How do female CEOs affect their company’s gender wage gap?

The labour market gender gap has a long history in economics research; its persis­tency is one of the great puzzles in labour economics and one of the most important issues in labour market policy. Over the years, the gap has evolved considerably as women have increased their labour force participation and have overcome men in terms of educational attainment. Despite the convergence between men and women in many labour market indicators, women are still vastly underrepresented in the higher levels of firms’ hierarchies (See for example, Bertrand and Hallock 2001, Wolfers 2006, Gayle et al 2012, Dezsö and Ross 2012 for the US. For other countries, see Cardoso and Winter-Ebmer 2010 (Portugal), Ahern and Dittmar 2012 and Matsa and Miller 2013 (Norway), Smith et al 2006 (Denmark)). Specifically, recent US data from the 2012 Current Population Survey and Exe­cuComp show that even though women are a little more than 50% of white collar workers, they represent only 4.6% of executives. Our own Italian data show that about 26% of workers in the manufacturing sector are women compared with only 3% of executives and 2% of CEOs.

In a recent paper, we exploit a detailed matched employer-employee longitudinal data set for Italy to uncover two original pieces of empirical evidence that may help our understanding of the causes of the underrepresentation. We first investigate what is the effect of the CEO’s gender on firms’ wage policies. Our data allows us to analyse the impact on the entire wage distribution within the firm while accounting for various observed and unobserved firm and workforce differences.

Our regressions by wage quantiles show that firm and workforce heterogeneity are relevant both by gender and by wage levels. The impact of female CEOs is positive on the wages of women at the top of the wage distribution but negative on wages of women at the bottom of the wage distribution. The impact on men is the opposite: female CEOs lower wages at the top and increase them at the bottom of the male wage distribution. As a result, female CEOs reduce the gender wage gap at the top and widen it at the bottom of the wage distribution, with essentially no effect on the average.

Figure 1. Coefficients of female CEO dummy on average wages by quantile of the female and male wage distributions

The estimated effects are reported in Figure 1. The figure reports the percentage differences between wages of workers employed in firms with a female and a male CEO, by quartiles of the female (continuous line) and male (dashed line) wage distribution. For example, females in a woman-led firm in the top quartile of the wage distribution earn approximately 10 percent more than females working for a male in the same quartile. Men, on the other hand, earn about 4 percent less if they work in firms with a female CEO. At the bottom of each distribution the effects are smaller, but of the opposite sign.

It is important to notice that these effects are estimated with firm fixed-effects, i.e. they are identified by firms switching from a male to a female CEO or vice versa. They are also robust across various specifications. In particular, they hold after taking into account industry effects, cohort effects, time trends, firm size, region, and individual workforce and executive’s observable characteristics and unobservable skills. They are also robust to specifying a different measure of female leadership – the proportion of female executives in the firm.

The second piece of empirical evidence we uncover refers to the impact of a female CEO on firm performance, an issue that has occupied the literature for some time. Prior evidence is mixed and depends on the estimation methods but overall seems to indicate that female CEOs neither improve nor worsen firm performance (Wolfers 2006 and Albanesi and Olivetti 2009). We find, however, that firms with female leadership perform better the higher the fraction of women in the workforce. This effect is large and highly statistically significant. We measure performance according to three different dimensions: sales per worker, value added per worker, and total factor productivity (TFP). In our preferred specification, sales per worker are 6 percent higher for every 10 percent increase in the share of women in the workforce, and value added per worker increases by approximately 8%.

What are the possible drivers of these wages and productivity differences? Our paper discusses three possible sources.

The ’glass ceiling’ literature emphasises the possibility that prejudice against women in male-dominated firms affects women’s wages and promotion prospects. We do indeed find that women at the top of the wage distribution earn less when employed by males, a result that is consistent with this hypothesis. However, this explanation is not consistent with the negative effect of female leadership on female wages at the bottom of the distribution, or the positive impact of female leadership at the bottom of the male wage distribution. Only ad-hoc preferences for or against males and females with different skill levels could reconcile a prejudice-based expla­nation with our evidence. Moreover, if female CEOs favoured female workers against male workers, especially when these workers have high skills, we would not observe the positive effect on productivity we find when female CEOs employ a larger share of females, as favouritism should be negatively associated with performance.

