A teacher-report measure of children's task-avoidant behavior: A validation study of the Behavioral Strategy Rating Scale

Page 1

Learning and Individual Differences 21 (2011) 690–698

Contents lists available at SciVerse ScienceDirect

Learning and Individual Differences

j ourna l homepage: www.e lsev ie r .com/ locate / l ind i f

A teacher-report measure of children's task-avoidant behavior: A validation study ofthe Behavioral Strategy Rating Scale☆

Xiao Zhang a,b,⁎, Jari-Erik Nurmi b, Noona Kiuru b, Marja-Kristiina Lerkkanen c, Kaisa Aunola b

a Beijing Key Lab of Applied Experimental Psychology, School of Psychology, Beijing Normal University, Chinab Department of Psychology, University of Jyväskylä, Finlandc Department of Teacher Education, University of Jyväskylä, Finland

☆ This study was funded by a grant from the NationalChina to Xiao Zhang (No. 31100751), a grant from the AcCenter of Excellence in Learning and Motivation ResearcNoona Kiuru (SA 7133146).⁎ Corresponding author at: Beijing Key Lab of Applied E

of Psychology, Beijing Normal University, 100875 Beijingchology, P.O. Box 35, 40014 University of Jyväskylä, Finlan

E-mail address: [email protected] (X. Zhang).

1041-6080/$ – see front matter © 2011 Elsevier Inc. Alldoi:10.1016/j.lindif.2011.09.007

a b s t r a c t

a r t i c l e i n f o

Article history:Received 8 December 2010Received in revised form 12 September 2011Accepted 13 September 2011

Keywords:EffortMotivationOff-task behaviorTask avoidanceTask persistenceTask-focused behaviorMethod effectTeacher report

This study aims to validate a teacher-report measure of children's task-avoidant behavior, namely the Behav-ioral Strategy Rating Scale (BSRS), in a sample of 352 Finnish children. In each of the four waves from Kinder-garten to Grade 2, teachers rated children's task-avoidant behavior using the BSRS, children completedreading and mathematics tests, and trained testers rated children's task-avoidant and social-dependent be-havior after the test situation. Mothers also rated children's task-avoidant behavior in the last two waves.The results showed that a two-factor model including one factor representing task avoidance and one meth-od factor accounting for wording effects among the negatively phrased items fitted the data best. The scaledemonstrated strict measurement invariance across all four waves and satisfactory convergent and discrim-inant validity. Teacher-reported task avoidance was negatively associated with performance in reading andmathematics. The results suggest that the BSRS is a reliable, valid, and developmentally suitable instrument.

Natural Science Foundation ofademy of Finland to the Finnishh (Nr. 213486) and a grant to

xperimental Psychology, School, China, and Department of Psy-d.

rights reserved.

© 2011 Elsevier Inc. All rights reserved.

Students' behavior in achievement settings has an impact on theirfuture academic performance (Diener & Dweck, 1978; Grimm, Steele,Mashburn, Burchinal, & Pianta, 2010). Different behavioral patternshave been described in the literature: some students show engage-ment, persistence, and active effort in challenging tasks, while othersresist challenging situations, avoid difficult tasks, and withdraw effortin the face of failure (Cain & Dweck, 1995; Turner et al., 2002). Suchtask-focused and task-avoidant behaviors have been linked to variousantecedents, including achievement goals (Dweck & Leggett, 1988),efficacy beliefs (Bandura, 1997), and task value (Wigfield, 1994). Al-though these behaviors have been widely investigated, systematic re-search on the validity of related measures is scarce. To fill this gap thepresent study examined the validity and reliability of a teacher-reportmeasure of children's task-avoidant behavior, namely the BehavioralStrategy Rating Scale (Aunola, Nurmi, Parrila, & Onatsu-Arvilommi,2000).

1. Task-avoidant behavior

The theoretical basis of investigating task avoidance originates froman interest in the role of dynamic mechanisms in learning. Task avoid-ance refers to maladaptive behaviors that students display in responseto academic difficulties or challenges (Aunola, Nurmi, Niemi, Lerkkanen,& Rasku-Puttonen, 2002; Onatsu-Arvilommi&Nurmi, 2000).Many sim-ilar, though not identical, constructs, such as self-handicapping (Berglas& Jones, 1978), off-task behavior (Nottelmann & Hill, 1977), and workavoidance (Seifert & O'Keefe, 2001), have been introduced in the field.Such behaviors are typified by decreased task enjoyment (Gottfried,1990), low concentration (Weinstein, Schulte, & Palmer, 1987), low ef-fort (Dweck & Leggett, 1988), and low persistence (Cain & Dweck,1995). Yet not all students engage in task-avoidant behavior; someapply task-focused strategies showing persistence and engagement inchallenging situations.

Task avoidance may originate from several sources. First, it maydevelop because students have experienced repeated failures whichlead to a sense of low self-efficacy (Bandura, 1997) or helplessnessbeliefs (Diener & Dweck, 1978). On the basis of such beliefs, studentschoose to withdraw effort from the task (Bandura, 1997; Diener &Dweck, 1978), evidenced in “passive task avoidance.” Second, stu-dents may also actively avoid challenging tasks. One reason for such“active task avoidance” is that trying hard and failing in the task iscompelling evidence of low ability (Covington, 1984). Such avoidance

Page 2

691X. Zhang et al. / Learning and Individual Differences 21 (2011) 690–698

may protect students from negative perceptions of self-worth but isalso likely to undermine achievement (Covington, 1984). Similarly,avoidant behavior may give an attributional excuse for expected fail-ures in future tasks (Berglas & Jones, 1978). By referring to low effortstudents are able to buffer against negative feedback in the case offailure. A recent study has shown, however, that it is difficult to dis-tinguish between active and passive task avoidance as they correlatestrongly (Määttä, Stattin, & Nurmi, 2002).

When students avoid concentrating on or investing effort in achallenging task, they forfeit an important opportunity for learning(Dweck & Leggett, 1988). Task avoidance has been shown to nega-tively influence children's academic skills including reading (Aunolaet al., 2002; Onatsu-Arvilommi & Nurmi, 2000), spelling (Georgiou etal., 2011; Georgiou, Manolitsis, Nurmi, & Parrila, 2010), and mathe-matics (Aunola, Nurmi, Lerkkanen, & Rasku-Puttonen, 2003;Onatsu-Arvilommi & Nurmi, 2000). Hill and Hill (1982) even foundthat task avoidance was associated with learning difficulties. Astask avoidance is shown to play such an important role in academicperformance, the measurement of the construct becomes a crucialissue.

2. Measuring task avoidance

There is a considerable variety in methods of measurement in re-search on learning motivation and related achievement beliefs andbehaviors. Overall, self-report methodology has dominated the field(Fulmer & Frijters, 2009). Although self-report measures are easyand quick to administer and show good internal consistency, theyhave several methodological weaknesses. First, young children maynot be able to process the statements in self-report measures due totheir limited cognitive capacity (Fulmer & Frijters, 2009). Second, stu-dents' developmentally bound reports of motivation may threatenthe stability aspect of validity and reliability in their reports (Duncan& McKeachie, 2005). For instance, different factor structures mayemerge at different developmental stages of self-understanding(Duncan & McKeachie, 2005). Third, existing self-report measuresare often highly general, rather than contextually bound, although cur-rent theory and research suggests that context specificity is essential tomotivational processes. For example, some students attempt to hidetheir effort expenditure from peers but not from teachers (Juvonen &Murdock, 1995). Last but not the least, social desirability issues havebeen a critical concern in self-report measures (Duncan & McKeachie,2005).

