Journal of Educational Psychology2001, Vol. 93, No. 1, 23-34 Copyright 2001 by the American Psychological Association, Inc.

O022-0663/01/$5.00 DOI: 10.I037//0022-0663.93.1.23

Between- and Within-Domain Relations of AcademicMotivation Among Middle and High School Students:

Self-Efficacy, Task-Value, and Achievement Goals

Mimi BongUniversity of South Carolina

The author used confirmatory factor analysis to examine between-domain relations of self-efficacy,task-value, and achievement goal orientations among 424 Korean middle and high school students. AHmotivational constructs demonstrated strong subject specificity in both age groups. Strengths of between-domain associations differed substantially by individual constructs. Performance-approach and perfor-mance-avoidance goals were highly correlated across domains, whereas task-value and mastery goalswere more distinct across domains. Self-efficacy perceptions were moderately correlated across subjects.High school students' academic motivation was more differentiated than that of middle school students.Within-domain interrelations among these motivation constructs were generally consistent with previousresearch. More important, consistent patterns of relations were observed in four different academicdomains within each age group.

Contemporary academic motivation research tends to emphasizethe distinctiveness of students' motivational orientation acrossdifferent situations (Weiner, 1990). The emphasis on context spec-ificity is translated into motivation constructs being assessed inreference to particular academic tasks, activities, or domains ofinterest. Such assessment practices considerably improved accu-racy of behavioral prediction by accounting for differences inindividuals' perceptions across diverse situations (Bandura, 1997;Mischel, 1977; Pajares, 1996). Despite this apparent benefit, as-sessing motivation in reference to specific situations makes itdifficult to conjecture about the nature of relations between studentmotivation in different contexts. Patterns of interrelations amongdifferent motivation constructs observed in a particular domainalso may or may not emerge in other academic domains. Meece(1994) aptly observed this when she mentioned that "although thedomain specificity of these measures may increase their predictivevalidity (Assor & Connell, 1992, as cited in Meece, 1994), it is notclear how well the findings will generalize to other subject areas"(p. 37).

The present research pursued two primary purposes in light ofthese observations. The first objective was to investigate thebetween-domain relations of student motivation. This study exam-ined how motivation constructs such as self-efficacy, task-value,or achievement goal orientations in one subject domain relate tothe same construct assessed in the contexts of different academicsubjects. Specific school subjects were chosen as the basic mea-surement level because they are known to act as principal psycho-

An earlier version of this article was presented at the annual meeting ofthe American Educational Research Association, New Orleans, Louisiana,April 2000.

Correspondence concerning this article should be addressed to MimiBong, Department of Educational Psychology, 135 Wardlaw Hall, Univer-sity of South Carolina, Columbia, South Carolina 29208. Electronic mailmay be sent to mimibong@sc.edu.

logical organizers of school-related cognition and affect (Gottfried,1985; Marsh & Yeung, 1996). The second objective of this studywas to examine the within-domain relations of these motivationalconstructs. Interrelations among self-efficacy, task-value, andachievement goal orientations were investigated in four differentacademic domains. It was of particular interest to determinewhether there is any notable difference in these construct relationsas a function of domain. The study also allowed comparison offindings across middle and high school years.

Between-Domain Relations of Motivation Constructs

Among a host of academic motivation constructs, the issue ofcross-domain association has been most frequently probed withacademic self-concept (e.g., Byrne & Shavelson, 1986; Marsh,1990, 1992; Marsh, Byrne, & Shavelson, 1988). With confirma-tory factor analytic techniques, researchers have shown that stu-dents' self-evaluations contain strong subject-specific compo-nents. Yet these subject-specific self-concepts were highlycorrelated within the broader boundaries of verbal and math do-mains, attesting to the hierarchical nature of academic self-concept. Unfortunately, it is difficult or even dangerous to applythese findings directly to other ostensibly related constructs with-out empirical testing (Bong, 1996). For example, whereas self-concepts are clearly divided along the line of verbal and mathdomains, self-efficacy beliefs in these two areas are often highlycorrelated (Bong, 1997; Marsh, Walker, & Debus, 1991). Al-though whether and how these two constructs differ are beyond thescope of the present investigation (interested readers should referto Bong & Clark, 1999), this demonstrates the need to study thebetween-domain relations separately for each motivation construct(Gottfried, 1985).

Between-Domain Relations of Self-Efficacy

Academic self-efficacy refers to students' beliefs about theircapabilities to perform given academic tasks at designated levels

23

24 BONG

(Schunk, 1991). The standard method used in self-efficacy re-search is to assess students' confidence toward specific tasks andexamine how well these perceptions predict performance on thevery tasks. The task-specific self-efficacy appraisal methods some-times leave researchers with the false impression that beliefs ofself-efficacy are only relevant in the context of carrying out asingle minute task. Quite the contrary, however, self-efficacyresearchers have acknowledged that one can face a wide range oftasks and situations with comparable self-efficacy and that percep-tions of efficacy developed toward a particular task may generalizeto other tasks of interest (Bandura, 1997; Pajares, 1996; Schunk &Swartz, 1993; Smith, 1989).

Bong (1997) provided evidence that students' self-efficacyjudgments contain strong subject-specific components. As was thecase with self-concept, some of these subject-specific efficacyperceptions were highly correlated, showing a certain degree ofbetween-domain generalization. According to Bandura (1997),individuals are likely to generalize their self-efficacy when differ-ent activities share similar subskills, when skills in dissimilardomains are developed concurrently, when generic self-regulatorycapabilities are acquired, when powerful personal triumphs areexperienced, or when commonalties across diverse activities andsituations are cognitively structured. Skills taught in differentschool subjects often share similar subskills, especially when theyare dependent on strong linguistic or quantitative competencies.Most skill development in school also takes place concurrently andis in large part based on common self-regulatory capabilities. Thepresent study sought to replicate Bong's findings on the hierarchi-cal nature of self-efficacy with different measures.

Between-Domain Relations of Task Value

Eccles and her colleagues defined task value as an incentive forengaging in different tasks (Eccles, Wigfield, Harold, & Blumen-feld, 1993; Wigfield & Eccles, 1992; Wigfield et al., 1997).Interest in and perceived importance and usefulness of the taskscomprise important dimensions of subjective task value. Researchfound that children as young as first grade distinguish their per-ceptions of task value toward different activity domains such asreading, math, music, and sport (Eccles et al., 1993; Wigfield et al.,1997). Although these results are certainly indicative, there is notenough evidence in the literature that permits sound speculationregarding the between-domain associations of academic task valueamong adolescents. There are several reasons for this. First, re-searchers have been more interested in the internal composition ofthe construct such as interrelations among importance, usefulness,interest, and cost. Second, investigations have been conductedmostly in English or math, seldom including other subject areas.Third, studies that did assess task values across multiple domainstypically involved younger children. Hence, it is difficult to gen-eralize these results to middle and high school students. In alongitudinal study with young adolescents, Wigfield, Eccles, MacIver, Reuman, and Midgley (1991) found that students' liking ofmath and English correlated at .07, whereas their self-concept ofability correlated at .37. This suggests that the cross-domain asso-ciations of task value may be weaker than those of otherconstructs.

Between-Domain Relations of AchievementGoal Orientations

The cross-domain relations have not been dealt with adequatelywith respect to various achievement goal orientations. Achieve-ment goals commonly refer to reasons for engaging inachievement-oriented behaviors (Ames, 1984; Dweck, 1989;Nicholls, 1984; Urdan & Maehr, 1995). Students are said todemonstrate learning or mastery goals when they undertake chal-lenging tasks for the sake of learning and, by doing so, improvingtheir competence. In contrast, students pursue performance or egogoals when they are overly conscious about how others evaluatethem. Those with performance-approach orientations try to vali-date their superior ability, whereas those with avoidance orienta-tions strive to conceal their incompetence. Most research onachievement goals to date has been occupied with effects andrelationships of different goals within a single academic context.

