-
The Effect of Rationale Provision on Motivation and
PerformanceOutcomes: A Meta-Analysis
Rebecca R. Steingut, Erika A. Patall, and Scott S.
TrimbleUniversity of Texas at Austin
A meta-analysis of 23 experimental studies examined the effect
of rationale provisionon subjective task value, autonomous
motivation, engagement, performance, perceivedautonomy, perceived
competence, perceived relatedness, and controlled
motivationcompared with a control condition. Results suggested that
rationale provision enhancedsubjective task value, engagement,
performance, and perceived autonomy to a small tomoderate extent (d
� .16 to d � .40) under fixed- and random-effects models.
Resultsalso suggested that rationale provision diminished perceived
competence under bothfixed- and random-effects models (d � �.19),
but did not impact autonomous moti-vation, controlled motivation,
or perceived relatedness. Moderator analyses for subjec-tive task
value, autonomous value, motivation, engagement, and performance
sug-gested that rationales were most effective (a) when they were
prosocial or autonomouscompared to controlling (for 3 outcomes),
(b) for samples with a higher proportion offemales (for 3
outcomes), and (c) for uninteresting tasks (for 3 outcomes).
Implicationsfor theory, future research, and practice are
discussed.
Keywords: meta-analysis, motivation, rationale,
self-determination theory, subjectivetask value
People are regularly faced with the challenge ofmotivating
another individual. Parents, teachers,employers, and doctors
regularly use one or moreof the following strategies to motivate
anotherindividual to engage in a task: (a) external contin-gencies
such as deadlines, rewards, punishment,(b) setting goals, or (c) an
emphasis on autonomyin the form of providing choices. Indeed,
compre-hensive syntheses of research show the successand
limitations of these strategies for motivatingothers (Deci,
Koestner, & Ryan, 1999; Locke &Latham, 2002; Patall,
Cooper, & Robinson, 2008).This meta-analysis adds to this body
of researchby synthesizing the effects of providing a ratio-nale,
another promising motivational strategy.
Rationale refers to the “verbal explanation ofwhy putting forth
effort during the activity
might be a useful thing to do” (Reeve, Jang,Hardre, & Omura,
2002, p. 185). Examples ofrationale provision include: a teacher
discussingthe way students may use information in thefuture,
parents referring to rewards associatedwith a chore, or a doctor
encouraging a patientto quit smoking by referring either to the
in-creased risk of cancer or to associated changesin physical
appearance. Across these differentexamples, rationale provision is
social: the ex-planation for the task’s value is provided byanother
individual. Though rationale is a moti-vational tool employed in
educational, work,and health contexts, and a central component
inseveral motivational theories, including Self-Determination
Theory (SDT; Ryan & Deci,2000b), there is little consensus
regarding char-acteristics that may enhance or diminish its
ef-fects. The present analysis examines the overalleffect of
rationale on motivation and perfor-mance outcomes and moderators of
that effect.
Theoretical Perspectives onRationale Provision
Self-determination theory (SDT; Ryan &Deci, 2000b) proposes
that satisfaction of thepsychological needs for autonomy,
competence,
Rebecca R. Steingut, Erika A. Patall, and Scott S.Trimble,
Department of Educational Psychology, Univer-sity of Texas at
Austin.
Erika A. Patall is now at University of Southern Califor-nia,
Rossier School of Education.
Correspondence concerning this article should be ad-dressed to
Rebecca R. Steingut, Department of Educa-tional Psychology,
University of Texas at Austin, 1 Uni-versity Station D5800, Austin,
TX 78712. E-mail:[email protected]
Thi
sdo
cum
ent
isco
pyri
ghte
dby
the
Am
eric
anPs
ycho
logi
cal
Ass
ocia
tion
oron
eof
itsal
lied
publ
ishe
rs.
Thi
sar
ticle
isin
tend
edso
lely
for
the
pers
onal
use
ofth
ein
divi
dual
user
and
isno
tto
bedi
ssem
inat
edbr
oadl
y.
Motivation Science © 2017 American Psychological
Association2017, Vol. 3, No. 1, 19–50 2333-8113/17/$12.00
http://dx.doi.org/10.1037/mot0000039
19
mailto:[email protected]://dx.doi.org/10.1037/mot0000039
-
and relatedness underlie motivation, particu-larly powerful
autonomous or intrinsic forms ofmotivation as opposed to
controlling or extrin-sic forms that may not sustain behavior as
ef-fectively across time or contexts. As such, SDTprioritizes an
examination of the contextual fac-tors that support satisfaction of
these needs.
SDT has identified several practices that sup-port the need for
autonomy, a sense of volitionor ownership over one’s behavior
(Deci,Eghrari, Patrick, & Leone, 1994; Reeve & Jang,2006),
including providing a rationale, whichexplains the value of the
task and, therefore,gives the individual a reason to endorse it.
Inother words, hearing about the task’s value isthought to
encourage individuals to feel moreownership, or to feel that their
work on the taskis more closely aligned with their own
personalgoals. Even if the activity is not freely chosen,as is
often the case with school assignments orwork responsibilities, SDT
predicts that ratio-nale provision will facilitate the sense that
onewould choose the task if it were not required,because that
individual understands the value ofthe task.
Although SDT primarily refers to rationaleprovision as an
autonomy supportive practice,SDT theorists argue that rationale may
alsosupport satisfaction of the need for competence,that is, the
sense of being able to succeed at atask, by providing structure or
informationabout the link between behaviors and outcomes(Grolnick,
Deci, & Ryan, 1997; Grolnick, Gur-land, Jacob, & Decourcey,
2002). For example,Grolnick et al. (2002) suggest that parents
ex-plain cleaning to a young child in terms ofhaving enough “space
to play on the floor” (p.159), thus connecting cleaning up to the
child’sown goals (autonomy support) and helping thechild understand
how to achieve desired out-comes (competence support). Thus, SDT
ex-pects rationale provision to support motivationprimarily because
it supports an individual’sexperience of autonomy and, secondarily,
be-cause it supports his or her sense of compe-tence.
In contrast to SDT, which is intended toapply across contexts,
expectancy value theory(EVT), as proposed by Eccles, Wigfield
andcolleagues (Eccles et al., 1983; Wigfield &Eccles, 2000),
theorizes motivation withinachievement contexts, but, like SDT,
providespredictions about the effects of rationale provi-
sion. EVT holds that subjective task value, orbeliefs about the
reasons for undertaking agiven task as well as an individual’s
expectan-cies, beliefs about how well he or she willperform on a
given task, are central to achieve-ment motivation, predicting
engagement, per-sistence, and performance (Wigfield &
Eccles,2000). To the extent that rationales lead to in-creased
subjective task value, EVT suggeststhat motivation and performance
will also in-crease. Thus, the EVT model suggests that sub-jective
task value is an important outcome andmechanism of effect of
rationale.
Although both SDT and EVT suggest thatrationale provision likely
supports subjectivetask value, motivation, engagement, and
perfor-mance, the theories differ in their explanationsfor the
mechanism by which rationales havesuch benefits. SDT emphasizes the
mediatingrole of enhanced autonomy and competence inexplaining why
the provision of rationales isexpected to have benefits for more
distal out-comes like intrinsic motivation, engagement,and
performance, whereas EVT emphasizes themediating role of enhanced
subjective task val-ue. Moreover, while SDT suggests that en-hanced
competence may be an outcome andmechanism of effect for rationale
provision,empirical research emanating from EVT hasfound that the
effect of rationale on perceivedcompetence may depend
characteristics of theindividual, including individual interest
and/orperceived competence (Durik & Harackiewicz,2007; Durik,
Shechter, Noh, Rozek, & Harack-iewicz, 2015). Given the
conflicting theory andevidence, it is difficult to make a
predictionregarding whether the provision of a rationalewill
enhance or diminish perceptions of compe-tence overall. However,
taking both theoriesinto consideration, we expected that the
provi-sion of rationale would enhance perceptions ofcompetence more
often that it does not.
Empirical findings generally support theoreticalpredictions that
rationale yields motivational ben-efits. In both field and
laboratory studies withindividuals of varied ages, the provision of
a ra-tionale has been found to lead to: Subjective taskvalue (Durik
& Harackiewicz, 2007), interest/enjoyment (Deci et al., 1994),
engagement (Jang,2008; Sansone, Wiebe, & Morgan, 1999),
learn-ing (Jang, 2008), and performance (Kuczynski,1983). For
instance, Jang (2008) provided a ratio-nale for learning about
correlations that pointed
20 STEINGUT, PATALL, AND TRIMBLE
Thi
sdo
cum
ent
isco
pyri
ghte
dby
the
Am
eric
anPs
ycho
logi
cal
Ass
ocia
tion
oron
eof
itsal
lied
publ
ishe
rs.
Thi
sar
ticle
isin
tend
edso
lely
for
the
pers
onal
use
ofth
ein
divi
dual
user
and
isno
tto
bedi
ssem
inat
edbr
oadl
y.
-
out how learning about correlations would be use-ful for
participants’ future role as teachers. Partic-ipants who heard the
rationale reported higherlevels of autonomous motivation,
demonstratedmore behavioral engagement, and learned more.Despite
these positive effects, a thorough reviewof the empirical
literature does not suggest a clearpicture. Some research has found
rationales tohave null effect on outcomes including free-choice
engagement (Deci et al., 1994), perfor-mance (Durik &
Harackiewicz, 2007; Shin, 2010),and other motivational variables
(Gillison,Standage, & Skevington, 2013; Shin, 2010). Fur-ther,
it would seem that the generally positiveeffects of rationale may
be enhanced or dimin-ished by a number of factors.
