Pervasive Negative Effects of Rewards on Intrinsic Motivation : The Myth Continues

The Behavior Analyst
2001, 24, 1–44
No. 1 (Spring)
Pervasive Negative Effects of
Rewards on Intrinsic Motivation:
The Myth Continues
Judy Cameron, Katherine M. Banko,
and W. David Pierce
University of Alberta
A major concern in psychology and education is that rewards decrease intrinsic motivation to
perform activities. Over the past 30 years, more than 100 experimental studies have been conducted
on this topic. In 1994, Cameron and Pierce conducted a meta-analysis of this literature and con-
cluded that negative effects of reward were limited and could be easily prevented in applied settings.
A more recent meta-analysis of the literature by Deci, Koestner, and Ryan (1999) shows pervasive
negative effects of reward. The purpose of the present article is to resolve differences in previous
meta-analytic ?ndings and to provide a meta-analysis of rewards and intrinsic motivation that per-
mits tests of competing theoretical explanations. Our results suggest that in general, rewards are not
harmful to motivation to perform a task. Rewards given for low-interest tasks enhance free-choice
intrinsic motivation. On high-interest tasks, verbal rewards produce positive effects on free-choice
motivation and self-reported task interest. Negative effects are found on high-interest tasks when
the rewards are tangible, expected (offered beforehand), and loosely tied to level of performance.
When rewards are linked to level of performance, measures of intrinsic motivation increase or do
not differ from a nonrewarded control group. Overall, the pattern of results indicates that reward
contingencies do not have pervasive negative effects on intrinsic motivation. Theoretical and prac-
tical implications of the ?ndings are addressed.
Key words: meta-analysis, rewards, reinforcement, intrinsic motivation, intrinsic interest
Most parents, educators, and behav-
but performance and interest are main-
ior analysts would agree that the ideal
tained only as long as the rewards keep
student is one who performs academic
coming. In other words, rewards are
tasks at a high level, shows high inter-
said to undermine intrinsic motivation.
est and involvement in school activi-
This premise is based on the view that
ties, is willing to take on challenging
when individuals like what they are
assignments, and is a self-motivated
doing, they experience feelings of
learner. To instill interest and to height-
competence and self-determination.
en student performance, many practi-
When students are given a reward for
tioners implement reward and incen-
performance, the claim is that they be-
tive systems in educational settings. In
gin to do the activity for the external
recent years, the wisdom of this prac-
reward rather than for intrinsic reasons.
tice has been debated in literature re-
As a result, perceptions of competence
views, textbooks, and the popular me-
and self-determination are said to de-
dia. Many writers and researchers
crease and motivation to perform the
claim that giving students high grades,
activity declines.
prizes, money, and even praise for en-
Those who decry the use of rewards
gaging in an activity may be effective
support their position by citing exper-
in getting students to perform a task,
imental studies on rewards and intrin-
sic motivation conducted in social psy-
This work was supported by a research grant
chology and education. Since the
from the Social Sciences and Humanities Re-
1970s, dozens of experiments have
search Council of Canada.
been designed to assess the impact of
Correspondence concerning this article should
rewards on intrinsic motivation. A cur-
be addressed to Judy Cameron, Department of
sory examination of the studies, how-
Educational
Psychology,
6-102
Education
North, University of Alberta, Edmonton, Alber-
ever, reveals a mixed set of ?ndings.
ta, T6G 2G5 Canada.
That is, in some studies, extrinsic re-
1