Duda and Nicholls's (1992) study is one of few studies thatexamined the between-domain associations of goals. They as-sessed high school students' task, ego, and work avoidance goalsacross classroom and sport. Students displayed similar goal orien-tations in these two areas. More interesting, cross-domain relationswere considerably stronger with achievement goals (r = .51 to .67)than with perceived ability (r = .32) or satisfaction/enjoyment(r = .15). The investigators argued that goals should generalizemore than perceived ability or satisfaction/enjoyment becausegoals reflect "the type of quality of one's personal criteria ofsuccess" (p. 291). Although this study showed some generality instudents' goals, its distinction between contexts of schoolwork andsport is nonetheless too broad. Consequently, its results cannot tellmuch about the associations of achievement goals across differentacademic domains. Goal adoption is influenced by students' viewsof ability as well as salient evaluation criteria (Dweck, 1989;Nicholls, 1984). Compared with younger children, older studentstend to endorse differentiated conceptions of ability. School envi-ronments in which they function also emphasize normative supe-riority. Thus, strong associations of performance goal orientationsacross different academic contexts may be expected.

Age Differences in Between-Domain Relations

It is generally agreed that even very young children differentiatetheir beliefs of competence and task value in different domains offunctioning (e.g., Eccles et al., 1993; Marsh, Craven, & Debus,1991; but see Harter & Pike, 1984). Studies with middle and highschool students often assess students' motivational orientationstoward specific academic domains, with an understanding that theyhold more or less differentiated perceptions toward these areas.What researchers do not yet know is how these specific beliefsrelate to each other and how such relations change with age. In thepresent study, high school students were hypothesized to demon-strate relatively more distinct motivational beliefs compared withmiddle school students. They have more academic experience,which can help them better attune to the demands and possibilitiesof each domain, which would in turn contribute to finer differen-tiation between domains. In particular, as a result of their heavierconcern with future college majors and career choices, high schoolstudents are believed to hold more differentiated task-value beliefscompared with middle school students.

BETWEEN- AND WITHIN-DOMAIN MOTIVATION 25

Within-Domain Relations of Motivation Constructs

Within-domain relations among motivation constructs have of-ten been subjected to empirical interrogation. Achievement goalsof mastery are typically positively related to the sense of self-efficacy (Meece, Blumenfeld, & Hoyle, 1988; Middleton, Kaplan,& Midgley, 1998; Middleton & Midgley, 1997; Roeser, Midgley,& Urdan, 1996; Skaalvik, 1997; Turner, Thorpe, & Meyer, 1998).Mastery goals also work to increase intrinsic motivation (Elliot &Church, 1997; Elliot & Harackiewicz, 1996). Perceived compe-tence and intrinsic motivation are positively correlated (Berndt &Miller, 1990; Feather, 1988; Meece, Wigfield, & Eccles, 1990). Inshort, the positive interdependence among mastery goals, self-efficacy, and task value has been well documented. Performancegoals' links to other motivation constructs are more problematic.Some studies have reported positive relations between perfor-mance goals and mastery goals (e.g., Roeser et al., 1996), whereasothers have reported negative relations (e.g., Turner et al., 1998).Noting this inconsistency, researchers have demonstrated that per-formance orientation can be reliably differentiated into approachand avoidance components (Elliot & Harackiewicz, 1996; Middle-ton & Midgley, 1997; Skaalvik, 1997).

Relations of performance orientations to other constructs havebecome somewhat clearer with this distinction. However, with theexception of a positive correlation between performance-approachand performance-avoid goals, these two performance goals' rela-tions with other adaptive motivational orientations have been lessthan unequivocal. In general, performance-approach goals demon-strate positive (Elliot & Church, 1997; Skaalvik, 1997) to nonsig-nificant relations (Middleton & Midgley, 1997) with mastery goalsand self-efficacy. Performance-avoidance goals demonstrate non-significant (Skaalvik, 1997) to negative relations (Elliot & Church,1997; Middleton & Midgley, 1997). This study aimed to providefurther empirical evidence on the within- and between-constructrelations of achievement goals, along with consistency of theserelations across domains and school levels.

The present investigation contributes to the current academicmotivation research in several ways. More specifically, it candemonstrate (a) the degree of between-domain associations ofpopular and important motivation constructs, (b) the nature ofwithin-domain relations of these constructs, (c) stability of theirinterrelations across different academic areas, and (d) potentialage-related differences in the generality of and interrelations be-tween academic motivation constructs.

during regular classroom hours. They were assured of confidentiality oftheir responses.

Measures

The present study used scales that are well established by previousresearch. All measures were assessed with respect to Korean, English,mathematics, and science. Items were strictly parallel across the fouracademic subjects. Students rated each statement on a response scaleranging from 1 (not at all true) to 5 (very true) throughout the survey.

Self-efficacy. Subject-level academic self-efficacy items were adaptedfrom the Patterns of Adaptive Learning Survey (PALS; Middleton &Midgley, 1997; Roeser et al., 1996) and the Self-Efficacy subscale of theMotivated Strategies for Learning Questionnaire (MSLQ; Pintrich & DeGroot, 1990). The five self-efficacy items were "I can master even thehardest material in [a specific subject] if I try," "I can do almost all thework in [a specific subject] if I don't give up," "I'm certain that I can doan excellent job on the problems and tasks assigned for [a specific subject]class," "I know that I will be able to learn the material for [a specificsubject] class," and "I'm confident that I will receive a good grade in [aspecific subject] this semester."

Task value. As in previous research (e.g., Berndt & Miller, 1990;Pokay & Blumenfeld, 1990), task value was operationalized as encom-passing perceived importance, perceived usefulness, and intrinsic interestin the subject. Items included were "I think what I learn in [a specificsubject] class is important," "I think [a specific subject] is a useful subject,"and "I find [a specific subject] interesting."

Achievement goals. Orientations toward mastery, performance-approach, and performance-avoidance goals were assessed with three itemseach. Items were adapted from the PALS. Mastery goal items were "I likeproblems and tasks that I can learn from during [a specific subject] class,even if I make a lot of mistakes," "The main reason why I study [a specificsubject] is because I like it,"1 and "In [a specific subject], I like problemsand materials the best that really make me think." Performance-approachgoal items were "I feel good if I'm the only person who can answer theteacher's question in [a specific subject] class," "I would like to show my[specific subject] teacher that I am smarter than the other students," and "Ifeel successful in [a specific subject] when I get better grades than others."Those for the performance-avoidance goal were "The reason I study [aspecific subject] is so the teacher doesn't think that I know less than othersin my class," "One of my main goals in [a specific subject] class is to avoidlooking like I'm stupid or I do worse than others in my class," and "I worryabout doing worse than the other students in my class."

Overview of Data Analysis Strategy

Confirmatory factor analysis (CFA) provides an effective means totest both the between- and within-domain interrelatedness of motivation

Method

Participants and Procedures

Four-hundred twenty-four students (50% girls) from three middleschools and two high schools in Seoul and Kyung-gi Province (in thevicinity of Seoul, Korea) participated. Korea has a 6-3-3 system fromelementary to high school. Therefore, all Korean middle and high schoolsoffer 3 years of schooling. There were 229 middle school students (48%boys, 52% girls; 49% freshmen, 49% juniors, 2% seniors) and 195 highschool students (53% boys, 47% girls; 54% freshmen, 46% juniors). Veryfew middle and high school seniors participated because senior years aretypically devoted to preparing for important nationwide entrance exami-nations. Data were collected as part of a larger research project on schoolinformation literacy. Students completed the motivation questionnaires

1 Although this item is consistent with the present study's conceptualdefinition of achievement goals (i.e., reasons for engaging in achievement-oriented behaviors), it may nonetheless appear to overlap with one of thetask-value items (i.e., intrinsic interest). Because item overlap works toinflate construct relations, 8 additional analyses were performed (i.e., 4domains X 2 samples) with this particular item loading on the task-valuefactor instead of the mastery goal factor. Goodness-of-fit indexes of thesesubsequent models were all uniformly lower than those of the originalmodels. Correlation coefficients among factors stay essentially the samewith minor fluctuation. The only noticeable difference was in the relationsbetween the task-value and the mastery goal factors of the middle schoolsample. However, these relations became stronger, not weaker, when theparticular item was included as a task-value rather than a mastery goalitem. Together, these results support the initial conceptualization of thisitem as a mastery goal orientation measure.