Factors That May Influence theEffectiveness of Rationale
Mixed findings suggest that the effects ofrationale are complex
and may not always bepositive. First, we discuss factors for
whichstrong theoretical logic exists for expectingvariation in the
effects of rationale along thefactor. We then discuss factors that
may beimportant to explore as moderators of the ef-fects of
rationale, though theoretical reasons forexpecting variation along
these factors are morelimited.
Type of Rationale
A central question regarding the effects ofrationale is what
practitioners should say whenproviding a rationale. SDT provides a
usefulframework for categorizing these statements ofrationale into
two types: autonomous and con-trolling and suggests that autonomous
rationaleswill lead to larger effects on adaptive outcomesthan
controlling rationales. Specifically, theysuggest that because of
their association withthe fundamental needs of autonomy,
compe-tence, and relatedness, both autonomous formsof regulation
and intrinsic goal contents lead tomore adaptive outcomes than
controlled regula-tion or extrinsic goal contents. Next, we
defineand discuss various forms of regulation and goalcontents and
their application to explainingvariation in the effects of
rationale.
According to SDT (Ryan & Deci, 2000a),motivation can be
conceptualized on a contin-uum ranging from autonomous to
controlled.
Autonomous motivation includes intrinsic,identified, and
integrated forms of motivationand reflects doing a task because it
is interestingor enjoyable, personally meaningful, or becauseit has
been “brought into congruence with one’sother values and needs”
(Ryan & Deci, 2000b,p. 73). In contrast, more extrinsic or
controlledmotivation reflects doing a task because of ex-ternally
imposed incentives or consequences(external motivation) or because
of internalfeelings of obligation or pride (introjected
reg-ulation). Whereas all types of motivation maylead an individual
to engage in a task, a keyinsight of SDT and corresponding
empiricalresearch is that the quality of motivation differsacross
these types of motivation, autonomousmotivation being more strongly
associated withadaptive outcomes like learning and well-beingthan
controlled motivation (Grolnick & Ryan,1987). Given that some
rationales express valuein ways that correspond to these types of
regu-lation, rationales referring to autonomous rea-sons for
engagement are expected to lead tolarger effects on adaptive
outcomes than ratio-nales referring to more extrinsic or
controlledreasons, which we refer to as controlling
ratio-nales.
In addition, another aspect of SDT relates tothis question of
what practitioners should saywhen providing a rationale: goal
contents,which is another important factor impactingmotivation and
performance outcomes. Specif-ically, pursuit of intrinsic goals,
which are “con-gruent with actualizing and growth tendenciesnatural
to humans,” and include “health andgrowth” and prosocial goals like
“communityand helpfulness” (Kasser & Ryan, 1996, p.280), is
associated with well-being and reduceddistress. In contrast, the
pursuit of extrinsicgoals, which depend on the reactions of
others,and include “money . . . , social recognition . . ., and . .
. appearance” (Kasser & Ryan, 1996, p.280), is associated with
less self-actualizationand vitality and more physical symptoms.
Thegoal framing hypothesis applies these otherwiseintraindividual
goal contents directly to ratio-nale provision: rationales
referring to intrinsicgoals may promote motivation and
perfor-mance, whereas rationales referring to extrinsicgoals may
undermine such outcomes (Vans-teenkiste, Niemiec, & Soenens,
2010). In fact,evidence has supported goal framing predic-tions
that autonomous rationales have greater
21EFFECTS OF RATIONALE
Thi
sdo
cum
ent
isco
pyri
ghte
dby
the
Am
eric
anPs
ycho
logi
cal
Ass
ocia
tion
oron
eof
itsal
lied
publ
ishe
rs.
Thi
sar
ticle
isin
tend
edso
lely
for
the
pers
onal
use
ofth
ein
divi
dual
user
and
isno
tto
bedi
ssem
inat
edbr
oadl
y.
-
effects on adaptive outcomes than controllingrationales (Reeve
et al., 2002; Vansteenkiste,Simons, Lens, Sheldon, & Deci,
2004; Vans-teenkiste, Simons, Lens, Soenens, et al.,
2004;Vansteenkiste, Simons, Soenens, & Lens,2004).
The distinction among types of rationalesmay be further
complicated by the fact thatsome rationales reflect a prosocial
value, mean-ing they refer to the value that task engagementwill
accrue to someone other than the partici-pant him or herself. In
the present analysis, wedistinguish between prosocial rationales
and au-tonomous rationales focused only on benefits tothe self in
light of some findings that prosocialrationales may bring about the
biggest boost tomotivation and performance.1 For example,Grant and
colleagues (e.g., Grant, 2008b, 2012;Grant et al., 2007) argued
that focusing on thebeneficiaries of a given task may promote
mo-tivation to a greater extent compared with fo-cusing on oneself
and found in one study inwhich autonomous and prosocial
rationaleswere explicitly compared (Grant & Hofmann,2011) that
a prosocial rationale emphasizing thebenefits to patients’ health
led to more handwashing among health care professionals than
arationale that emphasized the benefits to partic-ipants’ own
health. Though few other studieshave explicitly compared autonomous
andprosocial rationales, studies have certainly var-ied on this
dimension (see Table 1 for defini-tions and examples of each type
of rationale). Inother words, whereas prosocial rationales
areother-focused or “self-transcendent” (Yeager etal., 2014), both
autonomous and controlling ra-tionales refer to value that is
self-focused. Interms of the fundamental needs theorized bySDT,
prosocial rationales may have a particu-larly pronounced positive
effect because theymay also support feelings of relatedness.
Whilethe goal contents theory of SDT considersprosocial goals as a
subtype of intrinsic goals,more recently, researchers have
theorized con-sidering the two as separate but complementary:Grant
(2008a) found that intrinsic motivationstrengthens the effect of
prosocial motivationand Yeager et al. (2014) found that
interven-tions referring to both prosocial and self-oriented
reasons for learning may provide ben-efits greater than
intereventions that refer onlyto self-oriented reasons. Given these
findingsand theoretical debate regarding the categoriza-
tion of prosocial goals/motivation, the presentwork will compare
the effects of prosocial andautonomous rationales.
Finally, it is worth noting at this point that theabove
predictions regarding differential effectsacross types of rationale
are in some contrast topredictions that might be derived from
EVT.EVT defines subjective task value as a multidi-mensional
construct composed of four sub-types: personal importance or
significance of atask to one’s sense of self or identity
(attainmentvalue), usefulness of the task to obtaining per-sonal
goals in the future (utility value), enjoy-ment associated with a
task (intrinsic value),and resources needed or negative
consequencesof task engagement such as loss of time, exer-tion of
effort, negative affect, or inability toengage in other valued
tasks (cost; Eccles et al.,1983). Whereas attainment value, utility
value,and intrinsic value each increase subjective taskvalue, cost
decreases subjective task value.
Despite the distinction between these types ofsubjective task
value, EVT makes little distinc-tion between forms of subjective
value in termsof the nature of their effects (Vansteenkiste,Lens,
& Deci, 2006; Vansteenkiste et al., 2010).Hulleman, Barron,
Kosovich, and Lazowski(2016) suggest that the types of subjective
taskvalues vary along the same dimension as thetypes of motivation,
from more intrinsic or au-tonomous (e.g., intrinsic value) to less
so (e.g.,utility value or attainment value). Given thatEVT does not
differentiate the benefits of vari-ous forms of value (or the
rationales that refer toeach form) and SDT predicts that the
effects ofrationales will vary along a continuum fromcontrolling to
autonomous, the predictions ofeach theory are somewhat in conflict.
Thismeta-analysis provides an opportunity to com-pare and test the
predictions of each theory.
1 By type of rationale, we refer to the type of valuereferred to
in the rationale expression itself. This variable isdistinct from
characteristics of the setting in which therationale was provided
and from the task for which therationale was provided. For example,
a task that one mightconsider prosocial, such as recycling, may be
justified witha controlling rationale, such as by referring to the
moneythat subjects will get if they recycle (Vansteenkiste,
Simons,Lens, Sheldon, & Deci, 2004, Study 1). In this example,
therationale, would be considered to be controlling although itwas
provided for a prosocial task. See Table 1 for furtherexamples and
definitions.
22 STEINGUT, PATALL, AND TRIMBLE
Thi
sdo
cum
ent
isco
pyri
ghte
dby
the
Am
eric
anPs
ycho
logi
cal
Ass
ocia
tion
oron
eof
itsal
lied
publ
ishe
rs.
Thi
sar
ticle
isin
tend
edso
lely
for
the
pers
onal
use
ofth
ein
divi
dual
user
and
isno
tto
bedi
ssem
inat
edbr
oadl
y.
-
Tab
le1
Typ
eof
Rat
iona
le
Typ
eD
efini
tion
Exa
mpl
eT
ask
(Set
ting)
Aut
onom
ous
Stat
emen
tsof
valu
eth
atem
phas
ize
doin
ga
task
beca
use
ofbe
nefit
sto
the
self
,inc
ludi
ngho
wa
task
is:
“Han
dhy
gien
epr
even
tsyo
ufr
omca
tchi
ngdi
seas
es”
(Gra
nt&
Hof
fman
,20
11).
