
we consider how user-to-app CNEs and app-to-user CNEs
differ in value creation/capture processes and how these value
processes are influenced by software platform governance.
We carry out a time-series analysis using a unique dataset
about a leading web browser: Mozilla Firefox. We construct
a longitudinal weekly-level dataset spanning a period from
2008 to 2013. Using a vector auto-regressive (VAR) analysis,
we examine the dynamics of mutual influences between
browser usage and third-party app releases, and how these
mutual influences are affected by Firefox’s policies on app
review and platform updates. Our findings reveal temporal
 Song et al./Asymmetric Cross-Side Network Effects & Platform Governance
asymmetry between user-to-app CNEs and app-to-user CNEs:
specifically, growth in the quantity and variety of apps and
variety of apps drives end-user platform usage in the short-
term rather than in the long-term. In contrast, growth in end-
user platform usage drives growth in the quantity and variety
of apps in the long-term rather than in the short-term. These
results advance our understanding of the temporal dynamics
underlying the interdependence between the two sides of a
software platform.
We also find that the app review time influences the app-side
and the user-side differently. In particular, long app review
time significantly weakens the long-term user-to-app CNEs.
However, we do not find a significant impact from long app
review time on the short-term app-to-user CNEs. In addition,
frequent platform updates significantly weaken the long-term
user-to-app CNEs. However, regarding the short-term app-to-
user CNEs, frequency platform updates significantly weaken
the effect of app quantity, but not that of app variety, on end
users’ platform usage. These findings generate important
implications as to how the management of a software platform
can be improved.
Theory and Hypothesis Development
CNEs in Software Platform and
Value in Network Effect
In the platform literature, CNEs have long been concep-
tualized as the way in which the attractiveness of the platform
on one side is dependent on the size of the other side (Rysman
2009; Weyl 2010). In the context of software platforms,
positive CNEs manifest when growth in the size of the user
base helps the platform motivate more third-party app devel-
opment, and when the growth in size of the app base helps the
platform attract more users. However, recent research on
network effects suggest that network effects are not all about
size, but are also determined by the nature of network value
and the value creation/capture processes of network parti-
cipants (Afuah 2013). Therefore, to better conceptualize
CNEs based on value, we adopt the theoretical lens of value
creation/capture from the strategy literature (Bowman and
Ambrosini 2000; Lepak et al. 2007), which has been
employed in recent studies on network effects (Afuah 2013),
and contextualize this perspective to software platforms.
The perspective of value creation/capture distinguishes
between two types of value to various stakeholders (e.g., indi-
viduals, organizations, and society): use value and exchange
value (Bowman and Ambrosini 2000; Lepak et al. 2007). Use
value refers to the quality of a product or service as perceived
by users in relation to their needs. Exchange value refers to
the tangible or intangible monetary value that can be realized
by the seller when the exchange of a product or service takes
place, or the payment by the user to the seller for the use
value of the product or service. When the exchange does not
entail monetary payment (e.g., in the case of free software),
exchange value may be reflected in intangible benefits (such
as reputation gains) to the seller. The distinction between use
value and exchange value can be illustrated using the example
of a typical security protection app in Firefox, Permit
Cookies. As one of the earliest apps in Firefox, the app
allows end users to explicitly permit (or deny) the use of
cookies by the sites they visit. When Permit Cookies was
available in Firefox, Firefox users could directly use the app
to derive value from its functionalities, and this value capture
process did not require much additional effort by users
(except an easy procedure for installation). Likewise, when
other competing security protection apps were made available
in Firefox, directly using these apps allowed users to derive
value from their additional security protection functionalities.
From the standpoint of the developers of Permit Cookies, the
emergence of a large number of Firefox users with security
protection needs provided them with a potential market. They
could not derive value directly by simply accessing or “using”
these Firefox users through the platform. Instead, they
needed to undertake various time/effort-consuming processes
to identify app functionalities that serve user needs and even-
tually realize them in Permit Cookies. For example, detecting
a potential vulnerability in cookie use and implementing func-
tionalities to disable cookies from designated sites were
among the processes that developers of Permit Cookies under-
took for their product to be eventually favored by Firefox
users. Only when Firefox users adopted and liked the app
could developers derive value (e.g., in reputation) as an ex-
change for the app they provided. Likewise, for developers
of competing security protection apps, they could only derive
value (tangible payments or intangible reputation) from
Firefox users after the developers invested effort to identify
and eventually realize other security protection functionalities
in their apps provided to users for exchange.
Based on the theoretical distinction between use value and
exchange value, we distinguish between the CNEs that the
user-side has on the app-side (i.e., user-to-app CNEs) and the
CNEs that the app-side has on the user-side (i.e., app-to-user
CNEs). In user-to-app CNEs, app developers find the soft-
ware platform a more valuable place to participate when there
are more platform users into which they can tap. Therefore,
user-to-app CNEs are essentially based on exchange value, as
app developers do not derive value from directly “using”
platform users but derive value from serving users in ex-
change for their tangible or intangible payments. In contrast,
MIS Quarterly Vol. 42 No. 1/March 2018 123
Song et al./Asymmetric Cross-Side Network Effects & Platform Governance
in app-to-user CNEs, users find the platform a more valuable
option to use when there are more complementary apps for
them to use on the platform. Therefore, app-to-user CNEs are
essentially based on the use value, as users can derive value
(e.g., perceive aesthetic or functional quality) from directly
using these apps.
The distinct value bases of user-to-app CNEs and app-to-user
CNEs also lead to distinct processes of realizing and cap-
turing value by the two sides. For the app-side, since it can-
not directly “use” the platform’s users, it needs a process of
converting user interest in the platform to interest in apps.
Such a process should entail identifying user demand (for
certain apps), developing apps, and making apps satisfactory
to users so that exchange value can be generated. For the
user-side, as apps available on the platform are ready to be
used, end users need minimum additional processes other than
using these apps to derive use value. Such a distinction in
value creation/capture processes on the two sides can even-
tually lead to temporal asymmetric features of user-to-app
CNEs and app-to-user CNEs.
Temporal Asymmetry in CNEs
Our characterization of the temporal features of CNEs is
based on the concepts of short-term effect and long-term
effect on market participants as discussed in the marketing
literature (Dekimpe and Hanssens 1995; Pauwels and Weiss
2008). Conceptually, a short-term effect refers to the case in
which market participants exhibit immediate reactions to
certain marketing events or actions. A long-term effect, in
contrast, refers to the case in which market participants’
reactions to marketing events or actions are persistent over
time. Extending these concepts to our context of software
platforms, we consider a short-term effect as the immediate
response in growth of one side (either the app-side or the
user-side) to the growth of the other side. Likewise, we con-
sider a long-term effect as the persistent response in growth of
one side to the growth of the other side. Conceptually, short-
term effect and long-term effect are not mutually exclusive.
For example, if the growth-fostering effect of one side on the
other side takes effect immediately and is also persistent over
time, such a CNE is characterized by both a short-term effect
and a long-term effect.
Considering the process of realizing exchange value by the
app-side, we expect that the response of the app-side to the
user-side takes effect in the long-term rather than in the short-
term. This assertion is supported by the theoretical perspec-
tive of commitment in the strategy literature (Ghemawat
1991). On the one hand, using app development to realize
exchange value is a process of resource commitment, which
124 MIS Quarterly Vol. 42 No. 1/March 2018
is not a simple process that can be done instantaneously.
Developers need to figure out platform users’ underlying de-
mand for specific types of apps, organize required resources
based on their own expertise and constraints, engage in actual
development, and make their products accessible to users.
Many processes make it difficult for developers to respond
immediately to the growth of a platform’s user base by
providing ready-to-use apps. For example, it may not always
be a straightforward process to identify the user demand for
specific apps (Eisenmann et al. 2011). Developers may be
heterogeneous in terms of their resource constraints and
development capabilities, making it less likely for the app-
side as a whole to quickly respond with a large number of
apps (Boudreau 2012). Upon completion, developers also
have to go through the app review processes imposed by the
platform, which further delays the release of apps (Ghaz-
awneh and Henfridsson 2013). As a result, the responses of
the app-side to the user-side are less likely to take effect
immediately, and therefore user-to-app CNEs are less likely
to manifest as a short-term effect.
On the other hand, the resource commitment perspective sug-
gests a tendency of persistent app development as a response
to the user-side growth. When developers make development
decisions, they are less likely to give up their committed
development plans for several reasons. Irreversible invest-
ments in specialized resources (e.g., development teams) can
lock developers into development processes (Harrigan and
Porter 1983). Difficulty in redeploying allocated resources
(e.g., specialized tools) can lock out developers from alter-
native opportunities. Lags in adjusting development stra-
tegies often constrain developers from deviating from current
development plans. Organizational inertia built over time can
also lead to developers to having built-in biases that maintain
the status quo of development plans (Harter and Slaughter
2003). As a consequence, although app development and app
releases are less likely to be completed instantly, they will
likely occur eventually, but with a time lag. The growth in
the app base will therefore be realized with a delay in time
and in the long-term. The heterogeneity of developers in their
development pace may even smooth this long-term trajectory
of growth. Therefore, we hypothesize that
H1a: User-to-app CNEs in a software platform are char-
acterized by a significant long-term effect, but with
no significant short-term effect.
In contrast, the realization of use value by the user-side is a
relatively simple process and requires many fewer resource
commitments by end users. When apps are available on the
platform, users just need to respond by using the platform to
try the apps that interest them. Users do not need any extra
value-creation process to capture the use value and therefore
 Song et al./Asymmetric Cross-Side Network Effects & Platform Governance
their responses do not have to be delayed. Some other factors
may further expedite user-side responses. For example, app
centers and search functions offered by software platforms get
newly released apps quickly exposed to users (Haigu and
Spulber 2013). Word-of mouth generated on the user-side
also helps draw new users to the platform to quickly try new
apps (Zhang et al. 2012). Therefore, the response of the user-
side to app-side growth is likely to take effect immediately.
Such a response should result in a quick increase in platform
usage, based on which users can use apps. Consequently,
app-to-user CNEs are likely to manifest a significant short-
term effect.
App-to-user CNEs, however, are less likely to be charac-
terized by a significant long-term effect. The rationality of
this assertion can be theoretically justified by the wear-out
effect, which the marketing literature conceptualizes as the
quick decay in the effectiveness of marketing-related actions
over time (Bass et al. 2007). Prior research has found that the
effectiveness of various marketing actions, such as advertising
and product-line extension, can die out quickly (e.g., after one
or two weeks) (Leone 1995; Pauwels 2004). The growth of
the app-side is a product-line extension process on the plat-
form. In this case, the wear-out effect can be explained by
various factors. First, changes in user knowledge may reduce
user interest in new apps (Calantone and Sawyer 1978), and
the easy conditions of trial apps and available online informa-
tion can expedite this process. Second, multihoming allows
competing platforms to offer similar apps (Evans 2011),
which can generate a so-called “copy wear-out effect” (Naik
et al. 1998). Third, rapid app development and expansion on
a platform may also generate cognitive challenges to users.
The user tendency of forgetting (Bass et al. 2007) can con-
tribute to the decreasing attractiveness of new apps over time.
Proliferation of available options may also pose challenges for
users to identify apps that truly fit their needs and therefore
affect the long-term commitment of users (Grime et al. 2002).
Consequently, even if the increase in new apps draws users to
the platform, short-term user interests are not enough to guar-
antee users’ long-term commitment to the platform. Fourth,
many apps have short lifespans and may only be appealing to
users in the short-term. For example, from tracking over
830,000 iOS and Android apps, Flurry Insights found that
only 4 of 19 categories of iOS and Android apps (e.g.,
weather, health and fitness, news and magazines) have a
median user-retention rate above 25% after 30 days (Klotz-
back 2016). Most apps only incur short-term user interest,
making app-to-user CNEs unlikely to lead to long-term user
commitment to the platform. We therefore hypothesize that
H1b: App-to-user CNEs in a software platform are char-
acterized by a significant short-term effect, but no
significant long-term effect.
Software Platform Governance
Platform governance in software platforms refers to all
policies and mechanisms through which a software platform
operator exerts influence over participants on both sides and
coordinates operations in the ecosystem (Tiwana et al. 2010).
In general, platform governance includes three broad cate-
gories of policies and mechanisms. First, although pricing is
often linked with profitability, it is also an inherent gover-
nance mechanism. Decisions regarding which side to subsi-
dize and the use of pricing as an incentive mechanism all
influence the interaction between the two sides (Rochet and
Tirole 2006; Tiwana 2014). Second, mechanisms used to
coordinate and control platform participants are the central
elements of platform governance. Third, some policies of a
platform’s self-development play an important role in gover-
nance (Cusumano and Gawer 2002). The design and
improvements in a platform’s own architecture and features
shape the general environment in which cross-side inter-
actions take place. Therefore, certain self-development of the
platform (e.g., APIs) cannot be completely separate from
platform governance (Tiwana 2014).
We differ from prior empirical studies on platform gover-
nance (which have primarily considered pricing and platform
competition issues) by focusing on two non-pricing mech-
anisms in software platform governance and considering how
they influence CNEs. The first mechanism is app review,
which reflects the way in which the platform addresses gover-
nance through regulation and control of third-party devel-
opers’ development behavior (Ghazawneh and Henfridsson
2013; Maurer and Tiwana 2012). We specifically examine
the duration of app review processes. Platforms often impose
standards on the content and technical specifications of third
party apps. The review processes ensure that apps meet
performance standards, reliability, and compatibility. For
example, Apple’s App Store imposes a wide range of rules
that cover a range of requirements from user interface design,
functionality, and content to the use of specific technologies.
App review by the App Store can take up to two weeks. Sub-
mitted apps can be rejected for various reasons, such as lack
of information, functional errors, and poor interface design
(Starr 2014). Developers who fail app review are required to
the fix identified problems before resubmitting their apps.
Reviews on the revised versions further delay final app
releases.
The second mechanism we consider is updates to the platform
itself, and we specifically examine the frequency of platform
updates. Since the platform serves as the basis for cross-side
interactions, platform updates are an important way for the
platform managers to address governance by improving the
platform’s design and architecture (Tiwana 2014). Platform
MIS Quarterly Vol. 42 No. 1/March 2018 125
Song et al./Asymmetric Cross-Side Network Effects & Platform Governance
updates often need both the app-side and the user-side to
make adaptive actions, such as conforming to changing APIs
and addressing compatibility issues. In this regard, platform
updates should also influence how the two sides respond to
each other through CNEs.
Software Platform Governance and CNEs
In user-to-app CNEs, the app-side responds to user-side
growth mainly by creating and capturing the exchange value
of apps. Given the growth of the user-side, however, long app
review time can generate three effects that weaken the app-
side response in growth. First, long app review time directly
delays the release of new apps (Ghazawneh and Henfridsson
2013), making app developers less able to respond in a timely
fashion to user demands for certain types of apps. For ex-
ample, in a recent survey, iOS app developers revealed that
they often have to wait more than a week for Apple to ap-
prove the fixing of bugs in their original app versions. Devel-
opers expressed concern that these time-consuming review
processes undermine their ability to respond to the expanding
market and satisfy the needs of a growing user base (Graham
2016).
Second, long review times can escalate the burden on devel-
opers and result in more uncertainties in developers’ resource
commitments. Long review times are often associated with
more thorough reviews, which require developers to carry out
more demanding revisions of their original app versions.
Developers need to allocate more resources and adjust their
other development plans to complete revisions. For example,
iOS developers reported that they often plan for up to a month
of slack time in their app release plans due to the long and
demanding review processes (Graham 2016). Consequently,
developers who have constrained resources or are uncertain
about whether they are able to make satisfactory revisions are
also more likely to forgo their app development. The app-side
can therefore be constrained from developing apps in re-
sponse to user-side growth because of long app review times.
Third, long app review processes may negatively affect
developers’ resource commitment and incentives for future
development (Hilkert et al. 2010). With long app review
times, the platform generates high opportunity costs for
developers in their future development. When developers
anticipate these opportunity costs that are generally out of
their control, they will be more reluctant to commit future
resources to app development for the platform. Given com-
peting development alternatives, the length of a focal plat-
form’s review time can drive developers to switch to
alternatives. As a result, the app-side will be less able to
126 MIS Quarterly Vol. 42 No. 1/March 2018
respond to user-side growth with a steady increase in app
development over time. Considering these effects, we expect
that long app review times weaken user-to-app CNEs.
H2a: User-to-app CNEs in a software platform are
weakened by long app review time.
In app-to-user CNEs, the user-side responds to app-side
growth mainly by capturing the use value of apps. Long app
review times can negatively affect user interest in released
apps. First, by delaying the release of apps, time-consuming
review processes can make apps obsolete and reduce their use
value. Given the volatility of users’ online tastes (Sun 2012),
delayed apps can become less attractive to users. For
example, a developer team previously submitted an app for
Apple TV app about a week before the release of Apple TV.
However, a two-week app review time eventually resulted in
an app being barely noticed by users because the excitement
for Apple TV apps had already died down by the time some
apps were released (Graham 2016). Second, long app review
times can slow down developers’ processes for releasing
patches for bugs reported by users (Dave 2015). As a result,
users get hurt due to the reduced reliability of apps. When
users lose interest in apps, they also lose the incentive to
participate in the platform and therefore app-to-user CNEs are
weakened. We therefore hypothesize
H2b: App-to-user CNEs in a software platform are
weakened by long app review time.
In juxtaposition to H2b, we take into consideration some
mechanisms in app review that make long app review times
strengthen app-to-user CNE. Long app review times are
likely to be associated with more thorough and diligent
reviews, which should help improve the quality of apps and
their use value (Fisher and Oberholzer-Gee 2013). Thorough
app reviews can help correct more errors or potential weak-
nesses in submitted apps, and also ensure that apps better
conform to platform standards and generate complementary
value. For example, when a formal app review process was
absent, users on Google Play were more likely to encounter
apps that did not live up to their expectations and violated
Play Store policies. This motivated Google to formalize its
processes for app review and approval (Arce 2015). App
quality improvement driven by thorough app reviews can also
make apps more valuable to users. With more valuable apps,
the platform may also become more attractive to users and
app-to-user CNEs can therefore be reinforced. We therefore
develop a competing hypothesis to H2b:
H2c: App-to-user CNEs in a software platform are
strengthened by long app review time.
 Song et al./Asymmetric Cross-Side Network Effects & Platform Governance
As a software platform serves as the base infrastructure on
which developers build their apps, the software platform’s
frequent update policy can be justified from the perspective of
quality improvement. Frequent updates suggest that the
platform is constantly improved and adapted to the changing
environment by fixing issues and incorporating advanced
features (Arora et al. 2006). As a result, developers are more
likely to build confidence about the long-term robustness and
adaptiveness of the platform infrastructure, which incentivizes
their development decisions. Frequent platform updates
therefore can have a reinforcing effect on the user-to-app
CNE. When developers observe a user-side growth and they
also trust the reliability of the platform, they will be more
willing to develop more apps to appeal to the increasing user
base. Accordingly, we hypothesize that
H3a: User-to-app CNEs in software platform are
strengthened by frequent platform updates.
In juxtaposition to H3a, we take into consideration potential
tradeoffs in the frequent releases of platform updates. Despite
the benefits of platform updates, the frequent releases of up-
dates may also generate counterproductive effects. Table 1
illustrates the tradeoffs by summarizing the benefits of
platform updates and the downside of frequent releases of
platform updates. Updates on the platform inevitably change
the basis on which users and developers interact with each
other, and the frequent introduction of these changes may
constrain user-to-app CNEs for several reasons. First, devel-
opers have to adjust their app development to the new
platform functionalities (Boudreau 2010). If the updates are
released over-frequently, some developers that are less
capable may lack a sufficient time buffer to adapt (Boudreau
2012). Recent research also suggests that generational tran-
sitions initiated by platforms often make it difficult for com-
plementors to sustain their performance (Kapoor and Agarwal
2016). Consequently, the app-side may not be able to respond
effectively to the user-side growth in a frequently changing
platform environment, weakening the user-to-app CNE.
Second, frequent platform updates can introduce uncertainties
for developers about constantly changing platform configu-
rations (Mullins and Sutherland 1998) and the extra resources
they should commit to adapt. Developers need a time buffer
to resolve these uncertainties. The frequent releases of
updates can escalate developers’ concern about uncertainties
about the platform’s configurations and disincentivize their
commitment to the platform. Such disincentives can also
undermine the app-side response to the user-side growth and
weaken user-to-app CNEs. Third, platform updates can affect
the accessibility of older apps and force developers to commit
resources to app updates (Kingsley-Hughes 2012). Frequent
platform updates can therefore create more technical debt
(Ramasubbu and Kemerer 2015), that is, a buildup of software
maintenance obligations over time. Such technical debt for
developers may constrain their resource commitment for new
app development (Bavota et al. 2015), and weaken the app-
side expansion to serve the growing demand from the user-
side.
Moreover, releasing updates frequently can put the focal
platform at competitive disadvantage, especially when
developers have multihoming options (Evans et al. 2006;
Hyrynsalmi et al. 2016). The frequent changes of the focal
platform can make it appear to be less stable than competing
platform environments. Developers who prefer more manage-
able adaptation burdens and limited uncertainties about the
platform configuration may find the platform’s competitors
more attractive. The availability of multihoming options
further makes it more difficult for the focal platform to use its
user base to keep and expand its developer base (Zhu and
Iansiti 2012). Therefore, frequent platform updates are likely
to negatively affect user-to-app CNE in a competitive envi-
ronment. Considering the countervailing forces that frequent
platform updates may generate (i.e., the benefits of updates
based on which we formulated H3a and the downside of fre-
quent updates that we enumerated), we develop the following
competing hypothesis to H3a:
H3b: User-to-app CNEs in a software platform are
weakened by frequent platform updates.
Frequent platform updates can also generate countervailing
effects on how the user-side reacts to the app-side growth
(i.e., app-to-user CNEs). As Table 1 indicates, platform up-
dates may benefit users by providing them additional platform
features and an enhanced environment to use apps. In this
regard, platform updates may reinforce app-to-user CNEs
over time. However, the frequent releases of platform updates
can also be detrimental to the user-side. First, frequent up-
dates of the platform may create usage interruptions for
individual users. For instance, Firefox updates often require
restarting the browser, introduce backward incompatibility
with old add-on apps (Mandel 2012), and even change the
user interface (Kingsley-Hughes 2012). When these issues
occur too frequently, they can disrupt end users’ habitual
usage. Users are also forced to commit additional cognitive
resources to adapting to the changes. Users with lower capa-
bilities to overcome the usage interruption are likely to be
discouraged from continuing to use the apps on the platform,
leading to weakened app-to-user CNEs. This logic is in line
with insights from research on software updates in the
working environment (such as the enterprise software envi-
ronment). For example, update reminders, alerts, notifica-
tions, and installations can cause interruptions in workflow
(Hodgetts and Jones 2007), and these system-initiated inter-
MIS Quarterly Vol. 42 No. 1/March 2018 127
Song et al./Asymmetric Cross-Side Network Effects & Platform Governance