Due to these limitations researchers have begun to use newmethods to investigate students' achievement beliefs and behaviors(Fulmer & Frijters, 2009). A good example is the measurement oftask-avoidant behavior. For instance, Nurmi and colleagues devel-oped a measure, namely the Behavioral Strategy Rating Scale (BSRS;Aunola et al., 2000; Onatsu-Arvilommi & Nurmi, 2000), in whichteachers rate their students in the classroom on the basis of theirtask-avoidant behavior (see also Georgiou et al., 2010; Hirvonen,Georgiou, Lerkkanen, Aunola, & Nurmi, 2009; Stephenson, Parrila,Georgiou, & Kirby, 2008). The BSRS contains 2 positively and 3 nega-tively worded items that assess the same construct (task avoidance)and shows adequate internal consistency. Although it has been usedin several studies, it has not been validated thoroughly. The limitedvalidation of the BSRS may confound the measurement of task avoid-ance and lead to non-replicable findings. As the BSRS method is easyto use even among young students and provides a good tool to studymotivation in classroom contexts, the validation of the measure is im-portant for future research on task avoidance.

3. The present study

The purpose of this studywas to examine the BSRS's (1) factor struc-ture validity, (2) convergent and discriminant validity, (3) criterion

validity, (4) factor and item reliability, and (5) stability in these validityand reliability sources. To this end, we used a longitudinal sample, withthe expectation of determining the developmental appropriateness ofthe BSRS for use across different ages throughout children's early schoolyears.

For the BSRS, a single substantive factor measuring a continuum oftask avoidant vs. task-focused behavior has been repeatedly reported inthe literature (e.g., Aunola et al., 2002; Georgiou et al., 2010; Stephensonet al., 2008). Yet recent researchers emphasize the importance of takinginto account method effects associated with positively or negativelyworded items (e.g., Motl &DiStefano, 2002).We thus expect a two-factorstructure, with one content factor representing task avoidance and onemethod factor representing positive or negative wording of the items(Hypothesis 1a). With regard to the BSRS's reliability, we expect highitem reliabilities and factor-score scale reliabilities (Hypothesis 1b). Totest the convergent and discriminant validity multiple sources (i.e.,teachers, parents, and testers) were used to evaluate each student'stask-avoidant behavior in different contexts (classroom learning, reme-dial learning, homework, and evaluative contexts). Testers also rated stu-dents' social-dependent behavior. Convergent validity requires that thesame constructs have strong correlations between raters (in our caseclassroom teachers, remedial teachers, mothers, and testers; Campbell& Fiske, 1959). As children often exhibit different task-avoidant behav-iors in different contexts (Mägi, Lerkkanen, Poikkeus, Rasku-Puttonen,& Nurmi, 2011), we expect that classroom teacher-reported task avoid-ance would bemoderately convergent with remedial teachers' (Hypoth-esis 2a), mothers' (Hypothesis 2b), and testers' reports (Hypothesis 2c).As discriminant validity requires that the correlations between thesame constructs between raters be significantly stronger than those be-tween different constructs between raters (Campbell & Fiske, 1959),we assume that teacher-reported task avoidance would be divergentfrom tester-reported social dependence (Hypothesis 2d). When testingcriterion validity, we expect that teacher-reported task avoidancewould be negatively associated with both reading andmathematics per-formance (Hypothesis 3; see also Aunola et al., 2003; Stephenson et al.,2008). When assessing the stability of the BSRS over time, we anticipatelongitudinal invariance of the measurement structure (i.e., invariance offactor loadings, item intercepts, and item uniqueness over time; Hypoth-esis 4a) and that the convergent, discriminant, and criterion validitywould be supported in all assessment waves (Hypothesis 4b).

4. Method

4.1. Participants

This study is part of the ongoing First Steps Study (Lerkkanen, Niemi,Poikkeus, Poskiparta, Siekkinen, & Nurmi, 2006) inwhich 1880 childrenwere followed from Kindergarten to Grade 2. The sample was recruitedfrom four municipalities in Finland, two in Central, one inWestern, andone in Eastern Finland. The children comprised about a half of the agecohort from one municipality and the whole age cohort from theother three. The representativeness of the children's family backgroundwith respect to the general Finnish population was good.

The participants in this study were a subsample of 352 children(175 girls, 177 boys; age at kindergarten entry: M=74.0 months,SD=3.4 months). To obtain the subsample, we selected 378 childrenby randomly sampling a small number of children from each kinder-garten classroom. The aim of the sampling was to select by chance 3children from each classroom. Due to small or particularly largeclass size, the actual number of children included from each class-room ranged between 1 and 4 (M=2.5, SD=0.7). We then excluded26 children who were enrolled in special education classrooms. Pa-rental consent was received for all the remaining 352 children. Themajority (77.6%) of the children came from nuclear families, 11.5%from single parent families, 9.6% from blended families, and 1.3%from families where parents were divorced and the child lived in

Page 3

692 X. Zhang et al. / Learning and Individual Differences 21 (2011) 690–698

two homes. A total of 26.8% of the children's mothers had a Master'sdegree or higher, 36.3% had a BA or vocational college degree, 30.2%had a vocational school degree, and 6.7% had no education beyondcomprehensive school. Of the 352 children, 106 children in the fallof Grade 1, 64 in the spring of Grade 1, and 85 in Grade 2 received re-medial teaching at school.

A total of 236 teachers in Kindergarten, 136 in Grade 1, and 133 inGrade 2, participated in the study. All the teachers changed from Kin-dergarten to Grade 1, 128 teachers remained the same from fall tospring in Grade 1, and 114 teachers remained the same from Grade1 to Grade 2. All the Kindergarten teachers had at least a BA degree.Their teaching experience varied considerably: 61% had more than15 years, 25% from 6 to 15 years, 12% from 1 to 5 years, and 2% lessthan 1 year of experience. In Grade 1 and Grade 2, 96% of the teachershad at least a Master's degree, and the remaining 4% had a BA degree.Half (50%) of them had more than 15 years of teaching experience,20% had from 11 to 15 years, 16% from 6 to 10 years, and 13% had1–5 years of experience. All the teachers were asked for their writtenconsent to participate.

4.2. Procedures

In the spring of Kindergarten (Wave 1), the fall (Wave 2) and spring(Wave 3) of Grade 1, and the spring of Grade 2 (Wave 4), teachers filledout several questionnaires for each target child. For children receivingremedial teaching, classroom teachers and remedial teachers indepen-dently filled out several questionnaires. Trained testers (researchers orstudents of psychology and education) administered individual teststo the target children in the first wave and both group and individualtests in the last three waves. In each wave, testers filled out a question-naire for each target child after the individual test situation. In the lasttwo waves, mothers filled out several questionnaires for their children.