Han
d-w
ashi
ng(W
orkp
lace
setti
ngof
heal
thca
repr
ofes
sion
als)
•in
tere
stin
gor
enjo
yabl
e,pe
rson
ally
mea
ning
ful,
oral
igne
dw
ithpe
rson
al“v
alue
san
dne
eds”
(Rya
n&
Dec
i,20
00b,
p.73
)•
alig
ned
with
the
goal
ofgr
owth
,(e
.g.
pers
onal
impr
ovem
ent,
lear
ning
)
“Doi
nga
little
tae-
bohe
lps
you
rem
ain
phys
ical
lyfit
and
prev
ents
you
from
beco
min
gsi
ckat
ala
ter
age”
(Van
stee
nkis
te,
Sim
ons,
Soen
ens,
etal
,20
04).
Tae
-bo
(Hig
hSc
hool
Phys
ical
Edu
catio
nC
lass
)
“..
.if
you
choo
seto
eat
heal
thy
food
slik
eap
ples
you
will
have
mor
een
ergy
topl
ayan
dbe
activ
e,es
peci
ally
with
your
frie
nds”
(Ban
non
&Sc
hwar
tz,
2006
)
Hea
lthy
snac
kch
oice
(Kin
derg
arte
ncl
assr
oom
)•
alig
ned
with
the
goal
ofhe
alth
Con
trol
ling
Stat
emen
tsof
valu
eth
atem
phas
ize
doin
ga
task
beca
use
ofbe
nefit
sto
the
self
,in
clud
ing:
•ex
tern
ally
impo
sed
ince
ntiv
esor
cons
eque
nces
(e.g
.m
oney
,gr
ades
)as
soci
ated
with
the
task
•in
tern
alfe
elin
gsof
oblig
atio
nor
prid
e(a
lso
incl
udin
ggu
ilt)
•al
igne
dw
ithth
ego
alof
soci
alre
cogn
ition
•al
igne
dw
ithth
ego
alof
apo
sitiv
eap
pear
ance
“Rea
ding
the
text
coul
dte
ach
you
how
tosa
vem
oney
byre
usin
gm
ater
ials
”(V
anst
eenk
iste
,Si
mon
s,L
ens,
etal
.,20
04).
Lea
rnin
gab
out
recy
clin
g(C
olle
gest
uden
ts,
labo
rato
ryse
tting
)
“Doi
nga
little
tae-
bohe
lps
you
rem
ain
phys
ical
lyap
peal
ing
toot
hers
and
prev
ents
you
from
gain
ing
wei
ght
ata
late
rag
e.”
(Van
stee
nkis
te,
Sim
ons,
Soen
ens,
etal
.,20
04).
Tae
-bo
(Hig
hSc
hool
Phys
ical
Edu
catio
nC
lass
)
“The
reas
onw
ear
eas
king
you
totr
yha
rddu
ring
conv
ersa
tiona
lC
hine
seis
beca
use
we
are
goin
gto
give
you
ate
ston
the
mat
eria
lto
eval
uate
how
wel
lyo
ust
udie
dth
ein
form
atio
n.”
(Ree
veet
al.,
2002
,St
udy
1)
Lea
rnin
gco
nver
satio
nal
Chi
nese
(Col
lege
stud
ents
,la
bora
tory
setti
ng)
Pros
ocia
lSt
atem
ents
ofva
lue
that
emph
asiz
edo
ing
ata
skbe
caus
eof
bene
fits
toot
hers
,in
clud
ing
alig
nmen
tof
the
task
with
:
“Han
dhy
gien
epr
even
tspa
tient
sfr
omca
tchi
ngdi
seas
es”
(Gra
nt&
Hof
fman
,20
11).
Han
d-w
ashi
ng(W
orkp
lace
setti
ngof
heal
thca
repr
ofes
sion
als)
•th
ego
alof
com
mun
ity•
the
goal
ofhe
lpfu
lnes
s“R
eadi
ngth
ete
xtco
uld
help
you
know
how
tote
ach
your
futu
reto
ddle
rsth
atth
eyca
ndo
som
ethi
ngto
help
the
envi
ronm
ent”
(Van
stee
nkis
te,
Sim
ons,
Len
s,et
al.,
2004
).
Lea
rnin
gab
out
recy
clin
g(C
olle
gest
uden
ts,
labo
rato
ryse
tting
)
“We’
vere
ceiv
edle
tters
from
the
dean
’sof
fice
wri
tten
byst
uden
tsw
hoha
vebe
nefit
edfr
omsc
hola
rshi
psth
atw
ere
mad
epo
ssib
leby
the
alum
nido
natio
nsth
atyo
u’ve
solic
ited.
We
wan
ted
tosh
are
thes
ele
tters
with
you
togi
veyo
ua
sens
eof
the
impa
ctth
atyo
urw
ork
isha
ving
onst
uden
ts.”
(Gra
ntet
al.,
2007
,Stu
dy1)
Fund
rais
ing
calls
(Col
lege
stud
ents
empl
oyed
asfu
ndra
iser
s)
23EFFECTS OF RATIONALE
Thi
sdo
cum
ent
isco
pyri
ghte
dby
the
Am
eric
anPs
ycho
logi
cal
Ass
ocia
tion
oron
eof
itsal
lied
publ
ishe
rs.
Thi
sar
ticle
isin
tend
edso
lely
for
the
pers
onal
use
ofth
ein
divi
dual
user
and
isno
tto
bedi
ssem
inat
edbr
oadl
y.
-
Autonomy-Support
Beyond type of rationale, SDT suggests thatthe effects of
rationales may also vary to theextent that they are delivered alone
or in com-bination with other autonomy-supportive prac-tices.
Research examining the causal effects ofother autonomy-supportive
practices, includingchoice provision, acknowledgment of
negativeaffect, and noncontrolling language, either indi-vidually
or together as a whole, has often foundsuch practices to support
motivation, engage-ment, and performance (Reeve, Jang, Carrell,Jeon
& Barch, 2004; Vansteenkiste, Simons,Soenens, & Lens,
2004). SDT suggests a gestalthypothesis, that the effects of
autonomy sup-portive practices are synergistic and specificallythat
the effects of each practice are heightenedwhen combined (Deci et
al., 1994). As such,providing rationales within the context of
anautonomy-supportive environment overall isexpected to lead to
more positive effects thanrationales delivered in an otherwise
controllingcontext. Previous research supports this predic-tion.
Vansteenkiste, Simons, Lens, Soenens,and Matos (2005) found that,
regardless of ra-tionale type (autonomous, controlling),
ratio-nales provided using noncontrolling languageled to more
positive effects on autonomous mo-tivation, performance, and
engagement than ra-tionales delivered with controlling
language.Similarly, Deci et al. (1994) found that theeffects of
rationales on motivation and engage-ment were greatest when
combined with non-controlling language and acknowledgment
ofnegative affect. This synergy or gestalt effectsuggested that the
effects of autonomy support-ive practices were amplified as more
practiceswere provided simultaneously. That is, the ef-fect of all
three practices provided simultane-ously was larger than that of
one or two prac-tices.
Uninteresting Tasks
In addition to the characteristics of the ratio-nale,
characteristics of the task may also influ-ence the effects of
rationale provision. Specifi-cally, if the value of the task in
question isevident to the individual, then the rationale mayhave a
smaller effect because, regardless of anintervention such as
rationale, individuals arelikely to be highly motivated for the
task. Ac-
cording to scholars from a variety of perspec-tives including
interest theories (Sansone, Weir,Harpster, & Morgan, 1992;
Sansone et al.,1999) and SDT (Deci et al., 1994; Jang, 2008;Reeve
et al., 2002), rationales will be moreeffective when the task or
activity is relativelyboring or uninteresting because more
interest-ing activities do not require as much externalsupport for
motivation (Hidi & Renninger,2006; Ryan & Deci, 2000a).
Some researchsuggests that an individual’s personal interest inthe
domain may also moderate the effect ofrationale on motivation and
performance (Durik& Harackiewicz, 2007; Shechter, Durik,
Miy-amoto, & Harackiewicz, 2011) whereas morerecent work has
suggested that such moderationmay be an artifact of the
relationship betweeninterest and perceived competence (Durik et
al.,2015). Though this research is important forunderstanding the
effects of rationale, the roleof personal interest could not be
tested in thepresent analysis due to lack of reporting
orvariability in individual’s level of personal in-terest. Rather,
the present analysis examinesinterest as a feature of the task.
Exploratory and Methodological Factors
Given heterogeneity in research findings, wethought it might
also be useful to explorewhether the effects of rationale provision
varyby type of task, gender, or outcome measure-ment, although the
theoretical reason to expectsuch variation is more limited. First,
we exam-ined if the effects of rationales depend on thetype of task
for which the rationale is provided.Studies included in the
analysis included ratio-nales for school tasks (Shin, 2010) as well
aslearning tasks outside of the classroom (Jang,2008; Reeve et al.,
2002), health tasks (Bannon& Schwartz, 2006; Vansteenkiste,
Simons,Soenens, & Lens, 2004; Williams et al., 2001),and work
tasks (Grant, 2008b, 2012; Grant etal., 2007). We therefore sought
to determine ifthe effects of rationale for academic tasks
differfrom effects for work or health-related tasks.
Next, we tested whether or not the effectsof rationale depend on
the proportion of thesample that is female. Research suggests
thatwomen report a higher need for affiliation(Hill, 2009) and a
higher priority on commu-nal goals (Diekman & Eagly, 2013)
comparedwith men. Therefore, given the social nature
24 STEINGUT, PATALL, AND TRIMBLE
Thi
sdo
cum
ent
isco
pyri
ghte
dby
the
Am
eric
anPs
ycho
logi
cal
Ass
ocia
tion
oron
eof
itsal
lied
publ
ishe
rs.
