ORGANIZATIONAL BEHAVIOR AND HUMAN PERFOP,~MANCE 20, 3 1 - 5 3
(1977)
Further Examinations of the Relationship between Reward Contingency and Intrinsic Motivation JAMES L . FARR, ROBERT J. VANCE, AND ROBERT M . MCINTYRE
The Pennsylvania State University A series of laboratory studies were conducted to examine the effect of contingent monetary reward on the occurrence of behavior presumably determined by intrinsic task motivation. A general hypothesis that contingent monetary rewards would reduce the frequency of such behaviors was derived from a pos relate of cognitive evaluation theory. For all of the studies, intrinsic task motivation was operationally defined as the amount of free time a subject spent working on a task without receiving reward. Study 1 investigated the effects of reward contingency and reward magnitude upon the behavioral measure of intrinsic motivation and upon related attitudinal variables. Study 2 examined the effects of reward contingency and two individual difference variables (self-esteem and perceived locus of control) on intrinsic motivation and attitudinal measures. Although Study 2 showed no significant results, Study 1 demonstrated support for the hypothesis with the behavioral measure of intrinsic motivation, but not with the attitudinal measures. Further examination of the behavioral measure of intrinsic motivation suggested that this variable was bimodally distributed in both Studies 1 and 2. A test for normality indicated that the distributions were significantly non-normal and that parametric analyses of mean differences were inappropriate. Study 3 reported the results of nonparametric reanalyses of Studies 1 and 2 and of data previously published by Deci as supportive of the postulate under examination. Deci's data were also found to be generally bimodal. The nonparametric analyses were nonsignificant for all data sets. Theoretical implications of these results are discussed and future directions for research are suggested.
Traditional work motivation theories, including those in the expectancy framework (e.g., Porter and Lawler, 1968; Vroom, 1964), have assumed that intrinsic and extrinsic rewards are additive in their effects upon motivation. Recently, however, this assumption has been challenged by a number of researchers (most notably Deci, 1975, 1976) who contend that, at least under certain circumstances, there is an inverse relationship between extrinsic: rewards and intrinsic motivation. Intrinsically motivated behavior may be defined as behavior which ocThis research was supported in part by an institutional grant from the National Science Foundation to The Pennsylvania State University and by a grant from the Central Fund for Research, College of Liberal Arts, Pennsylvania State University. The assistance of Thomas Mitchell, Jeffrey lVlcClafferty, and Michael Ward is greatly appreciated. Requests for reprints should be sent to James L. Farr, Department of Psychology, Pennsylvania State University, University Park, PA 16802. 31 Copyright © 1977 by Academic Press, Inc. All rights of reproduction in any form reserved.
ISSN 0030-5073
32
FARR, VANCE, AND MCINTYRE
curs for which there are no apparent extrinsic rewards. Regarding human motivation, Deci (1975) has suggested that it would be more appropriate to define intrinsically motivated behaviors as those which are involved with the human need for being competent and self-determining (cf. deCharms, 1968; White, 1959). Cognitive evaluation theory (Deci, 1975) suggests that there are two processes by which intrinsic motivation is affected by extrinsic rewards. First, an individual's perceived locus of causality may be changed. If a behavior is intrinsically motivated, the individual perceives the locus of causality for the behavior to be his own intrinsic needs. When an individual receives extrinsic rewards for such behavior, however, the perceived locus of causality becomes external and the individual will then perform the behavior only if he views the behavior as instrumental for the attainment of the extrinsic reward. Second, intrinsic motivation may be changed through a shift in feelings of competence and self-determination. If rewards convey to individuals that they are competent and self-determining, then intrinsic motivation is enhanced by the rewards. However, if the rewards convey to individuals that they are not competent, then intrinsic motivation is decreased. Deci (1975) argues that all extrinsic rewards have both a controlling aspect and an informational aspect. The controlling aspect of extrinsic rewards changes the perceived locus of causality of the behavior. Thus, the individual receiving an extrinsic reward for certain behavior perceives the reward as the cause for that behavior. The informational aspect of rewards provides the individual with information regarding one's performance on the rewarded task or behavior, thereby enhancing feelings of personal competence. Thus, the cognitive evaluation theory developed by Deci (1975) implies that extrinsic rewards may either decrease or enhance intrinsic motivation dependent upon whether the controlling or informational aspect of the rewards is predominant. Deci (1972b, 1975, 1976) suggests that the controlling aspect is likely to be dominant when the extrinsic rewards are contingent upon the behavior. This behavior-reward contingency would strengthen the perception of the locus of causality for the behavior as extrinsic to the individual, viz., the reward itself. Thus, an individual, who has been contingently rewarded for an activity and has developed an extrinsic locus of causality, would be predicted to display subsequently less intrinsic motivation for that activity when no reward is expected when compared with an individual not previously rewarded or noncontingently rewarded for the activity. The deduction that contingent extrinsic rewards (particularly pay) may decrease intrinsic task motivation has important practical and theoretical implications. Industrial organizations have used incentive pay systems since the advent of the scientific management approach in an attempt to increase the task motivation (and thereby the performance) of their employees. While incentive plans have not been universally successful
REWARD CONTINGENCY AND INTRINSIC MOTIVATION
33