4.3. Measures

4.3.1. Behavioral Strategy Rating Scale — teachers' formIn each of the four waves, teachers evaluated the task-avoidant be-

havior of each target child in their classes using the Behavioral StrategyRating Scale (BSRS; Aunola et al., 2000). They were asked to considerhow the child typically behaved in classroom situations and then torate his or her behavior on a 5-point scale (1=not at all; 5=to a greatextent) using 5 statements consisting of 2 positivelyworded items (Doesthe student actively attempt to solve even difficult situations and tasks?Does the student demonstrate initiative and persistence in his/her activitiesand tasks?) and 3 negatively worded items (Does the student have a ten-dency to find something else to do instead of focusing on the task at hand?Does the student give up easily? If the activity or task is not goingwell, doesthe student lose his/her focus?). The BSRS is based on a more extensivescale (Onatsu-Arvilommi & Nurmi, 2000) that was planned to assessvarious aspects of task-avoidant vs. task-oriented achievement strate-gies (e.g., failure expectations, persistence, task avoidance, helplessness,and social support seeking). As the preliminary analysis (exploratoryfactor analysis) showed that all (positively and negatively worded)items formed a single factor in themore extensive scale, a short versionof the BSRS was developed. Another rationale for shortening the scalewas that we would like to decrease teachers' workload when ratingchildren in their classroom.

4.3.2. Behavioral Strategy Rating Scale — mothers' formIn the last two waves, mothers evaluated their child's task-avoi-

dant behavior using the same 5 statements as used in the teacher-re-port BSRS. Each mother was asked to consider how her childtypically behaved when doing homework and then to rate his orher behavior on a 5-point scale (1=not at all; 5= to a great extent).The alpha coefficients were 0.90 and 0.89 for the third and fourthwaves, respectively.

4.3.3. Observer-rating Scale of Achievement StrategiesIn each of the four waves, testers assessed each target child's task-

avoidant and social-dependent behaviors using the Task Avoidanceand Social Dependence subscales in the Observer-rating Scale ofAchievement Strategies (OSAS; Nurmi & Aunola, 1998). Task Avoidancemeasures to what extent the child diverts his or her attention from thetask rather thanmakes an active effort as task difficulty increases. SocialDependencemeasures to what extent the child constantly looks for sig-nals of approval or support for his or her task performance from the tes-ter. Task Avoidance contains 4 items comprising 2 positively wordeditems (Although the task turns difficult for the child, s/he tries hard to finishit; The child tries persistently to do the task) and 2 negatively wordeditems (If there are problems with the task, the child starts doing somethingelse; If the child cannot cope with the task, s/he becomes interested in otherthings in the room). Social Dependence contains 2 items (If there areproblems with the task, the child turns to the tester; The child seeks foryour support when doing the task). After testing each child on an individ-ual basis, the testers were asked to consider how the child's behaviorbecame manifest as the task became more difficult, and then to ratehis or her behavior on a 7-point scale (1=not at all; 7=always or al-most all the time this kind of behavior). The alpha coefficients for thefirst, second, third, and fourth waves were 0.80, 0.78, 0.79, and 0.70, re-spectively, for the Task Avoidance scale, and 0.96, 0.92, 0.85, and 0.90,respectively, for the Social Dependence scale.

4.3.4. Mathematics performanceIn each of the four waves, children's arithmetic skills were assessed

by the Basic Arithmetic test (Aunola & Räsänen, 2007). In the test, amaximum of 28 items containing 14 addition items (e.g., 2+1=; 3+4+6=) and 14 subtraction items (e.g., 4–1=; 20−2−4=) can beattemptedwithin a 3-minute time limit. Task difficulty increases acrossthe test. The test was administered on an individual basis in the firstwave and on a group basis in the last three waves. The score wasthe total number of correct answers (max. 28 points). The split-half re-liability for the test was 0.76, 0.91, 0.80, and 0.85 at the first, second,third, and fourth waves, respectively. The means for the test were3.13 (SD=2.09), 3.82 (SD=2.47), 10.86 (SD=4.09), and 16.53(SD=4.68) at the first, second, third, and fourth waves, respectively.

4.3.5. Reading performanceIn each of the last three waves, a group administered subtest of the

nationally normed reading test battery (ALLU; Lindeman, 1998) wasused to assess word-level reading. In the test, a maximum of 80 trialscan be attempted within a 2-minute time limit. On each item, thechild was asked to read 4 phonologically similar words and draw aline connecting a picture and the word semantically matching it.The following alternative forms of the subtest were used: Form Bwith capital letters in Wave 2, Form A with small letters in Wave 3,and Form B with small letters in Wave 4. According to the test manual(Lindeman, 1998), the Kuder–Richardson reliability coefficient was0.97 in Grade 1 and 0.82 in Grade 2. Alternate-form reliability be-tween forms A and B was 0.84. The means for the test were 8.95(SD=6.87), 19.05 (SD=8.83), and 25.14 (SD=7.31) at the second,third, and fourth waves, respectively.

4.4. Data analytical strategy

Analyses were conducted in four phases using confirmatory factoranalysis (CFA). First, we conducted CFAs separately for each of thefour waves to compare four alternative measurement models (i.e., sin-gle task-avoidant component, task-avoidant and task-focused compo-nents, task-avoidant and positive-wording components, task-avoidantand negative-wording components). The item and factor-score scale re-liabilities were also computed using the fourmodels. Second, we exam-ined the measurement invariance of the teacher-report BSRS across thefour waves by testing and comparing four nested models that imposed

Page 4

Model 1a Model 1bP1 P2 N3 N4 N5

Task

Avoidance

P1 P2 N3 N4 N5

Task

Avoidance

Positive

Method

P1 P2 N3 N4 N5

Task

Avoidance

Negative

Method

Model 1c Model 1d

Task

Avoidance

Task

Focus

P1 P2 N3 N4 N5

Fig. 1. Alternative models of the teacher-report BSRS. Model 1a = single task-avoidantcomponent; Model 1b = task-avoidant and task-focused components; Model 1c =task-avoidant and positive-wording components; Model 1d = task-avoidant and neg-ative-wording components.

Table 1Evaluation of the factor structure within time.

