5.4 Results

The results of the meta-analysis are given in Table 5-2. In order to achieve independence between the samples, not all the reported correlations can be used at once for the calculation of an average correlation. Therefore, one single performance measure is selected from each study. As ROA and Tobin’s Q are the most frequently used performance measures, Panel A of the table shows correlations calculated with predominantly ROA correlations, if available. This means that if a study reports two performance measures of which one was ROA, this particular correlation was used for the calculation of the numbers in Panel A. The same technique is applied for Panel B with Tobin’s Q as the most important performance measure. If a study reports a ROA correlation as well as a Tobin’s Q correlation, the former is used for Panel A and the latter for Panel B. If ROA is not available, it has been replaced by another accounting measure such as ROE or return on investment. The replacement has been made in this order of importance. If no other accounting measures are available, market performance measures such as Tobin’s Q, market-to-book ratio or stock return correlations are used to replace it. This is also done in this order of importance. For the replacement of Tobin’s Q correlations, the other market performance measures are used first and thereafter the accounting measures.

Table 5-2: Results of the meta-analysis of the relationship between board independence and financial performance. Panel A consists of aggregated results of primarily ROA correlations. If ROA is not available for a certain study, another accounting-based performance measure was used; without such an accounting-based performance measure in a study, a market-based performance measure is used. Panel B does the same thing for Tobin’s Q. Panel C uses accounting and market-based performance measures only and does not use another measure if the required correlation is not available. The attenuation factor to correct r is 0.8. 95% confidence and credibility intervals that do not include zero are given in italics. Panel D consists of all the critical ratios (CR) and P-values (P-val) of the independence definition moderator of Panels A and B.

The table shows in Panel A an overall corrected correlation of -0.040 with a corrected variance of 0.006 for the ROA performance measure. The Tobin’s Q performance measure yielded a weighted average of the correlation of -0.046 and a corrected variance of 0.008, as shown in Panel B. The confidence intervals of both weighted average correlations do not include zero and the Z-score test statistic convincingly rejects the null hypothesis, which assumes that there is no relationship. This indicates that a significant relationship between board independence and financial performance does exist, according to this meta-analytic study. In contrast to conventional reasoning and beliefs about a positive impact of board independence on performance, which is mainly based on the agency theory, the average weighted correlations point to a negative relationship. However, these average correlations are calculated using board composition measures that are based on various definitions of independence, different board composition measures, different performance measures and a different geographical orientation. The 2K1 test statistic indicates the existence of heterogeneity within the outcomes of the studies in this analysis. Therefore, several moderators are tested for their impact.

Panels A and B use the information of all the samples in this study. Panel A uses primarily ROA information, but also other accounting-based performance measures and – if accounting- based information is not available –market-based information in some cases. The situation in Panel B is the other way around. Therefore, Panel C splits up accounting and market- based performance measures. If one study does not report correlation results for one of these two types of performance measures, it is excluded from that particular sample. This results in 31 studies that report at least one accounting-based performance measure and 25 studies that report at least one market-based performance measure. The corrected correlations for both performance measures show a corrected average correlation that is negative. The accounting-based measure is significantly negative, which is not the case for the market-based performance measure. The critical ratio does not provide any evidence for rejecting the null hypothesis of no difference between the two groups. This means that there is no significant difference between the results of accounting and market-based performance measures.

In order to investigate whether geographical focus has a moderating influence on the weighted average correlations, the sample is split into two subsamples: one with data only from the United States and one with data only from outside the United States. Studies with both American and non-American data were excluded from this analysis. The ROA correlations in panel A of the table show an average corrected correlation for the United States that is significantly negative; the average corrected correlation for studies with data from outside the United States is significantly positive. The critical ratio, which compares the two corrected correlations of the samples, does not conclude that a significant difference between the two geographical areas exists. The same conclusion can be drawn from panel B with Tobin’s Q correlations. The United States sample in this case appears to have a significant negative relationship between board independence and performance and the values for data from outside the United States are positive, albeit non-significant. The critical ratio in this case gives no indication of a significant difference between the United States and non-United States data samples. Panel C provides the results of the geographical moderator as well. Here, accounting information is significantly negative in the United States and significantly (p [amp]lt; 0.05) positive outside the United States. However, the results for market-based measures are not significant.

Although the critical ratios are not significant, the direction of the results indicates that accounting-based performance measures are negative for the United States and positive for outside the United States. These results are all significant and stronger than for market-based performance measures. This is contrary to the initial thoughts that investors would be more willing to pay a premium for board independence outside the United States than inside, because good corporate governance is less common outside the United States due to the lack of rule-based legislation and strong supervisors. Therefore, the relationship with market-based performance measures, which is based on the valuations of investors, was expected to be influenced more than in the case of accounting-based measures. Although the direction of the results tends to confirm the hypothesis, it cannot be concluded that shareholders do indeed have more confidence in the American corporate governance practices and value these companies higher. The limited number of studies on which this difference is based might be a reason for the lack of evidence for the moderating effect of some variables.

Another moderating effect tested in this study is the definition of independence. Five different categories are used. (1) Definitions that name at least NEDs in their definition of independence; (2) definitions that regard (any percentage) of shareholdings as a hindrance for independence; (3) definitions that have included the requirement that (prior) employees are not independent; (4) definitions that regard a business relationship as a problem for independence; and (5) definitions that consider ties with management as a reason to label a director as non-independent. The results in Panels A and B show a number of calculated average correlations that have become insignificant. Critical ratios of consecutive average correlations are given in Panels A and B, the critical ratios of all pairs are given in Panel D. Critical ratios do not indicate any significant differences. However, these groups are not mutually exclusive, because some studies have used more than one criterion in their independence definition. Therefore, the critical ratios do not measure the difference between totally different subgroups. However, the results are not convincing in the sense that more stringent definitions would lead to a stronger relationship between board independence and performance, because some explanations for an inability to discover that relationship is the lack of real independence. But this split might not be good either, as real (in fact) independence is hard to measure.

The last moderator is the board composition measure. Whether board composition was measured by the percentage of independent directors or outsiders or the percentage of insiders does not have any moderating influence on the relationship between board independence and performance.