To test for non-response bias, we separated the main respondent sample into two
groups, based on their return dates, and considered late surveys representative of
non-respondents (Lambert and Harrington, 1990; Armstrong and Overton, 1997).
We performed a univariate analysis of variance (ANOVA) and the results show no
statistical significant differences ( po5 percent) between early and late respondents in
terms of firm-specific characteristics and a subset of operations-related responses.
Thus, non-response bias is not an issue in this study.