UNIVERSITY OF GRONINGEN – FACULTY OF ECONOMICS AND BUSINESS
Trust as control mechanism
in Inter-organizational
relationships
An empirical analysis: goodwill trust, calculative
trust and alliance outcome
by
Roman Schneider S2995557
M.Sc. Organizational & Management Control
Word Count: 8463
Supervisor: A.R. Abbasi
Second assessor: Dr. M.P. van der Steen
Table of Content
Abstract ... 1
1 Introduction ... 1
2 Theoretical Framework ... 3
2.1 Transaction-Cost Economics & Opportunism ... 3
2.2 The Role of Trust in Inter-Organizational Relationships ... 4
2.3 Calculative Trust and Opportunistic Behaviour ... 5
2.4 Goodwill trust and Opportunistic Behaviour ... 5
2.5 Moderator - Behaviour Uncertainty ... 6
2.6 Controls and Conceptual Model... 7
3 Research Methodology ... 7
3.1 Sample, Data Collection & Measurement of the variables ... 8
3.2 Analysis ... 8
3.2.1 Data Preparation & Test for Internal Consistency ... 9
3.2.2 Test for Multinomial Distribution ... 11
3.2.3 Structural Equation Model – Calculation & Validity ... 12
4 Results ... 14
5 Discussion, Limitations and Conclusion ... 16
5.1 Limitations ... 17
5.2 Conclusion ... 17
References... 19
1
Abstract
This paper contributes to the discussion of trust as control mechanism in inter-organizational relationships. Building on Transaction-Cost-Economics, previous findings and recent developments in the literature, this paper finds empirical evidence that goodwill trust and calculative trust are significant predictors for alliance outcomes. The data set collected from companies located within Indonesia, Netherlands and Germany show mixed results. After controlling for asset specificity and alliance age, I found that goodwill trust is negatively related towards opportunistic behaviour and consequently alliance outcome. In contrast, calculative trust was found to be the opposite. This contradicts previous findings and TCE literature. Furthermore I found that calculative trust affects opportunistic behaviour stronger in environments characterized by high behaviour uncertainty.
Keywords: Inter-Organizational Relationships, Goodwill Trust, Calculative Trust, Alliance Outcome, Opportunistic Behaviour, Transaction Cost Economics, Behaviour Uncertainty
1 Introduction
In the last decades, the competitive environment of business became more intense and global orientated. One way to deal with this phenomenon for companies is to collaborate in inter-organizational relationships. In addition to the traditional buyer-supplier relationships, IORs are often conceptualized to increase competitive advantages and have a variety of forms. These forms are for instance franchises, joint ventures, supplier- relationships, outsourcing and so on (Das and Teng, 1996).
Unfortunately for companies, there are also risks associated with collaborations. The academic literature distinguishes between two types of risks, the Performance Risk and the Relational Risk. Performance risk refers to the probability that objectives of the collaboration cannot be achieved because of market forces, such as competition. Relational risk relates to the possibility of failure because of lack of cooperation and hence, this risk is unique to collaborations. Relational risk increases with opportunistic behaviour among collaborators, for instance by cheating, distorting information, shirking etc.
2 is still immature and lacks empirical evidence. As a result, we know very little about the relationship between trust and control and the effect on alliance outcomes.
Drawing on TCE, Judge and Robert (2006) found empirical evidence for trust and contractual safeguards as predictors for alliance outcomes. In particular, trust was found to be a strong predictor for successful alliance outcomes. However, the data was limited to strategic alliances in the United States healthcare sector. Furthermore scholars argue that distinct types of trust exist (e.g. Lewicki and Bunker, 1996; Rosseau, Sitkin, Burt and Camerer, 1998; Poppo, Zou and Julie, 2015; Zaheer, McEvility and Perrone, 1998). Two of these types are “calculative trust” and “goodwill trust”. Calculative trust refers to expectation that another actor will behave in a certain way because overall benefits of that behaviour outweighs the benefits of not doing so. Additionally costs of misconduct are included. In contrast, goodwill trust arises from the belief in the other person’s moral character (goodwill) and enhances inter-organizational trust between partners.
Based on the previous findings of Judge and Robert (2006), I aim to contribute to the academic literature by finding empirical evidence about the relationship of the two distinctive concepts of trust (goodwill and calculative) and alliance outcomes in IORs.
RQ: What is the relationship between the two types of trust (goodwill and calculative) and alliance outcomes in IORs?
3
2 Theoretical Framework
2.1 Transaction-Cost Economics & Opportunism
Transaction cost economics (TCE) belongs to the new institutional theories and describes company governance based on the costs of transactions. A transaction hereby is simply the exchange of goods or services between two or more partners. Transaction costs occur when companies gather information, negotiate prices or contracts and adapt, enforce or monitor contractual agreements. The partners will try to optimize that exchange by increasing the efficiency of their resources. To do that, the actors will choose an appropriate governance mechanism that must be matched with the nature of the transaction (Williamson, 1985). In principle, the transactions are governed by three general mechanisms: market mechanism where prices govern, hierarchical mechanism where managers govern within the boundaries of the firm and a hybrid form (Williamson, 1991). TCE predicts that the mechanism associated with the lowest transaction cost will be chosen. The transaction costs depend on a combination of transaction and human factors. Transaction factors relate to the frequency, the level of uncertainty and the asset specificity (i.e. investments made for the IOR with little alternative use) of the transaction. The two human factors, bounded rationality and opportunism, refer to characteristics of the human nature (Dekker, 2004) and occur when the actors perceive potential hazards in governance structures (Williamson, 1998). Bounded rationality refers to the fact that actors have limited access to information, limited abilities to process the information and the unpredictability of all possibilities in the future. As such, written contracts or agreements between partners are limited by contingencies that are unforeseen. Williamson (1975) originally defined opportunism as “self-interest seeking with guile”. Later he elaborated guile as “calculated efforts to mislead, distort, disguise, obfuscate, or otherwise confuse” (Williamson, 1985). Other scholars have suggested that opportunistic behaviour can demonstrate itself in many forms, in particular by appropriating knowledge, evading obligations, breaching contracts, overstating capabilities, or holding up the partner (e.g. Dickson, Weaver and Hoy, 2006; Hamel, 1991). Opportunism arises in settings with high uncertainty when partners work for their own personal interests and/or contracts are flawed and leave room for opportunism (Dekker, 2004). TCE suggests that the potential damage caused by opportunism must be matched with appropriate governance mechanisms.
