ArticlePDF Available

Psychological Safety and Learning Behavior in Work Teams

Authors:

Abstract and Figures

This paper presents a model of team learning and tests it in a multimethod field study. It introduces the construct of team psychological safety—a shared belief held by members of a team that the team is safe for interpersonal risk taking—and models the effects of team psychological safety and team efficacy together on learning and performance in organizational work teams. Results of a study of 51 work teams in a manufacturing company, measuring antecedent, process, and outcome variables, show that team psychological safety is associated with learning behavior, but team efficacy is not, when controlling for team psychological safety. As predicted, learning behavior mediates between team psychological safety and team performance. The results support an integrative perspective in which both team structures, such as context support and team leader coaching, and shared beliefs shape team outcomes.
Content may be subject to copyright.
Psychological Safety and Learning Behavior in Work Teams
Author(s): Amy Edmondson
Reviewed work(s):
Source:
Administrative Science Quarterly,
Vol. 44, No. 2 (Jun., 1999), pp. 350-383
Published by: Sage Publications, Inc. on behalf of the Johnson Graduate School of Management, Cornell
University
Stable URL: http://www.jstor.org/stable/2666999 .
Accessed: 07/12/2012 15:29
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
.
Sage Publications, Inc. and Johnson Graduate School of Management, Cornell University are collaborating
with JSTOR to digitize, preserve and extend access to Administrative Science Quarterly.
http://www.jstor.org
This content downloaded by the authorized user from 192.168.52.71 on Fri, 7 Dec 2012 15:29:43 PM
All use subject to JSTOR Terms and Conditions
Psychological Safety
and
Learning
Behavior
in
Work
Teams
Amy Edmondson
Harvard University
?
1999
by
Cornell
University.
0001
-8392/99/4402-0350/$1
.00.
I
thank Richard Hackman
for extensive
advice and feedback on the
design
of this
study
and on
several
versions of
this
pa-
per.
Keith
Murnighan,
Rod
Kramer,
Mark
Cannon,
and
three
anonymous
reviewers
provided
feedback that
greatly
benefited
the final version of the
paper.
I
gratefully
acknowledge
the Division of Research at
the Harvard Business School for
provid-
ing
financial
support
for this research.
This paper presents a model of team learning
and tests
it
in a multimethod field study. It introduces
the construct
of team psychological safety-a shared
belief held by
members of a team that the team is safe
for interper-
sonal risk taking-and models
the
effects
of
team
psy-
chological safety and team efficacy together
on learning
and performance
in
organizational work
teams. Results
of
a study
of
51
work
teams
in
a manufacturing com-
pany, measuring antecedent, process, and
outcome vari-
ables, show that team psychological safety
is associated
with
learning behavior,
but team efficacy is not, when
controlling
for
team psychological
safety. As predicted,
learning behavior mediates between team
psychological
safety
and team
performance.
The results support an
in-
tegrative perspective
in
which both team
structures, such
as context
support and
team
leader coaching,
and shared
beliefs shape team
outcomes.'
A
growing reliance on teams
in
changing
and uncertain orga-
nizational environments creates a managerial
imperative to
understand
the
factors
that enable team
learning. Although
much has been written about teams and
about learning
in
organizations,
our
understanding
of
learning
in
teams re-
mains limited.
A
review of
the team effectiveness and orga-
nizational learning literatures reveals markedly
different ap-
proaches and a lack of cross-fertilization between
them.
An
emerging literature on group learning, with
theoretical papers
on
groups
as
information-processing systems
and
a number
of
empirical
studies
examining
information
exchange
in
labo-
ratory groups, has not investigated the learning
processes of
real work teams (cf. Argote, Gruenfeld, and
Naquin, 1999).
Although most studies
of
organizational
learning have been
field-based, empirical research on group
learning has primar-
ily taken place
in
the laboratory, and little
research has been
done to understand the factors
that influence
learning
behav-
ior in
ongoing
teams in real
organizations.
Studies of work teams
in
a
variety
of
organizational
settings
have
shown
that team effectiveness is enabled by
structural
features such as a
well-designed
team task, appropriate
team
composition,
and
a context that ensures the
availability
of
information, resources,
and rewards
(Hackman,
1987).
Many
researchers have concluded
that structure and
design,
including equipment, materials, physical
environment,
and
pay systems,
are
the
most
important
variables for
improving
work-team
performance (Goodman,
Devadas,
and
Hughson,
1988; Campion, Medsker,
and
Higgs,
1993;
Cohen and Led-
ford, 1994)
and have
argued against
focusing
on
interper-
sonal factors
(e.g., Goodman, RavIin,
and
Schminke, 1987).
According to
this
research, organization
and
team
structures
explain
most of
the variance
in
team effectiveness.
In
contrast, organizational learning
research has
emphasized
cognitive
and
interpersonal
factors to
explain
effectiveness,
showing,
for
example,
that
individuals'
tacit beliefs about
in-
terpersonal
interaction
inhibit
learning
behavior
and
give
rise
to ineffectiveness
in
organizations (e.g.,
Argyris, 1993).
This
cognitive emphasis
takes different forms.
Organizational
learning
theorists
have
offered
both
descriptive theory
ex-
350/Administrative Science Quarterly, 44 (1999):
350-383
This content downloaded by the authorized user from 192.168.52.71 on Fri, 7 Dec 2012 15:29:43 PM
All use subject to JSTOR Terms and Conditions
Psychological Safety
plaining the failure of organizations to adapt
rationally due to
cognitive biases that favor existing routines
over alternatives
(e.g., Levitt and March, 1988) and
prescriptive theory propos-
ing interventions that alter individuals'
"theories-in-use" to
improve organization effectiveness (e.g.,
Argyris and Schdn,
1
978). The former theorists suggest that
adaptive learning in
social
systems is fundamentally problematic
and rare,
and
the
latter, only slightly more sanguine,
propose that expert
intervention is necessary to bring it about
(cf. Edmondson
and Moingeon, 1998). This paper takes a
different approach
to
understanding learning
in
organizations
by examining to
what extent
and under what conditions
learning
occurs natu-
rally
in
organizational work groups.
Much
organizational learning research has
relied on qualita-
tive studies that
provide
rich
detail about
cognitive
and inter-
personal processes but do not allow explicit
hypothesis test-
ing (e.g., Senge, 1990; Argyris, 1993;
Watkins and Marsick,
1993). Many team studies,
in
contrast, have
used large
samples and quantitative data
but have not
examined
ante-
cedents
and consequences
of
learning
behavior (e.g.,
Good-
man, Devadas,
and
Hughson, 1988;
Hackman, 1990;
Cohen
and
Ledford, 1994).
I
propose
that to understand
learning
behavior
in
teams,
team structures and shared beliefs must
be
investigated jointly, using
both
quantitative
and
qualitative
methods.
This
paper presents
a
model
of
team
learning
and tests
it in
a
multimethod field study. The results
support an integrative
perspective
in
which both team structures,
such as context
support
and
team leader
coaching,
and shared beliefs
shape
team
outcomes.
Organizational
work
teams
are
groups
that
exist within the context
of
a
larger
organization,
have
clearly
defined
membership,
and share
responsibility
for
a
team
product or service (Hackman, 1987;
Alderfer, 1987).
Their
learning
behavior
consists of
activities
carried out
by
team
members
through
which a team
obtains and
processes
data
that
allow
it
to
adapt
and
improve. Examples
of
learning
be-
havior include
seeking feedback, sharing
information, asking
for
help, talking
about
errors,
and
experimenting.
It
is
through
these activities that teams
can
detect
changes
in
the
environment,
learn about
customers'
requirements,
im-
prove
members'
collective
understanding
of a
situation,
or
discover
unexpected consequences
of their
previous
actions.
These useful
outcomes often
go
unrealized
in
organizations.
Members of
groups
tend not
to
share
the
unique knowledge
they hold,
such that
group
discussions
consist
primarily
of
jointly
held
information
(Stasser
and
Titus,
1987), posing
a
dilemma for
learning
in
groups.
More
centrally,
those
in a
position
to initiate
learning
behavior
may
believe they
are
placing themselves at risk; for example, by
admitting an
er-
ror
or
asking
for
help,
an individual
may appear
incompetent
and
thus
suffer a
blow
to
his
or her
image.
In
addition,
such
individuals
may
incur more
tangible
costs
if
their actions
cre-
ate unfavorable
impressions
on
people
who
influence
deci-
sions about
promotions, raises,
or
project
assignments.
Im-
age
costs have been
explored
in research
on face
saving,
which has
established
that
people
value
image
and
tacitly
abide by social expectations
to save their own and
others'
351/ASQ, June 1999
This content downloaded by the authorized user from 192.168.52.71 on Fri, 7 Dec 2012 15:29:43 PM
All use subject to JSTOR Terms and Conditions
face (Goffman, 1955). Asking for help, admitting errors, and
seeking feedback exemplify the kinds of behaviors that pose
a threat to face (Brown, 1990), and thus people in organiza-
tions are often reluctant to disclose their errors (Michael,
1976) or are unwilling to ask for help (Lee, 1997), even when
doing
so would
provide
benefits for the team or
organization.
