Journal of Educational Psychology1995, Vol. 87, No. 3, 386-398 Copyright 1995 by the American Psychological Association, Inc.

0022-0663/95/$3.00

Self-Efficacy, Attribution, and Outcome Expectancy Mechanisms inReading and Writing Achievement: Grade-Level and

Achievement-Level Differences

Duane F. ShellUniversity of Texas at Austin

Carolyn ColvinUniversity of Iowa

Roger H. BruningUniversity of Nebraska—Lincoln

This study examined grade- and achievement-level differences in 4th-, 7th-, and lOth-gradestudents' control-related beliefs and relations between students' beliefs and their reading andwriting achievement. MANOVA results indicated grade- and achievement-level differencesin self-efficacy, causal attribution, and outcome expectancy beliefs but no interaction betweengrade and achievement level. Canonical correlations identified a single dimension linkingstudents' beliefs to achievement in both reading and writing. Quadratic relations to achieve-ment were found for outcome expectancy and intelligence attributions. As grade increased,beliefs for reading were more highly related to comprehension skill relative to componentskills, whereas beliefs for writing were more highly related to component skills relative tocommunication skills. At all achievement levels, a similar pattern of beliefs was related toachievement.

People's beliefs about their abilities to exercise personalcontrol of important events in their lives are thought to playa major role in motivating the self-regulation of cognitiveperformance and learning (see Bandura, 1986; Schunk,1991; Weiner, 1985; Zimmerman, 1989). Three particularcontrol-related beliefs have received extensive theoreticalformulation and empirical examination: (a) self-efficacy(Bandura, 1986; Schunk, 1991; Zimmerman, 1989), definedas confidence in one's capability for organizing and imple-menting the cognitive, behavioral, or social skills necessaryfor successful performance of a task; (b) causal attributions(Weiner, 1985), defined as one's judgments about the cau-sality of success or failure in achievement situations; and (c)outcome expectancy (Bandura, 1986; Schunk, 1991), de-fined as beliefs about the contingencies between one's suc-cessful task performance and possible outcomes or theexpectation that a behavior will result in particular out-comes. These beliefs are thought to play a foundational rolein motivating behavior for tasks that require high levels ofpersonal self-initiation and active self-regulation (see Ban-dura, 1986; Schunk, 1991; Zimmerman, 1989).

A growing body of research has shown that self-efficacy,causal attributions, and outcome expectancy are related toreading and writing achievement (Ehrlich, Kurtz-Costes, &

Duane F. Shell, Department of Educational Psychology, Univer-sity of Texas at Austin; Carolyn Colvin, Department of Curriculumand Instruction, University of Iowa; Roger H. Bruning, Depart-ment of Educational Psychology, University of Nebraska—Lincoln.

Correspondence concerning this article should be addressed toDuane F. Shell, Department of Educational Psychology, SZB 504,University of Texas, Austin, Texas 78712-1296. Electronic mailmay be sent via Internet to duane.shell@mail.utexas.edu.

Loridant, 1993; McCarthy, Meier, & Rinderer, 1985; Paris& Oka, 1986; Schunk & Rice, 1991; Schunk & Swartz,1993; Shell, Murphy, & Bruning, 1989; Wagner, Spratt,Gal, & Paris, 1989). These relations, however, are affectedby developmental and achievement-level differences inbeliefs.

Previous studies have found that self-efficacy, causalattribution, and outcome expectancy beliefs undergo devel-opmental change (Bandura, 1986; Hiebert, Winograd, &Danner, 1984; Paris & Oka, 1986; Schunk, 1991; Stipek,1993; Weiner, 1985). Young children commonly (a) haveinaccurate perceptions of causality (e.g., Stipek, 1993;Weiner, 1985), (b) overestimate the contingency betweentheir behaviors and outcomes (Stipek, 1993), and (c) over-state their self-efficacy or ability (Paris & Oka, 1986; Sti-pek, 1993). Also, as children age, their beliefs increase inaccuracy (Paris & Oka, 1986; Stipek, 1993). A particularlysalient change in causal attributions has been identified.Young children tend to equate effort and ability as causes,whereas older children tend to see effort and ability asinversely related (e.g., Stipek, 1993; Weiner, 1985). As aresult, young children tend to attribute success to effortmore than do older children (Stipek, 1993).

In addition to developmental changes in beliefs them-selves, developmental changes in the relations between be-liefs and achievement have been identified. As children age,their beliefs become more highly related to achievement(Paris & Oka, 1986; Stipek, 1993; Wagner et al., 1989).Also, with increasing age, there are changes in the relationsbetween specific beliefs and achievement. In particular, forolder children, self-efficacy or perceptions of ability andachievement are more highly related to achievement (Ban-dura, 1986; Paris & Oka, 1986; Stipek, 1993), and attribu-

386

BELIEF MECHANISMS IN READING AND WRITING ACHIEVEMENT 387

tion of success to ability is more highly related to achieve-ment relative to attribution to effort (Stipek, 1993).

Along with developmental differences, numerous studieshave found differences between the self-efficacy, causalattribution, and outcome expectancy beliefs of high achiev-ers and low achievers (Bandura, 1986; Ehrlich et al., 1993;Hiebert et al., 1984; Paris & Oka, 1986; Schunk, 1991;Stipek, 1993; Weiner, 1985). Compared with low achievers,high achievers tend to (a) have higher self-efficacy, (b)attribute causality for success more to internal causes (abil-ity or effort) relative to external causes (luck, task difficulty,or help), and (c) have higher outcome expectancy (e.g.,Bandura, 1986; Schunk, 1991; Stipek, 1993; Weiner, 1985).Also, beliefs have been found to be more strongly related toachievement for high achievers than for low achievers (e.g.,Bandura, 1986; Ehrlich et al., 1993; Paris & Oka, 1986;Stipek, 1993).

Although both developmental and achievement-level dif-ferences in beliefs have been studied, the interaction be-tween development and achievement has not been exten-sively examined. Because increased achievement is oftenassociated with increasing age, particularly for subjectsstudied in school, achievement-related differences poten-tially could be an artifact of developmental differences.Conversely, developmental differences could be an artifactof increased competency as knowledge and skills are mas-tered. Studies have found independent effects for develop-ment and achievement on beliefs, however, suggesting thatdifferences related to development are distinct from thoserelated to achievement (e.g., Hiebert et al., 1984; Paris &Oka, 1986). Even if development and achievement haveindependent effects on beliefs, they still may interact.Hiebert et al. (1984) found that high-achieving third graderswere more similar in causal attribution patterns to sixthgraders than to their low-achieving peers, and Paris and Oka(1986) identified different regression equations for studentsof different ability levels across grade levels. These findingssuggest possible interactions between development andachievement; however, there has been little direct empiricalexamination of possible interactions.

Previous studies (Ehrlich et al., 1993; Hiebert et al., 1984;Paris & Oka, 1986; Wagner et al., 1989) have examinedprimarily elementary and middle school children. Thus,there is only limited knowledge about how beliefs for read-ing continue to develop through high school and howachievement-level differences might interact with develop-mental differences in the higher grades. Also, few studies(e.g., Schunk & Swartz, 1993) have examined children'sbeliefs about writing. As a result, developmental or achieve-ment-level differences in children's beliefs about writinghave not been systematically examined. Studies examiningadult, skilled participants have found that beliefs are lessstrongly related to achievement for writing than for reading(McCarthy et al., 1985; Shell et al., 1989); however, howthis difference in the strength of the relations for readingand writing develops and whether this difference also oc-curs for persons at other achievement levels presently arenot known.

