Employment capital, board control, and the problem of misleading disclosures *.

The independent measures included multiple indicia of performance pressure, executive power, and board control. For performance pressure, we included each firm's return on assets, total market return (price appreciation and dividends), and bankruptcy risk for the year preceding the restatement year. The return measures indicate the extent of positive performance pressure, which, as theorized above, may lead to increasingly high performance targets. Conversely, bankruptcy risk was included in order to capture the notion of negative performance. We utilized Altman's Z score (Altman, 1983, 1988) as a proxy for this construct, a discriminant function that predicts the likelihood of subsequent bankruptcy based on a number of preceding financial indicators. As developed in Altman's work, lower Z scores indicate higher bankruptcy risk.

CEO tenure and equity ownership percentage, the measures of CEO power, were calculated as the CEO's years of service in that capacity and shares owned divided by total shares outstanding. In the case of the latter, an additional squared term was calculated in order to test the hypothesized curvilinear relationship (Finkelstein and Hambrick, 1989; McConnell and Servaes, 1990).

Board power, as suggested by the discussion above, involved the use of three measures. Outside director percentage was calculated as the ratio of unaffiliated outsiders to total board members, and equity ownership was calculated by finding the average shares owned by directors, both of which measures follow previous research (e.g., Hoskisson et al., 1994; Johnson et al., 1996). Average board tenure was entered in order to address the power of the board relative to that of the CEO (Golden and Zajac, 2001).

Four controls were employed: total assets, number of employees, CEO contingent compensation and CEO duality. Firm size has been a generally accepted control variable in strategic management research (e.g., Singh, 1986). Although assets and employees were matching criteria, we viewed it as important to enter these values as controls in order to protect against any residual between-pair variance and the influence that these variables may have on any of the independent measures. We included a measure of contingent compensation, calculated on the basis of each CEO's proportion of options and bonuses to salary, which is analogous to previously utilized measures of pay structure (Carpenter and Sanders, 2002). This variable permits us to capture and control for the extent to which a CEO would have a direct financial incentive to boost reported performance. Finally, duality, the simultaneous vesting in one individual of the positions of CEO and board chair, also intuitively relates to the balance of power between the board and the CEO. However, the existing literature has failed to establish unambiguous effects for duality (e.g., Johnson et al., 1996; Mahoney et al., 1997). Therefore, we did not believe a clear theoretical justification for its inclusion as an independent variable existed, and instead decided to include it as a control for any potential power effects.

Analytical Method

Given the use of a dichotomous dependent variable, logistic regression represented the most appropriate methodology for testing the hypothesized effects. This technique offers the benefit of avoiding violation of standard multiple regression assumptions, given the presence of a dichotomous dependent variable. Testing followed a hierarchical procedure, starting with a regression model including only the controls. A second regression model, including the control and independent variables, was estimated and compared to that obtained in the control model. A significant increment over the control model supports the validity of the main effects model and the interpretation of the individual coefficients (Cohen and Cohen, 1983).

After testing for main effects, we constructed two additional regression equations, the first of which added product terms calculated by multiplying the board variables found to be significant from the baseline model with each of the significant independent variables measuring performance and power. The second model incorporated the squared CEO ownership term to test for curvilinearity. Analysis began with confirmation that each additional model significantly incremented the main effects model and that the interaction and squared term models significantly incremented the main effects model (Cohen and Cohen, 1983; Jaccard et al., 1990).

RESULTS

Following the procedures outlined above, we obtained results providing at least partial support for most of our hypotheses. In this section, we will discuss the results and provide additional methodological detail for each hypothesis in turn. We begin by noting that Table 1 presents the correlations and descriptive statistics for the variables used in this study, and Table 2 presents the separate regression results for each of the models employed in the study. As shown in Table 2, the control model explains less than 10% of total variance (Nagelkerke [R.sup.2] = .08), although the contingent compensation variable is positive and significant (b = .362, p < .05). Model 2, the main effects model, significantly incremented Model 1 (Nagelkerke [R.sup.2] = .384, F = 12.769, p < .001), thereby supporting the validity of interpreting individual coefficients.

Hypothesis 1 predicted that employment capital risk, resulting from either bankruptcy risk or high firm performance, would increase the likelihood of misleading disclosures. Market return was positively associated with restatement activity (b = .011, p < .01), while return on assets, although positively signed as expected failed to attain statistical significance. Hypothesis 1 thus receives partial support, at least in the case of market return.

Hypotheses 2 and 3 were concerned with the relationship between CEO power and misleading disclosures, respectively defined in terms of tenure and ownership. As indicated in Model 2, Table 2, CEO tenure is not significantly related to the issuance of misleading disclosures (b = .001, n.s.), thereby failing to support Hypothesis 2. CEO ownership, on the other hand, is positively and curvilinearly related to misleading disclosures, as shown in Models 2 and 4 of Table 2. The main effect coefficient (b = .037, p < .05) indicates an entrenchment effect as ownership increases, and the test of curvilinearity in Model 4 establishes the U-shaped relationship (b = .003, p < .05) consistent with increasing disclosure manipulation at either low or high levels of ownership. Thus, Hypothesis 3 is supported.

Finally, hypotheses 4a and 4b predicted that board control would moderate the foregoing relationships by increasing the likelihood of misleading disclosures when control was low and decreasing misleading disclosures when control was high. From Model 2, we determined that a significant main effect existed only for average board tenure. Neither outside board percentage nor equity ownership attained statistical significance. Using board tenure as the basis for the test of moderation, we calculated separate product terms by multiplying board tenure by both CEO ownership and market return, the significant main effects relating to employment capital risk and CEO power. These product terms were entered in a separate regression, reported as Model 3.

As can be seen from Table 2, Model 3 significantly increments the main effects model (Model 2), permitting individual coefficient interpretation (F = 15.299, p < .001). The product term relating to CEO ownership was not significant, but the board tenure-market return interaction attained a high level of significance (b = .003, p < .01). We can conclude, therefore, that CEO ownership exists as a significant main effect unaffected by board influence.

Having established the existence of significant interaction terms, we evaluated the form of the interaction at high and low levels of the independent variables (Cohen and Cohen, 1983; Jaccard et al., 1990). Decomposition of the initial results revealed that, as predicted, board tenure was lower among restating firms with higher returns. In other words, a positive relationship between firm performance and restated earnings occurred only when board tenure was significantly lower than that found at firms that did not restate earnings. Unexpectedly, however, the converse held in cases of low firm returns: restating companies had higher levels of board tenure than was true of the matching companies. On balance, then, Hypothesis 4a receives partial support with respect to the effect of board tenure among firms with high market return levels.

DISCUSSION AND CONCLUSIONS

This research sought to extend our understanding of the phenomenon of misleading disclosures. We relied upon agency theory (e.g., Jensen and Meckling, 1976) and prospect theory (e.g., Kahneman and Tversky, 1979) to develop and test a model incorporating executives' motivations to protect their employment capital and to exploit their power. We also suggested that the existence of board control would moderate the relationships between these factors and the incidence of misleading disclosures. In this section we will discuss the research implications of our findings, managerial implications, and the limitations of this study and directions for future research.

Research Implications