271 /304
271
臺大管理論叢
第
28
卷第
1
期
Table 4 presents the contingency tables for the different regime samples. Panels A
and B show that the Type II error rates are 97.6% (568/582) and 79.0% (1,004/1,271) for
the SOX 302 and SOX 404 samples, respectively. The Type I error rates are 0.3%
(7/2,140) and 2.4% (204/8,634), respectively. Moreover, Panels C and D reveal that the
Type II error rates are 71.5% (512/716) and 88.6% (492/555) for the AS2 and AS5
samples, respectively. The Type I error rates are 3.7% (132/3,529) and 1.4% (72/5,105),
respectively. Additional information about the ratios of effective ICFR disclosures among
the restatement and non-restatement samples is provided in Table 5. Panel A reveals that
restatement companies subject to SOX 404 are significantly less likely to conclude that
their internal controls are effective, compared to restatement companies subjected to SOX
302 (79% versus 97.6%, p-value < 0.001). Non-restatement companies under the SOX
404 regime, however, are also less likely to conclude that their internal controls are
effective, compared with non-restatement companies under the SOX 302 regime (97.6%
versus 99.7%, p-value < 0.001). The results in Panel A support our H1a and H1b,
indicating that the enactment of SOX 404 reduces the likelihood of Type II errors but
increases the likelihood of Type I errors in public ICFR disclosures.
Panel B presents the ratios of effective ICFR disclosures among the restatement and
non-restatement companies for AS2 and AS5 samples. Restatement companies are
significantly more likely to conclude that their internal controls are effective in the AS5
period than in the AS2 period (88.6% versus 71.5%, p-value < 0.001). This result indicates
that the implementation of AS5 results in higher Type II error rates of public ICFR
disclosures. However, non-restatement companies are also significantly more likely to
conclude that their internal controls are effective in the AS5 period as in the AS2 period
(98.6% versus 96.3%, p-value < 0.001), which supports our H2b and indicates that the
implementation of AS5 has resulted in a lower Type I error rate.