30 Years of Research on Entry Mode and Performance Relationship: A Meta-Analytical Review

Abstract

Almost 30 years after Anderson and Gatignon systematically conceptualized various entry modes, the scholarly research on their performance effects has accumulated a large amount of empirical evidence. However, research has progressed in a somewhat fragmented manner where the cumulative impact of entry modes is difficult to discern and the entry-performance relationship still remains a conundrum in IB research. This study consolidates and meta-analyzes the empirical findings reported from 44 independent studies. While the results show an overall significant effect of entry modes on performance, more fine-tuned meta-analyses unravel that the effects of entry modes vary with the performance types and a number of plausible contingencies. Findings are discussed in terms of implications for future research and limitations.

1 Introduction

Entry mode refers to the forms of operations that firms employ to enter into foreign markets (Hill et al. 1990; Brouthers and Hennart 2007) and it has been one of the most important strategic decisions of multinational enterprises (MNEs). However, which entry mode leads to superior performance? For almost 30 years since Anderson and Gatignon (1986)’s landmark work that conceptually categorizes firms’ foreign market entry modes, this question has attracted a great attention of international business (IB) scholars. Yet, this fundamentally important inquiry still remains as a conundrum in international business research (Martin 2013). Despite the fact that a variety of explanations for why the choice of entry modes affect performance has been offered, the extant research evidence has progressed in a manner where its cumulative impact is difficult to discern. Some studies have suggested that entry modes with complete or high equity ownerships that confers great control by MNEs lead to good performance (e.g., Li et al. 2009). However, other researchers have argued exactly the opposite and suggest that entries through joint ownership allow foreign entrants to access local knowledge benefits MNEs (e.g., Lu and Beamish 2006) and ensure the mutual forbearance necessary for a stable relationship (e.g., Steensma and Lyles 2000).

Theoretically, a number of interesting questions have not been fully addressed. These include, but are not limited to: does entry mode truly matter in determining the performance? Does it matter if a subsidiary is wholly-owned or jointly owned? Are the incongruent findings contingent on other conditions? Thus, the proliferation of the E–P studies documenting the mixed findings presents a need for a more comprehensive and rigorous assessment of the existing literature in order to offer an overview of this stream of literature and provide insights into the power of different theoretical perspectives.

This article attempts to achieve three objectives. First, we comprehensively summarize the underlying reasoning with the supporting theories and weave together the accumulated empirical evidence to divulge patterns of entry and performance relationships in order to take stock of what we know. Second, entry mode researchers have taken three approaches to examine the relationship between entry mode and performance. One group of scholars looks at the direct relationship between entry modes and firm performance. Another group examines the choice of entry mode and firm survival. More recently, researchers have used a strategic fit approach and looked at the alignment of entry mode with mode experience. We use these three research approaches to determine which approach provides consistently significant results to help us understand better the relationship between entry mode choice and performance. Third, looking ahead, we explore the plausible contingency explanations for the next wave of research on entry mode. Specifically, we examine whether firms’ specific and contextual factors might influence the relationship between mode-choice and performance.

We believe that such a meta-analytic synthesis is timely and important for three reasons. First, given the fact that the documented evidence remains inconclusive with divergent and conflicting findings, there is a need to ascertain if entry modes matter in performance, and specifically, which entry mode(s) performs superiorly. Second, while a number of studies has excellently reviewed the studies on the antecedents to entry mode choice (Tihanyi et al. 2005; Zhao et al. 2004), up to date, there is no such systematic review on the consequences of entry modes, given a large number of empirical studies on this topic being undertaken. Thus, taking stock of the manner in which entry modes and performance are conceptually and empirically investigated is a critical next step in the progression of entry strategy research. Third, unlike a traditional literature review, which is intractable and does not consider and correct for sampling errors, meta-analytic approach enables researchers to examine a more broad set of factors and provide a powerful tool for quantitatively synthesizing empirical research over a variety of studies (Hunter and Schmidt 2004). Particularly in situations where there are large numbers of studies with mixed and inconsistent findings, meta-analytic approach can play an important role in providing comprehensive and rigorous assessments of evidence with data that are closer to definitive conclusions than those reported in any single study (Miller and Cardinal 1994).

This study is organized as follows: we begin with an overall review of the reasoning and the supporting theories, and the descriptive summary of the aggregated empirical findings reported in the existing studies. In the following section we provide a quantitative assessment of the existing findings. Then, we explore the entry-performance contingencies in the next section. The last section concludes with a discussion of key findings of our meta-analytic review and suggestions for future research.

2 E–P Relationship: An Overview

In this section, first, we use the classification of entry modes developed by Anderson and Gatignon (1986) to provide a summary review of extant studies’ logic and the related theoretical lenses in reasoning entry-performance relationships (Table 1). Second, we summarize the empirical findings of the study sample (Table 2) to highlight where entry-performance research stands.