Models Fit indexes

χ2 df CFI TLI RMSEA (90% CI) SRMR

Wave 1Model 1a 132.68, pb .0001 5 .79 .57 .27 (.23–.31) .11Model 1b 17.74, p=.001 4 .98 .94 .10 (.06–.15) .04Model 1c 13.30, p=.004 3 .98 .94 .10 (.05–.16) .04Model 1d 0.65, p=.72 2 1.00 1.01 .00 (.00–.08) b.01

Wave 2Model 1a 93.25, pb .0001 5 .81 .62 .24 (.20–.29) .09Model 1b 9.63, p=.05 4 .99 .97 .07 (.01–.13) .03Model 1c 7.22, p=.07 3 .99 .97 .07 (.00–.14) .03Model 1d 2.04, p=.36 2 1.00 1.00 .01 (.00–.12) .01

Wave 3Model 1a 92.46, pb .0001 5 .86 .72 .25 (.20–.29) .10Model 1b 14.77, p=.005 4 .98 .96 .10 (.05–.15) .04Model 1c 11.07, p=.01 3 .99 .96 .10 (.04–.16) .04Model 1d 0.40, p=.82 2 1.00 1.01 b.00 (.00–.07) b.01

Wave 4Model 1a 111.58, pb .0001 5 .78 .57 .27 (.23–.31) .10Model 1b 11.96, p=.02 4 .98 .960 .08 (.03–.14) .03Model 1c 8.97, p=.03 3 .99 .96 .08 (.02–.15) .03Model 1d 4.12, p=.13 2 1.00 .98 .06 (.00–.14) .01

Note. CFI = comparative fit index; TLI = Tucker–Lewis index; RMSEA = root meansquare error of approximation; SRMR = standardized root mean square residual;CI = confidence interval; Wave 1 = spring of Kindergarten; Wave 2 = fall of Grade1; Wave 3 = spring of Grade 1; Wave 4 = spring of Grade 2; Model 1a = singletask-avoidant component; Model 1b = task-avoidant and task-focused components;Model 1c = task-avoidant and positive-wording components; Model 1d = task-avoidant and negative-wording components.

693X. Zhang et al. / Learning and Individual Differences 21 (2011) 690–698

successive restrictions (i.e., no constraint, constraint of factor loadings,constraint of item intercepts, and constraint of item uniqueness) onthe model parameters. Third, after the constraints of factor loadingswere in place, convergent and discriminant validity was investigatedby relating teacher-reported latent dimensions to mother- and tester-reported latent dimensions. Fourth, CFAs were used to evaluate the cri-terion validity of the teacher-report BSRS by relating teacher-reportedlatent dimensions to the children's concurrent reading and mathemat-ics performance.

All the CFA models were evaluated with Mplus 5.21 (Muthén &Muthén, 1998–2009) by using maximum likelihood estimation withrobust standard errors (MLR). To evaluate the model fit, we usedchi-square values, comparative fit index (CFI), Tucker–Lewis index(TLI), root mean square error of approximation (RMSEA), and stan-dardized root mean residual (SRMR). A non-significant chi-square in-dicates good fit. The general cutoffs for accepting a model for CFI andTLI were equal to or greater than 0.95, and equal to or less than 0.05for the RMSEA, and less than 0.08 for the SRMR (Hu & Bentler, 1999).Two procedures were used to assess measurement invariance. Thefirst procedure used chi-square difference tests. Because we usedMLR, Satorra and Bentler (2001) scaled chi-square difference testswere conducted when testing nested models. Because multiple testswere applied, a significance level of .01 was used to adjust the alphalevel and control for type I error. The second procedure used Chen's(2007) recommendations (i.e., a change of ≥.01 in CFI supplementedby a change of ≥.015 in RMSEA or a change of ≥.01 in SRMR wouldindicate noninvariance at a particular step for a sample size of N300).

5. Results

5.1. Validity of the factor structure

We tested first the four CFA models presented in Fig. 1. Model 1arepresented a single-factor model of the substantive content, taskavoidance. Model 1b represented a model of two substantive contentfactors, task avoidance and task focus. Models 1c and 1d representedmodels with two uncorrelated factors, namely a content factor repre-senting task avoidance and a method factor representing the direc-tion of item wording. In Model 1c the method factor accounted forwording effects among the positively worded items, whereas inModel 1d the method factor accounted for wording effects amongthe negatively worded items. Table 1 presents the goodness-of-fit in-dexes for the four models tested in each of the four waves.

Model 1a showed a poor fit in each of the four waves. NeitherModel 1b nor Model 1c had a good fit: the chi-square statistics weresignificant or approaching significance, and the RMSEAs were notwithin acceptable ranges in any assessment wave. In contrast,Model 1d had a very good fit: all fit indexes were within acceptableranges in each wave. These results suggest that the best-fitting mea-surement model for the teacher-report BSRS included one substan-tive content factor measuring task-avoidant behavior and onemethod factor accounting for wording effects among the negativelyphrased items. Thus, Hypothesis 1a was supported.

The item reliability and validity of the BSRS were further exploredin all four models for each wave. Cronbach's alpha coefficients and thereliabilities of the factor-score scales were also calculated to evaluateinternal consistency. Table 2 presents the item reliabilities, item va-lidity coefficients (standardized factor loadings), and the reliabilitiesof the factor-score scales. The item reliabilities and validity coeffi-cients in Model 1a were lower than those in Model 1b, Model 1c,and Model 1d, especially for the 2 positively worded items. The fac-tor-score scale reliabilities were very high in all four models acrossall four waves. In turn, alpha coefficients across 5 items (2 reversed)were 0.86, 0.87, 0.88, and 0.88 for the first, second, third, and fourthwaves, respectively. Although alpha coefficients were high, theywere lower than the reliabilities of the factor-score scales. To

summarize, all the items included in Model 1d were good indicatorsof the latent factors, and the BSRS showed high internal consistency.Hypothesis 1b was therefore supported. Consequently, Model 1dwill be used in further analyses.

Page 5

Table2

Estimated

item

relia

bilities,stan

dardized

valid

ityco

efficien

ts(inpa

renthe

ses),a

ndfactor-sco

rescalerelia

bilities.

Item

sMod

elsat

Wave1

Mod

elsat

Wav

e2

Mod

elsat

Wav

e3

Mod

elsat

Wav

e4

Mod

el1a

Mod

el1b

Mod

el1c

Mod

el1d

Mod

el1a

Mod

el1b

Mod

el1c

Mod

el1d

Mod

el1a

Mod

el1b

Mod

el1c

Mod

el1d

Mod

el1a

Mod

el1b

Mod

el1c

Mod

el1d

RC(TA)

RC(TATF)

RC(TAPM

)RC

(TANM)

RC(TA)

RC(TATF)

RC(TAPM

)RC

(TANM)

RC(TA)

RC(TATF)

RC(TAPM

)RC

(TANM)

RC(TA)

RC(TATF)

RC(TAPM

)RC

(TANM)

P1.15

(.39

).48

(.69

).62

(.37

.69)

.48

(.70

).18

(.42

).49

(.70

).58

(.40

.69)

.52

(.72

).27

(.52

).65

(.81

).67

(.50

.65)

.65

(.80

).20

(.45

).57

(.75

).63

(.42.67)

.56

(.75

)P2

.27

(.52

).87

(.93

).69

(.50

.66)

.86

(.93

).25

(.50

).74

(.86

).64

(.49

.59)

.70

(.84

).29

(.54

).70

(.84

).68

(.52

.64)

.71

(.84

).25

(.50

).72

(.85

).65

(.48.65)

.72

(.85

)N1

.93

(.96

).94

(.97)

.94

(.97

).93

(.52

.81)

.90

(.95

).91

(.95

).91

(.95

).93

(.53

.81)

.92

(.96

).93

(.96

).93

(.96

).91

(.60

.74)

.86

(.93

).88

(.94

).88

(.94

).91

(.50

.81)

N2

.52

(.72

).51

(.72)

.51

(.72

).54

(.54

.51)

.52

(.72

).52

(.72

).52

(.72

).53

(.52

.51)

.57

(.75

).56

(.75

).56

(.75

).59

(.61

.47)

.56

(.75

).56

(.75

).56

(.75

).56

(.51

.55)

N3

.83

(.91

).83

(.91)

.83

(.91

).84

(.46

.80)

.86

(.93

).86

(.93

).86

(.93

).84

(.54

.74)

.84

(.92

).84

(.92

).84

(.92

).86

(.53

.76)

.86

(.93

).85

(.92

).85

(.92

).83

(.53

.75)

Factor-score

scales

.98

.98.94

.98.85

.94.94

.97

.97.90

.97.81

.89.92

.97

.98.91

.98.84

.91.91

.97

.97.90

.97.83

.90.92

Note.

