An Estimate of Variance Due To Traits in Leadership: David A. Kenny Stephen J. Zaccaro

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 8

Journal of Applied Psychology Copyright 1983 by the

1983, Vol 68, No 4, 678-685 American Psychological Association, Inc

An Estimate of Variance Due to Traits in Leadership


David A. Kenny Stephen J. Zaccaro
University of Connecticut Virginia Polytechnic Institute and
State University

The rejection by psychologists of the trait explanation of leadership emergence can


be attributed to the results of two major types of studies, attempts to identify
leadership traits and the use of rotation designs Numerous reviews of the literature
consistently note the failure to isolate a specific leadership trait Using a rotation
design, Barnlund (1962), after varying both the task and member composition of
groups and computing the correlation of leadership rank in one group with the
average leadership ranks received in all other groups, reported that leadership
emergence varied across group situations Thus, his data suggested that leadership
is not a stable characteristic The present study, however, uses the Social Relations
Model (Kenny, 1981) to reexamine Barnlund's conclusion. This reanalysis suggests
that between 49% and 82% (representing lower- and upper-bound estimates) lead-
ership variance can be attributed to some stable characteristic It was speculated
that this characteristic, rather than being a traditional personality trait, may actually
involve the ability to perceive the needs and goals of a constituency and to adjust
one's personal approach to group action accordingly

Groups pervade many aspects of social life is the crux of the interactional theory), "the role he achieves
The behavior of individuals interacting in is determined not by his personal qualities in the abstract
but by his standing in relation to his fellow members in
classrooms, in work groups, on juries, on ath- the special qualities required by the particular group goal
letic teams, and in therapy sessions is heavily or situation" (Gibb, 1969, p 268)
influenced by group-related phenomena (e.g.,
norms, influence pressures, and cohesiveness). The interactional interpretation states then
The effectiveness of such groups often depends that leadership occurs as a function of the in-
on the behavior of group leaders. Whether ap- teraction between the group's needs and the
pointed to the group or emerging from the member resources available to satisfy these
group process, leaders exert a major influence needs (Jenkins, 1947; Murphy, 1941) The
on directions their groups may take. member who possesses those resources that
can best aid the group in reaching its ordained
The current leadership literature empha-
goals is more likely to emerge as the leader.
sizes an interaction hypothesis in explaining
The initial processes involved in the early
who emerges as a group leader. The interaction
stages of group development concern the iden-
hypothesis as formulated by Sherif (1948, p.
tification of such member resources and sub-
456) states-
sequent group coalescence around the chosen
Leadership is an interactional phenomenon arising when
leader (Geier, 1967; Hollander & Julian, 1970).
group formation takes place The relative role an
There is the explicit rejection by interaction
individual member assumes within the group is determined
both by the role needs of the group and by the particular theorists of a purely trait explanation of lead-
attributes of personality, ability, and skill which differentiate ership emergence. In an early statement of the
him from other members of the group However (and this interaction hypothesis, Murphy (1941) stated
that, "Leadership does not reside in the person.
It is a function of the whole situation" (p.
Support for this article was provided in part from Na- 674). Contrary to the "great men" (or women)
tional Science Foundation Grants BNS-7913820 and BNS- theory of leadership, the specific circumstances
8210137 The MacArthur Foundation also supported the call for certain qualities in a leader. In an al-
first author at the Center for Advanced Study in the Be- ternate set of circumstances, this leader may
havioral Sciences, 1982-1983
Requests for reprints should be sent to David A Kenny,
become a follower. The trait approach has been
Department of Psychology U-20, University of Connec- evaluated by two major types of studies: at-
ticut, Storrs, Connecticut 06268. tempts to identify so-called leadership traits
678
VARIANCE IN LEADERSHIP 679