Table 1. Tradeoffs of Frequent Platform Updates

Benefits of Updates Downside of Frequent Update Releases
(1) Less time buffer for developer adaptation
(1) Constant improvement of the
(2) Developers’ uncertainties about the changing platform
platform infrastructure
App-Side configurations
(2) Adaption to the changing
(3) More technical debt and resource commitments to updating
external environment
existing apps
(1) Interruption to user habitual usage
(1) More platform features
(2) Additional cognitive resources required for users to adapt to
User-Side (2) Enhanced environment for app
changes
usage
(3) Negative impact on user-side externality

ruptions are a major impediment to the productivity of IT
users in the workplace (Ortiz de Guinea and Webster 2013).
Also, system updates that do not introduce new technical
features are generally not favored by users and may even
affect their continued system usage (Fleischmann et al. 2016).
Second, the usage interruptions can also lead to reduced
externality among users themselves. Updates inevitably make
some users lag in adaption and gradually lose momentum of
using the platform (Venkatesh et al. 2008). This causes tem-
porary shrinking of the user base for word-of-mouth (WOM)
of apps. When platform updates occur too frequently, this
temporary shrinking can accumulate and eventually hurt how
apps attract users through WOM. As a result, app-to-user
CNEs can become weaker. Considering the tradeoffs, we
expect that frequent platform updates generate an overall
weakening effect on app-to-user CNEs. The main reason is
that we expect app-to-user CNEs to take place primarily in the
short-term (as in H1b). The benefits of platform updates to
users may not manifest immediately in the short-term as they
also require user adaption processes. In contrast, usage
disruption and the shrinkage of user WOM base are likely to
be caused immediately by usage barriers. Thus, we expect
that the downsides of the frequent releases of platform
updates dominate the benefits of updates in the short-term.
We therefore hypothesize
H3c: App-to-user CNEs in software platform are
weakened by frequent platform updates.
Methodology
Research Context
Our empirical setting is a major web browser: Mozilla’s
Firefox. Since its debut in late 2004, Firefox remained the
second-most used web browser until late 2011 (when Google
Chrome surpassed it). Currently, Firefox is the third most
widely used web browser, with approximately 16% in market
share. As an open-source system, Firefox relies more on a
platform ecosystem to sustain its operations than do other
browsers. It extends its product boundaries by encouraging
a large number of third-party developers to supply comple-
mentary apps. These apps provide add-on functionalities,
such as videoconferencing and privacy protection, within
Firefox’s core framework.
To ensure app quality and user experience, Firefox mandates
most hosted apps to be reviewed by a team of editors. Devel-
opers are required to upload their apps to the developer hub
and provide related information, such as descriptions and
preview images. Uploaded apps are then scanned by editors
using a variety of tools, which can warn editors about poten-
tial flaws in the source code. Developers may be asked
(usually through emails) to provide additional information or
improve certain aspects of the apps before they are further
evaluated. The editors’ original review notes may be used in
the evaluation of resubmitted apps that have been previously
rejected. Firefox also updates its own browser platform to
add features and improve quality. These updates may also
require third party apps to be updated for compatibility. For
end users, Firefox can automatically check for available
updates and inform users to manually upgrade their installed
browsers to the latest version. The upgrading, however, may
require time-consuming processes of downloading updates
and rebooting computers.
The platform ecosystem of Firefox is an appropriate setting
for our research for several reasons. First, the free-of-charge
feature of Firefox makes it less likely that our examination of
CNEs would be confounded by any pricing effect. Second,
the editors of the Firefox platform offer weekly app review
data through the app forum. The longitudinal data collected
from the app forum can be used to assess the frequency and
duration of app reviews and to quantify their effects on CNEs.
Third, Firefox’s platform update policies are largely exoge-
nous to CNEs, as Firefox’s update decisions are generally not

