Board diversity and organizational valuation: unravelling the effects of ethnicity and gender

As will be discussed further, some of the reforms set specific targets of non-white (i.e., people of Black African, Chinese, Indian and Mixed Race backgrounds) board representation for organizations to comply with. For example, the 1998 Employment Equity and 2003 Black Economic Empowerment Acts set aspirational (not legally enforceable) target of 40 to 50% non-white senior management and/or board membership.

As an illustration (and will also be discussed further in Sects. 5, 7), the Appendix contains examples of annual report disclosures of progress made by some SA companies on complying with the 1998 employment equity and 2003 black economic empowerment acts.

This implicitly suggests that the average SA organizations should ideally have a workforce consisting of about 80% non-whites and 20% whites at all levels of the organizational hierarchy, including the board of directors in order to reflect the ethnic diversity and composition of the SA populace.

Board diversity has been broadly defined as the various attributes that may be represented among directors in the boardroom in relation to board process and decision-making, including age, gender, ethnicity, culture, religion, constituency representation, independence, knowledge, educational and professional background, technical skills and expertise, commercial and industry experience, career and life experience (van der Walt and Ingley 2003, p. 219). In this paper, we focus only on ethnic and gender aspects of the board.

This also suggests that the association between board diversity and organizational value can be non-linear, whereby initial increases in ethnic and gender board representations lead to declines in market valuation up to a point, beyond which additional increases in diversity improve market valuation (Roberson and Park 2007). Therefore, we investigate whether there is a curvilinear relationship between board diversity and market valuation, as part of our robustness analyses.

Organizational board decisions take time in order to be reflected in market value (Guest 2009; Ntim et al. 2012). Thus, to circumvent potential endogenous relationship between board diversity and market valuation, we introduce a one year lag between board diversity and market valuation such that this year’s organizational value depends on last year’s governance structure similar to Yermack (1996) and Ntim et al. (2012), as specified in equation (1). The sample also starts from 2002 for two reasons. First, King II came into operation in 2002, and secondly, data coverage in Perfect Information/DataStream on SA listed organizations is very limited until 2002. The sample ends in 2007 because it is the year for which data is available.

Due to insufficient number of observations in 3 industries, namely health care, oil and gas, and telecoms industries with three, one and three listed organizations, respectively, were merged with the closest remaining five major industries. Consequently, the three health care organizations were included in the consumer services industry, the one oil and gas organization was added to the basic materials industry, whilst the three telecoms organizations were also shared out to the technology industry.

However, we note that our choice is between random and fixed-effects estimation techniques. Therefore, to ensure that fixed-effects model is appropriate, we first carry out Hausman (1978) specification test by estimating both fixed and random-effects models for the BDIV proxies separately using Eq. (1) and comparing their respective coefficients. Under the null hypothesis of consistent random unobserved organizational-level heterogeneity (i.e., unobserved organization-specific effects or the regressions errors are uncorrelated with the independent variables), random-effects estimates will be both consistent and efficient, whilst fixed-effects coefficients will be consistent, but inefficient (Hausman 1978; Wooldridge 2010). In contrast, if the null hypothesis is rejected, then the fixed-effects approach will provide both consistent and efficient estimates, whereas random-effects estimates will be both inconsistent and biased (Hausman 1978; Gujarati 2003). The test consistently rejects the null hypothesis of consistent random effects for both models at the 1% level, providing further empirical support for our decision to rely primarily on fixed-effects models.

We carried out similar non-monotonic examination for the BDIVE, BDIVG and BDIVGNW measures and found statistically insignificant non-linear association between board diversity and market valuation. We further investigate other forms of non-linear transformations, such as cubing the variables (i.e., BDIVE, BDIVG, BDIVGNW and BDIV), but we found statistically insignificant link between board diversity and market valuation.

As estimating a lagged structure will jeopardise the validity of the Durbin-Wu-Hausman test (Gujarati 2003; Wooldridge 2010), we estimate Eq. (2)) as un-lagged structure. An additional advantage is that it allows us to ascertain the robustness of our results against estimating an un-lagged structure.