4 type of opportunism is especially a big problem in strategic IORs (i.e. IORs that aim to increase their competitiveness) because of a built-in bias towards goal conflict (Judge and Roberts, 2006). It is not surprising that opportunistic behaviour is often cited as primary cause for alliance failure because rising potential hazards from opportunism and it accompanying increasing transaction costs eventually exceed the overall benefits of the IOR, leaving no rational reason to continue the relationship. Hence, my first hypothesis is:
H1: Opportunistic behaviour is negatively related to alliance outcome
2.2 The Role of Trust in Inter-Organizational Relationships
As stated above, opportunism is a particularly big problem in IORs. From the TCE perspective, the actors need to adapt to appropriate governance mechanism to protect the company from this potential hazard. The two most accepted and debated mechanisms in the literature are the control and trust mechanism (e.g. Long and Sitkin, 2006). Especially for the latter, academics opinions are diverse. This is not surprising because earlier studies about trust in IORs often considered trust as a “rational prediction” (Lewis and Weigert, 1985) which promises to create favourable outcomes for the relationship (i.e. lower transaction costs). For example, Das and Teng (1998) define trust as "positive expectations about another's motives with respect to oneself in situations entailing risk".
Although the rational prediction of trust is important for its understanding, it provides an incomplete understanding of trust on its own (Wicks, Berman and Jones., 1999). Therefore it is important to consider additional characteristics of trust to complete this understanding. To begin with, one of this characteristic is the emotional bond between people. Over time, trust develops because people form emotional bonds with each other, enabling them to move beyond rationality to replace it with a “leap of faith” (Lewis and Weigert, 1985). Another important element of trust is the moral character of the trustee. This assumption have been emphasized by Ethicists with respect that the emotional bond between partner only evolve with the belief in the other person’s moral character or “good will” (Becker, 1996; Baier, 1994).
These developments in the perception of trust lead to two types of distinction between different types of trust in IORs. The first development distinguishes trust on an personal and inter-organizational level, goodwill trust and competence trust (Das and Teng, 1998; Lui and Ngo, 2004). Goodwill trust refers to expectation that the partner intends to fulfil his role in the relationship based on mutual perceptions and key characteristics of the trustee. Goodwill trust is linked to relational risk. On the other hand, competence trust refers to the expectation that the partner’s ability to fulfil his role. Competence trust strongly derives from reputation effects (Zaheer and Harris, 2006).
5 prediction) and assumes that incentives with rewards can lead to predictable outcomes (Zaheer and Harris, 2006; Poppo, Zou and Julie, 2015). In other words, calculative trust predicts the behaviour of the trustee in relation to the overall benefits and costs associated with that behaviour. Relational trust refers to social element of trust developed over time. Parties in long-lasting, stable relationships collect experiences with consequence to development of mutual expectations, shared values and normative conventions. The result of the development is a mutual agreed working environment which defines how both parties work together (Bercovitz and Nickerson, 2006). With respect to the feasibility in the short time of the master thesis, this paper focusses on 2 of the 4 commonly accepted types of trust: calculative trust and goodwill trust. 2.3 Calculative Trust and Opportunistic Behaviour
As briefly mentioned above, calculative trust refers to expectation of parties in a relationship that the other party will fulfil its obligations because of rewards and punishments. As such, it requires specific calculations and specification of benefits and costs that are related towards certain behaviours (Bromiley and Harris, 2006; Williamson, 1993). When partners believe that benefits outweigh costs in a specified transaction, it motivates them to fulfil their performance obligations (Poppo, Zou and Julie, 2015). Thus, suggesting that when goodwill trust increases, the behaviour of actors in IORs is more likely to converge towards the mutual agreements of the relationship and reduces the overall transaction cost. The relation supports the following hypothesis:
H2: Calculative Trust is negatively related to opportunistic behaviour
2.4 Goodwill trust and Opportunistic Behaviour
Previous literature found that the key to develop inter-organizational trust is the relationship between boundary role persons (Curall and Judge, 1995; Lui and Ngo, 2004). This suggests that managers must trust each other on an inter-personal level (at least to a minimum degree) and thus, enhances the importance of goodwill trust. Goodwill trust refers to inter-personal trust between persons based on mutual perceptions and key characteristics of the trustee. Goodwill trust occurs because an emotional bond between partners is developed over time. However, the emotional bond is not just only the relationship between trustor and trustee, but for the most part, the belief in the others person`s moral character (Wicks, Berman and Jones., 1999).
6 TCE relates goodwill trust with relational risk in IORs (Das and Teng, 1998). Previous literature about inter-organizational trust has shown that IORs with a higher level of inter-organizational trust have positive implications. Inter-organizational trust can improve overall communication between partners and increases information flow in IORs (Zaheer, Mcevily and Perone, 1998). Trusting partners are also more cooperative, show more commitment towards each other and the quality of interactions increases (Morgan and Hunt, 1994; Moorman, Zaltman and Deshpande, 1992). Furthermore, inter-organizational trust is associated with higher performance and lower transaction costs (Sako, 1998; Zaheer, McEviliy and Perone, 1998; Dyer and Chu, 2003).
