Data Analysis Techniques for Testing the Influence of New Venture Strategy on Performance.
Gibbons (1985) provided guidelines for the selection of the appropriate type of parametric and nonparametric test procedures for testing hypotheses regarding differences in location parameters when using independent samples. Table 1 provides a summary of these guidelines.
Guidelines for Selecting Parametric and Nonparametric Location Parameter Tests
Fisher's least significant differences (LSD) procedure is recommended by the SAS Institute (1989) as a multiple comparison technique which minimizes the comparisonwise error rate. J.H. Reeves (personal communication, October 4, 1994) of The University of Georgia Department of Statistics recommended raking the dependent variables of interest, and then using the Fisher's least significant differences procedure as a nonparametric multiple comparison technique.
This study expected to find nonnormal distributions for the
dependent variables of interest examined in this study. Table 2 contains the critical values of the
tests statistics which are required to accept the null hypothesis
of normality, and the actual statistics computed.
The entirety of these three statistical measures provide strong support for the hypothesis that the data examined in this study are not normally distributed. In fact, extreme departures from normality were found for both of the venture performance variables examined. This study also utilized: (1) normal probability plots; (2) histograms; and (3) box plots to examine the underlying distributions of the measures of performance examined in this study. The results of these plots provide further strong support for the rejection of the null hypothesis regarding the assumption of normality vis-à-vis the two dependent variables examined above.
Test Statistics for Assessing the Normality Assumption
In short, these results indicate that the use of parametric tests of location are inappropriate for this study, and may be inappropriate for other new venture performance studies as well, using the two performance measures examined in this study due to the significant departures from normality for these measures.
Equal Variance Assumptions
This research also expected to find that the assumption of equal variances for the sampled populations examined in this study required for the appropriate usage of parametric tests of location would be violated by the data utilized. The results of the pairwise testing for equality of variances for the sampled populations of the scope variable are shown in table 3.
Equality of Variance Test for Scope Classifications
The results of the equality of variance tests for the sampled
populations for the strategic scope variable provides strong
support for the rejection of the null hypothesis of equal
variances for the sampled populations for both measures of
economic performance examined in this study.
Similar results were generated for the sampled populations of the strategic weapons variable for both performance measures. Thus, there is strong evidence that the variances of the sampled populations examined in this study are not equal for both measures of new venture performance.
Continuous Distribution Assumption
This study expected that the two measures of economic performance examined in this study, which are assessed on a ratio scale, would have continuous distributions. An examination of the ranks of the two economic performance variables utilized in this research supported this hypothesis/ Therefore, the assumption of continuous distributions does not preclude the valid usage of the parametric or nonparametric tests of location in this research.
Discussion of Variable Transformations
This research did not transform the data utilized in this study, as Deakin (1976) and Frecka and Hopwood (1983) found that transformations of performance variables had little impact on the normality of the distributions or the equality of variances. Thus, there is little reason to expect that transformations would significantly impact this study's dependent variable distributions.
Summary of Findings on Distributional Characteristics
The results of the three hypotheses regarding the distributional characteristics of this study'' dependent variables provide strong support that the normality and equal variance assumptions required for the appropriate usage of the parametric statistical techniques employed in this research are violated by the data utilized in this study. Thus, on distributional grounds alone, there is significant evidence that the parametric statistical tests of location are not valid for this research. By contrast, the assumptions underlying the theoretical development of the nonparametric tests of location employed in this research were satisfied by this study's data.
Unfortunately, the vast majority of research in the field of entrepreneurship has utilized parametric statistical techniques, without ever explicitly checking or fully discussing whether the assumptions underlying the valid usage of such techniques are indeed satisfied by the data utilized.
© 1997 Babson College All Rights Reserved
Last Updated 03/03/98