Two factors that potentially affect the understanding of

developmental and achievement-level differences in the re-lations between beliefs and achievement for reading andwriting were identified by Shell et al. (1989). First, theyfound that for adult, skilled participants, there was a singlecanonical dimension linking beliefs to reading and writingachievement. This indicated generalized, reciprocal rela-tions between beliefs and achievement across the two do-mains. Shell et al.'s findings are consistent with studiesshowing that reading and writing component skills are re-ciprocally related and exert generalized influences on eachother's development (e.g., Shanahan, 1984; Shanahan &Lomax, 1986). These generalized, reciprocal relationsacross the two domains potentially could influence thedevelopment of the relations between beliefs and readingand writing achievement. Also, they could be different forchildren at different levels of achievement. No studies,however, have examined generalized relations across thetwo domains with children at different ages or with personsat different levels of achievement.

Second, Shell et al. (1989) found that for adult, skilledreaders and writers, outcome expectancy had a quadraticcurvilinear relation to reading and writing achievement.When curvilinearity is present, an exclusively linear modelunderestimates correlations. Therefore, if curvilinearity alsoexists in the relations between beliefs and achievement forchildren at different ages and achievement levels, previousstudies that examined only linear relations (e.g., Ehrlich etal., 1993; Paris & Oka, 1986; Wagner et al., 1989) may havedrawn inaccurate conclusions about the comparativestrengths of the relations between self-efficacy, causal at-tributions, and outcome expectancy and reading or writingachievement. No examinations of curvilinearity, however,have been done with children of different ages or achieve-ment levels.

A final consideration for understanding developmentaland achievement-level differences is the multifaceted natureof reading and writing. Previous studies have tended toexamine the relations between beliefs and single indicatorsof achievement, usually reflecting reading comprehensionor overall writing quality (e.g., Ehrlich et al., 1993; Hiebertet al., 1984; McCarthy et al., 1985; Paris & Oka, 1986; Shellet al., 1989; Wagner et al., 1989). Reading and writing,however, involve multiple component skills (e.g., Shana-han, 1984; Shanahan & Lomax, 1986). For these differentcomponent skills, the relations between beliefs and achieve-ment may be different across different grades and at differ-ent levels of achievement. Studies using a multivariateframework to examine the relations between beliefs andmultiple component skills, however, have not been done.

The purpose of the present study was to examine thenature of grade-level and achievement-level differences inself-efficacy, causal attribution, and outcome expectancybeliefs for reading and writing and the relations betweenthese beliefs and achievement in reading and writing. Spe-cifically, we sought to clarify and extend the existing re-search literature by (a) examining students over a broaderrange of grades from elementary school through high schoolthan has previously been studied, (b) testing for interactionsbetween grade level and achievement, (c) examining beliefs

388 D. SHELL, C. COLVIN, AND R. BRUNING

and achievement for both reading and writing to determinethe cross-domain generality of beliefs and to identify anyunderlying dimensionality present in the relations betweenbeliefs and achievement across grade and achievement lev-els, (d) testing for the presence of curvilinearity in therelations between beliefs and achievement, and (e) conduct-ing the study using a multivariate approach.

Method

Participants

Participants were 364 students (155 boys, 193 girls, and 16 ofunknown sex) from a midwestern public school system who vol-unteered to participate with the approval of their parents or guard-ians. The sample consisted of 105 fourth graders (43 boys, 59 girls,and 3 of unknown sex), 111 seventh graders (43 boys, 63 girls, and5 of unknown sex), and 148 tenth graders (69 boys, 71 girls, and8 of unknown sex). The students were predominantly White andfrom middle-class families, although students from families at alllevels of socioeconomic status participated in the study.

Measurement of Belief Variables

Self-efficacy. The self-efficacy instruments for reading and forwriting were adapted from the instruments reported in Shell et al.(1989) that were used with a college-age population. Each instru-ment contained two subscales: (a) a task subscale consisting ofreading or writing tasks of varying difficulty and (b) a skillsubscale consisting of component skills involved in reading orwriting. The reading and writing tasks used were consistent withthose used by Shell et al. but were adapted to better reflect thecommon types of reading and writing done by students in thegrades studied. Also, because of concerns about possible fatigueand loss of attention during administration, particularly for fourth-grade students, the instruments were shortened.

For the reading task subscale, students were provided with thefollowing five tasks: (a) "read a letter from a friend," (b) "read oneof your textbooks," (c) "read the daily newspaper," (d) "read abook from the library," and (e) "read a magazine like Newsweek."For the reading skill subscale, students were provided with thefollowing four component skills: (a) "know all the words on a pagein one of your school books," (b) "know the meaning of plurals,prefixes, and suffixes," (c) "identify parts of speech," and (d)"understand the main idea of a story." For the writing task sub-scale, students were provided with the following five tasks: (a)"write down the rules for a game you like," (b) "write a letter to afriend," (c) "write a 2-page report for a class," (d) "write a 1-pagesummary of a book you read," and (e) "write a story about whatyou did on summer vacation." For the writing skill subscale,students were provided with the following four component skills:(a) "correctly punctuate a sentence," (b) "use correct parts ofspeech in your writing," (c) "use correct plurals, prefixes, andsuffixes in your writing," and (d) "get your point across in yourwriting."

Because the subjective probability scale used by Shell et al.(1989) was considered to be too developmentally advanced for thestudents examined, particularly the 4th and 7th graders, a simplerresponse scale was developed and used in all grades. Students wereasked to indicate how sure they were that they could do each taskor component skill on a 5-point scale as follows: 1 (I'm sure Ican't), 2 (don't think I can), 3 (maybe I can), 4 (pretty sure I can),and 5 (I'm sure I can). Self-efficacy scores were computed by

calculating the mean scores for the items in each subscale. Coef-ficient alpha reliability estimates for the self-efficacy scales were.72 for the reading task subscale, .62 for the reading skill subscale,.69 for the writing task subscale, and .76 for the writing skillsubscale.

Outcome expectancy. The outcome expectancy instruments forreading and for writing were adapted from the instruments re-ported in Shell et al. (1989). In separate instruments, students wereasked to rate the importance of reading and writing for achievingsix goals on a 5-point scale as follows: 1 (not important at all), 2(not very important), 3 (kind of important), 4 (pretty important),and 5 (very important). Outcome expectancy scores were com-puted by calculating mean scores for each instrument. Five goalswere the same for both instruments: (a) "getting a job when youare an adult," (b) "having a lot of friends," (c) "getting along withyour parents," (d) "doing good in school," and (e) "going tocollege." The sixth goal for the reading instrument was "learningnew things." The sixth goal for the writing instrument was "tellingpeople things you know." Coefficient alpha reliability estimatesfor the outcome expectancy instruments were .50 for the readinginstrument and .56 for the writing instrument.1

Causal attributions. Students' beliefs about the causality ofsuccess were assessed in separate instruments for reading andwriting. In each, students were asked to rate the importance of sixdifferent causes (effort, ability defined as general intelligence,enjoyment, luck, task difficulty, and teacher help) for being a goodreader or writer on a 5-point scale as follows: 1 (not important atall), 2 (not very important), 3 (kind of important), 4 (prettyimportant), and 5 (very important). Each rating score was usedindividually in the analyses, resulting in 12 causal attributionscores, 6 for reading and 6 for writing.

Measurement of Reading and Writing Achievement

California Achievement Test scores. Students' scores on theCalifornia Achievement Test (CAT; Forms C and D) were used asmeasures of reading and writing achievement. CAT standardscores were used for all measures. These scores are standardizedon a continuous scale across all levels of the CAT, allowing directcomparison of scores from tests at different grade levels. Scoreswere obtained for Reading Comprehension, Reading Vocabulary,Language Mechanics, Language Expression, and Spelling.

Two of the CAT tests measure components of reading. TheReading Comprehension test measures three categories of com-prehension: (a) literal comprehension (recall of facts), (b) inter-pretive comprehension (inferred meaning, character analysis, andfigurative language), and (c) critical comprehension (author atti-tude or position and techniques of persuasion). The Reading Vo-cabulary test measures understanding of words with similar mean-ings, words with opposite meanings, and words with multiplemeanings.