and rotation studies in which member com- ers may be, in fact, situation specific because
position or group tasks are varied. Data from most group situations under study were per-
rotation designs demonstrating variability in formance tasks.
leader emergence as well as the failure to un- Two other proposed traits that have received
cover a trait that is consistently associated with substantial support in some quarters are dom-
leader emergence have led researchers to reject inance and personality adjustment. Stogdill
dispositional explanations for leader emer- (1948) reported that 11 studies (out of 17)
gence. We argue in this article, however, that indicated that leaders had higher scores on
such rejection is unwarranted or, at best, pre- various dominance scales. Gibb (1969) re-
mature. In his revision of StogdiU's Handbook ported several studies published after the Stog-
of Leadership, Bass (1981) maintained that dill review that indicated a similar relationship.
"some of the variance in who emerges as leader However, Mann (1959) again questioned the
and who is successful and effective is due to strength of these correlations. No correlation
traits of consequence in the situation; some of between dominance and leadership exceeded
the variance is due to situational effects and .42 and the median correlation was .20. The
some of the variance is due to the interaction same conclusions can be drawn about per-
of traits and situations" (pp. 81-82) In this sonality adjustment. Whereas several studies
article we will use a quantitative model of were reported as supporting a positive rela-
leadership that allows a researcher to isolate tionship between leadership and adjustment
the amount of variance attributable to traits (Bird, 1940; Gibb, 1969; Jenkins, 1947; Mann,
Furthermore, applying this model to one of 1959, Stogdill, 1948), Mann indicated that the
the more elaborate rotation designs (Barnlund, median correlation was 15 and no correlation
1962), we hope to mitigate the conclusion that exceeded .53
leader emergence is not a function of traits. All major reviews of the leadership literature
We first begin, however, by reviewing in turn have noted the overall failure to find either a
both trait identification studies and rotation consistent and strong trait or a constellation
design experiments. of traits that differentiate leaders from follow-
ers. One explanation for this failure offered
The Failure to Identify Traits Correlated by Gibb (1969) was the inadequacy of trait
With Leadership and personality measures. Measures used in
reported research may not have been sensitive
Reviews of the leadership literature all tend enough to highlight the leadership trait. An
to demonstrate the point that "no single trait alternate method, the rotation design, can be
or group of characteristics has been isolated used to test the trait theory of leadership with-
which sets off the leader from the members of out positing specifically what the trait is.
his group" (Jenkins, 1947, pp 74-75). Bird
(1940), after reviewing 20 studies, noted 79 Rotation Designs
leadership traits; however, only 5% of these
traits were common to four or more investi- Rotation designs are based on the hypothesis
gations (Geier, 1967). Subsequent reviews by that if leadership is a function of personal
Jenkins (1947) and Stogdill (1948) echoed this qualities of the leader, then the same person
discouraging state of affairs. will emerge as a leader when aspects of the
A trait that generally has been found to cor- situation are varied. If, however, leadership is
relate with leader emergence has been intel- a function of either situational factors or of
ligence. Stogdill (1948) reported 23 studies (out an interaction between qualities of the poten-
of 33) that reported a significant positive cor- tial leader and needs of the group, then varying
relation between leadership and intelligence situational factors will change who emerges as
Although Mann's (1959) review supported this the group leader Thus, to contrast a trait hy-
relationship, it cast doubt on its strength. In pothesis with an interactiomst hypothesis of
the studies surveyed, none yielded a correlation leader emergence, rotation designs are used to
higher than .50 and the median correlation vary some aspects of the group (e.g., member
was .25. Furthermore, Gibb (1969) suggested composition and task) over a series of sessions.
that the reportedly higher intelligence of lead- The relationship between leadership rank in
680 DAVID A KENNY AND STEPHEN J ZACCARO