26 BONG

constructs. Each survey item functioned as an indicator and was hy-pothesized to load on the only factor it was intended to measure. Toprobe the cross-domain associations of each motivation construct, abasic first-order CFA model with four subject-specific latent factorswas fitted to the data. When this model demonstrated acceptable fit withsubstantial first-order factor correlation, two competing higher ordermodels were imposed (see Figure 1). Decisions regarding the absoluteand relative effectiveness of CFA models were made on the basis ofmultiple goodness-of-fit indexes as well as model parsimony. All CFAswere performed separately with middle and high school samples todetermine age-related variations. Therefore, testing the cross-domainrelations could involve up to 30 analyses (i.e., 5 constructs X 3models X 2 samples) and their post hoc modifications (if necessary).

For within-domain relations of motivation, a CFA model with correlatedmotivation factors was fitted within the context of each academic domain.After goodness-of-fit indexes were examined, patterns of factor intercor-relation were compared across domains for consistency. Again, all CFAswere performed separately with middle and high school samples to dis-cover age-related discrepancy. Testing the within-domain relations thusinvolved 8 analyses (i.e., 4 domains X 2 samples). All CFAs were per-formed with the EQS program (Bentler, 1992).

ResultsTable 1 reports descriptive statistics. All scales demonstrated

acceptable reliability with standardized coefficient alphas rangingabove .70 (Mdn = .80). Zero-order correlation coefficients amongmeasures are presented in Table 2.

Between-Domain CFA

First-order CFA. Because items with parallel wording wereused across the four school subjects, correlated uniquenesses(CUs) were incorporated to more accurately estimate constructrelations (Marsh, Byrne, & Yeung, 1999). In almost all models,adding CUs between parallel items considerably improved themodel fit. In the case of academic self-efficacy, four additionalCUs (i.e., between the first and second indicators across the foursubjects) were needed to achieve satisfactory model fit with bothmiddle and high school samples. These items dealt with overcom-ing difficulties with effort and persistence.

Three a priori CFA models were posited (see Figure 1). ModelA was a basic first-order factor structure where each indicator

loaded on a single factor and where all factors were presumed tobe correlated. The Bentler-Bonett nonnormed fit index (NNFI),comparative fit index (CFI), and magnitude of residuals wereconsidered along with the chi-square statistics in determining themodel fit. As Table 3 presents, Model A demonstrated very goodfit to the empirical data with all motivational constructs and withboth middle and high school samples. Most chi-square values hadprobability levels greater than .05, indicating that the hypothesizedmodel did not differ significantly from the empirical data. Excep-tions were academic self-efficacy (bothps < .001) and high schoolperformance-avoidance goal models (p < .05). But even in thesemodels, ratios of chi-square values to their degrees of freedomwere satisfactory (2.24 being the largest). Also, all NNFI and CFIvalues were well above .90, with magnitude of residuals rangingbelow .05 {Mdn = .033).

As can be seen from Table 4, none of the first-order factorcorrelations approached 1, adding to the multidimensionality orsubject specificity of these motivational constructs. With the mid-dle school sample, median values of correlation coefficients were.55 for self-efficacy, .46 for task value, .47 for mastery goal, .67for performance-approach goal, and .67 for performance-avoidance goal. Median coefficients with the high school samplewere .42, .25, .13, .52, and .51, respectively, in the same order.Overall, magnitude of factor correlation tended to decrease in thehigh school sample. In particular, whereas middle school students'subject-specific mastery goal factors were all significantly corre-lated, with the high school sample, only the correlations betweenKorean and English and between math and science factors reachedsignificance. Correlations of mastery goal factors in verbal sub-jects (i.e., Korean and English) with those in quantitative subjects(i.e., math and science) all dropped to nonsignificance. Note-worthy reduction in factor correlations for the high school samplewas also observed with task-value factors. Compared with those inthe middle school sample, the correlations between Korean andmath and between Korean and science task-value factors fellsubstantially in the high school sample.

Higher order CFA. Because Model A demonstrated accept-able fit, and significant correlations among first-order factorswere obtained with most motivation constructs, higher orderstructures were subsequently imposed. Model B specified ver-

Model A Model B Model C

Figure 1. Structures of first- and second-order cross-domain confirmatory factor analysis models. Quanti =quantitative.

BETWEEN- AND WITHIN-DOMAIN MOTIVATION 27

Table 1

Descriptive Statistics of Scales

Scale

Self-efficacyKoreanEnglishMathematicsScience

Task valueKoreanEnglishMathematicsScience

Mastery goalKoreanEnglishMathematicsScience

Performance-approach goalKoreanEnglishMathematicsScience

Performance-avoidance goalKoreanEnglishMathematicsScience

(n

M

3.433.373.403.31

3.353.583.313.42

3.173.203.193.20

3.193.213.173.03

2.742.802.842.81

Middle

Boys= 109)

SD

0.860.890.960.96

0.880.940.881.03

1.000.990.931.09

0.970.890.940.95

0.940.891.001.03

school

Girls(« =

M

3.483.433.073.06

3.423.712.953.18

3.383.302.803.01

3.323.212.872.94

2.312.532.422.44

120)

SD

0.820.981.011.01

0.850.921.041.13

0.891.101.071.10

1.021.091.171.12

0.960.990.930.97

("

M

3.193.453.433.46

2.873.523.243.55

2.923.033.353.33

3.002.913.243.10

2.272.422.592.47

High

Boys= 103)

SD

0.791.041.081.01

0.921.021.191.03

0.891.021.171.11

0.991.061.101.08

0.890.921.040.99

school

(«

M

2.933.153.163.12

2.673.372.903.21

2.882.803.083.13

2.642.812.792.78

2.122.472.522.48

Girls= 92)

SD

0.930.960.920.92

0.880.801.020.93

0.880.931.010.94

0.951.111.061.07

0.750.930.890.97

a

.86

.90

.91

.91

.81

.75

.80

.85

.73

.81

.83

.84

.71

.76

.79

.78

.77

.73

.75

.80

Table 2Zero-Order Correlation Coefficients Among Measures

Measure 1 10 11 12 13 14 15 16 17 18 19 20

1. Korean self-efficacy2. English self-efficacy3. Math self-efficacy4. Science self-efficacy5. Korean task value6. English task value7. Math task value8. Science task value9. Korean mastery goal

10. English mastery goal11. Math mastery goal12. Science mastery goal13. Korean performance-approach goal14. English performance-approach goal15. Math performance-approach goal16. Science performance-approach goal17. Korean performance-avoidance goal18. English performance-avoidance goal19. Math performance-avoidance goal20. Science performance-avoidance goal

