Leadership Skills and Wages*!
Peter Kuhn and Catherine Weinberger
University of California, Santa Barbara
First draft: September 2000
This revision: November 2004
Controlling for cognitive skills, men who occupied leadership positions in high school
earn more as adults. The pure leadership-wage effect varies, depending on definitions and
time period, from 4 percent to 33 percent. This effect is not an artifact of measurement
error in cognitive skills or differences in a wide array of other physical or psychological
traits. High-school leaders are more likely to occupy managerial occupations as adults,
and leadership skills command a higher wage premium within managerial occupations
than elsewhere. Finally, it appears that leadership skills may be fostered by exposure to
high school leadership opportunities.
* We thank the Institute for Social, Behavioral and Economic Research at UC Santa
Barbara and the Spencer Foundation for research support. Dan Hamermesh, and
Christopher Jencks provided helpful comments, as did seminar participants at University
of British Columbia, the Society of Labor Economists, Universitat Pompeu Fabra,
Universitat Autonoma Barcelona, University of Heidelberg, the Institute for the Study of
Labor (IZA), the University of Arizona, McMaster University, UCSB, the Santa Fe
Institute and the Goleta Rotary Club. Chau Do provided excellent research assistance.
The authors can be reached at [email protected], and [email protected]
Today’s businesses, workers and educational institutions are making large
investments in identifying and developing a personal characteristic called “leadership”. A
recent search of books in print revealed over 5000 titles containing the word “leadership,”
the majority of which are training or self-help manuals aimed at a business audience.
Academic articles on the development and measurement of leadership have filled the
Leadership Quarterly since 1990. Elite universities are placing increased emphasis on
high school leadership in their admissions decisions (Morse, 2001), top business schools
send MBA students to Marine-run “boot camps” for the express purpose of fostering
leadership skills (Scannell, 2001), and Fortune 1000 companies are offering leadership
skills training to rapidly growing numbers of employees (Lawler, et. al. 2001).1
Compared to the kinds of cognitive skills measured by tests like the AFQT —on
which economists have focused almost exclusively in studies of earnings determination—
interpersonal skills including leadership rank much higher in the list of employee
attributes employers say they seek in workers. For example, employers of new college
graduates report that communications skills, motivation/initiative, teamwork skills, and
leadership skills are all more highly valued than academic achievement/GPA (National
Association of Colleges and Employers, 2000). Similarly, employers seeking less
educated entry-level workers report a growing importance of “soft skills” (Moss and Tilly
2001). Labor economists have noted a widespread change in work organization towards
increased employee involvement and teamwork (Lindbeck and Snower 2000). Goleman
(1997) argues persuasively that, relative to academic ability, “emotional intelligence”—a
constellation of social skills including leadership—is an increasingly important factor in
Is there such a thing as “leadership skill”? Is it rewarded in labor markets, and
can it be measured and distinguished from analytical or cognitive skills as well as from
other forms of human capital such as education and experience? To examine these
issues, we use three data sets that are representative of the national population of high
school students in 1960, 1972 and 1982. All three surveys include information on family
background, scores from standardized cognitive tests taken in high school, and labor
market outcomes approximately ten years after high school. All three include a
“behavioral” measure of leadership taken during high school (acting as a team captain or
club president) while two also contain other measures of leadership skill.
Our basic methodology regresses adult wages on indicators of leadership skills
taken before labor market entry (i.e. in high school), controlling for standard measures of
cognitive ability, family background, and (in some specifications) high school fixed
effects. Our use of pre-labor market measures of leadership skill allows us to avoid
certain kinds of endogeneity, for example the possibility that individuals who do well in
the labor market for some unrelated reason (e.g. a “lucky” promotion) might begin to
develop those very leadership skills whose effects we wish to measure.2 The control for
cognitive skills ensures that we are capturing only the additional effect of a
1 Between 1990 and 1999, the proportion of Fortune 1000 companies offering leadership skills training to at least 40
percent of their employees tripled, from 11 percent to 32 percent (Lawler, et. al. 2001).
2 Even more to the point, in wage regressions using concurrent self-assessed measures of leadership, accidentally-
successful individuals might simply interpret their success as evidence of exceptional leadership ability.
“noncognitive” skill such as leadership.3 We conduct a variety of tests to determine
whether leadership, as measured here, is in fact a distinct skill that is related to the
management of people. These include examining interaction effects between leadership
skills and other determinants of earnings, adding controls for other psychological and
physical characteristics for which leadership may be acting as a proxy, and estimating the
effect of leadership skills on occupational assignment.
Our main results are as follows. First, controlling for family background, for
standard measures of human capital, for mathematics test scores, and for all factors
associated with the high school attended (via high-school fixed effects), individuals who
exhibited leadership propensities in high school earn significantly more about ten years
later. This effect is observed in all three data sets and in all econometric specifications.