---

2
JUDY CAMERON et al.
wards produce negative effects on
vocated—the primary negative effect
measures of intrinsic motivation. Other
of rewards is that they tend to forestall
studies ?nd positive effects of reward;
self-regulation’’ (p. 659).
still others show no effect. A number
The assertion that rewards decrease
of reviewers have noted the contradic-
intrinsic motivation has captured the
tory nature of the ?ndings and have at-
attention of cognitive researchers,
tempted to identify the conditions un-
practitioners, and the general public
der which extrinsic rewards produce
because such a claim (a) seems to offer
decrements on measures of intrinsic
an empirical basis for psychological
motivation (Bates, 1979; Bernstein,
theories that assume that self-determi-
1990; Carton, 1996; Dickinson, 1989;
nation and freedom from control are
Flora, 1990; Morgan, 1984).
fundamental human motives, (b) ap-
In 1994, Cameron and Pierce pub-
pears to question basic behavioral con-
lished a meta-analysis of 96 experi-
ceptions of human nature, and (c) sug-
mental studies on the topic (with ad-
gests that rewards used in schools, hos-
ditional analyses by Eisenberger &
pitals, the workplace, and so on are
Cameron, 1996). Based on their re-
more harmful than bene?cial. A re-
sults, they argued that negative effects
viewer of this manuscript suggested
of reward were minimal and could be
that the claim that rewards are harmful
easily prevented in applied settings.
may be attractive to some practitioners
The research and recommendations
and educators because detecting and
made by Cameron and Pierce and by
rewarding performance improvements
Eisenberger and Cameron generated
is hard work and the negative effect
considerable debate (Hennessey &
claim relieves us of a dif?cult and de-
Amabile, 1998; Kohn, 1996; Lepper,
manding task.
1998; Lepper, Keavney, & Drake,
Clearly, Deci et al.’s (1999) ?nding
1996; Ryan & Deci, 1996; Sansone &
of general negative effects of reward
Harackiewicz, 1998) and seemingly
has important theoretical and practical
spurred the publication of a new meta-
implications and calls for a careful
analysis on the topic. Deci, Koestner,
analysis of contradictory empirical
and Ryan (1999) presented a meta-
claims. In this article, we argue that
analysis that claimed to support the
pervasive negative effects of reward
view that rewards have pervasive neg-
are not a necessary outcome of a meta-
ative effects on intrinsic motivation.
analysis of this literature. We contend
Deci et al. (1999) identi?ed 128 ex-
that a careful examination of Deci et
periments on rewards and intrinsic mo-
al.’s meta-analysis reveals several con-
tivation, including 20 unpublished
ceptual and methodological shortcom-
studies from doctoral dissertations.
ings. The disparate conclusions of the
They outlined a number of concerns
two major meta-analyses on the effects
they had with the meta-analyses con-
of reward on intrinsic motivation
ducted by Cameron and Pierce (1994)
(Cameron & Pierce, 1994; Deci et al.)
and Eisenberger and Cameron (1996).
suggest the value of correcting the
Deci et al.’s meta-analysis was de-
?aws in each and building on their
signed to rectify these concerns, to test
strengths to draw more de?nitive con-
cognitive evaluation theory, and to
clusions. In this article, we offer a re-
provide a more comprehensive review
analysis of the effects of rewards on
of the literature. Their ?ndings sup-
intrinsic motivation. Our reanalysis is
ported cognitive evaluation theory and,
informed by a consideration of Deci et
in general, rewards were found to have
al.’s decisions and procedures. In ad-
a substantial negative effect on intrin-
dition, the concerns raised by Deci et
sic motivation. Deci et al. concluded
al. about our previous research are ad-
that ‘‘although rewards can control
dressed. The purpose of the present ar-
people’s behavior—indeed, that is pre-
ticle is to resolve differences in previ-
sumably why they are so widely ad-
ous meta-analytic ?ndings and to pro-