128 MIS Quarterly Vol. 42 No. 1/March 2018

 Song et al./Asymmetric Cross-Side Network Effects & Platform Governance
based on interactions between the app-side and the user-side.
This feature should minimize endogeneity concerns about
platform update decisions and enable us to better assess the
impact of platform updates on CNEs.
Data and Measurement
We tested our hypotheses using a longitudinal week-level
dataset about the Firefox ecosystem from July 2009 to
October 2013 (223 weeks). We use data aggregated at the
weekly level mainly because daily data on CNEs and platform
governance does not entail sufficient variations, and monthly
data does not provide the granularity to reveal ongoing
patterns of CNEs and platform governance. Our data also
indicates that the status of third-party apps often changes on
a weekly basis.
We collected information about end users and third-party apps
from the official website for Firefox add-on apps (https://
addons.mozilla.org/en-us/firefox). The site records all third-
party add-on apps that have run on Firefox since it started in
2004. For each app, the site provides detailed information,
including the initial launch time, update history (e.g., when
and how many times), category, and the developer’s name.
As complementary components of the Firefox browser plat-
form, these apps are free to end users. As of October 2013,
there were more than 10,000 apps available on this platform,
spanning 14 categories.
In examining CNEs, we focus on two key aspects of the app-
side: the quantity of apps and the variety of apps. The exam-
ination of app quantity reflects the direct CNEs through which
the increased user base drives the growth of the overall app
base. The examination of app diversity better reflects the
indirect CNEs through which the increased user base drives
the development of complementary apps. To assess app
quantity, we use a measure app number (AN), which repre-
sents the total number of apps available on Firefox on a
weekly basis. We log-transformed AN values to address the
skew in distributions. To assess app variety, we use a mea-
sure app diversity (AD), which is operationalized using a
Herfindahl-like index measure. Specifically, we define

N
HI =
i =1
(xi x )2 , where xi is the weekly number of apps
in the ith category and x is the weekly total number of apps.
Therefore, xi/x is the weekly share of the ith category of apps.
A lower level of HI suggests a higher degree of weekly app
diversity. To make the resulting interpretation more intuitive,
we let AD = 1 – HI (as 1 is the maximum HI value). There-
fore, a higher value of AD means a higher level of weekly app
diversity.
Regarding the user-side, we use a measure user usage (UU)
defined as the weekly total number of times Firefox is used by
all Firefox users. Firefox tracks end users’ usage behavior.
One use is defined as each time a user launches the Firefox
web browser. User usage provides more objective and accu-
rate information about user demand than do other alternative
measures, such as downloads. Because of the skew in
distributions, we use the natural logarithm of UU values.
To examine app review time, we collected data from the
official Firefox app forum (https://forums.mozilla.org/
addons/), where the editors of Firefox post the weekly statuses
of the app review queues. We use this data to calculate the
average waiting time of newly released apps each week, and
construct a measure app review time (AR), which is defined
as the natural logarithm of the average number of days it takes
for a new app to pass review. Finally, to examine platform
updates, we leverage an important policy change event by
Firefox. On March 21, 2011, Firefox changed its update
policy to significantly reduce the intervals between browser
updates. Before this date, Firefox updated its browser plat-
form approximately once a year. After this date, the average
time interval between updates was reduced sharply to about
six weeks. We use a binary indicator as the measure of plat-
form update frequency (PU), with a value of 0 for all weeks
before March 21, 2011 (i.e., the period of less frequent
updates), and a value of 1 for all weeks after March 21, 2011
(i.e., the period of more frequent updates). Our entire sample
covers 99 pre-change weeks and 123 post-change weeks.
We also include some control variables. The first one is entry
of a key competitor (CE), which may affect the platform’s
performance on both sides. It is measured using a dummy
variable that indicates whether or not the key competitor,
Google Chrome, had entered the browser market. The second
factor is platform market share (PMS), which controls for
important marketing condition changes (e.g., entry into a new
foreign market) that may influence CNEs. PMS is measured
using the proportion of all web browser users who use Fire-
fox, and this data was obtained from StatCounter, a leading
web traffic and usage analysis tool. We also control for sea-
sonal effects by using a set of three seasonal dummy variables
SEA = [Sea1, Sea2, Sea3] in the model. Table 2 shows the
definitions and summary statistics of key variables. The
average values for user usage, app number, and app diversity
are 291 million (unlogged value), 5,831 (unlogged value), and
0.846 respectively.
Model Specification and Estimation
In our empirical analysis, we adopt a time-series analysis
technique, vector autoregression with exogenous variable
MIS Quarterly Vol. 42 No. 1/March 2018 129
Song et al./Asymmetric Cross-Side Network Effects & Platform Governance

Table 2. Constructs, Measures, and Descriptive Statistics

Variables Measurement Mean S.D. Min Max
Total quantity of apps on Firefox in each
App Number (AN) 8.577 0.449 7.717 9.229
week (log)
Hirfindahl index of all apps on Firefox in
App Diversity (AD) 0.846 0.002 0.841 0.848
each week
Weekly times of usage by all Firefox users
User Usage (UU) 19.385 0.459 18.697 20.016
(log)
App Review Time The average # of days for reviewing a new
2.047 0.345 1.609 3.023
(AR) app (log)
Platform Update Dummy (1: frequent update period, 0:
0.601 0.491 0.000 1.000
Frequency (PU) infrequent update period)
Dummy (1 – post-entry of Google Chrome;
Competitor entry (CE) 0.729 0.445 0.000 1.000
0 – pre-entry)
Platform market share Market share of Firefox, obtained from
26.691 4.485 17.650 32.401
(PMS) StatCounter

(VARX) (Dekimpe and Hanssens 1995; Luo 2009; Pauwels
2004). This approach has been employed in recent IS and
marketing literature (e.g., Adomavicius et al. 2012; Luo et al.
2013; Trusov et al. 2009) and it allows us to capture dynamic
interactions and feedback effects between the two sides of the
platform. In our research context, VARX has several advan-
tages over alternative modeling techniques. First, it can
assess both the short-term (i.e., immediate) effects and the
long-term (i.e., persistent) effects of endogenous variables on
each other. It therefore allows us to characterize app-to-user
CNEs and user-to-app CNEs based on their temporal features
(Chang and Gurbaxani 2012). Second, VARX models can
simultaneously capture the dynamic and intricate mutual
influences among different variables, helping better illustrate
CNEs between the two sides of the platform.
We adopt a standard VARX procedure as in the literature
(e.g., Dekimpe and Hanssens 1999), which includes the
following steps. First, we determine the appropriateness of
using VARX based on Granger causality tests (Granger
1969). Second, we determine the model specification (VARX
in levels, VARX in differences, or error-correction forms)
based on the unit-root and cointegration test results. Third,
using information criteria, we determine the number of lags in
model specification. Finally, we derive generalized impulse
response functions (GIRFs) to assess both short-term and
long-term effects and test hypotheses.
lags of up to 20 periods (Trusov et al. 2009), we find Granger
causal relationships among endogenous variables (e.g., app
diversity Granger causes user usage (p = .00)). These results
suggest the need to use a full dynamic model, such as a
VARX model.
Model Specification. We use unit root tests to determine
whether endogenous variables (i.e., AN, AD, and UU) are
evolving or stationary. Stationarity of endogenous variables
implies that the fluctuation of these variables caused by any
unexpected changes will eventually dissipate and these vari-
ables will revert back to their deterministic patterns without
a permanent regime shift. The variances of stationary vari-
ables are finite and time invariant (Dekimpe and Hanssens
1999). We use an augmented Dickey-Fuller test (ADF) to
check stationarity. With first-difference values, the ADF tests
of all endogenous variables reject the null hypothesis of a unit
root (see Table 3) and suggest stationarity. We thus estimate
the VARX model using the first-difference values of
endogenous variables. Also, in model specification, we need
to determine the appropriate number of lags used for endoge-
nous variables. The Akaike information criterion (AIC) and
Schwarz-Bayesian information criterion (SBIC) suggest that
the lag of two periods is the optimal choice (AIC = -21.93,
SBIC = -19.33). Our VARX models therefore include lags of
up to two weeks.

In line with prior studies (e.g., Adomavicius et al. 2012;
Trusov et al. 2009), we conduct a series of Granger causality
tests to explore whether explanatory variables explain the
variation of dependent variables, in addition to the lagged
values of dependent variables. Granger causality tests offer
valid arguments on causality in time-series data. By using
We first propose a VARX system to capture dynamic inter-
actions between the user-side and the app-side. We include
a vector of the exogenous variables, an intercept C, and a
deterministic-trend variable T that captures the impact of the
omitted, gradually changing trend of the dependent variables.
The VARX specification is given in Model 1:

130 MIS Quarterly Vol. 42 No. 1/March 2018

 Song et al./Asymmetric Cross-Side Network Effects & Platform Governance

Table 3. Unit Root Test Results after First Differences

Variable Test Statistic p-value
App Number -5.540 <.0001
App Diversity -12.653 <.0001
User Usage -13.868 <.0001
App Review Time -14.513 <.0001

ϕ j ϕ j ϕ j   AN functions (GIRFs) because in this case, the sequences of

 AN t  C AN  δ AN  j  11 12 13   t − j  moves of endogenous variables in the system become less an
 AD  = C  + δ  × T +
 t   AD   AD   ϕ21

j j j 
ϕ22ϕ23  ADt − j  +

j = 1  j j j  UU t − j 
issue of concern (Dekimpe and Hanssens 1999; Luo et al.
UU t  CUU  δUU  ϕ31ϕ32ϕ33 
   2013). Using GIRFs, we simulate the fitted VARX system
(1) using a Monte Carlo approach with 1,000 runs. The resulting
τ j τ j τ j  CE   
j  11 12 13   t − j  ε AN ,t  estimated means and standard errors of endogenous variables
 
j j j 
τ 21τ 22τ 23  PMSt − j  + ε AD,t 

j = 1  j j j   SEA   
are used to test the significance of the changes of these
τ 31τ 32τ 33  t − j  εUU ,t 
variables over time. The short-term effects are assessed by
  examining whether the one-week lagged changes of other
endogenous variables following a shock to a specific
In Model (1), t is the index of week, J is the maximum endogenous variable are significantly different from zero. To
number of lags, and ε is a vector of white-noise disturbances assess the long-term effects, we trace out a time period of 20
with a normal distribution of N(0, Σ). Then, (C, δ, φ, τ) are weeks (Dekimpe and Hanssens 1999), as GIRF values usually
the parameters to be estimated. We also test the existence of stabilize after such a long period. We examine whether the
serial correlation (using a LM test) and heteroskedasticity lagged changes of other endogenous variables after following
(using a White test) (Joshi and Hanssens 2010). The a shock to a specific endogenous variable are still signifi-
nonsignificant LM test (χ2 = 101.14; p = 0.26) indicates that cantly different from zero after 20 weeks.
error terms are not associated with serial correlation. The
significant White test (χ2 = 7374.90; p < 0.01) suggests the To examine the influence of app review time (AR), we
existence of heteroskedasticity. We therefore employ the specify the following model:
weighted least squares (WLS) estimator to address hetero-
skedasticity (Wooldridge 2002). φ j φ j 
 1,1 1,7   AN t − j 
 AN t  C AN  δ AN  φ j φ j   
In the VARX model (1), the user-to-app CNEs are reflected  AD  C  δ   2,1 2,7   ADt − j 
 t   AD   AD  φ j φ j   
in how the app-side variables (ANt, ADt) change over time  ARt  C AR  δ AR  j  3,1 3,7   ARt − j 
     
following the change in the user-side variable UUt. Likewise,  AN t × ARt  = C AN × AR  + δ AN × AR  × T +
 AD × AR  C  δ 
 φ j φ j 
 4,1 4,7 
j =1  j
 AN × AR 
 t − j t − j 
app-to-user CNEs are reflected in how the user-side variable j   ADt − j × ARt − j 
 t   AD × AR   AD × AR  φ5,1φ5,7 
t
 
UU changes over time following changes in the app-side UU t  CUU  δUU   j j  UU t − j 
UU × AR  C  
t   UU × AR  δUU × AR 
 φ6,1φ6,7   
variables (ANt, ADt). The inclusion of endogenous variables’  t  j UU t − j × ARt − j 
j 
φ7,1φ7,7 
own past values (i.e., lag terms) also helps control for intrinsic
τ j τ j 
(2)
change trends in endogenous variables. To test hypotheses on  1,1 1,3 
the short-term and long-term effects of CNEs, we employ the τ j τ j  δ AN 
 2,1 2,3  ε 
impulse response functions (IRFs) derived from the estimated τ j τ j   AD 
j  3 ,1 3 , 3   CE t− j   ε 
parameters and variance-covariance matrices from the full j    AR 
VARX model. The main reason is that the VARX estimated
+   τ j
τ
 4,1 4,3  
j =1  j 
PMS + ε
t − j   AN × AR 
 ε 
j  SEA
coefficients are not usually directly interpretable due to τ 5,1τ 5,3   t − j   AD × AR 
 j j   εUU 
general multicollinearity issues with including lagged terms τ 6,1τ 6,3  ε 
 j  UU × AR 
(Sims 1980). IRFs are used to simulate the over-time impact j 
τ 7,1τ 7,3 
of a change (or shock) to one endogenous variable (over its
baseline) on the dynamics of the full multi-equation system
(Enders 2010). The IRF results depict the net effects of all In model (2), we include in the VARX model both the
the modeled actions and reactions of endogenous variables singular term of AR and the interaction terms between AR
over time. We adopt the generalized impulse response and endogenous variables. Using GIRFs, we can also access

