You are on page 1of 13

Applied Economics

ISSN: 0003-6846 (Print) 1466-4283 (Online) Journal homepage: https://www.tandfonline.com/loi/raec20

Human capital and leadership: the impact of


cognitive and noncognitive abilities

Tingting Tong, Haizheng Li & Samuel Greiff

To cite this article: Tingting Tong, Haizheng Li & Samuel Greiff (2019): Human capital
and leadership: the impact of cognitive and noncognitive abilities, Applied Economics, DOI:
10.1080/00036846.2019.1619022

To link to this article: https://doi.org/10.1080/00036846.2019.1619022

Published online: 22 May 2019.

Submit your article to this journal

Article views: 42

View Crossmark data

Full Terms & Conditions of access and use can be found at


https://www.tandfonline.com/action/journalInformation?journalCode=raec20
APPLIED ECONOMICS
https://doi.org/10.1080/00036846.2019.1619022

Human capital and leadership: the impact of cognitive and noncognitive


abilities
Tingting Tonga, Haizheng Lib and Samuel Greiffc
a
International business college, Dongbei University of Finance and Economics, Dalian, China; bSchool of economics, Georgia Institute of
Technology and Hunan University, Atlanta, GA, USA; cThe Institute of Cognitive Science and Assessment, University of Luxembourg,
Luxembourg, Luxembourg

ABSTRACT KEYWORDS
We conduct an economic analysis about the impact of human capital on an individual’s potential Human capital; leadership;
of becoming a leader based on data from the Programme for the International Assessment of cognitive ability;
Adult Competencies Survey (PIAAC). Our human capital indicators include not only traditional noncognitive ability
measures such as education and experience, but also various measures of cognitive and non- JEL CLASSIFICATION
cognitive ability. Our cognitive ability measures include numeracy, literacy, and problem solving I21; J24
abilities, and noncognitive ability measures include perseverance, motivation to learn, and social
trust. We specifically investigate the effect of measurement error and reverse causality on the
estimation results. We find that problem-solving ability is the most important in affecting leader-
ship among cognitive ability measures, and perseverance shows the strongest impact among
noncognitive ability measures. As a leader supervises more employees, the role of cognitive and
noncognitive ability becomes more critical.

I. Introduction Chen, Grove., and Hussey 2017). Hanushek et al.


(2015) finds that cognitive ability is significant
It is believed that leadership is essential for busi-
predictor of wage, i.e. a one-standard deviation
nesses and organisations, because leaders are deci-
increase in numeracy skills is associated with an
sion-makers who choose the right direction for an
18% wage increase among prime-age workers. In
organisation. Leaders require a ‘package’ of skills
addition, studies found that noncognitive ability
as they are problem solvers who must identify the
explains a significant variation of the
problem, analyse causes, form plans and give
labour market outcome (e.g. Lundberg 2013;
instructions (Mumford et al. 2017). However,
Glewwe, Huang, and Park 2017).
only between 31% and 55% of large corporations
Based on Lazear (2012), leaders tend to be gen-
have a specific mechanism for systematically iden-
eralists rather than specialists because they con-
tifying leadership potentials of their employees in
front a wide variety of choices and these choices
the United States (Dries and Pepermans 2012).
span many fields. Therefore, it is a general skill,
One possible reason is that talents and skills pro-
not one specific talent or a particular technique,
cessed by leaders are difficult to measure and
that makes an individual a leader. Those general
quantify. Traditional human capital measures
skills can be represented by the human capital one
based on education level and experience cannot
possesses. Gibbons and Waldman (2006) shows
fully capture the effect of human capital on
that a worker is assigned to the high-level job
leadership.
(i.e. leadership position) if and only if her
In recent years, research on human capital has
expected human capital is greater than a certain
paid much attention to cognitive and noncognitive
critical level.
abilities as important factors in influencing one’s
However, it is challenging to measure those
labour market outcomes (e.g. Heckman, Stixrud,
general skills. Traditional measures are mostly
and Urzua 2006; Lindqvist and Vestman 2011;
based on education and on-the-job learning repre-
Eren and Ozbeklik 2013; Hanushek et al. 2015;
sented by experience. For example, Lazear (2012)

CONTACT Haizheng Li Haizheng.li@econ.gatech.edu School of economics, Georgia Institute of Technology, Atlanta, GA, USA
© 2019 Informa UK Limited, trading as Taylor & Francis Group
2 T. TONG ET AL.