To summarize, goodwill trust is positively associated with inter-organizational trust. Inter-organizational trust is desirable for IORs because it has several positive implications for the relationship. As such, it is expected to reduce the opportunistic behaviour of the partners, decreases transaction costs and improves the alliance outcome.
H3: Goodwill Trust is negatively related to opportunistic behaviour.
2.5 Moderator - Behaviour Uncertainty
Behaviour uncertainty refers to the “extend to which one party cannot effectively observe or evaluate the activities of the other party” (Poppo, Zou and Ryu, 2008). TCE assumes that in situations with high uncertainty, opportunistic behaviour increases because of actor’s bounded rationality (i.e. limitation to make accurate predictions about the future event). The results are that limited contracts are created that do not cover unexpected contingencies, increasing the possibility for opportunistic behaviour (Dekker, 2004). Therefore alternative control mechanisms must be found.
With Calculative trust, actors adapt to unobserved behaviour through expected rewards and punishment. When behaviour uncertainty is high, parties cannot observe the inputs of the other party. However, calculative trust allows them to measure the final outcome when rewards and punishments are successfully applied (Poppo, Zou and Ryu, 2015). Thus, calculative trust can reduce the relational risk of opportunism and increases the outcome of the alliance. This suggests the following hypothesis:
H4: The relationship between Calculative trust and opportunistic behaviour is stronger when behaviour uncertainty is high than when it is low
7
H5: the relationship between goodwill trust and opportunistic behaviour is stronger when behaviour uncertainty is high than when it is low
2.6 Controls and Conceptual Model
In addition to the two independent variables and the moderator, I choose two variables to control for heterogeneity. The first variable controls the data set for alliance age. Previous literature suggests that over time, IORs partners develop a habituation effect in which it is less likely for the IOR to experience opportunism (Judge and Dooley, 2006; Dyer and Singh, 1998).
The second control variable is asset specificity. Previous studies about performance in IORs often control for asset specificity because of TCE literature. TCE argues that relationships that made transaction specific investments which are hard to use for other purposes, are less likely to experience opportunism and hence favour alliance outcomes (Williamson, 1985).
The conceptual Model is illustrated in figure 1.
Figure 1: Conceptual Model
3 Research Methodology
8 3.1 Sample, Data Collection & Measurement of the variables
The sample of this study contains a questionnaire consisting of eleven sections compromised in a database under the supervision of Prof. A.R. Abbasi. The questionnaire was translated into Dutch, Indonesian and German and the companies were selected at random by seven master students with the restriction of active and commercial operating and a minimum size of at least 10 employees and an annual turnover of 2,5 million Euros. The data collection was conducted during March 2017. In my case, 35 companies were contacted by email and 27 personal contacts were contacted by telephone or email. Out of this 62 contacted companies, 27 replied (respond rate= 41,54%). Overall, 23 companies agreed and conducted the survey. The respondents from Germany, Indonesia and Netherlands had in-depth knowledge about contracts between their company and the partner in the IOR. The data from the Dutch firms and partly from the German firms were collected by conducting face-to-face interviews to increase reliability. Depending on the individual, the interviews were set for 60 to 90 minutes and the sections included questions about the contracts, control mechanisms, trust, synergies, performance, risks, uncertainties and dependencies related to the IOR.
The variables of this study were provided by the supervisor A.R. Abbasi and measured by a seven-point Likert-type scale ranging from “strongly disagree” to “strongly agree”, “completely inaccurate” to “completely accurate” and “very dissatisfied” to “very satisfied”. Because of confidential reasons, no further information was granted for these questions. In addition, several questions were aimed to asked specific information about the persons and company (e.g. sex, age). This included the question about the alliance age. The variables I investigated comprised several items as measurements. In total, 35 items were provided, conceptualized as follow:
• Alliance outcome - 12 items
• Opportunistic behaviour - 6 items
• Goodwill trust - 9 items
• Calculative trust – 3 items
• Behaviour uncertainty – 2 items
• Asset specificity – 3 items 3.2 Analysis
9 both approaches is that the covariant analytical model follows a holistic, simultaneous estimation of the causal structure by reproduction of variance-covariance-matrix calculation whereas the full information approach is more distinct. Therefore, it is more appropriate in reflective measurement models (Weiber and Mühlhaus, 2015). The most common estimation procedure within the covariant analytical approach is the maximum-likelihood approach.
Because the SEM maximum-likelihood approach needs several requirements (Weiber and Mühlhaus, 2015), I divided the upcoming sections into several parts. In the first part, I explain prerequisite work of data preparation and tests about internal consistency reliability in SPSS. The next section explains the methodology around the test for multinomial distribution. The last section explains tests for validity and for hypothesis, using a two-step approach.
3.2.1 Data Preparation & Test for Internal Consistency
The data set consisted of 140 lines with several empty entries and out of the 35 variables, 5 were reverse coded. In order to conduct reliable tests and calculations, it was therefore required to clean up the data and prepare it for future calculations. After cleansing for empty entries as well as 4 partly incomplete entries because of questionable reliability, a total sample size of 95 entries remained. Secondly, the 5 reverse coded variables were re-coded and a frequencies analysis (Appendix A) was performed to identify remaining missing entries. In sum, 23 fields missed entries across the variables and maximum 2 per variable. The remaining missing entries were later replaced by using the “estimate mean” function during maximum-likelihood calculations.
After the data preparation, I conducted exploratory factor analyses (EFA) with Principal Axis Factoring Method and Promax Kaiser Rotation (Weiber and Mühlhaus, 2015). Firstly, the measurements were observed isolated to control for one-dimensionality. Secondly, after removing the initial measures from the isolated EFA, a comprehensive EFA for all variables was performed to eliminate indicators and ensure only the most suitable and important variable are used (Homburg and Giering, 1996). To assess the validity of the measures, the Bartlett’s test of sphericity and the Kaiser-Meyer-Olkin (KMO) were calculated during the comprehensive EFA (table 1). The KMO explains the cohesiveness of the variables and should not fall below 0,6 (Kaiser and Rice, 1974). The Bartlett’s test of sphericity measures the overall significance of the variables by testing the null hypothesis and should be rejected (Dziuban and Shirkey, 1974). Both criteria’s reported by the EFA are sufficient with 0,773 exceeding 0,6 and the p value of 0,000 which rejects the null hypothesis.