Similarly, research has shown that the sense of threat
evoked
in
organizations by discussing problems
limits indi-
viduals' willingness to engage in problem-solving activities
(Dutton, 1993; MacDuffie, 1997). The phenomenon of threat
rigidity has been explored at multiple levels of analysis,
showing that threat has the effect of reducing cognitive and
behavioral flexibility and responsiveness, despite the implicit
need for these to address the source of threat (Staw, Sand-
elands, and Dutton, 1981). In sum, people tend to act
in
ways
that inhibit
learning
when
they
face the
potential
for
threat or embarrassment (Argyris, 1982).
Nonetheless,
in
some environments, people perceive
the
career and interpersonal threat as sufficiently
low
that they
do ask for
help,
admit
errors,
and discuss
problems.
Some
insight into this may be found in research showing that fa-
miliarity among group members can reduce the tendency to
conform and suppress unusual information (Sanna and
Shot-
land, 1990); however,
this
does not directly
address the
question of when group members
will
be comfortable
with
interpersonally threatening actions. More specifically,
in
a
recent study of hospital patient-care teams, I found signifi-
cant differences
in
members' beliefs about
the
social conse-
quences of reporting medication errors;
in
some teams,
members openly acknowledged them and discussed ways
to
avoid
their
recurrence;
in
others, members kept
their
knowledge of a drug error to themselves (Edmondson,
1996). Team members' beliefs about the interpersonal con-
text
in
these teams could be characterized as tacit; they
were
automatic, taken-for-granted
assessments of
the
"way
things
are around here."
For
example,
a
nurse
in
one team
explained matter-of-factly,
"Mistakes are
serious,
because of
the
toxicity
of
the
drugs [we use]-so you're
never afraid to
tell the Nurse
Manager";
in
contrast,
a nurse
in
another
team
in
the same
hospital reported,
"You
get put
on trial!
People get blamed for mistakes
. .
.
you
don't want to
have
made one." These
quotes
illustrate
markedly
different be-
liefs about the
interpersonal context;
in the first
team,
mem-
bers saw
it
as self-evident
that
speaking up
is natural and
necessary, and
in
the other, speaking up
was viewed as
a
last resort.
An
aim
of
the present study was
to
investigate
whether
be-
liefs about the
interpersonal
context
vary
between teams
in
the
same
organization,
as well
as to examine
their effects
on
team
outcomes. Existing
theories do not
address
the
issue
of
how
such beliefs
may
affect
learning
behavior
in
teams,
instead
focusing primarily
on structural
conditions associated
with
overall team effectiveness
(e.g., Hackman, 1987)
or
on
the
skills that must be
learned
by
individuals
to
enable
learn-
ing
in
difficult
interpersonal
interactions
(e.g., Argyris, 1982).
Similarly,
research
on
group training
has focused
primarily
on
task
knowledge
and
has
paid
little attention to
the role of
social
knowledge (Levine
and
Moreland, 1991). Thus,
the
352/ASQ, J une 1999
This content downloaded by the authorized user from 192.168.52.71 on Fri, 7 Dec 2012 15:29:43 PM
All use subject to JSTOR Terms and Conditions
Psychological Safety
role of beliefs about the interpersonal context in individuals'
willingness to engage
in
otherwise-threatening learning
be-
havior has been largely unexamined. This is the gap
I seek
to
fill
with a model and study of team learning.
A MODEL OF TEAM LEARNING
Team Learning Behavior
Organizational learning is presented
in the
literature
in two
different
ways:
some
discuss learning
as an
outcome;
others
focus
on a
process they
define as
learning.
For example,
Levitt and March
(1
988: 320) conceptualized organizational
learning as the outcome of a process of organizations
"en-
coding inferences from history into routines that guide
be-
havior";
in
contrast, Argyris and Sch6n (1978) defined
learn-
ing as a process of detecting and correcting error.
In this
paper
I
join the latter
tradition
in
treating learning
as a pro-
cess and attempt to articulate
the
behaviors
through which
such outcomes
as
adaptation
to
change, greater
understand-
ing,
or
improved performance
in
teams
can
be achieved.
For
clarity,
I
use the
term
"learning behavior" to avoid
confusion
with the
notion
of
learning
outcomes.
The conceptualization of learning as
a
process
has roots
in
the work of educational
philosopher
John
Dewey,
whose
writing
on
inquiry
and reflection
(e.g., Dewey,
1938)
has had
considerable influence
on
subsequent learning
theories (e.g.,
Kolb, 1984; Schbn, 1983). Dewey (1922)
described
learning
as an iterative
process
of
designing, carrying out,
reflecting
upon,
and
modifying actions,
in
contrast to what
he
saw as
the human
tendency to rely excessively
on habitual or
auto-
matic behavior.
Similarly,
I
conceptualize learning
at the
group
level of
analysis
as
an
ongoing process
of reflection
and
action,
characterized
by asking questions,
seeking
feed-
back, experimenting, reflecting
on
results,
and
discussing
errors or
unexpected
outcomes of actions. For a
team to dis-
cover
gaps
in
its
plans
and make
changes
accordingly,
team
members
must
test
assumptions
and discuss differences of
opinion openly
rather
than
privately
or
outside the
group.
I
refer
to
this set
of activities as
learning behavior,
as it
is
through
them that
learning
is enacted
at the
group
level.
This
conceptualization
is
consistent with a definition of
group
learning proposed recently by Argote, Gruenfeld,
and
Naquin
(1999)
as both
processes
and outcomes
of
group
interaction
activities
through
which
individuals
acquire, share,
and com-
bine
knowledge,
but it
focuses
on the
processes
and leaves
outcomes of these
processes
to be
investigated
separately.
The
management
literature
encompasses
related
discussions
of
learning,
for
example, learning
as
dependent
on attention
to
feedback
(Schon, 1983), experimentation (Henderson
and
Clark, 1990),
and discussion
of failure
(Sitkin,
1992;
Leonard-
Barton, 1995).
Research
has
demonstrated
performance
ben-
efits
for
feedback
seeking by
individual
managers
(Ashford
and
Tsui, 1991),
for teams
seeking
information and feedback
from
outside the team
(Ancona
and
Caldwell,
1992),
and for
research
and
development
teams that
experiment
frequently
(Henderson
and
Clark, 1990). Similarly,
because
errors
pro-
vide a source of information about performance by
revealing
353/ASQ, June 1999
This content downloaded by the authorized user from 192.168.52.71 on Fri, 7 Dec 2012 15:29:43 PM
All use subject to JSTOR Terms and Conditions
that something did not work as planned, the ability to
dis-
cuss them productively has been associated with organiza-
tional
effectiveness (Michael, 1976; Sitkin, 1992;
Schein,
1993). On one hand, if feedback seeking, experimentation,
and discussion of errors individually promote effective perfor-
mance, learning behavior-which includes all of these activi-
ties-is also likely to facilitate performance, whether for indi-
viduals or teams. On the other hand, learning behavior
consumes
time
without assurance of results, suggesting
that there are
conditions
in
which
it
may reduce efficiency
and detract from performance, such as when teams are re-
sponsible
for
highly
routine
repetitive
tasks with little need
for
improvement or modification. For teams facing change or
uncertainty, however, the risk of wasting time may be small
relative to the potential gain;
in
such settings, teams must
engage
in
learning behavior to understand
their
environment
and their
customers
and to
coordinate
members' actions ef-
fectively. Moreover, teams that perform routine production
tasks
may
still
require learning
behavior
for
effective self-
management as a team and for intermittent process
im-
provement:
Hypothesis
1
(H1): Learning behavior
in
teams
is
positively associ-
ated with team
performance.
Team Psychological Safety
Team
psychological safety
is defined as a shared belief that
the team
is
safe for interpersonal risk taking. For the most
part,
this belief
tends to
be
tacit-taken for granted and not
given
direct
attention either by individuals or by
the team as
a whole.
Although
tacit beliefs about
interpersonal
norms are
sometimes
explicitly discussed
in
a
team,
their
being
made
explicit
does
not alter
the
essence
of team
psychological
safety.
The construct has
roots
in
early
research on
organiza-
tional
change,
in
which Schein and
Bennis
(1965)
discussed
the
need
to
create
psychological safety
for
individuals
if
they
are to feel secure and
capable
of
changing.
Team
psycho-
logical safety
is
not the
same
as group cohesiveness,
as
re-
search has
shown
that cohesiveness
can
reduce
willingness
to
disagree
and
challenge
others'
views,
such
as
in
the
phe-
nomenon of
groupthink (Janis, 1982), implying
a lack of inter-
personal
risk
taking.
The term is meant
to
suggest
neither
a
careless sense
of
permissiveness,
nor
an
unrelentingly posi-
tive affect
but, rather,
a sense
of
confidence that the team
will
not
embarrass, reject,
or
punish
someone for
speaking
up.