Three of the CAT tests measure components of writing. TheLanguage Mechanics test measures such skills as capitalization,adjective use, and punctuation. The Language Expression testmeasures students' skills related to written expression in the areas

1 Although low, the coefficient alpha measure may have under-estimated the reliability for the outcome expectancy scales becauseof the short scale length of six items and ceiling effects on theitems "going to college" in both scales and "learning new things"in the reading scale. We remind readers that low reliability lowerspower (Pedhazur, 1982); therefore, low reliabilities on these mea-sures may have resulted in a failure to identify significant effectsbut could not have caused spurious effects to be found.

BELIEF MECHANISMS IN READING AND WRITING ACHIEVEMENT 389

of (a) word usage, (b) sentence structure, and (c) paragraph orga-nization. Although the test is designed to assess application ofwriting knowledge, no writing is done on the test. The Spellingtest measures spelling using the phoneme-grapheme-morphemeapproach.

Holistic writing score. Because the CAT does not contain a testwhere actual writing is done, a writing essay was used to obtain aperformance-based measure of writing. For the essay, studentswere asked to complete the following writing task:

Write two paragraphs describing your favorite TV program. Inthe first paragraph, name the program and write about who isin it, what they do in the program, and why they do the thingsthat they do. In the second paragraph, write about why youlike the show and why someone else would enjoy watching it.

Students were allowed as much time as they needed to completethe essay.

The writing essays were independently scored by one of theresearchers and a second rater using the holistic scoring methoddescribed by Shell et al. (1989). The raters were unaware of eachother's rating scores or students' identities, grade levels, or scoreson other measures. Raters assigned subscores for the followingfive categories: realization, clarity, organization, quantity-density,and mechanics. Because Shell et al. (1989) reported range limita-tions with their 4-point scale, each subscore category was ratedfrom 0 to 20. The five subscores were then summed, resulting ina total score ranging from 0 to 100. The final holistic writing scorewas computed as the mean score of the two raters. To facilitatecomparison with the CAT standard scores, the holistic score wasconverted to a standard score with a mean of 500 and a standarddeviation of 100.

Interrater reliability was assessed by correlating the scores of thetwo raters. A Pearson product-moment correlation of .74 wasobtained between the scores of the two raters. This reliabilityestimate was consistent with the interrater reliability obtained byShell et al. (1989) and was within the range of reliability estimatescommonly obtained for holistic scoring reliabilities (.68 to .89)when adequate statistical methods are used, as reported by White(1985). Because the mean score of the two raters was used foranalysis, the correlation between the raters is equivalent to asplit-half estimate of the reliability of the mean rating. Using theSpearman-Brown formula, we estimated the reliability of the meanholistic score to be .85.

Achievement-Level Groups

Using a composite literacy score created by calculating themean of the six reading and writing achievement scores, studentswere assigned to high-, average-, and low-achievement groupswithin their respective grades on the basis of cutoff scores atapproximately the 70th and 30th percentiles. The actual separa-tions for each grade were made near the cutoff scores at pointswhere there were relatively large differences between consecutivescores. The resulting achievement groups were as follows: highachievement (n = 104; 38 boys, 62 girls, and 4 of unknown sex;30 in Grade 4, 32 in Grade 7, and 42 in Grade 10); averageachievement (n = 156; 64 boys, 85 girls, and 7 of unknown sex;46 in Grade 4, 47 in Grade 7, and 63 in Grade 10); and lowachievement (n = 104; 53 boys, 46 girls, and 5 of unknown sex;29 in Grade 4, 32 in Grade 7, and 43 in Grade 10).

Procedures

Students were administered the instruments for self-efficacy,causal attribution, and outcome expectancy, and the writing task

during their first (fall) semester of 4th, 7th, or 10th grade for therespective samples. For the 4th-grade sample, data were collectedin the students' regular classrooms, and, for the 7th- and 10th-grade samples, data were collected in either the students'homeroom or study hall classrooms. Data collection was done bythe classroom or homeroom teachers or by one of the researchers.The belief instruments were administered first and were followedby the writing task. Scores from the most recent administration ofthe CAT were obtained from school records.

Data Analysis

We conducted multivariate analyses of variance (MANOVA)analyses and canonical correlation analyses using the MANOVAprocedure of SPSSPC, Version 3.1 (Norusis, 1988). Because gen-der differences in the beliefs studied have been identified in somestudies but not others (e.g., Stipek, 1993; Wagner et al., 1989), apreliminary 3 X 3 X 2 (Grade X Achievement Level X Gender)MANOVA was conducted to determine whether there were genderdifferences that interacted with any other effect. The MANOVArevealed no significant interactions between gender and the othereffects, although there was a significant gender main effect,Wilks's A = .86, Rao's F(18, 313) = 2.93, p < .001. Becausegender did not interact with grade or achievement level, genderwas collapsed across the other effects in all analyses.

Results

Means and standard deviations for the reading and writingachievement variables are provided in Table 1. MANOVAanalyses with pairwise follow-up tests (as described byLevin, Serlin, & Seaman, 1994, and explained further in thenext section) confirmed that all of these variables weresignificantly different for all pairs of grades and all pairs ofachievement levels.

Analysis of Differences in Beliefs

Means and standard deviations for the belief variables areprovided in Table 2. To examine grade-level and achieve-ment-level effects on beliefs and to determine whethergrade level and achievement level interacted, we conducteda 3 X 3 (Grade X Achievement Level) MANOVA with analpha level of .05. Significant main effects were found forgrade, Wilks's A = .56, Rao's F(36,676) = 6.37, p < .001,and achievement level, Wilks's A = .71, Rao's F(36,676) = 3.59, p < .001. The interaction effect was notsignificant, Wilks's A = .81, Rao's F(72, 1331) = 1.02,p = .43. These results indicated that there were (a) differ-ences among 4th, 7th, and 10th graders that were consistentacross ability levels and (b) differences among high, aver-age, and low achievers that were consistent across grade.

To examine the significant main effects, we used fol-low-up tests based on Fisher's least significant differenceprocedure with a familywise alpha of .05 (see Levin et al.,1994). For the significant main effects for grade andachievement level, we determined which groups were sig-nificantly different by conducting pairwise multivariate Ho-

390 D. SHELL, C. COLVIN, AND R. BRUNING

Table 1Means and Standard Deviations for Reading and Writing Achievement Variables

Variable

ReadingComprehension

MSD

VocabularyMSD

WritingWriting score

MSD

Language mechanicsMSD

Language expressionMSD

SpellingMSD

All

57394

55192

500100

58681

57174

57687

10th

64865

62971

56684

64167

61761

63472

Grade

7th

57559

54447

49087

59153

58253

57665

4th

46535

44931

41862

50143

49542

49456

High

623102

595105

56894

63887

61773

63686

Achievement

Average

57680

55478

49886

58164

57663

57173

Low

52074

50374

43580

54065

51854

52471

telling's T2 tests with an alpha of .05. For each significantpairwise multivariate difference, we determined which in-dividual variables contributed to the difference by conduct-ing univariate t tests using a Bonferroni adjustment tocontrol familywise alpha at .05 (see Table 2 for the exactBonferroni p values). We conducted t tests on only thosevariables that had pairwise effect sizes (d) of .25 or larger.This was done because we wished both to reduce the num-ber of comparisons done and to examine only those differ-ences that were likely to have practical meaningfulness.