Table 1 Reasoning and main underlying theories

Table 2 Descriptive summary of study sample and the effect of entry mode on performance

2.1 Review of Major Theories Underlying E–P Relationship

There has been a diverse set of theories applied in guiding the empirical examinations of the performance effects of entry modes. We organized the key reasoning and the underlying theories along the entry modes to provide an overarching view.

Table 1 summarizes the reasoning in the hypotheses development and the underlying theories that connect the various entry modes and performance. As shown in Table 1, typically the existing studies developed the E–P reasoning by drawing on one or combined theories of transaction cost economics (TCE), resource-based view, Dunning’s eclectic view, institutional theories, and organizational learning theory.

Table 1 reveals two noteworthy points. First, it appears that the directions of the predicted E–P relationships are fairly consistent among TCE-based reasoning. The studies that used TCE arguments in developing E–P relationships generally poised that better performance of MNEs’ foreign affiliates associates with the high control mode of entries including merger and acquisitions (M&As), greenfield investment, wholly-owned subsidiaries (WOS), and majority joint ventures (majority JVs). The primary reasoning for the positive effects of high-control modes rests on the cost minimization and control arguments rooted in TCE (Williamson 1991). Following this key tenet of TCE, studies contend that majority or wholly-owned equity positions provide a parent firm with a greater degree of control over the systems, methods, and decisions of its subsidiary so that MNEs are less reluctant in transferring their firm-specific resources and intangible resources to gain better performance (Gaur and Lu 2007; Lee and Beamish 1995; Chang et al. 2013). WOS can also benefit the entrants since the costs of preventing relational hazards in JVs can exceed the benefits of reducing environmental uncertainty through local partner (Dikova 2009). Additionally, high ownership-based entries give foreign entrants high degrees of control to avoid various frictions at operational level, thus achieving better performance (Boyd et al. 2012; Quer and Claver 2008; Deng 2001; Chen 1999; Merchant 2002; Murray et al. 2012). This TCE-based reasoning is also echoed in the studies using Dunning’s eclectic framework, arguing that the combinations of ownership, location and internalization take into account of a broad set of costs/risk in providing the optimal performance outcome (e.g., Woodcock et al. 1994; Brouthers et al. 1999).

Second, while studies drawing on agency theory and ownership-location-internalization (OLI) paradigm with consistent predictions, there are many variations in predicting E–P relationships employing more diverse theoretical lenses in examining joint ventures with distinctions of majority and minority equity (JVs), equal JVs and the levels of foreign ownership as entry strategies. Some even proposed contradicting arguments. For instance, studies applying the RBV view argued that WOS or higher levels of foreign ownership motivate parent firms to commit more resources such as advanced technology to support their subsidiaries, thus leading to higher IJV productivity and performance (Anand and Delios 1997; Luo 2003; Luo and Zhao 2004; Li et al. 2009). Also drawing on RBV views, however, other studies argued that JVs can outperform WOS since JVs enjoy relatively easy access to resources such as cheaper labor, materials, and the government policy (Luo 1997). JVs with local partner through whom foreign entrants can pool complementary resources and gain supports from local government in receiving faster approval and tax incentives can contribute to performance (Pangarkar and Lim 2003). Conversely, the opposite argument is made that with higher parent ownership, foreign parent firms are more motivated to commit resources. This creates more reliance of foreign subsidiaries on parent firms’ specific assets and makes subsidiaries less motivated in considering the host market particulars, hence resulting negative effects on their performance (Tang and Rowe 2012).

In the same vein, a number of studies using either institutional theory or organizational learning theory also made contradicting predictions. For instance, studies found that the high level of local involvement allowed JVs to better meet the local demands of the local institutional environment (Magnusson et al. 2009) and that in the less developed institutional context, having a large local partner in a JV arrangement helped JVs to establish legitimacy faster, hence positively impacted on JVs’ growth and survival (Lu and Xu 2006). On the opposite, other institution-based studies found that majority ownership increased MNEs’ bargaining power in negotiating with the host government and reduce the liability of foreignness to positively influence subsidiary performance (Zhao and Luo 2002). Studies drawing on organizational learning theory shifted from the control- and efficiency-centered explanations to the role of learning in establishing the E–P linkages. They argued that JVs can perform better than WOS because local partners represent the primary sources of local knowledge for successful operations (Lane et al. 2001; Lu and Beamish 2006).

