Analysis

Initially, descriptive statistic and binomial correlations were prepared to give an overview of the data. The results are presented in Table One.

TABLE 1

Descriptive Statistics and Correlations

 Variables N Mean Std. Dev. Fail? Fund Age Loc. # Emp Sex Occ. Fail? 23 1 0.5022 0.5011 1.000 Funding 13 8 0.4420 0.8196 -.118 1.000 Age 10 6 4.0000 5.3666 -.173 .126 1.000 Location 22 6 1.2800 0.4500 *-.135 .159 -.122 1.000 # Emp. 18 4 1.5200 0.9700 -.094 ** .347 .037 .085 1.000 Sex 22 4 1.5313 0.5264 ** -.181 .166 .050 ** .371 * .176 1.000 Occupation 19 0 2.0105 0.8482 ** -.403 ** .244 .085 ** .388 .132 ** .335 1.000

To test the hypothesis that not all firm disappearances are failures, Chi–Square Goodness of Fit test was used. The Chi–Square clearly rejects the null hypothesis that all disappearances are failures, as can be seen in Table Two.

TABLE 2

Chi-Squared Goodness of Fit

 Disappearance Observed N Expected N Residual Test Statistics Closed 115 0 115 Chi–Square 5724936 Failed 116 231 -115 df 1 Total 231 231 Asymp. Sig. *** p<0.000

Once the first hypothesis was supported, a Kruskal–Wallis one–way analysis of variance test was employed to show the significance of the association between each of the independent variables and the outcomes of failure or other type of closure. These results are presented in Table Three.

Discriminant analysis was used to test the effect and significance of the independent variables on whether a disappeared firm had actually failed or had closed for other reasons. The independent (discriminant) variables were entered in a stepwise method, selecting the variable that resulted in the smallest Wilkes’ lambda, because the relationship between variables was not predicted by prior research. While there was some concern about the assumption of the multivariate normality of the variables, other assumptions of the test were met. Box’s M, insignificant at p=.9139, did not allow the rejection of the null hypothesis that the SSCP matrices were equal, therefore homogeneity of variance among the variables was confirmed.

Kruskal–Wallis One–Way Analysis of Variance

 Independent Variables Dependent Variable (Disappearance) N Mean Rank Test Statistics Source of Funding Closed Failed Total 49 89 138 74.86 66.74 Chi–Square df Asymp. Sig. 2.465 1.000 p=0.116 Age of Business Closed Failed Total 38 68 106 58.84 50.64 Chi–Square df Asymp. Sig. 1.773 1.000 p=0.183 Location of Business Closed Failed Total 110 116 226 121.56 105.85 Chi–Square df Asymp. Sig. 5.352 1.000 * p=0.021 Number of Employees Closed Failed Total 75 109 184 97.87 88.80 Chi–Square df Asymp. Sig. 1.777 1.000 p=0.183 Sex of Owner(s) Closed Failed Total 108 116 224 123.10 102.63 Chi–Square df Asymp. Sig. 7.348 1.000 ** p=0.007 Occupation after Closure Closed Failed Total 83 107 190 119.16 77.15 Chi–Square df Asymp. Sig. 30.807 1.000 *** p<0.001

Discriminant Analysis: Classification Results

 Actual Group No. of Cases Predicted Group 0 Predicted Group 1 Closure (0) 83 50 = 60.2% 33 = 39.8% Failure (1) 107 19 = 17.8% 88 = 82.2% Ungrouped Cases 6 02 = 33.3% 04 = 66.7% Percent of "grouped" cases correctly classified: 72.63%

A total of 95 cases were included in the analysis (cases with at least one missing variable were eliminated). The resulting discriminant function was significant (p<0.001) and explained more than 22% of the variance in firm disappearance based on a cannonical correlation of 0.473. The overall hit rate was 72.63%. These results are in Table Four. The discriminant function is:

Di = -2.59 + 0.71(JOB)

As can be seen, only independent variable entered into the equation was the occupation of the owner after the business disappeared. Interestingly, we found that the owners of closed (rather than failed) firms were more likely to have started other firms or to be employed elsewhere, while the owners of failed firms were not. This is consistent with the predictions of the efficiency wage theory and is explained more fully in the next section.