Grade-level differences in beliefs. The pairwise multi-variate tests indicated significant differences between 10thand 7th graders, Hotelling's T2 = .19, F(18, 240) = 2.57,p < .001; 10th and 4th graders, Hotelling's T2 = .99, F(18,234) = 12.90, p < .001; and 7th and 4th graders, Hotell-ing's T2 = .52, F(18, 197) = 5.69, p < .001. Thus, eachgrade differed significantly from the other two grades. Theresults of the univariate follow-up tests for each beliefvariable are shown in Table 2. Fourth-grade students dif-fered from both 7th and 10th grade students in a number ofbeliefs. Compared with both 7th and 10th graders, 4thgraders had (a) lower task self-efficacy for both reading andwriting, (b) higher ratings of effort as a cause of success forwriting, and (c) higher ratings of intelligence as a cause ofsuccess for reading. Also, compared with 10th graders, 4thgraders had higher ratings of luck as a cause of success forboth reading and writing. Seventh-grade students differedfrom lOth-grade students only in having lower task self-efficacy. For task self-efficacy and attribution to luck, theresults were the same for reading and writing. For attribu-tions to effort and intelligence, the results were different forreading and writing; effort attributions were significant onlyfor writing, and intelligence attributions were significantonly for reading. There were similar trends in the means for

effort attributions for reading and intelligence attributionsfor writing, but these were not significant.

Achievement-level differences in beliefs. The pairwisemultivariate tests indicated significant differences betweenthe high- and average-achievement groups, Hotelling'sT2 = .18, F(18, 241) = 2.47, p < .001; the high- andlow-achievement groups, Hotelling's T2 = .69, F(18,189) = 7.21, p < .001; and the average- and low-achieve-ment groups, Hotelling's T2 = .20, F(18, 241) = 2.73, p <.001. Each achievement group, therefore, differed signifi-cantly from the other two achievement groups. The resultsof the univariate follow-up tests for each belief variable areshown in Table 2. High and low achievers differed substan-tially in their beliefs. Compared with low achievers, highachievers had (a) higher task and component skills self-efficacy for both reading and writing; (b) lower ratings ofluck, task difficulty, and teacher help as causes of successfor both reading and writing; (c) lower ratings of intelli-gence as a cause of success for reading; and (d) loweroutcome expectancy for reading and writing. Averageachievers tended to have scores between those of high andlow achievers; however, they differed more from lowachievers than from high achievers. Compared with lowachievers, average achievers had (a) higher componentskills self-efficacy for both reading and writing, (b) lowerratings of teacher help as a cause of success for both readingand writing, (c) lower ratings of task difficulty as a cause ofsuccess for reading, and (d) lower ratings of luck as a causeof success for writing. Compared with high achievers, av-erage achievers had only lower task and component skillsself-efficacy for both reading and writing. With small vari-ation in attributions for intelligence, luck, and task diffi-culty, the results were similar for both reading and writing.

BELIEF MECHANISMS IN READING AND WRITING ACHIEVEMENT 391

Table 2Means, Standard Deviations, and Results of Pairwise Comparisons for Belief Variables

Grade Achievement

Variable

ReadingTask efficacy

MSD

Skill efficacyMSD

Outcome expect.MSD

AttributionsEffort

MSD

IntelligenceMSD

EnjoymentMSD

LuckMSD

Task difficultyMSD

Teacher helpMSD

WritingTask efficacy

MSD

Skill efficacyMSD

Outcome expect.MSD

AttributionsEffort

MSD

IntelligenceMSD

EnjoymentMSD

LuckMSD

Task difficultyMSD

Teacher helpMSD

All

4.610.48

3.990.62

4.300.49

4.450.89

3.801.15

4.320.96

2.251.27

2.381.12

3.271.22

4.530.48

4.200.61

4.070.55

4.390.90

3.891.13

4.390.83

2.211.26

2.731.12

3.261.13

10th

4.84a0.34

3.990.63

4.24,0.45

4.39a0.99

3.66a1.13

4.320.90

2.07a1.19

2.451.16

3.351.15

4.70a0.36

4.240.60

4.04,0.48

4.31.0.95

3.861.03

4.430.85

2.01a1.18

2.691.12

3.401.05

7th

4.63a0.41

3.980.57

4.290.48

4.3520.82

3.67b1.17

4.321.02

2.14,1.20

2.381.04

3.141.21

4.53a0.47

4.150.61

4.0220.59

4.26b0.90

3.791.16

4.330.88

2.15,1.18

2.811.08

3.141.09

4th

4.26a0.50

4.010.67

4.39,0.55

4.64, 20.77

4-12a,b1.12

4.320.98

2.60a ,1.38'

2.291.14

3.291.31

4.31.0.56

4.190.62

4.19,20.57

4-64a,b0.79

4.031.22

4.390.75

2.55 a ,1.37'

2.691.15

3.181.25

High

4-76a>b0.33

4.26a0.49

4.16.,!0.49

4.460.87

3.58a1.01

4.45,0.85

1-87.11.02

2.01. i0.92

3.13a1.12

4-69ab0.33

4.46a0.45

3.94a0.53

4.370.97

3.69,1.02

4.380.78

1-85.!1.01 '

2.37a ,1.12

3.14.1.04

Average

4.60a>,0.48

3.99a0.62

4.33,0.47

4.440.91

3.73,1.27

4.360.94

2.28,1.25

2-33b,,1.07

3.12,,1.26

4-53a>10.50

4.20a0.54

4.06,0.55

4.380.91

3.901.21

4.420.89

2.15 b l1.18 '

2 .74, 21.08

3.10,,1.19

Low

4-47b)10.55

3.73a0.64

4.39a0.51

4.440.88

4.12a>11.05

4.13,1.06

2.59a1.42

2-84a>b1.22

3-64a>b1.17

4.39b>10.55

3.94a0.71

4-24a,0.53 '

4.430.83

4.08,1.09

4.360.80

2-67ab1.44'

3.06a21.07

3.60a>b1.07

Note. Within each grouping of grade or ability level across rows, means with the same alphabeticsubscript differ significantly at the following Bonferroni adjusted p values for pairwise tests ofvariables with effect size (d) greater than .25: Grades 4 and 7, p < .006; Grades 4 and 10, p < .006;Grades 7 and 10, p < .025; low and average achievement, p < .005; low and high achievement,p < .003; average and high achievement, p < .006. Within each grouping of grade or ability levelacross rows, means with the same numeric subscript have pairwise effect sizes (d) greater than .25and differ at p < .05 with no Bonferroni adjustment.

392 D. SHELL, C. COLVIN, AND R. BRUNING

Analysis of the Relations Between Beliefs andAchievement in Reading and Writing

Examination of quadratic curvilinearity. To test eachbelief variable for quadratic curvilinearity, we used a mul-tivariate equivalent to a powers approach in regression. Foreach belief variable, two canonical analyses were done onthe total sample. First, the original belief variable wascanonically correlated to the set of reading and writingachievement variables to establish the baseline linear rela-tion. Second, the original belief variable (linear component)and its square (quadratic component) were canonically cor-related to the same set of achievement variables to identifythe quadratic curvilinear relation. The results of these twoanalyses were then compared to determine whether thequadratic canonical correlation was meaningfully higherthan the linear-only correlation, on the basis of the signifi-cance of the quadratic relation and the increase in explainedvariance due to the addition of the quadratic component.

From these analyses we identified four variables as hav-ing meaningful quadratic curvilinear relations to achieve-ment: reading outcome expectancy, writing outcome ex-pectancy, reading attribution to intelligence, and writingattribution to intelligence. For these four beliefs, the linearcanonical correlations were .24, .25, .30, and .17, respec-tively, and the quadratic canonical correlations were .37,.31, .38, and .28, respectively. The increase in explainedvariance was 7.5%, 3.6%, 5.2%, and 5.2%, respectively.The quadratic curvilinear trends for each variable werecomputed from the canonical coefficients. For outcomeexpectancy for both reading and writing, higher ratings wereassociated with higher achievement up to a rating of ap-proximately 4, beyond which, higher ratings were associ-ated with lower achievement. For attributions to intelligencefor both reading and writing, higher ratings were associatedwith higher achievement up to a rating of approximately 3,beyond which, higher ratings were associated with lowerachievement.

