The two principal purposes of this study are to: (1) explicitly examine and test whether the assumptions underlying the theoretical development of analogous parametric and nonparametric statistical data analysis techniques are satisfied with a sample size of 199 ventures, and (2) examine the usefulness of analogous parametric and nonparametric techniques for exploring relationships among new venture strategies and venture performance. In this context, this study compares and contrasts the findings generated by these analogous techniques to determine the robustness of parametric procedures when the underlying assumptions are violated.

This study is important because most prior quantitative research studies in the field of entrepreneurship have utilized parametric statistical data analysis techniques without explicitly checking or fully discussing whether the assumptions underlying the theoretical development of such techniques were fully satisfied with the data utilized in such studies.

STATISTICAL DATA ANALYSIS ISSUES IN ENTREPRENEURSHIP RESEARCH

The majority of the current quantitative research in social sciences and entrepreneurship utilize parametric statistical data analysis techniques for evaluating data and testing hypotheses (Bygrave, 1989; Krauth, 1988). Krauth stated, "This seems strange, for most researchers will probably remember from their statistics courses that these procedures will yield correct statistical results only if certain very restrictive [italics added] assumptions are fulfilled" (p. v). Krauth noted that examinations of real data reveal that the assumptions underlying parametric techniques are violated in most instances. Kemery discussed the tendency of most prior entrepreneurship studies to either overlook or completely disregard methodological issues (as cited in Smith, Gannon, & Sapienza, 1989). Furthermore, Bygrave (1989) and Hofer and Bygrave (1992) stated that parametric techniques are inappropriately used in most entrepreneurship research studies.

__Distributional
Assumptions Underlying Statistical Data Analysis Techniques__

The assumptions underlying the theoretical development of the parametric statistical data analysis techniques employed in this research require that the dependent variables have: (1) normal distributions; (2) equal variances for each of the sampled populations (3) symmetric distributions; (4) continuous distributions; and (5) independence of observations.

By contrast, the assumptions underlying the theoretical development of the parametric statistical data analysis techniques employed in this research require that the dependent variables have: (1) measurement on at least an ordinal scale; (2) continuous distributions; and (3) independence of observations (Daniel, 1990; Gibbons, 1985).

__Prior
Research on Distributional Characteristics of Dependent Variables__

Friedman (1937) disputed the likelihood that the normality and equality of variance assumptions would be satisfied with "real world" data. stated, "This is especially apt to be the case with social and economic data, where the normal distribution is likely to be the exception rather than the rule" (p. 675). Gibbons (1985), Krauth (1988), and Daniel (1990) also concluded that there was minimal likelihood that "real world" data would conform to the stringent assumptions required for the valid usage of parametric techniques.

There have been a number of studies in finance and accounting that have examined the underlying distributions of the variables of interest. In general, these studie3s have found that most financial data sets do not support the assumptions required for the valid usage of parametric statistical data analysis techniques. Deakin (1976) investigated the normality and variances of 11 with sample sizes of 454 to 1114. He found that only one of the 11 ratios satisfied the normality assumption at a .05 level of significance. Deakin also found that square root and lognormal transformations of the data did not result in substantive differences in the distributions of the 11 ratios he examined. Finally, Deakin found that more recent data, with larger samples, were less likely to be normally distributed. Frecka & Hopwood (1983) also examined the distributional properties of commonly used financial ratios. Their study covered a 30 year period of time for each fiscal year of manufacturing firms. With sample sizes ranging from 346 to 1243, they found that only the working capital/total assets ratio was normally distributed for the majority of the 30 year period they examined. They also found that transformation of the data using a square-root transformation had little impact on the results obtained when compared to the raw data analyses.

Based on the mass of evidence of prior studies regarding the nonnormality of financial ratios, it is expected that the distributions of the change in sales and shareholder value created measures of new venture performance examined in this study will have nonnormal distributions:

H1: The dependent variables examined in this research will have nonnormal distributions.

There has been relatively little examination of the relationships among variances in studies utilizing measures of economic performance. Friedman (1937) found that the assumption of homogeneous variances was violated with the data he examined. More recently, Deakin (1976) and Fieldsend, Longford, and McLeay (1987) found evidence of departures from the assumptions of equality of variances. Thus, in tests between two or means, it is expected that the sampled populations of the dependent variables will have unequal variances:

H2: All sampled populations of the dependent variables examined in this research will have unequal variances.

This research examines economic performance variables measured on a ration scale, which typically have continuous distributions. Thus, it is hypothesized:

H3: The dependent variables examined in this research will have continuous distribution

Top of page | Chapter Listing | Return to 1997 Topical Index |

© 1997 Babson College All Rights Reserved

Last Updated 03/03/98