臺大管理論叢 NTU Management Review VOL.28 NO.3

157 NTU Management Review Vol. 28 No. 3 Dec. 2018 and Tate, 2005). CEO ownership was measured by the ratio of number of shares owned by the CEO to total outstanding shares in the fiscal year. Finally, about the control of CEO tenure, the age and tenure of CEOs seem to be highly interrelated, but recent studies indicate that they are not closely linked with each other (Barker and Mueller, 2002; Musteen et al., 2006; McClelland et al., 2012). In other words, CEO age and tenure may have different effects on corporate misconduct. We thus measured CEO tenure by the total number of years in the CEO position at the focal firm. 4. Results Table 1 shows the means, standard deviations and correlations of all the variables. Similar to the findings of prior studies (Beitel, Schiereck, and Wahrenburg, 2004), the correlation between ROE and ROA was also found to be highly correlated. The other correlation coefficients are within acceptable limits, suggesting that multicollinearity is not a problem in this work. Poisson regression and negative binomial regression analyses are widely used to test hypotheses when the dependent variable is an event count variable. However, negative binomial regression is used to estimate count models when the Poisson estimation is inappropriate due to over-dispersion. By comparing the mean and standard deviation of the dependent variable, the value of standard deviation is obviously greater than the double value of the mean, indicating that a negative binomial regression model could be applied to handle the effect of over-dispersion. We therefore tested our hypotheses by using negative binomial regression analyses. Considering the data structure of our sample (i.e., panel data), a series of longitudinal negative binomial regression models were thus adopted to test our hypotheses. Before performing regression analyses, a Hausman test was executed to compare random- and fixed-effects estimations of the model. This test rejected the appropriateness of using random-effects estimation (Prob>chi2 = 0.000).