that y=3, firm success. This approach is not appropriate for evaluating the effect of a dummy variable. Dichotomy
variables are analyzed by comparing the probabilities that result when the variable takes its two different values with
those that occur with the other variables held at their sample means. See Greene (2000) for further details.
Regarding the cut point or threshold parameters interpretation, Daykin (2002) suggested that first, if the
dependent variable measure shows that most firms are in either one extreme or the other (for example, very poor
performance or very good performance), one would expect that the threshold would be tightly bunched in the middle of
the distribution, very close to one another. If, on the other side, firms appear to be more balanced, it would be expected
that the cut point be widely dispersed.
Second, it could happen that the cut points adjust to the questionnaire wording, in order to obtain the dependent
variable, and might be doubtful and hard to understand. If this is the case, one would expect that the middle threshold to
be far apart, reflecting an indifference on the part of the respondents who may not understand the question. This is
important, as the questionnaire wording can be improved between studies, and a contraction toward the middle might be a
sign of improvement. In addition to probit regression analysis, descriptive statistics and test of mean and proportion
differences between successful, mediocre and failed firms were also run.
4 Results
4.1 Control variables
Control variables that affect performance include size (number of employees), age of the business, and industry (Lussier
and Pfeifer 2001; Reynolds 1987; Shane 1996). Small firms are more likely than large firms to fail. In the sample, the
average size, as measured by number of employees, of the failed firms was 17 (s.d. 31), of the mediocre firms was 17
(s.d. 25) and 30 (s.d. 51) for successful firms. The sample was based on SME and the means of successful, mediocre, and
failed firms is not significantly different at the .05 level. Therefore, firm size should not bias the results. The age of a
business also affects performance because new firms have a higher probability of a poor performance than established
businesses. New businesses often lose money. However, the mean age, with similar medians, of failed businesses were
14.24 (s.d. 11.3) years, 15.2 (s.d. 13.7) years for mediocre firms and 15.16 (13) successful companies. Therefore, all
groups are mature and the mean difference is not significant at the .05 level. Thus, age should not bias the results.
Industry can also affect success, as more service and retail firms tend to have higher failure rates (Lussier 1996a, 1996b).
However, all industry sectors were included in the sample; also, Chi-square testing found no significant differences
between the successful, mediocre, and failed businesses by industry. Thus, there are relatively equal numbers of firms that
performed well, mediocre or poorly by industry, and industry should not bias the results.
7
Convention