.39

.27

.40

.58

.23

.13

.24

.57

.25

.10

.25

.44

.22

.17

.23

.02

.02

.04

.04

.47—.43.39.25.69.33.25.22.68.24.17.22.51.20.08

- .06.00

-.02- .03

.54

.48—.59.05.24.74.48

- .03.17.69.51.12.27.53.35

- .08- .02

.10

.10

.58

.44

.59—.14.16.43.71.13.23.33.72.09.14.28.45.03

- .03.14.19

.63

.32

.34

.42—.27.06.11.68.27.01.12.35.15.03.11.05.02

- .05- .10

.30

.69

.38

.37

.40—.17.17.19.67.19.06.17.44.12.06

- .05.12.02

- .03

.38

.36

.71

.49

.36

.37—.44

- .08.16.69.40.09.23.51.25

- .01- .00

.07

.08

.40 .57

.34 .27

.36 .30

.74 .36

.38 .57

.35 .24

.45 .29— .31.11 —.20 .27.42 - .04.76 .18.06 .35.06 .06.25 - .03.41 .05

- .05 - .06- .05 - .12

.12 - .10

.11 - .13

.24

.70

.35

.37

.29

.62

.32

.32

.36

.13

.14

.21

.49

.05

.05

.01- .05

.00- .01

.39

.25

.67

.53

.31

.26

.66

.40

.38

.32—.48.11.22.57.31

- .02- .01- .04

.05

.40 .54 .45 .42 .44

.40 .31 .58 .36 .39

.49 .35 .45 .68 .43

.73 .38 .47 .43 .65

.37 .46 .40 .34 .38

.41 .17 .45 .27 .27

.49 .17 .30 .49 .33

.70 .20 .28 .23 .50

.45 .46 .28 .31 .33

.46 .26 .57 .26 .33

.57 .23 .34 .57 .39— .26 .36 .37 .58.05 — .58 .53 .54.07 .56 — .56 .61.30 .48 .50 — .64.52 .41 .44 .67 —.03 .20 .12 .15 .22

- .03 .10 .23 .18 .18.04 .06 .13 .28 .33.09 .05 .18 .29 .45

.01 .08 .18 .13

.14 .16 .19 .15

.00 .09 .26 .20

.05 .01 .20 .21

.09 .20 .20 .12

.03 .16 .20 .12

.07 .08 .27 .15

.08 .08 .21 .24

.14 .14 .16 .12

.16 .16 .19 .14- .05 .05 .22 .18

.04 .06 .24 .22

.15 .17 .16 .21

.20 .32 .27 .21

.13 .24 .30 .23

.22 .19 .32 .40— .62 .49 .58.58 — .57 .52.44 .43 — .68.47 .44 .69 —

Note. Middle school sample is shown (n = 229) above the diagonal; high school sample is shown (n = 195) below the diagonal. Coefficients greaterthan .13 in absolute value are significant ntp < .05 for the middle school sample. Coefficients greater than .14 in absolute value are significant atp < .05for the high school sample.

28 BONG

Table 3Goodness-of-Fit Indexes ofBetween-Domain Confirmatory Factor Analysis Models

Model

Self-efficacyModel AModel BModel C

Task valueModel AModel BModel C

Mastery goalModel AModel BModel C

Performance-approach goalModel AModel BModel C

Performance-avoidance goalModel AModel BModel C

x2

290.68292.00292.49

33.8235.4938.64

41.9743.3443.94

40.2940.5841.84

25.8432.9249.73

Middle

df

130131132

303132

303132

303132

303132

NNFI

.93

.93

.93

.99

.99

.99

.98

.98

.98

.98

.98

.98

1.001.00.97

school

CFI

.95

.95

.95

1.001.00.99

.99

.99

.99

.99

.99

.99

1.001.00.99

res.

.040

.041

.041

.026

.026

.030

.027

.029

.030

.020

.020

.020

.016

.020

.029

TC

1.00.99.99

1.00.99.97

1.00.99.99

1.001.00.99

1.00.97.91

X2

266.55271.99276.70

36.3636.5546.15

38.8547.31"54.39"

42.2743.3365.30

45.5445.6469.12

df

130131132

303132

303132

303132

303132

High school

NNFI

.94

.94

.93

.99

.99

.97

.98

.96

.95

.98

.98

.94

.97

.97

.94

CFI

.96

.96

.95

.99

.99

.99

.99

.98

.98

.99

.99

.97

.99

.99

.97

res.

.047

.050

.053

.033

.032

.043

.037

.044

.057

.032

.033

.045

.034

.033

.048

TC

1.00.97.93

1.001.00.85

1.00.84.70

1.00.99.86

1.001.00.90

Note. All factor loadings and factor variances were significant atp < .05. NNFI = Bentler-Bonett nonnormedfit index; CFI = comparative fit index; res. = average absolute standardized residuals; TC = target coefficient(Ix2 for the model with no factor correlation - x2 for the model being tes ted] /^ for the model with no factorcorrelation — x2 for Model A])." Variance of one disturbance term was constrained to be 0.

bal and quantitative second-order factors. It postulated that atleast two higher order factors were necessary to effectivelyaccount for relations among subject-specific factors. Model C,in contrast, specified a general second-order factor. It was basedon the premise that all subject-specific motivation factors sharea sizable amount of variance through a common higher orderfactor. Among Models A, B, and C, Model A should demon-strate the best fit because it is the least constrained of the three.If Models B or C display comparable fit to Model A, theyshould be preferred to Model A because of their relativeparsimony.

There is little reason to test higher order structures whenfirst-order factors are not sufficiently correlated. Goodness-of-fit indexes, such as NNFI and CFI, of higher order models canbe misleading because they reflect the capability of the entire

model with both lower and higher order factors to account forindicator variances. One way to ascertain the necessity of lowerorder factor correlation in model definition is to compare the fitof correlated and uncorrelated lower order factor structures(Marsh, 1990; Vispoel, 1995). In the present research, this basictest resulted in statistically significant {p < .05) chi-squaredifference statistics for all constructs, attesting to the indispens-ability of first-order factor correlations. The target coefficient(TC) is another useful index that reflects the proportion oflower order factor variances that is accounted for by the higherorder factors (Marsh & Hocevar, 1985). As with other fitindexes, TC values greater than .90 are generally consideredacceptable. Although the degree of first-order factor correlationwas, on the whole, sufficient to warrant higher order analysis,some of the mastery goal and task-value factors of the high

Table 4Standardized Path Coefficients for Model A for Middle School and High School

1.2.3.4.5.6.

Path

Korean—EnglishEnglish—MathMath—ScienceKorean—MathEnglish—ScienceKorean—Science

Self-efficacy

Middle

.52*

.51*

.62*

.58*

.47*

.62*

High

.38*

.44*

.63*

.24*

.43*

.41*

Task

Middle

.50*

.46*

.52*

.44*

.36*

.46*

value

High

.34*

.25*

.51*

.14

.24*

.17*

Mastery

Middle

.38*

.34*

.60*

.43*

.50*

.50*

goal

High

.28*

.11

.48*- .14

.10

.14

Performance-approach

Middle

.67*

.64*

.70*

.58*

.68*

.66*

goal

High

.66*

.51*

.74*

.53*

.48*

.42*

Performance-avoidance goal

Middle

.73*

.66*

.79*

.60*

.58*

.68*

High

.68*

.47*

.73*

.49*

.51*

.50*

< .05.

BETWEEN- AND WITHIN-DOMAIN MOTIVATION 29

school sample demonstrated nonsignificant relations with eachother. Consequently, testing for a general factor model (i.e.,Model C) for these constructs cannot be justified with the highschool sample. Nevertheless, results for both Models B and Cwere presented for all motivational constructs for the sake ofcompleteness.

With self-efficacy, results of higher-order CFAs differed be-tween middle and high school samples. With the middle schooldata, all fit indexes (including the TC) were acceptable and iden-tical across Models B and C (see Table 3). Nonetheless, thecorrelation coefficient between the verbal and quantitative second-order factors of Model B was .96, almost approaching 1 (see Table5). The general second-order factor of Model C was clearly de-fined by all four first-order factors (see Table 6). Together withconsideration of model parsimony, Model C should be viewed asthe best representation of the middle school self-efficacy data.With the high school self-efficacy data, Models B and C againshowed similar fit indexes. In contrast to the middle school data,however, the correlation between the verbal and quantitative fac-tors of the high school sample was only .77. Table 6 also showsthat the general factor of Model C was not well represented by theKorean and English first-order factors. In particular, roughly 78%of the Korean (i.e., 1 - [.46]2) factor variances remained unac-counted for by the general second-order factor. When two separatesecond-order factors were included in Model B, loadings of theKorean and English first-order factors improved noticeably, bring-ing considerable reduction in their residual variances. Model B ishence considered the better illustration of the high school self-efficacy data.