one session with rankings in other sessions is groups having different memberships did so
then observed. because of a special skill for that task. They
Three different rotational schemes have may have had special abilities or resources that
been employed. The first type of rotational fellow members recognized. A more complete
study holds the task constant but varies mem- test would thus involve varying the group task.
ber composition. Generally, each person is in Gibb (Note 1) did vary group tasks, while
a group with each other person in the study keeping membership constant. Leadership was
only once. Setting group size at n, to execute found to be consistent across the different
such a study there must be n2 subjects and tasks. Carter and Nixon (1949), using a similar
each subject must be in n + 1 groups. The design, had the same dyads work on three con-
second type of rotational study keeps member secutive tasks, an intellectual task, a clerical
composition constant but varies the task A task, and a mechanical task. Although lead-
third type of rotation study varies both mem- ership was associated in the intellectual and
ber composition and task. clerical situations, it seemed to be independent
Bell and French (1950) varied only member on the mechanical task Whereas these two
composition to study leader stability Twenty- studies appear, in part, to offer additional sup-
five male subjects interacted in five-member port for a trait interpretation of leader emer-
groups over six group sessions. In each session, gence, it is reasonable to assume an mertial
each individual interacted with group mem- effect whereby a leader who emerges in one
bers who were not a part of any other group group situation will continue to be perceived
that included the individual. During each ses- as a leader in subsequent group sessions by
sion, each group had to discuss a case history the same fellow members.
of a student who had a personal adjustment Only Barnlund (1962) varied both member
problem m college. After the group members composition and group task Further, the tasks
reached a solution consensus, they rank or- used were different enough so that each one
dered all fellow members according to leader required a different "constellation of skills."
preference. Bell and French then averaged the As m Bell and French (1950), each of the 25
rankings given each individual by his fellow subjects worked with four fellow members in
members and correlated this average with the six group sessions structured so that no in-
mean status score given this individual m the dividual worked with another member m more
other five groups of which he was a member. than one session. Subjects worked on motor,
The range of these correlations was —.03 to artistic, mathematical, literary, social, and
+.98, and the average correlation was 75 This spatial problems At the conclusion of each
result led Bell and French to conclude that session subjects were asked to rank order their
varying group composition did not have major preferences for a leader Using the same sta-
effects on consistency of leadership position tistical procedures as Bell and French, Barn-
and that the "recent trend toward emphasis lund reported a correlation of .64, which he
upon situational factors in leadership may re- judged to be statistically nonsignificant.1 He
quire some reevaluation" (p. 767) concluded that his results lent "credibility to
Borgatta, Bales, and Couch (1954) also var- the idea that leadership grows out of the special
ied member composition Three-member problems of coordination facing a given group
groups planned a future role play situation and the available talent of the participants"
and each subject participated in four group (P 51).
sessions. Using a variety of indices, including
leadership rankings by fellow members, as-
sertiveness ratings, and sociometric scales, 11
leaders were discerned in the first group ses- ' Because Barnlund assumed n equaled only five, a cor-
sion. The authors found that in the second relation of 878 would be needed to achieve conventional
and third sessions 8 of these 11 leaders re- statistical significance However, because 64 is an average
mained m the top position. of 30 nonindependent correlations each based on an n of
five, it would seem likely that it is statistically significant
In both studies only group composition was The interest in this article, however, is estimation of the
varied, and the task was kept constant. It is percent of variation due to traits and not significance
possible that leaders who emerged across testing
VARIANCE IN LEADERSHIP 681

A Model of Leadership Because X2 is the parameter of interest, we


must be clear what it is. The term <P$ represents
Barnlund's study appears to be somewhat the variance in leadership in a particular group
convincing against a trait hypothesis of emer- working on a task The term X2 represents the
gent leadership. Although rotation design pro- overlap of the j8s across two different groups.
vides a reasonable method to determine the It then represents the percent of the variance
extent to which leadership is stable, by itself, of the j8s that is stable across groups and tasks.
it provides us with only a qualitative basis of Formally it is the covariance of the /3s across
ascertaining leadership stability and not a two tasks divided by the variance within a
quantitative one. A quantitative model of task
leadership applied to Barnlund's data will give
Researchers who have employed the rotation
us a more accurate estimate of leadership sta-
design have computed correlations that they
bility.
have implicitly assumed to estimate X2. The
To review, for Barnlund's design each of the major correlation that Barnlund (1962) and
25 subjects works with four members in six others have computed is the within-group cor-
group sessions structured in such a way so that relation The leadership scores of individuals
no one individual works with another indi- in one group for task m are correlated with
vidual in more than one session. After each the average of their leadership scores for the
session subjects rank order the four other group other tasks When there are five groups for
members in terms of leadership We can refer each task and six tasks, there are a total of 30
to a data point as Xljkm, where i refers to rater within-group correlations each with an n of
(1 through 5), j to ratee (1 through 5), k to five. Given the model that is presented in
group (1 through 5), and m to session or task Equation 1, what will the within-group cor-
(1 through 6). Kenny (1981) presented a gen- relation be?
eral model for two-person data structures that
The model in Equation 1 must be modified
he called the social relations model. We will
because subjects are forced to rank order the
use this model for the ratings of leadership.
other members in the group. One consequence
Person i's rating of person j's leadership in
of using a rank order is that the mean remains
group k working on task m is
constant across raters (I e , Ms of 1, 2, 3, and
4 must equal 2.5 for all raters) Such a con-
- Hkm + <*ikm + fijkm + 1ijkn (1) straint modifies the rating of person i of person
The term ixkm represents a constant term for j in the following way (see Appendix):
group k and task m. The term alkm represents
the extent to which person i tends to see others M+
in the group as leaders.2 The term @jkm rep- 1 n- 1
resents the extent to which person j is seen by + 1Vkm ~ It km- (2)
others in the group to be a leader. It then is
The term _M IS now a constant across group
the measure of leadership The term yljkm rep-
and task; 0km is the mean fi for all persons in
resents the interaction of rater and ratee It
group k on task m, and ylkm is the mean y
represents disagreement in the ratings and so
for person i m group k working on task m.
can be considered an error of measurement
For Expression 2 the mean for each rater is
In sum, a is a rater bias term, /3 is the person's
the same. One benefit of this is to remove the
true leadership in the group, and y is an error
rater effect of a from the equation Note that
of measurement
if person i is a leader and so $l)km is large, then
Of key interest is the correlation of lead- i's ratings of j and others are inflated because
ership or |8 for two different tasks (tasks m and they are compared to each other and not to
m') The correlation pgmsm or X2 denotes the person i.
extent to which leadership is stable across two
tasks m and m' If it is 1.0, someone high on
leadership for one task will also be high for 2
Because rank orders are used, it may seem that the
another task If zero, then leadership is un- alpha term can be dropped As will be seen, any scheme
stable across tasks. It is then the key correlation that forces the mean to be the same for all raters affects
that needs to be estimated. beta
682 DAVID A KENNY AND STEPHEN J ZACCARO