This confidence stems from mutual
respect and
trust
among
team members.
The importance of trust
in
groups and organizations has long
been
noted
by
researchers
(e.g.,
Golembiewski
and
Mc-
Conkie, 1975; Kramer, 1999).
Trust
is
defined as the
expec-
tation
that
others' future actions
will
be favorable
to
one's
interests,
such
that one
is
willing
to be
vulnerable to those
actions
(Mayer, Davis,
and
Schoorman, 1995; Robinson,
1996).
Team
psychological safety
involves but
goes beyond
interpersonal trust;
it
describes
a team climate characterized
by interpersonal
trust and mutual
respect
in which
people
are comfortable
being
themselves.
For team
psychological safety
to
be
a
group-level construct,
it must
characterize
the
team rather
than individual members
354/ASQ, J une 1 999
This content downloaded by the authorized user from 192.168.52.71 on Fri, 7 Dec 2012 15:29:43 PM
All use subject to JSTOR Terms and Conditions
Psychological Safety
of the
team,
and team members
must
hold
similar percep-
tions
of it.
Previous researchers have studied the similarity
of beliefs
in
social
systems
such as
organizations
and
work
groups (for reviews, see Klimoski and Mohammed, 1994;
Walsh, 1995). Perceptions
of
psychological safety,
like
other
such
beliefs, should converge
in
a team, both because
team
members are
subject to
the same
set of structural
influ-
ences and
because these perceptions develop out of salient
shared experiences.
For
example,
most members
of a
team
will conclude
that making a mistake does not lead to rejec-
tion
when
they
have
had team experiences
in
which appre-
ciation
and
interest are expressed
in
response to discussion
of their own
and others' mistakes.
Team psychological safety should facilitate learning
behavior
in work teams because it alleviates excessive concern
about
others'
reactions
to
actions
that have
the potential
for em-
barrassment or
threat,
which
learning behaviors often
have.
For example, team members may be unwilling to bring up
errors that could
help
the team make
subsequent changes
because
they
are concerned about
being
seen as
incompe-
tent,
which
allows
them
to
ignore
or discount the
negative
consequences
of
their
silence for team
performance.
In
con-
trast,
if
they respect
and feel
respected by
other
team mem-
bers
and feel
confident that
team
members
will not
hold the
error
against them,
the benefits of
speaking up
are
likely
to
be
given
more
weight. Support
for the
centrality
of
interper-
sonal
inferences
in
groups
is
found
in research on
distribu-
tive
justice,
which shows that
people
are more focused on
relational
than
instrumental considerations
in
their assess-
ments of
allocation decisions made
by authority figures;
people
are
very
attentive to the tone and
quality
of
social
processes
and are more
willing
to
comply
with these when
they
feel
valued
(Tyler
and
Lind, 1992). Argyris
and
Schdn
(1
978)
made a
connection
between
interpersonal
threat and
learning
when
they posited
that
interpersonally threatening
issues impede learning behavior,
but
they
did not address
the
possibility
that
dyads
or
groups may
differ
in
their
tacit
beliefs about
interpersonal threat, thereby giving
rise
to
dif-
ferent
levels
of
learning.
In
contrast,
I
propose
that
psycho-
logical safety
varies from team
to
team,
such that otherwise
interpersonally threatening learning
behavior can occur
if
the
team has a
sufficiently
safe environment:
Hypothesis
2
(H2):
Team
psychological safety
is
positively
associ-
ated with
learning
behavior
in
organizational
work teams.
Psychological safety
does
not
play
a
direct
role
in the
team's
satisfying
customers'
needs,
the core element of
perfor-
mance; rather,
it facilitates the team's
taking appropriate
ac-
tions
to
accomplish
its work.
Thus, learning
behavior should
mediate
the
effects of team
psychological safety
on
perfor-
mance outcomes:
Hypothesis
3
(H3):
Team
learning
behavior
mediates between
team
psychological safety
and team
performance.
Team
Efficacy
and Team
Learning
Building
on earlier work on the role of
self-efficacy
in en-
hancing
individual
performance (Bandura, 1982),
a
body
of
research has established group efficacy as a group-level
phe-
355/ASQ,
June 1999
This content downloaded by the authorized user from 192.168.52.71 on Fri, 7 Dec 2012 15:29:43 PM
All use subject to JSTOR Terms and Conditions
nomenon (e.g., Guzzo et al., 1993) and also reported a rela-
tionship between group efficacy and performance (Lindsley,
Brass, and Thomas, 1995; Gibson, 1996). This work has not
specified mechanisms through
which
shared perceptions
of
efficacy lead to good performance, and one possibility is that
efficacy fosters team members' confidence, which promotes
learning behavior and helps accomplish desired team goals:
Hypothesis
4
(H4): Team efficacy is positively associated with
team
learning behavior.
Team members deciding whether to reveal errors they have
made
are
likely to be motivated to speak up if two condi-
tions are satisfied: first, they believe they will not be
re-
jected (team psychological safety) and, second, they believe
that the
team is capable of using this new information to
generate useful results (team efficacy). Team psychological
safety
and
team efficacy are thus complementary shared
beliefs, one pertaining to interpersonal threat and
the
other
characterizing
the team's
potential
to
perform.
Team
efficacy
thus should supplement team psychological safety's positive
effect on team
learning:
Hypothesis 5 (H5): Team efficacy is positively associated
with
team
learning behavior, controlling
for the effects
of
team psycho-
logical safety.
Team Leader
Coaching
and
Context Support
as
Antecedents of Team Psychological Safety
A
set of structural features-consisting of a clear compelling
team
goal,
an
enabling
team
design (including
context
sup-
port
such as
adequate resources, information,
and
rewards),
along
with team leader
behaviors
such
as
coaching
and di-
rection
setting-have
been
shown to
increase team effec-
tiveness
(Hackman, 1987; Wageman, 1998).
These structural
features
provide a starting point for examining
antecedents
of team
psychological safety.
The
extent of
context
support
experienced by
a team should be
positively
associated
with
team
psychological safety because
access
to
resources
and
information
is likely
to reduce
insecurity
and defensiveness
in
a
team.
Team leader coaching
is
also likely to
be an im-
portant
influence
on
team
psychological safety.
A
team
lead-
er's behavior is
particularly salient;
team
members are
likely
to attend to each other's actions and
responses
but to
be
particularly
aware of the behavior of the leader
(Tyler
and
Lind, 1992).
If the leader is
supportive, coaching-oriented,
and has non-defensive
responses
to
questions
and
chal-
lenges,
members are
likely
to conclude that
the
team consti-
tutes a
safe environment.
In
contrast,
if
team leaders act
in
authoritarian
or
punitive ways,
team members
may
be reluc-
tant to
engage
in the
interpersonal
risk involved
in
learning
behaviors such
as
discussing errors,
as was
the
case
in
the
study
of
hospital
teams mentioned above
(Edmondson,
1996). Furthermore,
team leaders themselves can
engage
in
learning behaviors, demonstrating
the
appropriateness
of and
lack of
punishment
for such
risks.
Hypothesis
6
(H6):
Team leader
coaching
and context
support
are
positively
associated with team
psychological safety.
Through enhancing psychological safety,
team leader coach-
ing
and context
support
are
likely
to facilitate team
learning.
356/ASQ, J une
1 999
This content downloaded by the authorized user from 192.168.52.71 on Fri, 7 Dec 2012 15:29:43 PM
All use subject to JSTOR Terms and Conditions
Psychological Safety
Team psychological safety thus
serves as a
mechanism
translating structural features
into behavioral outcomes:
Hypothesis 7 (H7): Team psychological
safety mediates
between
the antecedents of team leader
coaching
and
context support
and
the outcome of team learning
behavior.
Context
support
and team leader coaching
should also
affect
team efficacy. Effective coaching
is likely to contribute to
members' confidence in the
team's ability to do its job, as
is
a
supportive context,
which reduces obstacles to progress
and allows team members to feel confident about
their
chances
of
success. If coaching
and context support pro-
mote team efficacy and team
efficacy promotes team learn-
ing, this suggests that team
efficacy also functions
as a
me-
diator:
Hypothesis
8
(H8):
Team efficacy mediates between
the anteced-
ents of
team leader coaching
and context support and
the outcome
of team
learning
behavior.
Team Type
Organizations
use a
variety
of
types
of teams.
Team
type
varies across several dimensions,
including cross-functional
versus single-function,
time-limited versus
enduring,
and
manager-led versus self-led.