Thi
sar
ticle
isin
tend
edso
lely
for
the
pers
onal
use
ofth
ein
divi
dual
user
and
isno
tto
bedi
ssem
inat
edbr
oadl
y.
-
of rationale, women may be more likely toendorse rationales
provided, and in turn ex-perience motivational benefits than men.
De-spite this reasoning, few studies have exam-ined variation in
the effects of rationale bysex. One study tested this question and
foundthat a prosocial rationale led to a significantlyhigher rate
of work among boys who receivedthe rationale compared to the
control group,but not for girls (Kuczynski, 1983).
As for methodological characteristics, the ef-fects of rationale
on motivation and perfor-mance may also depend on the type of
measureused to assess the outcomes. In a classic meta-analysis of
the effects of rewards on intrinsicmotivation, Deci et al. (1999)
found that type ofmeasure was a significant moderator, with
thenegative effects of reward being stronger forbehavioral compared
to self-report measures ofintrinsic motivation. Given the finding
of thisrelated meta-analysis, it may be important toexplore if
outcome measurement moderates theobserved effects of
rationales.
Need for a Synthesis on the Effectsof Rationale
Rationales are used in a variety of settings tomotivate others.
Indeed, a substantial body ofresearch has accumulated testing the
effects ofrationale on motivation and performance out-comes,
although there has not yet been an at-tempt to synthesize this
evidence. Further, ex-isting research suggests that the effects
ofrationale on motivation and performance out-comes may depend on
characteristics of therationale, task, sample, and outcome,
thoughconclusions about the importance of variousfactors is
difficult to decipher without synthe-sizing the existing evidence.
Thus, given theimportance of rationale in theories of motiva-tion
and applied settings, a meta-analysis of theeffects of rationale on
motivation and perfor-mance outcomes was needed.
The purposes of this meta-analysis were two-fold. One purpose
was to simply answer themost basic question of the extent to which
ra-tionales produce a positive effect on motivationand performance
outcomes. Perhaps more im-portantly, the second purpose was to
examinefactors that may influence the effects of ratio-nale. A
number of predictions regarding mod-erators have emerged based on
existing theory
and research on rationale provision, though ev-idence has often
been mixed across studies orlimited within a single study to draw
firm con-clusions regarding the role of most of thesefactors. The
present meta-analysis provides anopportunity to use variation
between studies, inaddition to that within studies, to examine
therole of characteristics of the rationale, task,sample, or
outcome as moderators of rationaleeffects. Resolving the conflict
between compet-ing hypotheses of SDT and EVT related todifferential
effects of various types of rationalesis a particular aim of this
meta-analysis.
To summarize our hypotheses given the the-ory and empirical
evidence previously re-viewed, we predicted the following:
1. Rationale provision will lead to an overallpositive effect on
all adaptive outcomessuch as subjective task value,
autonomousmotivation, engagement, performance,perceived autonomy,
perceived compe-tence, and perceived relatedness and neg-ative
effects on maladaptive outcomessuch as controlled motivation.
2. Prosocial and autonomous rationales willlead to larger
positive effects than control-ling rationales, with prosocial
rationaleshaving the largest effects.
3. Rationales will lead to larger effects whencombined with
other autonomy-support-ive practices, including acknowledgmentof
negative affect or noncontrolling lan-guage compared to rationale
alone. Fur-ther, the effect of all three practices com-bined will
be greater than the effect twopractices or rationale alone.
4. Rationales will lead to larger effects whenthe task is
uninteresting.
We were less certain of predictions regard-ing type of task, sex
of participants and out-come measurement. Nonetheless, we
sus-pected that:
5. Rationales will lead to larger effects foracademic tasks than
for health or worktasks.
6. Rationales will lead to larger effectsamong samples with a
higher proportionof females compared to samples with asmaller
proportion of females.
25EFFECTS OF RATIONALE
Thi
sdo
cum
ent
isco
pyri
ghte
dby
the
Am
eric
anPs
ycho
logi
cal
Ass
ocia
tion
oron
eof
itsal
lied
publ
ishe
rs.
Thi
sar
ticle
isin
tend
edso
lely
for
the
pers
onal
use
ofth
ein
divi
dual
user
and
isno
tto
bedi
ssem
inat
edbr
oadl
y.
-
7. Rationale will lead to larger effects onbehavioral measures
of outcomes com-pared to self-report measures.
Method
Literature Search Procedures
Four complementary search strategies wereused in order to
retrieve as many published andunpublished tests of the effect of
rationale aspossible: (a) systematic electronic databasesearches,
(b) ancestry and descendant searches,(c) contact with researchers
through profes-sional networks and organizations, and (d) con-tact
with prolific researchers. The first strategyinvolved searches of
the ERIC, PsycINFO, Ac-ademic Search Complete, and ProQuest
Disser-tation & Theses, for documents catalogued be-fore
January 1, 2014. The following terms wereentered simultaneously
linked with “OR”s: ra-tionale, instrumental, instrumentality,
rele-vance, meaningful, meaningfulness, explana-tion, explanatory,
autonomy-support, purpose,goal content, motivation, engagement,
effort,persistence, performance, achievement, andlearning. The
search of the two main databases(ERIC & PsycINFO) returned
approximately7,730 results.
Next, reference sections of relevant docu-ments were examined to
determine whether anycited works had titles that also might be
relevantto the topic. We also used Web of Science(Reuter’s Web of
Knowledge) to identify rele-vant reports that had cited 7 seminal
papers onthe effects of rationales (e.g., Deci et al., 1994;Durik
& Harackiewicz, 2007; Jang, 2008;Reeve et al., 2002;
Vansteenkiste et al., 2005;Vansteenkiste, Simons, Soenens, &
Lens,2004).
Finally, members of several professional or-ganizations in
fields related to the topic, includ-ing Educational Psychology
(Division 15 of theAmerican Psychological Association; DivisionC of
the American Educational Research Asso-ciation [AERA]; Motivation
in Education Spe-cial Interest Group [associated with AERA])and
Social Psychology (Society for Personalityand Social Psychology),
were contacted andasked to provide information on relevant
re-search. Likewise, researchers who our searchindicated had
conducted more than two studies
examining the effects of rationale were askedfor other relevant
research.
Criteria for Including Studies
Each study meeting the following six criteriawere included in
the analysis. First, studies wererequired to have used an
experimental design inwhich the provision of rationale was
manipu-lated and participants were randomly assignedto conditions,
or cluster-randomized experi-ments in which clusters were randomly
as-signed to condition (Bannon & Schwartz, 2006;Gillison et
al., 2013; Grant, 2012; Grant &Hofmann, 2011; Vansteenkiste,
Simons,Soenens, & Lens, 2004). Second, studies wererequired to
have included at least one conditionthat received a rationale and a
control conditionthat did not receive a rationale.
Both gain-framed and loss-framed messageswere included in the
analysis (Bannon &Schwartz, 2006; Williams et al., 2001).
How-ever, the literature examining these two types ofmessages have
predominantly compared gain-framed to loss-framed messages and
omitted acontrol group. Therefore, most were excludeddue to a lack
of control group. Akl et al. (2011)synthesized the effects of gain
versus lossframed messages on behavior, but limited theiranalysis
to effects on health outcomes. In addi-tion to gain- and
loss-framed messages, otherstudies that compared different types of
ratio-nales and omitted a true control group were alsoexcluded
(Benware & Deci, 1984; Sheldon,Ryan, Deci, & Kasser, 2004;
Simons, Dewitte,& Lens, 2003; Vansteenkiste, Simons,
Lens,Sheldon, & Deci, 2004; Vansteenkiste, Simons,Lens,
Soenens, et al., 2004; Wang, Hu, & Guo,2013).
Third, given the definition of rationale provi-sion within the
SDT framework as an external orcontextual factor in which another
individual pro-vides a reason for putting forth effort, only
studiesthat operationalized rationale provision as an ex-ternal
explanation of the value of an activity wereincluded. Manipulations
of utility value thatprompted participants to self-generate an
explana-tion for the benefits of the task or activity wereexcluded
(Hulleman, Godes, Hendricks, & Har-ackiewicz, 2010). Although
evidence suggeststhat these interventions indeed influence
subjec-tive task value and other adaptive outcomes,
theseinterventions do not involve the external provision
26 STEINGUT, PATALL, AND TRIMBLE
Thi
sdo
cum
ent
isco
pyri
ghte
dby
the
Am
eric
anPs
ycho
logi
cal
Ass
ocia
tion
oron
eof
itsal
lied
publ
ishe
rs.
Thi
sar
ticle
isin
tend
edso
lely
for
the
pers
onal
use
ofth
ein
divi
dual
user
and
isno
tto
bedi
ssem
inat
edbr
oadl
y.
-
of a rationale and thus, may function differently(Canning &
Harackiewicz, 2015). Although ex-ternal provision of rationales can
provide ideasabout the value that is “otherwise hidden” (Jang,2008,
p. 798) to individuals, such rationales de-pend on the individual
endorsing this externallyexpressed value as something important to
thempersonally. In contrast, self-generated utility
valueinterventions depend on existing knowledge aboutthe value of a
task and capitalizes on that knowl-edge by encouraging individuals
to adopt or en-dorse the value.