---

THE MYTH CONTINUES
3
vide a meta-analysis of rewards and in-
money) for their work (Porter & Law-
trinsic motivation that permits tests of
ler, 1968; Vroom, 1964). Other theo-
competing theoretical explanations.
rists challenged the additive assump-
We begin with a general description
tion, suggesting instead that extrinsic
of the experiments conducted on re-
rewards might interfere with intrinsic
wards and intrinsic motivation. This is
motivation (DeCharms, 1968).
followed by a brief description of the
The idea that extrinsic rewards could
procedure and logic of meta-analysis.
disrupt intrinsic motivation instigated a
The meta-analyses by Cameron and
series of experiments carried out in the
Pierce (1994), Eisenberger and Cam-
early 1970s (Deci, 1971; Lepper,
eron (1996), and Deci et al. (1999) are
Greene, & Nisbett, 1973). In the initial
described, and criticisms of each are
studies, researchers tested the hypoth-
presented. We then provide a detailed
esis that external rewards would un-
account of how our reanalysis is de-
dermine intrinsic motivation either by
signed to resolve differences between
subverting feelings of competence and
Deci et al.’s and our earlier reviews of
self-determination or by de?ecting the
this literature. Results of our new
source of motivation from internal to
meta-analysis are presented, and dif-
external causes. Intrinsic motivation
ferences between our ?ndings and pre-
was inferred from changes in time
vious reviews are explained. Finally,
spent on an activity once rewards were
our discussion focuses on theoretical
removed, performance during the non-
and practical implications of the ?nd-
rewarded phase, or expressed task in-
ings.
terest. When rewards were found to
lower time on task, performance, or in-
THIRTY YEARS OF
terest, the researchers claimed that re-
RESEARCH ON REWARDS
wards undermined intrinsic motivation.
AND INTRINSIC
Results from the early studies appeared
MOTIVATION
to offer some support for the under-
mining hypothesis. That is, when in-
The term intrinsic motivation is gen-
dividuals were promised a material re-
erally understood in contrast to extrin-
ward, their performance, time on task,
sic motivation. Intrinsically motivated
and interest decreased once the reward
behaviors are those in which there is
was no longer forthcoming. Because of
no apparent reward except with the ac-
the implications for education, busi-
tivity itself (Deci, 1975). Extrinsic mo-
ness, and the psychology of motiva-
tivation, on the other hand, is said to
tion, the early ?ndings led to a great
occur when an activity is rewarded by
deal of research on the topic.
incentives not inherent in the task. Al-
Since the 1970s, over 100 experi-
though these terms have been criticized
ments have been performed to inves-
and debated (e.g., Bandura, 1986;
tigate alleged undermining effects of
Dickinson, 1989; Flora, 1990), they are
rewards. The vast majority of the stud-
accepted by many researchers. The dis-
ies on rewards and intrinsic motivation
tinction between intrinsic and extrinsic
have been conducted using a between-
motivation led psychologists to specu-
groups design. In a typical study, par-
late about the relation between these
ticipants are presented with an inter-
two sources. One view was that intrin-
esting task (e.g., solving and assem-
sic and extrinsic motivation combined
bling puzzles, drawing with magic
in an additive fashion to produce over-
markers, playing word games). Partic-
all motivation. For example, in work
ipants are rewarded with money or
settings, organizational psychologists
grades, candy, praise, good-player cer-
argued
that
optimal
performance
ti?cates, and so forth for performing
would occur when jobs were interest-
the activity. Rewards are tangible (e.g.,
ing and challenging and employees
money, candy, gold stars) or verbal
were externally rewarded (e.g., with
(e.g., praise, approval, positive feed-

---

4
JUDY CAMERON et al.
back). In addition, the rewards may be
over a number of sessions in a baseline
offered beforehand (expected reward)
phase, reinforcement procedures are
or presented unexpectedly after the ac-
implemented over several sessions, and
tivity (unexpected reward). In some ex-
?nally, reward is withdrawn and time
periments, reward is offered simply for
on task is assessed on repeated occa-
doing an activity; in other studies the
sions. An increase or decrease in in-
rewards are given for completing a task
trinsic motivation is measured by the
or for each puzzle or unit solved. In a
difference in time spent on the task be-
number of experiments, the rewards
tween baseline and postreinforcement
are offered for meeting or exceeding a
phases. In the ?ve studies employing
speci?c standard. Participants in a con-
this type of design (Davidson & Buch-
trol condition engage in the activity
er, 1978; Feingold & Mahoney, 1975;
without receiving a reward.
Mawhinney, Dickinson, & Taylor,
The reward intervention is usually
1989; Skaggs, Dickinson, & O’Connor,
conducted over a 10-min to 1-hr peri-
1992; Vasta, Andrews, McLaughlin,
od. Rewarded and nonrewarded groups
Stirpe, & Comfort, 1978), participants’
are then observed during a nonreward
performance during the postreward
period (typically, 2 min to 1 hr) in
phase either exceeded or remained at
which participants are free to continue
the same level as performance in the
performing the target task or to engage
prereward sessions. In other words,
in some alternative activity. The time
when the rewards were shown to func-
participants spend on the target activity
tion as reinforcement and multiple-tri-
during this nonreward phase, their per-
als procedures were used, there was no
formance on the task during the free-
evidence of a decremental effect of re-
choice period, or self-reported task in-
ward.
terest are used as measures of intrinsic
Those who argue that rewards de-
motivation. If rewarded participants
crease intrinsic motivation are critical
spend less free time on the activity,
of the single-subject designs. For ex-
perform at a lower level, or express
ample, Deci et al. (1999) state that one
less task interest than nonrewarded
can conclude very little from the sin-
participants, reward is said to under-
gle-subject designs because there are
mine intrinsic motivation.
too few participants and none of the
The ?ndings from the studies on re-
studies had control groups. Instead,
wards and intrinsic motivation have
claims about negative effects of reward
been diverse (positive, negative, and
are based on results from the group-
no effects have been reported). None-
design studies. As noted, however, the
theless, the results from these studies
?ndings from such studies have not
are often cited as evidence that rewards
been entirely clear-cut. Although most
and positive reinforcement can back?re
researchers have found that verbal re-
(e.g., Kohn, 1993). External rewards
wards do not decrease measures of in-
are said to be controlling and to inter-
trinsic motivation, the results with tan-
fere with a basic human desire for self-
gible rewards have been more contra-
determination.
dictory. To understand such diverse ef-
Because the detrimental effects of
fects, Cameron and Pierce (1994),
rewards have been interpreted as a
Eisenberger and Cameron (1996), and
challenge to behavioral conceptions of
Deci et al. used the methodology of
human nature and to the bene?ts of be-
meta-analysis to assess the group-de-
havioral technology for education and
sign experiments and to determine
business, a few behaviorally oriented
when and under what conditions re-
researchers have used single-subject
wards have detrimental effects on task
designs to assess the generality of the
performance and interest. Despite the
?ndings. In this type of study, partici-
seeming objectivity of this technique,
pants serve as their own controls. Mea-
these meta-analytic reviews reached
sures such as time on task are taken
markedly different conclusions. Cam-