Table 1: KMO and Bartlett's Test
Kaiser-Meyer-Olkin Measure of Sampling Adequacy. ,773
Bartlett's Test of Sphericity Approx. Chi-Square 762,460
df 153
Sig. ,000
10 of items per variable to 5 for alliance outcome, 3 for opportunistic behaviour, 4 for goodwill trust, 1 for calculative trust, 3 for asset specificity, 2 for behaviour uncertainty.
Table 2: factor loadings based on the Pattern Matrix in the EFA
Factor 1 2 3 4 5 6 ALLIANCEOUTCOME_4 ,787 ALLIANCEOUTCOME_5 ,522 ALLIANCEOUTCOME_10 ,857 ALLIANCEOUTCOME_11 ,802 ALLIANCEOUTCOME_12 ,919 OPPORTBHVR_1 ,873 OPPORTBHVR_2 ,739 OPPORTBHVR_4 ,750 GWILTRUST_1 ,537 GWILTRUST_4 ,692 GWILTRUST_5 ,755 GWILTRUST_9 ,367 CALCTRUST_2 ,773 ASSETSPEC_1 ,778 ASSETSPEC_2 ,555 ASSETSPEC_3 ,657 BHVRUN_1 ,577 BHVRUN_2 ,814
Extraction Method: Principal Axis Factoring.
Rotation Method: Promax with Kaiser Normalization.a a. Rotation converged in 6 iterations.
The last step checked the variables for internal consistency reliability using the Cronbach alpha test. The formula for the calculation is:
11 Results from the Cronbach alpha can range from -∞ to +1 and the literature suggest that a Cronbach alpha ≥ 0,7 indicates a good internal consistency of the construct whereas values below 0,5 are unacceptable (Nunally, 1994). The results of the Cronbach alpha tests are summarized in Table 3. Overall, the estimated results of the Cronbach alpha test deliver good results with asset specificity slightly below 0,7, but still acceptable (> 0,5). However, the reliability of calculative trust seems to be questionable. During the comprehensive EFA, two of the three initial items had strong loadings on other factors and thereby were removed. The Cronbach alpha originated from the initial three items was very low (0,288), suggesting reliability issues.
Table 3: Descriptive statistics, internal consistency and zero-order correlations
Variables Mean Std. Dev. alpha 1 2 3 4 5 6 1. Alliance Age 9,59 7,75 2. Asset Specifity 4,10 2,27 0,665 0,102 3. Goodwill Trust 4,96 1,17 0,748 0,015 -0,191 4. Calculative Trust 3,58 1,78 0,026 0,223* 0,165 5. Behaviour Uncertainty 3,01 0,69 0,744 -0,235 0,161 -0,426* -0,032 6. Opportunistic Behaviour 2,49 1,06 0,841 -0,114 0,157 -0,249 0,204 0,575** 7. Alliance Outcome 5,52 1,09 0,878 0,217 -0,181 0,62** -0,053 -0,473* -0,455** Notes: n=95 ; *p<0,05; **p<0,01
3.2.2 Test for Multinomial Distribution
As stated above, the maximum-likelihood method in AMOS is the most common used method for SEM estimations. However, the method requires a multinomial distribution of the data because of potential distortions of the results (Weiber and Mühlhaus, 2015). The examination of the multinomial distribution should take place in two steps, a univariate observation of each variable and an overall check for multinormality using “Mardia's Tests of Multinormality” (Mardia, 1970). Criterions for the univariate assessment are the skewness and kurtoses. Thresholds for violation of normal distribution, is debated in the literature (Temme and Hildebrandt 2009; West, Finch and Curran, 1995). I followed West’s, Finch’s and Curran’s (1995) proposition who suggest that the values for skewness (>2) and kurtosis (<7) should not be exceeded. In addition, C.R. values above 2,57 are considered as violation for both univariate and multivariate distribution.
Table 4: assessment of normality
Variable min max skew c.r. kurtosis c.r.
12
Variable min max skew c.r. kurtosis c.r.
CALCTRUST_2 1,000 7,000 ,296 1,177 -1,369 -2,723 BHVRUN_1 1,000 6,000 ,591 2,352 -,712 -1,416 BHVRUN_2 1,000 7,000 ,403 1,602 -1,068 -2,124 ASSETSPEC_1 1,000 7,000 ,034 ,135 -1,266 -2,520 ASSETSPEC_2 1,000 7,000 -,523 -2,081 -,679 -1,351 ASSETSPEC_3 1,000 7,000 -,086 -,341 -,995 -1,980 GWILTRUST_1 1,000 7,000 -1,164 -4,631 1,596 3,175 GWILTRUST_4 1,000 7,000 -,808 -3,217 ,199 ,396 GWILTRUST_5 1,000 7,000 -,823 -3,276 -,081 -,161 GWILTRUST_9 1,000 7,000 -,641 -2,550 -,279 -,555 OPPBHVR_1 1,000 7,000 1,011 4,021 ,531 1,057 OPPBHVR_2 1,000 7,000 1,353 5,382 1,307 2,601 OPPBHVR_4 1,000 7,000 ,863 3,432 -,173 -,343 ALLIANCEOUTCOME_4 1,000 7,000 -1,275 -5,073 2,021 4,021 ALLIANCEOUTCOME_5 2,000 7,000 -,753 -2,997 -,292 -,582 ALLIANCEOUTCOME_10 2,000 7,000 -,835 -3,323 ,361 ,718 ALLIANCEOUTCOME_11 2,000 7,000 -,735 -2,925 ,238 ,473 ALLIANCEOUTCOME_12 2,000 7,000 -1,137 -4,525 1,081 2,150 Multivariate 46,652 8,048
Table 4 illustrates the assessment of normality from SPSS Amos. The values show sufficient results for both skewness and kurtosis. However, for 11 variables the critical ratio of 2,57 exceeds either the skew or kurtosis. Additionally both values (Kurtosis = 46,652; C.R. 8,048) for the multivariate are critical. Elimination for outliers based on AMOS recommendations didn’t produce more positive results and overall model fits (described in the next section) decreased. Further, a transformation of the items is not recommended because of room for manipulation and difficulties with the interpretation of the results (Weiber and Mühlhaus, 2015).