Different from the above, a few studies of equal JVs as entry modes used TCE to predict a positive performance effect arguing that while shared-control structure may be more difficult to manage in terms of coordination than JVs with dominant control structure, equal JVs also ensure the mutual forbearance necessary for stable relationship with limited conflict (Steensma and Lyles 2000). Moreover, a few studies even poised nonlinear E–P relationships. For example, Ramaswamy et al. (1998) found that a U-shape relationship with the increased performance associated with majority equity ownership positions and the declining performance linked to shared equity ventures. On the contrary, Li et al. (2009) proposed an inverted U-shaped relationship arguing that the increases of foreign ownership to a very high level result in a smaller incremental benefits than a balanced ownership structure.

In sum, the existing studies have employed diverse theoretical lenses to derive the relationships between various entry modes and performance. While it appears that TCE has provided consistent predictions on the positive effects of high-control entry modes on performance, studies using other theories tend to offer contradictory reasoning. This could be one of the potential reasons that result in the mixed empirical findings as it is elaborated in the following section.

2.2 Descriptive Summary of Empirical Studies

Having reviewed the reasoning and the foundational theories behind the E–P relationships, we now summarize and describe the empirical studies to provide an overall view of the status of E–P research. In constructing the sample for the review and the subsequent quantitative analysis, first, we use Anderson and Gatignon (1986) classification scheme (A&G) that organized entry modes along a continuum of increasing control, resource commitment and the associated risks to classify entry modes.

Second, we follow the practices of recent meta-analysis articles that are closely related to our topics (Grinstein 2008; Bausch and Krist 2007; Tihanyi et al. 2005; Zhao et al. 2004; Kirca et al. 2011) to use only top-rated academic journals as sources to gather studies with a specific focus on E–P relationships. These journals include the leading IB journals identified by DuBois and Reeb (2000) and the top-rated journals in management and marketing given the increasing influences of these two disciplines (Chandy and Williams 1994). Namely, the source journals include: Journal of International Business Studies, Journal of World Business, Management International Review, and International Business Review, Administrative Science Quarterly, Academy of Management Journal, Strategic Management Journal, Organization Science, Management Science, Journal of Management Studies, Journal of Management, Journal of Business Venturing, Journal of Marketing, Journal of Marketing Research, Journal of International Marketing, and International Marketing Review.

Additionally, we use four parameters to establish the boundary to identify clearly the current research on E–P relationship. First, the entry-performance relationships must occur in the context of international business. Second, studies examining entry-performance relationships at firm or subsidiary unit are included in this paper. Third, we use the time frame of 1986–2015. This initial year is chosen because Anderson and Gatignon (1986)’s work marks specifically the beginning of systematic research on entry mode. Fourth, to ensure the representativeness, we follow a two-step searching procedure to identify studies to be included in the meta-analysis using the following two screening criteria: (a) any studies included must hypothesize on E–P relationship and be quantitatively empirical in nature, and (b) the study must report pairwise correlation coefficients involving one or more related types of entry mode and performance. These procedures yielded 45 studies with reported 100 correlations as the sample for our meta-analyses.

Table 2 summarizes the 45 studies and displays the reported findings on E–P relationships, which forms our study sample. We group these studies into three research approaches: Direct E–P analysis, Survival Analysis and Fit Analysis (column 1) and use them to base our subsequent analyses. As indicated, more than half of the studies (28) fall into Direct E–P approach. There are 13 studies focused on Survival analysis and only 4 studies approached it from Fit analysis. Column 6 in Table 2 also shows the entry mode examined in each study. For various types of entry modes examined, three patterns emerged from synthesizing these existing studies.

First, as shown in of Table 2 (column 6), the extant studies almost exclusively focused on the ownership-based entry modes with the only exception to three studies (Zhang et al. 2007; Blesa and Ripolles 2008; Ganotakis and Love 2012) that explicitly examined export and franchising as entry modes in addition to other entries.

Second, out of the eight ownership-based entry modes (JV without distinction between majority and minority ownership; equal JV; M&A; WOS; ownership level (percentage); majority JV; minority JV; and Greenfield), the four mostly examined entry modes are the ownership level in terms of equity percentage (44.4%), followed by JV without distinction (20%), M&A (15.6%), WOS (13.3%), and equal JV (13.3%), which ostensibly reflects the overall research focus around these entry types. As a result, what we know about the E–P relationship is likely to be overly influenced by the ownership-centered research results.

Third, at the aggregated study level, the majority of the existing studies, 38 out of 45 studies, focused on examining and comparing the performance effect of only one type of entry mode with other reference groups.¹ Moreover, among those these 38 studies, 31 of them focused on ownership-based entries (ownership percentage = 19, JV = 4, WOS = 5, and M&A = 2). Other types of non-ownership-based entry modes, such as export, franchising, and licensing, are thus underrepresented by the existing research. Thus, without comparing performance effects between and among specific entry modes which is a fundamentally important inquiry in this research area (Brouthers 2013; Martin 2013), the current research practice apparently limits our understanding of superior modes of entry and calls this trajectory into question (Martin 2013; Brouthers 2013; Shaver 2013).