(Lawler, 1973), the criticisms of and problems associated with these pay systems have been confined to the employees' neglect of areas of performance not explicitly related to pay, the reporting of invalid performance data, and the perception of a relationship between good performance and negatively valued outcomes (such as co-worker dislike). The reduction of intrinsic motivation by contingent pay would be another serious negative side effect of these pay plans. At a more theoretical level, cognitive approaches to work motivation such as expectancy theory (Dachler & Mobley, 1973; Lawler, 1973; Porter & Lawler, 1968) and behavioristic approaches (Jablonsky & DeVries, 1972; Nord, 1969) have both stressed the desirability of strengthening the relationship between performance and pay. The finding that contingent pay decreased the frequency of behaviors presumably determined by intrinsic task motivation would force significant modifications of these theoretical formulations concerning work motivation. What is the empirical status of the proposition that contingent extrinsic rewards (especially pay) decrease the rate of occurrence of intrinsically motivated behaviors? Empirical data relevant to deductions from this postulate have been interpreted differently by various writers. Deci (1975, 1976) has been the most ardent supporter of the postulate, arguing that his research (Deci, 1971, 1972a, 1972b) and that of others (e.g., Calder and Staw, 1975b; Greene and Lepper, 1974; Kruglanski, Alon, & Lewis, 1972; Lepper & Greene, 1975; Lepper, Greene, & Nisbett, 1973) gives the postulate unequivocal support. However, other researchers have criticized the methodology of a number of these studies (Calder & Staw, 1975a; Feingold & Mahoney, 1975; Scott, 1976) and nonsupportive data have been reported in other studies (Farr, 1976; Feingold & Mahoney, 1975; Hamner & Foster, 1975). Thus, it appears that more research is necessary before the relationship between extrinsic rewards and behaviors presumably determined by intrinsic motivation is well understood. A potentially useful line of research would appear to be the investigation of variab}es which may moderate the relationship between the two sets of variables. One variable which appeared to differentiate the research of Farr (1976) and Deci (1971, 1972b) was reward magnitude. In the Fan" (1976) study, which failed to support the postulate that contingent monetary rewards would decrease the frequency of occurrence of behaviors presumably determined by intrinsic task motivation, the average subject earned about $1.80. Subjects in the research of Deci could earn as much as $4.00. Reward magnitude could affect the saliency of the financial incentives. Ross (1975) has found that increasing the saliency of rewards increased the detrimental effect of extrinsic rewards upon intrinsically motivated behaviors. The purpose of the first study reported here was to investigate in a
34
FARR, VANCE, AND MCINTYRE
laboratory setting the effects of reward magnitude in both contingent and noncontingent conditions upon a behavioral operationalization of intrinsic motivation and upon a number of questionnaire measures of variables related to that construct. The specific hypotheses of the study were: 1. Contingent rewards adversely affect behavior presumably determined by intrinsic motivation and task satisfaction more than noncontingent rewards. 2. As reward magnitude is increased, the frequency of behavior presumably determined by intrinsic motivation and the level of task satisfaction are decreased. 3. Reward magnitude and reward contingency interact so that the differential effects of contingent and noncontingent rewards become greater as reward magnitude increases. STUDY 1 Method
Subjects Subjects were recruited from introductory psychology classes at The Pennsylvania State University. The subjects received course credit for their participation, as well as monetary compensation as described below. The 48 subjects (16 males and 32 females) were randomly assigned to six experimental conditions, with eight subjects per condition.
Task The task used in the present experiment was Soma, a puzzle manufactured by the Parker Brothers Co. The puzzle consists of 27 one-inch plastic cubes which are permanently joined to form 7 irregularly shaped pieces. The object of the puzzle is to reproduce a given configuration using two or more puzzle pieces.
Independent Variables The two independent variables in the present experiment were reward contingency and reward magnitude. There were two levels of reward contingency: subjects were paid contingently, receiving a certain amount of money for each puzzle solved, or noncontingently, in which case subjects received a fixed amount of money for their participation regardless of their ability for solving the puzzles. There were three levels of reward magnitude: subjects paid contingently received either $.50, $1.00, or $1.50 for each puzzle which they successfully solved; those subjects paid noncontingently received either $1.00, $2.00, or $3.00 for participating, depending on the condition to which they were assigned. These rates were based upon data presented by Deci, Cascio, and Krusell (1975) which indicated that the average subject in their research solved about one-half of the puzzles. Thus, the pay levels
REWARD
CONTINGENCY
AND INTRINSIC
MOTIVATION
35
used in the present research were intended to equate approximately the average amount of pay earned by subjects in the contingent and noncontingent pay conditions.
Dependent Variables A number of dependent variables were used in this study. Intrinsic motivation was measured behaviorally by the amount of time that the subject worked on the task during an unrewarded, free time period. The number of puzzles correctly solved during the experimental session and the time required to solve the puzzles were used as measured of task performance. Several attitudinal variables were measured by a questionnaire. These variables included three item scales designed to measure task interest, perceived task difficulty, intrinsic task motivation, pay fairness, and performance satisfaction. Slightly modified JDI scales (Smith, Kendall, & Hulin, 1969) measured satisfaction with the task, pay, and the experimenter (using the JDI supervisory scale). Subjects' attributions of the causes of task motivation were also assessed. Subjects were asked to rate on a seven-point scale the contribution of each of nine factors to task motivation (defined as "how hard you worked on the task"). Three of these factors were extrinsic (being accepted by the experimenter, praise from the experimenter, and money received), three were task-oriented (task interest, task difficulty, and knowledge of results), and three were intrinsic (satisfaction from working on the task, feeling of self-competence, and feeling of self-esteem).