It is, however, also important to note, that for most rating-Likert scaled measurement models, the distribution of data is not given. Therefore the extent of the violation has to be evaluated (Scholderer and Balderjahn, 2006). Under the condition of a moderate violation of the normality distribution, the maximum-likelihood calculation method is still appropriate. Therefore I chose to continue with analysis using the maximum-likelihood method.
3.2.3 Structural Equation Model – Calculation & Validity
13 During the first step, confirmatory factor analyses (CFA) were performed to assess the discriminant validity of the measures and evaluate the model fit. First I used an unrestrictive measurement model that allowed each latent variable to correlate freely and restricted each item to load on its presumed latent variable. To evaluate the model, I used model fit measures provided by AMOS. These measures compare the default model with an independent model. Little differences between the independent model and the default model lead to indices near zero whereas clear improvements of the model lead to indices near 1. The indices I used are the normed fit index (NFI), compared fit index (CFI) and Tucker-Lewis-Index (TLI). The normed fit index equals the difference between the chi-square of the default model and the independent model, divided by the chi-square of the independent model. The TLI is similar to the NFI but also considers the degrees of freedom for both models. The CFI represents the ratio of the discrepancies between the default model and the independence model. For all three fit indices, thresholds ≥ 0,9 indicate a good model fit. In addition, the Root Mean Square Error of Approximation (RMSEA) and chi-square to degrees of freedom ratio (x²/df) were used to further evaluate the model fit. For the RMSEA, values below 0, 05 indicate good, values between 0,05 and 0,08 indicate reasonable and values above 0,1 indicate unacceptable model fits (Browne and Cudeck, 1993). The x²/df ratio should be smaller than 2 (Byrne, 1989). Lastly, I compared the unrestrictive measurement model with a restrictive model that changed all correlations to 0 and performed chi-square difference test to assess discriminant validity. Significant Changes between the unrestrictive model and restrictive model (Δx² > 8,36; p< 0,05) support discriminant validity (Jöreskog, 1971).
14
4 Results
The confirmatory factor analysis of the latent variables produced following model fit results: x²= 196,712, 133 degrees of freedom, x²/DF= 1,479, comparative fit index (CFI) = 0,905, normed fit index (NFI) = 0,767, Tucker Lewis Index (TLI) = 0,878 and RMSEA = 0,071. The item factor loadings for the respective variables were significant at p< 0,001 except for asset specificity 2 and 3 (p< 0,05). The good results for CFI ≥ 0,9 and x²/DF ≤ 2, reasonable results for RMSEA ≤ 0,08 indicate a good model fit. However, the NFI ≤ 0,9 and TLI ≤ 0,9 express issues. This is not surprising because compared to the CFI, NFI and TLI don’t consider distribution distortions (Raykov, 2000) and the NFI also don’t consider small sample sizes. Furthermore, the Chi-square difference test between the unrestrictive model and the model that constrained each correlation to 0 suggests discriminant validity among the constructs (Δx²=100,813, ΔDF=20, p <0,001).
Table 5: Results and Model Fit from the Nested Model analysis
Path Theoretica
l Model
Constrained Models
1 2 3 4
opportunistic behaviour <--- alliance age -0,154 -0,174 -0,158 0,166 0,17 alliance outcome <--- alliance age 0,175* 0,178* 0,181* 0,179* 0,232* opportunistic behaviour <--- asset specifity 0,177 0,205 0,155 0,166 0,17 alliance outcome <--- asset specifity 0,057 0,063 0,044 0,048 -0,04 opportunistic behaviour <--- goodwill trust -0,33** 0 -0,256* -0,262* -0,377** opportunistic behaviour <--- calculative trust 0,236* 0,198 0 0,236* 0,226* opportunistic behaviour <--- bhvUncXGwillTr -0,059 0 -0,123 0 -0,081 opportunistic behaviour <--- bhvUncXCalcTr 0,35 0,344 0 0 0,232 alliance outcome <--- opportunistic
behaviour -0,377** -0,41** -0,371** -0,383** 0
Model Fit and Comparison
x² 366,629 375,211 379,714 375,84 378,496 df 298 300 300 300 299 X²/df 1,23 1,251 1,266 1,253 1,266 CFI 0,937 0,931 0,927 0,931 0,927 NFI 0,746 0,74 0,737 0,74 0,738 TLI 0,926 0,92 0,915 0,919 0,915 RMSEA 0,049 0,052 0,053 0,052 0,053 Δx² 8,6* 13,09** 9,211* 11,867** Δdf 2 2 2 1
15 The results of standardized coefficients and several model fits from the nested model analysis between theoretical model and the four constrained models are reported in table 5. The chi-square difference test between the theoretical model and the four models show, that each of the four models provides significantly worse results than the theoretical model. Furthermore, the model fit criteria’s (i.e. CFI, NFI, TLI and RMSEA) display the best results in the theoretical model. This suggests that the theoretical model deliver the best results and no path in the theoretical model should be omitted.
A comparison between model 4 and the theoretical model indicate significantly worse results in Model 4 (Δx² ≥ 8,36; p < 0,01). In addition, a negatively significant relationship (p< 0,01) between opportunistic behaviour and alliance outcome has been reported in the theoretical model. This suggests strong support for Hypothesis 1.