---

THE MYTH CONTINUES
5
eron and Pierce and Eisenberger and
nitude of the relation (e.g., type of re-
Cameron reported minimal negative ef-
ward, reward contingency). Conduct-
fects of tangible reward, whereas Deci
ing a meta-analysis entails specifying
et al. found tangible rewards to be det-
the criteria for including and excluding
rimental under a wide range of condi-
studies, collecting all experiments that
tions.
meet the criteria, and coding the stud-
Although the usefulness of meta-
ies.
analysis and statistical testing in gen-
Once all relevant studies are identi-
eral has been questioned by behavioral
?ed, the statistical results of each study
researchers
(e.g.,
see
Baron
&
are transformed into a measure called
Derenne, 2000; Derenne & Baron,
an effect size. An effect size is found
1999), research summaries based on
by converting the ?ndings from each
meta-analyses have become valued
study into a standard deviation unit. In
sources of information for both policy
the rewards and intrinsic motivation
makers and researchers. Deci et al.’s
literature, an effect size indicates the
(1999) meta-analytic ?nding of general
extent to which the experimental group
negative effects of reward has impor-
(rewarded group) and the control group
tant implications. Thus, to understand
(nonrewarded group) differ in the
why the meta-analyses by Cameron
means on measures of intrinsic moti-
and Pierce (1994) and Deci et al. re-
vation (e.g., free time on task after re-
sulted in different ?ndings, it is impor-
wards are removed, task interest). In its
tant to be familiar with the technique
simplest form, the effect size (g) is the
and logic of meta-analysis. The meta-
difference between the means of the re-
analytic procedures described below
warded group and the nonrewarded
are based on Hedges and Olkin (1985);
control group divided by the pooled
these were the basic procedures used
standard deviation of this difference. In
by Cameron and Pierce and by Deci et
a meta-analysis, the effect size from
al.
each study, rather than the individual
participants within a study, becomes
THE TECHNIQUE AND LOGIC
the unit of analysis. If the effect sizes
OF META-ANALYSIS
from all the studies present a random
pattern, they will hover around zero,
Meta-analysis is a technique for
indicating no evidence for an effect.
combining the results of a large num-
On the other hand, the effect sizes may
ber of studies on the same topic. It in-
cluster in a positive or negative direc-
volves combining data from concep-
tion, indicating that something is going
tually related studies to reach general-
on.
izations based on statistical criteria.
One problem in meta-analysis arises
Quantitative analyses, similar to meta-
when studies do not provide enough
analysis, have been conducted on sin-
information to calculate effect sizes.
gle-subject designs (e.g., see Kollins,
When means and standard deviations
Newland, & Critch?eld, 1997); how-
are not available, effect sizes can be
ever, meta-analysis is typically used
calculated from t tests, F statistics, and
with between-groups designs in which
p values (see Hedges & Becker, 1986).
a treatment group (e.g., a rewarded
However, in some cases, there may still
group) is compared to a control group
be insuf?cient information to obtain an
(nonrewarded group) on a common de-
effect size. The meta-analyst can con-
pendent measure (intrinsic motivation).
tact the researchers and try to obtain
The goals of a meta-analysis are to es-
the missing data. When the data cannot
tablish the relation between indepen-
be procured, the study can be excluded
dent and dependent variables (in this
from the analyses or assigned an effect
case, the relation between rewards and
size of 0.00 (indicating no difference
intrinsic motivation) and to determine
between experimental and control
what factors moderate or alter the mag-
groups). It has been argued that includ-