Investigators compute the average rating for where /3<m) is the mean of person / s |3s across
each person in a group. We must then compute all tasks but task m and the y term is similarly
this term from Expression 2. It is for person denned.
j in group k and task m: If we denote n(n - 2)/(« - I) 2 as p and
s (« - 2) of as a^/(n — 1), the covariance between
M + Pjkm — P kn>n ;
(« - I)2 Expressions 3 and 4 can be shown to equal
p ^ X 2 , and the variances of Expressions 3 and
2 "tl km 4, respectively, are p^aj + p<rt and
jkm (3)
n- 1 erf
where 0 km is the average /3 for all persons in
group k (see Appendix for proof.) The average
rating of person j m the other tasks besides m
X2 is the correlation of leadership across tasks,
can be approximated by (see Appendix)'
of represents error variance due to y, and
dp represents leadership variance.3 Then the
correlation between Expressions 3 and 4 equals

(5)
+

This is what the within-group correlation


theoretically estimates. .89 =
1 + X2(« - 1)
It does not exactly estimate X2 for two rea-
sons. First, unreliability attenuates the cor-
relation. Thus, to the extent that of does not which is derived from Expression 5 by setting
equal zero, the withm-group correlation of = 0 The solution for X2 is 82 Even if we
underestimates X2. Second, X2 and the withm- do not disattenuate .64 (and so assume no
group correlation are not equivalent. The error variance), the estimate of X2 is .49. Thus,
method of correlating the rating on a single 49% of variance represents a lower-bound es-
task with the average ratings on the remaining timate for the leadership-trait variance for the
n tasks results in an overestimation of X2. The Barnlund study. The comparable lower-bound
parameter we seek to measure, X2, is the cor- estimate for the Bell and French (1950) study,
relation between leadership in one other task which did not vary task, is .63.
Given the formal analysis that we performed
based on the social relations model (Kenny, Discussion
1981), it is possible to reexamine the rotation One of the oldest topics m psychology is the
studies to determine the value of X2. Because extent to which leadership is a trait. Research-
Barnlund (1962) varied both member com-
position and group task, we will use his study.
3
He reports the within-group correlation as .64. We are assuming that the 7s are independent It might
As the preceding analysis has shown, the seem plausible that y,j is correlated 7,,. For instance, if
person 1 thinks person j is more of a leader than others,
within-group correlation does not exactly es- j sees 1 as more (or less) of a leader For peer rating data,
timate the stability of leadership across group this correlation has not been found (Kenny, Lord, & Garg,
situations. Adjustments must be made for un- Note 2)
4
reliability and equivalence. Given the reli- Barnlund's (1962) use of the Spearman-Brown proph-
ecy formula to estimate the reliability of leadership across
ability estimates reported by Barnlund,4 the n tasks presumes that X2 is one Because X2 is less than
disattenuated correlation is .89. To determine one, his estimate of reliability is inflated, making the dis-
what X2 is, we need to solve for it from attenuated correlation somewhat deflated
VARIANCE IN LEADERSHIP 683