MIS Quarterly Vol. 42 No. 1/March 2018 131

Song et al./Asymmetric Cross-Side Network Effects & Platform Governance

how the interactions between AR and one side may influence Table 4 shows that user usage has significant long-term
the other side in both the short-term and the long-term. effects on both app number and app diversity. The short-term
effects of user usage on the app-side, however, are not signi-
To examine the impact of frequent platform updates, we ficant. These results suggest that when the installed base of
follow the approach in Pauwels and Weiss (2008) to opera- end users expands, the app-side does not respond with quick
tionalize update effects using two variables: a pulse variable growth in either app number or app diversity. Instead, in the
labeled Move and a step variable labeled PU. The pulse long-run, the app-side exhibits slow but persistent growth in
variable Move is equal to 1 at the time of the move from 0 both the number and variety of apps. Such patterns are
(infrequent updates) to 1 (frequent updates), and measures the illustrated in Figure 1, which shows the accumulated impulse
direct impact of changes in platform update policy. The step responses of endogenous variables and also illustrates the
variable PU is equal to 0 for the pre-change period with short-term and long-term effects. In Panel A, the solid curves
infrequent updates, and 1 for the post-change period with represent the over-time mean changes of app number and app
frequent updates. The step variable interacts with both the diversity following a one-unit shock to user usage. The
user-side and the app-side measures, and reveals how CNEs dotted curves represent the change levels that are one-
differ between the pre-change and the post-change periods. standard-deviation above or below the corresponding mean
The model specification is levels. Panel A shows that in the short-term (after 1 week
following the growth shock to user usage), the increases in
φ j  φ j 
 1,1 1,6   AN t − j 
app number and app diversity do not significantly deviate
 AN t  C AN  δ AN 
 AD  C  δ  φ j  φ j    from zero, meaning that the user-side growth does not
 2,1 2,6   ADt − j 
 t   AD   AD  j  j  AN × PU  immediately trigger a significant growth of the app-side,
 AN t × PU t  C AN × PU  δ AN × PU  φ φ j 
  
 3,1 3,6  t − j t − j
 =
  +
   ×T + φ j  φ j   ADt × PU t − j  possibly because only a few new apps can be developed and
 ADt × PU t  C AD × PU  δ AD × PU  j =1  4 ,1 4 ,6   
UU  C  δ   j j  UU t − j  released so quickly. However, the app-side changes increase
 t   UU   UU  φ5,1 φ5,6   
UU t × PU t  CUU × PU  δUU × PU  UU t − j × PU t − j 
over time and in the long-term (e.g., after 20 weeks following
 j j 
φ6,1 φ6,7  the growth shock to user usage), the changes are significantly
(3)
γ j  τ j  τ j  different from zero. This means that new apps keep emerging
 1   Movet − j   1,1 1,3  δ AN 
γ 2j    τ j  τ j  ε 
over time as more developers complete their time-consuming
   Movet − j   2,1 2,3  
j j
γ 3  0   j  j CE   AD
τ 3,1 τ 3j,3   t − j  ε AN × PU 
 app development and eventually the app-side response in
+  
  0
j
j =1 γ 4 
+
  
 j j 
  PMS  + 
j =1 τ 4 ,1 τ 4,3   SEA
t− j
 ε PU


growth manifests in the long-term.
  Move  
 t − j  ε 
γ 5j   t− j   j j   UU

  0  τ5,1 τ5,3   εUU × PU  In contrast, Table 4 shows that both app number and app

γ 6j     j j 
diversity have a significant short-term effect on user usage.
τ 6,1 τ 6,3 
However, their long-term effects on user usage are not
Again, using GIRFs, we can also access how the interactions significant. In Panel B of Figure 1, the solid curves represent
between PU and one side may influence the other side in both the over-time mean changes of user usage following a one-
the short-term and the long-term. unit shock to app number and app diversity. The dotted
curves represent the change levels that are one-standard-
deviation above or below the corresponding mean levels.
Panel B shows that in the short-term (after 1 week following
Results the growth shocks to app number and app diversity), the
increase in user usage significantly deviates from zero,
Main Results meaning that the more available apps can immediately attract
more users to use the platform possibly because users do not
All VARX models have reasonably good model fit and ex- need extra steps to use the platform. However, such increase
planatory power. For example, F-statistics for models (1) – declines over time and in the long term (after 20 week fol-
(3) are 54.03, 29.54, and 30.63, respectively (p < 0.01). The lowing the growth shocks to app number and app diversity),
adjusted R-squares for user usage in models (1) – (3) are it becomes nonsignificant. It means that user momentum
0.37, 0.38, and 0.38, respectively. The adjusted R-squares for gradually disappears and does not sustain in the long term.
app number in models (1) – (3) are 0.57, 0.63, and 0.58, Therefore, H1 is supported.
respectively. The adjusted R-squares for app diversity in
models (1) – (3) are 0.26, 0.25, and 0.26, respectively. With regard to the effect of app review time on CNEs, Table
5 reveals that the interaction between app review time and
Table 4 reports the hypothesis test results on the short- and user usage has a negative long-term effect on both app num-
long-term effects of CNEs. Regarding app-to-user CNEs, ber and app diversity. These results suggest that the longer

132 MIS Quarterly Vol. 42 No. 1/March 2018

 Song et al./Asymmetric Cross-Side Network Effects & Platform Governance

Table 4. User-to-App CNEs and App-to-User CNEs

Panel A: User-to-App Effects
Paths Short-Term Effect Long-Term Effect
User Usage  App Number 0.02 (0.03, p = 0.26) 0.18 (0.06, p = 0.01)**
User Usage  App Diversity 0.01 (0.01, p = 0.18) 0.07 (0.03, p = 0.04)*

Panel B: App-to-User Effects

Paths Short-Term Effect Long-Term Effect
App Number  User Usage 0.65 (0.31, p = 0.04)* 0.35 (0.89, p = 0.35)
App Diversity  User Usage 0.48 (0.22, p = 0.03)* 0.16 (0.66, p = 0.42)

Notes: *p < 0.05, **p < 0.01. Estimated coefficients are in percentage values. In parentheses are standard errors and corresponding p-values.

Panel A: User-to-App Effects (with One-Unit Shock on User Usage)

Panel B: App-to-User Effects (with One-Unit Shock on App Number (Left) and App Diversity (Right))

Figure 1. Accumulated Impulse Response Functions of Cross-Side Network Effects

the average time of app review, the weaker the growth in both
number and diversity of apps as a response to the user-side in
the long-term (i.e., less sustainable). Given that the user-to-
app effects are primarily long-term effects, H2a is supported.
In addition, Table 5 also shows that longer app review time
even weakens the short-term effects of user usage on app
number, as indicated by the negative short-term effect of the
interaction between app review time and user usage. This
suggests that longer app review time makes the response of
app number (to the user-side) even less immediate in the
short-run. However, app review time does not influence the
effect of user usage on app diversity in the short-term.
Table 5 also shows that the interactions between app review
time and both app number and app diversity do not signifi-
cantly influence user usage in either the short-term or long-
term. Therefore, neither H2b nor H2c is supported. These
results suggest that the overall influence of app review time
on app-to-user CNEs is neither positive nor negative. A
potential explanation is that the two countervailing effects
hypothesized in H2b (i.e., app-obsoleteness) and H2c (i.e.,
quality-improvement) cancel each other. We conduct an
additional analysis (as explained below) to further explore this
issue.

MIS Quarterly Vol. 42 No. 1/March 2018 133

Song et al./Asymmetric Cross-Side Network Effects & Platform Governance
Table 5. Influences of App Review Time on Cross-Side Network Effects
Panel A: User-to-App Effects
Paths
User Usage × App Review Time  App Number
User Usage × App Review Time  App Diversity
Panel B: App-to-User Effects
Paths
App Number × App Review Time  User Usage
App Diversity × App Review Time  User Usage
Notes: *p < 0.05, **p < 0.01. Estimated coefficients are in percentage values. In parentheses are standard errors and corresponding p-values.
Table 6 shows how frequent platform updates influence user-
to-app CNEs and app-to-user CNEs. Specifically, the inter-
action between platform update frequency and user usage has
a negative long-term effect on both app number and app
diversity. The implication is that when the platform updates
itself more frequently, the growth of both app number and app
diversity as a response to the user-side becomes weaker in the
long-term (i.e., less persistent and sustainable). In other
words, frequent platform updates prevent the app-side from
committing to persistent growth in both the number and
diversity of apps. In contrast, frequent platform updates have
no significant influence on the short-term effects of user-to-
app CNEs. Given that the user-to-app CNEs are primarily
long-term effects, H3a is not supported, and H3b is supported.
To help understand the interaction effect, we use Figure 2 to
visually illustrate how app number and app diversity change
over time after a one-unit shock on user usage, in both the
case of infrequent platform updates and the case of frequent
platform updates. The left portion (right portion) of Figure 2
shows that as a response to the one-unit shock on user usage,
app number (app diversity) increases more significantly in the
case of infrequent platform updates than in the case of
frequent platform updates. These results illustrate that
frequent platform updates significantly reduce the impact of
user usage on the app-side (i.e., the user-to-app CNEs).
Table 6 also shows that the interaction between platform
update frequency and app number has a negative effect on
user usage in the short-term. In other words, when the plat-
form updates itself more frequently, the short-term response
of the user-side to the increased number of apps becomes
weaker. However, frequent platform updates do not signifi-
cantly weaken the short-term response of user usage to app
diversity. Therefore, H3c is partially supported. A potential
explanation is that the growth in app diversity provides more
app complementarity that generates additional value to end
users. Such complementarity-based value creation may also
counteract the hassles of platform updates to end users and
help attract users to respond to the app-side in the short-term.
134 MIS Quarterly Vol. 42 No. 1/March 2018
Short-Term Effect Long-Term Effect
-0.02 (0.01, p = 0.04)* -0.02 (0.00, p = 0.00)**
-0.01 (0.02, p = 0.32) -0.06 (0.03, p = 0.04)*
Short-Term Effect Long-Term Effect
-0.16 (0.56, p = 0.39) -0.18 (0.71, p = 0.41)
-0.19 (0.57, p = 0.35) -0.17 (0.41, p = 0.34)
To provide additional insights into the moderating effect of
frequent platform updates, we split the sample and estimate
the model for both the time period before the change in
platform update policy and the time period after the change in
policy. We examine the difference between these two periods
in terms of the short-term and long-term effects of both user-
to-app CNEs and app-to-user CNEs. The results, summarized
in Appendix A, are consistent with those of the main analysis.
In the pre-change period (with infrequent platform updates),
user usage has a significant long-term effect on both app
number and app diversity (see Table A1 of Appendix A). In
the post-change period (with frequent platform updates),
neither the short-term effect nor the long-term effect of user
usage is significant (see Table A2 of Appendix A). Similarly,
in the pre-change period, number of apps has a significant
short-term effect on user usage (see Table A1 of Appendix
A). In the post-change period, this effect is nonsignificant
(see Table A2 of Appendix A). App diversity, however, has
a short-term effect on user usage in both the pre-change
period and the post-change period. In this regard, frequent
platform updates weaken user-to-app CNEs and partially
weaken app-to-user CNEs.
Robustness Checks
We conduct several additional analyses to verify the robust-
ness of our results. First, following Pauwels and Weiss
(2008), we evaluate the parameter stability of VARX models
using a bootstrapping approach. From the original dataset, we
draw a sample with replacement 100 times and rerun the
VARX models on each sample. The comparison of the
estimates in the main analysis with their corresponding distri-
butions of bootstrap estimates suggests that all significant
results hold. Thus, there is no concern regarding the stability
of the parameters of interest.
 Song et al./Asymmetric Cross-Side Network Effects & Platform Governance

Table 6. Influences of Frequent Platform Updates on Cross-Side Network Effects

Panel A: User-to-App Effects
Paths Short-Term Effect Long-Term Effect
User Usage × Platform Update Frequency  App Number -0.02 (0.03, p = 0.27) -0.06 (0.03, p = 0.04)**
User Usage × Platform Update Frequency  App Diversity -0.01 (0.02, p = 0.31) -0.07 (0.03, p = 0.02)*
Panel B: App-to-User Effects
Paths Short-Term Effect Long-Term Effect
App Number × Platform Update Frequency  User Usage -1.12 (0.45, p = 0.02)** -1.26 (1.27, p = 0.17)
App Diversity × Platform Update Frequency  User Usage -1.13 (1.15, p = 0.18) -1.16 (2.77, p = 0.31)

Notes: *p < 0.05, **p < 0.01. Estimated coefficients are in percentage values. In parentheses are standard errors and corresponding p-values.

