COMPARISON OF NONPARAMETRIC AND PARAMETRIC RESULTS
Table 10 provides a comparison of the results generated by analogous nonparametric and parametric statistical data analysis techniques for examining the influence among various new venture strategies and new venture performance.
The nonparametric results of this research provided strong support for differences in new venture performance based on the strategy which such ventures pursued. As shown in Table 10, the nonparametric procedures produced statistically significant findings for 35 of the 40 tests of hypotheses regarding differences in new venture performance based on the new venture's strategy.
By contract, parametric procedures generated results substantially different from those generated by the nonparametric procedures. Thus, only 7 of the 40 parametric procedures generated statistically significant results for differences in new venture performance based on the strategy a new venture pursues.
Overall, the nonparametric procedures produced statistically significant findings for 31 tests which the parametric procedure failed to find statistically significant differences as shown in Table 10. In addition, the parametric procedures produced four apparently statistically significant findings that were not supported by the nonparametric results, and that additional analysis showed were spurious.
Comparing the results obtained by these analogous nonparametric and parametric procedures indicates that the nonparametric procedures were far superior to the parametric procedures in their ability to differentiate between successful and unsuccessful ventures based on the strategies pursued by such ventures. Put differently, had one naively assumed that parametric statistical data analysis techniques were appropriate with this sample size of 199 new ventures, there would have been little, if any, evidence that the strategies of these ventures influence their economic performance. Overall, these results provide strong support that the parametric statistical data analysis techniques are not robust with regard to the validity of the inferences produced when the assumptions underlying the theoretical development of such techniques are violated by the data utilized. Thus, these results strongly support the need to check the underlying distributional characteristics of the variables studied in entrepreneurship research, even with large sample sizes, before selecting the types of statistical data analysis technique(s) to be utilized.
Comparison of Nonparametric and Parametric Results for New Venture Strategy Classes
Space limitations do not permit a detailed analysis of the reasons for the "failure" of parametric statistical data analysis techniques in this paper. This issue will be explored in depth in a future paper. For now, we would just note that the mean may not be an adequate measure of central tendency for nonnormal, nonsysmmetric distributions.
SUMMARY AND CONCLUSIONS
This research found that the unquestioning use of parametric statistical data analysis techniques is inappropriate for most research on the determinants of new venture performance. In particular, this study found that the distributional assumptions required for the appropriate usage of parametric procedures were substantially violated by the data utilized in this large sample.
This study also found that the failure to satisfy such assumptions has a substantial impact on the number and robustness of the findings generated using parametric statistical data analysis techniques. Put differently, parametric procedures failed to identify approximately 90% of the statistically significant findings that were identified using appropriate nonparametric procedures, because the assumptions underlying the theoretical development of parametric were violated. In addition, nearly 50% of the statistically significant findings generated by parametric procedures were not consistent with the nonparametric findings, and therefore probably spurious. In short, the unquestioning (and probably unjustified) use of parametric statistical data analysis techniques may be responsible for the lack of robust findings in many of the studies done in the field of entrepreneurship. Consequently, future entrepreneurship research should utilize nonparametric statistical data analysis techniques if the stringent assumptions underlying the appropriate usage of parametric procedures are not satisfied by the data utilized in such studies.
© 1997 Babson College All Rights Reserved
Last Updated 03/03/98