With regard to performance (column 7), the existing studies used a variety of performance measures ranging from typical financial indicators such as ROA to market performances, and to survival rates. Though there is hardly a consensus on the conceptualization and measurements of performance of foreign subsidiaries, we concur with the previous performance research (Venkatraman and Ramanujam 1986; Ariño 2003) that multiple indicators of performance help enhance the understanding of antecedents of performance construct. Thus, we follow recent suggestions by Hult et al. (2008) who conducted a comprehensive assessment of the performance measures in IB research to group the performances into four categories: financial (FP), operating (OP), overall effectiveness (OE) and survival.² It appears that the mostly used performance measures were financial performance (19 studies), followed by overall effectiveness (12 studies) and survival (12 studies). Lastly, operational performance was used in eight studies. In addition, four studies used more than two different measures of firm performance, with two studies using financial and operational performance, one study using financial performance and overall effectiveness, and one study using all three performance measures. What is more salient is that the information exhibited in column 7 of Table 2 shows that, overall, most studies documented insignificant effects of entry mode choice. For the 18 effects that documented significant effects of entry mode choice, there is a nearly even split between positive (N = 10) and negative effects (N = 8).

Table 2 also presents information for each study about the author(s), journal outlets, the year a study was published, the sample size used in each study, the data source (primary vs secondary), the data type (panel, cross-section or longitudinal), home country (country of origin of the study sample), host country (location of the study sample), and the industry type. These empirical studies spanned over 20 years from 1994 to 2015. There are only 4 papers published in the last 5 years (2011–2015) and the most studies appeared in the period from 1994 to 2010, indicating the declined interests of IB scholars in this area. The sample size used in these studies varied substantially ranging from 71 to 241,069 observations. For the sources and the nature of data used in these studies, the primary data (19 studies, 42.2%) is comparably less used than secondary data (57.8%). With only two studies using longitudinal data for their empirical tests, most of the studies relied either on cross-sectional data (25 studies = 55.6%) or panel data (18 studies = 40%). For the country of origin of samples used by the existing studies, only 7 studies used the multiple country-of-origin in both home and host country settings and 25 studies used the multiple country-of-origin samples either from home and host countries. The study samples used in the existing research is dominated by Japanese firms with 28.9% of them (13 studies) used sample firms originating from Japan. However, for the host countries wherein the sample firms operated, it is fairly diverse. Finally, the distribution of industry of the samples used by the existing studies shows that 14 studies sampled firms from manufacturing sector, 21 from multiple industries and only 2 specifically from service sector. Obviously, aside from samples from multiple industries, the study samples were biased to manufacturing sectors and the service sector was under-represented.

In sum, our review of the theoretical foundations underlying the entry-performance research and the extant empirical studies brings out several interesting characteristics presented below. First, while diverse theoretical lenses have been employed by the existing studies to examine E–P linkages, TCE stands out as the theory that offers consistent reasoning and predictions on the positive effects of high-control entry modes on performance and other theory-based reasoning tend to offer contradicting views. Second, despite the large number of empirical studies, the documented empirical evidence remains inconclusive and is at best characterized by divergent and conflicting findings subject to further clarification. It occurs that the studies analyzing the direct E–P relationship tend to not find significant effects of entry modes; rather most of the significant performance effects are found in the research approaches in Survival analyses and Fit analyses. Third, most of the studies focused on the effects of ownership-based entry modes. This unbalanced focus without specifying and incorporating explicitly other non-ownership-based entry modes may limit our knowledge on whether the ownership-based entry modes outperform or are outperformed by other types of entries. Finally, in their research design, most studies did not address the endogeneity issue that has been considered a major issue in the research design in entry-performance studies (Brouthers 2013; Shaver 1998). There are only 4 out of 45 studies that addressed this issue.

3 Quantitative Assessment of E–P Relationship

Quantitatively evaluating the performance effects of entry modes is our second objective. In meta-analytic study, two most commonly used effect size estimates are r and d (Hunter and Schmidt 1990). We use r in this study because it represents a standardized measure of association between topic variables and is easily interpretable and scale free (the calculation formula is given in the following section). The variety of entry modes that have been documented in the existing studies provides us the opportunity to statistically test and compare the performance effects of different entry modes to discern whether entry mode(s) are significant predictors of performance and which entry mode leads to better performance.