Note: Solid lines: with infrequent platform updates (PU = 0); Dotted lines: with frequent platform updates (PU = 1)

Figure 2. User-to-App CNEs Before and After Platform Update Policy Change (with One-Unit Shock on
Use Usage)

Second, we also collect data on user download (UD) for the
platform and use it as an alternative measure to characterize
the user-side. We consider the total number of times the
platform is downloaded and log-transform this variable to
address the distribution skew. We rerun the VARX models
with the user usage measure replaced with UD. The results
based on UD, as summarized in Appendix B, are qualitatively
consistent with the main results. Specifically, the increase in
UD leads to long-term growth of both number and diversity
of apps (in Table B1), and this long-term effect is weakened
both by long app review times (in Table B2) and by frequent
platform updates (in Table B3). The increase in number of
apps leads to the short-term growth of UD (in Table B1), and
this short-term effect is weakened both by long app review
times (in Table B2) and frequent platform updates (in Table
B3). However, app diversity has no significant effect on UD.
A potential explanation is that new users who have relatively
less knowledge about different types of apps are more likely
to download than are experienced users. Therefore, the
aggregate information such as total number of apps is cogni-
tively more appealing to new users than detailed information
such as the diversity of apps.
calculate the measures of developer number (DN) and devel-
oper diversity (DD) using the same approaches as for app
number and app diversity, respectively.4 We rerun the VARX
models with the measures app number and app diversity
replaced by DN and DD, respectively. The results, as
summarized in Appendix C, are qualitatively consistent with
the main results. The growth of user usage leads to long-term
growth in both DN and DD, and the growth of DN and DD
lead to a short-term growth of user usage. The long-term
effects of user usage on DN and DD are weakened by both
app review time and frequent platform updates. The short-
term effects of DN and DD on user usage are weakened by
frequent platform updates, but not by app review time.
Fourth, in the main analysis, we use a binary variable to
distinguish between the period with frequent platform updates
and the period with infrequent updates. To verify the robust-
ness of our results, we also try a continuous variable. Speci-
fically, we derive the time intervals between the releases of
different Firefox versions. We reverse-code platform update
time intervals by subtracting the interval values from the
maximum interval value, so that a high measure value means

Third, we try alternative measures that characterize the app-

side. We collect additional data on app developers and 4
We thank an anonymous referee for this suggestion.