ers have attempted to answer this question in situations This implication, however, can be
two ways The most common way has been harmonized with a leader-trait perspective by
an attempt to isolate the particular leadership proposing that persons who are consistently
trait As is well known, this approach has not cast in the leadership role possess the ability
isolated any leadership trait The second ap- to perceive and predict variations in group
proach has been to examine the stability of situations and pattern their own approaches
leadership across different situations. Persons accordingly Such leaders may be highly com-
are rotated across different tasks or groups. petent in reading the needs of their consti-
Only the Barnlund study varied both task and tuencies and altering their behaviors to more
member composition Using the social rela- effectively respond to these needs
tions model (Kenny, 1981), we estimated the The methodological and measurement dif-
percent variance in leadership due to trait as ficulties of demonstrating such an ability in
between 49% and 82% In conjunction, the leaders are obvious These difficulties, however,
rotation design and the social relations model rather than causing an abandonment of a trait
seem to suggest that leadership is much more notion of leadership, should spur the devel-
stable across situations than our introductory opment and use of innovative techniques that
texts would indicate can more effectively evaluate this notion.
Regarding our findings, the distinction be- Those who have so strongly endorsed the in-
tween leader emergence and leader effective- teractions theory of leadership have done so
ness needs to be emphasized We have focused because they have learned the wrong lesson
only on factors influencing leader emergence from the failure of past research attempts to
and not effectiveness Traits may determine isolate the leadership trait. This failure should
who emerges as leaders but may not be as not be interpreted as meaning no traits exists
important when considering the subsequent but that our research and measurement strat-
effectiveness of these leaders Likewise, the m- egies were inadequate.
teractionist perspective may be more relevant
m predicting leader effectiveness than leader Reference Notes
emergence. 1 Gibb. C A The emergence ofleaders in small temporary
Because our major result is derived from groups of men (Publication 1392) Ann Arbor Uni-
Barnlund's data, a few cautions regarding his versity of Michigan Microfilms, 1949
2 Kenny, D A , Lord, R G , &Garg, S A social relations
method are m order. Even though Barnlund analysis ofpeer ratings Unpublished manuscript, Uni-
varied both task and group membership, the versity of Connecticut, 1983
setting and experimental instructions were the
same across groups. These aspects of the group References
situation may be as important as task content
in determining leader emergence Further- Barnlund, D C Consistency of emergent leadership in
groups with changing tasks and members Speech
more, the subjects were homogeneous, all being Monographs, 1962, 29, 45-52
male college students. From the interactionist Bass, B M Slogdill's handbook of leadership A survey
perspective, which emphasizes group diversity, of theory and research New York Free Press, 1981
this homogeneity may have artificially raised Bell. G B , & French, R L Consistency of individual
leadership position in small groups of varying mem-
the estimate of X2. These limitations in Barn- bership Journal of Abnormal Social Psychology, 1950,
lund's study make replication all the more 45, 764-767
necessary. Bird, C Social psychology New York Appleton-Century-
Although our analysis reveals that leadership Crofts 1940
is stable, it in no way illuminates what per- Borgatta, E F , Bales, R L , & Couch, A S Some findings
relevant to the great man theory of leadership American
sonality trait or set of behaviors bring about Sociological Review 1954, 19, 755-759
this stability We suspect, however, that re- Carter, L F , & Nixon, M An investigation of the rela-
opening the search for a specific leadership tionship between four criteria of leadership ability for
personality trait may, as before, prove un- three different tasks Journal of Psychology, 1949, 27,
245-261
fruitful. The variability of group situations and Geier, J G A trait approach to the study of leadership
group needs does imply that different leader in small groups Journal of Communication, 1967, 17,
approaches may be needed for different group 316-323
684 DAVID A KENNY AND STEPHEN J. ZACCARO