---

6
JUDY CAMERON et al.
ing zero effect sizes is a conservative
pothesis is that the effect sizes are ho-
strategy; if a signi?cant effect is de-
mogeneous (i.e., effect sizes in a given
tected in spite of the inclusion of zeros,
analysis are viewed as values sampled
the contention is that the results would
from a single population; variation in
not be altered if missing data were
effect sizes among studies is merely
available (for a discussion of this issue,
due to sampling variation). When Q is
see Light & Pillemer, 1984). On the
statistically signi?cant, the implication
other hand, if one’s bias is toward no
is that moderator analyses should be
effect (i.e., we are satis?ed if the treat-
conducted. The original set of studies
ment is not harmful), including zeros
is then broken into subsets until the
favors this conclusion. One strategy for
chi-square statistics within the sub-
dealing with this issue is to conduct the
groups are nonsigni?cant. When the
analyses with zeros included and ex-
researcher has exhausted potential
cluded.
moderators and homogeneity is still
After effect sizes (g) are calculated
not obtained, outliers (studies with ex-
for each relevant study, an overall
treme effect-size values) are examined
mean effect size (d ) is obtained.
independently and the analysis is con-
First, g is converted to d by correcting
ducted with outliers removed.
for bias (g is an overestimation of the
population effect size, particularly for
DIFFERENCES BETWEEN
small samples; see Hedges, 1981). The
CAMERON AND PIERCE’S (1994)
overall mean effect size is obtained by
AND DECI ET AL.’S (1999)
weighting each effect size by the recip-
META-ANALYSES
rocal of its variance and averaging the
weighted d. This procedure gives more
Although Deci et al. (1999) and
weight to effect sizes that are more re-
Cameron and Pierce (1994) used the
liably estimated. The calculation of
same meta-analytic procedures to eval-
mean effect sizes provides a signi?-
uate the research on rewards and in-
cance test (whether the value differs
trinsic motivation, their results dif-
signi?cantly from 0.00) and a 95%
fered. Cameron and Pierce conducted a
con?dence interval (CI) (when the CI
hierarchical meta-analysis of the re-
contains 0.00, the results suggest that
wards and intrinsic motivation litera-
there is no evidence of a statistically
ture. Studies were included if they had
signi?cant effect).
a rewarded group and a nonrewarded
In a hierarchical meta-analysis, all
control group and if they used one of
studies are included in an overall anal-
the two main measures of intrinsic mo-
ysis. The researcher then searches for
tivation (free time on the task after the
moderator variables. The studies are
reward was removed or self-reported
broken out by one key moderator, then
task interest). The effects of reward on
another, and so on. The moderators that
the two dependent measures (free time
the researcher chooses to examine may
and task interest) were assessed sepa-
be based on theoretical considerations
rately. When a study did not provide
or on differences between the studies
enough information to calculate an ef-
(e.g., different procedures used in the
fect size, it was not included in the
studies, different characteristics of the
analyses.
samples used, year of publication,
Cameron and Pierce (1994) were
etc.).
?rst interested in whether rewards,
Hedges and Olkin (1985) recom-
overall, produced negative effects on
mend using homogeneity tests to as-
measures of intrinsic motivation. Their
certain whether a moderator analysis is
?ndings indicated no overall negative
necessary. Essentially, the procedure is
effects on either measure of intrinsic
to use a chi-square statistic, Q, with K
motivation. However, the set of effect
1 degrees of freedom, where K is the
sizes was signi?cantly heterogeneous;
number of effect sizes. The null hy-
thus, the researchers conducted a num-