showed that the number of previous roles an indi- leadership by employing new measures of cogni-
vidual held significantly increases the probability tive and noncognitive abilities that are more spe-
of being a leader. Grönqvist and Lindqvist (2016) cific to the leadership capability. Our cognitive
found that officer training during military service ability measures include numeracy, literacy, and
has a strong positive effect on the probability of problem solving, and noncognitive ability mea-
being a leader. Such human capital measures can- sures include perseverance, motivation to learn,
not distinguish skills for those with the same edu- and social trust. Moreover, we measure general
cation level and job experience. Therefore, the leadership based on the levels of the leadership
measure of human capital beyond education is in a variety of ways, from the occupation position
desirable in studying an individual’s to the number of employees supervised, to get
labour market outcomes. In this article, we inves- a better understanding of the impact of human
tigate how human capital such as cognitive and capital. Additionally, we specifically investigate the
noncognitive ability affects an individual’s prob- potential endogeneity problem caused by mea-
ability of becoming a leader, in addition to the surement error of the abilities and the potential
traditional human capital measures. reverse causality between the measured abilities
In recent years, cognitive and noncognitive abil- and leadership experience.
ities are increasing recognized as important factors We find that, among cognitive abilities, pro-
in influencing one’s labour market outcomes.1 For blem-solving ability, compared to numeracy and
instance, Kuhn and Weinberger (2005) showed that literary ability, has the largest effect in determin-
people who held leadership positions in high school ing the probability of being in a leadership posi-
are more likely to earn higher wages and take man- tion, and perseverance appears to be important
agerial positions as adults. Borghans, Well and noncognitive ability in determining leadership.
Weinberg (2008) showed that personality at age 16 Moreover, the role of cognitive and noncognitive
can help predict job assignment in the future, i.e. abilities becomes more important as the number
relatively caring (direct) people will end up working of employees supervised increases. Our results also
at jobs that require more caring (direct) personal- confirm that traditional human capital measures
ities. Similarly, Krueger and Schkade (2008) showed such as education and experience are both signifi-
that individuals who are more gregarious tend to be cant predictors of leadership.
employed in jobs that involve more social interac- The rest of the article is organized as follows. In
tions. These studies show that one’s human capital Section 2, we set up a simple theoretical framework
traits, even before completing education, help pre- on human capital and leadership. Section 3 intro-
dict his/her future career. duces the data, ability measures, and relationships
Human capital traits can be represented by cogni- among human capital measures. Section 4 presents
tive and noncognitive abilities and are quite stable in estimation results. In Section 5, we investigate the
life. They play an important role in determining one’s potential endogeneity issue, and Section 6 concludes.
labour market outcome. Lindqvist and Vestman
(2011) found that noncognitive ability is an important
determinant of productivity. Moreover, Lindqvist II. Theoretical framework and empirical
(2012) found that about one half of the height- strategy
leadership relationship is due to a positive correlation We follow Gibbons and Waldman's (2006) con-
between height and cognitive and noncognitive abil- ceptual framework of job assignment and human
ity. Moreover, Adam et al. (2018) found that noncog- capital. Assume that all firms have two kinds of
nitive ability is the best predictor of an appointment to jobs: managerial jobs ðj ¼ 1Þ, which are relatively
a CEO position, followed by cognitive ability. more complex, and ordinary jobs ðj ¼ 0Þ. We
This study conducts a more detailed investiga- define leaders as those individuals who conduct
tion about the effects of human capital on managerial jobs in a firm (j ¼ 1).

1
There is a large literature about leadership in the area of psychology and organizational behaviour literatures (See Zaccaro 2007; Yukl 2008,; Dinh et al. 2014
for a detailed review).
APPLIED ECONOMICS 3

Job assignment is based on productivity. In d1  d0 þ G1 ðSÞ  G0 ðSÞ


a firm, if an individual is more productive in hi  ¼ (2)
c0  c1
a managerial job than in an ordinary job, then
the individual will be assigned to managerial Given a fixed education level, when an individual
level jobs. Specifically, we define the output of has zero non-schooling human capital (i.e.
worker i assigned to job j as yij , hi ¼ 0), he/she will be relatively more productive
at ordinary jobs, i.e. d0 þ G0 ðSÞ > d1 þ G1 ðSÞ. As
yij ¼ dj þ Gj ðSi Þ þ cj hi þ uij (1) hi increases, the productivity growth in managerial
jobs is relatively faster than in ordinary jobs
where dj is the output of job j that is independent (c1 > c0 ). To be a leader, in general, an individual
of the worker’s human capital. Si represents indi- must be more productive in a managerial job than
vidual i’s level of education and Gj ðSi Þ is in an ordinary job, and thus an individual must
a function of education, and G0 j ðSi Þ > 0 for have a non-schooling human capital higher than
h . According to Figure 1, given hi > h , at
j ¼ 0 or 1. In the model, hi represents individual
a particular school level, the higher the non-
i’s human capital in addition to education. We will
schooling human capital, the larger the difference
refer to hi as non-schooling human capital, includ-
between productivities between a managerial and
ing on-the-job learning (Learni ), cognitive ability
an ordinary-level job, and thus the higher the
(Cogi ), and noncognitive ability (NonCogi ). In the
probability of being a leader. If education level
above equation, cj is the sensitivity of the output of
increases, the productivities
 are expected to
job j to the individual’s non-schooling human
increase (i.e. E yij shifts upwards).
capital, and uij is the error term.
Therefore, we specify the probability of being
Job characteristics include cj and dj , and we
a leader below,
assume d0 > d1 and c0 < c1 . As can be seen, h is
the level of non-schooling human capital at which ProbðleaderÞi ¼ f ðSi ; hi Þ þ ui
a worker is equally productive at jobs 0 and 1 (i.e. ¼ f ðSi ; Learni ; Cogi ; Noncogi Þ þ ui
Eðyi0 Þ ¼ Eðyi1 ÞÞ. Given a fixed schooling level, (3)
Figure 1 displays the relationship between non-
schooling human capital and output. We can show where the probability of being a leader depends on
that an individual’s education level, on-the-job learn-
ing, cognitive ability and noncognitive ability. We
estimate the model using various econometrics
techniques such as linear probability model and
Probit model. We further investigate the potential
endogeneity issue of ability measures using multi-
ple indicator approach and instrumental variable
estimation.

III. Data and measurement for leadership and


abilities
Our data originate from the Programme for the
International Assessment of Adult Competencies
(PIAAC) survey that was initiated by the OECD.2
The PIAAC survey collects internationally com-
parable data about key cognitive and workplace
Figure 1. Non-schooling human capital and output, given skills of adults between the ages of 16 and 65.
a fixed education level. Compared with other adult ability surveys, the