Fourth and similarly, only studies in which therationale
expressed value of the task that partici-pants were to take part in
were included. Studiesin which the rationale was not relevant to
the taskor which explained why the task was not relevant(Roser,
1990) were excluded. Fifth, included stud-ies needed to have
measured the effect of rationaleon at least one of the outcomes of
interest, namelysubjective task value, autonomous motivation,
en-gagement, performance, perceived autonomy,competence,
relatedness, or controlled motivation.However, these outcome
categories each includeda set of related constructs. The next
section defineseach outcome category and provides examples
ofincluded constructs. Sixth and finally, in order tobe included,
each study needed to contain suffi-cient information to estimate
the effect of rationaleon at least one outcome of interest.
Dependent Measures
This synthesis assessed the effect of rationaleon eight related
outcomes: value, autonomousmotivation, engagement, and performance,
per-ceived autonomy, competence, or relatedness,and controlled
motivation. Studies varied inhow these constructs were
operationalized andmost outcomes had conceptual
subcategories.Subjective task value included importance(Reeve et
al., 2002), usefulness or utility value(Deci et al., 1994; Shechter
et al., 2011). Com-petence valuation, the perceived importance
oftask success, was also considered a form ofvalue due to the
academic nature of the task inquestion, there was little
distinction between theimportance of task success and the
importanceof the task (Durik & Harackiewicz, 2007;Shechter et
al., 2011).
Autonomous motivation included forms of mo-tivation that are
generated by the self or are en-dorsed by the self. Intrinsic value
and interest
value were categorized as autonomous motiva-tion, as they are
conceptually more related tointrinsic motivation than to other
forms of value.Other autonomous motivation outcomes
includedinterest or enjoyment (Durik & Harackiewicz,2007;
Gillison et al., 2013; Shechter et al., 2011),intrinsic regulation,
and identified regulation(Vansteenkiste, Simons, Soenens, &
Lens, 2004).
Engagement was conceptualized as a multi-dimensional construct
with three components:behavioral, cognitive, and emotional
(Fredricks,Blumenfeld, & Paris, 2004). Although thesecomponents
are useful in explicating the varietyof manifestations of
engagement, the distinc-tions among them are not always clear.
Thepresent analysis included behavioral engage-ment outcomes,
including effort, free choicepersistence (Deci et al., 1994), time
on task(Grant et al., 2007), and behavioral engagement(Jang, 2008),
as well as cognitive engagement,such as interest-enhancing
strategies (Jang,2008; Sansone et al., 1999).
Performance outcomes included learning,which were commonly
measured with an experi-menter-created test (e.g., Durik &
Harackiewicz,2007; Vansteenkiste et al., 2005). In other
studies,performance was a measure or observation of thebehavior of
interest (e.g., Bannon & Schwartz,2006).
Few studies measured the effects of ratio-nale on the three
needs for autonomy, compe-tence, or relatedness. All studies that
mea-sured effects on needs referred to the needconstruct exactly,
with a few exceptions: au-tonomy included perceived
self-determina-tion (Reeve et al., 2002), competenceincluded
self-efficacy (Hall, Bishop, & Mar-teau, 2003), and relatedness
included affec-tive commitment to beneficiaries (Grant et
al.,2007). Finally, controlled motivation in-cluded introjected and
external regulation(Vansteenkiste, Simons, Soenens, &
Lens,2004).
Information Retrieved From Studies andCoder Reliability
Numerous characteristics of each study wereincluded in the
database. These characteristicsencompassed six categories: (a)
research report,(b) rationale manipulation, (c) task, (d)
sample,(e) outcome measure, and (f) estimate of effectsize. Table 2
lists the characteristics retrieved
27EFFECTS OF RATIONALE
Thi
sdo
cum
ent
isco
pyri
ghte
dby
the
Am
eric
anPs
ycho
logi
cal
Ass
ocia
tion
oron
eof
itsal
lied
publ
ishe
rs.
Thi
sar
ticle
isin
tend
edso
lely
for
the
pers
onal
use
ofth
ein
divi
dual
user
and
isno
tto
bedi
ssem
inat
edbr
oadl
y.
-
from each study. Lack of reporting was com-mon among the
studies, especially with respectto characteristics of the sample,
such as thegender of participants.
Every report was independently coded by thefirst and third
authors. Of 12,159 codes, onlyapproximately 2.4% were discrepant.
Discrep-ancies were noted, discussed, and, if agreementcould not be
reached, the second author re-solved the disagreement.
Effect Size Estimation
We used the d-index (Cohen, 1988), a stan-dardized mean
difference, as our index of effect.For the present analysis, we
computed the d-in-dex for each effect by subtracting the
controlmean from the experimental mean and dividingby the pooled
standard deviation. As such, pos-itive effects indicate that
participants in the ra-tionale condition had higher outcome
scoresthan those in the control condition. Whenmeans, standard
deviations, and sample sizes
were unavailable, an inference test or propor-tion was used to
calculate the d-index. Whenmeans and standard deviations were
reportedand the overall, but not group, sample size wasreported, we
assumed equal sample sizes amongthe groups. When effects could not
be calcu-lated from information provided in the report,we contacted
the author(s) for the required in-formation.
Methods of Data Integration
For each outcome, the set of effects wasexamined for values
greater than 3 standarddeviations from the mean, which were
Win-sorized and replaced with the effect size valuethree standard
deviations in the same direc-tion from the mean. Also, although
both pub-lished and unpublished studies were includedin the
analysis, we took additional steps toensure that publication bias
did not undulyinfluence analysis results. To test for publica-tion
bias, two complementary methods were
Table 2Complete List of Information Retrieved From Studies
Report characteristics1. Author2. Year3. Type of research report
(journal article, dissertation, unpublished data)
Sample1. Country2. Setting (laboratory, school, health care
center, work)3. Developmental stage of the sample (elementary,
middle, high, college, adults)4. Socioeconomic status (low, middle,
high, mixed)5. Age6. Grade level, when appropriate7. % Female8.
Ethnicity (Caucasian, African-American, Asian, Native American,
Hispanic)
Rationale manipulation1. Type of task (academic, work, health,
environmental, other)2. Task reported as interesting (yes/no)3.
Type of rationale (autonomous, controlling, prosocial)4. Language
(noncontrolling, controlling)5. Acknowledgment of negative affect
(yes/no)
Outcome measure1. Outcome category (subjective task value,
autonomous motivation, engagement,
performance, autonomy, competence, relatedness, controlled
motivation) anddescription
2. Type of outcome measure (self-report, behavioral)3. Time
point of outcome measurement (delay from the provision of
rationale)
Estimate of the effect1. Direction of the effect2. Magnitude of
the effect3. Sample size
28 STEINGUT, PATALL, AND TRIMBLE
Thi
sdo
cum
ent
isco
pyri
ghte
dby
the
Am
eric
anPs
ycho
logi
cal
Ass
ocia
tion
oron
eof
itsal
lied
publ
ishe
rs.
Thi
sar
ticle
isin
tend
edso
lely
for
the
pers
onal
use
ofth
ein
divi
dual
user
and
isno
tto
bedi
ssem
inat
edbr
oadl
y.
-
used. First, the trim and fill procedure sug-gested by Duval and
Tweedie (2000) wasused on the set of effects for each
outcomecategory before Winsorization. This tech-nique tests whether
the distribution of ob-served effect sizes in an analysis
represents asymmetrical distribution. If it does not, thetechnique
imputes missing values to create anormal distribution and adjusts
the effect es-timate given the imputed studies. In additionto trim
and fill, publication status was testedas a moderator of the effect
of rationale oneach outcome with more than 10 effect sizes.This
technique tests whether the effects aresignificantly smaller for
unpublished studies,which would indicate publication bias.
Calculating average effect sizes. We cal-culated the average
effect of rationale on eachoutcome by weighting each effect size by
theinverse variance of the effect (Hedges & Olkin,1985). In
addition to the average effect, 95%confidence intervals around the
average effectsize were calculated. The null hypothesis thatthe
effect was equal to zero was rejected if the95% confidence interval
did not contain zero.We determined whether the effect sizes for
agiven outcome category varied significantly us-ing a within-class
goodness of fit statistic (Qw)which follows a chi-square
distribution having(k-1) degrees of freedom where k is the numberof
effect sizes in the analysis.
Identifying independent hypothesis tests.Many studies include
multiple estimates of ef-fects of rationale on different outcomes
that aregrouped in the same outcome category. Thishappens for a
number of reasons. For example,multiple experimental groups might
have beencompared to a control group, groups might havebeen
compared on several measures of the sameconstruct, or the same
outcome might have beenmeasured at multiple time points. For
instance,de Young et al. (1993) compared three experi-mental groups
with a common control group.To maintain the assumption that effect
sizes areindependent, effects coming from the samesample within
each outcome category were av-eraged prior to calculation of the
overall effectof rationale on the outcome category. However,when
testing moderators, this approach limitsvariability and obscures
true differences in ef-fects. Thus, for moderator analyses, we used
ashifting unit of analysis approach (Cooper,2010) and allowed a
single sample to contribute
one effect to each level of the moderator forwhich an effect was
available. Effects withinthe same level of a moderator were
averaged sothat each sample only contributed one effectwithin a
level of a moderator.