Wav

e1=

spring

ofKinde

rgarten;

Wav

e2=

fallof

Grade

1;W

ave3=

spring

ofGrade

1;W

ave4=

spring

ofGrade

2;M1a

=sing

letask-avo

idan

tco

mpo

nent;M1b

=task-avo

idan

tan

dtask-foc

used

compo

nents;

M1c

=task-

avoida

ntan

dpo

sitive

-wording

compo

nents;

M1d

=task-avo

idan

tan

dne

gative

-wording

compo

nents;

RC=

relia

bilityco

efficien

t;TA

=task-avo

idan

tco

mpo

nent;TF

=task-foc

used

compo

nent;PM

=po

sitive

-wording

compo

nent;

NM

=ne

gative

-wording

compo

nent.

694 X. Zhang et al. / Learning and Individual Differences 21 (2011) 690–698

5.2. Evaluation of longitudinal measurement invariance

Fig. 2 shows a longitudinal factor model based on which aspects ofmeasurement invariance over time were investigated. This modelconsisted of 8 factors: 4 content factors and 4 method factors. Corre-lations among task-avoidant factors from different waves and amongmethod factors from different waves were allowed (see also Motl &DiStefano, 2002). Content and method factors were assumed to be in-dependent and thus no correlations were specified between them.Residual autocorrelations were not estimated for any observed item,as preliminary analyses showed that most (83.3%) of them were notsignificant.

Following Vandenberg and Lance's (2000) procedure, we nextevaluated the longitudinal measurement invariance of the teacher-report BSRS by comparing the goodness-of-fit of the four nestedmodels that imposed successive restrictions on the model parame-ters. Model 2a (a baseline configural invariance model) tested wheth-er the patterns of the overall factor structure were the same across allfour waves. Model 2b (a metric invariance model) included the re-strictions from Model 2a plus the additional constraints of equal fac-tor loadings for both content and method factors. A comparable fit ofModel 2b to Model 2a would suggest invariance of the factor loadingsfor both content and method factors across all four waves and, thus,satisfy weak invariance. Model 2c (a scalar invariance model) includ-ed all the restrictions fromModel 2b plus the additional constraints ofequal item intercepts. A comparable fit of Model 2c to that of Model2b would suggest invariance of the item intercepts across all fourwaves and, in turn, satisfy strong invariance. Finally, Model 2d (anitem uniqueness invariance model) included all the restrictionsfrom Model 2c plus the additional constraints of equal item unique-ness. A comparable fit of Model 2d to Model 2c would suggest invari-ance of item uniqueness across all four waves and, in turn, satisfystrict invariance. Table 3 presents the fit indices for the models.

The results presented in Table 3 showed that all the invariancemodels fitted to the data relatively well, although the chi-square sta-tistics were statistically significant. At each step when assessing mea-surement invariance, the chi-square difference test did not reach thesignificance level of .01, and the changes in CFI, RMSEA, and SRMRwere all less than .01. Strict measurement invariance was thus satis-fied across all four waves, and Hypothesis 4a was supported. Inother words, the results supported equality of the factor patterns, fac-tor loadings, item intercepts, and item uniqueness for the BSRS acrossall four waves.

The stability coefficients were also computed for both content andmethod factors across time using Model 2d and are presented inTable 4. Both content and method factors showed moderate to highstability over time. The stability coefficients between the first and sec-ond waves were smaller than those between the second and thirdwaves and between the third and fourth waves (see a similar findingin Georgiou et al., 2010). It is notable that during the first and secondwaves children made a transition from kindergarten to elementaryschool and were evaluated by different teachers across the twowaves. The stability coefficients for raw task-avoidance compositeswere also computed separately for students with and without achange of teachers from Grade 1 to Grade 2. The results indicatedthat the stability coefficients for students with a change of the teacher(from Wave 2 to Wave 3: r=.70, pb .01, n=16; from Wave 3 toWave 4: r=.70, pb .001, n=62) were close to those for studentswithout such a change (from Wave 2 to Wave 3: r=.71, pb .001,n=264; from Wave 3 to Wave 4: r=.72, pb .001, n=206).

5.3. Convergent and discriminant validity

To evaluate convergent and discriminant validity, unconditionalcorrelations were estimated between teacher-reported dimensionsand mother- and tester-reported dimensions. Because the evaluation

Page 6

695X. Zhang et al. / Learning and Individual Differences 21 (2011) 690–698

of the relations between latent dimensions requires weak measure-ment invariance of each dimension (i.e., equality of factor loadings),we first assessed the weak invariance of the mother-report BSRSand tester-report OSAS following the same steps as when assessingthe weak invariance of the teacher-report BSRS.

For themother-report BSRS,method effects associatedwith the neg-ativelyworded itemswere present in eachwave. The best-fittingmodelinvolved a content factor representing task avoidance and amethod fac-tor accounting for wording effects among the negatively phrased items,χ2(2, N=260–271)=3.28–4.70, psN0.10, CFIs=.99–1.00,TLIs=.97–.99, RMSEAs=.05–.07, SRMRs=.01. Evaluation of weakmeasurement invariance indicated that imposing the constraints ofequal factor loadings on the configural invariancemodel did not changethe fit significantly: ΔSatorra–Bentler-scaled χ2(6)=5.29, pN .01,ΔCFIb .01, ΔRMSEAb .015, ΔSRMRb .01. The weak invariance modelhad a good fit to the data, χ2(33, N=292)=41.22, p=.15, CFI=.99,TLI=.99, RMSEA=.03, SRMR=.03. Hence, the weak measurement in-variance of the mother-report BSRS was supported over time.

For the tester-report OSAS, method effects associated with the pos-itively worded itemswere present. The best-fittingmodel involved twocontent factors representing task-avoidant and social-dependent be-haviors and a method factor accounting for wording effects among thetwo positively phrased items, χ2(6, N=336–350)=6.12–9.39,psN0.15, CFIs=.99–1.00, TLIs=.98–1.00, RMSEAs=.01–.04,SRMRs=.01–.04. Evaluation of weak measurement invariance indicat-ed that imposing the constraints of equal factor loadings on the config-ural invariance model did not change the fit significantly, ΔSatorra–Bentler-scaled χ2(15)=20.90, pN .01, ΔCFIb .01, ΔRMSEAb .015,ΔSRMRb .01. The weak invariance model had a good fit to the data, χ2

(225, N=352)=241.58, p=.21, CFI=1.00, TLI=1.00, RMSEA=.01,SRMR=.04. Hence, the weak measurement invariance of the tester-report OSAS was also supported over time.