These dimensions combine to
form different
types of
teams,
such as a time-limited
new
product development team
or an ongoing self-directed pro-
duction team. The
team
learning
model should
be
applicable
across
multiple types
of
teams,
because the social
psycho-
logical
mechanism
at
the
core
of
the model
concerns
people
taking
action
in
the
presence
of
others, and the salience of
interpersonal threat should
hold across settings. Therefore,
although
the
utility
of
learning
behavior
may vary
across
types
of
teams,
the
association
between
team psychological
safety
and team
learning
behavior
should
apply
across differ-
ent team
types. Thus,
the
effects
of
team type
on
learning
behavior should be
insignificant
when assessed
together
with
the
other variables in
the
team
learning model,
shown
in
figure
1. For
example,
new product development teams
might
be
expected
to
exhibit
more
learning
behavior
than
production teams because
of the nature of their task; none-
theless,
mean
differences
in
learning
behavior that
might
be
observed
across
types
of teams should be
explained by
team
psychological safety
and team
efficacy,
as shown
in
figure 1,
rather than
by
team
type.
Figure
1. A
model of
work-team
learning.
Antecedent
Tea
m
Team
Conditions Beliefs
Behaviors
Outcomes
TEAM LEARNING
TEAM
BEHAVIOR
STRUCTURES
TEAM
PERFORMANCE
TEAM SAFETY
Seeking
feedback,
Context
support
TEAM EFFICACY
discussing
errors,
seeking
information
Satisfies
customer
needs
Team leader
and feedback from and
expectations
coaching
customers and others
357/ASQ, June 1999
This content downloaded by the authorized user from 192.168.52.71 on Fri, 7 Dec 2012 15:29:43 PM
All use subject to JSTOR Terms and Conditions
METHODS
To test
the hypotheses
in
the
team learning
model,
I studied
real work
teams
in an organization
that
has a variety
of team
types,
using
a
combination
of qualitative
and quantitative
methods
to investigate
and
measure
the constructs
in the
model.
Preliminary
observation
and
interviews in
the organi-
zation suggested
that there
was considerable
variation
in
the
extent
to which teams
engaged
in
learning
behavior,
making
it a good
site in
which to explore
the
phenomenon
and to
investigate
factors
associated
with team
learning.
Research
Site and
Sample
"Office
Design
Incorporated"
(ODI),
a manufacturer
of office
furniture with approximately
5,000 employees
and a reputa-
tion for
product
and
management
innovation, provided
the
research site for
this
study.
Teams
in
this company,
imple-
mented
in
1979
to promote
employee
participation
and
cross-functional
collaboration,
consisted
of four
types.
Most
were functional teams,
made
up
of
managers
or
supervisors
and direct reports,
and
these included
sales
teams, manage-
ment
teams,
and
manufacturing
teams; this type
of team
existed
within and
supported
the work of a
single
functional
department.
Although
encompassing
dyadic reporting
rela-
tionships,
functional teams
had shared goals, and
members
were
interdependent
in
reaching
them. As
with
other
teams
at
ODI,
they
also typically
had some
training in
teamwork.
Second,
ODI had
a
growing
number
of
self-managed
teams
in
both
manufacturing
and
sales; these
teams
consisted
of
peers
from the
same
function. The third
type
was time-lim-
ited cross-functional
product
development
teams, and the
fourth
was time-limited cross-functional project
teams,
con-
vened
to work on
other projects
that
involved
multiple de-
partments.
The
company
was
willing
to
participate
in this
research to obtain
feedback
on how well its teams
were
working.
My primary
contact at
ODI
was
a
manager
in
an
internal
or-
ganization
development group
who worked
closely
with me
to facilitate data collection.
She scheduled
interviews and
meetings,
recruited
teams to
participate
in the
study,
and
identified recipients
of the
work of
each
of
these teams.
As
ODI
did
not have
a central roster
of
all work teams,
she dis-
tributed
a memo to
managers
throughout
the
company
de-
scribing
the
goal
of the
study (to
assess
team effectiveness
at
ODI)
and
asking
for
lists of teams
in
their area.
This
yielded
a list of 53
teams,
encompassing
differences
in
orga-
nization
level,
department,
type, size,
self- versus
leader-
managed,
and tenure
or team
age.
At the
time of
survey
data
collection,
the oldest team
had
been
together
for about
seven
years,
and the
newest had been
in
place
for four
months;
both the oldest
and newest teams
were
production
teams. These 53
teams included
34 functional
teams
(in
sales,
manufacturing,
and staff services
such as
information
technology
and
accounting),
nine
self-managed
teams
(in
manufacturing
and sales),
five
cross-functional
product
devel-
opment
teams,
and three cross-functional
project
teams.
As
the
purpose
of the
study
was to test
a theoretical
model
rather than to describe
properties
of this particular
organiza-
358/ASQ,
J une 1 999
This content downloaded by the authorized user from 192.168.52.71 on Fri, 7 Dec 2012 15:29:43 PM
All use subject to JSTOR Terms and Conditions
Psychological Safety
tion
or to characterize teams of different types, this
sample
was not selected to ensure representativeness of the
popu-
lation of all teams at ODI, nor were the four subgroups
of
team types selected to ensure that they were
representative
of
each type. The sample did satisfy the essential
criterion to
achieve
the
purposes of this study, however, which
was to
include sufficient variance on the variables in the model
to
test hypothesized relationships. Despite using a
process
characterized by voluntary participation
in
the research,
the
resulting sample was not a self-selected group of
high-per-
forming
or
highly satisfied teams; instead, there was
sub-
stantial variance
for
all variables studied, including for
such
key measures as team psychological safety, learning
behav-
ior, and performance.
Procedure
The study involved three phases of data collection.
First, I
conducted interviews and observation that involved
eight
teams,
selected from
among those available during my
two
first visits to ODI, to ensure variance in team type.
Second,
I
designed
and
administered two
surveys
and
a
structured
in-
terview
instrument to obtain quantitative
data
for all
teams
in
the
sample. Third,
I
interviewed
and observed seven
teams,
selected
according
to
survey
results
as
high
or low
in
learn-
ing
behavior.
Phase 1:
Preliminary qualitative
research.
In
two
four-day
visits
to ODI,
I
observed eight team meetings, each of which
lasted one to three hours, and conducted
17
interviews
last-
ing
from 45 minutes
to an
hour with
members or observers
of
these
eight
teams.
The eight
teams included
five
product
development teams, two management teams, and one self-
managed production
team.
I
interviewed at
least
one and as
many
as six members of each
team,
as
well
as one senior
manager responsible
for
reviewing
the work of one
of
the
product development teams. The
objectives of
this phase
of
the
study
were to
verify
that the theoretical constructs
of
team
psychological safety
and team
learning
behavior
could
be
operationalized
at ODI
and,
if
so,
to
develop survey
items
to assess these constructs
in
language
that would
be
mean-
ingful
in this
setting-a
modified
empathic strategy (Alderfer
and
Brown, 1972).
In
team
meetings,
I
took notes
and lis-
tened for
examples
of
learning behavior,
such as
asking
for
feedback, asking
for
help, admitting errors,
and
proposing
or
describing
instances of
seeking help
or
information
from oth-
ers
outside the
team.
In
interviews,
I
asked
team members
to describe features of their
team,
such as the
goal
and the
nature of its
task,
and to describe how the team
organized
its
work and what
challenges
it faced. These
general ques-
tions
allowed me
to
listen
for
examples
of
learning
behavior.
I
taped
most
interviews, except
for
some
in
the
factory
where
noise levels made
it
difficult
to do
so,
and reviewed
tapes
and notes to
identify
data that
provided
evidence
of
team
psychological safety
and
learning
behavior and
to
as-
sess whether these constructs varied
across teams.
Ex-
amples
of
learning
behavior and
quotes
that
suggested
the
presence
or
absence
of
team
psychological safety
were
tran-
scribed, and these
data suggested that both
psychological
safety and learning behavior varied across
teams.
359/ASQ, June 1999
This content downloaded by the authorized user from 192.168.52.71 on Fri, 7 Dec 2012 15:29:43 PM
All use subject to JSTOR Terms and Conditions
Table 1
Construct Development
from
Preliminary Qualitative
Data*
Constructs
Positive form Negative form
Beliefs about the team interpersonal
context (inferred
from informant quotes)
Members
of
this
team respect
each
"I
trust the people here
that they're "The
[other]
team has a lot of trust in the
other's abilities.
making the right decision,
for the expertise
of other
[memberls,
unlike
function
and
for
ODI. And they feel the this one."
(Engineering member,
same
way
about me." [Finance NPD 2)
member,
New
Product Development
Team 1
(NPD 1)]
"Each person is important.
Everyone is
respected." (Marketing
member,
NPD
1)
Members of
this team
are interested in "There's much greater
openness on this "What gets in
the way is guys who hold
each other as
people.
team-it's
intangible
...
.
We
have
a
information close to
their
chests,
so
personal interest
in each other. We're knowledge
doesn't get filtered out to
comfortable outside
the realm of work,
the team." (Management team 2)
we've
shared personal
information ...
if
you don't know
anything about
people, you don't
know how to react
to them."
