Women in the boardroom: a bottom–up approach to the trickle-down effect

Abstract

This paper argues that role modeling can explain the impact of boardroom gender diversity on corporate performance. It theorizes that female workers are boosted by female leadership, gain increased motivation, and achieve greater productivity, thereby making their female directors more effective. We test this bottom–up approach to the trickle-down hypothesis on data hand-collected among local cooperatives providing microcredit in Senegal. All the organizations surveyed are similar and small, which allows us to use a homogenous performance metric. All of them outsource their human resource management to the same third party, which mitigates the risk of endogeneity. The data cover over 100,000 triads composed of gender dominance on the board, gender of CEO, and gender of credit officer. A better financial performance is achieved when the triad is gender-uniform—be it male or female—confirming the importance of role modeling and suggesting that the performance of female board members depends on the gender composition of the workforce.

Plain English summary Women’s leadership styles differ from men’s. But we still ignore whether the styles adopted by male and female directors make any difference in terms of financial performance. Scholars hold controversial views about whether and how the financial performance of firms depends on gender diversity in the boardroom. This article speculates that female directors act as role models on their subordinates (“trickle-down effect”) and their impact is “bottom–up” in the sense that female workers gain increased motivation when working under female leadership. We hand-collected data from financial cooperatives in Senegal. These organizations enabled us to observe the unlikely situation of boards including over 50% of women. We measured financial performance with loan repayment. Our results confirm that female-dominated boards achieve a better financial performance when they work with female CEOs and female subordinates. The principal implication is that the performance of female board members depends on the gender composition of the workforce.

Appendices

Appendix

Table 5 Correlation matrix

Robustness checks

To assess the robustness of our results, we propose three extensions of the model. First, for comparison purposes, we run OLS regressions where the dependent variable is Y = log(1 + Recovery rate) with robust standard errors to account for heteroscedasticity. The results in Table 6 confirm the baseline results. The only changes concern the loss of significance of coefficients associated with the dummy for a female subordinate in specifications (C) and (D) in Table 6. However, these changes do not affect our conclusions about the importance of gender combination and the fact that the financial performance of female-dominated boards is sensitive to the presence of female staff members along the production chain.

Second, we conduct the Tobit analysis in Table 3 with clustered standard errors. One may wonder why we introduce clustered standard errors in a robustness check rather than in the baseline regressions. The reason is that, in our setting, clustering itself might seem ad hoc because it can be performed at several levels. For instance, clustering standard errors at the credit officer level would be questionable since we wish to test whether a gendered environment affects the lending and monitoring behavior of female officers. If this hypothesis is true, an officer-based clustering might well misleadingly hide behavioral differences in the granting and monitoring of loans. The same holds for female CEOs. Overall, we decided to cluster standard errors at the triad level but keep this refinement for the sake of robustness.

Since each triad (board, CEO, and credit officer) monitors more than one loan, we check the robustness of our results by clustering the standard errors for loans granted during the same month by the same triad—to account for unobserved variance. Table 7 shows the results, which are close to those obtained in our baseline regressions. The only changes concern the loss of significance of four coefficients in Table 7: the dummy for a female-dominated board in specifications (B) and (D) and the dummy for a female subordinate in specifications (C) and (D). These changes do not affect our conclusions. The main insight from this robustness check is that the gendered impact of credit officers on financial performance uncovered in previous work (Mersland and Strøm 2009) might be context-dependent rather than systematic. By contrast, in our setting, the negative impact of female CEOs is strong, confirming the role congruity hypothesis (Eagly and Karau 2002), according to which formal authority is viewed as masculine and is therefore less effective when exercised by female CEOs.

In the third robustness check, we use a specification where the dependent variable is the probability of full repayment.⁸ Both the recovery rate used in our baseline estimate and the probability of full repayment are popular performance indicators in banking. However, we preferred to use the recovery rate in our baseline estimates (see Tables 3 and 4) because of its higher granularity, which takes into account the existence and level of partial reimbursements. The recovery rate provides a more accurate picture of the actual cash flow of the lender. Here, we use probit models where repayment takes the value of 1 when the loan is repaid in full and zero otherwise. The results presented in Table 8 are similar to those obtained with Tobit models explaining Y = log(1 + Recovery rate) in the baseline specifications, demonstrating the robustness of our results with respect to how repayment performance is measured. Overall, the robustness checks confirm our previous findings.

Table 6 OLS model with robust standard errors

Table 7 Tobit model with clustered standard errors

Table 8 Probit model with robust standard errors