Similarly results are reported in Model 1 and Model 2. Both chi square differences are significant at the 0,05 and 0,01 level. The standardized path coefficient between goodwill trust and opportunistic behaviour is negative at the 0, 01 level and suggest robust support for hypothesis 3. However, the relationship between calculative trust and opportunistic behaviour indicate a positive relationship (p< 0,05). Thus, the data do not support hypothesis 2.
A model assessment between the model that constrained both moderating effects to 0 and the theoretical model also show a significant change in x² (p < 0, 05). The results indicate that the relationship between goodwill trust and opportunistic behaviour is not affected by rising behaviour uncertainty. Therefore hypothesis 5 is not supported. Lastly, a significant moderating effect (p< 0,01) between the relationship of calculative trust and alliance outcome is reported. This suggests support for hypothesis 4.
16
Figure 2: derived path coefficients based on the analysis of the theoretical model
5 Discussion, Limitations and Conclusion
The results of this study found mixed support for the constructed hypothesis derived from TCE and IOR literature. To begin with, I found that opportunistic behaviour is a significant predictor for alliance outcomes. Though I expected the relationship to be stronger because of its strong relation to transaction costs and relational risk (Das & Teng, 1998), the direction was predicted by TCE. I also found strong support that trust from different sources have distinct effects on opportunistic behaviour. Furthermore both types of trust seem to have similar strong influence on opportunism (the estimated coefficients for good trust with -0,33 is not much stronger than the +0,236 of calculative trust). Nonetheless, it is surprising that calculative trust was found to be positively associated with opportunism whereas TCE literature predicts the opposite and similar studies confirmed the TCE assumption (Poppo et al, 2015; Judge et al, 2006). A possible explanation is that the variable calculative trust has been measured by one item.
17 (e.g. monitoring) because they are easier to justify (i.e. monitoring generate facts which helps managers to get support within their firm).
Although the respective loadings between the control variables and the two dependent variables opportunistic behaviour and alliance outcome were considerable, only the coefficient between alliance age and alliance outcome was found significant. I argued that over time, relationships experience habituation and therefore it is very likely that opportunistic behaviour decreases and hence alliance outcome increases. This habituation, however, seems to be have a stronger effect on alliance outcome than for opportunism. Furthermore asset specificity was found as insignificant predictor for opportunistic behaviour and alliance outcome. A possible explanation is that a considerable amount of companies studied in this paper were buyer-supplier relationships. This form of IORs is characterized with a lower level of asset specificity and hence lower influence on opportunism and performance.
5.1 Limitations
The study is subject to some limitations. First, the sample collected from the survey is heterogeneous. Companies were selected across different industries and countries (i.e. Indonesia, Germany and Netherlands). Similar studies that investigated performance or outcomes in IORs chose their data selection more carefully. Since I controlled for asset specificity and alliance age, possible distortions of the results are possible. Second, calculative trust has been measured by one item. Hence reliability and validity of the data is suspect and should be viewed with caution. Third, since the data sampled three countries across the globe, generalizability to other countries may be limited. Fourth, the data set used in this study had multinomial distribution issues and thus changes in the results are possible. Lastly, since the data is cross-sectional in nature, alternative relationships may exist.
5.2 Conclusion
Despite of the mentioned limitations above I conclude, that the research question is at least partially answered. Like the previous findings of Judge and Robert (2006), this study shows that opportunistic behaviour is negatively related to alliance outcomes. This confirms the TCE assumption that opportunism increases the costs of transactions and reduces performance. Furthermore the results indicate that different types of trust have distinct effects on opportunism. In fact, the results indicate that both types, calculative trust and goodwill trust, are similar strong predictors.
With reference to my previously mentioned limitations, future research might want to choose the data collection and construct measurements more carefully. A bigger and a more homogeneous data set (i.e. controlling for industry and country) is recommended. Second, future research should also consider TCE based control mechanisms (e.g. contractual safeguards) to give a more comprehensive view of the predictors. Lastly, because the research design is cross-sectional with single respondents, it does not highlight the dynamics between the mechanisms. Recently, researchers began to emphasize the possibility dynamic relationships of trust and control (e.g. Vosselman and van der Meer-Koolstra, 2009). Therefore future research can focus on the
18 The findings also have managerial implications. The results emphasize the importance for
manager’s expertise in building and expanding trust between IORs. Furthermore it seems important to know, that trust has distinct forms which should be recognized and treated differently by managers. Clear communication structures, efficient conflict management can facilitate trust and hence improve alliance outcomes.
19
References
Anderson, J. C., & Gerbing D. W. (1988). Structural equation modeling in practice: A review and recommended two-step approach. Psychological Bulletin, 103, 411–423
Backhaus, K., Erichson, B., Plinke, W., & Weiber, R. (2011). Multivariate Analysemethoden (13. edition). Berlin: Springer. (in german)
Baier, A. C. (1994). Moral prejudices. Cambridge, MA: Harvard University Press. Becker, L. C. (1996). Trust as noncognitive security about motives. Ethics, 107, 43-61.
Bercovitz, B., Jap, S., & Nickerson J. (2006). The antecedents and performance implications of cooperative exchange norms. Organization Science 17, 724–742.
Bromiley P., & Harris J. (2006). Trust, transaction cost economics, and mechanisms. In Handbook of Trust Research, Bachmann R, Zaheer A (eds). Edward Elgar Publishing: Northampton, MA, 124– 143.
Browne, K. A., & Cudeck, J. S. (1993). Alternative ways of assessing equation model fit. Testing structural equation models. Newbury Park: Sage.
Byrne, B. M. (1989). A primer of LISREL: Basic applications and programming for confirmatory factor analytic model. New York: Springer.