---

THE MYTH CONTINUES
7
ber of moderator analyses to determine
calculate effect sizes, Deci et al. im-
when and under what conditions re-
puted effect sizes of 0.00 and included
wards produced negative effects. Re-
these in each of their analyses.
wards were broken down by reward
In terms of initial task interest, Deci
type (tangible and verbal). Tangible re-
et al. (1999) noted that ‘‘the ?eld of
wards were subdivided into expected
inquiry has always been de?ned in
and unexpected, and expected tangible
terms of intrinsic motivation for inter-
rewards were further separated by the
esting tasks and the undermining phe-
reward contingency. Cameron and
nomenon has always been speci?ed as
Pierce used a behavioral framework to
applying only to interesting tasks in-
categorize rewards by reward contin-
sofar as with boring tasks there is little
gency; in addition, they used the cate-
or no intrinsic motivation to under-
gories suggested by Deci and Ryan’s
mine’’ (p. 633). Given that cognitive
(1985) cognitive evaluation theory
evaluation theory has little to say about
framework. Their results indicated
the effects of rewards on low-interest
negative effects on the free-time mea-
tasks, Deci et al.’s meta-analysis fo-
sure only when the rewards were tan-
cused on reward effects on high-inter-
gible, expected, and not contingent on
est tasks. Studies or conditions within
meeting a performance standard. The
studies were included only if the tasks
same ?ndings were reported by Eisen-
used were measured or de?ned to be
berger and Cameron (1996), who car-
initially interesting; studies or condi-
ried out some additional analyses of re-
tions within studies were excluded if
ward contingencies.
the tasks used were measured or de-
Deci et al. (1999) suggested that
?ned as initially uninteresting.
Cameron and Pierce’s (1994) and Ei-
Thus, Deci et al.’s (1999) meta-anal-
senberger and Cameron’s (1996) fail-
ysis began with the overall effects of
ure to detect more pervasive negative
rewards on intrinsic motivation for
effects was due to methodological in-
tasks of initial high interest only. Deci
adequacies. Speci?cally, they criticized
et al. analyzed the effects of reward on
Cameron and Pierce and Eisenberger
measures of self-reported task interest
and Cameron for the following: (a) col-
and free-choice intrinsic motivation.
lapsing across tasks with high and low
Their free-choice measure included
initial interest and omitting a modera-
time spent on a task after rewards were
tor analysis of initial task interest, (b)
removed. When a time measure was
including a study that used an inappro-
not reported in a study, Deci et al. used
priate control group (Boal & Cum-
measures of task persistence during the
mings, 1981), (c) omitting studies or
free-choice period (e.g., number of tri-
data as outliers rather than attempting
als initiated in a labyrinth game, num-
to isolate moderators, (d) omitting
ber of balls played in a pinball game,
studies that were published during the
number of successes on a task). Hence,
period covered by their meta-analysis,
Deci et al.’s analysis of the free-choice
(e) omitting unpublished doctoral dis-
measure was broader than the analysis
sertations, and (f) misclassifying stud-
by Cameron and Pierce (1994), who
ies into reward contingencies as de-
used only studies that assessed time
?ned by cognitive evaluation theory.
measures.
To rectify these issues in their recent
On tasks of high initial interest, Deci
meta-analysis, Deci et al. (1999) ex-
et al. (1999) found a signi?cant nega-
cluded the study by Boal and Cum-
tive effect of rewards on the free-
mings (1981), included studies that
choice measure and a non-signi?cant
were missed in the previous meta-anal-
effect on the self-report measure. Both
yses, and included unpublished doctor-
mean effect sizes were heterogeneous.
al dissertations. In addition, in contrast
To obtain homogeneity at each level of
to Cameron and Pierce (1994), for
analysis, Deci et al. tested a number of
studies with insuf?cient information to
moderator variables. When homoge-