In the quantitative assessment, first, a random subset of five studies from our study sample as given in Table 2 was initially read for the purposes of developing a standard method of coding effect sizes and study characteristics following the procedures recommended by Lipsey and Wilson (2001). Based on the methods described by Hunter and Schmidt (1990), all studies were read and effect sizes were coded independently by two trained researchers in this research area. Cohen’s kappa inter-rater reliability for the coding was 0.85. Consensus discussion took place to resolve any disagreements in coding. Correlations were harvested mostly among measures of constructs reflecting entry modes and performance.

Second, to estimate average correlations among variables weighted by sample size, we used Hunter and Schmidt (1990) procedures: ${\text{E}}({\text{r}}) = {\bar{\text{r}}} = \frac{{\mathop \sum \nolimits_{{{\text{i}} = 1}}^{\text{k}} {\text{n}}_{\text{i}} {\text{r}}_{\text{i}} }}{{\mathop \sum \nolimits_{{{\text{i}} = 1}}^{\text{k}} {\text{n}}_{\text{i}} }}$, where k is the number of studies, n is the sample size of studies, and r is the correlation coefficients. The 95% confidence intervals and Z values were reported to double confirm that the estimated effect size was within the range of estimation. After calculating a weighted average of all studies, we calculated the weighted average effect sizes for each category of entry mode as well as for each performance type. In calculating these statistics, we used the meta-analysis software program developed by Biostat™ that provides comprehensive meta-analytical statistics including estimates of effect size and confidence intervals. We also followed Hunter and Schmidt (1990)’s suggestion to correct sample errors in the reported “bare-bone” correlations. We first calculated the sampling error variance in the uncorrected correlation: ${\text{Var}}(e_{0} ) = [1 - \bar{r}_{0}^{2} ]^{2} /(N_{i} - 1)$, where $\bar{r}_{0}$ is the mean uncorrected correlation across studies and N _i is the sample size for the study in question. We then calculated the attenuation factor for each artifact ${\text{a}} = 1/[(U_{X}^{2} - 1)r_{0}^{2} + 1]$, where U _X is the ratio of the observed SD (standard deviation) in the unrestricted sample to the observed SD in the restricted sample. In this case, we collected the original SD in the studies as the observed SD in the unrestricted sample and the square root of the sampling error variance in the uncorrected correlation Var (e ₀) as the observed SD in the restricted sample. Finally, we computed the corrected study correlation r _Ci  = r _0i /a _i , where r _0i is the uncorrected correlation in each study and a _i is the attenuation factor for each study. This corrected r is used in the following analyses.

3.1 The Main Effects

Table 3 reports the main effect sizes (Panel 1) and the mean effect sizes of each entry mode based on A&G’s and B&H’s entry modes classifications (Panels 2 and 3). As Panel 1 shows, the overall combined mean effects of entry modes on firm performance across 100 observations with an accumulated sample size of 183,060 are significant (r _e  = 0.09; z = 2.76; p < 0.01; confidence interval 0.07–0.14). The 95% confidence interval does not include zero, indicating that entry modes do have an aggregated overall direct significant impacts on firm performance. With regard to each specific entry mode, high-control entry modes have significant positive effects on performance (r _e  = 0.11; z = 2.23; p < 0.05; confidence interval 0.08–0.16), while all medium- and low- control entry modes have no significant performance effects. However, within high-control mode, only the wholly-owned entry mode positively and significantly influences performance (r _e  = 0.11; z = 2.52; p < 0.05; confidence interval 0.02–0.20).

Table 3 Aggregated analyses of E–P relationships—overall effects

Following the work of Anderson and Gatignon (1986) (A&G) and Brouthers and Hennart (2007) (B&H), entry modes have been also classified into different types based on the degree of control and risks. Both classifications relied mainly on the accepted theories (respectively TCE for A&G and internalization for B&H). So far, they represent the comprehensive classifications of various entry modes and are influential in the realm of entry mode research. To understand further whether the E–P relationships differ due to the grouping of entry modes, we ran a one-way between subjects ANOVA test using respectively A&G and B&H classifications. The results reported in Panel 2 and 3 are very consistent. There is a significant effect of entry mode on overall performance at the p < 0.05 level when various entry modes are classified into high-, medium- and low-control types (For A&G: F (3, 174,345) = 124.71, p < 0.001; For B&H: F (3, 174,345) = 168.23, p < 0.001). To compare the specific performance effect across these two classifications as suggested by Shaver (1998), Brouthers (2013) and Martin (2013), we also conducted post hoc tests. Post hoc comparisons using the Tukey HSD test indicate that in both classifications the mean effect score for high control modes is most significant bearing the largest effect size (For A&G: M = 0.011, SD = 0.022; For B&H: M = 0.011, SD = 0.024). In both classifications, low control modes are equally ranked having the second largest effects (For A&G: M = 0.008, SD = 0.071; For B&H: M = 0.008, SD = 0.071), the medium-control modes under A&G and joint venture under B&H have smallest effects. These Tukey test results suggest that at the aggregated level, entry mode is a significant predictor of performance but the performance effects of entry modes vary depending on the degrees of control and risks.

