4. Research Results
4.3 Regression results
regression results on the relationship between whether the focal firm is connected to other call firms and its decision to hold conference calls. I find that the coefficient of DLINK is significantly positive, suggesting that firms connected to other call firms through interlocked directors are more likely to hold conference calls. Table 12 presents the regression result on the relationship between the number of call firms linked to the focal firm and the focal firm’s decision to hold conference calls. The significant positive coefficient of LINK indicates that firms connected to more other call firms are more likely to hold conference calls. The coefficients of DLINK and LINK remain significantly positive after controlling for corporate governance, year and industry fixed effects. Other control variables are also consistent with my prediction. ROA, SIZE, MB and ISHARE are positively, while LEV, DSHARE and LSHARE are negatively associated with the decision to hold conference calls. The pseudo R-square increases after including control variables, year and industry fixed effects.
For equations (1C) and (1D), I perform Zero-Inflated Poisson regression to mitigate the concern over the large number of companies without holding conference calls. The empirical results are reported in Tables 13 and 14 respectively. Table 13 demonstrates a significantly positive relationship between the focal firm’s connection to other call firms and its frequency of holding conference calls. This implies that firms connected to other call firms tend to hold conference calls more frequently. In Table 14, a significant positive coefficient of LINK is presented, suggesting that when focal firms are connected to more other call firms, they are more likely to hold conference calls frequently. The results in both Tables 13 and 14 hold after
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
48
controlling for corporate governance, year and industry fixed effects. The results provided in Tables 11 to 14 support the first hypothesis.
‧
Table 11. The decision to hold conference calls and the focal firm’s link to other call firms
2. Variable definitions are in Table 1.
‧
Table 12. The decision to hold conference calls and the number of other call firms linked to the focal firm
2. Variable definitions are in Table 1.
‧
Table 13. The frequency of conference call and the focal firm’s link to other call firms NCALL = α + 𝛽1𝐷𝐿𝐼𝑁𝐾 + 𝛽2𝑅𝑂𝐴 + 𝛽3𝑆𝐼𝑍𝐸 + 𝛽4𝐿𝐸𝑉 + 𝛽5𝑀𝐵 + 𝛽6𝐷𝑆𝐻𝐴𝑅𝐸 +
Log-likelihood -4530.075 -4464.787 -4374.733 -4284.333 -4197.383 Pearson χ2 1279.476 1410.053 1590.161 1770.962 1944.861 Note:
1. *** indicates significant at 1% level, ** indicates significant at 5% level and * indicates significant at 10% level.
2. Variable definitions are in Table 1.
‧
Table 14. The frequency of conference call and the number of other call firms linked to the focal firm
Log-likelihood -4537.505 4468.940 -4375.935 -4285.788 -4196.952 Pearson χ2 1264.616 1401.746 1587.757 1768.052 1945.722 Note:
1. *** indicates significant at 1% level, ** indicates significant at 5% level and * indicates significant at 10% level.
2. Variable definitions are in Table 1.
‧
I further investigate whether different board positions among interlocking directors affect the decision to hold conference calls and the frequency of conference calls. I use Logistic regression in equation (2A) and Zero-Inflated Poisson regression in equation (2B). In equations (2A) and (2B), I use a dummy variable (INDILINK) that equals to 1 if the focal firm is connected to other calls firms through independent directors and 0 otherwise. For firms that have a 0 measurement for INDLINK, it represents two different situations: (1) the focal firm is connected to other focal firms through non-independent directors, and (2) the focal firm is not connected to any other call firms at all. Hence, to test Hypothesis 2, I first use the INDLINK as independent variable directly. The results are illustrated in Tables 15 and 16. To examine the effects of independent director links more precisely, I subsequently delete firms that have no connection to other call firms and perform analyses of equations (2A) and (2B). The results are presented in Tables 17 and 18.
Table 15 indicates a significant positive coefficient of INDILINK, implying that when the focal firms are connected to other call firms through independent directors, they are more likely to hold conference calls. The positive coefficient of INDLINK reported in Table 16 suggests that when the focal firms are connected to other call firms through independent directors, they tend to hold conference calls more frequently. After deleting firms connected to no other call firms, the sample size becomes 1,866 firm-year observations. The coefficient is positive and significant at one-tail 10 percent level in Table 17 and at two-tail 10 percent level in Table 18, indicating that the effect of interlocking independent directors on conference calls still exists when excluding firms that are not connected to other call firms. These results still hold after controlling for corporate governance, and year and industry fixed effects. Overall, the results in Tables 15 to 18 support Hypothesis 2.
‧
Table 15. The decision to hold conference calls and independent board link
CALL = α + 𝛽1𝐼𝑁𝐷𝐿𝐼𝑁𝐾 + 𝛽2𝑅𝑂𝐴 + 𝛽3𝑆𝐼𝑍𝐸 + 𝛽4𝐿𝐸𝑉 + 𝛽5𝑀𝐵 + 𝛽6𝐷𝑆𝐻𝐴𝑅𝐸 +
2. Variable definitions are in Table 1.
‧
Table 16. The frequency of conference calls and independent board link
NCALL = α + 𝛽1𝐼𝑁𝐷𝐿𝐼𝑁𝐾 + 𝛽2𝑅𝑂𝐴 + 𝛽3𝑆𝐼𝑍𝐸 + 𝛽4𝐿𝐸𝑉 + 𝛽5𝑀𝐵 + 𝛽6𝐷𝑆𝐻𝐴𝑅𝐸 +
Log-likelihood -4539.160 -4469.924 -4383.364 -4281.181 -4196.516 Pearson χ2 1261.307 1399.778 1572.898 1777.264 1946.595 Note:
1. *** indicates significant at 1% level, ** indicates significant at 5% level and * indicates significant at 10% level.
2. Variable definitions are in Table 1.
‧
Table 17. The decision to hold conference calls and independent board link, excluding firms without directors’ link
2. Variable definitions are in Table 1.
‧
Table 18. The frequency of conference calls and independent board link, excluding firms without directors’ link
Log-likelihood -1511.819 -1487.790 -1446.368 -1440.674 -1399.563 Pearson χ2 388.596 436.653 519.497 530.886 613.107 Note:
1. *** indicates significant at 1% level, ** indicates significant at 5% level and * indicates significant at 10% level.
2. Variable definitions are in Table 1.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
58