MIS Quarterly Vol. 42 No. 1/March 2018 135

Song et al./Asymmetric Cross-Side Network Effects & Platform Governance
frequent platform updates. Using this measure, we rerun the
VARX analysis. As shown in Table D1 of Appendix D, the
results are qualitatively consistent with the main results (i.e.,
frequent platform updates weaken both user-to app CNEs and
app-to-user CNEs).
Fifth, we verify our findings on frequent platform updates to
the software platform by taking into consideration the fre-
quency of updates by competitors to their software platform.
We collect additional data on the historical updates of Fire-
fox’s key competitor, Google Chrome. We derive the time
intervals between the releases of different Chrome versions,
and reverse-code the time intervals by subtracting the interval
values from the maximum interval value. Therefore, a high
(low) value of reverse-coded interval means Chrome engages
in frequent (infrequent) releases of its updates. We then
include the measure of reverse-coded intervals as an addi-
tional control variable in the model of Appendix D and rerun
the VARX analysis. Compared to that in Appendix D, this
analysis further takes into account the competitive environ-
ment and controls for the update frequency of the key
competing platform. The results, as summarized in Appendix
E, show that our findings remain robust—that is, frequent
updates to a software platform weaken both user-to app CNEs
and app-to-user CNEs for it.
Additional Insights
We conduct additional analyses to develop insights that
complement our main findings. First, our main findings
reveal that long app review times neither weaken nor
strengthen app-to-user CNEs. A potential explanation is that
although long app review times delay app releases, more
thorough app reviews may improve the quality of apps and
counteract the negative impact of delayed app releases. To
further investigate the impact of app review time on app
quality, we collect additional weekly data on the quality of
new apps. On the official website for Firefox apps, end users
can rate apps after they use them. Ratings are based on a five-
star scale, with one star indicating extremely low quality and
five stars indicating extremely high quality. For each new
app, we use the average user rating as a measure of quality.
We then create a separate VARX model including app review
time and app quality as endogenous variables, and other
exogenous variables. The results, as summarized in Table F1
of Appendix F, show that long app review times significantly
increase the quality of new apps in both the short-term and the
long-term. This finding provides a key insight about why app
review times may not necessarily weaken app-to-user CNEs.
That is, the app quality improvement associated with long app
review processes is likely to offset the negative impact of
delayed app releases. It is also worth noting, however, that
136 MIS Quarterly Vol. 42 No. 1/March 2018
such app quality improvements are still not strong enough to
dominate the wear-out effects and lead to significant long-
term increases in user usage.
Second, to gain more insights on the impact of app quality, we
conduct an additional analysis by dividing the sample into
two subsamples: a high-rating sample with app ratings above
the mean and a low-rating sample with app ratings below the
mean. Then, we examine the app-to-user effect separately for
the high-rating sample and the low-rating sample. Table F2
in Appendix F summarizes the results. The results show that
app-to-user CNEs are relatively stronger in the high-rating
subsample than in the low-rating subsample. These findings
provide additional support for the explanation that improved
app quality (as a result of long app review times) can make
apps and the platform more attractive to users and counteract
the negative effect of app review time on app-to-user CNEs.
Third, as mentioned, a potential reason for the negative
moderating effect of frequent platform updates on user-to-app
CNEs is that app developers may be forced to devote more
resources to updating their existing applications, making it
more challenging for them to respond to user-side growth
with new apps. To further verify this insight, we specify a
new VARX model that incorporates frequent platform up-
dates, the number of updated apps, and the number of new
apps as endogenous variables, along with other exogenous
variables. The results, as summarized in Table G1 of Appen-
dix G, show that frequent platform updates significantly
increase the number of updated apps in both the short-term
and the long-term. The number of updated apps significantly
decreases the number of new apps in both the short and long
terms. Therefore, these findings support the explanation that
frequent platform updates weaken user-to-app CNEs by
constraining developers’ efforts in new app development.
We also rerun the full VARX model (3) with aggregate mea-
sures for the number and diversity of apps replaced with
separate measures for the number and diversity of new apps
and updated apps. The results in Table G2 of Appendix G
show that the interaction between frequent platform updates
and user usage has a negative effect on both the number and
diversity of new apps in the long-term. This suggests that
frequent platform updates weaken the long-term effect of
user-to-app CNEs for new apps. In contrast, the results also
indicate that the interaction between frequent platform up-
dates and user usage has a positive effect on the number of
updated apps in the long-term. The effect of this interaction
on the diversity of updated apps, however, is not significant.
These findings suggest that frequent platform updates essen-
tially drive developers to commit more resources to update
their existing apps, and such resource commitments are
widely spread across diverse categories of apps. As a result,
 Song et al./Asymmetric Cross-Side Network Effects & Platform Governance
developers have limited efforts to develop more new apps as
a response to the user-side.
Discussion
Theoretical Implications
Our work builds on the platform literature that has shown the
existence of CNEs and contributes to it by surfacing the
nature of the asymmetry among different types of CNEs. Our
theorization of asymmetric CNEs based on a value creation/
capture perspective and the supporting empirical findings
generate two important theoretical implications for charac-
terizing the two-sided platform. First, our findings shed light
on the differential temporal effects that the app-side and the
user-side of a software platform have on each other due to
their underlying different value creation/capture processes.
The pursuit of exchange value requires the app-side to engage
in resource commitment processes, which make the app-side
less likely to grow immediately as a response to the user-side
expansion. However, the commitment of resources is likely
to drive developers to realize their development over time,
making the CNEs from the user-side to the app-side even-
tually manifest in the long-term. In contrast, the pursuit of
direct use value makes the resource commitment processes of
the user-side relatively simple. The user-side is therefore
more likely to grow immediately as a response to the app-side
expansion. But over time, the user-side momentum is less
likely to be persistent in the long-term due to the lack of com-
mitment. In this regard, the differential value pursuit and
resource commitment drive the distinct immediacy and dura-
tion of CNEs on the two sides. Our reliance on a value
perspective to understand network effects essentially echoes
the call in recent network studies to go beyond network size
explanations of network effects (Afuah 2013).
Second, our revelation of the differential temporal effects
provides insights on how feedback, specifically delays in
response to feedback and instability in the generated feedback
(Tiwana 2014), works differently on the two sides of a plat-
form. Our findings suggest that delays are more of an issue
for the supply-side (the app-side in our case) responding to
demand-side (the user-side in our case) feedback, possibly
due to the relatively more complex resource commitment
processes on the supply-side. However, instability in supply-
side feedback to the demand-side is less of an issue as supply-
side resource commitments foster a persistent supply-side
feedback to the demand-side. In contrast, instability in
demand-side feedback to the supply-side can be an issue due
to volatility in platform users’ taste for apps.
The evidence on the influence of platform governance
policies on CNEs extends the understanding of platform
models, as the extant research has largely taken CNEs as
exogenously given (e.g. Eisenmann et al. 2006; Parker and
Van Alstyne 2008; Seamans and Zhu 2014). Our findings
suggest that both user-to-app CNEs and app-to-user CNEs can
be weakened by frequent updates of a software platform, and
that user-to-app CNEs can also be weakened by a long app
review process. An important implication is that there are
potential trade-offs in platform governance policies that may
not be revealed by adopting a narrow frame of reference, say
the platform to the exclusion of CNEs, in the evaluation. For
instance, quality control and improvement through platform
updates may be viewed, from the platform owner’s local
perspective, as benefits. However, these policies may also
have nuanced implications on the interdependencies with and
among platform participants, such as app developers and
users. Indeed, governance policies that are considered to be
advantageous for the platform owner may impose mutual
adaptation challenges for other ecosystem participants. For
example, despite the benefits of platform updates, the process
of releasing updates too frequently may inevitably disrupt the
smooth interactions across sides. Therefore, the impacts that
these governance policies can have on the evolution of the
platform itself require careful consideration alongside the
impacts that these governance policies can have on cross-side
evolution and interdependencies in the platform ecosystem.
The insights from our study also contribute to the theoretical
understanding of software development strategies. First, past
literature has suggested the sequential provision of multiple
versions of software as an optimal way to signal quality and
gradually build installed bases in the consumer market with
network externality (e.g., Gallaugher and Wang 2002; Pad-
manabhan et al. 1997). Our study suggests that the new
development environment (e.g., the hyperspeed of the web/
mobile world), in which platforms are vital in bridging apps
and users, may challenge traditional sequential strategies to
develop versions of software. The CNEs in the platform eco-
system intensify app competition as users face a variety of
choices (Boudreau 2012). The short-term effect of app-to-
user CNEs also suggests the difficulty in drawing long-term
user interest. In this case, the quality of introductory versions
becomes more important. App developers may have to recon-
sider how they balance resource commitments between
seeding initial market share with high-quality introductory
versions of an app and refining quality of the app in subse-
quent sequential versions to maintain and grow the installed
bases.
Second, while past research has largely assumed that sequen-
tial development is under the control of developers (e.g.,
MIS Quarterly Vol. 42 No. 1/March 2018 137
Song et al./Asymmetric Cross-Side Network Effects & Platform Governance
Padmanabhan et al. 1997), we delineate that platform gover-
nance policies can potentially interfere with developers’
sequential development and app releases. Long app review
times may impede the new app releases that developers use to
seed initial installed bases. Platform updates may force app
updates and even make app updates a burden for developers.
In this regard, developers should take into account the
potential influences from platform governance actions, and
plan their sequential development strategies accordingly to
make their app upgrades and releases more adaptive to the
context of the platform ecosystem.
The consideration of platform governance policies also
extends the lens of IS governance to an ecosystem context
that is characterized by complex adaptiveness. The extant IS
governance perspectives have been primarily concerned with
internal organizational arrangements (e.g., Sambamurthy and
Zmud 1999) or with interorganizational arrangements of
dyadic relationships (Klein and Rai 2009; Rai and Tang
2010). In contrast, the platform governance perspective is
concerned with the platform ecosystem and diverse stake-
holders (not only the platform owner), and needs to uncover
characteristics of complex adaptive systems (Tanriverdi et al.
2010), such as interdependencies and coevolution in the
ecosystem. Our findings reveal that the trajectory and direc-
tion of coevolution of the app and user-sides of the platform
are affected by the governance actions of the platform owner.
Our study’s revelation of the temporal features of CNEs
suggests that platform owners may be able to dynamically
adapt platform governance to accelerate (or decelerate) CNEs
and realize temporary advantages over time—an idea that is
especially important in complex adaptive systems where con-
catenating temporary advantages is more likely to yield
competitive advantages than simply seeking a sustainable
source of competitive advantage (e.g., Tanriverdi et al. 2010).
Practical Implications
The characterization of asymmetric CNEs has implications for
platform managers who often face a coordination challenge
when CNEs enable participants on each side to form expecta-
tions about the other side (Spulber 2010; Tucker and Zhang
2010). To address this issue, platforms usually need to decide
which side to prioritize in resource commitment (Hagiu and
Spulber 2013). Our findings suggest that given the long-term
effects of user-to-app CNEs, platform managers need to be
patient and persistent in supporting the app-side growth.
They may also subsidize the app-side in a way that expedites
the app-side’s response to user-side growth. Given the short-
term effect of app-to-user CNEs, platform managers may
devote resources to sustain the user-side response to the app-
side and mitigate feedback decays on the user-side. This
138 MIS Quarterly Vol. 42 No. 1/March 2018
implication can also be generalized to non-software platform
contexts. For example, in platforms such as Airbnb, hosts
(suppliers) often engage in more complex processes of
resource commitment (e.g., property maintenance, safety
assurance, picture preparation, etc.) than renters. Therefore,
akin to our findings in the software platform context, the
responses of the host-side and the renter-side to each other
may differ in immediacy and duration. Airbnb can provide
services that assist hosts to prepare their properties and more
swiftly capture the fast-growing demand from renters. On the
renter side, Airbnb can place more weight on approaches
(e.g., community building, feedback management) to improve
renter engagement in the long term. Providing greater trans-
parency about future host-side development to the renter-side
(e.g., more development in amenities, house rules, and profes-
sional displays) may maintain or increase momentum of
renters and sustain long-term renter-side growth.
Another key practical implication for platform managers is to
build governance capabilities in order to reduce the conflict
between governance actions and the growth of the platform
ecosystem through network effects. Platforms should take
action to improve the efficiency of their governance pro-
cesses, such as diligently reducing app review times. For
example, recently Apple announced its promise to signifi-
cantly reduce the average time for app review and approval to
less than two days (Webb 2016). The intention was to bring
Apple more in line with its key competitor, Alphabet Inc.’s
Google. In contrast to Apple, which checks apps before
including them in the App Store, Google Play often reviews
apps after making them available to users for download so as
to avoid delays caused by app review. Platforms may also use
more transparent information on review status to better
manage the expectations of app developers and users and
alleviate their frustration. Differentiated review processes for
heterogeneous apps can be employed to expedite reviews of
apps that are heavily requested by users.
Platform managers should also build their capability in
managing frequent platform updates as it is critical for plat-
form growth and competition. When comparing Firefox and
Chrome in terms of update management, Jono DiCarlo, a
former Firefox developer, concluded that “doing rapid
releases poorly just made Firefox look like an inferior version
of Chrome” (Kingsley-Hughes 2012). Therefore, platforms
should make serious efforts to enhance their software archi-
tectures (e.g., compatibility, standardization of APIs, and
modulation), expand their support to both app developers and
users, and even rely on assistance from certified support
professionals and open communities to streamline their plat-
form update processes. Doing so can help minimize the nega-
tive impacts of frequent platform updates on both developers
and users, and sustain the growth of the platform ecosystem.
 Song et al./Asymmetric Cross-Side Network Effects & Platform Governance
Regarding practical implications to app developers, our study
suggests that app developers need to consider strategies and
techniques to overcome the potential adverse effects of
platform governance. For example, the dynamic software
update technique (e.g., Hayden et al. 2012; Seifzadeh et al.
2013) can facilitate and simplify app update processes.
Developers may consider using more of these techniques in
adapting to frequent platform updates. Second, developers
may also consider more flexible strategies in the development
stage to be better prepared for subsequent platform gover-
nance actions. For example, emerging hybrid development
approaches (Beckman 2012) allow developers to use more
generic tools (e.g., Javacript) to fuse their apps with the native
features of platforms. Adopting these app development
approaches in the first place provides developers the flexi-
bility and convenience in subsequent revisions and improve-
ments driven by platform governance.
Limitations and Future Research
This study has some limitations and creates opportunities for
future research. First, our field data is about a specific soft-
ware platform. Future research may consider generalizing the
insights on asymmetric CNEs by examining other software
platforms (e.g., operating systems and mobile platforms) and
non-software platforms (e.g., B2C and C2C transactional plat-
forms). The CNEs in emerging platform business models
(e.g., sharing economy) and the influences of many techno-
logical and non-technological platform governance factors are
also worth exploring.
Second, this study focuses on a free software platform and
therefore the role of pricing is not relevant in our context.
However, pricing can be an important governance mechanism
of a platform (Parker and Van Alstyne 2008), and is critical
to other types of software platforms (e.g., enterprise soft-
ware). The examination of pricing issues may help shed light
on how CNEs influence the competition between a platform
and third party developers for profit. Therefore, future
research may focus on for-profit platforms and specifically
consider the pricing effect.
Third, our study focuses primarily on the interactions between
a platform and its own two-sided markets. The data avail-
ability keeps us from fully endogenizing the inter-platform
competition in our research model. Future research may
explore how the evolving competition among platforms is
influenced by asymmetric CNEs and the interactions among
CNEs and other platform strategies. Future research can
evaluate how platform strategies may strengthen, rather than
weaken, CNEs and enable platforms to accrue competitive
advantages over others. Situations in which platforms may
intentionally weaken CNEs so as to gain more control over
the two sides also deserve attention.
Finally, future research may adopt different theoretical per-
spectives, such as ecology theory, to examine the evolution of
the ecosystems of software and non-software platforms.
These perspectives may help generate additional insights on
the ecology processes on the supply-side and the demand-
side, and the temporal dynamics of cross-side network effects.
The roles of many other sides (in addition to the supply-side
and the demand-side) that generate valuable products/services
and influence cross-side network effects can be taken into
consideration.
Conculsion
Our study surfaces the asymmetric features of CNEs in the
software platform ecosystem and the impacts of platform
governance mechanisms on CNEs. Our findings reveal that
user-to-app CNEs primarily take effect in the long-term, while
app-to-user CNEs primarily take effect in the short-term. We
find that long app review times weaken user-to-app CNEs, but
not app-to-user CNEs. Our results also suggest that long app
review times result in enhanced app quality that may counter-
act the negative impact of delayed app releases (caused by
long app review times). In addition, we find that frequent
platform updates weaken user-to-app CNEs, in part because
frequent platform updates force developers to commit more
effort to updating existing apps rather than developing new
ones. Frequent platform updates partially weaken app-to-user
CNEs by mitigating the attractiveness of increased app
numbers, but not that of increased app diversity, to end users.
Overall, our study elaborates our understanding about the
dynamics of network effects in two-sided platforms and how
governance practices of platforms can be leveraged to estab-
lish a virtuous cycle of CNEs for the vibrant growth of the
platform ecosystem.
Acknowledgments
The authors gratefully acknowledge the financial support by the
National Natural Science Foundation of China under Grant
#71472083, Grant #71372112, Grant #71490721, and Grant
#71232008.
References
Adomavicius, G., Bockstedt, J., and Gupta, A. 2012. “Modeling
Supply-Side Dynamics of IT Components, Products, and Infra-
structures: An Empirical Analysis Using Vector Autoregres-
sion,” Information Systems Research (23:2), pp. 397-417.
MIS Quarterly Vol. 42 No. 1/March 2018 139
Song et al./Asymmetric Cross-Side Network Effects & Platform Governance
Afuah, A. 2013. “Are Network Effects Really All about Size? The
Role of Structure and Conduct,” Strategic Management Journal
(34:3), pp. 257-273.
Anderson, E. G., Parker, G. G., and Tan, B. 2014. “Platform
Performance Investment in the Presence of Network Exter-
nalities,” Information Systems Research (25:1), pp. 152-172.
Arce, N. 2015. “Google Launches App Approval Process: What
Android Developers Must Know,” Tech Times, March 18
(http://www.techtimes.com/articles/40556/20150318/google-
launches-app-approval-process-what-android-developers-must-
know.htm).
Arora, A., Caulkins, J. P., and Telang, R. 2006. “Sell First, Fix
Later: Impact of Patching on Software Quality,” Management
Science (52:3), pp. 465-471.
Baldwin, C., and Woodard, J. 2009. “The Architecture of Plat-
forms: A Unified View,” in Platforms, Markets and Innovation,
A. Gawer (ed.), Cheltenham, UK: Edward Elgar, pp. 19-44.
Bass, F. M., Bruce, N., Majumdar, S., and Murthi, B. P. S. 2007.
“Wearout Effects of Different Advertising Themes: A Dynamic
Bayesian Model of the Advertising-Sales Relationship,”
Marketing Science (26:2), pp. 179-195.
Bavota, G., Linares-Vasquez, M., Bernal-Cardenas, C. E., Di
Penta, M., Oliveto, R., Poshyvanyk, D. 2015. “The Impact of
API Change- and Fault-Proneness on the User Ratings of
Android Apps,” IEEE Transactions on Software Engineering
(41:4), pp. 384-407.
Beckman, M. 2012. “Native, Web, or Hybrid: How to Choose
Your Mobile Development Path,” InfoWorld, October 8
(http://www.infoworld.com/article/2615122/).
Boudreau, K. J. 2010. “Open Platform Strategies and Innovation:
Granting Access vs. Devolving Control,” Management Science
(56:10), pp. 1849-1872.
Boudreau, K. J. 2012. “Let a Thousand Flowers Bloom? An Early
Look at Large Numbers of Software App Developers and
Patterns of Innovation,” Organization Science (23:5), pp.
1409-1427.
Bowman, C., and Ambrosini, V. 2000. “Value Creation Versus
Value Capture: Towards a Coherent Definition of Value in
Strategy,” British Journal of Management (11:1), pp. 1-15.
Calantone, R., and Sawyer, A. G. 1978. “The Stability of Benefit
Segments,” Journal of Marketing Research (15:3), pp. 395-404.
Ceccagnoli, M., Forman, C., Huang, P., and Wu, D. J. 2012.
“Cocreation of Value in a Platform Ecosystem: The Case of
Enterprise Software,” MIS Quarterly (36:1), pp. 263-290.
Chang, Y. B. and Gurbaxani, V. 2012. “The Impact of IT-Related
Spillerovers on Long-Run Productivity: An Empirical Analysis,”
Information Systems Research (23:3), pp. 868-886.
Chu, J., and Manchanda, P. 2016. “Quantifying Cross-Network
Effects in Online C2C Platforms,” Marketing Science (35:6), pp.
870-893.
Clements, and Ohashi, H. 2005. “Indirect Network Effects And
The Product Cycle: Video Games in the U.S., 1994–2002,”
Journal of Industrial Economics (53:4), pp. 515-542.
Cusumano, M., and Gawer, A. 2002. “The Elements of Platform
Leadership,” Sloan Management Review (43:3), pp. 51-58.
Dave, K. 2015. “Apple’s App Store Review Process Is Hurting
Users, but We’re Not Allowed to Talk about It,” Business
Insider, April 12 (http://www.businessinsider.com/apples-app-
140 MIS Quarterly Vol. 42 No. 1/March 2018
store-review-process-is-hurting-users-but-were-not-allowed-to-
talk-about-it-2015-4).
Dekimpe, M. G., and Hanssens, D. M. 1995. “Empirical Generali-
zations about Marketing Evolution and Stationarity,” Marketing
Science (14:3), pp. 109-121.
Dekimpe, M. G. and Hanssens, D. M. 1999. “Sustained Spending
and Persistent Response: A New Look at Long-Term Marketing
Profitability,” Journal of Marketing Research (36:4), pp.
397-412.
Dubé, J.-P. H., Hitsch, G. J., and Chintagunta, P. K. 2010.
“Tipping and Concentration in Markets with Indirect Network
Effects,” Marketing Science (29:2), pp. 216-249.
Eisenmann, T. R., Parker, G., and Van Alstyne, M. 2006.
“Strategies for Two Sided Markets,” Harvard Business Review
(84:10), pp. 92-101.
Eisenmann, T. R., Parker, G., and Van Alstyne, M. W. 2011.
“Platform Envelopment,” Strategic Management Journal (32:12),
pp. 1270-1285.
Enders, W. 2010. Applied Econometric Time Series(3rd ed.), New
York: Wiley.
Evans, D. S. 2011. Platform Economics: Essays on Multi-Sided
Businesses, Competition Policy International.
Evans, D. S., Hagiu, A., and Schmalensee, R. 2006. Invisible
Engines: How Software Platforms Drive Innovation and Trans-
form Industries, Cambridge, MA: MIT Press.
Fisher, W. W., and Oberholzer-Gee, F. 2013. “Strategic Manage-
ment of Intellectual Property: An Integrated Approach,” Califor-
nia Management Review (55:4), pp. 157-183.
Fleischmann, M., Amirpur, M., Grupp, T., and Benlian, Hess, T.
“The Role of Software Updates in Information Systems
Continuance—An Experimental Study from a User Perspective,”
Decision Support Systems (83), pp. 83-96.
Gallaugher, J. M., and Wang, Y. M. 2002. “Understanding Net-
work Effects in Software Markets: Evidence from Web Server
Pricing,” MIS Quarterly (26:4), pp. 303-327.
Ghazawneh, A., and Henfridsson, O. 2013. “Balancing Platform
Control and External Contribution in Third-Party Development:
The Boundary Resource Model,” Information Systems Journal
(23:2), pp. 173-192.
Ghemawat, P. 1991. Commitment, New York: The Free Press.
Graham S. 2016. “Developers: Apple’s App Review Needs Big
Improvements,” MacStories, March 1 (https://www.macstories.
net/stories/developers-apples-app-review-needs-big-
improvements/).
Granger, C. W. J. 1969. “Investing Causal Relations by Econo-
metric Models and Cross-Spectral Methods,” Econometrica
(37:3), pp. 424-438.
Grime, I., Diamantopoulos, A., and Smith, G. 2002. “Consumer
Evaluations of Extensions and their Effects on the Core Brand:
Key Issues and Research Propositions,” European Journal of
Marketing (36:11), pp. 1415-1438.
Hagiu, A. 2006. “Pricing and Commitment by Two-Sided Plat-
forms,” RAND Journal of Economics (37:3), pp.720-737.
Hagiu, A., and Spulber, D. 2013. “First-Party Content and Coor-
dination in Two-Sided Markets,” Management Science (59:4),
pp. 933-949.
Harrigan, K. R., and Porter, M. E. 1983. “End-Game Strategies for
Declining Industries,” Harvard Business Review (61), pp.
111-120.
 Song et al./Asymmetric Cross-Side Network Effects & Platform Governance
Harter, D. E., and Slaughter, S. A. 2003. “Quality Improvement
and Infrastructure Activity Costs in Software Development: A
Longitudinal Analysis,” Management Science (49:6), pp.
784-800.
Hayden, C. M., Smith, E. K., Hardisty, E. A., Hicks, M. and
Foster, J. S., 2012. “Evaluating Dynamic Software Update
Safety Using Systematic Testing,” IEEE Transactions on Soft-
ware Engineering (38:6), pp. 1340-1354.
Hilkert, D., Benlian, A., and Hess, T. 2010. “Motivational Drivers
to Develop for Social Software-Platforms: The Example of
Facebook,” in Proceedings of the 16th Americas Conference on
Information Systems, Lima, Peru, Paper 86.
Hodgetts, H. M., and Jones, D. M. 2007. “Reminders, Alerts and
Pop-Ups: The Cost of Computer-Initiated Interruptions,” in
Human–Computer Interaction, J. Jacko (ed.), Berlin: Springer-
Verlag, pp. 818-826
Hyrynsalmi, S., Suominen, A., and Mäntymäki, M. 2016. “The
Influence of Developer Multi-Homing on Competition between
Software Ecosystems,” Journal of Systems and Software (111),
pp. 119-127.
Joshi, A., and Hanssens, D. M. (2010). “The Direct and Indirect
Effects of Advertising Spending on Firm Value,” Journal of
Marketing (74:1), pp. 20-33.
Kaiser, U., and Wright, J. 2006. “Price Structure in Two-Sided
Markets: Evidence from the Magazine Industry,” International
Journal of Industrial Organization (24:1), pp. 1-28.
Kapoor, R., and Agarwal, S. 2016. “Sustaining Superior Perfor-
mance in Business Ecosystems: Evidence from Application
Software Developers in the iOS and Android Smartphone
Ecosystems,” Working Paper, The Wharton School, University
of Pennsylvania.
Kingsley-Hughes, A. 2012. “Firefox Developer: Everybody Hates
Firefox Updates,” Forbes, July 10 (http://www.forbes.com/sites/
adriankingsleyhughes/2012/07/10/firefox-developer-everybody-
hates-firefox-updates/).
Klein, R., and Rai, A. 2009. “Interfirm Strategic Information Flows
in Logistics Supply Chain Relationships,” MIS Quarterly (33:4)
pp. 735-762
Klotzbach, C. 2016. “Enter the Matrix: App Retention and
Engagement,” Flurry Analytics Blog, May 12 (http://
flurrymobile.tumblr.com/post/144245637325/appmatrix).
Leone, R. P. 1995. “Generalizing What Is Known about Temporal
Aggregation and Advertising Carryover,” Marketing Science
(14:3), pp. 141-150.
Lepak, D. P., Smith, K. G., and Taylor, M. S. 2007. “Value Crea-
tion and Value Capture: A Multilevel Perspective,” Academy of
Management Review (32:1), pp. 180-194.
Liu, H., 2010. “Dynamics of Pricing in the Video Game Console
Market: Skimming or Penetration?,” Journal of Marketing
Research (47:2), pp. 428-443.
Luo, X. 2009. “Quantifying the Long-Term Impact of Negative
Word of Mouth on Cash Flows and Stock Prices,” Marketing
Science (28:1), pp. 148-165.
Luo, X., Zhang, J., and Duan, W. 2013. “Social Media and Firm
Equity Value,” Information Systems Research (24:1), pp.
146-163.
Mandel, L. 2012. “Improving the Firefox Update Experience,”
Mozilla Blog (https://blog.mozilla.org/futurereleases/ 2012/02/
03/improving-the-firefox-update-experience/).
Maurer, C., and Tiwana, A. 2012. “Control in App Platforms: The
Integration-Differentiation Paradox,” in Proceedings of the 33rd
International Conference on Information Systems, Orlando, FL.
Mullins, J. W., and Sutherland, D. J. 1998. “New Product Devel-
opment in Rapidly Changing Markets: An Exploratory Study,”
Journal of Product Innovation Management (15:3), pp. 224-236.
Naik, P. A., Mantrala, M. K., and Sawyer, A. G. 1998. “Planning
Media Schedules in the Presence of Dynamic Advertising
Quality,” Marketing Science (17:3), pp. 214-235.
Ortiz de Guinea, A., and Webster, J. 2013. “An Investigation of
Information Systems Use Patterns: Technological Events as
Triggers, the Effect of Time, and Consequences for Perfor-
mance,” MIS Quarterly (37:4), pp. 1165-1188.
Padmanabhan, V., Rajiv, S., and Srinivasan, K. 1997. “New
Products, Upgrades, and New Releases: A Rationale for
Sequential Product Introduction,” Journal of Marketing Research
(34:4), pp. 456-472.
Parker, G., and Van Alstyne, M. W. 2005. “Two-Sided Network
Effects: A Theory of Information Product Design,” Management
Science (51:10), pp. 1494-1504.
Parker, G., and Van Alstyne, M. W. 2008. “Managing Platform
Ecosystems,” in Proceedings of the 29th International Conference
on Information Systems, Paris.
Pauwels, K. 2004. “How Dynamic Consumer Response, Competi-
tor Response, Company Support, and Company Inertia Shape
Long-Term Marketing Effectiveness,” Marketing Science (23:4),
pp. 598-610.
Pauwels, K., and Weiss, A. 2008. “Moving from Free to Fee: How
Online Firms Market to Change their Business Model Suc-
cessfully,” Journal of Marketing (72:3), pp. 14-31.
Rai, A., and Tang, X. 2010. “Leveraging IT Capabilities and
Competitive Process Capabilities for the Management of
Interorganizational Relationship Portfolios,” Information Systems
Research (21:3), pp. 516-542.
Ramasubbu, N., and Kemerer, C. F. 2015. “Technical Debt and the
Reliability of Enterprise Software Systems: A Competing Risks
Analysis,” Management Science (62:5), pp. 1487-1510.
Rochet, J. C., and Tirole, J. 2006. “Two-Sided Markets: A Pro-
gress Report,” RAND Journal of Economics (37:3), pp. 645-667.
Rysman, M. 2009. “The Economics of Two-Sided Markets,”
Journal of Economic Perspectives (23:3), pp. 125-143.
Sambamurthy, V., and Zmud, W. 1999. “Arrangements for Infor-
mation Technology Governance: A Theory of Multiple Contin-
gencies,” MIS Quarterly (23:2), pp. 261-290.
Seamans, R., and Zhu, F. 2014. “Responses to Entry in Multi-
Sided Markets: The Impact of Craigslist on Local Newspapers,”
Management Science (60:2), pp. 476-493.
Seifzadeh, H., Abolhassani, H. and Moshkenani, M. S. 2013. “A
Survey of Dynamic Software Updating,” Journal of Software
Evolution and Process (25), pp. 535-568.
Sims, C. A. 1980. “Macroeconomics and Reality,” Econometrica
(48:1), pp. 1-48.
Spulber, D. F. 2010. “Solving the Circular Conundrum: Communi-
cation and Coordination in Internet Markets,” Northwestern
University Law Review (104:2), pp. 537-591.
Sriram, S., Manchanda, P., Bravo, M. E., Chu, J., Ma, L., Song, M.,
Shriver, S., and Subramanian, U. 2015. “Platforms: A Multi-
plicity of Research Opportunities,” Marketing Letter (26:2), pp.
141-152.
MIS Quarterly Vol. 42 No. 1/March 2018 141
Song et al./Asymmetric Cross-Side Network Effects & Platform Governance
Starr, M. 2014. “Apple Lists the Top 10 Reasons Apps Get
Rejected,” CNET, September 2 (http://www.cnet.com/news/
apple-lists-the-top-10-reasons-apps-get-rejected/).
Sun, J. 2012. “Why Different People Prefer Different Systems for
Different Tasks: An Activity Perspective in Technology
Adoption in a Dynamic User Environment,” Journal of the
American Society for Information Science and Technology (63:1),
pp. 48-63.
Tanriverdi, H., Rai, A., and Venkatraman, N. 2010. “Reframing the
Dominant Quests of Information Systems Strategy Research for
Complex Adaptive Business Systems,” Information Systems
Research (21:4), pp. 822-834.
Tiwana, A. 2014. Platform Ecosystems: Aligning Architecture,
Governance, and Strategy, New York: Elsevier, Inc.
Tiwana, A., Konsynski, B., and Bush, A. A. 2010. “Platform
Evolution: Coevolution of Platform Architecture, Governance,
and Environmental Dynamics,” Information Systems Research
(21:4), pp. 675-687.
Tucker, C., and Zhang, J. 2010. “Growing Two-Sided Networks by
Advertising the User Base: A Field Experiment.” Marketing
Science (29:5), pp. 805-814.
Trusov, M., Bucklin, R. E., and Pauwels, K. 2009. “Effects of
Word-of-Mouth Versus Traditional Marketing: Findings from an
Internet Social Networking Site,” Journal of Marketing (73:5),
pp. 90-102.
Venkatesh, V., Brown, S. A., Maruping, L. M., and Bala, H. 2008.
“Predicting Different Conceptualizations of System Use: The
Competing Roles of Behavioral Intention, Facilitating Condi-
tions, and Behavioral Expectation,” MIS Quarterly (32:3), pp.
483-502.
Webb, A. 2016. “Apple Shortens App Review Times in Push to
Boost Service Sales,” Bloomberg Technology, May 12
(http://www.bloomberg.com/news/articles/2016-05-12/apple-
shortens-app-review-times-in-push-to-boost-service-sales).
Weyl, E. G. 2010. “A Price Theory of Multi-Sided Platforms,”
American Economic Review (100:4), pp. 1642-1672.
Wilbur, K. C. 2008. “A Two-Sided, Empirical Model of Television
Advertising and Viewing Markets,” Marketing Science (27:3),
pp. 356-378.
Wooldridge, J. M. 2002. Econometric Analysis of Cross Section
and Panel Data, Cambridge, MA: MIT Press.
Zhang, K., Evgeniou, T., Padmanabhan, V. and Richard, E. 2012.
“Content Contributor Management and Network Effects in a
UGC Environment,” Marketing Science (31:3), pp. 433-447.
Zhu, F., and Iansiti, M. 2012. “Entry into Platform-Based Mar-
kets,” Strategic Management Journal (33:1), pp. 88-106.
About the Authors
Peijian Song is an associate professor at the School of Business,
Nanjing University. His research interests include platform strategy,
electronic commerce, and IT innovation. His work has been
published by journals such as Decision Support Systems, IEEE
Transactions on Engineering Management, and Journal of Inter-
142 MIS Quarterly Vol. 42 No. 1/March 2018
tional Marketing, and conferences such as the International
Conference on Information Systems and Academy of Management.
He serves on the editorial review board of Journal of Global
Information Management and Electronic Commerce Research and
Applications.
Ling Xue is an assistant professor at the Robinson College of
Business, Georgia State University. He received his Ph.D. in
information systems from the McCombs School of Business at the
University of Texas at Austin. His research interests include IT-
related corporate governance issues, strategic business value of IT,
interorganizational systems, and the digital platform business
models. His papers have been published in Information Systems
Research, MIS Quarterly, Academy of Management Journal,
Journal of Operations Management, Production and Operations
Management, Journal of Management Information Systems, Journal
of Information Systems, Decision Support Systems, ACM Trans-
action on MIS, Information & Management, International Journal
of Electronic Commerce, and Journal of Global Information
Management, and proceedings of the International Conference on
Information Systems. Ling served as the corresponding author for
this paper.
Arun Rai is Regents’ Professor of the University System of Georgia
at the Robinson College of Business, Georgia State University, and
holds the Robinson and Harkins Chairs. His research has focused on
how firms can leverage information technologies in their strategies,
relationships, and processes; how complex systems can be
effectively developed and deployed; and how digital innovations can
create business value and address societal problems such as digital
inequality, poverty, health disparities, and infant mortality. Several
of his research projects have been in collaboration with major
corporations in automotives, hi-tech, healthcare, financial services,
and logistics and supply chains, and with not-for-profit organi-
zations. He is a Fellow of the Association for Information Systems
and a Distinguished Fellow of the INFORMS Information Systems
Society.
Cheng Zhang is a professor in the Department of Information
Management & Information Systems, Fudan University, China. His
current research interests include information strategies, electronic
markets, and IT diffusion. His works have been published in
journals including Computer and Human Behaviors, Decision
Support Systems, Electronic Markets, IEEE Transactions on Engi-
neering Management, IEEE Transactions on Professional Com-
munication, INFORMS Journal on Computing, Information &
Management, International Journal of Electronic Commerce, Inter-
national Journal of Production Economics, International Journal of
Production Research, Journal of Management Information Systems,
Marketing Science, Journal of Global Information Management, and
Simulation Modeling Practice and Theory. He serves on the
editorial review boards of Journal of Global Information
Management and IEEE Transactions on Professional Commu-
nication.
 RESEARCH ARTICLE