(Manufacturing
member,
NPD
1)
"Our efforts to
get
to know each other
led to our mutual respect.... At
the
core, these
are
outstanding
human
beings. (Finance
member,
NPD
1)
In
this team, you aren't rejected
for being "Sally and Sue both had
been getting a "People try to
figure out what [the team
yourself or stating
what
you
think. hard time on the first shift
for leadern
wants to hear [before saying
outperforming....
That's
why they
like
what
they
think]."
(Management
being
on
this team."
(Chair production
team
2)
team
2)
"[Members of this team
are] willing
to
state
what they believe
. . .
people, in
other
teams,
if
they
don't
get
their
way, they stay
silent." (NPD 1)
Members
of this
team believe
that other "They're not out
to corrupt my success."
". .
we
struggled
through the problem
members have positive intentions.
(NPD
1
team member,
referring to the statement,
because it [the
project]
was
other team
members) clearly
for ODI's internal
needs,
not
for
customers.
We had a lot of nay sayers
who
just
wanted to
do
[the assignment
from management]
and
not question
it.
They
were
worried about getting
their
hands slapped...." (NPD 2)
Team behaviors
(observed
by
researcher
or
reported by
team members or team
observers)
Seeking
or
giving
feedback
"We
talked
to
over
a
hundred
customers; "They
were too
methodical,
too
detailed
this
changed
the
project
goal slightly,
in
their
wandering
. . .
they
did not do
to make it
integrate
more
with
the
enough
checking
with
customers
until
[other]
product
as
a
top priority."
too
far
along."
(Senior
manager,
R&D,
(Marketing
team
member,
NPD
1) describing
NPD
2)
"We also
bring
in
people
from Advanced
NPD
2
hired
a
vendor to conduct
Applications
to bounce
ideas off
of,
to customer
interviews,
in
contrast to
get
a check
on
what
we're
doing."
NPD
Team
1
members,
who
frequently
(Engineering
team
member,
NPD
1) spoke
to customers
themselves.
"[NPD
1
team
leader]
asks
me
to
come
to certain
meetings;
she wants
my
view, my industry
experience,
and
how
[this
product]
fits with
ODI's
systems
strategy." (Senior
manager, R&D)
"Am I
missing
the mark with
how to
proceed?
Is there
anything you
can
add?"
(Team leader,
management
team
1,
in a
team
meeting)
Making changes
and
improvements
(vs. "Every
three months
we decide we
need
"We did make
changes,
but too
slowly."
avoiding change
or
sticking
with a to
improve
how we
get
our (NPD
2)
course
too
long)
information. We look
for
better
ways
to
"They
did
learn,
but
not fast
enough."
do
something
and we make
changes."
(Senior
manager describing
NPD
2)
(NPD 1)
.".
every
six
months,
they
take time
....
[
[there
were
a
lot
of]
blind
alleys....
out
to look
at
what works
. . .
and
a
lot
We
had
a
preconceived
notion of what
happens
in
those
meetings."
was
important
that
prevented
us from
(NPD
1)
seeing
it
.
.
."
(NPD
2)
"We found ourselves
going
around
in
circles
a
lot.
Sometimes
this
took
a lot
of time."
(NPD
2)
360/ASQ, June 1999
This content downloaded by the authorized user from 192.168.52.71 on Fri, 7 Dec 2012 15:29:43 PM
All use subject to JSTOR Terms and Conditions
Psychological Safety
Table 1 (continued)
Construct Development
from
Preliminary Qualitative
Data*
Constructs
Positive
form Negative form
"This team gets stuck.... It's
hard to
get
a decision. The
dynamics
are that
the conversation
gets
shut down."
(Management team 2)
Obtaining or providing help or expertise
"[NPD
11
used the applications specialists
[an
ODI internal
design group] more
than
any
other
team
I
know of."
(Senior manager, R&D)
"I've learned a lot about marketing a
product-about
how and
why
we make
decisions." (Finance
team
member,
NPD
1)
"Are there any concerns right now
on
regional fleets?" (Team leader,
management team 1)
Experimenting
"There's a lot of testing of new ways
to
do
stuff.
We're
doing design
and
engineering
at the
same
time. It's wild.
It's incredibly complex.
We need to be
constantly creative about the
mechanisms...."
(NPD 1)
"There have been a lot
of
iterations.
It's
like reducing a sauce by
half.
It's
a
more flavorful sauce,
a more
complex
group of ingredients, but the
end
result
is simpler. We made it easier to use
...
by continually challenging
ourselves
to
find what is essential."
(NPD 1)
One team member
called
the
other
eight
together
at the
beginning
of the shift
and asked who was interested
in
trying
which new task. She listened
carefully
to
responses
and
suggested
a
plan
that
she explained would
allow
several
people
to
learn a new role.
(Chair
production
team
1)
Another team member
raised
the
question
of what
goal
to
set
for the
shift;
after
discussion,
the team settled
on
a new
(ambitious) target
of
producing
83 chairs.
(Chair production
team
1)
Engaging
in
constructive conflict
or
"They bring
conflict
up directly; they
confrontation
don't let
it
fester
.
.
"
(Team
leader,
NPD
1)
"People speak openly
in
team meetings,
[whereas
in
other
teams] they
wait
until the
meeting
is over and
'speak
privately
in
the
hall
[about
their
frustrations]."' (Finance
team member,
NPD
1)
*
NPD
=
new
product development.
Text
in
italics became the basis
of a new
survey item.
A
set
of
related
beliefs about the
interpersonal
context
emerged
as
suggestive
of
the
presence (or absence)
of
team
psychological safety,
including
a belief that others
won't reject people
for
being themselves,
that
team
mem-
bers care about and are interested in
each
other
as
people,
that other members
have
positive intentions,
and that
team
members
respect
each other's
competence.
Table
1
pre-
sents
excerpts
of these data to illustrate
the
constructs
of
team
psychological safety
and
team
learning
behavior
and to
show
the elements that made
up
each
construct.
Phase
2:
Survey
research.
All
members
of the 53 teams
in
the sample (496 individuals)
were administered a
five-section
survey
developed-for
this
study. Most teams were re-
361/ASQ, June 1999
This content downloaded by the authorized user from 192.168.52.71 on Fri, 7 Dec 2012 15:29:43 PM
All use subject to JSTOR Terms and Conditions
quested to
complete
surveys before or
after a team
meeting
and to
enclose
them
in
sealed
envelopes collected
by ODI
staff
and
mailed to me. In a
few cases,
surveys were
mailed
to
team members
with
return
envelopes
attached and
were
then
returned to me
directly.
In
total, 427
team
members
from 51
teams completed
the surveys,
an
86-percent re-
sponse rate; of these 51
teams,
90
percent of
members
re-
sponded. Two teams did not return
any
surveys;
in
both
cases, the
teams
continued to express
a desire to
participate
but
ultimately
failed
to do
so,
attributing this to busy
sched-
ules. At the
same
time,
for each team,
two
or
three
manag-
ers
outside
of each team
were
identified as
recipients of the
team's work
and were given
a short
survey
I
developed to
assess team
learning
behavior and
performance; 135
of the
150
observers surveyed
returned
the
survey, a
91-percent
response
rate. Three
months after
completing the
survey,
each team
received an
individual report,
providing
feedback
about their
team and
department results
compared
with the
overall ODI
results, along
with a brief
explanation
of how to
interpret
these
data.
During
this
time, to obtain
independent
data that
could help
establish the
construct
validity
of
survey variables
assessing
team
design,
another
researcher-blind
to
the
survey
re-
sults-interviewed 31
managers
who were
familiar with the
design
of one or more
of
the
51 teams and
who
had
not
served as
team observers. The interview
instrument
in-
cluded
questions to elicit informants'
descriptions of
team
design (goal,
task,
composition,
and
context
support), prob-
ing for factual
descriptions
and examples
rather than
evalua-
tions of the
team. The interviewer
reviewed the
tapes,
made
notes
and-using
a
five-point
scale from
very low
to
very
high-assessed
four
variables:
(1) presence
of
a clear
goal,
(2) team
task
interdependence, (3)
appropriateness
of
team
composition, and
(4)
context
support.
Phase 3:
Follow-up qualitative
research. From the team
survey data,
I
identified teams
with the
six lowest and six
highest
means for
team
learning
behavior;
seven of these
twelve
(four
high
and three
low)
were
available for
follow-up
observation
and interviews. The set of
seven teams con-
sisted of three
functional teams
(one
high-
and
two low-
learning),
two
product
development
teams
(high
and
low),
one
self-managed
team
(high),
and
one
project
team
(high);
none
of these
overlapped
with
the
eight teams
I
studied
in
the
first
phase.
I
observed six of these
teams,
individually
interviewed
one
or
two members of
each, and conducted
interviews
with
every
member
of
the
seventh team. The ob-
jective
of
this
phase
was to
explore
differences between
high-
and
low-learning
teams and to learn more
about how
team
learning
behavior
works. I
reviewed
these
field
notes
and
tapes
to
construct short
cases
describing
each
team,
which
were
then
used
to
suggest patterns
related to team
learning.