---

8
JUDY CAMERON et al.
neity could not be obtained, Deci et al.
contingent rewards into studies of
followed the procedure used by Cam-
‘‘maximum’’ and ‘‘not-maximum’’ re-
eron and Pierce (1994) and identi?ed
ward. In studies of maximum reward,
and removed outliers. First, Deci et al.
participants were offered rewards grad-
tested whether verbal versus tangible
ed in terms of meeting a criterion or
rewards were a moderator. Verbal re-
performance standard; all met the cri-
wards were found to increase free-
terion and received the full amount of
choice intrinsic motivation for college
reward. Six studies were identi?ed by
students (a nonsigni?cant effect was
Deci et al. as involving not-maximum
found for children) and to enhance task
reward. In these studies, some partici-
interest for both children and college
pants failed to attain the criterion and
students. Tangible rewards produced
were given less than the maximum re-
negative effects on both the free-choice
ward. Deci et al. reported that relative
and self-report measures. In accord
to a nonrewarded control condition,
with Cameron and Pierce, tangible re-
participants receiving less than the
wards were broken down into expected
maximum reward showed a large de-
and unexpected rewards. Unexpected
cline in free-choice intrinsic motiva-
rewards had no signi?cant effects; ex-
tion. In fact, the value (d
0.88) was
pected tangible rewards were found to
the largest mean effect size in their en-
signi?cantly undermine both self-re-
tire analysis.
ported task interest and free-choice in-
As a supplemental analysis, Deci et
trinsic motivation.
al. (1999) analyzed studies with chil-
Using cognitive evaluation theory as
dren in which the free-choice assess-
their framework, Deci et al. (1999) fur-
ment of high-interest activities was
ther subdivided expected tangible re-
conducted immediately following the
wards into task-noncontingent, engage-
removal of reward, within a week, and
ment-contingent, completion-contin-
after a week. Deci et al. found negative
gent, and performance-contingent re-
effects at each time of assessment and
wards. Task-noncontingent rewards
suggested that the undermining effect
were ‘‘those given without speci?cally
is not a transitory phenomenon. An ad-
requiring the person to engage in the
ditional analysis of the effects of re-
activity’’ (p. 636); engagement-contin-
wards on low-interest tasks was con-
gent rewards were those offered to par-
ducted by Deci et al.; no statistically
ticipants for engaging in a task without
signi?cant effects were detected.
a requirement to complete the task, do
All in all, Deci et al.’s (1999) meta-
it well, or reach some standard. Com-
analysis produced numerous negative
pletion-contingent rewards were those
effects of the various reward contin-
offered and given for completing a
gencies. Given the discrepancies be-
task, and performance-contingent re-
tween Deci et al.’s and Cameron and
wards were de?ned as those ‘‘offered
Pierce’s (1994) ?ndings, it is important
dependent upon the participants’ level
to examine carefully the procedures
of performance’’ (p. 636). Deci et al.
used by Deci et al. The ?rst noteworthy
found no signi?cant negative effects
difference between the two meta-anal-
for task-noncontingent rewards; en-
yses occurs at the level of all rewards.
gagement-contingent rewards produced
Cameron and Pierce were interested in
signi?cant negative effects on both
assessing the overall effects of rewards
free-choice intrinsic motivation and
across all types of tasks. Deci et al. did
self-reported task interest. Completion-
not conduct this analysis; instead, they
contingent and performance-contingent
argued that the more theoretically rel-
rewards also resulted in signi?cant
evant question concerned the effects of
negative effects on the free-choice in-
rewards on tasks of high initial interest.
trinsic motivation measure.
We contend that an analysis of the
In addition, Deci et al. (1999) pro-
overall effect of reward is central to an
vided a breakdown of performance-
understanding of this complex area of

---

THE MYTH CONTINUES
9
research. On a practical level, many
a control group performed the task
educators, parents, and administrators
without the offer of reward. In Deci et
have taken the position that rewards
al.’s analyses, only one of the rewarded
and incentive systems are harmful. The
groups was included. For other studies
view is that rewards negatively affect
that used more than one level of re-
students’ intrinsic interest across all
ward magnitude (e.g., Earn, 1982;
types of activities (e.g., reading, math,
McLoyd, 1979; Newman & Layton,
science, computer games, etc.); no dis-
1984), Deci et al. included all reward
tinction is made between low and high
conditions. Their omission of certain
initial levels of task interest. Writers
conditions within studies does not ap-
who caution against the use of rewards
pear to be systematic (e.g., reward
and reinforcement frequently use ex-
magnitude was not examined by Deci
amples to illustrate their point. More
et al. as a moderator), yet there are a
often than not, activities such as read-
number of different types of cases in
ing, lawn mowing, and mathematics
which this occurs. In addition, as did
are cited as activities that people will
Cameron and Pierce (1994), Deci et al.
lose interest in if they are given re-
also missed a few experiments that met
wards for performing the activity. Most
their inclusion criteria and that were
of these activities are not ones that in-
published during the period covered by
dividuals begin doing with high levels
their meta-analysis. Also, several stud-
of initial interest. Importantly, policy
ies using high-interest tasks that mea-
makers who adopt the view that re-
sured self-reported task interest were
wards are harmful rarely distinguish
either excluded or inadvertently omit-
between high- and low-interest tasks.
ted from Deci et al.’s analyses. Many
Because of this, an analysis of the
of these studies found positive effects
overall effects of reward is warranted.
on the self-report measure of task in-
It is our contention that a more com-
terest; Deci et al.’s omission of these
plete hierarchical breakdown of the ef-
effects helps to explain why they found
fects of rewards on intrinsic motivation
either negative effects or no effects on
should begin at the level of all rewards
the task-interest measure. A list of
over all types of tasks. Following this,
studies not included in Deci et al.’s
a breakdown of reward effects on high-
analyses, dependent measures that
and low-interest tasks would be appro-
were precluded, and a description of
priate.
conditions omitted by Deci et al. are
A further dif?culty with Deci et al.’s
presented in Appendix A. Any com-
(1999) meta-analysis concerns their
putational differences in sample sizes
supplemental analysis of reward effects
and effect sizes are also outlined in
on low-interest tasks. Several studies
Appendix A.
that used low-interest tasks were ex-
A ?nal issue concerns the classi?-
cluded from their primary meta-analy-
cation of studies into various reward
sis of high-interest tasks (e.g., Freed-
contingencies. Deci et al. (1999) sug-
man & Phillips, 1985; Overskeid &
gested that Cameron and Pierce (1994)
Svartdal, 1996). The problem is that
miscategorized many experiments. Us-
these studies were not brought back
ing cognitive evaluation theory to
into their supplementary analysis of
guide their classi?cation of studies,
low-interest tasks.
Deci et al. established the categories of
Another concern is that for some
task-noncontingent, engagement-con-
studies in their analysis of high-interest
tingent, completion-contingent, and
tasks, Deci et al. (1999) omitted con-
performance-contingent rewards. Al-
ditions that were relevant to their anal-
though this categorization system may
yses. For example, in an experiment by
be informative for cognitive evaluation
Wilson (1978), one group was offered
theory, the problem is that the catego-
$0.50 to engage in the target activity,
ries are too broad. Studies that used
a second group was offered $2.50 and
very different procedures were pooled