Procedure The general procedure was patterned after that of Deci (1972b). The experimental room contained a table with two chairs at which the experimenter and the subject were seated across from each other, another table to the subject's left on which were placed current issues of Newsweek, Sports Illustrated, Glamour, and Esquire, and a two-way mirror through which the subjects could be observed by a second experimenter. Since many subjects could have been aware of the fact that the mirror was two-way, thus raising the possibility of their being observed, a heavy curtain was drawn across the mirror in order to give the appearance that it was not being used in the present experiment. However, one small corner of the mirror was left uncovered to afford the observer a view of the subject as he was seated at the table. Upon entering the experimental room the subject was informed that he was participating in a study designed to examine puzzle-solving strategies. Subjects in the contingent payment conditions were told that they would receive $0.50, $1.00, or $1.50 for each puzzle correctly solved (dependent upon the particular condition to which the subject had been assigned). Subjects in the noncontingent p~yment conditions were told
36
FARR, VANCE, AND MCINTYRE
that they would receive $1.00, $2.00, or $3.00 for participating in the study (again, dependent upon the experimental condition). The subject was then presented with the puzzle pieces and with a practice configuration (illustrated on a piece of paper) which he had 2 min to attempt to construct. If the subject failed to solve the practice problem in the allotted time, the experimenter read the solution to him from a set of directions, and the subject assembled it himself. The first of four configurations was then presented, and the subject was told that he had 10 min to solve it. If he failed to solve the puzzle, the experimenter read the solution to him and the subject constructed the configuration himself, thus allowing him to see that the solution was indeed possible. The remaining three trials were conducted in the same manner. The subjects who were in the contingent payment condition were informed of the total amount of money earned after each trial. After completion of the four puzzles, the subject was informed by the experimenter that the final task would be to fill out a questionnaire. He was told that there were several forms of the questionnaire, and the appropriate form would be determined by the data which the experimenter had been recording throughout the session. It was further explained that a computer was used for this purpose, and it would be necessary for the experimenter to leave the room to go to a terminal in another part of the building. The subject was informed that the experimenter would return in about 10 min, and he could do anything he liked in the interim except leave the room. The experimenter then left the room for 8 min (the "free-time period' '), during which the observer measured the amount of time the subject worked on the puzzle. The experimenter had left on the table two illustrations of figures which were impossible to construct (unknown to the subject). Ostensibly, these were configurations which simply were not being used for that subject. The fact that these puzzles were unsolvable ensured that the subject, if he chose to work on the puzzles, could be busy for the full 8 minutes. The subject was judged to be working on a puzzle if he was manipulating one or more of the pieces and/or focusing his attention on the puzzle illustration. The magazines were provided in order to give the subject an attractive alternative to working on the puzzles. After 8 min the experimenter returned to the room with the questionnaire, which the subject then completed. Following this the subject was paid and the nature of the study was explained to him.
Analysis A multivariate analysis of variance was performed on the data. For sources of variance which were found to have a significant multivariate F Co < .05), univariate analyses of variance were conducted for each dependent variable separately.
37
REWARD CONTINGENCY AND INTRINSIC MOTIVATION
Resu Its Table 1 presents the results of the multivariate and univariate analyses of variance for the fourteen dependent variables. The only multivariate F which reached statistical significance (p < .05) was that for the Reward Contingency factor. Three univariate analyses of variance reached statistical significance for this factor. Reward Contingency significantly affected the dependent variables of Free Time Spent on Puzzles, the questionnaire measure of Intrinsic Motivation, and Money Attribution, Table 2 presents the means for the two conditions of the Reward Contingency factor for the three dependent variables significantly affected by this factor. The data provide some support for Hypothesis 1. Subjects in the Contingent Reward condition spent significantly less time during the free period working on the task than the subjects in the Noncontingent Reward condition. However, the questionnaire measure of intrinsic task motivation ( a d ~ t e d from Hackman and Lawler, 1971) gave opposite resuits. Subjects in the Contingent Reward condition reported more intrinsic task motivation, as measured by the three item scale, than did subjects in the Noncontingent Reward condition. Subjects in the Contingent Reward condition attributed more importance to money as a cause of task motivation than did the noncontingently paid subjects. There were no
TABLE 1 MULTIVARIATE AND UNIVARIATE ANALYSES OF VARIANCE: STUDY 1
Multivariate F Univariate F ' s Number of Puzzles Solved Time Required to Solve Puzzles Free Time Spent on Puzzles Task Interest Intrinsic Motivation Task Difficulty Pay Fairness Performance Satisfaction Task Satisfaction Experimenter Satisfaction Pay Satisfaction Money Attribution Intrinsic Attribution Extrinsic (Total) Attribution * p < .05 * * p < .01
Reward contingency
Reward magnitude
A × B
1.98"
1.22
1.00
0.07
1,22 5.57* 0.60 5.47* 0.25 0.08 1.81 1.64 0.11 0.50 10.17"*
1.05 1.12
b
n
38
FARR~ VANCE, AND MCINTYRE
significant effects of Reward Contingency upon any other dependent variables including the theoretically important ones of task satisfaction and attribution of motivation to intrinsic causes. The data offered no support for either Hypothesis 2 or 3. A significant multivariate F statistic was not found for either the main effect of Reward Magnitude or for the Reward Contingency x Reward Magnitude interaction.
Discussion Although the results for the behavioral measure presumed to indicate intrinsic motivation were supportive of Deci's (1975) postulate, several findings of the present study raised questions. The inconsistency between the behavioral and self-report measures of intrinsic motivation suggest difficulties with the construct of intrinsic motivation. The self-report measure of intrinsic motivation was adapted (slightly) from one developed by Hackman and Lawler (1971) and subsequently used in several published research investigations (e.g., Brief & Aldag, 1975). It seems necessary to examine further the relationship between behavioral and self-report measures of intrinsic motivation. The data for the behavioral measure of intrinsic motivation for this study raise other issues. The frequency distribution for the dependent variable of Free Time Spent on Puzzles was markedly bimodal. Twentytwo of the subjects spent less than 1 min on the task during the free time period and fourteen spent more than 7 min. Only twelve subjects spent between 1 and 7 min on the task during this period. The data for the contingent pay and noncontingent pay conditions, considered separately, were also bimodal, especially for the noncontingent condition. In the contingent condition fourteen subjects spent less than 1 min of the free time period on the task, four spent more than 7 rain, and six spent between 1 and 7 rain on the task. In the noncontingent condition eight subjects spent less than 1 min on the task, 10 spent more than 7 min, and six
TABLE 2 CONDITION MEANS FOR THE REWARD CONTINGENCY FACTOR: STUDY 1
Condition Dependent variable Free Time Spent on Puzzles (min) Intrinsic Motivation-questionnaire measure Money Attribution
Contingent reward
Noncontingent reward
2.12
4.38
17.50 3.58
15.29 2.17
REWARD CONTINGENCY AND INTRINSIC MOTIVATION
39
spent between 1 and 7 rain. The bimodal nature of these data suggested that individual difference variables might be moderating the relationship between extrinsic reward and intrinsic motivation. It also suggested that data analyses based upon the mean might not be appropriate due to the distributional characteristics of the variable. On the basis of the results of this investigation, two things were decided. First, a research study would be conducted which explicitly attempted to consider individual differences as a moderator variable. It was also decided to obtain, if possible, some raw data from Deci (1972a,b; Deci, Cascio, & Krusell, Note 1) for purposes of nonparametric reanalysis along with the data of this research. These reanalyses are reported in Study 3. STUDY 2 Locus of control (Rotter, 1966) and self-esteem were chosen as potential moderator variables to be examined in this research. Locus of control refers to the individual's chronic belief that the occurrence of an event or outcome is contingent upon his own behavior or his own relatively permanent characteristics (internal locus of control) or that the occurrence is the result of chance, is under the control of other individuals, or is unpredictable due to multiple causal factors (external locus of control). Since an individual with a belief in an external locus of control would be unlikely to attribute task behaviors to intrinsic motivation, extrinsic rewards would be able to cause an attributional shift only for individuals with a belief in an internal locus of control. Chronic self-esteem might also function as a moderator of the relationship between extrinsic rewards and intrinsic motivation. High self-esteem individuals might be more resistant to the attributional shift proposed by Deci (1975) than low self-esteem individuals. The hypotheses of Study 2 were: 1. Extrinsic contingent rewards would decrease the frequency of behavior presumably determined by intrinsic motivation of individuals with a belief in an internal locus of control, but would not affect such behavior of individuals with an external locus of control. 2. Extrinsic contingent rewards would decrease the frequency of behavior presumably determined by intrinsic motivation of individuals low in self-esteem, but would not affect such behavior of those high in self-esteem. Method
Subjects One hundred and fifty-two introductory psychology students at The Pennsylvania State University served as subjects in this study.