Next, higher order CFA results for the middle school task valuewere examined. Both Models B and C showed satisfactory fitindexes, with Model B demonstrating slightly better fit (see Table3). Tables 5 and 6 reveal that all higher order factors werereasonably well defined by their first-order factors. The correlationcoefficient between the second-order verbal and quantitative fac-tors of Model B was .84. Given that this correlation was correctedfor unreliability and thus represented the highest end of correlationcoefficients that these data could afford, keeping the two correlatedsecond-order factors seemed warranted. Accordingly, Model Bwas viewed as a more accurate description of the middle schooltask-value data. Results for the high school task value are notdiscussed for the aforementioned reason.

With the middle school sample's mastery goal, goodness-of-fitindexes and TCs of both Models B and C were outstanding and

virtually the same (see Table 3). Although specifying two second-order factors somewhat improved the paths from the Korean andEnglish first-order factors to the higher order verbal factor (seeTables 5 and 6), these increments were not as substantial as theywere in the case of self-efficacy. Moreover, the correlation coef-ficient of .92 between the second-order verbal and quantitativefactors of Model B raises a question regarding their discriminantvalidity. Model C is also more parsimonious than Model B. There-fore, Model C should be considered as the most effective repre-sentation of the middle school mastery goal data. Again, results forthe high school mastery goal are not discussed.

Similar results were obtained with the middle school perfor-mance-approach goal. Fit indexes between Models B and Cwere almost identical, and both models demonstrated excellentTCs (see Table 3). The high correlation (.94) between the verbaland quantitative factors of Model B (see Table 5) and parsi-mony consideration render Model C as the best representationof the middle school performance-approach goal data. With thehigh school sample, however, a different conclusion is calledfor. Only Model B was associated with the TC at greater than.90, which is also superior to that of Model C. Loadings of theKorean and English first-order factors on their verbal second-order factor in Model B showed sizable improvement fromthose on the general second-order factor in Model C. Thecorrelation coefficient between the verbal and quantitativefactors was .70, supporting the separation of the two second-order factors.

In the case of performance-avoidance goal, both middle andhigh school samples demonstrated analogous patterns. BothModels B and C showed acceptable TCs, but those of Model Bwere superior to those of Model C (see Table 3). The verbal andquantitative factors of Model B as well as the general second-order factor of Model C were adequately defined by their lowerorder factors (see Tables 5 and 6). However, specifying twosecond-order factors accounted for considerably more variancein their first-order factors. Loadings of the Korean and Englishfactors improved substantially in Model B compared with thosein Model C. Although this phenomenon held true with bothmiddle and high school samples, it was especially pronouncedin the high school sample. Moreover, the correlation coeffi-cients between the verbal and quantitative factors were less than.90, substantiating their independent specification. Therefore,Model B is considered the most suitable hierarchical represen-

Table 5Standardized Path Coefficients for Model B for Middle School and High School

Path

1. Verbal to Korean2. Verbal to English3. Quantitative to Math4. Quantitative to Science5. Verbal—Quantitative

Self-efficacy

Middle

.80*

.65*

.77*

.80*

.96*

High

.55*

.69*

.74*

.85*

.77*

Task

Middle

.74*

.67*

.75*

.68*

.84*

value

High

.47*

.74*

.72*

.71*

.47*

Mastery

Middle

.63*

.60*

.68*

.89*

.92*

goal

High

.27*1.00*.66*.72*.15

Performance-approach

Middle

.79*

.84*

.80*

.87*

.94*

goal

High

.80*

.81*

.91*

.81*

.70*

Performance-avoidance

goal

Middle

.88*

.84*

.87*

.91*

.83*

High

.82*

.83*

.84*

.87*

.70*

* p < .05.

30 BONG

Table 6Standardized Path Coefficients for Model C for Middle School and High School

Performance- Performance-Mastery goal approach goal avoidance goalSelf-efficacy Task value

Path Middle High Middle High Middle High Middle High Middle High

1. General to Korean2. General to English3. General to Math4. General to Science

.78*

.64*

.77*

.79*

.46*

.56*

.74*

.84*

.69*

.63*

.72*

.66*

.28*

.38*

.69*

.71*

.59*

.56*

.68*

.88*

.12

.10

.47*1.00*

.78*

.82*

.79*

.86*

.65*

.65*

.87*

.79*

.79*

.77*

.86*

.87*

.67*

.67*

.82*

.83*

*p < .05.

tation of both middle and high school samples' performance-avoidance goals.

Within-Domain CFA

Interrelations among self-efficacy, task-value, and achievementgoals were explored with CFA. Of particular interest here were therelations of achievement goal orientations to other motivationconstructs and stability of these relations across different academicdomains and age groups. As in the cross-domain CFAs, eachmeasured variable was hypothesized to load on a single a priorifactor. CU paths were added between the first two self-efficacyvariables. The five motivation factors—self-efficacy, task value,mastery goal, performance-approach goal, and performance-avoidance goal—were hypothesized to correlate with one another.Analyses were conducted separately for each of the four schoolsubjects. Within each domain, the same factor structure was im-posed separately on the middle and high school data. As Table 7shows, high school models in English, math, and science demon-strated satisfactory fit. Other models demonstrated only marginallyacceptable overall fit.

Table 8 presents correlation coefficients among motivation fac-tors by domain and school level. As in previous research, self-efficacy and task-value factors were significantly and positivelycorrelated with each other across both domain and school level.Also consistent with previous research, mastery goal factorsshowed significant positive correlations with self-efficacy andtask-value factors in all four school subjects for both agegroups. The significant positive correlation between performance-approach and performance-avoidance goals, consistently observedin previous findings, was also witnessed across domains andschool levels. Consistent with Skaalvik (1997) and Elliot and

Church (1997), positive correlations were observed between per-formance-approach goals and self-efficacy and between masteryand performance-approach goals, regardless of domain and age.The performance-approach goal factor also showed positive rela-tions with the task-value factor.

Relations of performance-avoidance goals with other factorswere more ambiguous. With the high school sample, the perfor-mance-avoidance goal factor showed a nonsignificant correlationwith self-efficacy in all school subjects except science. Skaalvik(1997) also reported a nonsignificant relationship between the two,whereas Middleton and Midgley (1997) and Elliot and Church(1997) reported a negative relationship. Also with the high schoolsample, performance-avoidance goals were not significantly re-lated to either task-value or mastery goals across the four domains.The nonsignificant relation of performance-avoidance goals withmastery goals is consistent with previous research. In general, theperformance-avoidance goal factor exhibited empirical indepen-dence from all but the performance-approach goal factor in thehigh school sample.

Somewhat puzzling results were obtained with the middleschool sample. Performance-avoidance goals demonstrated signif-icant positive relations with both self-efficacy and task value in alldomains but Korean. It was also positively correlated with masterygoals in all school subjects. In previous research, performance-avoidance goals typically showed negative to nonsignificant rela-tions with these more adaptive motivational states. The perfor-mance-avoidance goals' relations to other motivation factorsconstituted the most marked difference, not only between themiddle and high school samples but also between the previous andthe present research. There were other minor differences betweenthe two age groups. For example, relations of the mastery goal

Table 7Goodness-of-Fit Indexes of Within-Domain Confirmatory Factor Analysis Models

Domain

KoreanEnglishMathScience

x2

284.14288.90298.02396.55

Middle school

df

108108108108

NNH

.86

.88

.89

.86

CFI

.89

.91

.92

.89

res.

.05

.05

.03

.04

x2

274.47253.41264.38240.57

df

108108108108

High school

NNFI

.86

.90

.91

.93

CFI

.89

.92

.93

.94

res.