---

10
JUDY CAMERON et al.
into overall categories of engagement-
A NEW META-ANALYSIS:
contingent, completion-contingent, and
RESOLVING DIFFERENCES
performance-contingent rewards. For
BETWEEN PREVIOUS
example, under performance-contin-
META-ANALYSES
gent reward, Deci et al. pooled exper-
iments in which participants were of-
To address criticisms and resolve
fered a reward for doing well, for each
discrepancies among Cameron and
problem or unit solved, for obtaining a
Pierce’s (1994), Eisenberger and Cam-
certain score, or for exceeding a norm.
eron’s (1996), and Deci et al.’s (1999)
Eisenberger, Pierce, and Cameron
meta-analyses, we provide a reanalysis
of the effects of rewards on intrinsic
(1999) examined some of these differ-
motivation. Our goal is to eliminate
ent reward procedures and found very
shortcomings of the prior meta-analy-
different effects on measures of intrin-
ses (including ours) while building on
sic motivation.
their strengths. Our current meta-anal-
Rather than argue about which stud-
ysis focuses on how reward affects
ies belong in which category, we sug-
measures of free-choice intrinsic mo-
gest that a more nuanced approach is
tivation and self-reported task interest.
to return to the methods section of the
In accord with Deci et al., free-choice
original studies and code the reward
measures included free time on task
procedures actually employed in the
when the rewards were removed and,
experiment. To the extent that one can
when time measures were not avail-
obtain consistency in coding, such a
able, performance during the free-
procedural categorization of reward
choice period. As did Deci et al., we
contingencies would allow an assess-
combined performance and time mea-
ment of the effects of the actual con-
sures to make up the free-choice intrin-
tingencies rather than those presumed
sic motivation index (we found no sig-
to be an effect by any theoretical per-
ni?cant differences in the analyses
spective. The literature on rewards and
when only time measures were ana-
intrinsic motivation is fraught with
lyzed). Our analysis begins with an as-
competing theories (e.g., cognitive
sessment of the overall effects of re-
evaluation theory, the overjusti?cation
ward.
hypothesis, social cognitive theory,
We then examine the effects of dif-
general interest theory, the competing
ferent moderator variables. To deal
response hypothesis, behavioral theo-
with Deci et al.’s (1999) criticism, our
ry). The problem with organizing stud-
?rst breakdown is in terms of high and
ies according to a particular theoretical
low initial task interest. Subdividing
stance is that each theory could be
the studies by high and low initial task
used to organize the literature and, us-
interest allows us to directly compare
ing categories appropriate to the theo-
our ?ndings with those of Deci et al.
ry, each theory could gain support. Us-
In doing so, we favor cognitive eval-
ing a theoretical approach to guide the
uation theory. On the other hand, fail-
classi?cation of the reward procedures
ure to ?nd pervasive negative effects
does not provide us with a de?nitive
even with high-interest tasks favors the
answer about the effects of reward con-
conclusion that reward contingencies
tingencies on measures of intrinsic mo-
do not destroy interest. In other words,
tivation. Instead, we propose that a
the strongest way to test Deci et al.’s
procedural description of reward con-
claims is to use their requirement that
tingencies not only allows us to assess
tasks used in the studies must be bro-
where we stand in terms of the effects
ken out by high and low initial interest.
of the actual reward contingencies but
At each level of our analysis, a ho-
also provides us with a test of alter-
mogeneity statistic (Q) was calculated to
native accounts of the effects of re-
determine whether the set of effect sizes
wards on intrinsic motivation.
could
be
considered
homogeneous.

---