Tests for moderators of effects. We usedhomogeneity analyses to
assess moderator ef-fects (Cooper, Hedges, & Valentine,
2009).Specifically, a between class goodness of fitstatistic (Qb)
was used to test whether the dif-ference between groups varied more
than wouldbe expected due to sampling error alone. Thebetween class
statistic follows a chi-square dis-tribution with p � 1 degrees of
freedom, wherep is equal to the number of groups in the mod-erator
analysis. A significant between-classgoodness of fit test indicates
significant varia-tion between the groups of effects for each
levelof the moderator. Moderator tests were runwhen at least two
independent samples wereavailable for at least two levels of the
moderatorvariable and levels that had only one effect weredropped
before analysis. Because each moder-ator was tested individually,
it is possible thatmoderators were confounded. Confoundedmoderators
might lead to interpreting a moder-ator as significant only because
it co-occurswith another moderator in the sample of effects(see
Cooper, 2010 for a review). To explore theextent to which
confounding among moderatorsmight have unduly impacted results or
interpre-tation, we conducted a series of chi-square teststo
examine the relations among moderator vari-ables.
Fixed and random error models. The cal-culation of overall
average effects and the test-ing of moderator effects both rely on
severalstatistical assumptions. Two models with differ-ing
assumptions are often used: fixed-error andrandom-error models. The
fixed-error model as-sumes that the variance in effect sizes is
dueonly to variance in sampling of participants. Incontrast, the
random-error model assumes thatthe effect sizes are affected by
other factors ofstudies, which are assumed to vary
randomly.Therefore, the random-error model is more con-servative
than the fixed-error model.
Because it is impossible to determine whichset of assumptions
reflects the true reasons forvariation in the distribution of
observed effects,analyses were conducted using both fixed
andrandom-error models in order to examine theimpact of each set of
assumptions on the results.
29EFFECTS OF RATIONALE
Thi
sdo
cum
ent
isco
pyri
ghte
dby
the
Am
eric
anPs
ycho
logi
cal
Ass
ocia
tion
oron
eof
itsal
lied
publ
ishe
rs.
Thi
sar
ticle
isin
tend
edso
lely
for
the
pers
onal
use
ofth
ein
divi
dual
user
and
isno
tto
bedi
ssem
inat
edbr
oadl
y.
-
Following Greenhouse and Iyengar (1994), weconducted all
analyses using both fixed andrandom error models in order to
provide a sen-sitivity analysis. This analytic choice
enablesinferences regarding the impact of different as-sumptions on
analysis results. Indeed, whilesome meta-analysts use the empirical
results ofheterogeneity tests to determine the appropriateerror
model, meta-analytic experts generallyagree that theory should be
used to guide thechoice of error model and either error modelmay be
appropriate for both main and modera-tor analyses and that the
findings for the overalltest of heterogeneity should not
determinechoice of model (Borenstein, Hedges, Higgins,&
Rothstein, 2010; Hedges & Vevea, 1998).
Results
The literature search uncovered 23 reportsthat tested the effect
of rationale on at least oneof the following outcomes: value,
autonomousmotivation, engagement, performance, per-ceived autonomy,
competence, relatedness, andcontrolled motivation between 1983 and
2013.From these reports, 37 independent samplescontributed 289
effects, with sample sizes rang-ing from 18 to 265. Characteristics
of includedstudies are listed in Table 3. Using Grubbs(1950) test,
one outlier was found among thesample sizes (western sample in
Study 1 ofShechter et al. (2011) and Winsorized to itsnearest
neighbor.
Overall Effects of Rationale Provision
Average effects of rationale on each outcomeare listed in Table
4. Results showed a signifi-cant average effect of rationale on
five of theeight outcomes under both fixed-effect (FE)
andrandom-effects (RE) models: subjective taskvalue, engagement,
performance, perceivedcompetence, and perceived autonomy.
Averageeffect sizes ranged from a small effect on per-formance (FE:
d � .16, p � .001; RE: d � .19,p � .001) to medium effects on
perceived au-tonomy (FE: d � .40, p � .001; RE: d � .38,p � .01)
and subjective task value (FE: d � .33,p � .001; RE: d � .34, p �
.001). In addition,and contrary to expectations, the average
effectof rationale on perceived competence was neg-ative under both
fixed and random-error as-sumptions (FE: d � �.19, p � .05; RE: d
�
�.19, p � .05). Effects of rationale on autono-mous motivation,
perceived relatedness, andcontrolled motivation were not
significant.There was significant heterogeneity in the ef-fects of
rationale on subjective task value, butnot on any other outcome.
However, given ourtheoretical reasons for exploring most
modera-tors and the low power of heterogeneity tests,we conducted
moderator analyses for outcomeswith more than 10 independent
effects includ-ing subjective task value, autonomous motiva-tion,
engagement, and performance.
Trim and fill analyses were conducted andresults are listed in
Table 5. Results suggest thatthe true average effects of rationale
on perfor-mance and engagement may be smaller than theobserved set
of effect sizes suggest, with 7missing effects being imputed on the
left side ofthe distribution for both outcomes. After includ-ing
imputed effect sizes, the effect of rationaleon engagement was no
longer significant undereither model while the effect on
performanceremained significant under the fixed effectsmodel but
not the random effects model. How-ever, publication status did not
significantlymoderate effects on any of the four outcomesfor which
moderators were tested, suggestingthat the publication status of
included studiesdid not significantly explain variability in
effectsizes (see Tables 6–9). Even so, results of thetrim and fill
analyses suggest that findings forengagement and performance
outcomes shouldbe interpreted with caution.
Moderator Analyses
Several variables that we intended to test asmoderators could
not be tested because of lackof variability and/or reporting in the
primaryliterature (developmental level, ethnicity, out-come measure
for all outcomes except engage-ment).
Subjective task value. Moderation by sixfactors was tested for
the effects of rationale onsubjective task value: rationale type,
autonomysupportive practices, interest level of the task,type of
task, proportion of the sample that wasfemale, and publication
status (see Table 6).The interestingness of the task and
percentagefemale of the sample each moderated the effectof
rationales on value under fixed effects but notrandom effects
assumptions. Results suggestedthat the effect of rationale on
subjective task
30 STEINGUT, PATALL, AND TRIMBLE
Thi
sdo
cum
ent
isco
pyri
ghte
dby
the
Am
eric
anPs
ycho
logi
cal
Ass
ocia
tion
oron
eof
itsal
lied
publ
ishe
rs.
Thi
sar
ticle
isin
tend
edso
lely
for
the
pers
onal
use
ofth
ein
divi
dual
user
and
isno
tto
bedi
ssem
inat
edbr
oadl
y.
-
Tab
le3
Cha
ract
eris
tics
ofE
xper
imen
tal
Stud
ies
Incl
uded
inth
eM
eta-
Ana
lysi
s
Aut
hor
(yea
r)R
epor
tty
peSu
bsam
ple
(n)
Rat
iona
lety
pe
Ack
now
ledg
ene
gativ
eaf
fect
/no
ncon
trol
ling
lang
uage
Uni
nter
estin
gta
skPr
opor
tion
fem
ale
Tim
epo
int
Out
com
eca
tego
ry(b
ehav
iora
lm
easu
re)
Spec
ific
outc
ome
dva
r(d)
Ban
non
and
Schw
artz
(200
6)J
32A
NC
L/N
LP
Snac
kch
oice
.7.1
536
AN
CL
/YL
PSn
ack
choi
ce.7
.13
Beh
rens
(199
9)M
71A
./NH
AM
Snac
kch
oice
.29
.06
./NP
Cog
nitiv
eL
�.1
9.0
6./N
VPe
rtin
ence
.82
.06
./NV
Rel
evan
ce.4
3.0
662
A./N
HA
M.2
4.0
7./N
PC
ogni
tive
L.0
2.0
6./N
VPe
rtin
ence
1.0
7./N
VR
elev
ance
.58
.07
66C
CL
/NH
AM
.18
.06
/NP
Cog
nitiv
eL
.45
.06
/NV
Pert
inen
ce.4
7.0
6/N
VR
elev
ance
.47
.06
Beh
rens
(199
9),
Pilo
t56
A./N
.V
Pert
inen
ce1.
05.0
8./N
VR
elev
ance
.69
.08
A./N
VPe
rtin
ence
.83
.08
./NV
Rel
evan
ce.8
4.0
8C
CL
/N.
VPe
rtin
ence
.23
.07
/NV
Rel
evan
ce.1
9.0
7de
You
nget
al.
(199
3)J
46.
./NH
P.5
.09
48.
./NP
1.27
.153
../N
P.4
1.0
8D
eci
etal
.(1
994)
J32
.N
CL
/NU
LA
MIn
tere
st.7
5.1
3N
CL
/NU
E(B
)Fr
eech
oice
.7.1
3N
CL
/NU
VU
tility
.87
.14
32.
CL
/YU
LA
MIn
tere
st�
.01
.13
CL
/YU
E(B
)Fr
eech
oice
.29
.13
CL
/YU
VU
tility
.4.1
364
.N
CL
/YU
LA
MIn
tere
st�
.26
.08
NC
L/Y
UE
(B)
Free
choi
ce.0
3.0
8N
CL
/YU
VU
tility
.81
.09
(tab
leco
ntin
ues)
31EFFECTS OF RATIONALE
Thi
sdo
cum
ent
isco
pyri
ghte
dby
the
Am
eric
anPs
ycho
logi
cal
Ass
ocia
tion
oron
eof
itsal
lied
publ
ishe
rs.
Thi
sar
ticle
isin
tend
edso
lely
for
the
pers
onal
use
ofth
ein
divi
dual
user
and
isno
tto
bedi
ssem
inat
edbr
oadl
y.