Based on the weak invariance models of the teacher-report BSRS,the mother-report BSRS and the tester-report OSAS, further analyseswere conducted to investigate the convergent and discriminant valid-ity of the teacher-report BSRS. The results indicated that the model inwhich the correlations were estimated among the task-avoidant and

P1

TaskAvoidanceWave 1

MethodEffectsWave 1

MethodEffectsWave 2

N5N4N3P2P1N5N4N3P2

TaskAvoidanceWave 2

Fig. 2.Model for the evaluation of longitudinal measurement invariance. Wave 1= spring ofof Grade 2.

method factors of different reporters provided an adequate fit to thedata, χ2(1166, N=352)=1499.00, pb .0001, CFI=.97, TLI=.97,RMSEA=.03, SRMR=.06. Table 5 presents the correlations betweenteacher-reported dimensions and mother- and tester-reported di-mensions at each wave of assessment.

The convergent correlations between teacher- and mother-reported task avoidance (Wave 3: r=.36; Wave 4: r=.48) were sig-nificant and moderate. These correlations were larger than the dis-criminant correlations between teacher-reported task avoidanceand mother-reported method effect. The convergent correlations be-tween teacher- and tester-reported task avoidance were also signifi-cant in the first three waves (Wave 1: r=.19; Wave 2: r=.22;Wave 3: r=.24) but not in the last wave. Although these correlationswere small, they were larger than the discriminant correlations be-tween teacher-reported task avoidance and tester-reported social de-pendence and method effect in the first three waves. Overall,Hypotheses 2b and 2d were supported, and Hypotheses 2c and 4bwere partially supported.

For children receiving remedial teaching, correlations betweentask-avoidance composites rated by classroom teachers and thoserated by remedial teachers were .51 (pb .001), .61 (pb .001), and .54(pb .001) at the second, third, and fourth waves, respectively, sug-gesting moderate convergence between their ratings. Thus, Hypothe-ses 2a and 4b were supported.

5.4. Criterion validity

To determine the criterion validity of the teacher-report BSRS, CFAswere conducted separately for eachof the fourwaves. Results suggestedthat themodels inwhich correlationswere estimated between teacher-reported dimensions and reading and mathematics performance pro-vided adequate fits to the data, χ2(5/8, N=352)=3.70–19.22,psN0.10, CFIs=.99–1.00, TLIs=.96–1.02, RMSEAs=.00–.06,SRMRs=.01–.02. Teacher-reported task avoidance was associated neg-atively with children's performance in reading (rs=−.21 to −.36,psb .001) and mathematics (rs=−.16 to −.39, psb .01) within each ofthe four waves. In contrast, method factors were not related

MethodEffectsWave 3

MethodEffectsWave 4

N5N4N3P2P1N5N4N3P2P1

TaskAvoidance

Wave 3

TaskAvoidanceWave 4

Kindergarten; Wave 2 = fall of Grade 1; Wave 3= spring of Grade 1; Wave 4 = spring

Page 7

Table 3Evaluation of longitudinal measurement invariance.

Models Fit indexes

χ2 df CFI TLI RMSEA(90% CI)

SRMR

Model 2a(configural invariance)

195.95,p=.004

146 .99 .98 .03(.02–.04)

.07

Model 2b(weak invariance)

219.46,p=.003

164 .98 .98 .03(.02–.04)

.07

Model 2c(strong invariance)

241.97,p=.001

176 .98 .98 .03(.02–.04)

.07

Model 2d(strict invariance)

250.75,p=.002

191 .98 .98 .03(.02–.04)

.07

Model comparisons Δχ2 df ΔCFI ΔTLI ΔRMSEA ΔSRMRModels 2a vs. 2b 23.44,

pN .0118 b.01 b.01 b.01 b.01

Models 2b vs. 2c 22.51,pN .01

12 b.01 b.01 b.01 b.01

Models 2c vs. 2d 13.21,pN .01

15 b.01 b.01 b.01 b.01

Note. CFI = comparative fit index; TLI = Tucker–Lewis index; RMSEA = root meansquare error of approximation; SRMR = standardized root mean square residual;CI = confidence interval; Model 2a = equality of the overall factor patterns;Model 2b = Model 2a plus equality of the factor loadings; Model 2c = Model 2bplus equality of the item intercepts; Model 2d = Model 2c plus equality of the itemuniqueness; Δχ2=ΔSatorra–Bentler-scaled χ2.

Table 5Correlations of teacher-reported dimensions with mother- and tester-reporteddimensions.

Teacher-reportedlatentdimensions

Mother-reported latentdimensions

Tester-reported latent dimensions

Taskavoidance

Methodeffect

Taskavoidance

Methodeffect

Socialdependence

Wave 1Taskavoidance

– – .19⁎⁎⁎ .16⁎ .09

Methodeffect

– – .01 .24⁎⁎⁎ −.02

Wave 2Taskavoidance

– – .22⁎⁎⁎ .19⁎⁎ .14⁎⁎

Methodeffect

– – −.03 .21⁎⁎ −.06

696 X. Zhang et al. / Learning and Individual Differences 21 (2011) 690–698

significantly to either reading or mathematics performance (rs=−.09to .10, ns) within any assessment wave. Thus, Hypotheses 3 and 4bwere supported.

6. Discussion

As task avoidance plays a crucial role in the development of stu-dents' academic skills, the measurement of the construct is an impor-tant issue. In this study, we validated a teacher-report measure ofchildren's task avoidance, the BSRS, with a representative sample ofFinnish children. The results indicated that the BSRS had good con-struct, convergent, discriminant, and criterion validity, as well asgood item and scale reliabilities. More importantly, the validity andreliability were supported across all four assessment waves from Kin-dergarten to Grade 2. These results suggest that the BSRS is a usefultool for investigating task avoidance in classroom contexts.

Because the BSRS was developed with both positively and nega-tively worded items, we expected the wording-effect factor to be em-pirically separable from the substantive task-avoidant factor. Insupport of this hypothesis, the BSRS demonstrated a two-factor struc-ture with a content factor representing task avoidance and a methodfactor representing wording effects among the negatively phraseditems. In other words, both positively and negatively worded itemsassess the same construct (task avoidance), although there is a needto account for the method effects associated with negative items.Moreover, the results showed that the expected two-factor modelwas supported across all four waves. These findings are consistent

Table 4Stability correlations of latent factors.

Waves of latent factors 1 2 3 4

1. Wave 1 – .41⁎⁎⁎ .43⁎⁎⁎ .46⁎⁎⁎

2. Wave 2 .37⁎⁎⁎ – .73⁎⁎⁎ .54⁎⁎⁎

3. Wave 3 .30⁎⁎⁎ .75⁎⁎⁎ – .63⁎⁎⁎

4. Wave 4 .38⁎⁎⁎ .68⁎⁎⁎ .71⁎⁎⁎ –

Note. Below the diagonal (lower left part): correlation coefficients of task-avoidancecomponents; above the diagonal (upper right part): correlation coefficients ofmethod components. Wave 1 = spring at Kindergarten; Wave 2 = fall at Grade 1;Wave 3 = spring at Grade 1; Wave 4 = spring at Grade 2.⁎⁎⁎ pb .001.

with recent analyses for Likert scales (see Motl & DiStefano, 2002for Rosenberg Self-Esteem Scale and DiStefano & Motl, 2006 for SocialPhysique Anxiety Scale). One might ask whether the distinction be-tween the content factor and the method factor may simply indicatemeaningful conceptual differences between positively (task focus)and negatively worded items (task avoidance). This is of course pos-sible and there is no absolute way to statistically test whether the sec-ond “method factor” is due to positive/negative wording of the itemsor conceptual differences between the items.

