Economic Modelling 28 (2011) 439–450

Contents lists available at ScienceDirect

Economic Modelling
j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / e c m o d

The choice of foreign market entry mode: An analysis of the dynamic probit model
Ming-Yuan Chen a,⁎, Jing-Yun Chang b
a
Graduate Institute of Human Resource Management, National Central University, No. 300, Jhongda Rd., Jhongli City, Taoyuan County 32001, Taiwan
b
Department of Industrial Economics, Tamkang University, No.151, Ying-Chuan Road, Tamshui, Taipei County 25137, Taiwan

a r t i c l e i n f o a b s t r a c t

Article history: Based on the official survey data of Taiwanese FDI in China, this study investigates the transaction cost
Accepted 11 August 2010 determinants of foreign market entry mode choice. In contrast to the existing literature, this paper constructs
a panel data set and makes a dynamic probit analysis on the mode choice between wholly-owned
JEL classification:
subsidiaries and joint ventures. Empirical results indicate that state dependence between current and past
L23
modes plays an important role in determining entry mode choice. However, the impact of state dependence
M21
will be reduced significantly if the mode adopted in the previous period is associated with poor performance
Keywords: of the parent firm. There is strong linkage between mode performance and future mode choice. Evidence also
Entry mode choice shows that the hypothesized influences of transaction cost factors on entry mode choice gain support in the
Dynamic probit model dynamic choice process.
Transaction cost theory © 2010 Elsevier B.V. All rights reserved.

1. Introduction Other models are also developed to explain the phenomena

One of the most critical decisions faced by multinational enter-
prises (MNEs) intending to undertake foreign direct investment (FDI)
is the choice of foreign market entry mode. A large number of studies
analyze the determinants of entry mode choice within the transaction
cost framework proposed by Anderson and Gatignon (1986). Based on
the transaction cost economics initiated by Williamson (1975, 1985),
the underlying theory claims that the organizational structure and
design are determined by the trade-off between market transaction
costs and hierarchical control costs. Depending on the factors
pertaining to transaction specific assets, external and internal
uncertainty, and the potential for opportunistic behavior, MNEs
choose a specific mode of entry in order to minimize transaction
costs. Transaction cost analysis has been extended by many research-
ers by integrating several factors into the given framework, such as
environmental and strategic factors, network relationship, and
institutional and cultural factors (i.e., Brouthers, 2002; Coviello and
Munro, 1997; Hennart, 1991; Hill, et al., 1990; Lu, 2002; Yiu and
Makino, 2002). In a recent study, Brouthers et al. (2008a) combine
transaction cost economics with insights from the real option theory
and suggest that this model effectively improves decision-making
outcomes. The firms choose the entry mode predicted by this
combined model have significantly satisfactory performance.
⁎ Corresponding author. Tel.: + 886 3 422 7151x66153; fax: + 886 3 422 7492.
E-mail address: mychen@mgt.ncu.edu.tw (M.-Y. Chen).
associated with entry mode choice. The eclectic framework intro-
duced and developed by Dunning (1980, 1988, 1998, 2000) is a model
commonly applied. This approach states that the entry mode
decisions are determined by three sets of advantages as perceived
by enterprises, namely, ownership advantages, location advantages,
and internalization advantages. The more advantages a firm has the
greater the tendency of adopting the mode with a high level of
control. The resource-based view is also among the most frequently
applied perspectives. This view, including knowledge-based and
organizational capabilities theories, regards a firm as a bundle of
capabilities and knowledge, where individual skill, organization and
technology are linked together (Aulakh and Kotabe, 1997; Madhok,
1998; Nelson and Winter, 1982). Entry mode decisions are made
under the consideration related to the development and acquisition of
firm-specific resources and capabilities. Brouthers et al. (2008b)
further argue that resource-based advantages have different levels of
applicability in different national institutional environment. The entry
mode predicted by a model that incorporates both resource-based
advantages and external institutional factors yields better subsidiary
performance. A survey article by Brouthers and Hennart (2007)
provides a review of issues and various approaches for the interna-
tional entry mode research.
This paper examines the transaction cost determinants of entry
mode choice between wholly-owned subsidiaries (WOSs) and joint
ventures (JVs) for Taiwanese FDI in China. Taiwan has been one of the
main contributors of FDI flows in the East Asian area since 1990,
particularly for the investment into the Chinese market. People in
Taiwan share common language, ethnic links, and cultural traits with

0264-9993/$ – see front matter © 2010 Elsevier B.V. All rights reserved.
doi:10.1016/j.econmod.2010.08.004
440 M.-Y. Chen, J.-Y. Chang / Economic Modelling 28 (2011) 439–450
Chinese. A lot of Taiwanese have strong affection for their original
home in China and have strong commitment to their family. Shared
culture and family relations produce more mutual trust and less
uncertainty between Taiwanese investors and Chinese local firms
(Wei et al., 2005). The influence of cultural distance between home
and host countries on entry mode choice is widely recognized in the
literature. Large cultural distance, implying high transaction costs for
MNEs investing overseas, may lead them to adopting JVs as the entry
mode (Hennart and Larimo, 1998). Foreign investors can share risk
with local partners and avoid costly mistakes in an environment with
unfamiliar culture. Cultural advantages for Taiwanese firms' entry into
China suggest that they inherently prefer WOSs to JVs as they are
more acquainted with local conditions and hence are less reliant on
local firms. Strategically, Taiwanese firms tend to enter this culturally
similar market with wholly-owned modes in order to obtain a high
return (Brouthers, 2002; Kim and Hwang, 1992). It is thus interesting
to know whether the transaction cost arguments gain support in
explaining the mode choice when WOSs seem to be a MNE's natural
choice.
The main feature of this paper is that we construct a panel data set
to make a dynamic binary choice analysis on mode choice. Previous
studies generally regard entry mode choice as a one-stage static
decision-making problem. However, the choice may be a multiple-
stage dynamic process in the sense that firms having started to enter
into a foreign market may change their original strategy due to
learning effects or unanticipated developments and results. There are
two merits in the panel data dynamic choice analysis. First, although
cross-sectional studies can test the relation between observed firm
characteristics and entry mode choice, they overlook that the choice
might be affected by unobserved heterogeneity across firms. The
panel data model has greater flexibility in modeling behavior
differences among individuals. Second, a dynamic specification,
where the effect of the lagged dependent variable is considered in
the binary choice model, is necessary since state dependence (or
habit-persistence) is important for a firm's entry mode decision. Some
firms may remain more likely to choose JVs than others in every
period due to firm heterogeneity. However, having JVs in one period
may also lower the probability of choosing WOSs in the next period
because of large adjustment costs in changing the organizational form
in relatively short periods. This reflects the effect of state dependence.
Heckman (1981a) has shown that state dependence is particularly
important in a panel that has only a few observations for each
individual, which, we believe, is a usual case for most of entry mode
data. By modeling the dynamic effect in the binary choice model, the
transaction cost influences (i.e., the impact of firm heterogeneity) on
entry mode choice can be tested in a more precise way that controls
for state dependence of mode choice.
Furthermore, as indicated by Brouthers and Hennart (2007),
although entry mode studies have made experience a standard
variable in mode choice models to consider the effect of learning from
past choices, they fail to distinguish between favorable and unfavor-
able experience. A firm that has tried JVs and has not performed well
will consider whether continue using JVs or try another mode type. As
performance reflects the success and failure of the selected mode, the
dynamic setting is able to identify the link between previous mode
performance and future mode choices. It allows us to examine how
mode performance affects state dependence of mode choice. In sum, a
dynamic binary choice analysis, with its capability of dealing with the
persistence effect of mode use and the impact of past performance on
mode choice dynamics, might help us obtain additional insights not
available from the cross-sectional techniques.
The dichotomous specification of WOSs and JVs used in this paper
deserves more notice. In the entry mode literature, three commonly
explored modes are wholly-owned mode, joint venture mode, and
non-equity contractual mode such as licensing, franchising, and
exporting. In the Chinese market, few Taiwanese firms use licensing
agreements or export to China prior to establishing plants there. This
feature makes it less necessary to consider contracts as an alternative
mode of servicing the Chinese market. Besides, according to the
perspective articulated by Hennart (1988, 1989, 2000), the remuner-
ation to input suppliers in the equity modes, either WOSs or JVs, is
paid ex post from the profit of the venture, which is different from
contracts with payments being specified ex ante. The variables that
determine the choice between contracts and equity modes are thus
not necessarily the same as those affecting the choice between WOSs
and JVs (Brouthers and Hennart, 2007). Gatignon and Anderson
(1988) and Brouthers et al. (2003) also emphasize that the role of
transaction cost variables in predicting entry mode choice when a
contractual partnership has been chosen is not as clear as for the
decision as to whether to set up a WOS or JV. Therefore, while
classifying entry modes in a binary way may reduce the variability of
mode choice, it allows for a more explicit and precise test of
transaction cost arguments. Examining the binary choice between
WOSs and JVs has been the methodology adopted by many entry
mode studies, e.g., Brouthers (2002), Chen and Hennart (2002),
Cleeve (1997), Hennart (1991), Hennart and Larimo (1998), Makino
and Neupert (2000), and Padmanabhan and Cho (1996).
This paper also distinguishes the entry mode, i.e., WOSs vs. JVs,
from the establishment mode, i.e., greenfields vs. acquisitions, and
focuses on the former. In the literature, some researchers have mixed
these two dimensions by considering the choice among, say, wholly-
owned greenfields, wholly-owned acquisitions, and JV greenfields.
However, in the survey article of international entry mode research,
Brouthers and Hennart (2007) do not include the studies as to the
choice between greenfield and acquisition entry. They argue that in
view of the theoretical reasons and the empirical evidence, the
decision of entry mode and the decision of establishment mode are
separate and independently made. This distinction is also recognized
by several studies, e.g., Dikova and van Witteloostuijn (2007),
Hennart (2000), Hennart and Park (1993), and Padmanabhan and
Cho (1996).
Based on the official survey data conducted by the Department of
Statistics of Taiwan, this paper shows that state dependence, which
suggests the persistence between current and previous entry modes,
is important in explaining an MNE's entry mode choice. It is also
evidenced that the transaction cost arguments of entry mode choice
gain support when state dependence has been controlled in a
dynamic probit model, provided that this model is appropriately
estimated by assuming the endogeneity of initial conditions. Trans-
action cost economics predict the choice of entry modes satisfactorily
even if the MNEs inherently have greater propensity of using WOSs
because of common cultural traits, at least for Taiwanese FDI in China.
Furthermore, there is evidence supporting an intriguing dynamic
phenomenon that poor mode performance lowers the degree of state
dependence. The possibility of continuing adoption of WOSs in the
current period due to mode dependence will be reduced significantly
if WOSs used in the previous period produces poor performance in the
MNE's parent firm.
The next section presents the empirical models, including static
and dynamic binary choice models. Section 3 reviews the literature
and discusses the transaction cost determinants of entry mode
choice. Section 4 describes data and variable measurement. The
empirical results are presented in Section 5, and finally Section 6
concludes.
2. Static and dynamic models of entry mode choice
Previous studies generally use cross-sectional data for probit or
logit models to analyze the choice of foreign entry mode. In this study,
since longitudinal data from official surveys are available to track a
firm's mode choices, the binary choice models and their estimation
are extended to the panel data format.
 M.-Y. Chen, J.-Y. Chang / Economic Modelling 28 (2011) 439–450
The structure of the static random effects probit model for entry
mode choice is specified as