Measures
Antecedent factors.
l
coded the team
survey
to
identify
re-
spondents
by
team
rather than
by
individual and to
identify
team
type
(functional,
self-managed, product
development,
or
project)
and
company
department
(operations,
sales,
staff
362/ASQ,
June
1999
This content downloaded by the authorized user from 192.168.52.71 on Fri, 7 Dec 2012 15:29:43 PM
All use subject to JSTOR Terms and Conditions
Psychological
Safety
Table 2
Chronbach's Alpha and Intercorrelations between Group-level
Survey Variables*
Variable Mean S.D. 1 2
3 4 5 6 7 8 9 10 11 12
1. Context support 4.78 .97 .65
2.
Team leader coaching
3.77
.81 .69 .80
3. Team
psychological
safety 5.25
1.03 .70
.63
.82
4. Team
efficacy
5.07
1.07 .70
.50 .50 .63
5. Team
learning
behavior 4.67
.93
.68 .63 .80 .50 .78
6. Team performance 5.10 1.03 .60 .45
.72 .50 .71 .76
7.
Internal motivation
6.11 .68
(.03) (-.06)
.15 (-.02) .12 .33 .64
8.
Job involvement 3.30
1.69 -.16 -.22 (-.07) -.26 -.09 (-.01) .31 t
9. Team
tenure
(in years) 2.40
1.70
(-.06)
.34 -.26 -.15 -.16 -.09 (.05) (.01) t
10. Average company
tenure
(in years)
9.00 6.70
.33
-.31 .26
.15
.17
.14
(.06) (-.01)
.16 t
11.
Team learning
(observer rated)
3.48
.77 .49
-.48 .60 .52 .60
.34 -.16
(-.02)
-.21 .30 .84
12. Team
performance
(observer rated) 4.95
1.29
.48 -.50 .47 .43 .52 .36 -.1
1
-.12 -.21 .22 .81 .87
*
Chronbach's alpha coefficients
are presented on
the diagnonal.
Correlations
in parentheses not
significant at p
<
.05;
all
other
correlations are
significant at p
< .05.
t
Only
1
survey item.
services,
or
cross-functional).
I
included
in
the
survey
scales
developed
by Hackman (1990)
to assess team
design fea-
tures,
including context support
and team
leader coaching.
Team
shared
beliefs. I developed scales to measure team
psychological
safety and team efficacy, using
items
designed
to assess several features of each theoretical construct.
In
doing this,
I
also drew from qualitative data obtained
in
phase-1
interviews. Sample
items for
psychological safety
include
"If
you
make a mistake
on
this
team,
it
is often held
against you"
(reverse scored),
"It is safe
to take a risk on
this
team,"
and "No one on
this team would
deliberately
act
in a
way
that would undermine my
efforts." Team
efficacy
was measured with items
such as
"With
focus and
effort,
this
team
can
do
anything
we set out to
accomplish."
As
in
other
sections
of the
survey,
a
mix of
negatively
and
posi-
tively
worded
items was used
to
mitigate response
set bias.
(See
the
Appendix
for all
items.)
The
survey
also
measured
team tenure
(the average
number of
years
each member
had worked
in
the
team)
and
company
tenure
(respondents'
years
of
employment
at
ODI).
Between-scale correlations
for
variables
in
the model are
shown
in
table
2,
at the
group
level of
analysis (N
=
51).
Team
behavior.
I
developed
scales to assess the extent of
learning
behavior for both
the team and observer
surveys.
Team learning behavior includes
items such as "We
regu-
larly
take time to
figure
out
ways
to
improve
our team's
work
process"
and "Team
members
go
out and
get
all
the
information
they possibly
can
from others-such as from
customers or other
parts
of
the
organization."
Performance.
Hackman's
team
performance
scale was
used
to obtain
self-report
measures
of team
performance,
and
I
developed
a similar scale
for the observer
survey, including
"This team meets or exceeds
its
customers'
expectations"
and "This team
does superb work."
363/ASQ, J une
1 999
This content downloaded by the authorized user from 192.168.52.71 on Fri, 7 Dec 2012 15:29:43 PM
All use subject to JSTOR Terms and Conditions
1
Discriminant
validity was
also
established
by creating a
multitrait
multimethod
(MTMM) matrix
(Campbell and Fiske,
1959) for each
group of
variables,
from
which I
confirmed that,
for
antecedent
and
outcome
variables,
correlations be-
tween
items designed to
measure
the
same construct were
larger
than
correla-
tions between
these items and all other
items in the
section. For
the
antecedent
variables,
the average
within-trait, be-
tween-method
correlation
was .35, and
between-trait, between-method correla-
tions (between
each item of
a given
scale and all
items
in
other
scales)
aver-
aged .25.
For
the outcome
variables,
the
average
within-trait,
between-method cor-
relation was
.36 and
between-trait, be-
tween-method
correlations
averaged
.25.
2
Factor
analyses (principal
components,
varimax
rotation), using a
cut-off criterion
of .40 for factor
loadings
and
eigenvalues
of 1.0 or
above, yielded
six factors for
the
antecedent
variables,
replicating most
of
the planned scales:
items for team
psychological
safety,
team
efficacy,
team
task,
and clear
goal
loaded
onto four
fac-
tors
exactly
as
planned,
while context
support
items loaded onto
two
factors,
both
conceptually
related
to
context
sup-
port,
and team
composition items loaded
onto
the
first three
factors. All items
were retained
in
the
planned scales be-
cause they made
a positive
contribution
to
Cronbach's
alpha.
For the
team
out-
comes
section,
factor
analysis replicated
the
planned
scales-team
learning
behav-
ior
and team
performance.
To
test
whether
team
learning behavior and
team
psychological
safety
items were
tapping
into the same
issues rather
than
into two
distinct
constructs,
I ran a
factor
analysis
on
all items from
both
scales.
Reassur-
ingly, two
clean
factors
resulted, replicat-
ing
the
planned scales
precisely.
3
Three
of
the
four interview
variables
were
more correlated with the
survey scale mea-
suring
the same
construct than with
any
other scale in
the
survey.
The
interview
measure
of
adequacy
of team
composition
was more
highly
correlated
with
the
survey
measure of team
composition
(r
=
.33,
p
<
.01) than with
any
of the other
survey
vari-
ables. The
degree
of
task
interdependence
and
wholeness was
most
correlated with
the
survey
variable
assessing
task
design
(r
=
.34, p
<
.01),
and
context
support
was
most correlated with context
support
(r
=
.33, p
<
.01).
Although
the
differences be-
tween correlation
values
were
in
some
cases
small,
the
overall
degree
of conver-
gence
between
the two
different
instru-
ments
is
striking.
The
fourth interview vari-
able,
clear team
direction,
is more
correlated with
context
support
(r
=
.28,
p
<
.05)
than
with
the
survey
measure of
team direction
(r
=
.17, p
=
.12);
however,
this
result
is,
in
fact,
reassuring
for mea-
surement
reliability,
as
the
survey
and
in-
terview
"direction"
variables measured
two
distinct constructs. The
survey
mea-
sured
the
extent
to which time and effort
had
been
spent
on
clarifying
team
goals,
and
the
interviews asked to what extent
the team had a clear
shared
goal;
the low
correlation between
the two is thus not
surprising.
Team
feedback
variables.
Additional
variables, not
included
in
the team
learning
model, such
as
presence of a
clear
goal,
adequacy of team
composition, team
task
design,
quality of
team
relationships, job
satisfaction, job
involvement, and
in-
ternal
motivation were
included
in
the
team
survey for
the
purpose of
providing
supplementary
feedback to
the teams.
Adequacy of
Measures
I
conducted
preparatory
analyses
to
assess
psychometric
properties of
the two
new
instruments,
including
internal
consistency
reliability
and
discriminant
validity of the
scales.
The
results
supported the
adequacy of
most of the
mea-
sures for
substantive
analysis,
although
Cronbach's
alpha
was
low for
both
context
support and
team
efficacy (see
table
2). Discriminant
validity
was established
through factor
analysis.1 As the team
antecedent
and
outcome
sections
yielded,
respectively, six
and
three distinct
factors
with
eigenvalues
greater
than
one,
these results
demonstrated
that the team
survey was not
hampered
by excessive
com-
mon-method
variance,
according to
Harman's
one factor test
for
common-method bias.2
I
computed two
scales
from
the observer
survey (team
learning behavior
and team
performance),
and
both showed
high
internal
consistency
reliability (see
table
2).
Discriminant
validity was
lacking;
many
team
learning
behavior items
were
as
correlated
or
more
correlated
with
team perfor-
mance items as with
themselves.
Some of
this
between-
scale
(multitrait) correlation
can
be
attributed
to a
substantive
relationship
between team
learning behavior and
team
per-
formance;
however,
because of the
lack
of
discriminant
va-
lidity,
I
avoided
analyses
that
tested
relationships between
the
two
variables
in
the observer
survey.
