KOREASCHOLAR

Introduction
An important decision that a manufacturer has to make in distributing a product to customers is the degree of forward channel integration (Aulakh & Kotabe, 1997; Coughlan et al., 2001; John & Weitz, 1988). Transaction cost economics (TCE) developed by Williamson (1975, 1985, 1986, 1999) has been one of the leading theoretical frameworks used to explain the channel integration decision (Frazier, 1999; Watson et al., 2015). TCE is generally a theory for explaining the choice of an efficient governance structure in transactions and includes asset specificity, uncertainty, and frequency as its explanatory variables. According to Williamson (1985, 1986, 1999), much of the explanatory power of TCE is driven by asset specificity. TCE-based channel integration studies argue that as asset specificity increases, firms are expected to increase the degree of channel integration. This study proposes to extend existing research in four important ways. First, existing studies have not examined individual dimensions of asset specificity. This study examines two important dimensions discussed by TCE: human asset specificity and physical asset specificity. Second, existing studies have tended to measure asset specificity in a particular way (i.e., with a particular set of questionnaire items). This study examines the robustness of the estimated asset specificity-integration relationship to alternative measures of asset specificity. Third, existing studies have focused on firms in one country such as the United States, Canada, or Germany. This study empirically examines the roles and relative importance of human and physical asset specificity in channel integration in two countries with different cultures, the United States and Japan. Fourth, existing studies have not investigated the possibility of endogeneity between asset specificity and channel integration. This study tests whether asset specificity is endogenous in explaining channel integration through an instrumental variables and two-stage least squares (IV-2SLS) approach.
Literature Review
In the context of distribution channels, asset specificity refers to the extent to which durable, transaction-specific investments in human and/or physical assets are needed to distribute the product in question (John & Weitz, 1988; Klein et al., 1990; Shervani et al., 2007). Examples of such investments include (1) the time and effort employed to acquire the firm-specific, product-specific, and customer-specific knowledge needed for distribution activities, and (2) specialized physical equipment and facilities (e.g., warehouses, deliver vehicles, refrigeration equipment, demonstration facilities, and repair and service centers) (Anderson, 1985; Bello & Lohtia, 1995; Brettel et al., 2011a, 2011b; John & Weitz, 1988; Shervani et al., 2007; Williamson, 1985, 1986). According to TCE, when the assets needed to distribute a product are non-specific, the use of independent channels is a priori more efficient than the use of integrated channels based on the benefits of distribution specialists and competition in the market place (Anderson, 1985). Conversely, a high level of specific assets, whether human or physical, has important implications for the degree of channel integration. The primary consequence is to reduce a large number of relationships between a manufacturer and independent channel members to a small number of relationships, which may expose the transaction in question to opportunistic behavior. Because the unique productive value created by a high level of specific assets makes it costly to switch to a new relationship, the use of independent channels will not be effective as a safeguard against opportunism (John & Weitz, 1988; Shervani et al., 2007). Channel integration provides a safeguard against opportunism by permitting (1) the better monitoring and surveillance of integrated channels relative to independent channels, and (2) the reduction of profits from opportunistic behavior since employees in integrated channels do not ordinarily have claims to profit streams (John & Weitz, 1988). As a result, as asset specificity increases, manufacturers are expected to increase the degree of channel integration to exercise greater control over the channels (John & Weitz, 1988; Shervani et al., 2007). This leads to the following basic TCE hypothesis concerning asset specificity and channel integration: TCE hypothesis. Asset specificity will be positively related to the degree of channel integration. Existing studies of channel integration tend to provide support or partial support for the hypothesized positive relationship between asset specificity and channel integration. One limitation of key studies is that they have not fully explored the dimensions of asset specificity because they treat asset specificity as unidimensional or examine only one dimension of asset specificity. Specifically, Anderson and Schmittlein (1984), Anderson (1985), Anderson and Coughlan (1987), and Krafft et al. (2004) focus on human asset specificity. While John and Weitz (1988), Shervani et al. (2007), and Brettel et al. (2011a) consider both human and physical asset specificity in their theoretical discussions, their empirical analyses focus only on human asset specificity. Klein et al. (1990), Aulakh and Kotabe (1997), and Brettel et al. (2011a) use a single measure of asset specificity that contains distinct items measuring human and physical asset specificity. Importantly, none of these studies has examined the dimension of physical asset specificity while controlling for the impact of human asset specificity. These observations suggest that further research is needed that explicitly measures and evaluates the relative importance of human and physical asset specificity in the channel integration decision.