-
Tab
le3
(con
tinu
ed)
Aut
hor
(yea
r)R
epor
tty
peSu
bsam
ple
(n)
Rat
iona
lety
pe
Ack
now
ledg
ene
gativ
eaf
fect
/no
ncon
trol
ling
lang
uage
Uni
nter
estin
gta
skPr
opor
tion
fem
ale
Tim
epo
int
Out
com
eca
tego
ry(b
ehav
iora
lm
easu
re)
Spec
ific
outc
ome
dva
r(d)
64.
CL
/NU
LA
MIn
tere
st.5
6.0
9C
L/N
UE
(B)
Free
choi
ce�
.33
.08
CL
/NU
VU
tility
1.05
.09
Dur
ikan
dH
arac
kiew
icz
(200
7)J
145
A./N
LA
MIn
tere
st.3
4.0
3./N
E(S
R)
Invo
lvem
ent
.07
.03
./NP
0.0
3./N
VC
omp.
val
.55
.03
./NV
Util
ity.4
6.0
3./N
PC�
.14
.03
Gill
ison
etal
.(2
013)
J19
4C
NC
L/Y
LA
MIn
tere
st�
.05
.02
NC
L/Y
AM
.28
.02
NC
L/Y
AM
Intr
insi
cgo
al.1
1.0
2N
CL
/NE
(SR
)E
ffor
t�
.33
.02
NC
L/Y
V.0
2.0
224
0C
CL
/NL
AM
.13
.02
CL
/NA
MIn
tere
st.2
7.0
2C
L/N
AM
.51
.02
CL
/NE
(SR
)In
trin
sic
goal
.12
.02
CL
/NV
.38
.02
200
AN
CL
/YL
AM
Inte
rest
�.0
4.0
2N
CL
/YA
M.1
.02
NC
L/Y
AM
Intr
insi
cgo
al.2
7.0
2N
CL
/YE
(SR
)E
ffor
t�
.25
.02
NC
L/Y
V.0
3.0
221
0A
CL
/NL
AM
Inte
rest
�.1
2.0
2C
L/N
AM
�.0
8.0
2C
L/N
AM
Intr
insi
cgo
al�
.14
.02
CL
/NE
(SR
)E
ffor
t�
.21
.02
CL
/NV
�.0
3.0
2G
rant
etal
.(2
007)
,1
J22
P./.
LE
(B)
Tim
e�
.53
.19
./.P
�.0
4.1
827
P./.
LE
(B)
Tim
e.7
.17
./.P
.9.1
7G
rant
etal
.(2
007)
,2
J30
P./N
LE
(B)
Free
choi
ce.8
.14
./NV
Sign
ifica
nce
.91
.15
32 STEINGUT, PATALL, AND TRIMBLE
Thi
sdo
cum
ent
isco
pyri
ghte
dby
the
Am
eric
anPs
ycho
logi
cal
Ass
ocia
tion
oron
eof
itsal
lied
publ
ishe
rs.
Thi
sar
ticle
isin
tend
edso
lely
for
the
pers
onal
use
ofth
ein
divi
dual
user
and
isno
tto
bedi
ssem
inat
edbr
oadl
y.
-
Tab
le3
(con
tinu
ed)
Aut
hor
(yea
r)R
epor
tty
peSu
bsam
ple
(n)
Rat
iona
lety
pe
Ack
now
ledg
ene
gativ
eaf
fect
/no
ncon
trol
ling
lang
uage
Uni
nter
estin
gta
skPr
opor
tion
fem
ale
Tim
epo
int
Out
com
eca
tego
ry(b
ehav
iora
lm
easu
re)
Spec
ific
outc
ome
dva
r(d)
Gra
ntet
al.
(200
7),
3J
64P
./NH
E(B
)Pe
rsis
tenc
e�
.17
.06
./NFR
Aff
ectiv
e.5
4.0
358
P./N
HE
(B)
Pers
iste
nce
.52
.07
./NFR
Aff
ectiv
e.5
4.0
3G
rant
etal
.(2
008)
,1
J21
AC
./NL
P#
Don
atio
ns.3
2.1
9./N
PD
olla
rsra
ised
.32
.19
23P
./NL
P#
Don
atio
ns1.
41.2
2./N
PD
olla
rsra
ised
1.19
.2G
rant
etal
.(2
007)
,3
J34
CP
./NH
P#
Don
atio
ns1.
4.1
5G
rant
(201
2)a
J38
CP
./NH
PD
olla
rsra
ised
.71
.13
./NP
Rev
enue
/Shi
ft�
.09
.12
./NP
Sale
s.6
.13
./NP
.02
.12
41C
P./N
HP
Rev
enue
1.5
.13
./NP
Rev
enue
/Shi
ft�
.27
.11
./NP
Sale
s1.
22.1
2./N
PSa
les/
Shif
t�
.13
.11
44C
P./N
HP
Rev
enue
3.35
.22
./NP
Rev
enue
/Shi
ft.4
4.1
./NP
Sale
s3.
42.2
3./N
PSa
les/
Shif
t.6
4.1
Gra
ntan
dH
ofm
ann
(201
1)J
43P
./N.
PSo
apus
age
.64
.144
A./N
.P
Soap
usag
e�
.27
.09
Hal
let
al.
(200
3)J
122
A./Y
HPC
Self
-effi
cacy
�.4
6.0
311
9A
./YH
PCSe
lf-e
ffica
cy�
.06
.03
Har
acki
ewic
zet
al.
(201
2)J
181
A./N
LP
STE
Mco
urse
s.3
1.0
2Ja
ng(2
008)
J13
6A
NC
L/Y
UH
1HE
(B)
.43
.03
NC
L/Y
UA
UT
.55
.03
NC
L/Y
U2H
E(B
).6
4.0
3N
CL
/YU
E(S
R)
Stra
tegi
es.5
6.0
3N
CL
/YU
E(S
R)
.42
.03
NC
L/Y
UP
Con
cept
ual
L.4
6.0
3N
CL
/YU
PR
ote
L.3
.03
NC
L/Y
UV
Impo
rtan
ce.7
.03
John
son
(200
4)D
265
A./N
.V
Rel
evan
ce.4
6.0
2(t
able
cont
inue
s)
33EFFECTS OF RATIONALE
Thi
sdo
cum
ent
isco
pyri
ghte
dby
the
Am
eric
anPs
ycho
logi
cal
Ass
ocia
tion
oron
eof
itsal
lied
publ
ishe
rs.
Thi
sar
ticle
isin
tend
edso
lely
for
the
pers
onal
use
ofth
ein
divi
dual
user
and
isno
tto
bedi
ssem
inat
edbr
oadl
y.
-
Tab
le3
(con
tinu
ed)
Aut
hor
(yea
r)R
epor
tty
peSu
bsam
ple
(n)
Rat
iona
lety
pe
Ack
now
ledg
ene
gativ
eaf
fect
/no
ncon
trol
ling
lang
uage
Uni
nter
estin
gta
skPr
opor
tion
fem
ale
Tim
epo
int
Out
com
eca
tego
ry(b
ehav
iora
lm
easu
re)
Spec
ific
outc
ome
dva
r(d)
Kuc
zyns
ki(1
983)
J18
(Fem
ale)
P./N
UH
PR
ate
ofw
ork
.47
.23
18(M
ale)
./NU
LP
Rat
eof
wor
k2.
09.3
4M
cCal
l(2
004)
J36
AC
./NL
1MP
.61
.12
./N2M
P.3
9.1
1./N
3MP
.32
.11
./NV
Ben
efits
.17
.11
33A
C./N
L1M
P.3
8.1
3./N
2MP
.47
.13
./N3M
P.1
2.1
3./N
VB
enefi
ts.1
1.1
3R
eeve
etal
.(2
002)
,1
J70
CC
L/N
UH
AU
TSe
lf-d
eter
m.
�.1
6.0
6C
L/N
UE
(B;S
R)
Eff
ort
.39
.06
CL
/NU
VIm
port
ance
�.0
7.0
670
AP
NC
L/Y
UH
AU
TSe
lf-d
eter
m.
.62
.06
NC
L/Y
UE
(B;S
R)
Eff
ort
.59
.06
NC
L/Y
UV
Impo
rtan
ce.5
2.0
670
CP
CL
/NU
HA
UT
Self
-det
erm
.�
.03
.06
CL
/NU
E(B
;SR
)E
ffor
t.4
9.0
6C
L/N
UV
Impo
rtan
ce�
.16
.06
Ree
veet
al.
(200
2),
270
AP
NC
L/Y
UH
AU
TSe
lf-d
eter
m.
.63
.06
NC
L/Y
UE
(B;S
R)
Eff
ort
.38
.06
NC
L/Y
UV
Impo
rtan
ce.7
.06
Sans
one
etal
.(1
999)
J11
1P
./YU
LE
(B)
Stra
tegy
use
.34
.04
Shec
hter
etal
.(2
011)
,1
J72
(Eas
tern
)A
./NH
AM
Invo
lvem
ent
.52
.06
./NE
(B)
Free
choi
ce.2
4.0
6./N
P.0
9.0
6./N
VU
tility
.21
.06
210
(Wes
tern
)A
./NL
AM
Invo
lvem
ent
.02
.02
./NE
(B)
Free
choi
ce.0
2.0
2./N
P�
.22
.02
./NV
Util
ity.0
8.0
2Sh
echt
eret
al.