40
FARR, VANCE, AND MCINTYRE
Subject Screening The subjects were administered in group settings a questionnaire which contained scales to measure locus of control (Rotter, 1966) and serfesteem (Ghiselli, 1971). The correlation between these two scales was found to be - . 11, which was sufficiently low that a factorial design using both of these individual difference variables was feasible. The median score for the locus of control distribution was 10. Subjects receiving scores of 8 or less were classified as internal locus of control, and those receiving scores of 12 or more were classified as external. The median self-esteem score was 24. Subjects receiving scores of 26 or more were classified as high self-esteem and those receiving scores of 22 or less were classified as low self-esteem. The joint classification of subjects in terms of both the locus of control and serf-esteem variables resulted in 24 subjects classified as external locus of control, low self-esteem; 26 as external and high serf-esteem; 22 as internal locus of control, low self-esteem; and 25 as internal, high serf-esteem.
Independent Variables Three independent variables were examined in this study: locus of control, self-esteem, and reward contingency. Two levels of locus of control were used which are designated internal and external (Rotter, 1966). High and low self-esteem were used as the levels of this variable. Reward contingency also had two levels as described in Study 1.
Dependent Variables The dependent variables in this study were identical to those in Study 1.
Procedure In general, the procedure for this study paralleled that of Study 1. There were, however, a few changes from the procedure of Study 1. Upon their arrival at the experimental room and before they knew the exact task to be completed in the experiment, subjects were asked to complete a 10-item rating scale concerned with subjects' preferences for various recreational and leisure activities. Only one item was of interest to the present research, that concerning interest in solving puzzles. This was included in this study in an attempt to get a pre-experimental measure of task interest. Subjects in the Noncontingent Reward condition were paid $2.00 for their participation in the study. Subjects in the Contingent Reward condition were paid $0.70 for each puzzle correctly solved. Data from Study 1 suggested that this pay rate for the Contingent condition would make comparable the average pay for subjects in the two reward conditions. Subjects were assigned to the Locus of Control and Self-Esteem conditions on the basis of their screening test scores obtained several weeks prior to the actual running of the experiment. There were 16 subjects
REWARD CONTINGENCY AND INTRINSIC MOTIVATION
41
selected in each of the four locus of control-self-esteem combinations (Internal L C - H i g h SE, Internal L C - L o w SE, External L C - H i g h SE, External LC--Low SE). The selection of these subjects from the pools of eligible subjects was quasi-random. The order of experimental treatments involving the various individual differences combinations was randomized among established time periods, and then subjects were chosen on the basis of their availability for these specific times. Half of the subjects in each individual differences condition were randomly assigned to the Contingent Reward condition and the other half to the Noncontingent Reward condition. This created a 2 x 2 x 2 factorial design with eight subjects per cell.
Analysis A multivariate analysis of variance was performed on the data. Univariate analyses of variance were conducted separately for each dependent variable for sources of variance which had a significant multivariate F (p < .05).
Results Table 3 presents the results of the multivariate analysis of variance for Study 2. No source of variance was found to have a significant multivariate F sl;atistic. Therefore, there was no support for any of the hypotheses of Study 2. In order to be certain that a significant (in the univariate sense) effect of Reward Contingency upon the Free Time Spent on Puzzles dependent variable was not hidden by the inclusion of many other dependent variables in the multivariate analysis, a univariate analysis of variance was conducted for this dependent variable. No source of variance significantly affected Free Time Spent on Puzzles. The distribution of values for the Free Time Spent on Puzzles was again markedly bimodal. For the contingent condition 15 subjects spent less than 1 min during the free time period on the task, 11 spent more than 7
TABLE 3 MULTIVARIATE ANALYSIS OF VARIANCE; STUDY 2 Source Reward contingency (A) Locus of control (B) Self-esteem (C) AxB AxC BxC AxBxC
F
p
1.55 0.86 0.60 1.30 1.07 0.58 1.31
.13 .61 .86 .25 .41 .87 .24
42
FARR, VANCE, AND MCINTYRE
min, and only six spent between 1 and 7 min on the task. For the noncontingent condition 12 subjects spent less than 1 min on the task during the free time period, 18 spent more than 7 min, and only two spent between 1 and 7 min on the task. Discussion The absence of any significant effect for reward contingency in Study 2 was contrary to the predictions of Deci (1975). There was no support for the hypothesis that contingent rewards reduce the frequency of occurrence of behavior presumably determined by intrinsic motivation in comparison to noncontingent rewards. The frequency distribution for the dependent variable of free time spent on puzzle was again quite bimodal, replicating this finding from Study 1. The bimodality within the reward contingency conditions again suggested that strong individual differences may be operating with regard to the effects of extrinsic reward upon the behavioral measure of intrinsic motivation. Study 2 made less likely the possibility that either self-esteem or locus of control was the individual differences variable of importance. Furthermore, the frequency data for the free time variable suggested that parametric analyses might not be most appropriate for these data due to their distributional properties. Study 3 presents the results of nonparametric reanalyses of the free time data of Studies 1 and 2, as well as reanalyses of some of Deci's data from previously reported studies (Deci, 1972a,b; Deci et al., Note 1). 1 Also in Study 3 are presented analyses related to possible reinforcement explanations of the findings of Studies 1 and 2. The finding that extrinsic rewards may decrease the occurrence of behaviors presumed to be determined by intrinsic motivation has been interpreted as being anomolous to reinforcement theory by proponents of cognitive evaluation theory (Deci, 1975). However, cognitive evaluation theory has focused upon extrinsic rewards as the sole class of possible reinforcers of these behaviors. It is possible that other factors may serve as reinforcers of such behavior. The solving of the puzzles during the experimental task period might be reinforcing to subjects. If this is true, then subjects who solve many of the puzzles during the task period might be predicted to spend more of the free time period working on the puzzle than those who solved few or none of the puzzles. Data supportive of this prediction would suggest that reinforcement theory is a viable explanation of the phenomenon of interest. Because reinforcement theory represents a more