2
The link provides access to the PIAAC data: http://www.oecd.org/site/piaac/publicdataandanalysis.htm.
4 T. TONG ET AL.

main advantage of PIAAC is that it offers several Numeracy: "the ability to access, use, interpret
in-depth measures of cognitive and noncognitive and communicate mathematical information and
abilities. In our study, we use United States survey ideas in order to engage in and manage the math-
data, and we include only full-time employees.3 ematical demands of a range of situations in adult
Our primary measure for leadership is defined life" (OECD 2013).7
based on the occupational classification of the Literacy: "the ability to understand, evaluate,
respondent’s job at the 2-digit level as defined by use and engage with written texts to participate
the International Standard Classification of in society, to achieve one’s goals, and to develop
Occupations (ISCO-08). If the occupation is one’s knowledge and potential" (OECD 2013).8
‘Manager’, then the individual is defined as a - Problem solving (in technology-rich environ-
leader.4 With this definition based on ‘Manager’, ments): "the ability to use digital technology, com-
we have 187 leaders and 1,306 non-leaders. munication tools, and networks to acquire and
In addition, we construct two additional leader- evaluate information, communicate with others
ship measures: ‘Lead5’ and ‘Lead10’, which are and perform practical tasks" (OECD 2013).9
defined based on the number of individuals an The sample statistics for the three cognitive
individual supervises.5 This measure has the abilities are reported in Table 1. Leaders have
advantage of excluding individuals who have higher average cognitive ability scores in all three
a manager title but do not actually supervise domains than non-leaders. The average scores for
a team. Specifically, Lead5 defines leaders as who leaders in numeracy, literacy and problem solving
supervise more than five employees. Based on this are 2.95, 3.00 and 2.98, respectively, whereas the
measure, we have 261 leaders and 1,087 non- scores for non-leaders are 2.64, 2.78 and 2.80,
leaders. Lead10 defines leaders as those who lead respectively; and the difference in all three mea-
more than 10 employees. With this definition, we sures between leaders and non-leaders are statisti-
have 154 leaders and 1,194 non-leaders. The cor- cally significant. In addition to this, the standard
relation between Manager and Lead5 is 0.37, deviations of the measured cognitive abilities of
between Manager and Lead10 is 0.34, and between the leaders are smaller than those of the non-
Lead5 and Lead10 is 0.73. leaders.
The PIAAC survey collects information on each Similar to cognitive ability, there are many dif-
individual’s cognitive abilities in three domains and ferent ways to measure noncognitive ability. Some
provides three new measures on cognitive abilities: commonly used indices of noncognitive ability
numeracy (NUM), literacy (LIT) and problem sol- include Rotter’s measure of locus of control
ving (PS). Each of the three skill domains measures (Rotter 1966), the Rosenberg self-esteem scale
related and yet somewhat distinct dimensions of an (Rosenberg 1965), the Five-Factor Model of
individual’s skill set and is represented by a 500- Personality (Muller and Plug 2006) and emotional
point scale (ranging from 0 to 500) with higher intelligence (Goleman 2000). For instance,
points denoting a higher level of desirable skills.6 Heckman, Stixrud and Urzua (2006) use Rotter
The definitions of each of the three domains pro- Locus of Control Scale and Rosenberg Self-
vided by the PIAAC are as follows (OECD 2013) : Esteem as noncognitive ability measures. The

3
We exclude individuals in agriculture industry and the military. These industries involve different mechanisms of becoming a leader, and are not considered
in the current study.
4
Individuals are defined as managers if they belong to one of these occupational groups: administrative and commercial managers (ISCO = 12), production
and specialized services managers (ISCO = 13), hospitality, retail, and other services manager (ISCO = 14).
5
These two measures are constructed based on two PIAAC questions: ‘Do you manage or supervise other employees?’ and ‘How many employees do you
supervise or manage directly or indirectly?’ The response options for the first question include ‘Yes’ and ‘No’ and for the second question include ‘1 to 5
people’, ‘6 to 10 people’, ‘11 to 24 people’, ‘25 to 99 people’ and ‘100 or more people’.
6
When presenting descriptive statistics and regression analyses, we divide all cognitive ability scores by 100. Thus, the numeracy, literacy, and problem
solving scores range from 0 to 5 in the analyses that we apply.
7
Numeracy tasks require, for instance, calculating the number of layers of tea candles packed in a box given other information or calculating the cost of
a trip from a motor-vehicle logbook.
8
The literacy test contains questions that require finding the right contact information in a simulated website, identifying the name of the author of
a particular book in a simulated library website, and extracting certain information from given paragraphs or tables.
9
Problem-solving questions include tasks such as reserving a meeting room on a particular date using a reservation system, organising a family get together,
and locating information on a spreadsheet and then e-mailing the requested information.
APPLIED ECONOMICS 5

Table 1. Descriptive statistics.


Manager = 1 Manager = 0
Variable Obs. Mean SD Obs. Mean SD Difference
Cognitive skills
Numeracy 187 2.950 0.446 1306 2.638 0.516 0.312*
Literacy 187 2.995 0.387 1306 2.783 0.453 0.213*
Problem solving 181 2.975 0.398 1144 2.801 0.403 0.174*
Noncognitive skills
Perseverance (yes = 1) 187 0.439 0.498 1306 0.340 0.474 0.099*
Openness to learning (yes = 1) 187 0.519 0.501 1306 0.442 0.497 0.077*
Social trust (yes = 1) 187 0.342 0.476 1306 0.263 0.441 0.079*
Education degree
Below high school 187 0.016 0.126 1306 0.077 0.267 −0.061*
High school 187 0.283 0.452 1306 0.508 0.500 −0.224*
Bachelors 187 0.471 0.500 1306 0.242 0.428 0.229*
Masters or Ph.D. 187 0.230 0.422 1306 0.173 0.378 0.057*
Others
Male (yes = 1) 187 0.647 0.479 1306 0.510 0.500 0.137*
Experience (years) 187 20.476 8.106 1306 19.087 9.133 1.389*
Married (yes = 1) 147 0.878 0.329 1061 0.744 0.437 0.134*
Either parent had a college degree (yes = 1) 187 0.492 0.501 1258 0.418 0.493 0.074*
Number of children 187 1.599 1.255 1305 1.530 1.314 0.069
Industry Dummies
Manufacturing (yes = 1) 187 0.241 0.429 1305 0.227 0.419 0.014
Trade (yes = 1) 187 0.214 0.411 1305 0.184 0.388 0.030
Service (yes = 1) 187 0.545 0.499 1305 0.589 0.492 −0.044
When presenting descriptive statistics and regression analyses, we divide all cognitive ability scores by 100. Thus, the numeracy, literacy, and problem
solving scores range from 0 to 5.
* indicates that the difference between leader and non-leader is significant at the 10% significance level.