Furthermore, our findings suggested that the two-factor model in-cluding a content factor and a negative method factor showed mea-surement invariance across time. Specifically, factor loadings, itemintercepts, and item uniqueness were all found equivalent across allfour waves. The invariance of the factor loadings and item interceptssuggests that at all points along the latent task-avoidant continuumthe same score on the latent factor indicates equivalent average rat-ings on task-avoidant behavior. The equivalence of item uniquenessindicates that like-item ratings are equally accurate (reliable) acrosstime and that different teachers of young children at differing agesused the same range on the task-avoidant dimension for their ratings.

To determine whether the BSRS actually measures the intendedconstruct – task avoidance – and does not confound it with other con-structs, we examined the convergent and discriminant validity of theBSRS. Our findings showed that the convergent correlations were sta-tistically significant between teacher- and mother-reported taskavoidance, suggesting interchangeability between teachers' andmothers' ratings. These convergent correlations were larger than thediscriminant correlations between teacher-reported task avoidanceand mother-reported method effects, indicating that the teacher-report BSRS measured task avoidance in a specific manner that differ-entiated the construct from other constructs (i.e., method effect).Moreover, the convergent correlations were also significant betweenteacher- and tester-reported task avoidance during the first threewaves. These convergent correlations were also larger than the dis-criminant correlations between teacher-reported task avoidanceand tester-reported social dependence and method effects. In

Wave 3Taskavoidance

.36⁎⁎⁎ .06 .24⁎⁎⁎ .18⁎⁎ .04

Methodeffect

.10 .19 −.16⁎ .29⁎⁎⁎ .00

Wave 4Taskavoidance

.48⁎⁎⁎ .13 .03 .12 .10

Methodeffect

.07 .19 −.09 .25⁎⁎⁎ −.08

Note. Wave 1 = spring of Kindergarten; Wave 2 = fall of Grade 1; Wave 3 = spring ofGrade 1; Wave 4 = spring of Grade 2.

⁎⁎⁎ pb .001.⁎ pb .05.

⁎⁎ pb .01.

Page 8

697X. Zhang et al. / Learning and Individual Differences 21 (2011) 690–698

addition, results also showed moderate convergence between task-avoidant behavior reported by a classroom teacher and that reportedby a remedial teacher in a subsample of children receiving remedialteaching, providing further validation information for the BSRS.

Notably, however, the convergent correlations between classroomteachers' reports and remedial teachers', mothers', and testers' re-ports were from low to moderate, rather than high, suggesting thatthe four sources were not fully interchangeable. These results mightbe due to the fact that the four sources stem from different contexts(classroom learning vs. remedial learning vs. homework vs. evalua-tion contexts). Children often exhibit different behaviors in differingcontexts, where they are observed by different people. Althoughboth parents and teachers observe a huge amount of children's be-havioral samples in their daily interactions with the children, class-room teachers and remedial teachers observe their students inclassroom learning and remedial learning contexts, respectively,whereas parents observe a variety of child behaviors at home. It istherefore not surprising that the convergent correlations were mod-erate between classroom teachers' reports and mothers' and remedialteachers' reports. Finally, the convergent correlations were low be-tween teacher- and tester-reports, which might be due to the factthat testers observed the children in an evaluative situation onlyonce at each wave.

In this study, we also evaluated the criterion validity of the BSRSby correlating children's task-avoidant behavior with their readingand mathematics performance. The results suggested that teacher-reported task avoidance was negatively associated with reading andmathematics performance at all waves. This finding is consistentwith recent research suggesting the important role of task-avoidantstrategies in academic skills (e.g., Aunola et al., 2003; Onatsu-Arvilommi & Nurmi, 2000). It also underscores that children's task-avoidant behavior might be of crucial importance throughout theirearly school years. The results showed also that method factorswere not related to either reading or mathematics skills within anywave. In other words, the method factor does not seem to includeany variance that is associated with academic skills, although itshould be methodologically taken into account as it may bring someerror variance in the model.

With respect to the stability aspect of the BSRS, the item reliabilityand factor-score scale reliability were adequate at all four waves, andthe test–retest stabilities were moderate to high across waves. Also,the expected two-factor measurement model was supported at allfour waves, and the parameters of the measurement model, includingfactor loadings, intercepts, and uniqueness, showed invariance acrossthe four waves. Moreover, classroom teachers' reports in the BSRSshowed satisfactory convergent and discriminant validity withmothers' and remedial teachers' reports in the BSRS and with testers'reports in OSAS across waves. Teacher-reported task avoidance wasalso associated with children's reading andmathematics performanceat each of the four waves. In sum, the reliability and validity of theteacher-report BSRS was remarkably stable over time, suggestingthat the measure is a developmentally appropriate instrument andcan be used to measure children's task avoidance throughout theirearly school years.

Certain limitations of this study should be noted when makinggeneralizations based on the results. First, although the BSRS turnedout to be a valid instrument among children from Kindergarten toGrade 2, it is unknown whether its psychometric properties can besupported in children's later elementary or secondary years. Researchthat further tracks the present sample will help answer this question.Second, it remains to be seen whether our results for Finnish childrencan be generalized to children elsewhere. Future studies that includelongitudinal samples from other countries would be helpful in an-swering this question. Third, this study focused on validating theteacher-report BSRS solely by other observer-based methods. Itwould be valuable for future studies to examine whether the BSRS

is associated with children's self-reports of task avoidance or otherestablished measures of motivation, such as task value, intrinsic mo-tivation, and interest in different academic subjects.

In conclusion, the results of this study showed that the teacher-report BSRS is a valid, reliable, and developmentally suitable instru-ment. This validation study was characterized by at least threestrengths. First, the BSRS was thoroughly validated by reference tovarious sources, including construct, discriminant, convergent, andcriterion validity. Second, we evaluated the stability aspect of the dif-ferent validity sources and demonstrated the developmental appro-priateness of the BSRS across different ages. To our knowledge, theBSRS is the only example of a scale assessing learning motivationand related beliefs and behaviors where measurement invarianceand convergent/discriminant validity have been established acrossdifferent age groups. Third, the BSRS was evaluated with a represen-tative sample of Finnish children, which allowed us to generalize ourfindings to a broader population of children.

References

Aunola, K., Nurmi, J. -E., Lerkkanen, M. -K., & Rasku-Puttonen, H. (2003). The roles ofachievement-related behaviors and parental beliefs in children's mathematicalperformance. Educational Psychology, 23, 403–421.