1 The authors would like to thank Edward Deci for supplying the raw data for these analyses.
REYVARD CONTINGENCY AND INTRINSIC MOTIVATION
43
parsimonious explanation of these data, there may be little need for cognitive evaluation theory if reinforcement theory is supported (Scott, 1976). A related question concerns the effects of extrinsic rewards versus no rewards upon behavior presumed to be determined by intrinsic motivation. Deci (1972b) has suggested that only contingent extrinsic rewards have a deleterious effect upon such behavior, but Calder and Staw (1975b) and Kruglanski et al. (1971), among others, have found that noncontingent extrinsic rewards may also have this effect when compared to noreward conditions. Since the present Studies 1 and 2 did not examine no-reward conditions, the main results of an additional experiment (Vance, Note 2) are described in Study 3. Vance (Note 2) compared contingent extrinsic rewards, noncontingent extrinsic rewards, and no extrinsic rewards for their effects upon behavior presumed to be determined by intrinsic motivation. STUDY 3 Normality o f the Free Time Measure
The frequency distributions for the free time dependent variable for Studies 1 and 2 are presented in Table 4. The Shapiro-Wilk test for normality was performed on each distribution (Ryan, Joiner, & Ryan, 1976). z All distributions were found to deviate significantly from normality (p < .01). The violations of the normality assumption, along with the fact that the two modes in each distribution fell at opposite extremes, strongly suggest that parametric analyses utilizing distribution means are inappropriate fbr these data. Kolmogorov-Smirnov two-sample tests (Siegel, 1956) were performed on the distributions, revealing no signifiCant differences (p > .05, two-tailed test) between the contingently and noncontingently paid groups of either study with respect to the amount of time spent working on the task during the free time period. The raw data for the free time dependent variable from the earlier work of Deci (1972a,b; Deci et al., Note 1) are also shown in Table 4. These data generally show bimodality, similar to the data of Studies 1 and 2. The Shapiro-Wilk test indicated that the normality assumption was violated (p < .01) for all distributions, except for that of the female subjects in the control group of the Deci et al. (Note 1) study (p < . 10 for this condition). Thus, parametric tests were considered inappropriate for these data also. Deci's first study reported here (1972a) consisted of six experimental conditions of which we received data for four. The four conditions for which we have data included a contingent pay condition, a contingent
2 Appreciation is extended to David Wahrenberger of the Department of Statistics, The Pennsylvania State University for his assistance regarding these analyses.
Noncontingent Control Male Subjects Contingent praise Control Female Subjects Contingent praise Control No Pay Contingent Noncontingent
Deci (1972b)
Vance (Note 2)
(Note 1)
e t al.
Deci
Contingent pay Contingent praise Control
Contingent Noncontingent
2
Deci (1972a)
Contingent Noncontingent
Condition
1
Study
TABLE 4
0 0 0 0 1 1 3
10 3 12 11 10
2 0
0 2 0
2 2
1 1
1-1:59
2 6
9 9
12 7 8
15 12
14 8
0-0:59
0 0 1 1 1
0 0
2 1
0 1 0
2 0
1 1
2-2:59
1 2 0 1 1
0 1
2 0
0 1 1
0 0
2 0
3-3:59
0 4 1 0 1
0 2
1 0
0 1 0
1 0
1 0
4-4:59
1 1 1 2 1
1 0
2 0
1 0 1
0 0
0 3
5-5:59
Minutes spent on task during free time period
DISTRIBUTIONS OF THE AMOUNT OF TIME SPENT ON TASK DURING FREE TIME PERIOD
0 1 2 0 3
1 0
1 0
1 0 0
1 0
1 1
6-6:59
4 5 6 8 4
12 7
5 6
2 4 6
11 18
4 10
7-7:8:00
r~
~q
z
z r~
4:*
REWARD CONTINGENCY AND I N T R I N S I C MOTIVATION
45
praise condition, a pay and praise condition, and a control condition. Deci made no predictions concerning the effect of the treatment for the pay and praise condition, so the data for this condition are not included in Table 4 and subsequent analyses. Deci did make specific predictions for the remaining conditions. He predicted that subjects in the contingent pay condition would work less on the task during the free time period than control subjects and that subjects in the contingent praise condition would work more than the control subjects. These data, shown in Table 4, were analyzed by ~he Birnbaum-Hall test for three independent samples (Conover, 1971). The test statistic, which is similar in concept to the Kolmogorov-Smirnov two-sample test statistic, was nonsignificant (p > .05, two-tailed test). Thus, Deci's (1972a) hypotheses were not supported by this analysis of his data. The next study of Deci (1972b) examined the effects of noncontingent monetary reward on intrinsic motivation. Deci hypothesized that there would be no significant difference between subjects who received noncontingent pay and subjects who received no pay with respect to the amount of time spent on the task during the free time period. His data revealed no difference when analyzed by a parametric test. The reanalysis with the Kolmogorov-Smirnov two-sample test again revealed no significant difference between the noncontingent pay and control conditions. However, these data should probably not be taken as support for the cognitive evaluation theory of Deci (1975) because only the null hypothesis has been affirmed (see Calder and Staw, 1975a, for further discussion of this point). These data did provide further evidence that the behavioral measure of intrinsic motivation was bimodally distributed and thus strengthened the argument that nonparametric analyses of these data are more appropriate than parametric ones. The data of Deci et al. (Note 1) were reanalyzed separately by subject sex by means of the Kolmogorov-Smirnov two-sample test. Deci et al. had predicted that contingent praise would enhance the intrinsic motivation for male subjects, but decrease the intrinsic motivation for females. The nonparametric reanalyses gave no support to these hypotheses. Extrinsic R e w a r d versus N o R e w a r d