3.2 Entry and Performance by Different Research Approaches

As shown in Table 2, the existing entry and performance studies undertake three approaches; the entry effects on performance (E–P Analysis), the entry effects on survival (Survival Analysis), and entry-performance fit-model (Fit-Model). To further discern which research approach best explains the performance impacts of entry modes, we grouped the studies into these three groups and compared mean effect sizes (Table 4).

Table 4 Entry mode and performance relationship by three research approaches

Table 4 reports the mean effect sizes among three research approaches: simple direct E–P analysis, survival analysis, and fit analysis. Panel 1 column 1 illustrates the effect size of each entry modes reported in the studies that examined direct entry and performance relationships. Judging by z values, the results indicate that there is no overall significant effects (K = 54, r_e (effect size) = 0.08, z = 1.80) and that none of the entry mode is significantly associated with performance. For the studies focusing on the survival analysis in column 2, there is no significant aggregated overall effect (K = 29, r_e (effect size) = 0.04, z = 1.31). WOS is the only entry mode that bears significant effects on performance (r_e (effect size) = 0.10, z = 2.12, p < 0.05). For studies that used fit analysis, the results in column 3 show positive and significant effects on performance (K = 11, r_e (effect size) = 0.64, z = 2.91, p < 0.01). Taken together, these results clearly suggest that the fit approach that takes into account of the antecedents to entry modes such as institutional and cultural distances and the endogeneity issue is more likely to unravel significant relationships between entry modes and performance as compared to the other research approaches.

4 Exploration of E–P Contingencies

While it is useful to summarize what we already know so as to economize on rediscovery, it is equally important and meaningful to spark new conversations by examining plausible moderating effects that condition E–P relationship. This will allow us to uncover potentially the new explanatory variables and ultimately develop more nuanced theoretical understandings of the E–P relationship. Given the observed heterogeneity in the reported findings and the results presented in the previous section, the idiosyncratic differences in the study contexts that received various prior research attentions (performance types, entry timing, parent size, subsidiary age, subsidiary size, cultural distance, developed home vs developed host country) and data-type factors (data source: primary vs secondary data and data types: cross-sectional, panel and longitudinal data) across the existing studies may play a role in understanding the E–P relationship. We concentrate on these moderators because they are either most studied or included in the research designs, and consequently are the moderators in greatest need of refined understanding with a sufficient number of studies amenable to moderator analyses. Therefore, this section explores these plausible E–P contingencies. Specifically, the contextual contingencies are those variables that have been studied in various ways in prior IB research including: timing of entry (Pan et al. 1999), subsidiary age often treated as a measure of local market experience (Delios and Beamish 2001), subsidiary size, parent size (Steensma et al. 2005), developed home/host country, and cultural distance (see Tihanyi et al. 2005 for a meta review). For data-type contingencies, the data source as an important phenomenon traditionally labeled as the file drawer effect (Rosenthal 1979) has been emphasized in a recent review and evaluation of meta-analytic practices (Geyskens et al. 2009). The data type reflects temporal dimension of the subject of research.

In exploring the moderating effects, we follow Hunter and Schmidt (1990) to conduct Chi square-based test of heterogeneity (Q) for categorical variables and weight-least square regression of continuous variables. The Hunter–Schmidt Q statistics is calculated as a test of homogeneity to assess whether sampling error is the primary explanation for variability in effect sizes. The Q test is computed by summing the squared deviations of each study’s effect estimate from the overall effect estimate, weighting the contribution of each study by its inverse variance. Rejection of the null indicates that sampling error does not account for all of the variability between studies, providing evidence for the existence of moderators. The Q test results show significant heterogeneity across studies ($\chi^{2}$ (87.5) = 366.81, df = 99, p < 0.001), indicating existence of moderators.