Because it is
likely
that the team
observers or
customers are
in
a better
posi-
tion
to
judge performance-defined
in
part
as
meeting recipi-
ents'
needs-than to
assess
specific
behaviors,
which
they
may
not
always
observe,
substantive
analyses reported
be-
low
rely
primarily
on
observers'
ratings
of
performance
and
members'
ratings
of
behavior. Observers'
ratings
of
learning
are used
in
certain
analyses
to illustrate
consistency
in
re-
sults across
different measures of the same
construct. Pear-
son
correlations between team members' and
independent
observers'
responses
about
team
learning
(r
=
.60,
p
<
.001)
and
team
performance
(r
=
.36,
p
<
.01) provided
one mea-
sure of
construct
validity
for the team
survey.
A
substantial
degree
of
correspondence
between
analogous
measures
in
the
team
survey
and
structured interview data also contrib-
uted to
establishing
the
construct
validity
of
the
survey
mea-
sures of
teams' structural
features;
correlations between
each
team-structure scale
in
the
survey
and the
correspond-
ing
variable in
the structured
interviews
were
positive
and
significant.3
Finally,
a
group-level
variable must
satisfy
two
criteria
(Kenny
and
LaVoie,
1985). First,
the
construct must be
conceptually
meaningful
at the
group
level;
for
example,
team
size
is a
meaningful
group
attribute,
internal motivation is not. Sec-
ond,
data
gathered
from
individual
respondents
to
assess
the
group
attribute
must
converge,
such that the
intraclass
correlation (ICC) is
greater
than zero.
Intraclass correlation
364/ASQ, J une
1999
This content downloaded by the authorized user from 192.168.52.71 on Fri, 7 Dec 2012 15:29:43 PM
All use subject to JSTOR Terms and Conditions
4
To
generate
each intraclass correlation
coefficient, one-way analysis of
variance
(ANOVA)
was
conducted
on
the
full data
set
of 427
cases, with team
membership
as
the independent variable
and a
team
survey
scale as the
dependent
variable.
Intraclass correlations are
significant
when the one-way ANOVA from
which
the coefficients are derived are
significant
(Kenny and LaVoie, 1985).
Psychological Safety
Table 3
Analysis of Variance and Intraclass Correlation Coefficients for
Group-Level
Scales
Team survey variables
F(50,427)
P
ICC
Context support 4.80 <.001 .29
Team leader coaching 4.88 <.001 .30
Team psychological safety 6.98 <.001 .39
Team efficacy 5.70 <.001 .34
Team learning behavior 5.79 <.001 .27
Team performance 6.02 <.001 .35
Internal
motivation* 1.13 .07 .03
Job involvement* 1.25
.06 .04
Observers'
survey
variables
F(50,135)
P
ICC
Team
learning
behaviors
2.27
<.001 .19
Team performance 2.90 <.001
.21
*
Two variables that are conceptually individual-level variables are included for
purposes
of
comparison,
to demonstrate the contrast between these results
and those for the variables from the same
survey
that
are conceptually group-
level.
One-way
ANOVA shows these two variables are
not significantly differ-
ent
across
teams,
in
contrast to the
group-level variables,
which are
significant
to
the
p
<
.0001 level.
coefficients, measuring
the extent to which
team
members'
responses agree
with each other and differ
from
other
teams,
were
calculated for all
group-level variables
in
the
team
survey;
all were
significant
at the
p
<
.0001 level.4
Table 3 shows the results. It is
particularly noteworthy that
new
measures such as team psychological safety and team
learning behavior have high ICCs (.39 and .33, respectively),
satisfying
the
methodological prerequisite
for
group-level
variables. In
contrast, ICCs were near zero
for
constructs
that are
conceptually meaningful
at the individual rather than
group
level of
analysis (internal motivation,
with
ricc
=
.03
and
job
involvement
with
rick
=
.04);
the
data
thus confirm
that these individual-level constructs are less
likely
to be
shared within and
vary
across teams. ICCs were calculated
for observer variables as measures of interrater
reliability
for
different observers of the same
team;
these were also
posi-
tive and
significant.
These results allowed the creation
of a
group-level
data set
(N
=
51)
that
merged group
means for
group-level
variables from both
surveys.
RESULTS
Team
Psychological Safety, Efficacy, Learning Behavior,
and Performance
To test
hypotheses relating
team shared
beliefs, learning
be-
havior,
and
performance,
I
conducted a series of
regression
analyses, using
customers'
ratings
of team
performance
as
the
dependent
variable and measures obtained from
team
members as
regressors.
Because
respondents belonging
to
the same team are
not
independent,
I
performed regression
analyses
on the
group-level
data set
(N
=
51)
to avoid
violat-
ing
the
regression assumption
of
independence.
The results
are shown
in
table 4.
First, regressing
team
learning (self-
reported)
on team
performance (observer-rated)
reveals that
learning behavior
is a significant predictor of team
perfor-
365/ASQ, June 1999
This content downloaded by the authorized user from 192.168.52.71 on Fri, 7 Dec 2012 15:29:43 PM
All use subject to JSTOR Terms and Conditions
mance, supporting HI (model 1). This minimal test of two
key variables in the model was utilized to increase power,
given the small team
N,
and the same strategy was used to
test other core relationships, such as between team psycho-
logical safety and team learning behavior. To explore alterna-
tive models, I then introduced additional regressors into the
model-specifically, context support and team leader coach-
ing, which
in
previous studies have been used to explain
team performance-and these provided no additional ex-
planatory value, nor did they, without learning behavior, ac-
count for
more variance than learning behavior alone (table
4, models
2
and 3). Similarly, a series of alternative regres-
sors
(team psychological safety, team efficacy, context sup-
port,
and
team
leader
coaching) individually accounted
for
less of the variance
in
team performance than was ac-
counted
for
by team learning behavior. Thus,
of seven
alter-
native
models, team learning behavior accounts
for the most
variance in observer-rated team performance, providing sup-
port for
HI.
I
conducted four regressions to test hypotheses relating
team
psychological safety and team efficacy to
team learn-
ing behavior. To assess the consistency of these predictions
for
differing
data
sources,
I
first used
self-reported
team
learning behavior as the dependent variable and then
re-
peated
the
same analyses using observers' ratings
of team
learning behavior. The results reveal a high degree of consis-
tency across the
two
sets
of
equations using the two inde-
pendent measures
of
team
learning,
as shown
in
table
5.
First, regressing psychological safety on self-reported
team
learning behavior shows a significant positive relationship,
providing initial support for
H2
(panel A, model 1).
I
then
re-
gressed team efficacy alone on team learning behavior
to
test
H4, and although team efficacy accounts
for
substan-
tially
less
variance
than team
psychological safety,
the
rela-
tionship
was
positive
and
significant (panel A,
model
2). H5,
that team
efficacy
is
positively
associated with
team
learning
behavior when
controlling
for team
psychological safety,
was
not
supported
for
self-reported
team
learning (panel A,
model
3).
With
observer-rated
team
learning
as
the
depen-
dent
variable,
the results for
H2
and
H4
were similar
to
those obtained
using self-reported
team
learning (panel B).
When team
psychological safety
and
team
efficacy
were en-
tered into the model
together, however,
team
efficacy
re-
mained significant (model 3, panel B), providing
some
sup-
port
for H5.
Finally,
to
explore
alternative
models,
I
regressed
other antecedent variables on
team
learning
be-
havior, and,
as shown
in
table
5
(models 4, 5,
and
6),
team
psychological safety
accounts
for more variance
in
both self-
reported
and observer-assessed team
learning
behavior
than
context
support
or team
leader
coaching.
Model
7
then
shows that when all
regressors
are entered
into
the model-
for either measure of
team
learning behavior-only
team
psychological safety
is
significant. Together,
these results
provide
substantial
support
for
H2,
that
team
psychological
safety
is
associated
with
team
learning behavior; support
for
H4,
that team
efficacy
is
associated
with team
learning
be-
havior;
and mixed
support
for
H5,
that
team
efficacy predicts
team
learning
behavior
when
controlling
for team
psychologi-
cal
safety.
366/ASQ, June 1999
This content downloaded by the authorized user from 192.168.52.71 on Fri, 7 Dec 2012 15:29:43 PM
All use subject to JSTOR Terms and Conditions
Psychological
Safety
Table 4
Regression Models of
Observer-assessed Team Performance
(N= 51)
Model
Variable
(1) (2)
(3) (4)
(5) (6) (7)
Constant
1.31 .44
.87 1.41
2.04- 1.48 2.32-
Team
learning
behavior .80-- .40
Context support
.22
.43 .75--
Team
leader coaching
.45
.57
.93--
Team psychological safety
.67--
Team efficacy
.60--
Adjusted P-squared
.26 .27
.26 .21
.17 .22 .23
*
p <
.05;
*-p
<
.01.
Table 5
Regression Models of
Team Learning (N= 51)
Model
Variable
(1) (2)
(3) (4)
(5) (6)
(7)
A.