Prob ½ MODEit = 1j xit ; ui  = Φ α + β ′ xit + ηt + σu ui ðfor t = 1; …; Ti Þ
where the subscript i indexes individual firms and t indexes time
period (year). The panel is unbalanced in this study. MODEit is a
binary variable for entry mode choice. MODEit = 1 indicating WOSs
and MODEit = 0 for JVs. The xit represents explanatory variables that
are hypothesized to determine mode choice. ηt is a set of year
dummies to filter macro-economic shocks that are common across
firms. ui ~ N(0,1) accounts for firm-specific unobserved heterogene-
ity and is assumed to be independent of the model's random error
term. Φ is the cumulative density function of the standard normal
distribution. As the variance of the random error term is normalized
to 1, the underlying variance of the model's disturbance is thus equal
to 1 + σu2. For the same individual firm, the disturbances in different
two periods are correlated because of their common component, ui,
and their correlation (ρ) is a constant given by ρ = σu2 /1+σu2. The
parameters for estimation in this model include the coefficient vector
β and the parameter ρ. The parameters will be fitted by maximum
likelihood estimation. A statistically significant estimate of ρ would
suggest the importance of controlling for unobserved heterogeneity,
and imply that the random effects probit model is a better
specification than the pooled probit model where ρ is assumed to
be zero.
The dynamic version (or habit-persistence version) of the random
effects probit model can be written as
h
Prob MODEit = 1 jMODE i;t−1 ; xit ; ui
i ð
 
= Φ α + γMODEi;t−1 + β′ xit + ηt + σu ui ðfor t = 2; …; Ti Þ
ð2Þ
where the observed entry mode in the previous period, MODEi,t − 1, is
included. The coefficient γ represents the degree of state dependence.
Correlation between ui (time-invariant term) and MODEi,t − 1 makes
the latter endogenous in (2). Estimation requires an assumption about
the relation between the first period observations (initial conditions),
MODEi1, and ui. If the initial conditions are assumed to be exogenous,
i.e., MODEi1 and ui are not correlated, which may be appropriate if the
start of the process is the start of the observation period for each firm,
the maximum likelihood random effects probit estimation mentioned
above can be used. If the initial conditions are correlated with ui, as
would be expected in the current study, this estimation method will
be inconsistent and overstate state dependence.
Heckman (1981b) suggests an approximation to the process
generating the first period observations using the same form of
equation as for the rest of the observations but with some restrictions.
Heckman's approach has been used in the studies of dynamic
(intertemporal) labor force participation (e.g., Hyslop, 1999; Stewart,
2007). In the current context, a suggested equation for the first period
observations is


Prob ½ MODEi1 = 1jxi1 ; ui  = Φ ξ + δ′ xi1 + τu ui ðfor t = 1Þ
The model of Eq. (2) is retained for t ≥ 2. The same random effect,
ui, appears in both Eqs. (2) and (3) but has different scaling
parameters, σu and τu, respectively. Exogeneity of the initial
conditions corresponds to su = 0 and can be tested accordingly.
Eqs. (2) and (3) specify a complete model for the dynamic process.
A simplified procedure to implement Heckman's estimator involves
the creation of two dummy variables: dit = 1 if the observations
belong to period 1, and dit = 0 otherwise; fit = 1–dit = 1 if the
441
observations belong to all periods except period 1, and fit = 0 other-
wise (Arulampalam and Stewart, 2007). This is also the procedure
adopted in LIMDEP program (version 9.0, 2007). The encompassing
model is thus
ð1Þ h i
Prob MODEit = 1jMODEi;t1 ; xit ; xi1 ; ui
½
= Φ γMODEi;t1 + β′ xit + δ′ xi1 + ηt + ðα + σ u ui Þfit ð4Þ