Research Hypotheses
Based on the above literature review, we seek to extend existing research by distinguishing between two types of asset specificity, human and physical asset specificity. As already explained, TCE and TCE-based channel integration studies argue that both human and physical asset specificity are positive drivers of the degree of channel integration. Thus, our research hypotheses are the following:
Hypothesis 1. Human asset specificity will be positively related to the degree of channel integration.
Hypothesis 2. Physical asset specificity will be positively related to the degree of channel integration.
Research Methodology
As shown in Table 1, previous empirical studies attempt to test the basic TCE hypothesis concerning asset specificity and channel integration using (1) a particular measure of asset specificity, (2) data from a single national survey of firms in the United States, Canada, or Germany, and (3) methods such as an ordinary least squares (OLS) regression analysis and a partial least squares structural equation modelling (PLS-SEM) approach. In contrast with these studies, we seek to test the above two hypotheses concerning two types of asset specificity and channel integration using (1) different measures of asset specificity, (2) data from parallel national surveys of firms in two countries with different cultures, the United States and Japan, and (3) the methods used in prior empirical analyses and an IV-2SLS approach, which is a widely accepted method for investigating the potential endogeneity problem of focal explanatory variables (Antonakis et al., 2010, 2014; Zaefarian et al., 2017). This research strategy is partly based on the guidelines for high-quality replication studies articulated by Bettis et al. (2016b). The aims are to assess the generalizability of important prior results using different survey data drawn from different research contexts and to assess the robustness of these results using different measures and methods, thereby providing important additional evidence that contributes to the establishment of repeatable cumulative knowledge (Bettis et al., 2016a, 2016b). We developed the survey questionnaire in several steps. Following John and Weitz (1988), Shervani et al. (2007), and Brettel et al. (2011b), the dependent variable, channel integration, was operationalized by the percentage of sales through direct channels. We measured the focal explanatory variable, asset specificity, in four ways: (1) a four-item scale of human asset specificity used by Shervani et al. (2007), (2) a four-item scale of physical asset specificity based on Bello and Lohtia (1995) and Klein et al. (1990), (3) a six-item scale of human and physical asset specificity used by Klein et al. (1990), and (4) a four-item scale of human and physical asset specificity used by Brettel et al. (2011a). We also included four control variables: environmental uncertainty, behavioral uncertainty, financial performance, and channel members’ capabilities. Based on existing studies, manufacturers of electronic and telecommunication, metal, and chemical products in industrial (business-to-business) markets were selected as the setting for the empirical test. The unit of analysis was the domestic channel integration decision made at a product-market level. Respondents were sales/marketing managers (or executives) knowledgeable about channel design and strategies. In the United States, a professional marketing research company administered the data collection. In Japan, respondents were surveyed by mail. In total, we obtained 235 usable responses from US managers and 279 responses from Japanese managers.
Results and Conclusions
Following similar studies (John & Weitz, 1988; Shervani et al., 2007), an OLS regression analysis was used to test the hypotheses. The results, shown in Table 1, exhibit significant explanatory power for each model. As expected, (1) human asset specificity exhibits significant positive relationships with the degree of channel integration in both the United States and Japan (Models 1 & 2). These findings support Hypothesis 1. Conversely, (2) physical asset specificity does not have the expected significant positive relationships with the degree of channel integration in both the United States and Japan (Models 1 & 3). These findings do not support Hypothesis 2. Also, (3) asset specificity (Klein et al., 1990) and (4) asset specificity (Brettel et al., 2011a), two composite measures of human and physical asset specificity, exhibit the expected significant coefficients (Models 4 & 5). Additionally, we conducted a similar analysis using a structural equation modelling approach. The results mirrored those of OLS regression, thus providing further support for it. To assess the problem of potential endogeneity between asset specificity and channel integration, we employed IV-2SLS. We used (1) the level of the product’s technical content and (2) the need for coordination between production and distribution activities as instruments for human/physical asset specificity. Our instruments were individually significant predictors of asset specificity and met the exclusion restriction. However, the endogeneity test revealed no evidence of endogeneity. Thus, asset specificity was treated as exogenous in the model. In summary, our preliminary results suggest that human asset specificity, not physical asset specificity, is relevant to the channel integration decision. This finding is significant in that TCE-based channel integration studies tend to measure only one type of asset specificity. We are currently conducting additional analyses to better understand the relationship between human and physical asset specificity, for example, (1) the effects of human and physical asset specificity on different kinds of direct distribution, and (2) a multiple equation model in which human asset specificity is a function of physical asset specificity and direct distribution is a function of both human and physical asset specificity. We believe that our results will have important implications for the ways in which managers approach the channel integration decision.

• Hidesuke Takata(Keio University, Japan)
• Mark E. Parry(University of Missouri-Kansas City, USA)