Rotter scale measures the degree to which indivi- response to the statement ‘I like learning new
duals feel they control their life and the Rosenberg things’. We set Learn = 1 if an individual responds
scale measures perceptions of self-value. ‘to a very high extent’. Otherwise, Learn = 0. Social
In our study, based on the information available trust is measured with two statements: ‘There are
in the PIAAC data, we construct three noncogni- only a few people you can trust completely’ and ‘If
tive ability measures including perseverance, you are not careful, other people will take advantage
openness to learning and social trust. They are of you.’ Responses to the statements are again mea-
reflected in the five-factor model of personality, sured on a 5-point scale with the anchor points
which includes extraversion, agreeableness, con- ‘strongly agree’, ‘agree’, ‘neither agree nor disagree’,
scientiousness and openness to experience, and ‘disagree’ and ‘strongly disagree’. We define that
are likely to be important for leadership. individuals have good social trust (i.e. Strust = 1)
In our data, perseverance is measured with the if they answer ‘disagree’ or ‘strongly disagree’ to at
question ‘I like to get to the bottom of difficult least one of the statements; otherwise, Strust = 0.
things.’10 The respondents select their response Table 1 presents the descriptive statistics for the
from a 5-point scale with the anchor points ‘not at noncognitive abilities of leaders and non-leaders.
all’, ‘very little’, ‘to some extent’, ‘to a high extent’ Among leaders, individuals with strong perseverance,
and ‘to a very high extent’. To be categorized in the openness to learning and good social trust account for
high-perseverance group, the response has to be ‘to 44%, 52% and 34%, respectively. The percentages
a very high extent’. Thus, if a person responds to the drop to 34%, 44% and 26% for non-leaders. The
statement ‘to a very high extent’, we set Perse = 1. largest difference in percentage points is observed in
Otherwise, Perse = 0. We place a strict rule on perseverance. It is likely that leaders generally need
defining a high level of perseverance because the a high level of perseverance to solve complex pro-
responses per se are subjective, and a person may be blems when completing their work duties.
more likely to be overly positive in a self-evaluation. Education is measured by individuals’ highest
Similarly, openness to learning is measured by the academic degrees such as high school diplomas,

10
Other studies use similar questions to measure perseverance such as ‘I have achieved a goal that took years of work’ and ‘I have overcome setbacks to
conquer an important challenge’ (Duckworth et al. 2007).
6 T. TONG ET AL.

bachelor’s, master’s and Ph.D. degrees. Similar to abilities are also positively correlated, but the correla-
the patterns of cognitive and noncognitive abil- tions are much smaller (all correlations < .20) than
ities, leaders have higher levels of education than those within the cognitive abilities.
non-leaders. Among leaders, 70% have bachelor’s Figure 2 depicts the distributions of the three
degrees or higher, whereas, among non-leaders, cognitive ability measures separated by education
only 42% have a similar level of education. In level. Clearly, as education level increases, the
addition, Table 1 shows that leaders generally mean ability levels increase for all cognitive ability
have more years of work experience, and men measures as well. On average, individuals with
are more likely to be a leader (i.e. 65% of the master’s or PhD degrees have the highest abilities,
leaders). Family background variables include and those with no high school diploma have the
marital status, number of children and parental lowest. The largest increase is observed for numer-
education. In general, leaders are more likely to acy. Individuals with master’s or PhD degrees
be married and have higher educated parents. have a 38.7% higher average score in numeracy
Table 2 shows that the cognitive ability measures than those with less than a high school education.
are highly correlated with each other (all correlations It is interesting that problem-solving scores show
> .82). The correlation between numeracy and lit- the smallest gap between different educational
eracy is the highest. It is possible that comprehending levels.
a numeracy task requires good literacy. Table 2 For noncognitive abilities, as shown in Table 3,
further indicates that cognitive and noncognitive the proportion of individuals with higher

Table 2. Correlations between abilities.


Problem
Numeracy Literacy solving Perseverance Openness to learning Social trust
Numeracy
Literacy 0.886*
Problem solving 0.824* 0.858*
Perseverance 0.097* 0.061* 0.065*
Openness to learning 0.043 0.057* 0.066* 0.354*
Social trust 0.153* 0.154* 0.187* −0.011 0.141*
The numbers are correlation coefficients between different ability measures.
* indicates the correlation is significant at the 10% significance level.

Figure 2. Kernel distributions of three cognitive abilities by education levels.


APPLIED ECONOMICS 7

Table 3. Noncognitive ability and education level. noncognitive abilities, Cobb-Clark and Schurer
Education Perseverance Openness to learning Social trust (2012) demonstrates the remarkable stability of the
Below High School 23.5% 45.1% 17.6%
High School 21.5% 41.7% 20.2% Big Five traits of personality. Specht, Egloff and
College 32.1% 52.0% 37.0% Schmukle (2011) reported rather high levels of stabi-
Masters or Ph.D. 31.6% 55.1% 39.9%
lity for several aspects of personality. The implication
The percentages represent proportions of individuals with high noncogni-
tive abilities (i.e. perseverance = 1) at each education level. of the stability means that those ability measures
should not change much over one’s life, and thus the
noncognitive abilities is larger for individuals with labour market experience such as being a leader will
college or graduate degrees. For instance, for those not change them to an extent that it may influence the
with graduate degrees, the proportion with high outcomes of this study. The reverse causality should
perseverance is 8.1 percentage points higher than not bias our estimation results.
for those with less than a high school education. Therefore, in our baseline estimation, we apply
A similar pattern is found for openness to learning, the Linear Probability Model (LPM) and Probit
where the difference between graduate education estimation, and the results are reported in Table 4.
and below a high school education is 10 percentage The advantage of LPM is that it does not require
points. For social trust, the gap between the higher any assumptions about the underlying distribution
and lower educational levels is even larger. of the error term, and it is robust to heteroscedas-
According to Hooghe, Marien and de Vroome ticity. We first include the three cognitive ability
(2012), individuals with an attitude of trust are likely measures separately in Models 1–3.11 We found
to develop good social relationships and a clear aca- each of the three cognitive ability measures has
demic orientation, which in turn leads to higher positive and significant impact on leadership. By
education. comparing the coefficients of three cognitive abil-
ities, we find that the problem-solving ability has the
largest impact on leadership. More specifically,
IV. Cognitive, noncognitive abilities and when problem-solving score increases by 10 points,
leadership the probability of being a leader increases by
1.03 percentage points, while the number for
The three measures of cognitive abilities adopted by
numeracy and literacy is 0.92 and 0.74, respectively.
the OECD discussed above are considered to be com-
It is generally believed that problem solving ability
prehensive measures of one’s cognitive ability (OECD
is vital for leadership, which comes along with the
2013). Therefore, following the literature, we treat
need to solve problems on a regular basis at different
them as adequate proxy of unobserved abilities. In
occasions and in a variety of circumstances. The
addition, both cognitive and noncognitive abilities
problems a leader faces at work are often complex
develop early in life and reach stability in adolescence
or non-transparent. A high level of problem-solving
or early adulthood at latest (e.g. Cunha and Heckman
ability can help leaders define exactly what the pro-
2007; Borghans, Well, and Weinberg 2008). Thus,
blem is and help them generate appropriate solutions
those abilities can be considered as antecedents of
to the specific problem at hand (Mumford et al.
later outcomes including occupational status and lea-
2000). Connelly et al. (2000) found that complex
dership position. More specifically, for cognitive abil-
problem-solving skills, together with social judge-
ities, a large number of studies in phychology have
ment skills and leader knowledge, accounted for sig-
shown that intelligence exhibits an impressive level of
nificant variance in leadership even after controlling
stability from childhood throughout adulthood all the
for general intelligence, motivation, and personality.
way to old age (e.g. Deary et al. 2000; Schalke et al.
Because problem solving has the largest impact on
2013). Based on Cunha and Heckman (2007), indivi-
leadership compared to numeracy and literacy, we
duals’ cognitive ability becomes stable after age 10 or
use problem solving ability as a indicator of cognitive
so, and thus early interventions are much more effec-
ability measure in the following estimation.
tive than later interventions. Regarding the
11
Because three cognitive ability measures are highly correlated, incorporating them together in the model causes significant multicollinearity issues. The
coefficient of problem solving ability remains to be positive and significant, but literacy becomes negative and significant.
8 T. TONG ET AL.