Gibb, C A Leadership In G Lindzey & E. Aronson in honor of Donald T Campbell San Francisco Jossey-
(Eds.), Handbook ofsocialpsychology (2nd ed.) Reading, Bass, 1981
Mass Addison-Wesley, 1969 Mann, R D A review of the relationships between per-
Hollander, E P , & Julian, J W Studies in leader legitimacy, sonality and performance in small groups. Psychological
influence, and innovation In L Berkowitz (Ed), Ad- Bulletin, 1959, 56. 241-270
vances in experimental social psychology (Vol 5) New
Murphy, A J A study of the leadership process. American
York Academic Press, 1970
Sociological Review, 1941, 6, 674-687
Jenkins, W O A review of leadership studies with particular
reference to military problems Psychological Bulletin, Shenf, M An outline of social psychology New 'York. Har-
1947, 44, 54-79 per & Brothers, 1948
Kenny, D A Interpersonal perception A multivanate Stogdill, R M Personal factors associated with leadership'
round robin analysis In M B Brewer & B E Collins A survey of the literature Journal of Personality, 1948,
(Eds.), Scientific inquiry and the social sciences A volume 25, 35-71

Appendix

Expression 2 is equal to X,jkm - X, km + fi and fijkm


has equal means across ratees for each rater The (» - I) 2 (B - I) 2
mean of Equation 1 across ratees for rater ; is
Making this substitution and gathering terms, we
2 $)km have
X, km = Hkm + Oilkm +' ' + 7, km (Al)
n(n - 2)
Because (n - I) 2 Jkm

2 ft/tm 2 Pjkm ~ fiiltn


',, (A2) (A6)
n— \ n —1 n—1 «— 1 n— 1
the right-hand side of Equation A1 becomes which is Expression 3
Expression 4 is an approximation of the mean
<X,km ylkm (A3) of Expression 3 across n tasks It ignores the fol-
n-1 n-1 lowing minor terms that are means of at least n2
terms, (a) The mean of all the /3s of ratees for groups
Thus X,Jkm - X,km + ix equals with person j except those persons in group k, task
m (b) The mean of all the 7s of rater-ratee com-
O n
ft km filkm . - r A A\ binations for groups with person j except those 7s
+ Pjkm —" + _ + y,jkm - "Ylkm, (A4) in group k, task m
_
To determine the variances of Expressions 3 and
which is Expression 2 4 in the text and their covanance, we define the
The mean of Expression A4 across raters for ratee following nonzero expected values
j equals

Pikm

and

Tikm
n'J
(A5) The mean of Expression 3 for any group of n
n- 1 persons is constant. The expected value of its vari-
ance then equals n/(n — 1) times the expected value
Given Equation A2 and Expression A3, the fourth of Expression 3 squared That expectation times «/
term in Expression A5 becomes (7Z - 1 ) IS
VARIANCE IN LEADERSHIP 685

because
where p equals n(n - 2)/(n - I)2 and a2 = oil ryottm^ 2 fl + X2(n - 1)"]
m
(B-l). ' ^i. » J
The mean of Expression 4 for a group of n persons The
is ordinarily nonzero The mean equals covanance of Expression 3 and 4 equals
because
no {km) 1 r, ((cm)

The expected value of the variance of Expression


4 minus its mean equals
+ X2(
"~ ^ 1 + ?± Received February 14, 1983
} n ' Revision received June 30, 1983

New Journal on Aging Call for Nominations

By action of APA's Publications and Communications Board and Council of Rep-


resentatives, the APA is publishing a quarterly journal called Psychology and Aging,
the first issue of which will appear in 1985 Psychology and Aging will contain original
articles on adult development and aging The articles may be reports of research or
applications of research, and they may be biobehavioral, psychosocial, educational,
methodological, clinical, applied, or experimental (laboratory, field, or naturalistic)
studies. For more information about the new journal, see the November issue of the
APA Monitor.
Nominations for the editor of Psychology and Aging are now open Candidates
must be members of APA and should be available to start receiving manuscripts in
mid-1984 to prepare for issues published in 1985 To nominate candidates, prepare
a statement of one page or less in support of each candidate. Submit nominations
no later than April 2, 1984, to
Martha A. Storandt
Department of Psychology
Box 1125
Washington University
St. Louis, Missouri 63130
(314)889-6508

You might also like