An additional study (Vance, Note 2), employing the general paradigm of Studies 1 and 2 and of Deci (1972b), was conducted to compare the effects of nonpayment for task performance with both contingent and noncontingent payment. Previous research (Calder & Staw, 1975b; Kruglanski, Friedman, & Zeevi, 1971) has indicated that noncontingent reward may lead to a decrease in behavior presumably determined by intrinsic motivation, relative to a no-payment condition. Thus, inclusion of a nopayment condition would allow the detrimental effect of reward, either
46
FARR, VANCE, AND MCINTYRE
contingent or noncontingent, on free time behavior to clearly manifest itself. The frequency distributions for the free time behavioral measure for this study appear in Table 4. The three distributions were found by the S h a p i r o - W i l k test to be significantly non-normal (p < .01). The t~irnbaum-Hall three-sample test revealed no significant differences among the distributions (p > .05). No support was found for the hypothesis that extrinsic reward decreases the frequency of behavior presumably determined by intrinsic motivation.
Reinforcement Theory Predictions Since individual differences in the amount of free time spent working on the puzzle were large, and since these differences could not be accounted for by the pay contingency variable, the data from Studies 1 and 2 and from Vance (Note 2) were examined for possible effects of task success upon the amount of free time spent on the task. Subjects who successfully solved all four puzzles may have found the task reinforcing, and, therefore, may have worked on the task significantly longer during the free time period than subjects who were not very successful in solving the puzzles. Free time distributions as a function of number of puzzles solved for Studies 1 and 2 and Vance (Note 2) are presented in Table 5. For each study, the number of subjects who solved four puzzles was determined, and then the poor-performance category was established so as to contain approximately the same number of subjects as the good-performance category. A Kolmogorov-Smirnov two-sample test was performed on each set of distributions. A significant difference between the distributions of Study 1 was obtained (p < .01, two-tailed). However, no such difference was found between the distributions in either Study 2 or Vance (Note 2) (p > .05, two-tailed). Finally, an attempt was made to determine if there were performance differences in terms of number of puzzles solved between subjects who subsequently worked on the task less than 1 rain and those who worked 7 min or longer. Behrens-Fisher t tests (Games and Klare, 1967) were performed on the data from Studies 1 and 2, and from Vance (Note 2). The results of these analyses are presented in Table 6. In Study 1, subjects who worked on the task for 7 rain or longer had solved significantly more puzzles than had subjects who worked on the task for less than 1 rain, thus supporting the finding reported in the previous paragraph. There was no significant differences in mean number of puzzles solved as a function of free time spent on the puzzle for the data from Study 2 and Vance (Note 2). These results also agree with those presented above. It may be noted that an optimal analysis of reinforcement theory predictions would have separated subjects into three categories: those receiving no reinforcement (no puzzles correctly solved), those receiving intermittent or partial reinforcement (some puzzles correctly solved), and those receiving continuous reinforcement (all puzzles correctly solved). How-
0-2 4 0-1 4
2
Vance (Note 2)
0-2 4
No, puzzles solved
1
Study
8 10 8 8
13 5
0-0:59
1 1 0 1
0 1
1 - 1:59
1 1 0 0
1 0
2-2:59
0 0 1 0
0 2
3-3:59
0 1 0 2
0 0
4-4:59
0 0 0 1
2 0
5-5:59
1 0 1 0
1 0
6-6:59
Minutes spent on task during free-time period
7 10 4 5
2 11
7-8:00
18 23 14 17
19 19
Total
TABLE 5 DISTRIBUTIONS OF THE AMOUNT OF TIME SPENT ON TASK DURING FREE TIME PERIOD AS A FUNCTION OF NUMBER OF PUZZLES SOLVED
4~
Z
.<
©
Z
g3
g3 ©
H
48
FARR, VANCE~ AND MCINTYRE TABLE 6 BEHRENS--FISHER t-TEsT RESULTS ON NUMBER OF PUZZLES SOLVED FOR EXTREME GROUVS on FREE TIME MEASURE Minutes Spent on Task during Free-Time Period 0-0:59
7-8:00
Study
N
Mean number solved
N
Mean number solved
t
p
1 2 Vance (Note2)
22 27
2.50 2.93
14 29
3.64 3.10
3.89 0.69
<.001 N.S.
33
2.39
18
2.50
0.31
N.S.
ever, in the present Studies 1 and 2 and Vance (Note 2), there was a total of only three subjects who solved no puzzles. Thus, the decision was made to analyze the data as presented above.
