Finance Research Letters 22 (2017) 129–135

Contents lists available at ScienceDirect

Finance Research Letters

journal homepage: www.elsevier.com/locate/frl

CEO age and CEO gender: Are female CEOs older than their
male counterparts?
Pradit Withisuphakorn a, Pornsit Jiraporn b,∗
a
National Institute of Development Administration (NIDA), NIDA Business School, Bangkok, Thailand
b
Pennsylvania State University, Great Valley School of Graduate Professional Studies, Malvern, PA 19355, USA

a r t i c l e i n f o a b s t r a c t

Article history: Motivated by the debate on gender inequality, we study CEO gender and CEO age. Be-
Received 7 November 2016 cause women face significantly more obstacles in advancing their careers, it may take them
Accepted 30 December 2016
longer to reach the top position, i.e. the chief executive officer (CEO). If this is the case, fe-
Available online 31 December 2016
male CEOs should be older than their male counterparts on average. Our evidence shows
that female CEOs are actually younger on average, approximately two full years younger
JEL Classification:
M14 than male CEOs, after controlling for firm and board characteristics. The two-year differ-
G30 ence represents as much as 26% of the standard deviation in CEO age.
© 2017 Elsevier Inc. All rights reserved.
Keywords:
Female CEOs
Gender diversity
CEO age
Women CEOs
Corporate culture

1. Introduction

Gender inequality has been increasingly in the spot light. However, in spite of the rising attention on this issue, it is
well-known that women are still at a disadvantage in several ways. For instance, women do not get paid as much as men
are for doing a similar job. The scarcity of female executives in large corporations also implies that it is substantially much
harder for women to be given an executive position. Motivated by the issue of gender equality, we investigate CEO gender
and CEO age. Women face many more obstacles than men do when they try to advance their careers. Women are less likely
to get promoted or get promoted later than men are. So, it is reasonable to expect that it takes more time for women to
reach a high-ranking position. In this study, we concentrate on the highest position in the firm, i.e. the chief executive offer
(CEO). Because it takes more time for women to advance their careers, by the time they become a CEO, they are expected
to be older than men of comparable quality. So, it can be argued that female CEOs should be older on average.
On the contrary, given the specific and rare combination of skills necessary to ascend to the CEO position, there should
be no difference between males and females among top executives (Faccio, Marchica, Mura, 2011). In fact, women who can
overcome so many obstacles and so much adversity in the male-dominated world must possess extraordinary skills. If this
is the case, women can reach the CEO position no later than men can and, perhaps, even earlier. As a result, female CEOs
are not older and may be even younger than their male counterparts. Our study is the first to shed light on the issue of
CEO gender and CEO age.

∗
Corresponding author.
E-mail address: pjiraporn@gmail.com (P. Jiraporn).

http://dx.doi.org/10.1016/j.frl.2016.12.026
1544-6123/© 2017 Elsevier Inc. All rights reserved.
130 P. Withisuphakorn, P. Jiraporn / Finance Research Letters 22 (2017) 129–135

