Data from a sample of Canadian manufacturing plants was collected through a mail survey conducted between September and December 2007, following a procedure inspired by Dillman (2000). The plant was chosen as the level analysis because it is generally where many environmental issues are evaluated and operational decisions are implemented. Using the National Pollutant Release Inventory (NPRI) and the Canadian Scott’s Directory, a sample frame of 503 randomly selected plants with more than 100 employees in the North American Industrial Classification Systems (NAICS) codes 332–335.1 A total of 94 usable responses were collected from that effort, leading to a response rate of 18.7%. Out of 94 plants, 22 have reported that they had more than 250 employees. Unfortunately due to missing data in some questionnaires, the sample was reduced to 85 for this study.
To minimize key-informant bias, each plant was contacted by phone prior to sending the survey in order to identify the individual most knowledgeable about the production, environmental and supply practices and performance (Kumar et al., 1993). Furthermore, although responses from multiple informants may have been preferred, the informants chosen for this study were positioned to make the assessment asked of them. In addition, the study was tested for common method bias, which could pose problems for survey research that relies on self-reported data — especially if the same person provides the data at the same time. One important concern in such cases is that common method bias may artificially inflate observed relationships between variables. To minimize common method variance, the dependent variables were placed after the independent variables in the survey; which helps to diminish, if not avoid, the effects of consistency artifacts (Podsakoff et al., 2003). A Harman’s single factor test was also conducted (Harman, 1967 and Shah and Ward, 2007). If common method variance existed, a single factor would emerge from a factor analysis of all questionnaire measurement items, or one general factor that accounted for most of the variance would result. The exploratory factor analysis revealed four factors with eigenvalues greater than 1.0 that accounted for 66.8% of the total variance. The first factor only accounted for 36.73% of the variance. These results suggested that common method variance was not a serious problem in our study.