+ ðξ + τ u ui Þdit ðfor t = 1; …; Ti Þ
The coefficients on fit and dit are viewed as random coefficients
that share the common random component, ui, as both σuui and τuui
appear in the equation. The contribution of observations from firm i to
the likelihood is the joint density
Li ðΛ jMODEi ; xi ; ui Þ
T
½
= ∏t i = 1 Φ γMODEi;t1 + β′ xit ð5Þ
′
+ δ xi1 + ηt + ðα + σu ui Þfit + ðξ + τu ui Þdit 
where Λ represents the full set of structural parameters, including
nonrandom slope parameters and random parameters. Since ui is
unobserved, it is necessary to operate the unconditional likelihood,
i.e., ui must be integrated out. The contribution of firm i to the
unconditional likelihood is thus given by
Li ðΛ jMODEi ; xi Þ
T
½
= ∫ui ∏t i = 1 Φ γMODEi;t1 + β′ xit + δ′ xi1 ð6Þ
+ ηt + ðα + σu ui Þfit + ðξ + τu ui Þdit Þ gðu Þdu
i i
where g(⋅) is the probability density function of the standard normal
distribution. The integrals will not exist in closed form, but since they
are of the form of expectation, they can be estimated by simulation.
The simulated log-likelihood is
n 1 R
R
T
½
logLs = ∑i = 1 log ∑r = 1 ∏t i = 1 Φ γMODEi;t1 + β′ xit + δ′ xi1
    ð7Þ
+ ηt + α + σu ui;r fit + ξ + τ u ui;r dit 
The simulation is over R draws on ui for each firm i. The maximum
simulated likelihood (MSL) estimator can be obtained by maximizing
Eq. (7) over the full set of structural parameters. We use LIMDEP
program, with the simulation over 500 draws (R = 500), to estimate
the model. The relevant theory of simulation based estimation can be
referred to Greene's (2007) textbook.
3. Determinants of entry mode choice
In this section, by reviewing the literature we discuss the
hypothesized relations between transaction cost factors and entry
mode choice. Besides, since this paper emphasizes state dependence
ð3Þ
of mode choice in a dynamic framework, previous mode performance
is also a key factor to determine the current mode choice.
3.1. Scale of the firm
Relative to large MNEs, smaller ones suffer from fewer productive
resources such as managerial skills, human resources, or financial
resources, and have limited complementary assets. They are also
exposed to more risks and uncertainty as a failure of foreign
investment could possibly lead the entire enterprise to bankruptcy.
442 M.-Y. Chen, J.-Y. Chang / Economic Modelling 28 (2011) 439–450
These inherent weaknesses may encourage small MNEs to make JV
agreements with local firms that have close relationship with local
environment and have easier access to information channels and
assets (Asiedu and Esfahani, 2001; Kogut and Singh, 1988; Mutinelli
and Piscitello, 1998). In addition, large MNEs with large worldwide
networks of subsidiaries are more capable of exploiting global
economics of scale than small ones. Global strategies may cause
global profit maximization by MNEs to diverge from local profit
maximization by their JV partners. The action of the partners will
affect MNEs' operation in other countries, while having no such
external effect on local partners' operation (Gomes-Casseres, 1989).
This spillover effect of JVs is more likely to appear in large MNEs that
operate subsidiaries in more than one host countries and have more
cross-border transactions within the firm. WOSs will thus be a
preferred entry mode for large MNEs that tend to have goal conflicts
with local firms.
3.2. Relative size of the subsidiary
The entry mode decision can be determined by the trade-off
between risks and returns and by resource availability and need for
control (Agarwal and Ramaswami, 1992). A large investment is
usually associated with high risks and returns and requires the ability
of MNEs to secure financial resources. When the size of the subsidiary
relative to the parent (or to the entire enterprise) is large, MNEs have
more resources committed to the local subsidiaries. They will have
greater incentives to choose WOSs with tight control because the
subsidiary's performance has more significant impact on the overall
performance of the firm (Bhaumik and Gelb, 2005). However, this
propensity of favoring WOSs tends to diminish as the MNE's scale
increases. For a given relative size of the subsidiary, the absolute size
rises as the MNEs grows larger. The incentives for firms to use JV
modes become stronger so as to acquire local partners' expertise of
running a large organization in the host country (Gatignon and
Anderson, 1988).
3.3. International experience
Many studies have argued that the more international experience
a firm possesses the more efficient it is to gain full control of
subsidiaries (Anderson and Gatignon, 1986; Chang and Rosenzweig,
2001; Delios and Beamish, 1999; Gatignon and Anderson, 1988;
Gomes-Casseres, 1989; Hennart, 1991; Mutinelli and Piscitello, 1998).
The firms with less experience in managing foreign production lack
the knowledge of dealing with local economic and environmental
conditions. That knowledge, which is tacit and subject to high
transaction costs, usually accrues to local firms. Less-experienced
firms are thus not willing to make significant resource commitments
and tend to have recourse to local firms. The tendency of using WOSs
increases as the parent accumulates experience and develops
expertise and knowledge in managing international operations.
Also, from the view of internal uncertainty in Williamson's (1975,
1985) transaction cost framework, unlike experienced MNEs that may
have developed strong internal control mechanisms to reduce
internal uncertainty, inexperienced MNEs are less capable of knowing
how to monitor and evaluate inputs appropriately and will make
considerable errors to depress efficiency if they try to exercise higher
control (Anderson and Gatignon, 1986; Gatignon and Anderson,
1988). Low-control modes such as JVs become more attractive for
inexperienced MNEs that cannot manage a wholly-owned integrated
structure efficiently.
On the contrary, Anderson and Gatignon (1986) indicate that
based on the ethnocentric orientation of international neophytes,
international experience would be negatively related to the level of
ownership control. Ethnocentrism leads inexperienced firms to
demand high ownership first to develop their own advantages by
holding key positions. Over time, when the firms have gained local
knowledge and have adapted to local conditions, they are more
willing to delegate control and to use JV modes. Davidson and
McFetridge (1984, 1985) also argue that experience encourages JVs
because experienced firms are able to enforce, monitor, and control
the JV agreements more effectively after acquiring that kind of
experience by running an integrated operation.
3.4. Technological capability and sources of the subsidiary's technology
The transaction cost theory traditionally explains the association
between technological capability and entry mode choice by focusing
on the tacit knowledge transfers among firms. MNEs with high R&D
capability tend to transfer proprietary technology of products or
processes to their subsidiaries. They should prefer a complete control
of subsidiaries. One reason is that it is difficult to obtain a fair return
for the technology contributed by the parent to the JVs (Hennart,
1982). Another reason is the fear of leaking proprietary technology
beyond the JVs and losing technological advantages to their rivals
(Anderson and Gatignon, 1986; Bhaumik and Gelb, 2005; Hennart,
1982, 1991; Hill et al., 1990; Mutinelli and Piscitello, 1998). For
Taiwanese FDI in China, while the main objective of Taiwanese firms is
to learn more about the local business conditions, their venturing
partners' objective usually is to access proprietary technology that
gives them competitive advantages. Thus, the threat of proprietary
technology leakage encourages Taiwanese MNEs having higher R&D
capability to choose WOSs in China.
It can be further argued that the entry mode decision is related to
the sources of the subsidiary's technology. An MNE will have strong
incentives to choose WOSs if the technology of its subsidiary is
proprietarily sourced from the parent, because it will especially
concern about the risk of dissemination of technological know-how.
On the other hand, JVs will be appropriate for an MNE whose
subsidiary's technology is outsourced to third-party providers. The
subsidiary can benefit from the opportunity of learning in cooperation
and from the creation of its own capability via interfirm knowledge
transfers.
3.5. Product markets and sources of materials
When the products of foreign subsidiaries are sold to the local
market, rather than exported to the home market or other
countries, MNEs will desire more local market knowledge in
order to deal with uncertainty in the product market, such as the
unexpected changes in product demand and customer's preferences
and the availability of substitute or complementary products
(Brouthers et al., 2002). That knowledge can be readily obtained
through venturing with local firms. Therefore, MNEs tend to choose
JVs if their subsidiaries primarily sell to the local market. Hennart
(1991) and Hennart and Zeng (2005) also indicate that the export
of the subsidiary's products may compete with domestic or third
country's sales of the parent and cause the conflict between global
and local profit maximization pursued by MNEs and their venturing
partners respectively. When the subsidiary's products are sold
locally, the goal conflict is expected to be smaller and JVs are likely
to be chosen.
The tangible nature of products for manufacturing MNEs makes
them perceptive to the variations in quality, quantity, and timing of
raw materials and intermediate inputs (Brouthers et al., 2002;
Richardson, 1993). When subsidiaries have to obtain materials from
local suppliers, uncertainty implied in such transactions will motivate
MNEs to engage in JVs so as to rely on local partners that already
possess reliable and constant provision of materials or have developed
networks of outside suppliers. Apparently, if the materials of
subsidiaries are directly provided by parent firms, the secured source
should lead to use of WOSs.
 M.-Y. Chen, J.-Y. Chang / Economic Modelling 28 (2011) 439–450
3.6. Resource-intensive industries
MNEs are likely to venture with local firms if their products are
intensive in the resources that make a host country attractive as a
location of foreign investment. Entering into a country in the form of
JVs can reduce transaction costs of acquiring local resources via
developing the business relation with local firms controlling the
access to resources. This will be the case in the industries where local
backward integration into natural resource is important. The higher
tendency towards JVs in resource-intensive industries may be due to
that local firms have exclusive or privileged access to domestic supply
of agricultural or mineral resources which can be vertically integrated
with processing, and that there may not be well-functioning
competitive markets for these resources because local institutions,
business regulations, and government policies discourage or prohibit
wholly-owned foreign investment in resource industries, particularly
in emerging economies (Chen and Hennart, 2002; Gomes-Casseres,
1989; Hennart, 1991; Mutinelli and Piscitello, 1998). Besides,
resources-intensive industries are less dependent on a firm's
complicated R&D and marketing capabilities so that the experience
of local firms is relatively important in such basic industries than in
more advanced industries (Gomes-Casseres, 1989). This also predicts
the higher possibility that JVs are chosen in resource-intensive
sectors.
3.7. Production relations between the subsidiary and the parent firm
In the production system the parent firm and its subsidiaries can
be linked vertically or horizontally. Vertically linkages are normally at
a stage of the production process that comes either before or after the
main activity of the parent firm. Subsidiaries either turn raw materials
into semi-manufactured products or turn semi-manufactured pro-
ducts into final products. Horizontal linkages involve use of the same
or modified technology to produce more or less the same product.
Brouthers and Hennart (2007) point out that Williamson's concept of
asset specificity is more appropriately applicable to vertical invest-
ments although many entry mode studies have used it to explain
horizontal expansions. Asset specificity implies bilateral dependence
between transacting partners. Coordinated efforts are required to
solve opportunism and self-interested bargaining. Due to the
increased costs associated with the coordinated efforts, transaction
cost economics suggests a positive relation between the degree of
asset specificity and the likelihood that WOSs are chosen (Yiu and
Makino, 2002). Therefore, vertical linkages between foreign subsid-
iaries and their parent firms, implying more attributes of asset
specificity, should encourage the use of WOSs.
The propensity of choosing WOSs when the parent firm and its
subsidiary are vertically connected may differ across industries.
Relative to less-advanced industries, such as resource-intensive
industries, advanced industries have more complicated vertical
transactions for material and intermediate goods or for marketing
and distribution, and hence are likely confronted with the contractual
holdup problems. One can argue that the vertically-linked production
relationship in motivating MNEs to choose WOSs tends to appear in
non-resource-intensive industries.
3.8. Diversification
Several studies have shown that MNEs will enter a foreign market
through JVs when they are diversified into a different industry
(Asiedu and Esfahani, 2001; Gomes-Casseres, 1989; Hennart, 1991;
Mutinelli and Piscitello, 1998). The parent firm may find that the
product-specific knowledge and distribution access needed to enter
into a new industry are subject to high transaction costs and are
difficult to acquire in the market. Joint venturing with local firms in
the new industry is the most efficient way to obtain relevant
443
resources. However, the structural problems faced by JVs (Hennart
and Zeng, 2005) are more likely to emerge when MNEs enter into an
unfamiliar industry. Venturing with local firms that have tacit
knowledge specific to the new industry implies greater asymmetric
dependence on local partners. This allows local firms to expropriate
the quasi-rent stream of foreign investors by taking advantage of
unspecified and unenforceable elements of the contract. Anticipating
this potential costly redistribution, MNEs may not choose JV modes.
Thus, because of two opposite influences, the relation between
diversified investment and mode choice may be ambiguous.
Nevertheless, the consideration of avoiding the threat of expro-
priation will play a more significant role in the FDI activities of small
and medium-sized enterprises (SMEs). Contrast to large enterprises,
SMEs are generally less-experienced internationally and do not have
well-developed systems and processes for information access and
capital availability. Since SMEs spend most of their resources in a
unique foreign market, they may be most concerned with the
unforeseen opportunistic behavior by the local partners when joint
venturing into a new industry. Thus, adoption of WOSs in diversified
investment is likely to appear in small SMEs. However, this tendency
decreases as firms grow and have greater abilities to solve holdup
problems in JV contracts.
3.9. Relational network
The social relational network is important for SMEs' internation-
alization process (Bell, 1995; Chen and Chen, 1998; Coviello and
Munro, 1997). The relational network, which is built on the
relationship among relatives, friends, alumni, and religious beliefs,
establishes trust norms and creates highly elastic external linkages.
These linkages encompass non-contractual transaction based on
inter-personal links and trust beyond pure business relations, and
provide technology, entrepreneurial know-how and market access to
support cross-border operations (Chen and Chen, 1998). Members of
the relational network value relationship rather than discrete
transactions so that opportunistic behavior is expected to be
controlled and minimized (Coviello and Munro, 1997). A firm short
of firm-specific assets might still succeed in making foreign
investment because the relational network helps it overcome entry
barriers of foreign markets, enables it to access into local comple-
mentary resources markets, and facilitates rapid acquisition of capital
and trade connections.
The relational network is important for Taiwanese FDI activities. As
in the footwear industry exemplified by Chen and Chen (1998), the
network that creates mutual trust and understanding facilitates SMEs
to maintain business by following Taiwanese or foreign clients to host
countries in response to rising of wage and manufacturing costs. The
relational network gives assistance akin to that from local partners,
and allows foreign investors to be less reliant on local firms. Thus, the
stronger are the relational network connections, the more likely an
MNE will choose WOSs.
3.10. Previous mode performance
MNEs can learn from the previous mode and determine current
mode choice depending on whether previous mode performance is
favorable or unfavorable (Brouthers and Hennart, 2007). In the
dynamic choice process, as current mode choice is assumed to be
affected by the previous mode, it is expected that previous mode
performance will influence the dependence between current and
previous mode choices. Poor performance should reduce the degree of
state dependence, in the sense that the likelihood of choosing WOSs in
the current period is reduced if the wholly-owned mode used in the
previous period produced poor performance. Such firms might think
that changing the mode to JVs will improve performance.
444 M.-Y. Chen, J.-Y. Chang / Economic Modelling 28 (2011) 439–450
Table 1
The number of FDI firms in 2004–2007 annual surveys.
Number of FDI firms in the survey
Number of manufacturing FDI firms in the survey
Number of manufacturing FDI firms with China being the major FDI country
Number of manufacturing FDI firms with China being the major FDI country as
a percentage of number of manufacturing FDI firms
Source: Department of Statistics, Ministry of Economic Affairs of Taiwan.
4. Data and variable measurement
Many of studies for Taiwanese FDI in China put focus on the effect
of investment towards China on Taiwanese domestic industries.
Relatively fewer academic studies in Taiwan address entry mode
choice in the Chinese market, and they are generally based on
questionnaires sent by researchers themselves to a selected number
of firms with a low recovery rate. In this paper, the sample is taken
from annual survey data on Taiwanese FDI conducted by Department
of Statistics, Ministry of Economic Affairs of Taiwan. The survey is
based on the questionnaires sent to all firms that are approved to
make foreign investment by the Investment Commission of Ministry
of Economic Affairs. It is implemented annually during May and June
and published in August the following year. The firms that have been
approved to make foreign investment may not be the ones recorded in
each year's survey because of closure, moving out of business, or
withdrawing the capital from the investing regions. Although the
official survey is not designed for the investigation of entry modes,
careful consideration on its questionnaire certainly provides us a
reliable source to analyze the choice of entry mode between WOSs
and JVs.
Table 1 reports the total number of FDI firms, the number of
manufacturing FDI firms, and the number of manufacturing FDI firms
of which China (including Hong Kong) is the major investing country
for the 2004–2007 surveys (2003–2006 FDI activities). As can be seen,
China undoubtedly is the main country for Taiwanese FDI. We use the
manufacturing firms with the major FDI country being China from
2004 to 2007 surveys to construct the panel data set.1 Furthermore, in
order to implement the estimation of the dynamic probit mode, the
final sample firms are restricted to those appearing in the surveys for
at least two consecutive years. Accordingly, there are 1240 FDI firms
with 3841 observations in an unbalanced panel. Among them, 568
firms have full observations for 4 consecutive surveys (2004–2007),
225 firms have observations for 3 consecutive surveys (either 2004–
2006 or 2005–2007), and 447 firms have observations for 2
consecutive surveys (2004–2005, 2005–2006, or 2006–2007). The
detailed descriptions for the measurement of entry mode and all
variables proxy for the determinants hypothesized in Section 3 are
presented in Table 2.
5. Empirical results
5.1. Descriptive statistics
The proportion of observations that choose WOSs as the entry
mode, either unconditional or conditional on mode at t − 1, is
presented in Table 3. The first row of the table shows that there is
considerable state dependence in entry mode choice. The proportion
of observations choosing WOSs at t for those having chosen WOSs at
t−1 is much higher than that for those having chosen JVs at t − 1
1
There are a lot of changes in the questions of the questionnaire since 2004 so that
the panel data set starts from 2004 and does not include the surveys prior to 2004.
2004 survey 2005 survey 2006 survey 2007 survey
1880 1711 1677 1770
1741 1434 1367 1485
1307 1145 1106 1198
75.07% 79.85% 80.38% 80.67%
(0.903 vs. 0.264). The remaining rows suggest that state dependence
appears in all subgroups distinguished by the factors that are
hypothesized to influence entry mode choice. The propensity of
choosing WOSs can also be due to heterogeneity across firms. As can be
seen, the unconditional proportion of using WOSs is higher for the
MNEs investing in other countries besides China, having more
employees and sales, and possessing less experience and high
technology capability. It is also higher for the subsidiaries with the
technology source being the Taiwanese parent, with a smaller
percentage of products sold to the local market and more reliance on
Taiwanese material supplies, and vertically linked to the parent's
production. The subsidiaries that operate in non-resource-intensive
industries or in different industries from the parent, or have an
intensive relational network are more likely to choose WOSs.
However, it seems that the heterogeneity-induced difference in
choosing WOSs within all subgroups is more evident when the mode
at t − 1 is JVs. That difference conditional on WOSs at t − 1 is trivial for
several subgroups. This might be due to that the effect of state
dependence causes the firms to maintain WOSs and mitigates the role
of firm heterogeneity in influencing mode choice.
Table 4 reports the means and standard deviations of the variables
for all observations and for observations classified by the entry mode.
Consistent with the implications of Table 3, the test for the equality of
means between WOS and JV observations suggests that on the
average the firms choosing WOSs tend to be larger in either the scale
of multinational operations, total number of employees, or total sales.
These firms possess higher technology capability, have the subsidi-
ary's technology directly transferred from Taiwanese parent, sell
fewer products to the local market, and obtain more materials from
Taiwan. They also have vertical parent–subsidiary production rela-
tion, operate in non-resource-intensive industries, and have more
relational network connections.
5.2. Regression results
We firstly report in Table 5 the results of static probit regressions
that focus on the impact of firm characteristics and do not consider the
dependence between past and current entry modes. In column (1),
results from the pooled probit model are presented since a static
model can also be estimated by pooling the data and ignoring the
correlation structure to obtain consistent parameter estimates
(Maddala, 1987). Column (2) contains the results from the probit
model that allows for individual-specific random effects. The
estimated value of ρ is 0.798, which is significantly greater than
zero, suggesting that the random effects probit model controlling for
unobserved heterogeneity is a better specification and that 79.8% of
the latent disturbance variance is due to unobserved heterogeneity. In
column (3), three interaction terms are included into the regression to
consider specific arguments mentioned earlier. The similar value of ρ
is found in column (3).
It must be noted that random effects probit and pooled probit
models have different normalizations in LIMDEP program to facilitate
easy estimation. In the random effects probit model, the estimates of β
 M.-Y. Chen, J.-Y. Chang / Economic Modelling 28 (2011) 439–450 445