There are several theoretical arguments as to why women are at a disadvantage in corporations. For instance, the human
capital theory (Becker, 1964) argues that women lack adequate human capital for a high-raking position. Gatekeepers, who
are mostly male, do not offer women the same organization rewards, such as training and development, nor promotion and
pay (Oakley, 20 0 0; Terjesen, Sealy, and Singh, 20 09). The social identity theory argues that individuals seeks to surround
themselves with people who share similar demographic profiles, perspectives, and values. Given that most executives are
male, women are probably viewed less favorably, according to the social identity theory (Tejfel and Turner, 1986). Because
of the disadvantages discussed above, it can be argued that it is much harder and therefore takes substantially longer for
women to reach the CEO position, resulting in female CEOs being older on average.
Based on more than 12,0 0 0 observations over 15 years, our empirical evidence shows that female CEOs are actually
younger than their male counterparts, approximately two full years younger to be specific. The two-year difference repre-
sents as much as 26% of the standard deviation of CEO age. The age difference remains even after controlling for several firm
characteristics, board structure, as well as variations over time and across industries. We also execute a fixed-effects analysis,
which control for any firm-specific time-invariant omitted variables, and obtain consistent results. Because only about 10%
of the CEOs in our sample are female, we conduct an analysis using propensity score matching. For each firm with a female
CEO, we identify a comparable firm with a male CEO based on a large number of firm and board characteristics. Thus, our
treatment and control groups are indistinguishable in terms of observable characteristics, and therefore should hire CEOs of
comparable age if gender were irrelevant. Nevertheless, even with the propensity score matching, we still find that female
CEOs are younger.
To mitigate reverse causality, we execute an instrumental-variable (IV) analysis. First, we employ CEO gender in the
earliest year for each firm as the instrument. CEO gender in the earliest year could not have resulted from CEO age in any
of the subsequent years, thereby reducing reverse causality. Our IV analysis shows that female CEOs are younger. Moreover,
we confirm the results using an additional instrumental variable, i.e. proportion of female CEOs of companies in the same
3-digit zip code. Because firms tend to recruit their employees from geographically-proximate areas, firms located close to
one another share the same employee pool. Using as our instruments CEO gender in the earliest year and proportion of
female CEOs of firms in the same area, we continue to find that female CEOs are younger than their male counterparts.
In addition, to ensure that our results are not driven by unobservable heterogeneity, we run a fixed-effects regression
analysis. This approach controls for unobservable characteristics that remain constant over time. The fixed-effects analysis
confirms our results. Further, we exploit the insight in Altonji et al. (2005) to estimate how strong selection from unobserv-
ables would have to be to render our results invalid. We find that selection from unobservables would have to be 9 times
stronger than selection on observables, an extremely unlikely probability. It does not appear that our results are vulnerable
to the omitted-variable bias. Finally, we execute a fixed-effect instrumental-variable analysis, which controls for both un-
observable heterogeneity as well as reverse causality. The results remain consistent and therefore appear to be robust to
endogeneity.
A number of recent studies investigate gender diversity in corporations. For instance, Ahern and Dittmar (2012) docu-
ment that the introduction of mandatory board member gender quotas led to an increase in acquisitions and performance
deterioration in Norway. Weber and Zulehner (2010) find that start-ups with female first hires are more likely to survive.
Other recent studies on gender diversity include Levi, Li, and Zhang (2011), Matsa and Miller (2012), and Faccio, Marchica,
and Mura (2016). Most studies in gender diversity focus on the effect of CEO gender or firm performance or corporate
outcomes. Our study is the first, however, to explore how CEO gender and CEO age are related. We find strong empirical
evidence that female CEOs are younger than their male counterparts.
Despite of so many obstacles and disadvantages, women reach the CEO position earlier than men do on average. Our
results can be explained by the status characteristics theory ((Biernat and Kobrynowicz, 1997). This theory posits that the
standards of ability for low-status groups (such as women) are higher than those for high-status group members (i.e. male).
Thus, to reach the same position, a woman must be of substantially higher quality. Because of their extraordinary skills and
ability, female CEOs do not reach the top of the company later than do their male counterparts. In fact, they reach the top
even earlier.

2. Sample selection and data description

Our data on CEO gender is from the EXCUCOMP database. Firm characteristics are from COMPUSTAT. The sample period is
from 1996 to 2010. We exclude financial and utility firms from the sample because they are regulated and therefore possess
unique firm characteristics. The final sample consists of 12,112 firm-year observations, representing 1787 unique firms (an
unbalanced panel data set).
Following the literature, we control for a number of firm characteristics. In particular, we control for firm size (total
assets), profitability (EBIT/total assets), leverage (total debt/total assets), investments (capital expenditures/total assets), in-
tangible assets (advertising and R&D), firm age, and year and industry dummies to control for variations over time and
across industries.
Table 1 shows the summary statistics. A few statistics are noteworthy. The proportion of firms with a female CEO is
9.9%. The average CEO age is 55.43 with a standard deviation of 7.476. The average percentage of independent directors on
the board is 68.775%. The average board size has 8.989 directors. Table 1 also shows the statistics for a number of firm
characteristics.
 P. Withisuphakorn, P. Jiraporn / Finance Research Letters 22 (2017) 129–135 131

Table 1
Descriptive statistics.