(201
1),
252
(Eas
tern
)A
./NL
AM
Invo
lvem
ent
.38
.08
./NPC
�.2
5.0
8./N
E(S
R)
Eff
ort
.5.0
8./N
E(S
R)
Invo
lvem
ent
.16
.08
34 STEINGUT, PATALL, AND TRIMBLE
Thi
sdo
cum
ent
isco
pyri
ghte
dby
the
Am
eric
anPs
ycho
logi
cal
Ass
ocia
tion
oron
eof
itsal
lied
publ
ishe
rs.
Thi
sar
ticle
isin
tend
edso
lely
for
the
pers
onal
use
ofth
ein
divi
dual
user
and
isno
tto
bedi
ssem
inat
edbr
oadl
y.
-
Tab
le3
(con
tinu
ed)
Aut
hor
(yea
r)R
epor
tty
peSu
bsam
ple
(n)
Rat
iona
lety
pe
Ack
now
ledg
ene
gativ
eaf
fect
/no
ncon
trol
ling
lang
uage
Uni
nter
estin
gta
skPr
opor
tion
fem
ale
Tim
epo
int
Out
com
eca
tego
ry(b
ehav
iora
lm
easu
re)
Spec
ific
outc
ome
dva
r(d)
./NP
.09
.06
./NV
Com
p.va
l.0
6.0
8./N
VU
tility
.24
.08
56A
./NL
AM
Inte
rest
�.3
3.0
7./N
PC�
.62
.08
./NE
(SR
)E
ffor
t�
.22
.07
./NE
(SR
)In
volv
emen
t�
.34
.07
./NP
.07
.07
./NV
Com
p.va
l�
.22
.07
./NV
Util
ity.1
8.0
761
(Wes
tern
)A
./NL
AM
Inte
rest
�.5
5.0
7./N
PC�
.35
.07
./NE
(SR
)E
ffor
t�
.17
.07
./NE
(SR
)In
volv
emen
t�
.1.0
7./N
P�
.49
.07
./NV
Com
p.va
l�
.13
.07
./NV
Util
ity�
.34
.07
60A
./NL
AM
Invo
lvem
ent
�.2
6.0
7./N
PC.3
2.0
7./N
E(S
R)
Eff
ort
.27
.07
./NE
(SR
)In
volv
emen
t.3
8.0
7./N
P�
.18
.07
./NV
Com
p.va
l.3
6.0
7./N
VU
tility
�.2
1.0
7Sh
in(2
010)
D43
A./N
.U
3A
M�
.2.1
./NU
5A
M�
.46
.1./N
AM
0.1
./NA
MIn
tere
st.1
8.1
./NA
MIn
trin
sic
valu
e�
.06
.141
./N.
PA
pplic
atio
n�
.93
.12
./NP
Defi
nitio
ns.5
7.1
2./N
PT
erm
inol
ogy
�.1
1.1
143
./N.
PG
rade
�.5
8.1
1./N
VU
tility
�.3
.1./N
AU
T�
.18
.1(t
able
cont
inue
s)
35EFFECTS OF RATIONALE
Thi
sdo
cum
ent
isco
pyri
ghte
dby
the
Am
eric
anPs
ycho
logi
cal
Ass
ocia
tion
oron
eof
itsal
lied
publ
ishe
rs.
Thi
sar
ticle
isin
tend
edso
lely
for
the
pers
onal
use
ofth
ein
divi
dual
user
and
isno
tto
bedi
ssem
inat
edbr
oadl
y.
-
Tab
le3
(con
tinu
ed)
Aut
hor
(yea
r)R
epor
tty
peSu
bsam
ple
(n)
Rat
iona
lety
pe
Ack
now
ledg
ene
gativ
eaf
fect
/no
ncon
trol
ling
lang
uage
Uni
nter
estin
gta
skPr
opor
tion
fem
ale
Tim
epo
int
Out
com
eca
tego
ry(b
ehav
iora
lm
easu
re)
Spec
ific
outc
ome
dva
r(d)
./NU
3C
M�
.09
.1./N
U5
CM
�.6
7.1
1./N
CM
�.3
5.1
./NFR
tope
ers
.28
.1./N
FRto
teac
hers
.23
.1./N
PC.4
1.1
44A
./N.
U3
AM
.23
.1./N
U5
AM
0.1
./NA
M.0
6.1
./NA
MIn
tere
st�
.06
.1./N
AM
Intr
insi
cva
lue
�.1
7.1
43./N
.P
App
licat
ion
.05
.1./N
PD
efini
tions
.45
.1./N
PT
erm
inol
ogy
.16
.144
./N.
PG
rade
.34
.1./N
VU
tility
.25
.1./N
AU
T.1
5.1
./NU
3C
M.2
1.1
./NU
5C
M�
.05
.1./N
CM
0.1
./NFR
tope
ers
.31
.1./N
FRto
teac
hers
�.7
6.1
./NPC
0.1
Stei
ngut
etal
.(2
012)
U34
(Bla
ck)
AC
./NH
AM
Inte
rest
.24
.12
./NPC
�.5
5.1
2./N
E(S
R)
.37
.12
./NP
.18
.12
./NV
.26
.12
./NFR
tope
ers
.02
.12
40(W
hite
)A
C./N
HA
Mto
teac
hers
�.2
7.1
./NPC
�.2
5.1
./NE
(SR
).4
4.1
./NP
.09
.1./N
V.2
2.1
./NFR
tote
ache
rs�
.17
.1
36 STEINGUT, PATALL, AND TRIMBLE
Thi
sdo
cum
ent
isco
pyri
ghte
dby
the
Am
eric
anPs
ycho
logi
cal
Ass
ocia
tion
oron
eof
itsal
lied
publ
ishe
rs.
Thi
sar
ticle
isin
tend
edso
lely
for
the
pers
onal
use
ofth
ein
divi
dual
user
and
isno
tto
bedi
ssem
inat
edbr
oadl
y.
-
Tab
le3
(con
tinu
ed)
Aut
hor
(yea
r)R
epor
tty
peSu
bsam
ple
(n)
Rat
iona
lety
pe
Ack
now
ledg
ene
gativ
eaf
fect
/no
ncon
trol
ling
lang
uage
Uni
nter
estin
gta
skPr
opor
tion
fem
ale
Tim
epo
int
Out
com
eca
tego
ry(b
ehav
iora
lm
easu
re)
Spec
ific
outc
ome
dva
r(d)
Van
stee
enki
ste
etal
.(2
005)
J47
CN
CL
/NL
1MP
Con
cept
ual
�.6
2.0
9N
CL
/NP
�.0
6.0
8N
CL
/N1M
PR
ote
.39
.09
NC
L/N
P.1
3.0
848
CC
L/N
L1M
PC
once
ptua
l�
1.33
.1C
L/N
P�
1.13
.1C
L/N
1MP
Rot
e.2
8.0
8C
L/N
P.3
9.0
948
AN
CL
/NL
1MP
Con
cept
ual
1.22
.1N
CL
/NP
1.18
.1N
CL
/N1M
PR
ote
�.6
4.0
9N
CL
/NP
�.2
.08
48A
CL
/NL
1MP
Con
cept
ual
1.22
.1C
L/N
P1.
18.1
CL
/N1M
PR
ote
�.6
4.0
9C
L/N
P�
.2.0
8V
anst
eeen
kist
eet
al.
(200
4)J
123
CN
CL
/NL
AM
Iden
tified
�.9
9.0
4N
CL
/NA
MIn
trin
sic
reg.
�.8
6.0
4N
CL
/NE
(SR
)�
1.47
.04
NC
L/N
1WE
(B)
Pers
iste
nce
�.3
7.0
3N
CL
/N1M
E(B
)Pe
rsis
tenc
e.4
5.0
3N
CL
/N4M
E(B
)Pe
rsis
tenc
e.4
8.0
3N
CL
/NP
Gra
de�
.48
.03
NC
L/N
PJo
inin
ggy
m0
0N
CL
/NC
M�
.07
.03
NC
L/N
CM
Intr
ojec
ted
�.0
6.0
312
1C
CL
/NL
AM
Iden
tified
�.1
3.0
3C
L/N
AM
Intr
insi
cre
g.�
.65
.03
CL
/NE
(SR
)�
.87
.04
CL
/N1W
E(B
)Pe
rsis
tenc
e�
.4.0
3C
L/N
1ME
(B)
Pers
iste
nce
.38
.03
CL
/N4M
E(B
)Pe
rsis
tenc
e�
.09
.03
CL
/NP
Gra
de�
.34
.03
CL
/NP
Join
ing
gym
00
CL
/NC
M.8
9.0
4(t
able
cont
inue
s)
37EFFECTS OF RATIONALE
Thi
sdo
cum
ent
isco
pyri
ghte
dby
the
Am
eric
anPs
ycho
logi
cal
Ass
ocia
tion
oron
eof
itsal
lied
publ
ishe
rs.
Thi
sar
ticle
isin
tend
edso
lely
for
the
pers
onal
use
ofth
ein
divi
dual
user
and
isno
tto
bedi
ssem
inat
edbr
oadl
y.
-
Tab
le3
(con
tinu
ed)
Aut
hor
(yea
r)R
epor
tty
peSu
bsam
ple
(n)
Rat
iona
lety
pe
Ack
now
ledg
ene
gativ
eaf
fect
/no
ncon
trol
ling
lang
uage
Uni
nter
estin
gta
skPr
opor
tion
fem
ale
Tim
epo
int
Out
com
eca
tego
ry(b
ehav
iora
lm
easu
re)
Spec
ific
outc
ome
dva
r(d)
CL
/NC
MIn
troj
ecte
d�
.39