THE ECOSYSTEM OF SOFTWARE PLATFORM: A STUDY

OF ASYMMETRIC CROSS-SIDE NETWORK EFFECTS
AND PLATFORM GOVERNANCE
Peijian Song
School of Business, Nanjing University, Nanjing CHINA {songpeijian@nju.edu.cn}

Ling Xue
J. Mack Robinson College of Business, Georgia State University, Atlanta, GA 30303 U.S.A. {lxue5@gsu.edu}

Arun Rai
J. Mack Robinson College of Business, Georgia State University, Atlanta, GA 30303 U.S.A. {arun.rai@eci.gsu.edu}

Cheng Zhang
School of Management, Fudan University, Shanghai CHINA {zhangche@fudan.edu.cn}

Appendix A
Cross-Side Network Effects Before and After Frequent Platform Updates

Table A1. Cross-Side Network Effects Before Frequent Platform Updates

Panel A: User-to-App Effects
Paths Short-Term Effect Long-Term Effect
User Usage  App Number 0.04 (0.22, p = 0.43) 0.31 (0.15, p = 0.03)*
User Usage  App Diversity 0.01 (0.02, p = 0.31) 0.09 (0.02, p = 0.00)*
Panel B: App-to-User Effects
Paths Short-Term Effect Long-Term Effect
App Number  User Usage 0.94 (0.49, p = 0.04)* 0.20 (1.14, p = 0.43)
App Diversity  User Usage 0.64 (0.29, p = 0.03)* 0.19 (0.78, p = 0.41)

Notes: *p < 0.05, **p < 0.01. Estimated coefficients are in percentage values. In parentheses are standard errors and corresponding p-values.