Mean Median Std. Dev. 25th 75th

Female CEO (Dummy) 0.099 0.0 0 0 0.298 0.0 0 0 0.0 0 0

CEO Age 55.430 55.0 0 0 7.476 50.0 0 0 60.0 0 0
% of Independent Directors 68.775 71.420 16.924 57.140 81.820
Board Size 8.989 9.0 0 0 2.370 7.0 0 0 10.0 0 0
Total Assets 56,610.430 1381.295 1.443 571.627 3799.394
EBIT/Total Assets 0.098 0.098 0.107 0.057 0.146
Total Debt/Total Assets 0.214 0.207 0.176 0.055 0.325
Capital Expenditures/Total Assets 0.057 0.042 0.053 0.023 0.071
Advertising/Total Assets 0.013 0.0 0 0 0.036 0.0 0 0 0.009
Firm Age 23.929 18.0 0 0 18.985 10.0 0 0 33.0 0 0

Table 2
Are female CEOs older than their male counterparts? female CEO is a dummy variable equal to one if the CEO is female and zero otherwise.

(1) (2) (3) (4) (5) (6)

OLS OLS Fixed-effects Fixed-effects Between Random-effects
CEO Age CEO Age CEO Age CEO Age CEO Age CEO Age

Female CEO −2.205∗ ∗ ∗ −2.095∗ ∗ ∗ -1.461∗ ∗ ∗ −1.351∗ ∗ ∗ −2.470∗ ∗ ∗ −1.594∗ ∗ ∗

(−6.474) (−6.127) (−6.249) (−5.575) (−3.297) (−7.169)
% Independent Directors −0.017∗ 0.022∗ ∗ ∗ 0.017∗ ∗ ∗ −0.018∗ 0.010∗
(−1.836) (3.749) (2.876) (−1.752) (1.945)
Ln(Board Size) 0.494 −0.175 −0.063 1.034 0.160
(0.602) (−0.360) (−0.124) (1.237) (0.382)
Ln(Total Assets) 0.105 1.046∗ ∗ ∗ 0.912∗ ∗ ∗ 0.122 0.363∗ ∗ ∗
(0.907) (6.054) (4.854) (0.861) (3.566)
EBIT/Total Assets −1.127 −1.585∗ −0.752 −1.657 −1.298∗
(−1.046) (−1.918) (−0.840) (−0.978) (−1.762)
Total Debt/Total Assets −0.764 0.530 1.151∗ −0.950 0.155
(−0.962) (0.903) (1.874) (−0.944) (0.307)
Capital Expenditures/Total Assets −5.604∗ ∗ −1.856 −1.359 −7.792∗ ∗ −2.872∗
(−2.284) (−0.975) (−0.647) (−1.991) (−1.676)
Advertising/Total Assets −10.251∗ ∗ ∗ −5.826 0.981 −9.228∗ ∗ −8.455∗ ∗ ∗
(−3.041) (−1.440) (0.222) (−2.516) (−3.074)
R&D/Total Assets −11.308∗ ∗ ∗ 1.311 2.050 −13.291∗ ∗ ∗ −5.716∗ ∗ ∗
(−4.411) (0.546) (0.836) (−4.353) (−3.113)
Firm Age 0.025∗ ∗ -0.002 0.002 0.036∗ ∗ ∗ 0.022∗ ∗ ∗
(2.505) (−0.055) (0.071) (3.698) (2.679)
Constant 55.643∗ ∗ ∗ 56.977∗ ∗ ∗ 47.874∗ ∗ ∗ 48.493∗ ∗ ∗ 53.248∗ ∗ ∗ 53.479∗ ∗ ∗
(371.759) (28.692) (29.665) (23.897) (13.371) (14.605)
Fixed effects None Year, Industry Year, Firm Year∗ Industry, Firm Year, Industry Year, Industry
Observations 12,112 12,112 12,112 12,112 12,112 12,112
R-squared 0.008 0.072 0.528 0.566 0.129

t-statistics in parentheses.
∗∗∗
p< 0.01, ∗ ∗ p< 0.05, ∗ p< 0.1.

