While much of the finance literature focuses on the oversight role of boards of directors, there is an increasing focus on the broader role that boards play in the operational and strategic management of the firm (Hillman et al., 2009; Masulis et al., 2012). . Closeness assesses how central a board is in a network by looking at the distance between a company and other companies in the network, measured by the shortest number of connections between two companies. Proximity is defined as the sum of the reciprocal of the shortest distance between company i and all other companies in the network (Freeman, 1979), i.e.

Our fourth measure is the Eigenvector, which captures the quality of the firms with which a firm is associated. Eigenvector is defined as the sum of Firm i's first-degree connections to all other firms (δ(i,j)) in the network, weighted by the connectedness of the firms to which it is connected, i.e. Tobin's Q has been widely used to measure a company's performance by expressing a ratio of the market value of a company.

## Empirical Design

12 We also control for various corporate governance characteristics known to be related to firm performance. Likewise, since busy directors may be ineffective monitors, we also control for director busyness by creating a dummy variable (BUSY) that equals one if a board has over 50% of directors with three or more many directorships in the same year (Andres et al., 2013; Fich & Shivdasani, 2006). To capture the experience the board has in monitoring a firm, we control for tenure (TEN), defined as the average length of time directors have served on the board (Horton et al., 2012) which is a proxy for management.

Finally, we control for CEO duality with a dummy variable (DUAL) that equals 1 when the CEO is also chairman and 0 otherwise.

## Data

We include a measure of board size, defined as the total number of board members sitting on the board (Core et al., 1999), since larger boards may be less effective monitors (Yermack, 1996) and smaller boards may be less connected. On average, about 70% of the firms in the sample were connected to the largest network, compared to Larcker et al. 2013), who report an annual average of 72% for US firms, although this percentage is also declining over time. We observe that both Degree and Closeness decrease over the sample period, consistent with Omer et al.

Taking the logistic transformation of AGE is also consistent with previous studies (e.g. Larcker et al., 2013).

## Results

### Preliminary analysis

To offset the skewed distributions of AGE, LEV, BSIZE, and TEN, we take the log of the variables for the regressions. In addition, we take the log (1+ FEM) (Wooldridge, 2009) to ensure that observations with zero committee members are included in the analysis. We then investigate the relationship between firm performance and connectivity by sorting firms into portfolios based on their connectivity scores, and then compare the portfolio averages of our firm performance and connectivity metrics.

We then test the significance of the differences in means between the fixed performance measures of the high and low portfolios. However, we see an average Betweenness of 0.0 for the bottom two quintiles, which suggests that at least 40% of the sample is in no way able to control or help the flow of information within the network. This suggests that at least 40% of the sample are either isolated firms or sit on the edges of the network, connected to only one connection to the main network.

Examining the average firms' performance measures by affiliation quintile, we observe that there appears to be a positive correlation between ROA and the four measures of affiliation, but a negative association with Tobin's Q. This may indicate that isolated firms that make the majority of the bottom quintile in most years, perform slightly better in terms of operating results, but worse when looking at market value of book assets than those companies in smaller networks and companies on the fringes of the main network. Specifically for three of the four connectivity measures there is a significant negative correlation, with Quintile 5 on average having lower returns than Quintile 1, Betweeness is insignificant.

The univariate results, including the correlation coefficients and the quintile analysis, suggest that there is evidence of a mixed relationship between connectivity and firm performance. There appears to be a positive relationship between ROA and connectivity, while TQ is lower for highly connected companies.

### Multivariate Analysis

*Analysis of the Overall Effect of Connectivity*

Among the board structure characteristics, there is a strong relationship between average board tenure (TEN) and firm performance. The results in Table 7 support Omer et al. 2013) who find a board's connectedness, when measured by Closeness, appears to be detrimental to firm performance. The results are consistent with the univariate results for Degree and Closeness and again support the correlation coefficients in Table 4.