Table 2
Variable measurement.

Variable name Description

Entry mode:
MODE MODE is a dummy variable for entry mode choice. It is equal to 1 for WOSs and 0 for JVs. In the survey, MNEs are asked to identify the organizational form of
the foreign subsidiaries in their major FDI countries from the following items: (1) WOSs; (2) JVs with Taiwanese firms; (3) JVs with local Taiwanese firms; (4)
JVs with local non-Taiwanese firms; (5) JVs with local governments; (6) JVs with foreign firms; (7) JVs with Taiwanese individuals; (8) JVs with local
individuals. In this study, the MNEs checking (1) is defined as having WOSs (MODE = 1) and those checking (3), (4), (5), (8) are categorized to have JVs
(MODE = 0). Some MNEs check (2) or (7) only and do not check either one of (1), (3), (4), (5), or (8). For these samples, since the hypotheses of entry mode
choice in the literature are largely based on the benefits and costs of venturing with local entities, we classify them to have WOSs, i.e., we include the cases of
multiple parents as Hennart (1991) has done. Finally, few samples checking (6) only are classified to have a JV mode because one can reasonably conjecture
that the foreign firms might have already established their relationships with local entities.
Scale of the firm:
LMNE, lnEMP The MNE's scale is measured by three variables. LMNE is a dummy variable, which is equal to 1 if the MNE makes investment in other foreign countries in
and lnSALE addition to China (including Hong Kong), and 0 if it invests in China only. This variable is proxy for the scope of an MNE's worldwide network. In the survey,
MNEs are asked to report the major, second major, and third major FDI countries. Most MNEs have only one FDI country. Thus we divide the sample firms, of
which the major FDI country is China, into two groups: one has other FDI countries in addition to China and the other does not. lnEMP is the log of total
number of employees of the MNE. lnSALE is the log of total sales (thousand Taiwan dollars) of the MNE. In response to the survey, MNEs report their total
number of employees and total sales.
Relative size of the subsidiary:
SUBSIZE SUBSIZE is the ratio of the subsidiary's sales to total sales of the MNE. Previous studies usually use assets to measure the relative size. Unfortunately, data for
the assets are not reported in the survey. Also, we do not use the ratio of the subsidiary's sales to the parent's sales in measuring the relative size. Some MNEs,
particularly traditional manufacturing firms, report very low parent's sales in the survey primarily because the focus of the entire enterprise has been moved
to China. This situation will make variable values using the ratio of the subsidiary's sales to the parent's sales fall into a very wide range and cause the model
estimation to be difficult and untraceable. Therefore, we use the ratio of the subsidiary's sales to total sales so that its values can be confined to the range
between 0 and 1.
International experience:
lnEXPER lnEXPER is the log of the number of years since the MNE has engaged in FDI. In the survey, MNEs are asked to fill in the beginning years of investing in major,
second major, and third major FDI countries. The beginning year of FDI is thus defined as the earliest one among them, and it is defined as the year of entering
into the major country for MNEs that report the major FDI country only. It might be possible that an MNE has completely exited from a specific country that is
its first foreign market. But such information is unavailable from the survey.
Technological capability:
RDLOW and RDLOW is a dummy variable equal to 1 if the MNE's R&D intensity, defined as the R&D expenditure as a percentage of annual sales, is less than 0.03 and greater
RDHIGH than zero. RDHIGH is a dummy variable equal to 1 if the MNE's R&D intensity is greater than 0.03. The omitted group is for zero R&D intensity. MNEs report
their annual R&D expenditure in the survey. Using 0.03 as the turning point seems an arbitrary choice; however, it produces better results to distinguish the
effects between low and high R&D intensity while other turning points also show a similar pattern of the R&D effect.
Sources of the subsidiary's technology:
TECHTW and TECHTW is a dummy variable equal to 1 if the subsidiary's technology is transferred directly from Taiwanese parent firm. TECHOS is a dummy variable equal to
TECHOS 1 if the subsidiary outsources its technology. The omitted group is for the subsidiaries that internalize technology development. In the survey, MNEs are asked
to identify the sources of the subsidiary's technology in the major FDI country from the following items: (1) provided by Taiwanese parent; (2) provided by
cooperative firms; (3) learned in the counseling process; (4) jointly developed with other firms; (5) developed by the subsidiary itself; (6) recruited from
other firms; (7) commissioned to outsiders; (8) purchased from outsiders; (9) learned from other firms. According to MNEs' responses, we define TECHTW as
the category of firms that check (1) as the technology source, and TECHOS as the category of firms that check (2), (6), (7), (8) as the technology source.
Internalization of technology development (the omitted group) is for the firms checking (3), (4), (5), (9).
Product markets and sources of materials:
LOCMKT and LOCMKT is a dummy variable equal to 1 if more than 50% of the subsidiary's products are sold to the local market, i.e., the Chinese market. MATTW is a dummy
MATTW variable equal to 1 if more than 50% of the subsidiary's raw materials or intermediate goods are provided by Taiwanese parent firm. In the survey, for the
subsidiary's product sales, MNEs are asked to write down the percentage sold to the local markets, the percentage sold back to Taiwan, and the percentage
exported to other countries. For either sources of materials or sources of intermediate goods, MNEs report the percentage provided by Taiwanese parent firm,
by local Taiwanese businesses, by local non-Taiwanese businesses, and by other countries.
Resource-intensive industries:
RESOURCE RESOURCE is a dummy variable equal to 1 if the subsidiary's main product in China belongs to a resource-intensive manufacturing industry defined by the
Standard Industrial Classification System of Taiwan, i.e., SIC 08 (food and beverages), 09 (tobacco), 10 (textiles mills), 11 (apparel, clothing accessories and
other textile product), 12 (leather, fur and allied product), 13 (wood and bamboo products), 14 (furniture and fixtures), 15 (pulp, paper and paper products),
19 (petroleum and coal products), 20 (rubber products), 21 (plastic products), 22 (non-metallic mineral products), 23 (basic metal industries), 24 (fabricated
metal products). In the survey, MNEs are required to indicate the industries of major, second major, and third major products in the FDI countries.
Production relations between the subsidiary and the parent firm:
VERTLINK VERTLINK is a dummy variable equal to 1 if the subsidiary's production is vertically linked to the parent's production. In the survey, MNEs are asked to identify
the production relation between the subsidiary and the parent firm from 13 items of the descriptions. We categorize the MNEs that identify the relation as one
of the following four items to have the vertical linkages: (1) Taiwanese parent does not have the production sector and is responsible only for selling the
foreign subsidiary's products; (2) The foreign subsidiary does not have the production sector and is responsible only for selling the products of Taiwanese
parent; (3) Taiwanese parent produces components and semi-manufactured products, and the foreign subsidiary assembles and produces final products; (4)
the foreign subsidiary produces components and semi-manufactured products, and Taiwanese parent assembles and produces final products.
Diversification:
DIVERSE DIVERSE is a dummy variable equal to 1 if the subsidiary's main product in China belongs to a different industry from Taiwanese parent, according to the
Standard Industrial Classification System of Taiwan.
Relational network:
RELCUST and Two variables are proxy for the relational network connections. RELCUST is a dummy variable equal to 1 if the MNE follows Taiwanese or foreign clients to
RELFIRM make FDI in China. An MNE is defined to have this kind of network if in the survey it identifies the motivation of FDI in China as “to make investment along
with Taiwanese customers” or “to meet the request of foreign customers”. The second variable, RELFIRM, is the log of the number of Taiwanese MNEs that
make FDI in China in the same industry defined by the Standard Industrial Classification System of Taiwan. A large number of firms from Taiwan may imply
that relational network connections are more likely to be established.
Previous mode performance:
PERF-SUB and Previous mode performance is an explanatory variable interacted with the previous mode in the dynamic model to examine whether MNEs adjust the mode in
PERF-PAR response to unfavorable previous performance. In the survey, MNEs are asked to subjectively evaluate the profitability of their parent firm and foreign
subsidiaries. We consider the subsidiary's performance (PERF-SUB) and the parent's performance (PERF-PAR) respectively. Both PERF-SUB and PERF-PAR are
dummy variables equal to 1 if MNEs identify the performance as having loss or breakeven, and equal to 0 for having surplus.
446 M.-Y. Chen, J.-Y. Chang / Economic Modelling 28 (2011) 439–450