Team learning (self-report)
Constant
.69 2.45--
.41 1.06
1.4100 1.9400
.42
Team
psychological
safety .76--
.70
.51 -
Team
efficacy
.45--
.11
-.05
Context support
.52--
.69--
.28
Team
leader coaching
.31
.75--
.14
Adjusted R-squared
.63 .23 .63 .52
.45
.38 .66
B.
Team
learning
behavior (Observer-assessed)
Constant
.990 1.5--
.48 1.27-
1.53-- 1.7300 .53
Team
psychological
safety .46--
.35 .330
Team
efficacy
.38--
.22-
.21
Context support
.27-
.40-- -.07
Team
leader
coaching
.23 .46--
.12
Adjusted R-squared
.35 .26
.40 .26
.23
.21
.36
p
<
.05;
*-p
<
.01.
To test
H3,
that team
learning
behavior mediates
the
effects
of
team
psychological
safety
on team
performance,
I
con-
ducted a
three-stage analysis
to test whether
three condi-
tions
for mediation
were
satisfied:
(1)
the
proposed
mediator
significantly
predicts
the
dependent
variable, (2)
the
indepen-
dent variable
predicts
the
mediator,
and
(3)
the contribution
of
the
independent
variables
drops
substantially
for
partial
mediation and becomes
insignificant
for
full mediation when
entered
into the
model
together
with the mediator
(Baron
and
Kenny,
1986).
In
these
analyses,
I
used
observers'
rat-
ings
of
performance
as
the
dependent
variable
and self-re-
ported
team
learning
behavior
as the
mediating variable,
be-
cause
this
created a
higher
hurdle for
demonstrating
a
relationship
between team
learning
and team
performance.
As
shown
above,
team
learning
behavior
is
significantly posi-
tively
associated with
team
performance,
supporting
the first
of
the
three conditions.
The second
condition,
that the inde-
pendent
variable
(team psychological
safety) significantly pre-
dicts
the
proposed
mediator
(team
learning behavior)
also
was
established above.
Finally,
the third condition
for
media-
tion is also satisfied: the contribution
of team psychological
367/ASQ, June 1999
This content downloaded by the authorized user from 192.168.52.71 on Fri, 7 Dec 2012 15:29:43 PM
All use subject to JSTOR Terms and Conditions
safety becomes insignificant (B
=
.25, p
=
.42)
when entered
into
the regression model together
with
team learning,
which remains
significant (B
=
.60, p
<
.05).
Context
Support,
Leader
Behavior, Psychological
Safety,
and
Learning
Behavior
Next,
I used
regression to
test
H6, H7, and H8,
followed by
GLM
analysis to
further
explore
the
relationships
in H6, that
team leader coaching and context support are positively
as-
sociated
with
team psychological safety. Results
are
shown
in
table 6. As a first test, I regressed these two
variables on
team psychological safety. Both were positively
related to
the dependent variable; context support was significant
and
team
leader coaching was close to significant (condition
2).
In
testing H7, that team psychological safety
mediates be-
tween
coaching,
context
support, and team
learning behav-
ior, the first two conditions of the three-step analysis
were
already
satisfied-team
psychological safety predicted
team
learning behavior, and context support
and
coaching pre-
dicted team
psychological safety.
The third condition was
also
satisfied;
when all three
predictors
were entered
into
the model
simultaneously,
the
effects
of
context
support
and
coaching
were
insignificant,
and team
psychological
safety
remained
significant.
This
result
supports
H7,
as did
repeating
the
three-step analysis for self-reported
learning
behavior.
In
contrast, the results shown
in
table
7
do
not support H8,
that team
efficacy
functions as a mediator.
Team
efficacy
predicted
observer-rated team
learning behavior,
but of the
two
independent variables, only
context
support
predicted
the
proposed mediator, team efficacy. Despite
insufficient
support
for the
second
condition,
I
checked
the
third
condi-
tion
by regressing
context
support
and
team efficacy
on ob-
server-rated team
learning
and found that
the
effects
of
con-
text
support
were
insignificant,
while team
efficacy
remained
barely significant. Finally, using self-reported
team
learning,
I
found
no
support
for mediation.
Next,
I examined
relationships
between context
support,
team leader
coaching,
team
psychological safety,
and team
learning behavior using
GLM
analyses
on the individual-level
data set
(N
=
427).
This allowed simultaneous
testing
of
ran-
dom effects of team
membership
and fixed effects
of team
type
while
exploring
the
relationship
between
predictor
vari-
Table
6
Tests of Team
Psychological Safety
as a
Mediator between Coaching, Context Support,
and
Learning
Observer-assessed
Self-report learning
learning
behavior behavior
Conditions to
demonstrate mediation*
Independent
variable
B t
P
R2 B
t
P
R2
1.
Does
psychological safety significantly Team psychological safety .46 5.26 <.001 .33 .76
9.16 <.001 .63
predict team learning?
2.
Do
coaching and context support Team leader coaching .33
1.89
.06 .52
.33
1.89
.06 .52
significantly predict
team
psychological
Context
support
.56
3.82
<.001
.56 3.82 <.001
safety?
3. Does
the effect of the
antecedents
drop
Team
psychological safety
.29 2.46 .02 .33 .51 4.56
<.001 .66
substantially
or
become insignificant? Context support .09
.66 .51 .24 1.83
.07
(Team learning)
Team leader
coaching
.12 .78 .21
.14 1.00
.32
*
Dependent
variables are in
italics.
368/ASQ, June 1999
This content downloaded by the authorized user from 192.168.52.71 on Fri, 7 Dec 2012 15:29:43 PM
All use subject to JSTOR Terms and Conditions
Psychological Safety
Table 7
Tests of Team Efficacy as a Mediator between Coaching, Context Support, and Learning
Observer-assessed Self-report learning
learning
behavior behavior
Conditions to
demonstrate mediation* Independent variable
B
t
P
R2 B
t
P
R2
1.
Does team efficacy significantly predict Team efficacy .45 3.97 <.001 .23 .38 4.27 <.001 .26
team learning?
2. Do coaching and context support Team leader coaching .01 .06 .95 .01 .06 .95
significantly predict
team
efficacy? Context support .77 5.01 <.001 .49 .77 5.01 <.001 .49
3.
Does
the
effect of the antecedents drop Team efficacy .26 2.05 .05 .02 .21 .83
substantially
or
become
insignificant? Context support .20 1.46
.15 .27
.67 4.44 <.001 .52
(Team learning)
*
Dependent variables are
in
italics.
ables and either team
psychological safety
or team
learning
behavior.
Despite mean
differences across team
types
in
both
team
psychological safety
and team
learning behavior,
GLM
analyses revealed
that these differences could
be
ex-
plained by context support and
team
leader coaching.
As
shown
in
table
8, context support,
team leader
coaching,
and
team
membership (random
effects
of
belonging
to the
same
team)
were
significant predictors
of individuals'
ratings
of team
psychological safety.
In
contrast,
the effect
of
team
type was insignificant. Controlling
for team tenure revealed
that
its
effects
on
team psychological safety were
also
insig-
nificant.
Similarly, team psychological safety,
team
efficacy,
and
team
membership
were
significantly
related
to
team
learning behavior, while
team
type and team tenure again
were insignificant. Although the GLM analyses allowed a
more
detailed apportioning of
the variance
in
individuals' re-
sponses
than the
group-level
data
set,
which
uses
team
means as data
points,
the direction and
magnitude
of the
results are consistent
with those
obtained
using regression
analysis.
Exploring
Differences between
High-
and
Low-
learning
Teams
I used the data from the seven teams identified
in
phase
3
as
high-
or
low-learning
to better
understand the
relationship
between team
psychological safety
and
learning
behavior.
Table 8
Results
of GLM
Analyses (N= 427)
Model Independent variable F-ratio
p
Team
psychological safety
Team
type F(3,51)
=
2.02
.12
P2 =
.60
Team membership* F(50,427)
=
3.25 <.001
Context
support
F(1
,427)
=
26.83 <.001
Team leader
coaching
R1,427)
=
39.81
<.001
Team tenure
R1,427)
=
0.10
.74
Team
learning behavior Team type F(3,51)
=
2.21 .10
R2
=
.53
Team
membership* F(50,427)
=
2.64
<.001
Team
psychological safety
F(1
,427)
=
42.21 <.001
Team
efficacy
F(1,427)
=
10.52 <.001
Team
tenure
F(1,427)
=
0.22 .49
*
Team
membership
is
the
categorical
variable
identifying
each
team.
The
result that team
membership
accounts
for
significant
variance
in
team
psychological safety
or in team
learning
behavior indicates that
variance is attributable
to
unexplained effects of belonging to the same team.
369/ASQ, J une 1999
This content downloaded by the authorized user from 192.168.52.71 on Fri, 7 Dec 2012 15:29:43 PM
All use subject to JSTOR Terms and Conditions
My goal
in
studying them
was to learn
more
about
how
they
functioned as teams
rather than to confirm or disconfirm a
model. Table 9
summarizes these qualitative