Aunola, K., Nurmi, J. -E., Niemi, P., Lerkkanen, M. -K., & Rasku-Puttonen, H. (2002). De-velopmental dynamics of reading skills, achievement strategies, and parental be-liefs. Reading Research Quarterly, 37, 310–327.

Aunola, K., Nurmi, J. -E., Parrila, R., & Onatsu-Arvilommi, T. (2000). Behavioral StrategyRating Scale. Jyväskylä, Finland: University of Jyväskylä.

Aunola, K., & Räsänen, P. (2007). The Basic Arithmetic Test. Jyväskylä, Finland: Universi-ty of Jyväskylä.

Bandura, A. (1997). Self-efficacy: The exercise of control. New York, NY: Freeman.Berglas, S., & Jones, E. E. (1978). Drug choice as a self-handicapping strategy in re-

sponse to noncontingent success. Journal of Personality and Social Psychology, 36,405–417.

Cain, K., & Dweck, C. S. (1995). The development of children's achievement motivationpatterns and conceptions of intelligence. Merrill-Palmer Quarterly, 41, 25–52.

Campbell, D. T., & Fiske, D. W. (1959). Convergent and discriminant validation by themultitrait–multimethod matrix. Psychological Bulletin, 56, 81–105.

Chen, F. F. (2007). Sensitivity of goodness of fit indices to lack of measurement invari-ance. Structural Equation Modeling, 14, 464–504.

Covington, M. V. (1984). The self-worth theory of achievement motivation: Findingsand implications. The Elementary School Journal, 85, 5–20.

Diener, C. I., & Dweck, C. S. (1978). An analysis of learned helplessness: Continuouschanges in performance, strategy, and achievement cognitions following failure.Journal of Personality and Social Psychology, 36, 451–462.

DiStefano, C., & Motl, R. W. (2006). Further investigating method effects associatedwith negatively worded items on self-report surveys. Structural Equation Modeling,13, 440–464.

Duncan, T. G., & McKeachie, W. J. (2005). The making of the Motivated Strategies forLearning Questionnaire. Educational Psychologist, 40, 117–128.

Dweck, C. S., & Leggett, E. L. (1988). A social-cognitive approach to motivation and per-sonality. Psychological Review, 95, 256–273.

Fulmer, S., & Frijters, J. C. (2009). A review of self-report and alternative approaches inthe measurement of student motivation. Educational Psychology Review, 21,219–246.

Georgiou, G. K., Hirvonen, R., Liao, C. -H., Manolitsis, G., Parrila, R., & Nurmi, J. -E.(2011). The role of achievement strategies on literacy acquisition across languages.Contemporary Educational Psychology, 36, 130–141.

Georgiou, G. K., Manolitsis, G., Nurmi, J. -E., & Parrila, R. (2010). Does task-focused ver-sus task-avoidance behavior matter for literacy development in an orthographical-ly consistent language? Contemporary Educational Psychology, 35, 1–10.

Gottfried, A. E. (1990). Academic intrinsic motivation in young elementary school chil-dren. Journal of Educational Psychology, 82, 525–538.

Grimm, K. J., Steele, J. S., Mashburn, A. J., Burchinal, M., & Pianta, R. C. (2010). Early be-havioral associations of achievement trajectories. Developmental Psychology, 46,976–983.

Hill, C., & Hill, K. (1982). Achievement attributions of learning-disabled boys. Psycho-logical Reports, 51, 979–982.

Hirvonen, R., Georgiou, G. K., Lerkkanen, M. -K., Aunola, K., & Nurmi, J. -E. (2009). Task-focused behavior and literacy development: A reciprocal relationship. Journal ofResearch in Reading, 33, 302–319.

Hu, L. T., & Bentler, P. M. (1999). Cutoff criteria for fit indexes in covariance structureanalysis: Conventional criteria versus new alternatives. Structural Equation Model-ing, 6, 1–55.

Juvonen, J., & Murdock, T. B. (1995). Grade-level differences in the social value of effort:Implications for the self-presentation tactics of early adolescents. Child Develop-ment, 66, 1694–1705.

Lerkkanen, M. -K., Niemi, P., Poikkeus, A. -M., Poskiparta, E., Siekkinen, M., & Nurmi, J. -E.(2006). The first steps study (Alkuportaat, ongoing). Finland: University of Jyväskylä.

Page 9

698 X. Zhang et al. / Learning and Individual Differences 21 (2011) 690–698

Lindeman, J. (1998). Allu - Ala-asteen lukutesti [Reading test for primary school]. Turku,Finland: University of Turku, Center for Learning Research.

Määttä, S., Stattin, H., & Nurmi, J. -E. (2002). Achievement strategies at school: Typesand correlates. Journal of Adolescence, 25, 31–46.

Mägi, K., Lerkkanen, M. -K., Poikkeus, A. -M., Rasku-Puttonen, H., & Nurmi, J. -E. (2011).The cross-lagged relations between children's academic skill development, task-avoidance, and parental beliefs about success. Learning and Instruction, 21, 664–675.

Motl, R.W., & DiStefano, C. (2002). Longitudinal invariance of self-esteem andmethod ef-fects associated with negatively worded items. Structural Equation Modeling, 9,562–578.

Muthén, L. K., & Muthén, B. O. (1998–2009). Mplus user's guide. Los Angeles, CA:Muthén & Muthén.

Nottelmann, E. D., & Hill, K. T. (1977). Text anxiety and off-task behavior in evaluativesituations. Child Development, 48, 225–231.

Nurmi, J. -E., & Aunola, K. (1998). Observer-rating Scale of Achievement Strategies. Jyväs-kylä, Finland: University of Jyväskylä.

Onatsu-Arvilommi, T., & Nurmi, J. -E. (2000). The role of task-avoidant and task-focusedbehaviors in the development of reading and mathematical skills during first schoolyear: A cross-lagged longitudinal study. Journal of Educational Psychology, 92, 478–491.

Satorra, A., & Bentler, P. M. (2001). A scaled difference chi-square test statistic for mo-ment structure analysis. Psychometrika, 66, 507–514.

Seifert, T., & O'Keefe, B. (2001). The relationship of work avoidance and learning goalsto perceived competency, externality and meaning. The British Journal of Education-al Psychology, 71, 81–92.

Stephenson, K., Parrila, R., Georgiou, G., & Kirby, R. (2008). Effects of home literacy, par-ents' beliefs, and children's task-focused behavior on emergent literacy and wordreading skills. Scientific Studies of Reading, 12, 24–50.

Turner, J. C., Midgley, C., Meyer, D. K., Gheen, M., Anderman, E. M., Kang, Y., et al.(2002). The classroom environment and students' reports of avoidance strategiesin mathematics: A multimethod study. Journal of Educational Psychology, 94,88–106.

Vandenberg, R. J., & Lance, C. E. (2000). A review and synthesis of the measurement in-variance literature: Suggestions, practices, and recommendations for organization-al research. Organizational Research Methods, 3, 4–69.

Weinstein, C. E., Schulte, A. C., & Palmer, D. R. (1987). The Learning and Study StrategiesInventory (LASSI). Clearwater, FL: H and H Publishing Company.

Wigfield, A. (1994). Expectancy-value theory of achievement motivation: A develop-mental perspective. Educational Psychology Review, 6, 49–78.