Table 3 Table 4
Unconditional and conditional proportions of observations choosing WOSs. Means and standard deviations of variables.

Number of Unconditional Conditional Conditional Variable All observations Observations Observations

observations on WOSs on JVs at for WOSs for JVs
at t − 1 t−1
(N = 3841) (N = 2673) (N = 1168)
All observations 2601 0.705 0.903 0.264
Mean Standard Mean Standard Mean Standard
LMNE = 1 698 0.739 0.885 0.335
deviation deviation deviation
LMNE = 0 1903 0.693 0.910 0.242
lnEMP ≥ 5.298a 1524 0.760 0.912 0.306 MODE 0.696 0.460
lnEMP b 5.298 1077 0.629 0.889 0.225 LMNE 0.268 0.443 0.283⁎ 0.451 0.232 0.422
lnSALE ≥ 13.816b 1095 0.752 0.907 0.309 lnEMP 5.627 1.588 5.777⁎ 1.562 5.283 1.592
lnSALE b 13.816 1506 0.672 0.900 0.239 lnSALE 13.291 2.029 13.416⁎ 1.996 13.005 2.075
SUBSIZE ≥ 0.3 1356 0.703 0.892 0.284 SUBSIZE 0.375 0.307 0.375 0.309 0.375 0.303
SUBSIZE b 0.3 1245 0.708 0.915 0.241 lnEXPER 1.915 0.638 1.912 0.633 1.922 0.650
lnEXPER ≥ 2.015c 1380 0.695 0.895 0.255 RDHIGH 0.207 0.405 0.230⁎ 0.421 0.155 0.362
lnEXPER b 2.015 1221 0.717 0.913 0.273 RDLOW 0.507 0.500 0.516 0.500 0.485 0.500
RDHIGH = 1 527 0.776 0.936 0.246 TECHTW 0.558 0.497 0.624⁎ 0.485 0.408 0.492
RDLOW = 1 1370 0.716 0.900 0.275 TECHOS 0.115 0.319 0.082⁎ 0.274 0.191 0.393
RDLOW = RDHIGH = 0 704 0.632 0.880 0.254 LOCMKT 0.452 0.498 0.436⁎ 0.496 0.491 0.500
TECHTW = 1 1428 0.785 0.926 0.309 MATTW 0.278 0.448 0.297⁎ 0.457 0.234 0.423
TECHOS = 1 311 0.534 0.832 0.188 RESOURCE 0.386 0.487 0.350⁎ 0.477 0.469 0.499
TECHTW = TECHOS = 0 862 0.636 0.877 0.252 VERTLINK 0.316 0.465 0.342⁎ 0.474 0.259 0.438
LOCMKT = 1 1203 0.679 0.900 0.232 DIVERSE 0.451 0.498 0.461 0.499 0.430 0.495
LOCMKT = 0 1398 0.728 0.906 0.295 RELCUST 0.615 0.487 0.634⁎ 0.482 0.573 0.495
MATTW = 1 676 0.747 0.910 0.276 RELFIRM 4.557 1.092 4.647⁎ 1.103 4.352 1.037
MATTW = 0 1925 0.691 0.900 0.260
⁎ Represents that the difference in mean between observations for WOSs and
RESOURCE = 1 1000 0.649 0.901 0.241
observations for JVs is statistically significant at the 0.01 level.
RESOURCE = 0 1601 0.741 0.904 0.284
VERTLINK = 1 842 0.757 0.902 0.309
VERTLINK = 0 1759 0.681 0.904 0.248
DIVERSE = 1 1170 0.724 0.912 0.289
International experience (lnEXPER) is negatively related to the
DIVERSE = 0 1431 0.690 0.896 0.244
RELCUST = 1 939 0.724 0.914 0.288 propensity of using WOSs in both columns (2) and (3). However, the
RELCUST = 0 1662 0.695 0.897 0.250 opposite implications of experience, as stated in Section 3, offset each
RELFIRM ≥ 4.605d 1258 0.746 0.906 0.303 other so that its impact is not very strong. The finding is more
RELFIRM b 4.605 1343 0.667 0.900 0.236 consistent with the argument of ethnocentrism. It also suggests that
Total number of observations is 2601, which is pooled data for period 2 to period Ti. for Taiwanese investment in China, experience is imperative for firms
a
The number of employees ≥ 200. to efficiently enforce JV agreements with Chinese local firms.
b
The value of sales ≥1 billion Taiwan dollars.
c Higher R&D intensity encourages Taiwanese firms to use WOSs in
The international experience ≥ 7.5 years.
d
The number of firms in the same industry ≥100. China. Compared with the group of firms without R&D investments,
the positive effect is significant for the group of RDHIGH but not
for RDLOW.2 This implies that the fear of local partners' access to
proprietary technology is evident when the MNE's technology
returned by the program is normalized by the standard deviation of capability is above a certain level. For an MNE only having low
the random error term, which is assumed to be 1 here. But parameter technology levels, the need of obtaining complementary technology
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
estimates of β = 1 þ σ u2 are obtained in the pooled probit model. To
from JVs becomes stronger and lowers the tendency of complete
control over the subsidiary. Consistent with these notions, it is found
make valid comparisons in coefficient estimates, it is necessary toffi
qffiffiffiffiffiffiffiffiffiffiffiffiffiffi that relative to those internalizing technology developments, MNEs
convert the random effects probit coefficients from β to β 1 þ σ 2u , = having the subsidiary technology proprietarily transferred from
that is, the random effects probit coefficients need to be multiplied by Taiwanese parent (TECHTW) are more likely to fully control their
pffiffiffiffiffiffiffiffiffiffi
ffi
the estimate of 1−ρ(Arulampalam, 1999), which is 0.449 in column subsidiaries. In contrast, MNEs outsourcing the subsidiary technology
(2). Consequently, compared with the pooled probit coefficients in (TECHOS) tend to choose JVs probably for the benefits from learning
column (1), as the random effect term captures the impact of and interfirm knowledge transfers.
unobserved heterogeneity, almost all “scaled” coefficients in the The coefficient on LOCMKT is negative. MNEs selling fewer
random effects probit of column (2) are reduced, except those of products to the local market are found to more likely choose WOSs.
SUBSIZE, lnEXPER and MATTW. In this case, JVs are not a preferred choice because MNEs are less
Regression results in Table 5 generally are consistent with the dependent on local partners to deal with the uncertainty in the local
predictions of the transaction cost theory. Let us focus on the results product market, or because they want to avoid the goal conflict with
from the random effects probit model. Column (2) shows that MNEs local partners when the subsidiary has more exports that might
operating subsidiaries in the countries besides China (LMNE = 1), compete with the sales of the parent firm. The positive coefficient on
hiring more employees (lnEMP), or having greater sales (lnSALE) tend MATTW shows that a more secured source of materials from
to choose WOSs. However, only the coefficient on lnEMP is significant Taiwanese parent will lead to use of WOSs due to less reliance on
at the 12% level. The coefficient on SUBSIZE is significantly positive, local firms' assistance for an assured provision of materials.
supporting the hypothesis that MNEs desire a higher level of control
over their subsidiaries when they have larger resources committed to
the subsidiaries. In column (3) where the interaction between
2
SUBSIZE and lnSALE is included, the positive effect of lnSALE turns The purpose of using two dummies for three levels of R&D intensity is to consider
to be significant and the interaction term has a significantly negative that the effect of R&D intensity may not be strictly linear. In unreported regressions,
we keep R&D intensity a continuous variable in running regressions and also include
coefficient. The positive effect of the relative size of the subsidiary on
its squared term. However, the results are not satisfactory. The significantly positive
the likelihood of choosing WOSs is shown to decrease as the scale of coefficient on RDHIGH suggests a threshold effect of R&D expenditures on entry mode
the firm (in total sales) becomes large. choice.
 M.-Y. Chen, J.-Y. Chang / Economic Modelling 28 (2011) 439–450 447