The quadratic analyses were done with the total sample.However, we did examine the extent to which similar resultsoccurred in each grade- and achievement-level group. Be-cause of reduced sample size in the specific groups, corre-lations of the magnitudes identified for these variables (.20to .40) were not always significant at the conventional alphalevel of .05; however, for all four beliefs, the comparisonsbetween the quadratic and linear analyses produced resultssimilar to those obtained for the total sample. For bothoutcome expectancy variables and reading attributions tointelligence, there were increases in the canonical correla-tions of .04 or greater in five of the six groups and increasesin explained variance of 2.5% or greater in four or more ofthe six groups. For writing attributions to intelligence, therewas an increase in the canonical correlation of .05 or greaterand an increase in explained variance of 2.5% or greater inthree of the six groups. Also, the quadratic trends computedfrom the canonical coefficients for each specific grade- andachievement-level group were similar to the trend obtainedfor the total sample.

The quadratic curvilinear relation is represented by thecombination of both the linear (original variable) and qua-dratic components (the square of the variable). Becausecanonical analysis does not provide a way to link these twocomponents together so that a single structure coefficient isobtained for the combined effect, we used an approachsimilar to that of Shell et al. (1989) to obtain a single valuefor the quadratic curvilinear relation in subsequent canoni-cal analyses. Predicted scores for reading outcome expect-ancy, writing outcome expectancy, reading intelligence at-tributions, and writing intelligence attributions werecomputed from the canonical coefficients obtained fromtheir respective quadratic canonical analyses. These pre-dicted scores then were used in place of the original vari-ables in all subsequent canonical analyses. We used thecanonical coefficients from the analyses done with the totalsample because these would be more stable than thoseobtained in each grade- and achievement-level group be-cause of the larger sample size. We also felt that the use ofa common predicted score in all analyses would facilitatecomparisons across different groups. We verified that thepredicted scores based on the total sample were accuraterepresentations of the specific quadratic relations in eachgrade and achievement group by comparing, in each group,(a) the results of a canonical analysis done with the commonpredicted scores and (b) results obtained from an analysisdone with predicted scores based on the canonical coeffi-cients obtained for that specific group. These comparisonsindicated that, in all groups, the magnitude of the overallcanonical correlation and variable loadings for all variableswere equivalent.

Canonical analyses. To examine how the relations be-tween beliefs and achievement changed as a function ofgrade level and achievement, canonical correlation analyseswere conducted for each grade (10th, 7th, and 4th) and eachachievement group (high, average, and low). For the canon-ical analyses, the reading and writing achievement variablesconstituted the dependent variable set, and the self-efficacy,attribution, and outcome expectancy belief variables consti-tuted the independent variable set.

The results of the canonical analyses are summarized inTable 3. In all groups, a single-dimensional canonical rela-tion was identified. Like the findings of Shell et al. (1989),this suggests that beliefs and achievement in reading andwriting were linked by a single underlying dimension. Theeffects of entering outcome expectancy and attributions tointelligence in quadratic form can be determined by com-paring (a) the results obtained with all variables in linearform and (b) the results obtained with the four variablesentered as quadratic predicted scores. In all groups except7th grade, there was an increase in the explained achieve-ment variance (Rc

2) of 3% to 4% when the quadratic pre-dicted scores were used, suggesting that inclusion of thequadratic components for the four variables resulted inmeaningful increases in explained achievement variance.

In the low-achievement group, two significant (p < .05)canonical correlations were found; however, the averagevariance in the achievement variable set accounted for bythe second canonical variate was extremely low (6.7%),

BELIEF MECHANISMS IN READING AND WRITING ACHIEVEMENT 393

Table 3Summary Statistics for Canonical Correlation Analysis

Group

10th gradeLinearQuadratic

7th gradeLinearQuadratic

4th gradeLinearQuadratic

High achievementLinearQuadratic

Average achievementLinearQuadratic

Low achievementLinearQuadratic

Rc

.63

.66

.64

.64

.67

.70

.79

.82

.69

.72

.78

.80

.39

.43

.41

.41

.44

.48

.62

.66

.47

.51

.61

.64

Wilks'sA

.366

.310

.280

.279

.259

.234

.143

.131

.291

.250

.114

.102

Rao'sF

1.271.51

1.161.17

1.161.26

1.741.84

1.701.94

1.982.11

df

108, 718108, 718

108, 506108, 506

108, 471108, 471

108, 466108, 466

108, 764108, 764

108, 466108, 466

P

.04

.001

.15

.14

.15

.06

<.001<.001

<.001<.001

<.001<.001

Averagevariance8

.53

.54

.35

.39

.39

.42

.73

.73

.62

.62

.59

.591 Average of the squared structure coefficients (loadings) for the achievement variable set.

suggesting that the second canonical variate did not providea good representation of the achievement variables. Also,redundancy analysis (Pedhazur, 1982) indicated that thesecond canonical correlation accounted for no meaningful,nonredundant variance in the set of achievement variables.As a result, we retained only the first canonical correlationfor analysis. The canonical correlations for the 4th-gradeand 7th-grade groups did not achieve significance at the p <.05 level. In both groups, however, the magnitude of thecanonical correlation, the Wilks's A, the average variance inthe achievement variable set accounted for by the canonicalvariate, and the results of redundancy analysis were consis-tent with those of the other grade and achievement groups.We therefore used the 4th- and 7th-grade results in subse-quent analyses, with the caution that structures of the ca-nonical relations for these grades may be less stable thanthose of other groups.

The structure of the identified dimension is described bythe correlations between the original variables and theirrespective canonical variates (the structure coefficients orloadings). These loadings describe the nature of the canon-ical relation between beliefs and achievement and are pro-vided in Table 4.

Grade-level differences in the relation between beliefsand achievement. With the exception of spelling and lan-guage mechanics in the 7th grade, all achievement variableshad loadings above .40, indicating that the canonical vari-ates in each grade provided a good representation of stu-dents' reading and writing achievement. All belief variablesexcept attributions to effort and enjoyment had meaningfulloadings (above .30) in at least two grades, with reading andwriting beliefs having nearly equivalent loadings.

For all grades, the basic structures of the canonical rela-tions were similar. Higher reading and writing achievementwere related to (a) higher self-efficacy for both tasks andskills, (b) the curvilinear pattern of attribution to intelli-

gence, (c) lower attribution to external causes, and (d) thecurvilinear pattern of outcome expectancy. There were,however, some differences across grades. For achievement,there was an increase in the predictability of reading com-prehension as grade increased. Also, there was a drop in thepredictability of the holistic writing score in 10th grade.Vocabulary, language mechanics, and spelling all weremore poorly predicted in 7th than in either 10th or 4th grade.For beliefs, there was a decrease in the contribution ofself-efficacy for reading in 10th grade, although this mayhave been related to ceiling effects on the efficacy ratings in10th grade (M = 4.84 out of 5.0). Also, attributions tointelligence for both reading and writing and to teacher helpfor reading did not contribute as strongly in 7th grade as ineither 10th or 4th grade. In 4th grade, attributions to taskdifficulty and teacher help for writing did not contribute.