Table 4. Ability and leader.


LPM LPM LPM LPM Probit
Numeracy 0.092***
(0.022)
Literacy 0.074***
(0.025)
Problem solving 0.103*** 0.100*** 0.097***
(0.031) (0.031) (0.028)
Perseverance 0.048* 0.037*
(0.025) (0.021)
Openness to learning −0.002 −0.001
(0.025) (0.021)
Social trust 0.023 0.018
(0.025) (0.022)
Education 0.135*** 0.148*** 0.138*** 0.132*** 0.124***
(0.024) (0.025) (0.025) (0.025) (0.023)
Experience 0.004*** 0.004*** 0.005*** 0.005*** 0.005***
(0.001) (0.001) (0.001) (0.001) (0.001)
Male 0.044** 0.057*** 0.061*** 0.060*** 0.057***
(0.022) (0.021) (0.023) (0.023) (0.021)
Married 0.030 0.036* 0.037* 0.035 0.047*
(0.020) (0.020) (0.022) (0.022) (0.027)
Parent education −0.005 −0.001 −0.002 −0.003 −0.006
(0.021) (0.021) (0.022) (0.022) (0.021)
Number of children 0.015* 0.014* 0.015* 0.014 0.010
(0.008) (0.008) (0.009) (0.009) (0.009)
Constants −0.349*** −0.325*** −0.417*** −0.425***
Industry dummies Yes Yes Yes Yes Yes
No. of observations 1162 1162 1040 1040 1040
R^2 0.080 0.074 0.074 0.076
F-statistics 10.535 10.015 9.503 7.225
Pseudo R^2 0.114
The Probit model presents the marginal effects evaluated at the means of the covariates.
For LPM estimation, the values in the parenthesis are robust standard errors.
Number of observations are different when we change the definitions of leadership.
*** denotes significance at 1%, ** denotes significance at 5%, and * denotes significance at 10%.

The results from Models 1 to 3 also show noncognitive ability measures, only persever-
that if an individual has a college degree, the ance shows a positive and significant effect.12
probability of being a leader increases by 14–- More specifically, a high degree of perseverance
15 percentage points, depending on different increases the probability of being a leader by
cognitive ability measures. Experience also has 4.8 percentage points. Perseverance (i.e. main-
a significant positive impact on leadership, taining high energy levels even in difficult cir-
which is in line with previous findings on the cumstances) is an essential trait for leaders.
relation between experience and leadership Leaders, who oftentimes face unexpected diffi-
(Lazear 2012). In addition, being married culties, obstacles and discouragement, need
increases the probability of being a leader, strong perseverance to lead the team and
while gender and parental education do not work toward the goal that is to be achieved.
seem to influence a person’s chance of being Additionally, Model 5 presents the results for
a leader. an alternative estimation method (i.e. Probit
After evaluating the impact of the cognitive estimation). The results of the Probit estima-
ability measures, we then add noncognitive tion in Model 5 are consistent with the OLS
ability measures in Model 4. The inclusion of estimation of the LPM model (Model 4).
noncognitive ability solves the omitted ability Considering the fact that many people may
problem to a large extent as some human capi- have a manager title but do not actually super-
tal variables that are included in Model 3 might vise a team, we test the robustness of our esti-
have been correlated with noncognitive ability mation results by constructing two additional
measures (e.g. perseverance). Out of the measures of leadership: supervising at least 5

12
If we include the noncognitive measures in the model separately, the results are similar. That is, perseverance is the only significant predictor out of the
noncognitive abilities.
APPLIED ECONOMICS 9

