Following previous studies of Agle et al. (1999), this study defined stakeholder salience as the degree of received priority from our management team to competing stakeholder claims. Each respondent was asked to rate “this stakeholder group was highly salient to our organization”on a Likert scale of 1 to 5, with 1 denoting no salient at all and 5 a very strong salient. The classification of stakeholders divided into primary stakeholders, secondary stakeholders, and regulatory stakeholder (Buysse & Verbeke 2003; Clarkson, 1995). The primary stakeholders include employees, shareholders customers, and suppliers, while secondary stakeholders include competitors, international treaties and agreements,
environmental NGOs, and the media. The regulatory stakeholders include the national (and local) governments and local public communities.
Environmental performance
This study follows the research of Judge and Douglas (1998) to define environmental performance as a firm’s effectiveness in meeting and exceeding society’s expectation with respect to concern for the natural environment.
Meanwhile, this research classifies environmental performance into process measures (organizational systems and stakeholder relations) and outcome measures (regulatory compliance and environmental impact) (Ilinitch et al., 1998).
To obtain a more holistic view of the respondents’perceptions of organizational environmental performance, a separate measuring item was included in the survey.
Using the 1-5 scale, respondents were asked, “How would you rate the overall
effectiveness of your environmental performance?”with 1 denoting no effective at all and 5 a very strong effective.
The measurement of process performance included organizational systems and stakeholder relations. This study modified the measures from the study of Chen et al. (2006), there are five items for organizational systems. Following the research of Buysse and Verbeke (2003); there are three items for stakeholder relation. The measurement of outcome performance included regulatory compliance and environmental impact. Following the research of Ilinitch et al.
(1998), there are five items for regulatory compliance. This study modified the measures from the study of Chen et al. (2006), there are four items for environmental impacts. The measurements of observed variables are provided in Table 1.
4. Data analysis
4.1 Measurement model
This study followed Delmas and Toffel (2008) recommendations concerning the multi-dimensionality of eco-innovation adoption, stakeholder pressure and organizational environmental performance constructs by using a structural equation model. Structural equation models are able to incorporate estimates of measurement error into the study with the result of reducing possible bias in the parameter estimates via the use of multiple indicators. Therefore, LISREL 8.54 was used to perform the necessary- structural equation estimations in our study.
LISREL is an analysis procedure that combines path analysis with factor and multiple regression analyses (Joreskog et al., 2003).
This study use measurement model analysis to refine the measuring scale. The measurement model refers to the construction of latent variables from observable
items. In our case, we constructed nine latent variables from 39 items. We tested the measurement model by examining individual item reliability, internal consistency, and discriminant validity (see Table 1 and 2).
This study appraises construct reliability by calculating Cronbach’s alpha coefficients for each of the eco-innovation adoption, stakeholder pressure and organizational environmental performance. According to the test result of the reliability test of the nine major constructs, showed in Table 1. All nine of the scales had alpha’s greater than 0.70 and, therefore, were considered reliable (Churchill, 1991; Nunnally and Bernstein, 1994).
[INSERT TABLE 1 ABOUT HERE]
We tested internal consistency for each latent construct using two methods.
First, we calculated composite reliability for each latent variable by dividing (a) the squared sum of the individual standardized loadings by (b) the sum of the variance of their error terms and the squared sum of the individual standardized loadings (Fornell and Larcker, 1981). The values calculated for each of our latent variables exceed the threshold value of 0.70 (Nunnally, 1978), which suggests that our measurement model demonstrates adequate internal consistency. Second, we calculated the average variance extracted (AVE), which measures the amount of variance captured by the construct in relation to the amount of variance attributable to measurement error. For each latent variable, average variance extracted is calculated as (a) the sum of the squared item standardized loadings divided by (b) the sum of the variance of the error terms and the squared item standardized loadings. Convergent validity is judged to be adequate when average variance extracted is at least 0.50, which indicates that the variance captured by the construct exceeds the variance due to measurement error (Fornell and Larcker,
1981). As displayed in Table 1, the average variance extracted values are satisfactory for all constructs.
Discriminant validity refers to the extent to which measures of different constructs are distinct. Discriminant validity is deemed adequate when the variance shared between two constructs is less than the variance shared between a construct and its measures (Fornell, Tellis, and Zinkhan, 1982). The variance shared by any two constructs is obtained by squaring the correlation between them.
The variance shared between a construct and its measures is the average variance extracted. Discriminant validity was assessed by comparing (a) the correlations between a given construct and all other constructs to (b) the average variance extracted for the focal construct. Table 2 shows the correlation matrix for the constructs; the diagonal elements have been replaced by the square root of the constructs’ average variance extracted. Our constructs demonstrate adequate discriminant validity because these diagonal elements are greater than the off-diagonal elements in the corresponding rows and columns.
[INSERT TABLE 2 ABOUT HERE]
4.2 Structural model Goodness of fit
We find that the χ2 is statistically significant (χ2 = 232.53, degree of freedom
= 92, p = 0.000), which could suggest some misspecification of the model, although it is well recognized that this statistic is sensitive to sample size (Arbuckle and Wothke, 1999). We consider other structural diagnostics for the overall fit of the model that are not sensitive to sample size (Bentler and Bonett, 1980). The root mean squared error of approximation (RMSEA) (Steiger, 1990) is an estimate of the discrepancy between the original and reproduced covariance
matrices in the population. Cudeck and Browne (1983) suggested that an RMSEA of 0.05 represents a close fit and that RMSEAs of less than 0.08 represent a reasonable fit. In our model, the RMSEA of 0.06 is within the acceptable range.
Likewise, the 0.905 incremental fit index (IFI) (Bollen, 1989), the 0.904 Tucker-Lewis index (TLI) (Tucker and Tucker-Lewis, 1973), and the 0.903 comparative fit index CFI (Bentler, 1990) are all above the common threshold of 0.90 that designates an acceptable fit. These structural diagnostics indicate a very good relative fit of the proposed theoretical model to the underlying data. The standardized residuals in Q-plot roughly present normal distribution, so the goodness of fit of the model of this study is acceptable.