.05

.05

.04

.03

Note. All factor loadings and factor variances were significant atp < .05. NNFI = Bentler-Bonett nonnormedfit index; CFI = comparative fit index; res. = average absolute standardized residuals.

BETWEEN- AND WITHIN-DOMAIN MOTIVATION 31

Table 8Standardized Correlation Coefficients Among Factors for Middle School and High School

Korean English Math

Path

Science

Middle

.77*

.77*

.73*

.05

.77*

.62*

.04

.64*

.17*

High

.65*

.73*

.51*

.02

.85*

.44*

.16

.51*-.02

Middle

.80*

.82*

.73*

.31*

.73*

.57*

.17*

.73*

.32*

High

.85*

.83*

.57*

.08

.95*

.59*

.16

.57*

.01

Middle

.85*

.82*

.77*

.34*

.64*

.86*

.34*

.70*

.32*

High

.87*

.78*

.63*

.12

.88*

.63*

.06

.69*-.01

Middle

.81*

.83*

.76*

.25*

.79*

.60*

.25*

.69*

.30*

High

.84*

.83*

.52*

.23*

.94*

.52*

.13

.61*

.14

Self-efficacy—task valueMastery goal—self-efficacyPerformance-approach goal—self-efficacyPerformance-avoidance goal—self-efficacyMastery goal—task valuePerformance-approach goal—task valuePerformance-avoidance goal—task valueMastery goal—performance-approach goalMastery goal—performance-avoidance goalPerformance-approach goal—performance-

avoidance goal .18* .33* .54* .39* .39* .35* .52* .57*

*p < .05.

with task value were uniformly stronger in the high school samplethan in the middle school sample. Relations of the performance-approach goal with self-efficacy, on the other hand, were consid-erably stronger for the middle school than the high school sampleacross the four academic domains.

Discussion

Subject-Specificity and Between-DomainRelations of Academic Motivation

The present results provide strong empirical support for thesubject specificity of self-efficacy, task-value, and variousachievement goals. Both middle and high school students ex-pressed motivational orientations that were sufficiently distinct—albeit correlated—across the core school subjects examined in thisresearch. For each of these motivational constructs, four a priorisubject-specific factors emerged. These first-order factors wereclearly defined by their respective indicators with statisticallysignificant and sizable factor loadings. The subject-specific factorswere, on average, moderately correlated among themselves. Al-though there were some notable differences in the magnitude ofthese relations by construct and age, none of the correlation coef-ficients was large enough to cast doubt on the multidimensionalnature of academic motivation. These results are consistent withthe existing theory and research and demonstrate further thatspecific school subjects indeed function as an important organiza-tional framework for school-age children's and adolescents' mo-tivation (Gottfried, 1985, 1990; Marsh & Yeung, 1996; Simpson,Licht, Wagner, & Stader, 1996).

Strengths of between-domain relations differed substantially byindividual construct. Performance-approach and performance-avoidance goals demonstrated the strongest cross-domain associ-ations, whereas task-value and mastery goals showed the weakestcorrelation. Self-efficacy perceptions were moderately correlatedacross subjects, consistent with previous findings (Bong, 1997).Although students' desires to outperform peers or to avoid nega-tive judgments were specific to each school subject, they werenonetheless least affected by the individual subject matter. Stateddifferently, those who express performance-approach or perfor-

mance-avoid goals in one achievement context would more likelypursue similar goals in other contexts. Ames (1992) argued thatone of the most salient classroom factors that affects studentmotivation is evaluation practices. As students progress from ele-mentary to middle to high school, evaluation standards becomeincreasingly normative. School environments that stress normativesuccess in turn orient students to performance goals (Anderman& Midgley, 1997; Roeser et al., 1996). The relatively strongbetween-domain correlations of performance-approach and perfor-mance-avoidance goals suggest that adopting these goals dependslargely on individual susceptibility to normative concerns.

How much value students attach to the subject matter and theirpreferences toward task mastery and challenge in the subject were,in contrast, more distinct across domains. In particular, high schoolstudents demonstrated mastery goals that were clearly differenti-ated between subjects. This extreme domain specificity of masterygoal orientation contradicts the view that achievement goal orien-tations originate from stable personal dispositions (Duda &Nicholls, 1992; Harackiewicz, Barron, Carter, Lehto, & Elliot,1997). Rather, results indicate that importance, usefulness, andintrinsic interest students perceive in the school subject may playa more meaningful role in guiding students to the mastery goaladoption. As expected, high school students' task-value percep-tions were clearly differentiated across diverse subjects, presum-ably reflecting their relatively imminent concern with future col-lege majors. Not only did mastery goal orientation show a similarpattern of cross-domain associations to that of task value, relation-ships between these two constructs became much stronger amongthe high school students than among the middle school students.One of the unanswered questions in the achievement goal researchis where the goals come from (Urdan & Maehr, 1995). Althoughcovariation does not imply causation, these results point to theneed to investigate whether the mastery and performance goals aredifferentially affected by different sources.

Differences in the cross-domain associations in turn determinedthe suitability of hypothesized hierarchical representations for eachmotivation construct. A hierarchical structure with the generalsecond-order factor most effectively illustrated relations amongmiddle school students' self-efficacy, mastery goals, and perfor-

32 BONG

mance-approach goals in the four school subjects. That a generalfactor taps all lower order factors should not be taken as evidencethat the particular construct lacks domain specificity (Bong, 1997;Marsh, 1990). Quite the contrary, 23% to 69% (i.e., 1 - [factorloading]2) of the variance in the subject-specific factors of theseconstructs was unique to themselves and was thus left unaccountedfor by the higher order factor. The good fit demonstrated by thegeneral factor model simply suggests that these early adolescentsexpressed perceptions of self-efficacy, mastery goals, and perfor-mance-approach goals that were fairly similar across differentcontent areas. Middle school students' value perceptions and per-formance-avoid orientations and high school students' self-efficacy, performance-approach, and performance-avoidance goalswere better represented by a hierarchical structure with separateverbal and quantitative factors.

High school students' subject-specific task-value and masterygoals in the four academic subjects were too weakly correlated torender acceptable any hierarchical representation. Therefore, as-sessing general academic value or general mastery orientationswithout referring to specific school subjects and tasks may behighly inappropriate at least for late adolescents. In general, highschool students demonstrated more differentiated motivational be-liefs than did middle school students. This pattern was consistentlyobserved across the five motivation constructs considered in thisstudy. The difference between the two age groups is mostly due tothe high school students' clearer distinction between primarilyverbal and primarily quantitative subjects. On a broad level, theincreased differentiation of academic motivation demonstrated byhigh school students corroborates findings from the self-conceptresearch. Shavelson, Hubner, and Stanton (1976) argued that "withincreasing age and experience (especially acquisition of verballabels), self-concept becomes increasingly differentiated" (p. 414).Similar mechanisms may be at work with other motivationalconstructs.

As Marsh and Yeung (1998) pointed out, in their discussionof the self-concept literature, results from the higher orderfactor analyses do not imply any direction of causality betweenthe more specific and more general components. The presentresults certainly do not indicate that the subject-specific factorscan be safely inferred from the higher order factors. In addition,they do not suggest, as Bong (1997) warned, that the moregeneral factors can function the same way as the more specificfactors. The results merely indicate that some motivationalconstructs appear to be more hierarchically structured than theothers and that the nature of this hierarchy differs betweendifferent constructs and age groups. Perhaps the most pressingneed for future research in this area involves uncovering thepsychological grounds that create such a hierarchy and itschange thereafter. There are many viable explanations of whystudents' academic motivation begins to differ across diversesubject areas as they grow older. However, whereas some ofthese mechanisms may be relevant to most academic motivationconstructs, others seem pertinent mainly to a subset of theseconstructs. More research is needed on the social-cognitiveprocesses underlying the differentiation of each motivationconstruct and on the differences in students' behavioral inten-tions before and after such differentiation.