Table 5 Finally, it is evidenced that the variables RELCUST and RELFIRM,

Static probit estimation of entry mode choice. proxy for relational network connections, are positively related to the
Variable Pooled probit Random effects probit probability of choosing WOSs. The existence of Taiwanese or foreign
customers that are followed by MNEs when investing in China and
(1) (2) (3)
more number of Taiwanese MNEs that make foreign investment in the
Constant −0.629 −1.071 (0.576)⁎ −2.772
same industry are shown to provide the relational network similar to
(0.232)⁎⁎⁎ (0.698)⁎⁎⁎
LMNE 0.029 (0.057) 0.004 (0.131) 0.224 (0.161) assistance from local partners of the JVs, and hence facilitate
lnEMP 0.121 (0.022)⁎⁎⁎ 0.075 (0.048) 0.059 (0.049) Taiwanese investors to choose complete control of their subsidiaries
lnSALE 0.028 (0.019) 0.043 (0.045) 0.176 (0.059)⁎⁎⁎ as an entry mode.
SUBSIZE 0.078 (0.082) 0.342 (0.174)⁎⁎ 4.250 (1.057)⁎⁎⁎ Table 6 presents the results of dynamic probit estimation that
SUBSIZE × lnSALE −0.300
(0.080)⁎⁎⁎
includes previous entry mode (MODEt − 1) in the model. Column (1)
lnEXPER −0.008 (0.040) −0.135 (0.087) −0.164 (0.088)⁎ reports the results from the random effects probit model when the
RDHIGH 0.237 (0.069)⁎⁎⁎ 0.387 (0.151)⁎⁎⁎ 0.345 (0.154)⁎⁎ initial conditions are assumed to be exogenous. Since the estimated
RDLOW 0.071 (0.057) 0.079 (0.119) 0.066 (0.121) value of ρ is very small (b0.001), the regression results are almost
TECHTW 0.475 (0.049)⁎⁎⁎ 0.585 (0.092)⁎⁎⁎ 0.580 (0.092)⁎⁎⁎
identical to those of the pooled probit model. The results of the
TECHOS −0.255 −0.441 −0.474
(0.072)⁎⁎⁎ (0.134)⁎⁎⁎ (0.135)⁎⁎⁎ Heckman approach estimated by MSL are given in columns (2) and (3).
LOCMKT −0.103 −0.190 (0.101)⁎ −0.198 (0.102)⁎ The estimated value of τu is 1.734 in column (2) and 1.805 in column
(0.047)⁎⁎ (3). Both are significantly greater than zero, implying the rejection of
MATTW 0.120 (0.051)⁎⁎ 0.278 (0.107)⁎⁎⁎ 0.268 (0.107)⁎⁎ the exogeneity of the initial conditions. The coefficient on MODEt − 1 is
RESOURCE −0.121 −0.194 (0.133) −0.085 (0.141)
(0.051)⁎⁎
significantly positive, apparently suggesting positive state dependence
VERTLINK 0.143 (0.050)⁎⁎⁎ 0.152 (0.092)⁎ 0.308 (0.112)⁎⁎⁎ in entry mode choice. Compared to column (1), the Heckman
VERTLINK × RESOURCE −0.449 estimator shows a fall in the coefficient on MODEt − 1 of about a half.
(0.192)⁎⁎ Let us focus on column (3) which includes the interaction terms.
DIVERSE 0.084 (0.045)⁎ 0.152 (0.098) 0.293 (0.113)⁎⁎⁎
Similar to the results obtained from Table 5, all coefficient estimates
DIVERSE × LMNE −0.465
(0.203)⁎⁎ are as expected and most of them are statistically significant. The
RELCUST 0.115 (0.048)⁎⁎ 0.200 (0.100)⁎⁎ 0.211 (0.101)⁎⁎ predictions for the influences of firm heterogeneity on entry mode
RELFIRM 0.071 (0.023)⁎⁎⁎ 0.132 (0.055)⁎⁎ 0.131 (0.055)⁎⁎ choice based on transaction cost economics are robust even if the
ρ (= σ2u /1+σ2u) 0.798 (0.018)⁎⁎⁎ 0.799 (0.018)⁎⁎⁎ cross-period dependence of mode choice is taken into account in the
Log-likelihood −2179.569 −1707.242 −1695.117
dynamic model. In columns (4) and (5) of Table 6, the impact of
Number of firms 1240 1240 1240
Number of observations 3841 3841 3841 previous mode performance on the state dependence of mode choice
⁎⁎⁎, ⁎⁎ and ⁎ represent that the coefficients are statistically significant at 0.01, 0.05 and is considered by including the interactions of previous mode choice
0.1 levels respectively. The year dummies are included in the regressions. with subsidiary's performance (MODEt − 1 × PERF-SUBt − 1) and with
parent firm's performance (MODEt − 1 × PERF-PARt − 1) respectively.3
As indicated earlier, MNEs will learn from previous mode performance
to determine the current mode by looking into whether performance
As expected, the negative coefficient on RESOURCE means that is favorable or unfavorable. It is our purpose to know whether the
MNEs would not like to choose WOSs if their subsidiaries operate in parent firm's performance, which is likely affected by the MNE's global
the resource-intensive industries. But the importance of this effect is plans and strategies, or the subsidiary's performance itself determines
captured by the firm-specific term as its significance drops substan- the dynamic mode adjustments. Regression results show that both
tially when compared to that appeared in the pooled probit model. MODEt − 1 × PERF-SUBt − 1 and MODEt − 1 × PERF-PARt − 1 have nega-
Consistent with transaction cost arguments, the vertical production tive coefficients, but the former is not significant. The likelihood of
link between the subsidiary and the parent (VERTLINK), which has choosing WOSs in the current period decreases significantly if WOSs
more attributes of asset specificity, causes the likelihood of choosing used in the previous period were associated with poor parent firm's
WOSs to increase for the saving of coordination costs. In column (3), performance (PERF-PARt − 1 = 1). That is, if the previous mode is
the interaction between VERTLINK and RESOURCE has a significantly WOSs, poor parent's performance will cause the degree of state
negative coefficient, supporting the prediction that asset specificity is dependence on WOSs to be reduced and encourage MNEs to change
more likely to be a factor in favor of WOSs in non-resource-intensive the mode in the current period. The subsidiary's performance itself
industries. seems not an important consideration for an MNE to adjust the
The significance for the positive effect of DIVERSE is weak in organizational form of that subsidiary.
column (2) (only at the 12% level), but turns to be strong after the One can also interpret the above findings as that the marginal
interaction with LMNE is included in column (3). The coefficient on effect of MODEt − 1, i.e., the effect for a change of MODEt − 1 from 0 to 1,
the interaction term is significantly negative. These results have depends on the parent firm's previous performance (PERF-PARt − 1).
particular implications for Taiwanese foreign investment. The Using the estimated coefficients in column (5), the magnitude of the
Taiwanese firms investing in China only (LMNE = 0) are usually impact of PERF-PARt − 1 on the marginal effect of MODEt − 1 can be
SMEs. As stated earlier, because of their inherent disadvantages in calculated. It is important to note that the estimated coefficients in a
managerial resources and asymmetric dependence on local firms probit model are not the marginal effects we are accustomed to
when entering a new industry, SMEs are likely to be expropriated by analyzing. Like those of any nonlinear regression, the marginal effects
local firms if they choose JVs. This consideration motivates them to vary with the values of explanatory variables. For the purpose of
use WOSs. The structural problems in JVs can be alleviated for large demonstration, consider an exemplified firm in 2004 that has
MNEs that also make foreign investment in other countries
(LMNE = 1) since they may have more expertise learned from
3
multinational operation to deal with holdup problems. The negative As described in Table 2, one may notice that PERF-SUB and PERF-PAR are
relationship between diversification and WOSs, a result frequently dichotomous measure of performance. Only subjective measures (managerial
evaluations) of performance are available in the survey. This is common in the
found in the literature, tends to appear in the MNEs that operate in literature. As indicated by Brouthers (2002), previous survey-based studies have found
multiple countries and is reflected in the coefficient of the interaction that firms are reluctant to provide objective measures of performance for their foreign
term. subsidiaries. Subjective measures are usually employed.
448 M.-Y. Chen, J.-Y. Chang / Economic Modelling 28 (2011) 439–450

Table 6
Dynamic probit estimation of entry mode choice.