Currall, S. C., & Judge T.A. (1995). Measuring trust between organizational boundary role persons. Organizational Behavior and Human Decision Processes, 64(2), 151–170.
Das, T. K., & B. Teng. (1996). Risk types and inter-firm alliance structures. Journal of Management Studies 33 (6), 827-843.
Das, T. K., & Teng, B. S. (1998). Between trust and control: Developing confidence in partner cooperation in alliances. Academy of Management Review, 23(3), 491-512.
Dekker, H. C. (2004). Control of inter-organizational relationships: Evidence on appropriation concerns and coordination requirements. Accounting Organizations and Society, 29(1): 27- 49. Dickson, P. H., Weaver K. M., & Hoy, F. (2006). Opportunism in the R&D alliances of SMES: The roles of the institutional environment and SME size. Journal of Business Venturing 21 (4), 487–513. Dyer, J., & Chu W. (2000). The determinants of trust in supplierautomaker relationships in the U.S., Japan, and Korea. Journal of International Business Studies, 31(2), pp. 259–285.
Dziuban, C. D., & Shirkey, E. C. (1974). When is a correlation matrix appropriate for factor analysis?. Psychological Bulletin, 81, 358–361.
20 Homburg, C., & Baumgartner, H. (1995). Beurteilung von Kausalmodellen. Marketing ZFP, 17(3), 162–176. (in german)
Homburg, C., & Giering, A. (1996). Konzeptualisierung und Operationalisierung komplexer Konstrukte – Ein Leitfaden für die Marketingforschung, Marketing: Zeitschrift für Forschung und Praxis, 18(1), 5–24. (in german)
Judge, W.Q, & Robert, D. (2006). Strategic Alliance Outcomes: a Transaction-Cost Economics Perspective. British Journal of Management 17, 23-37
Jöreskog, K. G. (1971). Statistical analysis of sets of congeneric tests. Psychometrika, 36, 109–133. Kaiser, H. F., & Rice, J. (1974). Little Jiffy, Mark IV. Educational and Psychological Measurement, 34, 111–117.
Lewicki R.J., & Bunker B.B. (1996). Developing and maintaining trust in work relationships. In Trust in Organizations: Frontiers of Theory and Research. Sage Publication: Thousand Oaks, CA; 114– 139.
Lewis, J., & Weigert, A. (1985). Trust as a social reality. Social Forces, 63: 967-985.
Long, C., & Sitkin, S. (2006). trust in the balance: How managers integrate trust-building and task control. Handbook of Trust Research, 87.
Lui, S. S., & Ngo, H.Y. (2004). The role of trust and contractual safeguards on cooperation in non-equity alliances. Journal of Management, 30(4), 471–486.
Mardia, K. V. (1970). Measures of multivariate skewness and kurtosis with applications. Biometrika, 57, 519–530.
Mjoen, H., & Tallman, S. (1997). Control and performance in international joint ventures. Organization Science, 8, 257–274.
Moorman, C., Zaltman, G., & Deshpande, R. (1992). Relationships between providers and users of market research: The dynamics of trust within and between organizations. Journal of Marketing Research, 29, 314–328.
Morgan, R. M., & Hunt, S. D. (1994). The commitment- trust theory of relationship marketing. Journal of Marketing, 58, 20–38.
Nunnally, J. C., & Bernstein, I. H. (1994). Psychometric theory (3. edition.). New York: McGraw-Hill. Poppo L, Zhou K.Z, Julie J. Li. (2015). When can you trust „trust“? Calculative trust, relational trust, and supplier performance. Strategic management journal 37(4), 724-741
21 Rousseau D.M., Sitkin S.B., Burt R.S., & Camerer C. (1998). Not so different after all: a
cross-discipline view of trust. Academy of Management Review, 23(3), 393–404.
Raykov, T. (2000). On the large-sample bias, variance, and mean squared error of the conventional noncentrality parameter estimator of covariance structure models. Structural Equation Modeling, 7, 431-441.
Sako, M. (1998). Does trust improve business performance? Trust within and between organizations. Oxford: Oxford University Press.
Scholderer, J., & Balderjahn, I. (2006). Was unterscheidet harte und weiche
Strukturgleichungsmodelle nun wirklich? Marketing ZFP, 28(1), 57–70. (in german)
Temme, D., & Hildebrandt, L. (2009). Gruppenvergleiche bei hypothetischen Konstrukten – Die Prüfung der Übereinstimmung von Messmodellen mit der Strukturgleichungsmethodik. Zfbf, 61(2),138–185. (in german)
Van Aken, J., Berends, H., & Van der Bij, H. (2012). Student Projects. In Problem solving in organizations: A methodological handbook for business and management students. Cambridge University Press.
Velez, M. L., Sanchez, J. M., & Alvarez-Dardet, C. (2008). Management control systems as inter- organizational trust builders in evolving relationships: Evidence from a longitudinal case study. Accounting Organizations and Society, 33(7-8), 968-994.
Vosselman, E. G. J., & van der Meer-Kooistra, J. (2006). Changing the boundaries of the firm - adopting and designing efficient management control structures. Journal of Organizational Change Management, 19(3), 318-334.
Weiber, R., & Mühlhaus D. (2014). Strukturgleichungsmodellierung – Eine anwendungsorientierte Einführung in die Kausalanalyse mit Hilfe von AMOS, SmartPLS und SPSS. v.2. Berlin: Springer Gabler (in german)
West, S. G., Finch, J. F., & Curran, P. J. (1995). Structural equation models with nonnormal variables: Problems and remedies. Structural equation modeling. London: Sage.
Wicks A.C., Berman S.L., Jones T.M. (1999). The structure of optimal trust: moral and strategic implications. Academy of accounting review, 24, 99-116
Williamson, O. E. (1975). Markets and hierarchies. New York: Free Press.
Williamson, O. E. (1985). Economic Institutions of Capitalism: Firms, Markets, Relational Contracting. Free Press, New York.