GENERAL DISCUSSION The interpretation of the data of Studies 1 and 2, Vance (Note 2), and the reanalyses of earlier work of Deci is obviously dependent upon whether one places primary emphasis on the parametric or nonparametric analyses. The consistent distributional properties shown in Table 4 cast grave doubts that the population is distributed normally, which would be required by the various parametric tests used. The Kolmogorov-Smirnov two-sample test has high power-efficiency (Siegel, 1956) when compared with the t test, and thus would be expected to yield comparable results as the parametric test if the assumptions of the parametric test have been met by the data. The differential results of the two types of analyses suggest that the results of the parametric tests may have been substantially affected by the failure to meet the test's assumptions. If the preference for the nonparametric approach is granted, then the data presented in these studies and reanalyses in general do not support the postulate that contingent extrinsic rewards decrease the frequency of occurrence of behaviors presumably determined by intrinsic task motivation. When results of research are negative, it is important to examine the research methods carefully to ascertain if methodological problems have prevented the emergence of significant results. In the present research the general methodology was modeled after that of Deci (1972b) in order to replicate his procedures. The important experimental manipulation was whether the money in the contingent reward condition was indeed viewed as contingent upon task performance. Manipulation check data revealed that, in Studies 1 and 2 and Vance (Note 2), subjects in the contingent reward condition saw the money as being a significantly more important
REWARD
CONTINGENCY
AND INTRINSIC
MOTIVATION
49
reason for working hard and performing well on the task than the subjects in the noncontingent condition. This suggests that the contingency manipulation was sufficiently powerful. It appears unlikely that any other procedural factor could have obviated the results. If methodological causes for the negative results can be ruled out, then the conceptual framework for the research must be examined. The rationale and variable operationalization for these studies follow clearly from Deci (1975). Misinterpretation of Deci's theoretical statements seems unlikely. One can question the conceptual adequacy of the theoretical arguments. Kruglanski (1975) has recently noted inherent ambiguities and inconsistencies in the internal-external distinction proposed by Deci (1975). For example, if an individual works hard on a task which he or she describes as interesting, is the cause of the task motivation properly classified as internal (due to personal enjoyment and satisfaction) or external (due to the task)? A related issue is the question of the attribution of :motivation or behavior to a single cause or single category of causes (e.g., internal). Behavior is generally complex and multiply determined. Why should task motivation be seen as caused by only internal o r only external factors and not by both? The data of Studies 1 and 2 offered no support for the decrease in intrinsic attribution hypothesized to occur when extrinsic rewards are contingent upon task performance, although there was an increase in the attribution of behavior to the money received in this condition. This suggests that attributions are not necessarily related to each other in a zero-sum fashion. An increase or decrease in importance of a causal element to a particular behavioral situation does not necessarily lead to the decrease e r increase in attributed importance of another causal element. The frequency data presented in Table 4 and the discussion above suggest that there may be important individual differences in the response to contingent extrinsic rewards. Study 2 attempted to isolate some of these individual difference variables but was not successful. Neither selfesteem nor locus of control appeared to be important moderators of the effect under investigation. It should be noted that the present research only examined possible linear effects of these individual difference variables, These variables might have nonlinear effects not investigated here. Also in Study 2 a post hoc analysis of the pre-experimental questionnaire item concerning preference for puzzle solving as a leisure activity failed to yield any significant findings. There was no difference between those subjects expressing strong preference for puzzle solving and those not preferring such leisure activity in terms of their comparative responses to contingent and noncontingent rewards. This seems particularly damaging to the arguments of Deci (1975). He would predict that the
50
FARR, VANCE, AND MCINTYRE
intrinsic motivation of individuals with strong preference for the task activities would be more negatively affected by contingent rewards than that of individuals with weak preferences. It should be noted that task preference was not explicitly used as an independent variable in this research and that any conclusions about this variable are especially tenuous. Although Deci (1972b) has argued that extrinsic rewards must be contingent in order to adversely affect behaviors presumably intrinsically motivated, other researchers (e.g., Calder & Staw, 1975b; Kruglanski et al., 1971) have found that noncontingent extrinsic rewards may also reduce the frequency of such behavior and decrease the favorability of related attitudes. If extrinsic reward, regardless of contingency, differs from nonreward with regard to its effect upon behavior, then the design of Studies 1 and 2 is not optimal because it did not include a nonreward condition. The previously unpublished data of Vance (Note 2) presented earlier suggested that reward conditions did not differ from nonreward conditions regarding the effects upon the frequency of behavior presumed to be dependent upon intrinsic motivation. Thus, the argument that no differences were found in our Studies 1 and 2 between contingent and noncontingent reward conditions (regarding intrinsic motivation) because any extrinsic reward adversely affects such motivation in comparison to nonreward conditions appears not to be viable. Data from Studies 1 and 2 and Vance (Note 2) were examined for possible reinforcement explanations of the free time behavior results. The solving of a puzzle may be considered reinforcing to the subject. Thus, those subjects who solved all four puzzles had a different reinforcement history regarding the task than those who solved none or few of the puzzles. If the more frequently reinforced subjects also tended to be those who spent large amounts of free time on the task, then the behavior in question would appear to follow rather simple conditioning principles and would require no special postulates about its determinants. The data presented in Tables 5 and 6 suggest that in Study 1 the reinforcement explanation has some support, but that in Study 2 and Vance (Note 2) it has no support. The failure to replicate the reinforcement effect makes the overall results equivocal in interpretation. However, the extensive history of successful reinforcement explanations of behavior suggests that more explicit attention be paid to this theoretical framework. Researchers in this area may have focused too narrowly on the extrinsic reward variable without considering other possible sources of reinforcement for the subjects (Scott, 1976). It can be argued that the above results, which give some support to a reinforcement interpretation of the data, also are supportive of cognitive evaluation theory. Task success should lead to increases in feelings of