A second explanation focuses on peer-group, role model effects, or other com­plementarities between female CEOs and their highly skilled female employees. For example, if females at the higher end of the skill distribution find it easier to interact with female firm leaders, or if the transmission of knowledge is easier between people of the same gender, we should find both the effects on productivity shown in our results, and positive impacts at the top of the female wage distribution. However, this alternative explanation is unable to generate the negative effect of female lead­ership on female wages at the bottom of the distribution, and any effect of female leadership on the male wage distribution.

While peer and role-model effects may play a positive role in breaking the glass ceiling and improving women’s condition in the workplace, they are consistent only with some of the evidence we uncover. We therefore propose a third explanation based on a model with statistical discrimination that reconciles all of our evidence. In our model, we assume that CEOs are better (i.e., more accurate) at assessing skills of workers of their own gender. This assumption may be motivated by gender dif­ferences in language, verbal and non-verbal communication styles and perceptions that may affect the assessment of personal skills and attitudes, improve conflict resolutions, and favour a correct job-task assignment. A large socio-linguistic literature has found support for this assumption. For example, Dindia and Canary (2006) and Scollon et al (2011) find differences in verbal and non-verbal communication styles between groups defined by race or gender that may affect economic and so­cial outcomes. Recent employee surveys also indicate that significant communication barriers between men and women exist in the workplace (Angier and Axelrod 2014, Ellison and Mullin 2014). We also assume that complex tasks require more skills to be completed successfully, and that there is a comparative advantage to employ higher human capital workers in complex tasks.

Two main empirical implications result from this model. First, thanks to the more precise signal they receive from female workers, female CEOs can reduce the mismatch between female workers’ productivity and job requirements. As a result, firm performance will increase more the higher the fraction of females employed. This is what we find in our data.

Second, and again because of the more accurate assessment of female workers’ skills, female CEOs will make more efficient task assignments of female workers, who will receive compensation closer to their actual productivity. Wages at different ends of the wage distribution are affected differently by information of different quality. When information is noisy (as we assume happens when female workers are employed by male CEOs), worker-task mismatches occur more frequently. Therefore, female workers with high skill will have on average lower wages when employed by male CEOs – some of them are mismatched to lower productivity jobs. The opposite happens at the bottom of the distribution – male CEOs assign some low productivity workers to higher paid tasks, increasing their pay, on average, relative to similarly skilled female workers employed by female CEOs. More precise information therefore results in higher wages at the top of the female skill distribution and lower wages at the bottom when they are employed by female CEOs, relative to the case when information is less precise, when female workers are employed by male CEOs. The opposite occurs to male workers, in line with our results summarised in the figure above.

This theory has implications on the effects of policies that have been recently implemented in the EU and other countries to reduce gender inequality at the top of firms’ hierarchy. These policies are justified on three grounds: justice for women who deserve the same opportunity to reach the top of the corporate ladder as men; improved company performance, which can be driven by several sources as we noted above; and a wider public interest in equality of opportunities for all (see Walby 2013). We can assess the second motivation by simulating how firm performance would change if a larger fraction of companies in our sample had a female CEO. While this is not exactly the policy that most countries have implemented, it is suggestive of the effect of policies aimed at increasing female representation at the top of the corporate ladder.

We performed two counterfactual experiments. In one experiment, we assign a female CEO randomly to one half of the firms in our sample. In the second experiment, we allocate female CEOs to the same number of firms, but targeting the assignment only to the firms that have the largest fraction of female employees. We do this to generate the largest productivity effects. Results show that when female CEOs are allocated randomly, the average percent change in firm performance is generally small. In contrast, our ‘targeted’ exercise delivers large positive effects in the firms that are assigned a female CEO, and also positive effects overall. For example, in this scenario sales per worker would increase by 14.2% in the firms whose CEO’s gender has changed, and by 6.7% in the overall sample of firms. Although our exercises ignore general equilibrium effects (including endogenous re-employment of workers of different gender to firms with different leadership), these results confirm that, based on our estimates, the order of magnitude of the efficiency gains from having a larger female representation in firm leadership can be quite large.

This article is published in collaboration with Vox EU. Publication does not imply endorsement of views by the World Economic Forum.

To keep up with the Agenda subscribe to our weekly newsletter.

Author: Luca Flabbi is a Senior Economist in the Research Department, Inter-American Development Bank. Mario Macis is an AssisAssistant Professor of Economics and Management at the Carey Business School, Johns Hopkins University. Andrea Moro is a Full Professor, Università Ca’ Foscari di Venezia. Fabiano Schivardi is a Professor of Economics, University of Cagliari; Research Fellow, EIEF.

Image: A woman with a book sits on a bench at the departure area at the Fraport airport. REUTERS/Lisi Niesner