Achievement-level differences in the relation between be-liefs and achievement. All achievement variables hadloadings above .57, indicating that the canonical variates ateach achievement level provided a good representation ofstudents' reading and writing achievement. Only a smallnumber of belief variables contributed to the relations. Self-efficacy beliefs for tasks had loadings above .60 for readingand above .40 for writing in all groups; attributions tointelligence and attributions to luck for both reading andwriting had loadings above .30 in at least two groups; andoutcome expectancy had loadings above .50 for reading intwo groups and above .40 for writing in one group. Al-though the number of belief variables contributing to thecanonical relations was small, the relation of the beliefvariable set to achievement was high in all achievement-level groups (see Table 3, column 1). The pattern of load-ings was similar for both reading and writing beliefs, al-though the loadings for task self-efficacy and outcomeexpectancy were somewhat higher for reading than forwriting. Also, the loadings for task self-efficacy were con-

394 D. SHELL, C. COLVIN, AND R. BRUNING

Table 4Correlations Between Variables and Their Respective Canonical Variates

Variable

ReadingComprehensionVocabulary

WritingWriting scoreLanguage mechanicsLanguage expressionSpelling

ReadingTask efficacySkill efficacyOutcome expectancy"Causal attributions

EffortIntelligence8

EnjoymentLuckTask difficultyTeacher help

WritingTask efficacySkill efficacyOutcome expectancy"Causal attributions

EffortIntelligence"EnjoymentLuckTask difficultyTeacher help

10th

Grade

7th 4th

Achievement variables

.97

.77

.47

.70

.85

.53

.76

.58

.73

.36

.83

.28

Belief variables

.24

.33

.45

- .02.63.16

- .38- .43- .36

.40

.54

.30

- .16.38.02

- .53- .46- .37

.59

.45

.58

.12

.17

.17-.44

-!04

.39

.71

.46

-.01.24.03

- .54-.39-.30

.58

.75

.75

.56

.76

.41

.58

.65

.39

.18

.37

.29- .42- .37- .41

.56

.42

.35

.14

.41

.10- .41-.14-.19

High

.94

.93

.59

.83

.92

.86

.70-.10

.27

- .20.39

-.20-.18

.20-.02

.44

.14

.10

- .27.31

- .03- .21-.07

.13

Achievement

Average

.86

.77

.75

.79

.91

.60

.71- .08

.51

-.12.33.06

- .31- .17

.04

.52

.05

.48

-.16.18.03

- .43- .14- .08

Low

.92

.90

.61

.84

.70

.57

.64

.05

.51

-.18.28

- .07- .33

.19

.01

.47

.10

.28

- .06.34.15

- .38.16.02

' Variable was entered as a quadratic predicted score.

sistently higher at all achievement levels than the loadingsfor either outcome expectancy or attributions, indicatingthat these self-efficacy beliefs were more strongly repre-sented in the canonical relation.

The structures of the canonical relations were similar forthe low- and average-achievement groups but differed forthe high-achievement group. For the low- and average-achievement groups, higher reading and writing achieve-ment were related to (a) higher self-efficacy for tasks, (b)the curvilinear pattern of attribution to intelligence, (c)lower attribution to luck, and (d) the curvilinear pattern ofoutcome expectancy. For the high-achievement group,higher reading and writing achievement were related only tohigher self-efficacy for tasks and the curvilinear pattern ofattribution to intelligence. There also were some other dif-ferences across the achievement levels. For achievement,writing variables, except for language mechanics, tended tobe more poorly predicted in the low-achievement groupthan in either the average- or high-achievement groups.There also was an increase in the predictability of theholistic writing score and a decrease in the predictability ofreading variables in the average group relative to the highand low groups. For beliefs, attribution to intelligence for

writing did not contribute to the relation for the averagegroup.

Discussion

Grade-Level and Achievement-LevelDifferences in Beliefs

The results help clarify the nature of grade-level andachievement-level differences in beliefs for reading andwriting. Significant main effects were found for both gradeand achievement, indicating that there were differences inbeliefs related to both of these factors. However, no signif-icant interaction was found, indicating that the grade-leveldifferences were distinct from the achievement-level differ-ences. These findings suggest that the beliefs characteristicof particular levels of achievement are not simply reflec-tions of students' age or grade level, and the beliefs char-acteristic of particular ages or grade levels are not merelyartifacts of knowledge and skill mastery. This implies thatthe beliefs and subsequent motivation of low-achievingreaders or writers are not necessarily the same as those of

BELIEF MECHANISMS IN READING AND WRITING ACHIEVEMENT 395

younger readers and writers, even if the performance of lowachievers is similar to that of younger students.

Grade-level changes and achievement-level differenceswere similar for reading and writing beliefs. This suggests,consistent with research indicating that reading and writingskills codevelop (e.g., Shanahan, 1984; Shanahan & Lomax,1986), that beliefs about reading and beliefs about writingfollow similar developmental courses. Also, this indicatesthat the patterns of beliefs typical of high- or low-achievingwriters are similar to those of high- or low-achievingreaders.

Grade-level differences in beliefs. The present studyprovided an important extension to the previous researchconcerning the development of children's motivational be-liefs (e.g., Hiebert et al., 1984; Paris & Oka, 1986; Stipek,1993; Wagner et al., 1989) by continuing the examination ofgrade-level differences through high school. The resultsindicated that not only were there no significant differencesbetween 7th and 10th graders in outcome expectancy andcausal attribution beliefs, but there were no effect sizes of.25 or greater. These findings suggest that by junior highschool, children have developed generally stable percep-tions of causality and outcome expectancy that undergolittle subsequent change through high school.

The only significant differences in either causal attribu-tions or outcome expectancy beliefs found between 4th and7th graders were for causal attributions to effort and intel-ligence. These beliefs have received the most emphasis indiscussions of developmental change in causal attributions(e.g., see Stipek, 1993; Weiner, 1985), and our findingssubstantiate that the changes in causal beliefs that occur inthe late elementary grades primarily involve changes inperceptions of the importance of effort and intelligence orability. For both effort and intelligence, the ratings of theirimportance were lower for 7th than for 4th graders. Theseresults are consistent with those of previous studies (e.g.,see Hiebert et al., 1984; Stipek, 1993). Decreases in chil-dren's attribution to effort and ability have been interpretedas indicating that these beliefs become more accurate aschildren age because the high ratings younger children giveto these causes are thought to overstate their likely actualcausal influence (e.g., see discussion in Stipek, 1993). Ourresults, therefore, also can be interpreted as indicating thatchildren in late elementary school are continuing to developincreased accuracy in their beliefs about the causal role ofeffort and intelligence in reading and writing.

Attributions to luck had a somewhat anomalous pattern ofdifferences. Although 10th graders significantly differedfrom 4th graders, 7th graders did not significantly differfrom children in either of the other grades. However, theeffect sizes (see Table 2) between 7th and 10th grade weremuch smaller (.06 for reading and .12 for writing) than theeffect sizes between 7th and 4th grade (.35 for reading and.31 for writing). The pattern of effect sizes and direction ofdifferences (lower ratings in higher grades) were similar tothose found for effort and intelligence attributions suggest-ing that luck attributions follow a course of developmentsimilar to effort and intelligence attributions. The failure tofind a significant difference between 4th and 7th grade,

however, indicates a need for further examination of thedevelopment of luck attributions.

Although outcome expectancy beliefs did not differ sig-nificantly between any of the three grades, there were effectsizes above .25 between 4th and 10th graders for readingand between 4th and both 7th and 10th graders for writing(see Table 2). These patterns were similar to those identifiedfor effort, intelligence, and luck attributions, suggestingpossible grade-level differences in outcome expectancy be-liefs similar to those found for these attributions. Our resultscannot confirm the presence of these differences, but theydo suggest that further study of the development of outcomeexpectancy beliefs is warranted.

Self-efficacy beliefs exhibited a different pattern ofgrade-level differences from causal attribution and outcomeexpectancy beliefs. Unlike these beliefs, task self-efficacywas significantly higher at each successive grade level.Bandura (1986) has proposed that the development ofhigher self-efficacy is directly associated with improvementin actual cognitive and behavioral skills and, therefore, canbe expected to continue to develop as long as these skills aredeveloping. Our findings support Bandura's claim, becausethe grade-level increase in task self-efficacy through highschool mirrored a grade-level increase in students' actualreading and writing skills (Table 1). Interestingly, grade-level differences occurred only for task self-efficacy; thescores for component skills self-efficacy were basically thesame in each grade. This suggests that the aspect of self-efficacy that develops across grades is belief in the capa-bility of successfully reading various types of materials andsuccessfully communicating in various formats rather thanbelief in the capability of doing specific reading or writingskills.