22 Williamson, O.E. (1993). Calculativeness, trust, and economic organization. Journal of Law and Economics 36(1), 453–486.
Zaheer, A., McEvily, B., & Perrone, V. (1998). Does trust matter? exploring the effects of
23
Appendix
Appendix A (frequency analysis – SPSS Output)
24 N Valid 94 94 94 94 95 Missing 1 1 1 1 0 Mean 2,67 2,59 2,61 5,45 5,81 Median 2,00 2,00 2,00 6,00 6,00 Statistics
GWILTRUST_3 GWILTRUST_4 GWILTRUST_5
GWILTRUST_6 R GWILTRUST_7 R N Valid 94 94 94 95 95 Missing 1 1 1 0 0 Mean 5,73 4,69 5,09 5,01 4,05 Median 6,00 5,00 5,00 6,00 4,00 Statistics GWILTRUST_8
R GWILTRUST_9 CALCTRUST_1 CALCTRUST_2 CALCTRUST_3
N Valid 95 95 94 93 95
Missing 0 0 1 2 0
Mean 4,84 4,95 5,56 3,58 5,99
Median 5,00 5,00 6,00 3,00 6,00
Statistics
ASSETSPEC_1 ASSETSPEC_2 ASSETSPEC_3 BHVRUN_1 BHVRUN_2
N Valid 94 94 94 94 94
Missing 1 1 1 1 1
Mean 3,76 4,35 4,18 2,81 3,20
25 Appendix B (results from the CFA – Chi square difference)
Nested Model Comparisons
Assuming model Default model to be correct:
Model DF CMIN P NFI
Delta-1 IFI Delta-2 RFI rho-1 TLI rho2 contrained 20 100,813 ,000 ,119 ,142 ,094 ,118
Appendix C (results from the CFA – model fit) Baseline Comparisons Model NFI Delta1 RFI rho1 IFI Delta2 TLI rho2 CFI Default model ,767 ,701 ,910 ,878 ,905 contrained ,648 ,606 ,791 ,760 ,785 Saturated model 1,000 1,000 1,000 Independence model ,000 ,000 ,000 ,000 ,000 RMSEA
Model RMSEA LO 90 HI 90 PCLOSE
Default model ,071 ,049 ,092 ,056
contrained ,100 ,083 ,117 ,000
Independence model ,205 ,191 ,219 ,000
Appendix D (results from the Analysis – standardized estimated coefficients) Standardized Regression Weights: (Group number 1 - Theoretical model)
26 Estimate ALLIANCEOUTCOME_4 <--- AllianceOutcome ,838 OPPBHVR_4 <--- OpportunisticBehaviour ,800 OPPBHVR_2 <--- OpportunisticBehaviour ,854 OPPBHVR_1 <--- OpportunisticBehaviour ,753 GWILTRUST_9 <--- GwillTrust ,551 GWILTRUST_5 <--- GwillTrust ,748 GWILTRUST_4 <--- GwillTrust ,538 GWILTRUST_1 <--- GwillTrust ,750 ASSETSPEC_3 <--- AssetSpecifity ,621 ASSETSPEC_2 <--- AssetSpecifity ,593 ASSETSPEC_1 <--- AssetSpecifity ,720 CALCTRUST_2 <--- CalcTrust ,870 ALLIANCE_AGE <--- Age ,992 Uncert2xcalc2 <--- BehaviourUncertaintyXCalcT ,872 Uncert1xcalc2 <--- BehaviourUncertaintyXCalcT ,706 Uncert1xGwilltrust1 <--- BehaviourUncertaintyXGWillTrust ,825 Uncert2xGwilltrust1 <--- BehaviourUncertaintyXGWillTrust ,463 Uncert1xGwilltrust4 <--- BehaviourUncertaintyXGWillTrust ,695 Uncert2xGwilltrust4 <--- BehaviourUncertaintyXGWillTrust ,414 Uncert1xGwilltrust5 <--- BehaviourUncertaintyXGWillTrust ,744 Uncert2xGwilltrust5 <--- BehaviourUncertaintyXGWillTrust ,549 Uncert1xGwilltrust9 <--- BehaviourUncertaintyXGWillTrust ,724 Uncert2xGwilltrust9 <--- BehaviourUncertaintyXGWillTrust ,425
Appendix E (results from the Analysis – Chi square difference) Nested Model Comparisons
Assuming model Theoretical model to be correct:
Model DF CMIN P NFI
Delta-1 IFI Delta-2 RFI rho-1 TLI rho2 Model 1 - goodwill trust to 0 2 8,582 ,014 ,006 ,007 ,005 ,007 Model 2 - Calc Trust to 0 2 13,084 ,001 ,009 ,011 ,009 ,011 Model 3 Moderator = 0 2 9,211 ,010 ,006 ,008 ,005 ,007 Model 4 - opportunistic = 0 1 11,867 ,001 ,008 ,010 ,009 ,011
27 Model NFI Delta1 RFI rho1 IFI Delta2 TLI rho2 CFI Model 2 - Calc Trust to 0 ,737 ,692 ,930 ,915 ,927 Model 3 Moderator = 0 ,740 ,696 ,934 ,919 ,931 Model 4 - opportunistic = 0 ,738 ,692 ,931 ,915 ,927
Saturated model 1,000 1,000 1,000
Independence model ,000 ,000 ,000 ,000 ,000 RMSEA
Model RMSEA LO 90 HI 90 PCLOSE
Theoretical model ,049 ,029 ,066 ,508
Model 1 - goodwill trust to 0 ,052 ,033 ,068 ,427 Model 2 - Calc Trust to 0 ,053 ,035 ,069 ,369
Model 3 Moderator = 0 ,052 ,033 ,068 ,419
Model 4 - opportunistic = 0 ,053 ,035 ,069 ,369
