The authors thank Sarah Kaplan, Chirag Kasbekar, Rajiv Krishnan Kozhikode, Peter Liesch, Fabrice Lumineau, Anoop Madhok, Anita McGahan, Klaus Meyer, Anand Nandkumar, Harbir Singh, Kannan Srikanth, Aks Zaheer, and seminar participants at the Indian School of Business, University of Sydney, and University of Queensland for helpful comments on earlier drafts of this paper.
Appendix. Steps Taken Against Single-Respondent and Common-Method Bias RemedyImplementationProcedural remedies1Protecting respondent anonymity reduces respondents’ propensity to be socially desirable, acquiescent or lenient when crafting their responses (Podsakoff et al. 2003).In our cover letter, we guaranteed respondents complete anonymity.2Scale-reordering decreases the tendency of respondents to speculate on the relationship between the dependent and independent variables and intentionally match their responses to both (Parkhe 1993).Our survey placed the items of the dependent and independent variables far apart from each other.3Reducing item ambiguity is made possible by careful attention to the wording of items (Tourangeau et al. 2000).Our survey did not use vague concepts or double-barreled or complex questions, all of which reduce item ambiguity. Pretests with Indian managers allowed us to recognize and change some of the unclear words.Statistical remedies4Partial correlation adjustment involves using a marker variable to control for common method bias. This variable is typically a variable theoretically unrelated to at least one other variable in a study, preferably the dependent variable. The results cannot be attributed to common-method bias if any of the zero-order correlations that were significant before the adjustment continue to be significant after partial correlation adjustment (Lindell and Whitney 2001).In our study, tenure of the respondent, which was theoretically unrelated to many other variables, was used as the marker variable. Common method bias was not a severe problem in our study, as all our zero-order correlations that were significant before the partial correlation adjustment remained significant after the adjustment (Lindell and Whitney 2001).5Significant interaction terms suggest that single-respondent bias has not affected the findings (Brockner et al. 1997).Most of our interaction effects are significant. Respondents are unlikely to have intentionally theorized the moderated relationships when filling out the questionnaire when interaction effects are significant.6Triangulating survey data with archival data is typically employed to examine convergent validity of a construct (Keats and Hitt 1988, Parkhe 1993, Dhanaraj et al. 2004).(1) The percentage of equity reported by the respondents for the dependent variable correlated near perfectly with the archival data on equity percentage in the joint venture. (2) We were able to obtain data on the number of employees of the foreign partner from secondary sources for 66 Indian firms in our sample. Correlation between the data obtained from the survey and from secondary sources was high, 0.98 (p<0.000). (3) Similarly, correlation between number of foreign employees for a subset of 52 foreign firms obtained from archival data and that obtained from survey data (for 68 foreign firms) correlated highly (i.e., 0.97).7Harman’s one-factor test suggests that a single factor that accounts for most of the variance will emerge when all variables are entered together if a substantial amount of common-method bias is present in the data (Podsakoffet al 2003).After an unrotated principal-components-factor analysis on all the variables included in the survey, four factors with eigenvalues greater than 1.0 emerged, which together accounted for 59% of the total variance. Besides, the first (largest) factor did not account for a majority of the variance (19.71%).