The results of contingency analyses are reported in Panel 1 and 2 of Table 5. The results in Panel 1 suggest that 3 out of 6 categorical moderators: industry type (manufacturing, service, and multiple), the data source (secondary vs primary), and time span of data (panel, cross-section, and longitudinal) significant influenced E–P relationships. E–P relationship is significantly stronger for manufacturing industries (r_e = 0.04, K = 14, Q = 21.38, p < 0.001) compared to service industry (r_e = −0.13). The data types significantly and positively moderate the E–P relationship (Q = 9.14, p < 0.001). It occurs that the relationship is stronger with primary data used in the E–P studies (r_e = 0.21, K = 45, Q = 9.14, p < 0.001) compared to studies using secondary data (r_e = 0.05, K = 55). Time span of data also turns out to be a significant moderator (Q = 18.01, p < 0.001). Specifically, the results show that the effects of entry modes on performance are larger for studies using cross-sectional data (r_e = 0.29, K = 64) than those using longitudinal (r_e = 0.21, K = 3) and panel data (r_e = 0.03, K = 33). However, the result for longitudinal data should be interpreted with caution as it is based on a fairly small number of effect sizes (K = 3). For continuous variables as possible moderators, the results in Panel 2 show that only subsidiary age (the number of years a subsidiary operating in a given foreign market) significantly and positively moderate the E–P relationship (K = 39, b = 1.561, p < 0.001).

Table 5 Exploratory analyses of E–P contingencies

5 Discussion and Conclusion

Given entry modes as a key enduring theme for IB research (Griffith et al. 2008) and its strategic importance with long-term consequences (Morschett et al. 2010), there is a need to comprehensively and systematically evaluate “where we are within this field of inquiry, and what is our trajectory” (Shaver 2013, p. 23) to inform us the current status in this specific research area and to highlight the possible areas for future research in order to move ahead. For this purpose, this current review systematically synthesizes and evaluates the empirical findings to take stock of the status of this research area, and explores further the plausible entry-performance contingencies.

5.1 Summary of Findings

Indeed, our review shows that, for almost 30 years’ research in addressing the central question of whether the choice of entry mode affects performance, the breadth and the depth of insights that have accumulated over time are impressive. The specific findings from this review are summarized below:

First, among the other theories that are used by our sample studies to support the E–P relationship, TCE appears to be able to provide the most consistent predictions. Those studies generally suggest that better performance of MNEs’ foreign affiliates associates with the high control mode of entries including merger and acquisitions, greenfield investment, wholly-owned subsidiaries, and majority joint ventures. Studies using other theories, such as RBV, OLI, institutional theory, organizational learning, and agency theory, however, tend to offer contradictory findings.

Second, in terms of performance measurement, financial performance was most commonly used compared to operational performance, overall effectiveness, and survival. A significant number of studies focused on ownership-based entry modes rather than export, franchising, or licensing. And within those studies of ownership-based entry modes, very few considered using multiple measurements of entry mode. Although a majority of studies looked into direct E–P relationship, they tend to not find significant effects, which were found more often in the Survival analyses and Fit analyses.

Third, the most studies in entry mode and performance appeared before 2011, indicating the declined interests of IB scholars in this area. Compared to cross-sectional and panel data, longitudinal data is much less used in entry mode research.

Fourth, though our meta-analytic review shows that at the aggregated level, entry modes do have significant effects on performance, the analyses based on the breakdown of each specific entry mode further reveal that wholly-owned subsidiary is the only one that significantly and positively influences firms’ performance. No significant relationships are found between the rest of entry modes and performance.

Fifth, interestingly our analyses of E–P relationships using three different research approaches show that the studies adopting Fit approach tend to find the significant and positive effects of the entry modes on performance. The survival approach shows that the significant performance effect is only associated with WOS; while simple E–P approach does not explain any significant relationships.

Sixth, the explorative contingency analyses based on meta-analytic approach reveals that the industry type, time span of data used, data sources, and subsidiary age are plausible moderators that explain the variations in entry-performance relationships. However, other context-specific factors do not affect the E–P relationships.

5.2 Implications for Future Research

Overall, our meta-analytic results show no solid evidence to support the direct and significant performance effects of most entry modes, suggesting the need to uncover under-explored areas. Two broad avenues await scholarly pursuit further. First, as identified by Hennart and Slangen (2015) for more entry mode studies, three identified research areas are relevant and can be extended to examination of the performance consequences of entry modes. These include the evolution of actual suboptimal choices over time vis-à-vis theoretically optimal choices, the replication of prior mode choice, and the processes of entry decision-making. We believe that future research in these directions represent the “bright future for entry mode research” (Hennart and Slangen 2015, p. 120) and can “meaningfully alter the trajectory of existing scholarship” (Shaver 2013, p. 24).

Second, these findings also suggest that entry modes as ownership structure neither necessarily nor naturally lead directly to the attainment of the intended performance goals by MNEs, rather the performance effects are contingent on or pass through other factors. Our initial explorative analysis of plausible moderators supports this conjecture and confirms that the documented inconsistent findings are attributed to the omission of important moderating variables that future research should theorize and test the relationships between entry modes and performance. This suggests further that future research should focus on what happens beyond the point of entry. The point at which entry modes are chosen and established should serve as a baseline for scholars to incorporate theory-based factors to explain the indirect relationships between the already established entry modes and performance or the effects of subsequent mode switch. For instance, the significant and positive moderating effect of subsidiary age indicates that the longer the operating lengths of subsidiaries are in a given market, the higher probability for high-control entry modes to generate significant impacts on performance. Does this imply that the cumulative local knowledge through many years of operation may have complementary effects for the choice of entry mode? If this is the case, as subsidiaries become more knowledgeable of local markets, the initially-chosen entry modes are more important. So far this and other contextual factors have been mainly kept in the background as controls by the existing studies focusing on E–P relationships.