Its size varies with the nature of the leadership measure and across years, ranging from
four percent for one indicator of leadership in 1971 to 33 percent for a more exclusive
indicator in 1991. Second, less than one quarter of the leadership effect on adult earnings
operates through differences in educational attainment after high school. This, together
with other results summarized below, suggests that our leadership measures are not acting
as proxies for other personal attributes, such as persistence, a low discount rate, or
measurement error in cognitive skills. Third, the skill identified by this leadership
measure is distributed in a manner that is largely orthogonal to cognitive skills, as
measured by mathematics test scores. Relatedly, we find that the marginal effect of
leadership skills on earnings does not vary with the respondent’s level of education, or
3 The term “non-cognitive skills” seems to have been adopted by economists to distinguish personality measures or
social skills from standard math and verbal skill measures, which have generally been referred to as “cognitive skills”.
Psychologists sometimes object to this distinction, rightly pointing out that cognition is involved in the use of these
other skills as well.
with the level of math scores, working just as strongly for low-math-score and low-
education individuals as for others.
In the course of our analysis, we conduct several additional exercises to ensure
that our measure of leadership is not acting as a proxy for some other physical or
psychological characteristic. In two of our data sets we are able to add controls for
“beauty” or physical attractiveness; in one we can also control for height. Even though
“beauty” affects earnings exactly as expected (Hamermesh and Biddle 1994, Averett and
Korenman 1996, Biddle and Hamermesh 1998), in neither case does accounting for the
effects of beauty change the estimated leadership coefficient. Adolescent height does not
have a significant effect on adult earnings in our data. We also present instrumental
variables evidence that argues against interpreting our estimated leadership effect as a
proxy for measurement error in cognitive ability.
Controlling for a very rich set of psychological characteristics available in the
Project Talent data (including “vigor”, “maturity”, “self-confidence”, and “tidiness”)
explains very little of the leadership coefficient. Significantly in our opinion, the largest
reduction in the leadership coefficient occurs when we add a control for “sociability” (a
self-assessed measure of enjoyment of being around people). We take this as
confirmation that our leadership variable is capturing some sort of social skill. The final
piece of evidence that our leadership measure truly captures an ability to lead other
people involves occupational outcomes: ceteris paribus, high-school leaders are more
likely to be managers as adults. Just as important, while high school leadership is
rewarded in both managerial and nonmanagerial occupations, the marginal market value
of extra leadership skills is considerably greater within managerial occupations than in
The research in this paper forms part of a growing literature on the role of non-
cognitive skills in wage determination. Much of this literature is reviewed in Bowles,
Gintis and Osborne (2001); it includes papers examining the effects on adult outcomes of
childhood measures of aggression and withdrawal (Osborne 1999); of both child and
adult measures of “locus of control” (Goldsmith, Veum and Darity 1997, Osborne 1999,
and Coleman and DeLeire 2000); of measures of motivation (Goldsmith, Veum and
Darity 2000), and of indicators of behavioral problems during high school (Cawley,
Heckman and Vytlacil 2001). The current paper differs from this research in its focus on a
skill that firms explicitly say they seek in new hires, and that firms actually pay to foster
among their incumbent workers.4 Our work also relates to a recent group of papers on
the labor market effects of high school athletic participation (Anderson 2000; Barron,
Ewing and Waddell (2000); Eide and Ronan 2000; Stevenson 2000). 5 All of these papers
find that participation in high school athletics has small positive effects on adult wages.
Our work differs from these papers in its focus on leadership within and beyond sports,
and in the richer set of measures and additional sensitivity testing that is possible in our
three data sets, especially the much larger and more comprehensive Talent data.
2. Data and Methods
As noted, we use three data sets: Project Talent (1960), The National Longitudinal
4 Our paper also focuses much more on the top of the wage distribution, while much existing research focuses on
problems employers encounter when employing low-wage workers.
Study of the Class of 1972 (NLS72), and the sophomore cohort of High School and
Beyond (HSB 1982 seniors). All are surveys of a representative sample of U.S. high-
school students, who were re-interviewed between 9 and 13 years after leaving high
school. In all three data sets, we restrict our attention to white men. This restriction
allows us to study the labor market valuation of leaders without the confounding effects
of race or gender discrimination or of the changing roles and expectations of women
during this time period. Aside from providing information from three different decades,
each of the data sets has some unique advantages that contribute to our understanding of
The earliest of the data sets, the Project Talent study of 1960 High School
Students, is also by far the largest. The students in this study were surveyed during high
school in 1960, and followed longitudinally for eleven years after high school. Talent has
a much more complete inventory of personality, behavioral, and ability measures than all
the more recent nationally-representative panel surveys, including the PSID, NLSY,
NLS72, and HSB. During the base year, over 400,000 students —approximately five
percent of all U.S. high school students—devoted two full days to the study, during which
they responded to a 400-question survey and were given comprehensive cognitive and
psychological assessments. Of these 400,000 students, approximately 150,000 were men
in grades 10 through 12. Our focus in this paper is on the representative sample of the
latter group who were selected for re-interview eleven years after finishing high school
5 This literature, in turn, relates to (but does not always cite) an older literature, mostly in psychology, sociology and
education journals, linking participation in high school extracurricular activities with positive educational, behavioral
and economic outcomes (see Eccles and Barber 1999 for a thorough review of the older literature; recent contributions
include Camp 1990, Gerber 1996, and Barber, Eccles and Stone 2001). In a wide-ranging study of the earliest of our
data sets (Talent), Jencks (1977) detected a wage effect of high school leadership but did not explore its origins or
structure in much detail.