3. Results

3.1. CEO gender and CEO age

We start by making a simple t-test comparison between male and female CEOs in term of age. The average age for the
male CEO is 55.64, whereas the average for the female CEO is 53.64. The t-statistic for the difference is highly significant at
the 1% level. So, the preliminary evidence shows that female CEOs are statistically younger. Then, we execute a regression
analysis. Table 2 shows the results of the regression analysis. The dependent variable is CEO age in years. The standard
errors are clustered by firm. Our focus is on the dummy variable equal to one if the CEO is female and zero otherwise. In
Model 1, the female CEO dummy is the only independent variable. The coefficient of the female CEO dummy is negative
and highly significant. In Model 2, we include firm and board characteristics as control variables. The female CEO dummy
still carries a negative and significant coefficient. The magnitude of the coefficient in Model 2 is 2.095, suggesting that, on
average, female CEOs are more than two full years younger than their male counterparts. The two-year difference represents
26% of the standard deviation in CEO age, which is considerable.
It can be argued that the results may be driven by unobservable characteristics that remain constant over time. To deal
with this issue, we perform a fixed-effects analysis. Essentially, a fixed-effects analysis remove any cross-sectional variation
and concentrates only on the variation over time within the same firm. So, this analysis would capture the change in CEO
age when the gender of the CEO changes within the same firm. Model 3 is a fixed-effects regression. Again, the coefficient
132 P. Withisuphakorn, P. Jiraporn / Finance Research Letters 22 (2017) 129–135

Table 3
Are female CEOs older than their male counterparts?: propensity score matching.

(1) (2) (3) (4)

OLS OLS Fixed-effects Fixed-effects
CEO Age CEO Age CEO Age CEO Age

Female CEO −2.299∗ ∗ ∗ −2.071∗ ∗ ∗ −1.341∗ ∗ ∗ −1.081∗ ∗

(−5.816) (−5.025) (−3.351) (−2.288)
% Independent Directors 0.003 −0.001 0.016
(0.198) (−0.043) (0.861)
Ln(Board Size) 1.122 0.612 0.410
(1.047) (0.495) (0.287)
Ln(Total Assets) 0.418∗ ∗ 0.324 0.050
(2.455) (0.682) (0.090)
EBIT/Total Assets −1.382 −4.069 −4.228
(−0.687) (−1.568) (−1.425)
Total Debt/Total Assets −0.550 −0.322 0.282
(−0.467) (−0.190) (0.145)
Capital Expenditures/Total Assets −8.053∗ −3.726 2.985
(−1.827) (−0.631) (0.429)
Advertising/Total Assets −12.311∗ −1.429 10.606
(−1.877) (−0.148) (0.894)
R&D/Total Assets −8.564 −11.945 −4.615
(−1.642) (−1.415) (−0.532)
Firm Age −0.005 0.042 0.006
(−0.466) (0.787) (0.118)
Constant 55.737∗ ∗ ∗ 54.893∗ ∗ ∗ 53.031∗ ∗ ∗ 54.242∗ ∗ ∗
(216.134) (18.780) (11.774) (9.542)
Fixed effects None Industry, Year Year, Firm Industry∗ Year, Firm
Observations 2390 2390 2390 2390
R-squared 0.028 0.133 0.664 0.807

t-statistics in parentheses.
∗∗∗
p < 0.01, ∗ ∗ p < 0.05, ∗ p < 0.1.

of the female CEO dummy is still negative and significant. In Model 3, because we include the firm fixed effects, we can-
not include the industry dummies. To account for industry variation as well as variation over time, we multiply the year
dummies by the industry dummies and include a separate dummy for each product. Model 4 is a fixed-effects regression
with the industry-year dummies as well as the firm fixed-effects. The female CEO dummies continue to show a negative
and significant coefficient.
In Model 5, we calculate the averages of all the variables for each firm over time and include them in the regression.
Essentially, this regression captures only the cross-sectional variable across firms. The coefficient of the female CEO dummy
remains significantly negative. Finally, in Model 6, we execute a random-effects model as a robustness check. The result re-
mains similar. Regardless of model specifications, we always obtain consistent results, showing that female CEOs are younger.

3.2. Propensity score matching

Only about 10% of the sample firms have a female CEO. So, the treatment and the control groups are lopsided. To alleviate
this concern, we execute propensity score matching. For each firm with a female CEO, we identify a comparable firm with
a male CEO based on nine variables, i.e. the nine control variables in the regression analysis. So, our treatment and control
groups are very similar in terms of observable characteristics, except for one dimension, i.e. CEO gender.
Table 3 shows the regression results based on propensity score matching. Model 1 has the female CEO dummy as the
only independent variable. The coefficient is negative and significant. In Model 2, we add the control variables. The female
CEO dummy still exhibits a negative and significant coefficient. Model 3 is a fixed-effects regression, which yields a similar
result. Finally, in Model 4, we include the products of the year and industry dummies to control for variations over time
and across industries, in addition to the firm fixed effects. Please note that the R2 in Model 4 is very high, 80.7%. So, our
model captures over 80% of the variation in CEO age.