Variable MSL Heckman estimation of random effects probit

Random effects probit Considering the impact of previous performance

(pooled probit)a
(1) (2) (3) (4) (5)

Constant −0.910 (0.344)⁎⁎⁎

LMNE −0.025 (0.082) 0.003 (0.092) 0.151 (0.118) 0.152 (0.118) 0.145 (0.119)
lnEMP 0.077 (0.032)⁎⁎ 0.083 (0.036)⁎⁎ 0.076 (0.037)⁎⁎ 0.075 (0.037)⁎⁎ 0.073 (0.037)⁎
lnSALE −0.018 (0.028) 0.011 (0.031) 0.067 (0.043) 0.065 (0.043) 0.059 (0.043)
SUBSIZE −0.016 (0.115) 0.097 (0.129) 1.775 (0.767)⁎⁎ 1.795 (0.770)⁎⁎ 1.979 (0.784)⁎⁎
SUBSIZE × lnSALE −0.127 (0.058)⁎⁎ −0.129 (0.058)⁎⁎ −0.138 (0.059)⁎⁎
lnEXPER −0.044 (0.062) −0.099 (0.064) −0.109 (0.065)⁎ −0.109 (0.065)⁎ −0.097 (0.066)
RDHIGH 0.115 (0.101) 0.238 (0.117)⁎⁎ 0.244 (0.119)⁎⁎ 0.244 (0.119)⁎⁎ 0.241 (0.120)⁎⁎
RDLOW 0.009 (0.084) 0.031 (0.094) 0.035 (0.096) 0.034 (0.096) 0.027 (0.097)
TECHTW 0.265 (0.070)⁎⁎⁎ 0.373 (0.076)⁎⁎⁎ 0.380 (0.077)⁎⁎⁎ 0.381 (0.077)⁎⁎⁎ 0.392 (0.078)⁎⁎⁎
TECHOS −0.153 (0.101) −0.299 (0.110)⁎⁎⁎ −0.327 (0.113)⁎⁎⁎ −0.327 (0.113)⁎⁎⁎ −0.326 (0.114)⁎⁎⁎
LOCMKT −0.104 (0.067) −0.157(0.074) ⁎⁎ −0.173(0.075)⁎⁎ −0.173 (0.076)⁎⁎ −0.190 (0.077)⁎⁎
MATTW 0.002 (0.075) 0.089 (0.089) 0.091 (0.091) 0.094 (0.091) 0.101 (0.091)
RESOURCE 0.009 (0.073) −0.017 (0.082) 0.070 (0.093) 0.070 (0.093) 0.071 (0.094)
VERTLINK 0.057 (0.071) 0.069 (0.077) 0.192 (0.095)⁎⁎ 0.193 (0.096)⁎⁎ 0.214 (0.098)⁎⁎
VERTLINK × RESOURCE −0.344 (0.162)⁎⁎ −0.344 (0.163)⁎⁎ −0.370 (0.164)⁎⁎
DIVERSE 0.122 (0.065)⁎ 0.131 (0.072)⁎ 0.233 (0.086)⁎⁎⁎ 0.234 (0.086)⁎⁎⁎ 0.239 (0.086)⁎⁎⁎
DIVERSE × LMNE −0.332 (0.158)⁎⁎ −0.332 (0.158)⁎⁎ −0.334 (0.159)⁎⁎
RELCUST 0.143 (0.069)⁎⁎ 0.203 (0.077)⁎⁎⁎ 0.222 (0.079)⁎⁎⁎ 0.222 (0.079)⁎⁎⁎ 0.242 (0.080)⁎⁎⁎
RELFIRM 0.020 (0.033) 0.077 (0.038)⁎⁎ 0.077 (0.039)⁎⁎ 0.076 (0.039)⁎ 0.074 (0.039)⁎
MODEt − 1 1.860 (0.065)⁎⁎⁎ 0.975 (0.080)⁎⁎⁎ 0.899 (0.082)⁎⁎⁎ 0.932 (0.099)⁎⁎⁎ 1.015 (0.098)⁎⁎⁎
MODEt − 1 × PERF-SUBt − 1 −0.062 (0.101)
MODEt − 1 × PERF-PARt − 1 −0.317 (0.107) ⁎⁎⁎

Coefficients for fit:

α −1.371 (0.378)⁎⁎⁎ −2.053 (0.495)⁎⁎⁎ −2.023 (0.497)⁎⁎⁎ −1.920 (0.501)⁎⁎⁎
σu 0.976 (0.057)⁎⁎⁎ 1.073 (0.061)⁎⁎⁎ 1.072 (0.061)⁎⁎⁎ 1.116 (0.062)⁎⁎⁎

Coefficients for dit:

ξ −1.771 (0.525)⁎⁎⁎ −4.355 (0.741)⁎⁎⁎ −4.361 (0.742)⁎⁎⁎ −4.228 (0.736)⁎⁎⁎
τu 1.734 (0.098)⁎⁎⁎ 1.805 (0.103)⁎⁎⁎ 1.809 (0.104)⁎⁎⁎ 1.777 (0.102)⁎⁎⁎
ρ (= σ2u /1+σ2u) b0.001 0.488 0.535 0.535 0.555
Log-likelihood −1007.219 −1683.281 −1670.105 −1669.950 −1666.487
Number of firms 1240 1240 1240 1240 1240
Number of observations 2601b 3841 3841 3841 3841
⁎⁎⁎, ⁎⁎ and ⁎ represent that the coefficients are statistically significant at 0.01, 0.05 and 0.1 levels respectively. The year dummies are included in the regressions.
a
Random effects probit estimation assuming exogenous initial conditions has results almost identical to those of pooled probit estimation, as can be seen that the estimate of ρ is very
small and insignificant.
b
Observations for each sample firm are from period 2 to period Ti.

following characteristics: investing in China only, possessing moder-
ate R&D intensity, outsourcing its subsidiary's technology, having high
proportion of raw materials provided by the parent firm and high
proportion of products sold to the local market, entering into
resource-intensive industries, having the subsidiary vertically linked
to the parent firm, making diversified investment, investing in China
following foreign clients, and having the number of employees, total
sales, the years of experience, and the number of same-industry firms
at the sample means. The marginal effect of MODEt − 1 can be
calculated as the difference in the expected probability of adopting
WOSs for previous modes (modes in 2003) being WOSs and JVs,
i.e.,Φ(α + γ + β′xit) − Φ(α + β′xit).4 We then compare the marginal
effect of MODEt − 1 for PERF-PARt − 1 equal to 0 and 1 respectively.
The result shows that for parent firms having superior previous
performance (PERF-PARt − 1 = 0), the expected probability of choos-
ing WOSs as the previous mode being WOSs is 0.695 and the
corresponding probability as the previous mode being JVs is 0.307.
This implies that the marginal effect of MODEt − 1 on the probability of
choosing WOSs, or the effect of state dependence, is 0.388. Similar
calculation for parent firms having poor previous performance (PERF-
PARt − 1 = 1) suggests that the marginal effect of MODEt − 1 is 0.270.
Apparently, poor previous performance in the parent firm reduces the
degree of state dependence on WOSs by 0.118. In other words, for the
previous mode being WOSs, the probability of continuing adoption of
WOSs in the current year will decrease by around 12% if the parent
firm experiences poor previous performance. This confirms the
importance of mode performance in affecting a firm's future mode
decisions.
Finally, in order to provide a descriptive assessment of the fit of the
models, following Heckman (1981a), Table 7 presents the observed
frequency of mode choice patterns for period 2 to period Ti, together
with the frequencies predicted by the models shown in columns (1),
(2), (3) and (5) of Table 6. The bottom row of the table records the χ2
Pearson goodness-of-fit statistics for each model. The statistic is
calculated as

2
14 ns − n̂s
Goodness of fit = ∑s = 1
n̂s

4
That is, the marginal effect, Φ(α + γ + β′xit) − Φ(α + β′xit), is computed when the where ns and n̂s are respectively the observed and predicted
values of variables in x are as the following: LMNE = 0, RDLOW = 1, RDHIGH = 0,
TECHTW = 0, TECHOS = 1, MATTW = 1, LOCMKT = 1, RESOURCE = 1, VERTLINK = 1,
frequencies of the sth pattern, and there totally are 14 choice patterns.
DIVERSE = 1, RELCUST = 1, and lnEMP, lnSALE, lnEXPER, and RELFIRM equal to the A lower value of the χ2 statistic implies a better fit for a model. As can be
sample mean. Time dummies are also assigned to represent year 2004. seen, the goodness-of-fit statistics indicate a poor fit to the observed
 M.-Y. Chen, J.-Y. Chang / Economic Modelling 28 (2011) 439–450 449

Table 7 production relationship, diversification of foreign investment, and

Comparisons of predicted with observed entry mode choice frequencies. relational network connections. On the other hand, the probability of
Choice Observed Predicted frequency choosing WOSs decreases with an MNE's international experience and
patterna frequency the proportion of the subsidiary's products sold to the local market.
Random Heckman estimation of random
(number Accordingly, this study provides a more comprehensive picture of
effects effects probit
of firms)
probit foreign market entry mode choice of Taiwanese manufacturing firms
Column (1) Column (2) Column (3) Column (5) than has been available to date in Taiwan.
Table 6 Table 6 Table 6 Table 6