Consistency of Within-Domain Relationsof Academic Motivation

Consistent with previous results (e.g., Berndt & Miller, 1990;Elliot & Church, 1997; Ethington, 1991; Meece et al., 1990;Middleton & Midgley, 1997; Pokay & Blumenfeld, 1990; Skaal-vik, 1997), academic self-efficacy, task-value, and mastery goalperceptions were positively correlated in all school subjects amongboth middle and high school students. Also consistent with previ-ous findings, performance-approach and performance-avoidancegoals showed a significant positive relation across domain andschool level. Performance-approach goals also demonstrated pos-itive correlations with self-efficacy, task-value, and mastery goalorientations. These latter findings challenge Nicholls's (1984)earner claim that ego goals work to lower intrinsic motivation.More recent research, based on the differentiated conception ofachievement goals into approach and avoidance motives, showedthe facilitative effects of performance-approach goals (Elliot &Church, 1997; Elliot & Harackiewicz, 1996; Skaalvik, 1997; butsee Middleton & Midgley, 1997). The present results provideadditional evidence in support of the approach-avoid distinction.The evidence is especially powerful because the positive associa-tions of performance-approach goals to other adaptive motiva-tional orientations were observed across multiple academic do-mains and different age levels.

Overall, the present research revealed more similarities thandifferences in how important motivation constructs relate to oneanother. Only a few researchers have examined the uniformity ofthese relations across diverse achievement contexts (Gottfried,1985; Mac Iver, Stipek, & Daniels, 1991; Meece et al., 1988).These researchers have generally reported a reasonable degree ofconsistency in construct relations, sometimes despite appreciabledifferences in mean-level motivation. For example, Mac Iver et al.(1991) found that although there were significant mean differencesin within-semester changes of intrinsic value, utility value, self-concept of ability, and effort investment across course types,relations among these changes were nonetheless parallel in differ-ent courses. The present results are compatible with their findingsin that self-efficacy, task-value, and achievement goals in discretedomains, within each age group, showed a very similar pattern ofinterconnectedness. It will be interesting to see whether this con-sistency of within-domain relations is maintained when moreconcrete outcomes such as task choice and performance enter theequation.

In contrast to the remarkable cross-domain resemblance, therewere several age-related differences. The most marked differenceinvolved the role of performance-avoidance goals. With highschool students, performance-avoidance goals demonstratedmostly nonsignificant relations with positive motivational orienta-tions (Middleton & Midgley, 1997; Skaalvik, 1997). Whether highschool students would display avoidance orientations in a givensubject was, therefore, independent of their perceptions of confi-dence, value, and mastery preferences in that subject. Oddlyenough, however, middle school students' avoidance goals showedsignificant positive relations with those same motivational con-structs in all subjects except Korean. In other words, as thesestudents feel more efficacious and perceive greater task-value inthe given subject, they not only put forth effort to improve their

BETWEEN- AND WITHIN-DOMAIN MOTIVATION 33

competence and document their superior ability but also try hard toavoid looking incapable.

This finding can be understood in light of Mac Iver et al.'s(1991) observation. The researchers suspected that relationsamong important motivational constructs would differ betweenmiddle and high school students. Specifically, they reasoned that,compared with high school students,, for whom utility value wouldplay a more critical role, middle school students would be influ-enced more by their willingness to please their parents. Perceivedimportance of extrinsic pressure indeed related significantly withincreased effort among middle school students but not among highschool students. Presumably, the young adolescents who partici-pated in this research possessed a strong desire to please signifi-cant adults, and this led them to manifest similar levels of approachand avoidance tendencies. The considerably stronger associationsbetween both types of performance goals and perceptions of self-efficacy exhibited by middle school students are in line with thisinterpretation. Interestingly, middle school students' motivationalpatterns in Korean resembled those of high school students. Com-pared with other school subjects whose demand characteristicschange dramatically as students transit to middle schools, Koreanmay be perceived by most Korean students as a relatively stablesubject. This might have contributed to the middle school students'discrimination of Korean from other school subjects. The proposedrelationship between task novelty/familiarity and performance ori-entation is speculative and warrants further probing.

The present research has several limitations that have implica-tions for future work in this area. First, it dealt only with students'academic motivation in core academic subjects. A different con-clusion may be reached when a more expanded set of schoolsubjects are included. For example, Marsh and Shavelson (1985)found that two higher order factors—verbal and math self-con-cepts—were sufficient to describe relations among lower orderself-concepts in the core academic subjects. However, additionalhigher order factors were required to adequately represent thecovariation among self-concepts in more diverse school subjects.Likewise, evaluation concerns would be significantly lower indomains that are viewed as less important. Between-domain asso-ciations of achievement goals may change accordingly. The resultsreported in this article, therefore, may show only part of the wholepicture for each motivational construct. Second, the present studywas conducted with Korean students. There may be importantdifferences in motivational patterns between Korean (or Asianstudents in general) and Western students. Eaton and Dembo(1997) reported that fear of academic failure predicted achieve-ment motivation of Asian American students but not that of theirnon-Asian peers. Although the results generally fit with the exist-ing theory, their generalizability may be limited. Third, this articlediscussed several age-related differences. However, firm conclu-sions regarding developmental changes in motivation generalityand interrelations should await a longitudinal investigation.

References

Ames, C. (1984). Competitive, cooperative, and individualistic goal struc-tures: A cognitive-motivational analysis. In R. Ames & C. Ames (Eds.),Research on motivation in education: Vol. I. Student motivation (pp.177-207). Orlando, FL: Academic Press.

Ames, C. (1992). Classrooms: Goals, structures, and student motivation.Journal of Educational Psychology, 84, 261-271.

Anderman, E. M., & Midgley, C. (1997). Changes in achievement goalorientations, perceived academic competence, and grades across thetransition to middle-level schools. Contemporary Educational Psychol-ogy, 22, 269-298.

Bandura, A. (1997). Self-efficacy: The exercise of control. New York:Freeman.

Bentler, P. M. (1992). EQS: Structural equations program manual. LosAngeles: BMDP Statistical Software, Inc.

Berndt, T. J., & Miller, K. E. (1990). Expectancies, values, and achieve-ment in junior high school. Journal of Educational Psychology, 82,319-326.

Bong, M. (1996). Problems in academic motivation research and advan-tages and disadvantages of their solutions. Contemporary EducationalPsychology, 21, 149-165.

Bong, M. (1997). Generality of academic self-efficacy judgments: Evi-dence of hierarchical relations. Journal of Educational Psychology, 89,696-709.

Bong, M., & Clark, R. E. (1999). Comparison between self-concept andself-efficacy in academic motivation research. Educational Psycholo-gist, 34, 139-154.

Byrne, B. M., & Shavelson, R. J. (1986). On the structure of adolescentself-concept. Journal of Educational Psychology, 78, 474-481.

Duda, J. L., & Nicholls, J. G. (1992). Dimensions of achievement moti-vation in schoolwork and sport. Journal of Educational Psychology, 84,290-299.

Dweck, C. S. (1989). Motivation. In A. Lesgold & R. Glaser (Eds.),Foundations for a psychology of education (pp. 87-136). Hillsdale, NJ:Erlbaum.

Eaton, M. J., & Dembo, M. H. (1997). Differences in the motivationalbeliefs of Asian American and non-Asian students. Journal of Educa-tional Psychology, 89, 433-440.

Eccles, J., Wigfield, A., Harold, R. D., & Blumenfeld, P. (1993). Age andgender differences in children's self- and task perceptions during ele-mentary school. Child Development, 64, 830-847.

Elliot, A. J., & Church, M. A. (1997). A hierarchical model of approachand avoidance achievement motivation. Journal of Personality andSocial Psychology, 72, 218-232.

Elliot, A. J., & Harackiewicz, J. M. (1996). Approach and avoidanceachievement goals and intrinsic motivation: A mediational analysis.Journal of Personality and Social Psychology, 70, 461-475.

Ethington, C. A. (1991). A test of a model of achievement behaviors.American Educational Research Journal, 28, 155-172.