3.3. Instrumental-variable analysis and the treatment effects model

To further confirm the results, we execute an instrumental-variable (IV) analysis. To mitigate concerns on reverse causal-
ity, we identify the earliest year when each firm appears in our sample. Then, we replace the value of the female CEO
dummy in each given year by the value in the earliest year. As CEO gender in the earliest year could not have resulted from
CEO age in any of the subsequent years, reverse causality is highly unlikely. The results are shown in Table 4. Model 1 is
the first-stage regression, where we regress the female CEO dummy on the female CEO dummy in the earliest year. The
coefficient of the CEO dummy in the earliest year is positive and highly significant. Model 2 is the second-stage regression,
 P. Withisuphakorn, P. Jiraporn / Finance Research Letters 22 (2017) 129–135 133

Table 4
Instrumental-variable analysis.

(1) (2) (3) (4)

First Stage Second Stage First Stage Second Stage
Female CEO CEO Age Female CEO CEO Age

Female CEO (Earliest Year) 0.540∗ ∗ ∗ 0.415∗ ∗ ∗

(34.596) (30.556)
Proportion of Female CEOs 0.900∗ ∗ ∗
in the same 3-digit zip code (65.317)
Female CEO (Instrumented) −2.567∗ ∗ ∗ −2.193∗ ∗ ∗
(−3.342) (−5.451)
% Independent Directors 0.001∗ ∗ ∗ -0.016∗ ∗ ∗ 0.001∗ ∗ ∗ −0.017∗ ∗ ∗
(8.193) (−3.448) (7.519) (−3.674)
Ln(Board Size) 0.052∗ ∗ ∗ 0.526 0.046∗ ∗ ∗ 0.500
(3.964) (1.452) (4.097) (1.393)
Ln(Total Assets) 0.020∗ ∗ ∗ 0.113∗ 0.015∗ ∗ ∗ 0.107∗
(8.844) (1.797) (7.475) (1.722)
EBIT/Total Assets 0.049∗ −1.108 0.035 −1.123∗
(1.947) (−1.624) (1.600) (−1.648)
Total Debt/Total Assets −0.057∗ ∗ ∗ −0.792∗ −0.031∗ ∗ −0.770∗
(−3.548) (−1.826) (−2.216) (−1.783)
Capital Expenditures/Total Assets −0.026 −5.599∗ ∗ ∗ −0.013 −5.603∗ ∗ ∗
(−0.431) (−3.461) (−0.249) (−3.464)
Advertising/Total Assets 0.045 −10.221∗ ∗ ∗ 0.127∗ ∗ −10.245∗ ∗ ∗
(0.602) (−5.034) (1.980) (−5.048)
R&D/Total Assets 0.123∗ ∗ -11.265∗ ∗ ∗ 0.050 −11.299∗ ∗ ∗
(2.332) (−7.975) (1.098) (−8.007)
Firm Age 0.001∗ ∗ ∗ 0.026∗ ∗ ∗ 0.001∗ ∗ ∗ 0.025∗ ∗ ∗
(7.271) (6.069) (4.497) (6.044)
Constant −0.449∗ ∗ ∗ 56.748∗ ∗ ∗ −0.284∗ ∗ ∗ 56.929∗ ∗ ∗
(−5.352) (24.981) (−3.925) (25.312)
Fixed Effects Industry, Year Industry, Year Industry, Year Industry, Year
Shea’s Partial R2 0.090 0.326
Sargan’s Statistic (p-value) 0.568
Observations 12,112 12,112 12,112 12,112
R-squared 0.169 0.071 0.384 0.072

t-statistics in parentheses.
∗∗∗
p< 0.01, ∗ ∗ p< 0.05, ∗ p< 0.1.