Firms with Ti = 4 Acknowledgments

0, 0, 0 113 52.59 82.80 86.51 88.15
1, 0, 0 15 34.14 23.28 22.44 21.75 This research is financially supported by the Taiwan National
0, 1, 0 8 26.70 21.93 20.57 20.14
0, 0, 1 20 29.36 22.66 21.87 21.50
Science Council (NSC-98-2410-H-032-034). We would like to thank
1, 1, 0 30 53.21 30.47 28.52 27.60 two anonymous referees and the Editor for their valuable comments
1, 0, 1 24 51.71 30.02 28.32 27.40 and insights that help us improve this paper.
0, 1, 1 22 48.89 29.18 27.53 27.08
1, 1, 1 336 271.40 327.67 332.24 334.37
Total 568 568 568 568 568 References
Firms with Ti = 3
Agarwal, S., Ramaswami, S., 1992. Choice of foreign market entry mode: impact of
0, 0 45 35.82 43.32 43.85 43.99
ownership, location, and internalization factors. Journal of International Business
0, 1 22 36.64 25.10 24.34 23.97
Studies 23, 1–27.
1, 0 17 32.70 22.31 20.85 20.46
Anderson, E., Gatignon, H., 1986. Modes of foreign entry: a transaction cost analysis and
1, 1 141 119.84 134.28 135.96 136.57 propositions. Journal of International Business Studies 17, 1–26.
Total 225 225 225 225 225 Arulampalam, W., 1999. A note on estimated effects in random effect probit models.
Firms with Ti = 2 Oxford Bulletin of Economics and Statistics 6, 597–602.
1 311 313.67 319.49 319.98 319.47 Arulampalam, W., Stewart, M.B., 2007. Simplified implementation of the Heckman
0 136 133.33 127.51 127.02 127.53 estimator of the dynamic probit model and a comparison with alternative
Total 447 447 447 447 447 estimators. The Institute for the Study of Labor (IZA) Discussion Paper No. 3039. .
Pearson 170.89 29.15 22.36 19.83 Asiedu, E., Esfahani, H.S., 2001. Ownership structure in foreign direct investment
goodness-of-fit projects. The Review of Economics 83, 647–662.
statistic Aulakh, P.S., Kotabe, M., 1997. Antecedents and performance implications of channel
(χ2 statistic) integration in foreign markets. Journal of International Business Studies 28, 145–175.
Bell, J., 1995. The internationalization of small computer software firms: a further
a
The choice pattern is for period 2 to period Ti. “1” corresponds to WOSs and “0” challenge to stage theories. European Journal of Marketing 29, 60–75.
corresponds to JVs. Thus “1, 1, 0” corresponds to a firm that chooses WOSs in periods 2 Bhaumik, S.K., Gelb, S., 2005. Determinants of entry mode choice of MNEs in emerging
and 3 and JVs in period 4. markets. Emerging Markets Finance and Trade 41, 5–24.
Brouthers, K.D., 2002. Institutional, cultural and transaction cost influences on entry mode
choice and performance. Journal of International Business Studies 33, 203–221.
Brouthers, K.D., Hennart, J.F., 2007. Boundaries of the firm: insight from international
frequency for the random effects probit assuming exogenous initial entry mode research. Journal of Management 33, 395–425.
conditions (or pooled probit model) (column (1) of Table 6), but a much Brouthers, K.D., Brouthers, L.E., Werner, S., 2002. Industrial sector, perceived
environmental uncertainty and entry mode strategy. Journal of Business Research
improved fit for the Heckman estimation of the random effects probit. 55, 495–507.
Particularly, the statistic indicates the best fit for the model taking Brouthers, K.D., Brouthers, L.E., Werner, S., 2003. Transaction cost-enhanced entry mode
account the impact of previous parent firm's performance on state choices and firm performance. Strategic Management Journal 24, 1239–1248.
Brouthers, K.D., Brouthers, L.E., Werner, S., 2008a. Real options, international entry
dependence of mode choice (column (5) of Table 6). mode choice and performance. Journal of Management Studies 45, 936–960.
Brouthers, K.D., Brouthers, L.E., Werner, S., 2008b. Resource-based advantages in an
6. Conclusions international context. Journal of Management 34, 189–217.
Chang, S.J., Rosenzweig, P.M., 2001. The choice of entry mode in sequential foreign
direct investment. Strategic Management Journal 22, 747–776.
Using the official survey data of Taiwanese FDI in China, this study Chen, H., Chen, T.J., 1998. Network linkages and location choice in foreign direct
examines the transaction cost determinants of entry mode choice investment. Journal of International Business Studies 29, 445–467.
Chen, S.S., Hennart, J.F., 2002. Japanese investors' choice of joint ventures versus
between WOSs and JVs. This study departs from previous researches
wholly-owned subsidiaries in the US: the role of market barriers and firm
in that it constructs a panel data set and focuses on an analysis of the capabilities. Journal of International Business Studies 33, 1–18.
dynamic binary choice model that controls for state dependence of Cleeve, E., 1997. The motives for joint ventures: a transaction costs analysis of Japanese
mode choice. The dynamic probit model is formulated according to MNEs in the UK. Scottish Journal of Political Economy 44, 31–43.
Coviello, N., Munro, H., 1997. Network relationships and the internationalization
Heckman's (1981b) approach and is estimated by the MSL method process of small software firms. International Business Review 6, 361–386.
under the assumption of endogenous initial conditions. Davidson, W.H., McFetridge, D.G., 1984. International technology transaction and the
Empirical results suggest the existence of state dependence theory of the firm. Journal of Industrial Economics 32, 253–264.
Davidson, W.H., McFetridge, D.G., 1985. Key characteristics in the choice of international
between current and previous entry modes. Having WOSs in the technology transfer mode. Journal of International Business Studies 16, 5–21.
previous period increases the probability that WOSs will be chosen in Delios, A., Beamish, P.W., 1999. Ownership strategy of Japanese firms: transactional,
the current period. However, there is also evidence to show that the institutional, and experience influences. Strategic Management Journal 20, 915–933.
Dikova, D., van Witteloostuijn, A., 2007. Foreign direct investment mode choice: entry
degree of state dependence decreases if the entry mode in the and establishment modes in transition economies. Journal of International Business
previous period did not produce good parent firm's performance. Studies 38, 1013–1033.
Echoing the argument of Brouthers and Hennart (2007), we find Dunning, J.H., 1980. Toward an eclectic theory of international production: some
empirical tests. Journal of International Business Studies 11, 9–31.
strong linkage between mode performance and future mode choice. Dunning, J.H., 1988. The eclectic paradigm of international production: a restatement
The hypothesized effects of transaction cost factors on entry mode and some possible extensions. Journal of International Business Studies 19, 1–31.
choice gain support even if Taiwanese MNEs inherently have greater Dunning, J.H., 1998. Location and the multinational enterprise: a neglected factor?
Journal of International Business Studies 29, 45–66.
propensity of using WOSs in investing China because of cultural
Dunning, J.H., 2000. The eclectic paradigm as an envelope for economic and business
similarity. Specifically, empirical findings indicate that the probability theories of MNE activity. International Business Review 9, 163–190.
of an MNE's choice of WOSs is positively associated with the MNE's total Gatignon, H., Anderson, E., 1988. The multinational corporation's degree of control over
number of employees, relative size of the subsidiary, and technological foreign subsidiaries: an empirical test of a transaction cost explanation. Journal of
Law, Economics and Organization 4, 305–336.
capability of the business. It also increases when there are proprietary Gomes-Casseres, B., 1989. Ownership structures of foreign subsidiaries. Journal of
technology transfer from parent to subsidiary, vertical parent–subsidiary Economic Behavior and Organizations 11, 1–25.
450 M.-Y. Chen, J.-Y. Chang / Economic Modelling 28 (2011) 439–450
Greene, W., 2007. Econometric analysis, 6th edition. Prentice Hall, New York.
Heckman, J.J., 1981. Heterogeneity and state dependence. In: Rosen, S. (Ed.), Studies in
Labor Markets. University of Chicago Press, Chicago.
Heckman, J.J., 1981. In: Manski, C.F., McFadden, D. (Eds.), The incidental parameters
problem and the problem of initial conditions in estimating a discrete time–discrete
data stochastic process. : Structural Analysis of Discrete Data with Econometric
Applications. Cambridge: MIT Press.
Hennart, J.F., 1982. A Theory of Multinational Enterprise. University of Michigan Press,
Ann Arbor.
Hennart, J.F., 1988. A transaction costs theory of joint ventures. Strategic Management
Journal 9, 361–374.
Hennart, J.F., 1989. Can the “New Forms of Investment” substitute for the “Old Forms?” A
transaction costs perspective. Journal of International Business Studies 20, 211–234.
Hennart, J.F., 1991. The transaction cost theory of joint ventures: an empirical study of
Japanese subsidiaries in the United States. Management Science 37, 483–497.
Hennart, J.F., 2000. In: Pitelis, C., Sugden, R. (Eds.), The transaction cost theory of the
multinational enterprise, 2nd ed. : The Nature of the Transnational Firm. Routledge,
London.
Hennart, J.F., Larimo, J., 1998. The impact of culture on the strategy of multinational
enterprise: does national origin affect ownership decisions? Journal of Interna-
tional Business Studies 29, 515–538.
Hennart, J.F., Park, Y., 1993. Greenfield vs. acquisition: the strategy of Japanese investors
in the United States. Management Science 39, 1054–1070.
Hennart, J.F., Zeng, M., 2005. Structural determinants of joint venture performance.
European Management Review 2, 105–115.
Hill, C.W.L., Hwang, P., Kim, W.C., 1990. An eclectic theory of the choice of international
entry mode. Strategic Management Journal 11, 117–128.
Hyslop, D.R., 1999. State dependence, serial correlation and heterogeneity in intertemporal
labor force participation of married women. Econometrica 67, 1255–1294.
Kim, W.C., Hwang, P., 1992. Global strategy and multinationals' entry mode choice.
Journal of International Business Studies 23, 29–54.
Kogut, B., Singh, H., 1988. The effect of national culture on the choice of entry mode.
Journal of International Business Studies 18, 411–432.
Lu, J.W., 2002. Intra- and inter-organizational imitative behavior: institutional influences on
Japanese firms' entry mode choice. Journal of International Business Studies 33, 19–37.
Maddala, G., 1987. Limited dependent variable models using panel data. Journal of
Human Resources 22, 307–338.
Madhok, A., 1998. The nature of multinational firm boundaries: transaction cost, firm
capabilities and foreign market entry mode. International Business Review 7, 259–290.
Makino, S., Neupert, K.E., 2000. National culture, transaction costs, and the choice
between joint venture and wholly owned subsidiary. Journal of International
Business Studies 31, 705–713.
Mutinelli, M., Piscitello, L., 1998. The entry mode choice of MNEs: an evolutionary
approach. Research Policy 27, 491–506.
Nelson, R.R., Winter, S.G., 1982. An Evolutionary Theory of Economic Change. Belknap
Press of Harvard University Press, Cambridge.
Padmanabhan, P., Cho, R.K., 1996. Ownership Strategy for a foreign affiliate: an empirical
investigation of Japanese firms. Management International Review 36, 45–65.
Richardson, J., 1993. Restructuring supplier relationships in U.S. manufacturing for
improved quality. Management International Review 33, 53–67.
Stewart, M.B., 2007. The interrelated dynamics of unemployment and low-wage
employment. Journal of Applied Econometrics 22, 511–531.
Wei, Y., Liu, B., Liu, X., 2005. Entry modes of foreign direct investment in China: a
multinomial logit approach. Journal of Business Research 58, 1495–1505.
Williamson, O.E., 1975. Markets and Hierarchies. Free Press, New York.
Williamson, O.E., 1985. The Economic Institutions of Capitalism. Free Press, New York.
Yiu, D., Makino, S., 2002. The choice between joint venture and wholly owned
subsidiary: an institutional perspective. Organization Science 13, 667–683.