Feather, N. T. (1988). Values, valences, and course enrollment: Testingrole of personal values within an expectancy-valence framework. Jour-nal of Educational Psychology, 80, 381-391.

Gottfried, A. E. (1985). Academic intrinsic motivation in elementary andjunior high school students. Journal of Educational Psychology, 77,631-645.

Gottfried, A. E. (1990). Academic intrinsic motivation in young elemen-tary school children. Journal of Educational Psychology, 82, 525-538.

Harackiewicz, J. M., Barron, K. E., Carter, S. M., Lehto, A. T., & Elliot,A. (1997). Predictors and consequences of achievement goals in thecollege classroom: Maintaining interest and making the grade. Journalof Personality and Social Psychology, 73, 1284-1295.

Harter, S., & Pike, R. (1984). The Pictorial Scale of Perceived Competenceand Social Acceptance for Young Children. Child Development, 55,1969-1982.

Mac Iver, D. J., Stipek, D. J., & Daniels, D. H. (1991). Explaining within-semester changes in students' effort in junior high school and senior highschool courses. Journal of Educational Psychology, 83, 201-211.

Marsh, H. W. (1990). The structure of academic self-concept: The Marsh/Shavelson model. Journal of Educational Psychology, 82, 623—636.

Marsh, H. W. (1992). Content specificity of relations between academic

34 BONG

achievement and academic self-concept. Journal of Educational Psy-chology, 84, 35-42.

Marsh, H. W., Byrne, B. M., & Shavelson, R. J. (1988). A multifacetedacademic self-concept: Its hierarchical structure and its relation to aca-demic achievement. Journal of Educational Psychology, 80, 366-380.

Marsh, H. W., Byrne, B. M., & Yeung, A. S. (1999). Casual ordering ofacademic self-concept and achievement: Reanalysis of a pioneeringstudy and revised recommendations. Educational Psychologist, 34, 155-167.

Marsh, H. W., Craven, R. G., & Debus, R. (1991). Self-concepts of youngchildren 5 to 8 years of age: Measurement and multidimensional struc-ture. Journal of Educational Psychology, 83, 377-392.

Marsh, H. W., & Hocevar, D. (1985). Application of confirmatory factoranalysis to the study of self-concept: First- and higher order factormodels and their invariance across groups. Psychological Bulletin, 97,562-582.

Marsh, H. W., & Shavelson, R. J. (1985). Self-concept: Its multifaceted,hierarchical structure. Educational Psychologist, 20, 107-123.

Marsh, H. W., Walker, R., & Debus, R. (1991). Subject-specific compo-nents of academic self-concept and self-efficacy. Contemporary Educa-tional Psychology, 16, 331-345.

Marsh, H. W., & Yeung, A. S. (1996). The distinctiveness of affects inspecific school subjects: An application of confirmatory factor analysiswith the National Educational Longitudinal Study of 1988. AmericanEducational Research Journal, 33, 665-689.

Marsh, H. W., & Yeung, A. S. (1998). Top-down, bottom-up, and hori-zontal models: The direction of causality in multidimensional, hierar-chical self-concept models. Journal of Personality and Social Psychol-ogy, 75, 509-527.

Meece, J. L. (1994). The role of motivation in self-regulated learning. InD. H. Schunk & B. J. Zimmerman (Eds.), Self-regulation of learning andperformance: Issues and educational applications (pp. 25-44). Hills-dale, NJ: Erlbaum.

Meece, J. L., Blumenfeld, P. C , & Hoyle, R. H. (1988). Students' goalorientations and cognitive engagement in classroom activities. Journalof Educational Psychology, 80, 514-523.

Meece, J. L., Wigfield, A., & Eccles, J. S. (1990). Predictors of mathanxiety and its influence on young adolescents' course enrollment in-tentions and performance in mathematics. Journal of Educational Psy-chology, 82, 60-70.

Middleton, M. J., Kaplan, A., & Midgley, C. (1998, April). Achievementgoal orientation and self-efficacy: Different goals, different relations.Paper presented at the annual meeting of the American EducationalResearch Association, San Diego, CA.

Middleton, M. J., & Midgley, C. (1997). Avoiding the demonstration oflack of ability: An underexplored aspect of goal theory. Journal ofEducational Psychology, 89, 710-718.

Mischel, W. (1977). On the future of personality measurement. AmericanPsychologist, 32, 246-254.

Nicholls, J. G. (1984). Conceptions of ability and achievement motivation.In R. Ames & C. Ames (Eds.), Research on motivation in education:Vol. 1. Student motivation (pp. 39-73). Orlando, FL: Academic Press.

Pajares, F. (1996). Self-efficacy beliefs in academic settings. Review ofEducational Research, 66, 543-578.

Pintrich, P. R., & De Groot, E. V. (1990). Motivational and self-regulatedlearning components of classroom academic performance. Journal ofEducational Psychology, 82, 33-40.

Pokay, P., & Blumenfeld, P. C. (1990). Predicting achievement early andlate in the semester: The role of motivation and use of learning strate-gies. Journal of Educational Psychology, 82, 41—50.

Roeser, R. W., Midgley, C , & Urdan, T. C. (1996). Perceptions of theschool psychological environment and early adolescents' psychologicaland behavioral functioning in school: The mediating role of goals andbelonging. Journal of Educational Psychology, 88, 408—422.

Schunk, D. H. (1991). Self-efficacy and academic motivation. EducationalPsychologist, 26, 207-231.

Schunk, D. H., & Swarte, C. W. (1993). Goals and progress feedback:Effects on self-efficacy and writing achievement. Contemporary Edu-cational Psychology, 18, 337-354.

Shavelson, R. J., Hubner, J. J., & Stanton, G. C. (1976). Self-concept:Validation of construct interpretations. Review of Educational Re-search, 46, 407-441.

Simpson, S. M., Licht, B. G., Wagner, R. K., & Stader, S. R. (1996).Organization of children's academic ability-related self-perceptions.Journal of Educational Psychology, 88, 387-396.

Skaalvik, E. M. (1997). Self-enhancing and self-defeating ego orientation:Relations with task and avoidance orientation, achievement, self-perceptions, and anxiety. Journal of Educational Psychology, 89,71-81.

Smith, R. E. (1989). Effects of coping skills training on generalizedself-efficacy and locus of control. Journal of Personality and SocialPsychology, 56, 228-233.

Turner, J. C , Thorpe, P. K., & Meyer, D. K. (1998). Students' reports ofmotivational and negative affect: A theoretical and empirical analysis.Journal of Educational Psychology, 90, 758—771.

Urdan, T., & Maehr, M. L. (1995). Beyond a two-goal theory of motivationand achievement: A case for social goals. Review of Educational Re-search, 65, 213-243.

Vispoel, W. P. (1995). Self-concept in artistic domains: An extension ofthe Shavelson, Hubner, and Stanton (1976) model. Journal of Educa-tional Psychology, 87, 134-153.

Weiner, B. (1990). History of motivation research in education. Journal ofEducational Psychology, 82, 616—622.

Wigfield, A., & Eccles, J. S. (1992). The development of achievement taskvalues: A theoretical analysis. Developmental Review, 12, 265-310.

Wigfield, A., Eccles, J. S., Mac Iver, D,, Reuman, D. A., & Midgley, C.(1991). Transitions during early adolescence: Changes in children'sdomain-specific self-perceptions and general self-esteem across the tran-sition to junior high school. Developmental Psychology, 27, 552-565.

Wigfield, A., Eccles, J. S., Yoon, K. S., Harold, R. D., Arbreton, A. J. A.,Freedman-Doan, C , & Blumenfeld, P. C. (1997). Change in children'scompetence beliefs and subjective task values across the elementaryschool years: A 3-year study. Journal of Educational Psychology 89,451-469.

Received February 15, 2000Revision received May 25, 2000

Accepted May 26, 2000 •