Table 5. Ability and leader-different measures of leadership. number of employee is challenging because it
Lead5 Lead10 needs more problem-solving, planning and
LPM Probit LPM Probit
working as well as cooperation. Our results
Problem solving 0.093** 0.099** 0.125*** 0.127***
(0.037) (0.039) (0.031) (0.030) show that a 10-point increase in problem-
Perseverance 0.050 0.048 0.047* 0.040* solving score increases the probability of
(0.031) (0.030) (0.025) (0.023)
Openness to learning −0.018 −0.017 −0.010 −0.006 Lead5 by 0.93 percentage points, and it
(0.030) (0.030) (0.025) (0.023) increases to 1.25 for Lead10 in LPM estimation.
Social trust 0.015 0.014 −0.006 −0.007
(0.031) (0.031) (0.025) (0.024) Additionally, perseverance changes from insig-
Education 0.055* 0.053* 0.055** 0.053**
(0.033) (0.032) (0.027) (0.024)
nificant to significant when we use the higher
Experience 0.004*** 0.005** 0.004*** 0.004*** criteria for leaders (i.e. Lead10)
(0.002) (0.002) (0.001) (0.001)
Male 0.063** 0.064** 0.014 0.013 Our results are consistent with previous litera-
(0.029) (0.029) (0.024) (0.022) ture. For instance, Adam et al. (2018) showed that
Married 0.046 0.054 0.004 0.004
(0.031) (0.034) (0.026) (0.026) both cognitive and noncognitive abilities of CEOs
Parent Education −0.034 −0.038 −0.017 −0.021 in large firms are higher than those in small firms,
(0.029) (0.029) (0.024) (0.022)
Number of children 0.017 0.016 0.016 0.013 which indicated that the importance of ability
(0.011) (0.012) (0.010) (0.009) increases as a leader supervises more employees.
Constants −0.256** −0.388***
Industry dummies Yes Yes Yes Yes
No. of observations 934 934 934 934
R^2 0.043 0.056 V. A further investigation of the endogeneity
F-statistics 4.657 4.830
Pseudo R^2 0.053 0.087 issue
For LPM estimation, the values in the parenthesis are robust standard
errors. The above baseline estimation provides strong evi-
*** denotes significance at 1%, ** denotes significance at 5%, and * dence that cognitive and non-cognitive abilities have
denotes significance at 10%.
a significant effect on an individual’s chance of being
in a leadership position and on the intensity of the
people (Lead5) and 10 people (Lead10). The leadership role measured by the number of people
results are presented in Table 5, with Models supervised. Our estimation is based on the exogeneity
1 and 3 showing the results of LPM, and assumption of those ability measures. We believe that
Models 2 and 4 displaying the results of the exogeneity assumption is reasonable given the
Probit estimation. stable nature of one’s abilities. Moreover, because
Overall, human capital measures in Table 5 we compare estimates across ability measures and
exhibit the same pattern with previous estima- across different intensity of leadership, any potential
tions in Table 4. Problem-solving ability, edu- bias would be mitigated or differenced out in such
cation and experience remain significant comparisons.
predictors of leadership. In addition, none of Yet it is desirable to investigate the potential endo-
the noncognitive ability measures is significant geneity problem as well as how sensitive the above
when Lead5 (i.e. leading more than 5 employ- results are. We will proceed in two steps: we first relax
ees) is used to measure leadership, and perse- the assumption that the ability variables are the mea-
verance becomes positive and significant when surement of one’s true ability and instead treat them
Lead10 (i.e. leading more than 10 employees) is as indicators for ability and then use Multiple
used to measure leadership. A high degree of Indicator (MI) approach; and second, we allow for
perseverance increases the probability of being the possibility of the reverse causality and apply the
a leader by 4.0–4.7 percentage points. Instrumental Variable approach to estimate the
Additionally, Table 5 shows that the effects model.
of both cognitive and noncognitive ability MI approach imposes a weaker assumption
become more critical when the number of because ability measures do not have to perfectly
employees supervised by the leader increases capture the true ability but only need to be indicators
from 5 to 10, which is to be expected given of it (see e.g. Hanushek et al. 2015). More specifically,
that we are more selective in defining what based on the MI technique, we treat one ability mea-
constitutes a leader. Supervising a greater sure as the indicator and others as instruments in the
10 T. TONG ET AL.

estimation.13 In theory, the roles of all three indica- Table 6. Multiple indicator and instrument variable approach.
tors, numeracy, literacy and problem solving are Manager
MI IV
interchangeable in the MI estimation. Because pro-
Problem solving 0.088** 0.046
blem-solving ability is likely to be a more compre- (0.037) (0.101)
hensive measure of one’s ability than the other two, Perseverance 0.048* 0.048*
(0.025) (0.025)
we use it as the indicator of cognitive ability, and use Openness to learning −0.002 0.000
numeracy and literacy scores as instruments in the (0.024) (0.024)
Social trust 0.025 0.022
estimation. For noncognitive measures, we keep them (0.025) (0.027)
College education 0.135*** 0.146***
in the model without applying the MI and IV treat- (0.025) (0.033)
ments. The main reason is that, based on the five- Experience 0.005*** 0.004***
(0.001) (0.001)
factor model of personality, an individual’s personal- Constants −0.209** −0.275
ity (noncognitive) can be classified into five categories Control variables Yes Yes
No. of observations 1040 1031
including extraversion, agreeableness, conscientious- Adjusted R2 0.076 0.074
ness, openness to experience and neuroticism (Muller Endogeneity test (Chi2) 0.315(0.575)
Over-identification test (Chi2) 2.282(0.320)
and Plug 2006). Each category represents distinguish- The values in the parenthesis are robust standard errors.
ing personality aspects that could have independent *** denotes significance at 1%, ** denotes significance at 5%, and *
denotes significance at 10%.
impact on one’s labour market outcomes. Therefore,
our three measures of noncognitive ability (i.e. perse-
verance, openness to learning and social trust) can fit
into the first four categories, and thus can be used as how such a possibility may affect our results, we
exogenous regressors.14 apply the Instrument Variable (IV) estimation.
The results of the MI estimation are reported in Moreover, the IV estimator can help resolve the
Table 6. The estimates are in line with previous endogeneity problem caused by both measurement
estimation in Tables 4 and 5. Human capital mea- error and reverse causality, although with stronger
sures including problem solving ability, persever- requirement for finding good instruments.
ance, education and experience remain to be The IV estimation requires that an instru-
significant predictors of leadership in the MI ment should be correlated with abilities mea-
approach, and the magnitudes are similar. The sured based on test scores but not with the
results indicate that our results above are quite error term. Due to data limitation, such instru-
robust. The three new measures of cognitive abil- ments are difficult to find. In particular, the
ities are multi-dimensional and comprehensive, endogenous variable is the problem solving
and they adequately capture one’s ability. ability measure in the model, and we need to
The above MI estimation helps resolve the poten- find instruments, that is, correlated with it but
tial measurement problem for true abilities. not with the error term. We first adopt two
Additionally, there is another potential endogeneity instruments based on the PIAAC questionnaire.
problem caused by the possible reverse causality, i.e. The first one is based on the question ‘whether
the measured ability may be related to the experi- the test language is the same as the language
ence of being a leader as one learns from the job. usually spoken at home’. Specifically, language
Because of the stability of one’s ability, such spoken at home may negatively affect test
a reverse influence should be relatively small, espe- scores if the home language is not English.
cially given the age when an individual was put in However, they should be exogenous to the indi-
a leadership position.15 However, in order to see vidual’s inner ability.