Achievement-level differences in beliefs. Consistentwith previous studies (Bandura, 1986; Hiebert et al., 1984;Schunk & Rice, 1991; Schunk & Swartz, 1993; Weiner,1985), we found substantial differences between the beliefsof high and low achievers. The large number of beliefdifferences identified suggests that belief differences be-tween high and low achievers are pervasive, encompassingall aspects of belief about personal control. Relative to highachievers, low achievers exhibited a potentially dysfunc-tional belief pattern of ascribing higher outcome expectancyto reading and writing while simultaneously expressinglower self-efficacy for their reading and writing and ascrib-ing higher causality to factors that are external or uncon-trollable. As noted in previous formulations of the effects ofbeliefs (e.g., Bandura, 1986; Schunk, 1991; Stipek, 1993;Weiner, 1985), this pattern of beliefs can have a strongnegative impact both on motivation and on feelings ofself-worth.

The scores of average achievers fell between those ofhigh and low achievers, but their beliefs generally weremore like those of high achievers. As with the differencesfound between 10th and 7th grade, the primary differencebetween high and average achievers was in self-efficacy.This substantiates previous findings that self-efficacy be-liefs are strongly linked to achievement and achievement-level differences (Bandura, 1986; Schunk, 1991; Schunk &

396 D. SHELL, C. COLVIN, AND R. BRUNING

Rice, 1991; Schunk & Swartz, 1993). Also, there werenonsignificant effect sizes above .25 between high andaverage achievers for both luck and task difficulty (seeTable 2), suggesting a need for further research on whetherbeliefs about external causality also play a role in differen-tiating high and average achievers.

As previously has been found (e.g., Hiebert et al., 1984;Weiner, 1985), low achievers thought that effort was asimportant as average and high achievers thought. Lowachievers, however, simultaneously ascribed causal influ-ence to a greater number of external and uncontrollablecauses. These findings suggest that low achievers may havea more contingent causality for effort. In essence, effort willcause success if "the task is easy enough," "I receive enoughhelp," "I am lucky," or "I am smart enough." If any of theseother causal conditions are not met, the positive effects ofattribution to effort could be undermined. As Stipek (1993)notes, this contingent causality may, in fact, be real for lowachievers, because they tend to experience more tasks thatare beyond their reading and writing skill level and mayreceive more teacher help. Thus, effort may not help thesestudents succeed unless other conditions are met. This sug-gests that to maintain positive attributions to effort andpositive motivation among low achievers, reading and writ-ing tasks need to be structured so that low achievers canachieve success through their efforts.

Relation Between Beliefs and Achievement

The results confirmed the importance of examining thetwo factors first identified by Shell et al. (1989): (a) gener-alized across-domain dimensionality of beliefs and achieve-ment and (b) quadratic curvilinearity in the relationsbetween beliefs and achievement. A single underlying ca-nonical dimension linking beliefs and achievement wasfound for all grades and achievement levels, indicating thatbeliefs and achievement in the domains of reading andwriting have a generalized reciprocal relation to each other.The pervasiveness of these findings across grade andachievement levels suggests that this dimensionality is afundamental property of the relations between beliefs andachievement in reading and writing. This indicates thatconsideration of possible reciprocal influences from beliefsin both domains will be necessary for full understanding ofthe motivational influences of beliefs on reading or writing.Also, approaches to literacy instruction that emphasize theconnections between reading and writing may be able to usethe reciprocal relations of beliefs to achievement in the twodomains to enhance motivation for both reading andwriting.

The quadratic curvilinear trends identified for reading andwriting outcome expectancy and reading and writing intel-ligence attributions suggest that there is a ceiling effect onthe positive contributions of these variables to motivationand achievement. Attributing either too little or too muchcausal influence to intelligence or having either too low ortoo high outcome expectancy appears to be associated withlower achievement and may be associated with lower mo-

tivation as well. Only at moderate levels do there appear tobe positive contributions from these beliefs. This suggestscaution in how attributional and outcome expectancy feed-back are delivered in classrooms. Feedback that over- orunderstresses either the importance of intelligence in stu-dent achievement or the contingencies between reading andwriting and the attainment of life goals may have negativemotivational and achievement consequences. These find-ings also suggest that previous studies, in which only linearrelations were examined (e.g., Ehrlich et al., 1993; Mc-Carthy et al., 1985; Paris & Oka, 1986), have likely under-estimated the contributions of these beliefs to explainingvariance in achievement.

Grade-level changes in the relations between beliefs andachievement. Unlike previous studies (e.g., Paris & Oka,1986; Wagner et al., 1989), neither the strength of therelation between beliefs and achievement nor the pattern ofbeliefs predictive of higher achievement differed substan-tially across grades. In all three grades, beliefs accounted forsimilar amounts of variance in achievement (Table 3). Also,although there were some differences in the magnitudes ofthe loadings for specific beliefs, in all grades the overallstructures of the canonical dimension for beliefs were sim-ilar. Higher achievement was associated with higher self-efficacy, lower attribution to external causes, and the cur-vilinear patterns of attribution to intelligence and outcomeexpectancy. These findings indicated that there were nomajor grade-level changes in either the types of beliefs thatwere related to higher achievement or the strength of therelations between these beliefs and achievement for thedomains of reading and writing. Additionally, counter to thefindings of Shell et al. (1989) that beliefs were less stronglyrelated to writing than to reading, there was equivalentprediction of reading and writing achievement in all grades.

The differences between the findings in the present studyand those of previous studies appear to be due to themultivariate assessment of reading and writing achievementthat we used. Although beliefs predicted similar overalllevels of achievement variance in all grades, the particularreading and writing variables that were best predictedchanged. Specifically, there was a difference between read-ing and writing in the predictability of higher order com-prehension or communication skills versus component sub-skills. Reading comprehension became better predictedrelative to vocabulary across grades, whereas the holisticwriting score was more poorly predicted relative to lan-guage mechanics, language expression, and spelling in 10thcompared with 4th grade. These trends suggest that, asstudents age, their beliefs for reading become more predic-tive of higher order comprehension skills, whereas theirbeliefs for writing become more predictive of componentsubskills.

These trends may be an artifact of literacy instruction.Instruction in reading and writing tends to shift from a focusin the early grades on development of reading and writingcomponent subskills to a focus in the higher grades on useof reading and writing as tools for learning or expressingideas. Thus, in elementary school, students' views of them-selves as a reader or writer and their beliefs about reading

BELIEF MECHANISMS IN READING AND WRITING ACHIEVEMENT 397

and writing may be based on an evaluation of how well theyexecute component subskills, whereas in high school, stu-dents' views and beliefs may be based more on an evalua-tion of how well they comprehend what they read andwhether they communicate effectively in their writing.

The shift in instructional focus, however, may differsomewhat for reading and writing. By high school, there islittle direct emphasis on or assessment of reading compo-nent subskills. For example, except for special remedialclasses, there are no courses devoted to teaching reading perse, and students do not practice specific reading subskills intheir courses. Student reading is done primarily in subjectmatter classes for the purpose of learning. However, therestill may be courses in writing, such as composition orcreative writing, that include writing skill components.Also, in addition to comments on content, teacher feedbackon students' written work is likely to include commentsabout spelling, mechanics such as punctuation and gram-mar, and proper word use or phrasing, and these maycompose a substantial part of students' grades. Thus, evenin high school, students may continue to hold a view ofgood writing that includes an emphasis on component skillproficiency. This would account for their beliefs aboutwriting being more related to component subskills relativeto higher order communication skill. More study, however,is needed to determine how instructional experiences influ-ence the development of students' beliefs about readingand writing and the relations between these beliefs andachievement.