Future research should also focus on comparative assessment of the performance effects of various entry modes. So far, what we know about E–P relationship in general is likely to be overly influenced by the characteristics of the most studied modes such as ownership level and WOS, and we cannot make firm conclusion on contract or low control modes since they are only covered by few existing studies. The reason that little research has been done in the contract-based entries may be the result of lacking of conceptual understanding of the boundaries and linkages between firm hierarchy and market.

The extant studies have significant risks of convergence around insights based on the most researched entry modes such as ownership level and WOS, as these insights may limit our ability to develop robust theory, and the most researched entry modes may not necessarily be the most empirically common nor the most managerially relevant. Therefore, future research endeavor should include multiple entry modes in a study particularly the contract-based modes to compare the performance impacts across them since without focusing on the comparative nature of entry modes, examining entry-performance relationships suffers endogenous problems that can be potentially misleading (Shaver 2013).

Another implication involves complementarity of theories used in examining E–P relationships. While we all benefit from multiple theories that have helped us comprehend this research area, the uneven focus on the control and risk aspects by transaction cost theory and on the resource capability by RBV theory, to certain degree, makes it difficult to understand the overall nature of the entry-performance relationships. A noticeable start of integrating different theories to explain E–P relationship is Brouthers and Xu (2002)’s study that won 2013’ Journal of International Business Studies decade award. However, our review reveals that few studies draw on multiple particularly complementary theories studying this phenomenon. Since the nature of entry-performance relationships are multifaceted and are inherently both economic and social, it can be fruitful to use theoretical lenses beyond the typical theories reviewed in this study to advance our knowledge. At least in a limited sense that the choice of entry mode is between hierarchy and market, both economic and social theories are needed to understand firm boundaries (Santos and Eisenhardt 2005) and need more diverse theoretical approaches (Combs et al. 2004).

Lastly, future research should consider the temporal dimension of E–P relationships. Our review reveals a troubling over-reliance on cross-sectional designs to demonstrate the impact of entry modes on performance. Although time is an important consideration in nearly every aspect of IB research, it is absolutely essential to entry-performance. The static approach relying heavily on cross-sectional data in the current literature can only offer a partial picture of a dynamic E–P relationship. This is because certain tangible performance outcomes related to entry modes may accrue quickly, while others may be more beneficial over a longer period of time. Initial entry modes adopted by MNEs to achieve certain performance outcomes may also change over time and their performance effects may decay, reverse, or accelerate as implied by our findings on varying moderating effects of panel and longitudinal data. These and other temporal elements in E–P relationship cannot be understood within the context of cross-sectional data. In order to make inferences and fully comprehend E–P phenomena, future E–P research needs to carefully incorporate various time constructs such as lags and longitudes to provide the potential to enrich our knowledge in this research area.

5.3 Limitations

Despite some of the benefits of the scope of this review and the power of meta-analytics, our work also has several limitations that highlight caveats to our meta-review approach should be noted. First, though we follow the previous practices to select empirical studies from top-rated IB journals and management and marketing journals, it does not mean that interesting, important, and highly relevant research in E–P research does not occur outside of the journals. Our findings may not representatively capture the entire field of E–P research. In this end, our goal is to be broad enough while focusing on the most highly regarded and impactful research outlets. Interpretation of results should be done in this light. Second, our meta-analytical sample based on the reported correlations is skewed towards ownership-based entry modes with a few observed correlations between contract-based entry modes such as franchising and licensing and performance. Thus, our findings should be interpreted with caution since the significant effects could be biased towards the dominated ownership-based entry modes. Third, in order to avoid the attrition of correlation observations, we only include the plausible contingencies that are identifiable in our E–P correlation-based sample studies. This results in the exclusion of some other important factors that should be considered in future research. Fourth, our capture of industries may not be precise given a large number of studies in our sample did not report the industry at all or briefly described the sample was from manufacturing industry. The same thing applies to country-of-origin information. Some studies did not clarify the specific home countries or host countries, instead, they alleged multiple countries were entered by firms from the same home country or firms from multiple countries entered the same host country. We believe that research on a single industry entering into a certain host country from another home country is needed to provide more accurate evidence of entry mode and firm performance relationship.