13
The Multiple Indicator approach is closer to classic approach of errors-in-variables but is less restrictive. More specifically, for the MI approach, in equation
y ¼ α0 þ xβ þ γq þ v, where q represents the omitted ability variable. Suppose we have multiple indicators of q from q1 to qn , and they are highly
correlated. Indicator q1 can be written as q1 ¼ δ0 þ δ1 q þ a1 , where covðq; a1 Þ ¼ 0 and covðx; a1 Þ ¼ 0. If we rewrite q as a function of q1 and then
substitute it back into the original equation, we then have y ¼ α1 þ xβ þ γ1 q1 þ v  γ1 a1 , where γ1 ¼ δγ1 . Then, we express each of the rest of the
indicators (i.e. q2 to qn ) as a function of q, so we obtain an error term for each indicator of q (i.e. a1 to an ). Since all the common component in indicators
has been controlled by q, MI approach assumes that the error terms are uncorrelated, then q2 to qn become valid instruments for q1 .
14
Because of data limit, we do not have useful information in investigating the potential endogeneity issue for noncognitive ability measures.
15
The average age of leaders in our sample is 42.
APPLIED ECONOMICS 11

The second instrument is based on the question VI. Conclusion


‘whether an individual was interrupted by some
In this study, we investigate the effects of an indivi-
other activity, task, or event’. If individuals were
dual’s human capital on leadership. We employ var-
interrupted during survey, they may not be able to
ious measures of human capital based on cognitive
concentrate on the tests, and, thus, have lower test
and noncognitive ability, in addition to the traditional
scores. Because these interruptions are likely to be
education-based and experience-based human capital
beyond the individual’s control, they are exogen-
measures. We also adopt different leadership mea-
ous to the person’s real ability and leadership
sures that can represent the degree of the leadership
related factors in the error term.
to investigate the human capital impact. We find that
In addition, based on previous literature, we use
problem-solving ability shows the strongest effect on
industry average abilities as instruments (Lin et al.
leadership among cognitive measures, and persever-
2011). Specifically, if the endogeneity problem is spe-
ance exhibits a positive and significant effect on lea-
cific for individuals, but not for industries, then net-
dership among noncognitive ability measures. In
ting out this individual-specific component yields
addition, as an individual supervises more employees,
a measure that only depends on the underlying char-
the effects of cognitive and noncognitive abilities
acteristics of particular industries. In our case, the
become more pronounced. Moreover, we explore
required ability may be particular to industries
the potential endogeneity issue caused by measure-
because of the nature of representative jobs of the
ment error and reverse causality via Multiple-
industry, thus an individual’s test scores may be cor-
Indicator estimation and Instrumental Variable esti-
related with others who also work in this industry.
mation to test the robustness of our results, and the
Moreover, according to previous studies (e.g. Cunha
tests generally support our findings.
and Heckman 2007), cognitive ability is well devel-
Leadership is vital to the development of any orga-
oped at a relatively young age at around age 10, test
nisation. By providing a quantitative analysis about
scores of other individuals in the same industries
the impact of human capital on leadership, our results
should not have a direct impact on this individual’s
should shed new light on identifying individuals with
real ability. Therefore, we use industry average test
strong leadership potentials. In the US, only around
score (excluding the individual himself) as instru-
30–50% of large corporations have a specific frame-
ments for the cognitive ability measure.
work for the identification of leadership potential
The results of the IV estimation are reported
(Dries and Pepermans 2012). Therefore, it is desirable
in model 2 in Table 6. First, the over-
to have a set of criteria and guidelines that firms can
identification test could not be rejected, indi-
use to assess leadership potential for efficient job
cating that there is no evidence against the
assignments and promotions.
validity of our instruments. Additionally, the
results are generally consistent with the above
estimation. In particular, the perseverance, edu-
cation and experience are significant predictors Disclosure statement
for leadership with positive signs, while the No potential conflict of interest was reported by the authors.
problem solving ability becomes insignificant,
possibly due to the inefficiency of the IV esti-
mation. Moreover, the Hausman tests cannot Funding
reject the null hypothesis of exogeneity of the
This study was funded by the National Natural Science
ability measures in the model. Thus, the poten-
Foundation of China (71773151) and the Fonds National
tial endogeneity bias does not seem to be de la Recherche Luxembourg (ATTRACT "ASKI21").
significant.
A note of caveat in our IV estimation is that we
cannot find best instrumental variables. Similar to References
the claim in Lin et al. (2011), our results at least Adams, R., M. Keloharju, and S. Knüpfer. 2018. “Are CEOs
demonstrate some strong and robust associations Born Leaders? Lessons from Traits of a Million
between human capital and leadership. Individuals.” Journal of Financial Economics.
12 T. TONG ET AL.