Achievement-level differences in the relations betweenbeliefs and achievement. In the high-achievement group,higher achievement was related only to higher self-efficacyand the curvilinear trend for attributions to intelligence,whereas at the other achievement levels, achievement alsowas related to lower attributions for luck and the curvilineartrend for outcome expectancy. These results, like those ofprevious studies (e.g., Bandura, 1986; Schunk, 1991; Sti-pek, 1993), confirm the importance of self-efficacy beliefsand beliefs about ability for motivating and sustainingachievement. The findings also suggest that self-efficacyand causal beliefs about intelligence may be especiallypotent for identifying individual differences in motivationand achievement for those at higher achievement levels.

The results substantiate the importance of maintainingpositive motivational beliefs even when achievement is low.Among low achievers, those who had (a) higher self-efficacy, (b) more positive outcome expectancy, (c) a real-istic, moderate view of the causal influence of intelligence,and (d) lower attribution to luck as a cause of success hadhigher achievement. Placing special emphasis on fosteringmore positive patterns of these beliefs in instruction andother interventions designed for low achievers, therefore,may be useful for enhancing motivation.

The finding that outcome expectancy was not related toachievement for high achievers suggests that beliefs aboutoutcome contingencies may play a smaller role in the mo-tivation of high achievers. As defined in the present study,outcome expectancy reflects the importance of reading andwriting as means for achieving other ends or goals (e.g., see

Bandura, 1986; Stipek, 1993). As suggested by high achiev-ers' lower ratings of outcome expectancy (Table 2), thistype of tool or utility value may be less important for highachievers than other more intrinsic motivations. It seemsreasonable that persons who read or write very well wouldbe more likely to derive intrinsic satisfaction from theseactivities or derive intrinsic motivation from their compe-tency (see Stipek, 1993, for a complete discussion of thesetopics). Thus, the attainment of other outcomes may be lessmotivating for high achievers, although more study isclearly needed to substantiate this speculation.

Conclusions and Directions for Future Research

The results suggest that self-efficacy, causal attribution,and outcome expectancy beliefs exert potentially importantmotivational influences on children's reading and writing.Consideration of these possible influences is necessary,therefore, for full understanding of how reading and writingskills develop and for the creation of effective reading andwriting instruction.

Although beliefs are related to reading and writingachievement, beliefs themselves cannot directly causeachievement. The next research step needs to be examina-tion of how beliefs affect the cognitive processes that arecausally related to achievement. Schunk and Swartz (1993)and Schunk and Rice (1991) have found that self-efficacy isassociated with the use of specific reading and writingstrategies. Their research, however, constitutes only a be-ginning of this examination. Also, because self-efficacy,causal attribution, and outcome expectancy beliefs presum-ably exert their influence by affecting motivation (e.g., seeBandura, 1986; Schunk, 1991; Stipek, 1993; Weiner, 1985;Zimmerman, 1989), there is a need to examine more di-rectly how beliefs relate to measures of motivation such aseffort and persistence and how these motivational influ-ences subsequently affect and are themselves affected bythe strategies and knowledge used during reading and writ-ing. Finally, there are additional beliefs (e.g., see Schunk,1991; Stipek, 1993), such as goals for reading and writing orattributions to factors such as strategy use, that need to beexamined and integrated with those considered in our study.

References

Bandura, A. (1986). Social foundations of thought and action: Asocial cognitive theory. Englewood Cliffs, NJ: Prentice Hall.

Ehrlich, M. F., Kurtz-Costes, B., & Loridant, C. (1993). Cognitiveand motivational determinants of reading comprehension ingood and poor readers. Journal of Reading Behavior, 25, 365-381.

Hiebert, E. H., Winograd, P. N., & Danner, F. W. (1984). Chil-dren's attributions for failure and success in different aspects ofreading. Journal of Educational Psychology, 76, 1139-1148.

Levin, J. R., Serlin, R. C, & Seaman, M. A. (1994). A controlled,powerful multiple-comparison strategy for several situations.Psychological Bulletin, 115, 153-159.

McCarthy, P., Meier, S., & Rinderer, R. (1985). Self-efficacy andwriting: A different view of self-evaluation. College Composi-tion and Communication, 36, 465-471.

398 D. SHELL, C. COLVIN, AND R. BRUNING

Norusis, M. J. (1988). SPSS/PC+ Advanced Statistics™ V2.0.Chicago, IL: SPSS.

Paris, S. G., & Oka, E. R. (1986). Children's reading strategies,metacognition and motivation. Developmental Review, 6, 25-56.

Pedhazur, E. J. (1982). Multiple regression in behavioral research(2nd ed.). New York: Holt, Rinehart, & Winston.

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

Schunk, D. H., & Rice, J. M. (1991). Learning goals and progressfeedback during reading comprehension instruction. Journal ofReading Behavior, 23, 351-364.

Schunk, D. H., & Swartz, C. W. (1993). Goals and progressfeedback: Effects on self-efficacy and writing achievement.Contemporary Educational Psychology, 18, 337-354.

Shanahan, T. (1984). Nature of the reading-writing relation: Anexploratory multivariate analysis. Journal of Educational Psy-chology, 76, 466-477.

Shanahan, T., & Lomax, R. G. (1986). An analysis and comparisonof theoretical models of the reading-writing relationship. Jour-nal of Educational Psychology, 78, 116-123.

Shell, D. F., Murphy, C. C , & Bruning, R. H. (1989). Self-efficacyand outcome expectancy mechanisms in reading and writingachievement. Journal of Educational Psychology, 81, 91-100.

Stipek, D. J. (1993). Motivation to learn: From theory to practice(2nd ed.). Needham Heights, MA: Allyn & Bacon.

Wagner, D. A., Spratt, J. E., Gal, I., & Paris, S. G. (1989). Readingand believing: Beliefs, attributions, and reading achievement inMoroccan schoolchildren. Journal of Educational Psychology,81, 283-293.

Weiner, B. (1985). An attributional theory of achievement moti-vation and emotion. Psychological Review, 92, 548-573.

White, E. M. (1985). Teaching and assessing writing. San Fran-cisco, CA: Jossey-Bass.

Zimmerman, B. J. (1989). A social cognitive view of self-regu-lated academic learning. Journal of Educational Psychology, 81,329-339.

Received November 29, 1993Revision received February 9, 1995

Accepted March 7, 1995 •

AMERICAN PSYCHOLOGICAL ASSOCIATION

SUBSCRIPTION CLAIMS INFORMATION Today's Date:.

We provide this form to assist members, institutions, and nonmember individuals with any subscription problems. With theappropriate information we can begin a resolution. If you use the services of an agent, please do NOT duplicate claims throughthem and directly to us. PLEASE PRINT CLEARLY AND IN INK D7 POSSIBLE.

PRINT FULL NAME OR KEY NAME OF INSTITUTION MEMBER OR CUSTOMER NUMBER (MA Y BE FOUND ON ANY PAST ISSUE LABEL)

DATE YOUR ORDER WAS MAILED (OR PHONED)

_CHAROECHECK/CARD CLEARED DATE:_

CITY STATE/COUNTRY

YOUR NAME AND PHONE NUMBER

TITLE

(If possible, send a copy, front and back, of your cancelled check to help us in our researchof your claim.)

ISSUES: MISSING DAMAGED

VOLUME OR YEAR NUMBER OR MONTH

Thank you. Once a claim is received and resolved, delivery of replacement issues routinely takes 4-6 weeks.

— — — ^ — ^ ^ — (TO BE FILLED OUT BY APA STAFF) — — — — — — — — —

DATE RECEIVED:.ACTION TAKEN: _STAFF NAME:

DATE OF ACTION: _INV.NO.&DATE:LABEL NO. & DATE:

Send this form to APA Subscription Claims, 750 First Street, NE, Washington, DC 20002-4242

PLEASE DO NOT REMOVE. A PHOTOCOPY MAY BE USED.