REWARD CONTINGENCY AND INTRINSIC MOTIVATION
51
competence antd self-determination which would enhance intrinsic motivation according to cognitive evaluation theory. However, cognitive evaluation theory also makes other predictions which were not supported by the data of the present research. In addition, reinforcement theory offers a more parsimonious explanation of this particular result. It appears at this time that general statements about the effects of contingent (and noncontingent) extrinsic rewards either upon behavior presumably determined by intrinsic motivation or attitudes related to the construct are not warranted. This is not to say that it should be presumed that there are no such effects. Data have been presented which suggest that contingent extrinsic rewards may decrease the frequency of behavior presumably determined by intrinsic motivation and other data suggest that such rewards may enhance this behavior (see reviews by Kruglanski, 1975, and Deci, 1975). What appears to be evolving is a need for more complex models than those previously presented. Types of variables which need to be included in a complete model concerned with the effects of rewards upon motivation include individual difference variables, task variables, and situational variables. It is doubtful whether theoretical or applied progress will be made unless these factors are considered in the research in this area. REFERENCES Brief, A. P., & A]dag, R. J. Employee reactions to job characteristics; a constructive replication. Journal of Applied Psychology, 1975, 60, 182-186. Calder, B. J., & Staw, B. M. The interaction of intrinsic and extrinsic motivation: some methodological notes. Journal of Personality and Social Psychology, 1975, 31, 76±80 (a). Calder, B. J., & Staw, B. M. Self-perception of intrinsic and extrinsic motivation. Journal of Personality and Social Psychology, 1975, 31, 599-605 (b). Conover, W. J. Practical nonparametric statistics. New York: Wiley, 1971. Dachler, H. P., & lVlobley, W. H. Construct validation of an expectancy-instrumentalitytask goal model[ of work motivation: some theoretical boundary conditions. Journal of Applied Psychology Monograph, 1973, 58, 397-418. deCharms, R. Personal causation. New York: Academic Press, 1968. Deci, E. L. Effects of externally mediated rewards on intrinsic motivation. Journal of Personality and Social Psychology, 1971, 18, 105-115. Deci, E. L. Intrinsic motivation, extrinsic reinforcement, and inequity. Journal of Personality and Social Psychology, 1972, 22, 113-120 (a). Deci, E. L, The effects of contingent and non-contingent rewards and controls on intrinsic motivation. Organizational Behavior and Human Performance, 1972, 8, 217-229 (b). Deci, E. L. Intrinsic motivation.. New York: Plenum. 1975. Deci, E. L. Notes on the theory and metatheory of intrinsic motivation. Organizational Behavior and Human Performance, 1976, 15, 130-145. Deci, E. L., Cascio, W. F., & Krusell, J. Cognitive evaluation theory and some comments on the Calder and Staw critique. Journal of Personality and Social Psychology, 1975, 31, 81-85.
52
FARR, VANCE, AND MCINTYRE
Farr, J. L. Task characteristics, reward contingency, and intrinsic motivation. Organizational Behavior and Human Performance, 1976, 16, 294-307. Feingold, B. D., & Mahoney, M. J. Reinforcement effects on intrinsic interest: undermining the overjustification hypothesis. Behavior Therapy, 1975, 6, 367-377. Games, P. A., & Klare, G. R. Elementary statistics: data analysis for the behavioral sciences. New York: McGraw-Hill, 1967. Ghiselli, E. E. Explorations in managerial talent. Pacific Palisades, California: Goodyear Publishing Company, 1971. Greene, D., & Lepper, M. R. Effects of extrinsic rewards on children's subsequent intrinsic interest. Child Development, 1974, 45, 1141-1145. Hackman, J. R., & Lawler, E. E. Employee reactions to job characteristics. Journal of Applied Psychology Monograph, 1971, 55, 259-286. Hamner, W. C., & Foster, L. W. Are intrinsic and extrinsic rewards additive: a test of Deci's cognitive evaluation theory of task motivation. Organizational Behavior and Human Performance, 1975, 14, 398-415. Jablonsky, S. F., & DeVries, D. L. Operant conditioning principles extrapolated to the theory of management. Organizational Behavior and Human Performance, 1972, 7, 340-358. Kruglanski, A. W. The endogenous-exogenous partition in attribution theory. Psychological Review, 1975, 82, 387-406. Kruglanski, A. W., Alon, S., & Lewis, T. Retrospective misattribution and task enjoyment. Journal of Experimental Social Psychology, 1972, 8, 493-501. Kruglanski, A. W., Friedman, I., & Zeevi, G. The effects of extrinsic incentive on some quality aspects of task performance. Journal of Personality, 1971, 39, 606-617. Lawler, E. E. Motivation in work organizations. Monterey, California: Brooks/Cole, 1973. Lepper, M. R., & Greene, D. Turning play into work: effects of adult surveillance and extrinsic rewards on children's intrinsic motivation. Journal of Personality and Social Psychology, 1975, 31, 479-486. Lepper, M. R., Greene, D., & Nisbett, R. E. Undermining children's intrinsic interest with extrinsic rewards: a test of the"overadjustification" hypothesis. Journal of Personality and Social Psychology, 1973, 28, 129-137. Nord, W. R. Beyond the teaching machine: the neglected area of operant conditioning in the theory and practice of management. Organizational Behavior and Human Performance, 1969, 4, 375-401. Porter, L. W., & Lawler, E. E. Managerial attitudes and performance. Homewood, Illinois: Dorsey Press, 1968. Ross, M. Salience of reward and intrinsic motivation. Journal of Personality and Social Psychology, 1975, 32, 245-254. Rotter, J. B. Generalized expectancies for internal versus external control of reinforcement. Psychological Monographs, 1966, 80, 1 (Whole No. 609). Ryan, T. A., Jr., Joiner, B. L., & Ryan, B. F. Minitab H: reference manual. North Seituate, Mass.: Duxbury Press, 1976. Scott, W. E., Jr. The effects of extrinsic rewards on "intrinsic motivation": a critique. Organizational Behavior and Human Performance, 1976, 15, 117-129. Siegel, S. Nonparametric statistics for the behavioral sciences. New York: McGraw-Hill, 1956. Smith, P. C., Kendall, L. M., & Hulin, C. L. The measurement of satisfaction in work and retirement. Chicago: Rand-McNally, 1969. Vroom, V. H. Work and motivation. New York: Wiley, 1964. White, R. W. Motivation reconsidered: the concept of competence. Psychological Review, 1959, 66, 297-333.
REWARD CONTINGENCY AND INTRINSIC MOTIVATION
53
REFERENCE NOTES 1. Deci, E. L., Cascio, W. F., & Krusell, J. Sex differences, positive feedback, and intrinsic motivation. Paper presented at the Eastern Psychological Association Convention, Washington, D.C., 1973. 2. Vance, R. J. The constructs of intrinsic and extrinsic motivation in conflicting models of work motivation. Unpublished Master's thesis, The Pennsylvania State University, 1977. RECEIVED: August 12, 1976