Borghans, L., B. Well, and B. A. Weinberg. 2008. Outcome and Social Behavior.” Journal of Labor
“Interpersonal Styles and Labor Market Outcomes.” The Economics 24 (3): 411–482.
Journal of Human Resources 43: 815–858. Hooghe, M., S. Marien, and T. de Vroome. 2012. “The
Card, D. 1999. “The Causal Effect of Education on Earnings.” In Cognitive Basis of Trust. The Relation between
Handbook of Labour Economics. Vol. 3, edited by Education, Cognitive Ability, and Generalized and
O. Ashenfelter and D. Card, 1801–1863. North-Holland, Political Trust.” Intelligence 40: 604–613.
Amsterdam: Elsevier. Krueger, A. B., and D. Schkade. 2008. “Sorting in the Labor
Chen, W., W. A. Grove., and A. Hussey. 2017. “The Role of Market: Do Gregarious Workers Flock to Interactive Jobs?”
Confidence and Noncognitive Skills for Post-Baccalaureate Journal of Human Resources 43: 859–883.
Academic and Labor Market Outcomes.” Journal of Kuhn, P., and C. Weinberger. 2005. “Leadership Skills and
Economic Behavior & Organization 138: 10–29. Wages.” Journal of Labor Economics 23: 395–436.
Cobb-Clark, D., and S. Schurer. 2012. “The Stability of Lazear, E. P. 2012. “Leadership: A Personnel Economics
Big-Five Personality Traits.” Economics Letters 115: 11–15. Approach.” Labour Economics 19 (1): 92–101.
Connelly, M. S., J. A. Gilbert, S. J. Zaccaro, K. V. Threlfall, Lin, C., P. Lin, F. M. Song, and C. Li. 2011. “Managerial
M. A. Marks, and M. D. Mumford. 2000. “Exploring the Incentives, CEO Characteristics and Corporate Innovation
Relationship of Leadership Skills and Knowledge to Leader in China’s Private Sector.” Journal of Comparative
Performance.” The Leadership Quarterly 11 (1): 65–86. Economics 39 (2): 176–190.
Cunha, F., and J. Heckman. 2007. “The Technology of Skill Lindqvist, E. 2012. “Height and Leadership.” Review of
Formation.” The American Economic Review 97 (2): 31–47. Economic Statistics 94 (4): 1191–1196.
Deary, I. J., L. J. Whalley, H. Lemmon, J. R. Crawford, and Lindqvist, E., and R. Vestman. 2011. “The Labor Market
J. M. Starr. 2000. “The Stability of Individual Differences in Returns to Cognitive and Noncognitive Ability: Evidence
Mental Ability from Childhood to Old Age: Follow-Up of the from the Swedish Enlistment.” American Economic
1932 Scottish Mental Survey.” Intelligence 28: 49–55. Journal: Applied Economics 3 (1): 101–128.
Dinh, J. E., R. G. Lord, W. L. Gardner, J. D. Meuser, Lundberg, S. 2013. “The College Type: Personality and
R. C. Liden, and J. Hu. 2014. “Leadership Theory and Educational Inequality.” Journal of Labor Economics 31
Research in the New Millennium: Current Theoretical (3): 421–441.
Trends and Changing Perspectives.” The Leadership Muller, G., and E. Plug. 2006. “Estimating the Effect of
Quarterly 25: 36–62. Personality on Male and Female Earnings.” Industrial
Dries, N., and R. Pepermans. 2012. “How to Identify Leadership and Labor Relations Review 60 (1): 3–22.
Potential: Development and Testing of A Consensus Model.” Mumford, M. D., E. M. Todd, C. Higgs, and T. Mclntosh. 2017.
Human Resource Management 51 (3): 361–385. “Cognitive Skills and Leadership Performance: The Nine
Duckworth, A. L., C. Peterson, M. D. Matthews, and Critical Skills.” Leadership Quarterly 28: 24–39.
D. R. Kelly. 2007. “Grit: Perseverance and Passion for Mumford, M. D., S. J. Zaccaro, F. D. Harding, T. O. Jacobs,
Long-Term Goals.” Journal of Personality and Social and E. A. Fleishman. 2000. “Leadership Skills for
Psychology 92 (6): 1087–1101. a Changing World: Solving Complex Social Problems.”
Eren, O., and S. Ozbeklik. 2013. “The Effect of Noncognitive The Leadership Quarterly 11 (1): 11–35.
Ability on the Earnings of Young Men: A Distributional OECD. 2013. OECD Skills Outlook 2013: First Result from the
Analysis with Measurement Error Correction.” Labour Survey of Adult Skills. OECD Publishing, France.
Economics 24: 293–304. Rosenberg, M. 1965. Society and the Adolescent Self-Image.
Gibbons, R., and M. Waldman. 2006. “Enriching a Theory of Princeton. NJ: Princeton University Press.
Wage and Promotion Dynamics inside Firms.” Journal of Rotter, J. B. 1966. “Generalized Expectancies for Internal
Labor Economics 24 (1): 59–107. versus External Control of Reinforcement.” Psychological
Glewwe, P., Q. Huang, and A. Park. 2017. “Cognitive Skills, Monographs: General and Applied 80 (1): 1–28.
Noncognitive Skills, and School-To-Work Transitions in Schalke, D., M. Brunner, C. Geiser, F. Preckel, U. Keller,
Rural China.” Journal of Economic Behavior & M. Spengler, and R. Martin. 2013. “Stability and Change in
Organization 134: 141–164. Intelligence from Age 12 to Age 52: Results from the
Goleman, D. 2000. “Leadership that Gets Results.” Harvard Luxembourg MAGRIP Study.” Developmental Psychology 49:
Business Review 78: 4-17. 1529–1543.
Grönqvist, E., and E. Lindqvist. 2016. “The Making of Specht, J., B. Egloff, and S. C. Schmukle. 2011. “Stability
a Manager: Evidence from Military Officer Training.” and Change of Personality across the Life Course: The
Journal of Labor Economics 34 (4): 869–898. Impact of Age and Major Life Events on Mean-Level
Hanushek, E. A., G. Schwerdt, S. Wiederhold, and and Rank-Order Stability of the Big Five.” Journal of
L. Woessmann. 2015. “Returns to Skills around the World: Personality and Social Psychology 101: 862–882.
Evidence from PIAAC.” European Economic Review 73: Yukl, G. 2008. “How Leaders Influence Organizational
103–130. Effectiveness.” The Leadership Quarterly 19: 708–722.
Heckman, J. J., J. Stixrud, and S. Urzua. 2006. “The Effects of Zaccaro, S. J. 2007. “Trait-Based Perspectives of Leadership.”
Cognitive and Noncognitive Abilities on Labor Market American Psychologist 62 (1): 6–16.

You might also like