(in 1971, 1972, or 1973) and in addition where white, employed, and had graduated from
high school by that time.6 Aside from its size and focus on multiple years of high school,
a key distinctive feature of Talent is the framing of its leadership activity measure: in
contrast to NLS72 and HSB (which refer to leadership in the senior year only), the Talent
survey counts all leadership activity in the past three years. The Talent survey also
includes an independent measure of leadership taken from the personality inventory.
The NLS72 followed 1972 high school seniors until 1986. HSB followed 1980
high school sophomores until 1992. In both of these studies, the students were asked
about leadership roles and club/sports participation during the senior year of high school
only. Other variables are similar to those available in Project Talent. Senior year math
scores, parents’ education and own educational attainment could be coded in exactly
comparable ways.7 Unfortunately, the 1992 HSB survey includes only annual, rather than
hourly, earnings. It does, however, contain annual earnings for several years (1988-1991)
and a question about physical attractiveness not available in the other surveys, as well as
information on participation in three other high school leadership activities that allows us
to construct a more continuous measure of leadership.
Two of the three data sets used in our analysis contain plausible measures of
leadership in addition to the team and club activity measures. Talent has a self-assessed
leadership scale (silead) which was constructed by the original Talent investigators from
the respondent’s indication that each of the following five statements described himself
“extremely well” or “quite well”:
6 Wage equations are restricted to those with hourly earnings between $1 and $50. This includes 99.5% of all
respondents with hourly earnings reported.
7 Math scores were made comparable by coding them as percentile scores among all white men in the same grade who
were tested in the base year.
1. I am the leader in my group.
2. I am influential.
3. I have held a lot of elected offices.
4. People naturally follow my lead.
5. I like to make decisions.
These questions were contained in a 150-question Student Activities Inventory designed
to assess 13 different personality traits. An individual’s score simply sums the number of
positive responses.8 About 38 percent of men in our (weighted) sample agreed with none
of the above statements, while only 3 percent agreed with all five. In some of our
analysis, five dummy variables distinguish the exact silead score. For other treatments,
silead is transformed into a standardized score with mean zero and standard deviation one
within each grade cohort.
In HSB, we were able to construct a similar leadership measure by summing the
number of distinct types of leadership activities in which an individual had participated
during the past year. Two of these are the team captain and club president activities
already described. The other three are:
1. Spoke before a group of 50 or more.
3. Headed a group problem-solving session.
3. Chaired a meeting.
Aside from checking the robustness of our results to the measurement of
leadership, these two alternative leadership measures have the advantage of being more
8 Unfortunately we do not have access to data on the five questions separately, so we are forced to use this “scale” as it
was constructed by the original Talent investigators. See note 21 below. Silead is correlated with the leadership
activities measure used in the previous section: those who were both captain and president were twice as likely as
members of the full sample to agree with four or five of the silead statements.
continuous than our indicators of club presidency or captainship. This will allow us to
explore the form of the leadership-earnings relationship in more detail, focusing
especially on those individuals with rare and very high endowments of leadership skills.
Because of the long time interval covered by each of these studies, sample
attrition could potentially be a serious problem. The treatment of attrition in the NLS-72
and HSB is already familiar to economists. Weights were constructed by the NCES
based on the assumption that non-respondents are similar to the respondents with the
same observable characteristics. The designers of the Project Talent study took a more
rigorous and costly approach. In this study, at each resurvey a representative subsample
of non-respondents was randomly selected and aggressively pursued. For the white men
in this special sample, the researchers achieved a 100% success rate, making it fully
representative of (initial) non-respondents.9 In our analysis we use weights derived from
this aggressively-followed sample to adjust for attrition bias in the larger sample. These
weights simply up-weight the responses of this aggressively-followed sample by the
inverse of their share of the (initially) non-responding population, to generate means that
are representative of the entire 1960 high school population.10
Descriptive statistics for all three data sets are presented in Table 1. The complete
Talent sample includes all 24041 white men who were working at the eleven year
interview date, whose measured wages fell between one dollar and fifty dollars per hour,
and who had non-missing data on math scores and team/club membership/leadership.11
9 As a last resort, collection agencies were used to locate recalcitrant respondents.
10 Special sample respondents represent 6.7 percent of individuals in the sample used for analysis, but 44.1 percent of
the weighted sample. To avoid artificially inflating the precision of the estimates, Stata’s p-weight option was used
throughout the paper.
11 Despite significant sample attrition, Table 1 shows that our usable Talent sample is more than eight times as large as
our NLS72 sample and more than ten times the size of our HSB sample.
- Peter Kuhn and Catherine Weinberger
- This revision: November 2004