where we include the female CEO dummy instrumented from the first stage as an independent variable. The coefficient
of the instrumented female CEO dummy is negative and significant. Thus, even after accounting for reverse causality, we
continue to find that female CEOs are younger.
For further robustness, we add one more instrumental variable so we can perform a test of over-identifying restrictions.
Firms tend to recruit employees from geographically proximate areas. So, it is likely that firms located in close proximity
share the same talent pool, particularly the female talent pool. In an area with a large number of high-quality female
individuals, firms in the area are more likely to have a female CEOs. We use as our instrument the proportion of firms in
the same 3-digit zip code with a female CEO. Zip codes are assigned to maximize efficiency in mail delivery and thus should
not be related to firm policies or outcomes. Zip code assignments are therefore plausibly exogenous.
In Model 3, we first regress the female CEO dummy on the proportion of females CEOs in the same 3-digit zip code, the
female CEO dummy in the earliest year, and the control variable. The proportion of female CEOs close by has a significantly
positive coefficient as does the coefficient of the female CEO dummy in the earliest year. Shea’s partial R2 is 38.4%. Model 4 is
the second-stage regression, where CEO age is the dependent variable. The instrumented female CEO dummy’s coefficient is
negative and significant, indicating that female CEOs are younger. Because the model is over-identified, we can now estimate
Sargan’s statistic. The p-value for the Sargan statistic is 0.568, not statistically significant. So, our instrumental variables are
acceptable.
As a further robustness check, we execute a fixed-effects instrumental-variable analysis. This approach is very rigorous
and controls for both unobservable heterogeneity, reverse causality, and possible measurement errors. Table 5 shows the
results. Model 1 is the first-stage regression. The proportion of female CEOs in the same area has significant explanatory
power. Model 2 is the second-stage regression. The coefficient of the instrumented female CEO dummy is still significantly
negative. It appears that our results are robust and clearly show that female CEOs are indeed younger.
Furthermore, to account for the fact that firms with a female CEO may be subject to a selection bias, we execute a vari-
ation of the Heckman two-step approach: the treatment effects model. The first stage predicts the likelihood that a firm
would have a female CEO. From the first stage, we estimate the inverse Mills ratio. In the second stage, we include the
inverse Mills ratio in the regression. In this manner, we can explicitly test whether CEO gender is related to CEO age af-
ter controlling for self-selection due to unobservable private information. In order to minimize muti-collinearity problems,
134 P. Withisuphakorn, P. Jiraporn / Finance Research Letters 22 (2017) 129–135

Table 5
Fixed-effects instrumental-variable analysis.

(1) (2)
First Stage Second Stage
Female CEO CEO Age

Proportion of Female CEOs 0.951∗ ∗ ∗

in the same 3-digit zip code (58.535)

Female CEO (instrumented) −1.197∗ ∗

(−2.532)
% Independent Directors 0.0 0 0∗ 0.017∗ ∗ ∗
(1.709) (2.856)
Ln(Board Size) 0.031∗ −0.068
(1.790) (−0.134)
Ln(Total Assets) −0.006 0.911∗ ∗ ∗
(−0.933) (4.849)
EBIT/Total Assets 0.054∗ −0.762
(1.681) (−0.851)
Total Debt/Total Assets −0.031 1.158∗
(−1.433) (1.886)
Capital Expenditures/Total Assets 0.086 −1.374
(1.153) (−0.655)
Advertising/Total Assets −0.287∗ 1.040
(−1.833) (0.235)
R&D/Total Assets 0.045 2.053
(0.511) (0.837)
Firm Age −0.002∗ ∗ 0.003
(−2.067) (0.090)
Constant 0.032 47.964∗ ∗ ∗
(0.445) (27.353)
Fixed Effects Industry∗ Year Industry, Year
Observations 12,112 12,112
R-squared 0.640 0.565

t-statistics in parentheses.
∗∗∗
p < 0.01, ∗ ∗ p < 0.05, ∗ p < 0.1.