MIS Quarterly Vol. 42 No. 1—Appendices/March 2018 A1

Song et al./Asymmetric Cross-Side Network Effects & Platform Governance

Table A2. Cross-Side Network Effects After Frequent Platform Updates

Panel A: User-to-App Effects
Paths Short-Term Effect Long-Term Effect
User Usage  App Number 0.06 (0.22, p = 0.39) 0.03 (0.04, p = 0.24)
User Usage  App Diversity 0.01 (0.01, p = 0.17) 0.01 (0.03, p = 0.37)
Panel B: App-to-User Effects
Paths Short-Term Effect Long-Term Effect
App Number  User Usage 0.37 (0.74, p = 0.30) 0.40 (1.08, p = 0.36)
App Diversity  User Usage 0.49 (0.24, p = 0.04)* 0.19 (0.78, p = 0.41)

Notes: *p < 0.05, **p < 0.01. Estimated coefficients are in percentage values. In parentheses are standard errors and corresponding p-values.

Appendix B
Download as An Alternative Measure of User-Side

Table B1. Cross-Side Network Effects

Panel A: User-to-App Effects
Paths Short-Term Effect Long-Term Effect
User Download  App Number 0.12 (0.16, p = 0.24) 0.45 (0.18, p = 0.02)**
User Download  App Diversity 0.00 (0.01, p = 0.83) 0.03 (0.01, p = 0.01)*
Panel B: App-to-User Effects
Paths Short-Term Effect Long-Term Effect
App Number  User Download 2.05 (0.92, p = 0.03)* 0.43 (2.31, p = 0.43)
App Diversity  User Download 0.04 (0.91, p = 0.48) 0.48 (1.62, p = 0.39)

Notes: *p < 0.05, **p < 0.01. Estimated coefficients are in percentage values. In parentheses are standard errors and corresponding p-values.

Table B2. Influences of App Review Time on Cross-Side Network Effects

Panel A: User-to-App Effect
Paths Short-Term Effect Long-Term Effect
User Download × App Review Time  App Number -0.02 (0.01, p = 0.04)* -0.03 (0.01, p = 0.01)**
User Download × App Review Time  App Diversity -0.03 (0.04, p = 0.24) -0.12 (0.05, p = 0.02)*
Panel B: App-to-User Effect
Paths Short-Term Effect Long-Term Effect
App Number × App Review Time  User Download -1.85 (2.79, p = 0.26) -3.62 (4.92, p = 0.24)
App Diversity × App Review Time  User Download -2.49 (2.23, p = 0.15) -3.75 (3.92, p = 0.18)

Notes: *p < 0.05, **p < 0.01. Estimated coefficients are in percentage values. In parentheses are standard errors and corresponding p-values.

A2 MIS Quarterly Vol. 42 No. 1—Appendices/March 2018

 Song et al./Asymmetric Cross-Side Network Effects & Platform Governance

Table B3. Influences of Frequent Platform Updates on Cross-Side Network Effects

Panel A: User-to-App Effect
Paths Short-Term Effect Long-Term Effect
User Download × Platform Update Frequency  App Number 0.00 (0.01, p = 0.71) -0.13 (0.03, p = 0.00)**
User Download × Platform Update Frequency  App Diversity 0.00 (0.02, p = 0.79) -0.10 (0.04, p = 0.02)*
Panel B: App-to-User Effect
Paths Short-Term Effect Long-Term Effect
App Number × Platform Update Frequency  User Download -1.61 (0.40, p = 0.00)** -0.19 (0.81, p = 0.41)
App Diversity × Platform Update Frequency  User Download -0.07 (0.46, p = 0.44) -0.90 (1.12, p = 0.22)

Notes: *p < 0.05, **p < 0.01. Estimated coefficients are in percentage values. In parentheses are standard errors and corresponding p-values.

Appendix C
Number and Diversity of Developers as Alternative Measures of App-Side

Table C1. Cross-Side Network Effects

Panel A: User-to-App Effects
Paths Short-Term Effect Long-Term Effect
User Usage  Developer Number 0.01 (0.02, p = 0.31) 0.07 (0.02, p = 0.00)**
User Usage  Developer Diversity 0.01 (0.01, p = 0.17) 0.03 (0.01, p = 0.01)**
Panel B: App-to-User Effects
Paths Short-Term Effect Long-Term Effect
Developer Number  User Usage 0.39 (0.14, p = 0.01)** 0.28 (0.94, p = 0.39)
Developer Diversity  User Usage 0.61 (0.19, p = 0.01)** 0.68 (0.63, p = 0.15)

Notes: *p < 0.05, **p < 0.01. Estimated coefficients are in percentage values. In parentheses are standard errors and corresponding p-values.

Table C2. Influences of App Review Time on Cross-Side Network Effects

Panel A: User-to-App Effect
Paths Short-Term Effect Long-Term Effect
User Usage × App Review Time  Developer Number -0.02 (0.01, p = 0.04)* -0.04 (0.02, p = 0.04)*
User Usage × App Review Time  Developer Diversity 0.01 (0.01, p = 0.17) -0.03 (0.01, p = 0.01)**
Panel B: App-to-User Effect
Paths Short-Term Effect Long-Term Effect
Developer Number × App Review Time  User Usage 0.31 (0.50, p = 0.28) 0.47 (1.05, p = 0.33)
Developer Diversity × App Review Time  User Usage 0.68 (0.61, p = 0.15) 0.62 (1.02, p = 0.28)

Notes: *p < 0.05, **p < 0.01. Estimated coefficients are in percentage values. In parentheses are standard errors and corresponding p-values.

MIS Quarterly Vol. 42 No. 1—Appendices/March 2018 A3

Song et al./Asymmetric Cross-Side Network Effects & Platform Governance

Table C3. Influences of Frequent Platform Updates on Cross-Side Network Effects

Panel A: User-to-App Effect
Paths Short-Term Effect Long-Term Effect
User Usage × Platform Update Frequency  Developer Number -0.02 (0.02, p = 0.17) -0.09 (0.02, p = 0.00)**
User Usage × Platform Update Frequency  Developer Diversity 0.00 (0.01, p = 0.50) -0.08 (0.01, p = 0.00)**
Panel B: App-to-User Effect
Paths Short-Term Effect Long-Term Effect
Developer Number × Platform Update Frequency  User Usage -1.03 (0.05, p = 0.00)** -0.79 (1.01, p = 0.23)
Developer Diversity × Platform Update Frequency  User Usage -1.14 (0.04, p = 0.00)** -0.61 (0.92, p = 0.26)

Notes: *p < 0.05, **p < 0.01. Estimated coefficients are in percentage values. In parentheses are standard errors and corresponding p-values.

Appendix D
Platform Update Interval as An Alternative Continuous Measure
of Frequent Platform Updates

Table D1. Influences of Frequent Platform Updates on Cross-Side Network Effects

Panel A: User-to-App Effect
Paths Short-Term Effect Long-Term Effect
User Usage × Platform Update Frequency  App Number -0.01 (0.01, p = 0.17) -0.17 (0.07, p = 0.02)*
User Usage × Platform Update Frequency  App Diversity 0.00 (0.01, p = 0.75) -0.02 (0.01, p = 0.04)*
Panel B: App-to-User Effect
Paths Short-Term Effect Long-Term Effect
App Number × Platform Update Frequency  User Usage -0.93 (0.47, p = 0.04)* -1.10 (1.51, p = 0.24)
App Diversity × Platform Update Frequency  User Usage -0.91 (0.43, p = 0.03)* -1.01 (1.58, p = 0.27)

Notes: *p < 0.05, **p < 0.01. Estimated coefficients are in percentage values. In parentheses are standard errors and corresponding p-values.

A4 MIS Quarterly Vol. 42 No. 1—Appendices/March 2018

 Song et al./Asymmetric Cross-Side Network Effects & Platform Governance

Appendix E
With the Update Intervals of Competing Platform as An Additional Control

Table E1. Cross-Side Network Effects

Panel A: User-to-App Effects
Paths Short-Term Effect Long-Term Effect
User Usage  App Number 0.01 (0.02, p = 0.28) 0.14 (0.07, p = 0.04)*
User Usage  App Diversity 0.01 (0.01, p = 0.14) 0.07 (0.03, p = 0.02)*
Panel B: App-to-User Effects
Paths Short-Term Effect Long-Term Effect
App Diversity  User Usage 0.65 (0.15, p = 0.01)** 0.08 (0.90, p = 0.47)
App Diversity  User Usage 0.49 (0.19, p = 0.01)** 0.31 (0.64, p = 0.32)

Notes: *p < 0.05, **p < 0.01. Estimated coefficients are in percentage values. In parentheses are standard errors and corresponding p-values.

Table E2. Influences of Frequent Platform Updates on Cross-Side Network Effects

Panel A: User-to-App Effects
Paths Short-Term Effect Long-Term Effect
User Usage × Platform Update Frequency  App Number -0.03 (0.03, p = 0.17) -0.12 (0.04, p = 0.01)**
User Usage × Platform Update Frequency  App Diversity -0.01 (0.01, p = 0.15) -0.10 (0.03, p = 0.01)**
Panel B: App-to-User Effects
Paths Short-Term Effect Long-Term Effect
App Number × Platform Update Frequency  User Usage -1.14 (0.45, p = 0.02)* -1.26 (1.30, p = 0.18)
App Diversity × Platform Update Frequency  User Usage -0.91 (0.41, p = 0.03)* -1.94 (2.42, p = 0.22)

Notes: *p < 0.05, **p < 0.01. Estimated coefficients are in percentage values. In parentheses are standard errors and corresponding p-values.

Appendix F
Additional Analyses on App Quality

Table F1. Additional Analysis on App Quality

Paths Short-Term Effect Long-Term Effect
App Review Time  Quality of New Apps 15.13 (2.72, p = 0.00)** 11.34 (3.51, p = 0.00)**

Notes: *p < 0.05, **p < 0.01. The coefficients are percentage values. In parentheses are standard errors and corresponding p-values.

MIS Quarterly Vol. 42 No. 1—Appendices/March 2018 A5

Song et al./Asymmetric Cross-Side Network Effects & Platform Governance

Table F2. App-to-User CNEs for High-Rating and Low-Rating Apps

Panel A: App-to-User Effects for High-Rating Apps
Paths Short-Term Effect Long-Term Effect
App Number  User Usage 2.41 (0.42, p = 0.00)** 0.60 (0.56, p = 0.16)
App Diversity  User Usage 0.75 (0.33, p = 0.02)* 0.35 (0.59, p = 0.28)
Panel B: App-to-User Effects for Low-Rating Apps
Paths Short-Term Effect Long-Term Effect
App Number  User Usage 0.41 (0.15, p = 0.01)* 0.15 (0.43, p = 0.37)
App Diversity  User Usage 0.78 (0.74, p = 0.16) 0.35 (0.41, p = 0.21)

Notes: *p < 0.05, **p < 0.01. Estimated coefficients are in percentage values. In parentheses are standard errors and corresponding p-values.

Appendix G
Additional Analysis on Platform Updates

Table G1. Impact of Frequent Platform Updates on the App-Side

Paths Short-Term Effect Long-Term Effect
Platform Update Frequency  Number of Updated Apps 7.36 (2.26, p = 0.00)** 8.94 (2.71, p = 0.00)**
Number of Updated Apps  Number of New Apps -11.36 (2.61, p = 0.00)** -11.33 (2.73, p = 0.00)**
Notes: *p < 0.05, **p < 0.01. Estimated coefficients are in percentage values. In parentheses are standard errors and corresponding p-values.

Table G2. Influences of Frequent Platform Updates on User-to-App Effect (for New and Updated Apps)
Panel A: User-to-App Effect for New Apps
Paths Short-Term Effect Long-Term Effect
User Usage × Platform Update Frequency  Number of New
-2.23 (2.89, p = 0.24) -4.60 (1.07, p = 0.00)**
Apps
User Usage × Platform Update Frequency  Diversity of New
-0.10 (0.15, p = 0.26) -0.25 (0.12, p = 0.03)*
Apps
Panel B: User-to-App Effect for Updated Apps
Paths Short-Term Effect Long-Term Effect
User Usage × Platform Update Frequency  Number of
2.27 (3.13, p = 0.24) 5.62 (1.58, p = 0.00)**
Updated Apps
User Usage × Platform Update Frequency  Diversity of
0.11 (0.67, p = 0.45) 0.08 (0.68, p = 0.45)
Updated Apps
Notes: *p < 0.05, **p < 0.01. Estimated coefficients are in percentage values. In parentheses are standard errors and corresponding p-values.

A6 MIS Quarterly Vol. 42 No. 1—Appendices/March 2018