Panel B of Table 9 reports the results for testing the model defined in equation 7, with Eigenvector as the connectivity variable. So far, the results provide significant evidence that board connectivity is negatively associated with company performance, both in the current year and in future year results. Using equation (7) and replacing the individual connectivity measure with the principal component score (CONN), we run OLS regressions of each firm performance.

The coefficient of CONN is negatively and significantly related to each of the current-year and subsequent-year firm performance variables for all but TSR. We follow Horton et al. 2012) and includes a lag of the firm's performance measure, FPi,t-1, as an independent variable to control for the firm's performance impact from the board selection process and director preferences. However, the results do not suggest that prior firm performance influences the negative relationship between firm connections and firm performance, as we reported in previous tests.

Of particular note, the results for both the main model and the dynamic model show a negative relationship between employee boards and firm performance. Finally, the inclusion of LDV noticeably improves the fit of the model based on the adjusted R2s, with the exception of the results for TSR, which increases by only a small amount.

### Robustness Tests

*Alternative Models**Structural Break Tests*

In Section 4.2 we used equation (7) including a lagged dependent variable, specifically the lag of the firm's performance measure, to control for unobserved factors, firms' board selection process and director preferences, to ensure robustness of the negative relationship between board bonding. and strong performance. Specifically, the network literature argues that if a company views connectedness as a useful governance mechanism, firms can increase connectedness by appointing more connected directors to a board to improve firm performance. To test this, we look at changes in linkage to changes in firm performance where we measure the change in linkage and measures of firm performance between years t0 and t-1, and also t-1 -t-2 for firm performance (Andres 2013 Yermack, 1996).8 Following previous research, we include changes in control variables (t0 -t-1) that may also determine connectivity changes.

We include the change in size ln(MV), market-to-book ln(MTB), firm risk (RISK) and board size ln(BSIZE) to control for the impact of changes in size, growth opportunities, risk of a firm ' operations, and size of the board. The results show no significant relationship between firm performance changes and the simultaneous changes in connectivity. However, there is weak evidence that changes in past performance increase connectivity, mostly limited to the Eigenvector coefficients.

From the controls, we observe the expected positive relationship between changes in board size and changes in board connectedness, and we observe a positive relationship between changes in firm risk and the aggregate measure of connectedness. 8 Andres (2013) uses changes in measures of board centrality on stock returns and Yermack (1996) looks at changes in board size on stock returns. In addition to the controls used in previous studies, I control for board size and risk.

A change in board size is expected to have an effect on connectivity and a positive change in risk may prompt firms to increase connections in search of a stronger resource base. These results are robust to alternative estimation methods, time periods and the inclusion of past firm performance or estimating the results using change variables, where the relationship generally becomes insignificant.

## Conclusion

Instead, we find that connected board members undermine the oversight role of directors, introduce poor quality information, or introduce too much information for the board to deal with.

Descriptive and sample statistics of characteristics of firms and management Descriptive statistics of characteristics of firms and management. Descriptive statistics on firm-level characteristics and governance are presented here for a sample of Australian listed firms from 2001 to 2011. Panels A, B, C and D provide averages of firm performance measured for portfolios, sorted by degree, proximity, betweenness and eigenvector or

Differences in business performance between the highest (5) and lowest (1) portfolios are tested for significance using the two-sample unpaired mean comparison test, assuming unequal variances. Panel B reports multivariate regression estimates of firm performance in terms of Grade using the model defined in Equation (7). Panel B reports multivariate regression estimates of firm performance on proximity using the model defined in equation (7).

Panel B reports multivariate regression estimates of firm performance on Betweenness using the model defined in Equation (7). Panel B reports the multivariate regression estimates of firm performance on the eigenvector using the model defined in Equation (7). Panel B reports multivariate regression estimates of firm performance on aggregate connectivity using the model defined in Equation (7).

This table presents pooled OLS regressions of firm performance on four measures of Connectivity and Aggregate Connectivity (CONN) using equation (7) and including a control for past performance. Pooled OLS regressions of actual firm performance on the combined measure of Aggregate Connectivity (CONN) are estimated using equation (7).