we include the proportion of female CEOs in the same area in the first-stage regression. The coefficient of the female CEO
dummy remains negative and significant (results not shown but available upon request). Thus, after accounting for unob-
served private information that makes certain firms more likely to select a female CEO, there is still a large effect of CEO
gender on CEO age. Female CEOs are indeed younger.
Finally, to ensure that our results are not driven by unobservable heterogeneity, we exploit the insight of Altonji et al.
(2005). Their study suggests that selection on observables can be used to estimate the potential bias generated by unob-
servables, that is, how much stronger the selection on unobservables, relative to the selection on observables, would have
to be to explain away the full estimated effect.1
This potential bias can be estimated in the following way. Consider two regressions: one with a restricted set of control
variables and the other with a full set of controls. Denote the estimated coefficient for the variable of interest from the
first regression β R (where R stands for Restricted) and the estimated coefficient from the second regression β F (where F
stands for Full). Then the ratio can be computed as β F /(β R - β F ).2 The intuition behind the formula is straightforward. First,
consider why the ratio is decreasing in (β R - β F ). The smaller the difference between β R and β F , the less the estimate is
affected by the selection of observables, and the stronger the selection on unobservables needs to be (relative to observables)
to explain away the entire effect. Then consider the intuition behind β F in the numerator. The larger β F , the more the effect
needs to be explained away by the selection on unobservables, and therefore the higher the ratio.
We use Model 1 in Table 2 as our restricted model. This model includes the female CEO dummy as the only independent
variable. We use Model 2 in Table 2 as our full model, which includes all the control variables. Then we estimate the ratio
as (−2.071)/(−2.299-(−2.071)) = 9.08. Consequently, to attribute the entire OLS estimate to selection effects, the selection on
unobservables would have to be at least 9.08 times stronger than the selection on observables. It appears unlikely that our
results are principally driven by unobservables. This provides a certain degree of comfort that our results are not spurious
due to possibly omitted variables.

1
Altonji et al. (2005) consider the situation where the explanatory variable is a binary variable. Bellows and Miguel (2009) develop an analogous test
for the case where the variable of interest is continuous. Full details of the test are provided by Bellows and Miguel (2008).
2
See Altonji et al. (2005) for details of the underlying assumptions.
 P. Withisuphakorn, P. Jiraporn / Finance Research Letters 22 (2017) 129–135 135

4. Concluding remarks

Gender inequality has engendered a great deal of debate in the literature as well in the society at large. In spite of the
substantial progress in reducing gender inequality in the past few years, women are still substantially at a disadvantage. Our
study contributes to the debate on gender inequality by investigating CEO gender and CEO age. Due the various obstacles
and disadvantages against women, women are less likely or take longer time to advance their careers. Therefore, by the
time a woman becomes a CEO, she may be older than an otherwise comparable male CEO. Based on more than 12,0 0 0
observations over 15 years, our evidence refutes this hypothesis. We show that female CEOs are actually younger than
their male counterparts by approximately two full years, even after controlling for firm and board characteristics. The two-
year difference represents about 26% of the standard deviation of CEO age. Additional analyses based on propensity score
matching as well as an instrumental-variable analysis strongly confirm our conclusion. Our results imply that women who
are able to make it to the top must possess extraordinary skills, which allow them to reach the CEO position younger.

References

Ahern, K.R., Dittmar, A., 2012. The changing of the boards: the impact on firm valuation of mandated female board representation. Q. J. Econ. 127, 137–197.
Altonji, J.G., Elder, T.E., Taber, C.R., 2005. Selection on observed and unobserved variables: assessing the effectiveness of Catholic schools. J. Political Econ.
151–184.
Becker, G., 1964. Human Capital. University of Chicago Press, Chicago, IL.
Biernat, M., Kobrynowicz, D., 1997. Gender- and race-based standards of competence: lower minimum standards, but higher ability standards for devalued
groups. J. Personality Social Psychol. 72, 544–557.
Faccio, M., M.T. Marchica, and R. Mura, 2016, CEO gender, corporate risk-taking, and the efficiency of capital allocation, J. Corporate Finance 39, 193–209.
Matsa, D.A., Miller, A.R., 2012. A Female Style in Corporate Leadership? Evidence from Quotas. Working paper, Northwestern University.
Oakley, J., 20 0 0. Gender-based barriers to senior management positions: understanding the scarcity of female CEOs. J. Bus. Ethics 27, 321–334.
Tejfel, H., Turner, J.C., 1986. The Social Identity Theory of Intergroup Behavior, Psychology of Intergroup Relations. Nelson-Hall, Chicago, IL.
Terjesen, S., Sealy, R., Singh, V., 2009. Women directors on corporate boards: a review and research agenda. Corporate Governance 17, 320–337.
Weber, A., Zulehner, C., 2010. Female hires and the success of start-up firms. Am. Econ. Rev. 100, 358–361.