The Developmental Dynamics between Interest, Self-concept of Ability, and Academic Performance

Only a few studies have examined the direction of associations between academic achievement, interest, and self-concept of ability simultaneously by using longitudinal data over several school years. To examine the cross-lagged relationships between students' interest, self-concept of ability, and performance in mathematics and reading, longitudinal data from Grade 1 to Grade 7 of comprehensive school was gathered from 216 students. The results showed that, in both reading and math, performance predicted students' subsequent self-concept of ability. Some evidence was also found that math performance predicts subsequent interest in mathematics, and that self-concept of math ability mediates the impact of math performance on interest. No evidence was found for the assumption that self-concept of ability or interest would predict subsequent academic performance.

students' interest in reading and mathematics, related self-concept of ability, and school performance from first to seventh grade.
Interest and Self-concept of Ability Achievement motivation theorists (Atkinson, 1964;Bandura, 1982Bandura, , 1997Weiner, 1985) have long suggested that individuals' achievement-related beliefs and attitudes play an important role in academic environments. According to Pintrich and Schunk (1996), for example, students with positive self-perceptions of their competence are more likely both to perform better and to engage in an adaptive manner in academic tasks than those with negative self-perceptions. Similarly, students who value and are interested in specific academic tasks are more likely to perform better, learn more and be more adaptively engaged in those tasks (Pintrich & Schunk, 1996).
The importance of beliefs, expectancies and interest was first introduced by Eccles and her colleagues (1983) in their expectancy-value model of achievement motivation. It emphasizes two theoretical concepts: beliefs and expectancies related to academic situations, and subjective task values concerning different subject areas (see also Eccles & Wigfield, 1995;Wigfield & Eccles, 2000). According to this theory, expectancies and ability beliefs refer to students' beliefs about their competence in upcoming tasks and in a given task at hand. Concepts closely similar to these are perceived competence (Harter, 1982) and selfconcept of ability (e.g. Nurmi & Aunola, 2005). The latter is used in the present study to refer to students' perceptions of their competence and task difficulty in a specific subject.
The value aspect of academic motivation in Eccles et al.'s (1983) theory, in turn, includes three components: Attainment value (the importance of doing well in a task in terms of selfschema), utility value (the instrumentality of a goal for reaching other goals), and intrinsic or interest value (the enjoyment one gets from engaging in an activity). Other concepts used to refer to motivation in a particular subject include intrinsic motivation (Deci, Vallerand, Pelletier, & Ryan, 1991;Gottfried, 1990;Harter, 1981) and interest (Schiefele, 1996). In the present study, we use the concept of interest to refer to how much students enjoy or like a certain school subject or a task, as it is close to the concept of intrinsic value in Eccles et al.'s (1983) theory.

Interest, Self-concept of Ability, and Academic Performance
There are a large number of studies on the relations between academic performance and self-concept of ability. These studies have shown on the one hand that self-concept of ability contributes to subsequent academic achievement (for a review see Valentine, DuBois, & Cooper, 2004). For example, studies by Eccles and her colleagues (Eccles et al., 1983; see also  found that among adolescents, self-perception of ability is one of the strongest predictors of subsequent performance even when controlled for the previous level of performance. Similarly, Marsh et al. (2005) found that self-concept of ability predicts both grades and standardized test scores among seventh-graders. On the other hand, academic achievement has also been found to provide a basis for positive self-concept. For example, Marsh et al. (2005) found that academic achievement predicted subsequent selfconcept of ability among seventh-graders. These results have been interpreted to mean that self-concept of ability and academic achievement form a reciprocal cycle with high selfconcept of ability, leading to increased investment and performance, which in turn leads to further increases in self-concept of ability in related domains (Eccles et al., 1983;Marsh et al., 2005). It has also been found that the associations between students' academic performance and self-concept of ability become stronger in their later school years (Denissen, Zarrett, & Eccles, 2007;Eccles, Wigfield, & Schiefele, 1998).
Two possible mechanisms may underlie the associations between self-concept of ability, interest, and academic performance. On the one hand, self-concept of ability provides a basis for interest in different school subjects, which then contributes to academic performance. Positive ability beliefs lead to higher motivation (e.g. Deci & Ryan, 1985;Wigfield & Eccles, 2000) and greater effort and investment (e.g. Wigfield et al., 1998) in academic tasks. This then leads to higher levels of performance (see Pintrich & Schunk, 1996). On the other hand, students' academic performance and related feedback may influence students' ability beliefs and interest. For example, Spinath and Spinath (2005) suggested that both competence beliefs and learning motivation rely strongly on normative ability feedback, such as grades. In addition, Eccles' model (Eccles et al., 1983 noted that earlier academic experiences and related feedback provide the basis for the development of students' self-concept of ability, which further influences their subjective task values: Students who are performing well in a certain subject also feel that they are competent in that subject, which then leads to higher interest and enjoyment of the subject (see also Deci et al., 1991;Gottfried, 1990;Köller et al., 2001).
One major limitation of the previous research is that only a few studies have examined the direction of influence between academic achievement, interest, and self-concept of ability simultaneously (Marsh et al., 2005;Skaalvik & Valas, 1999). In one study, Marsh et al. (2005) investigated the cross-lagged associations between math-related self-concept of ability, interest, grades, and test scores among seventh-grade students in Germany. The results showed, first, that the association between math-related self-concept of ability and math grades and test scores was reciprocal: self-concept of ability predicted subsequent math achievement, but also, math achievement predicted later self-concept of ability. However, the predictive power of test scores was stronger than the predictive power of grades. Second, self-concept of ability also predicted later interest in math, and interest in math, in turn, was found to predict later self-concept of ability marginally significantly. No associations were found between interest and grades or test scores. However, Marsh et al. (2005) conducted some additional-and, in this research field, more traditionalpath analyses in which they paired the different variables and examined the causal ordering among various pairs of the different constructs. The results of these models showed that some of the relatively small effects in the original model, including all the four variables (i.e. selfconcept of ability, interest, test scores and grades), were stronger when examining the constructs one pair at a time. One possible reason for this, according to Marsh et al. (2005), might be the multicollinearity of the different constructs. Although the study by Marsh et al. was longitudinal, it included only two measurement points across one year and focused only on math-related self-concept of ability, interest, and performance. In another study, Skaalvik and Valas (1999) investigated the cross-lagged associations between achievement, interest, and self-concept of ability in relation to both mathematics and language arts. Even though their study investigated three groups of students-third, sixth, and eighth-gradersstudents were studied only twice, at the beginning and end of each grade. They found that academic achievement affected both subsequent self-concept and motivation, whereas selfconcept and motivation did not predict academic performance.

Self-Concept of Ability as a Mediator between Academic Performance and Interest
To our knowledge there are no earlier studies examining whether self-concept of ability would act as a mediator between academic performance and interest. Studies that have examined the cross-lagged associations between all these three constructs simultaneously have included only two measurement points (Marsh et al., 2005;Skaalvik & Valas, 1999), which does not allow for testing a mediator hypothesis. However, there is a reason to assume that self-concept of ability may act as a mediator between performance and interest. Namely, it has been suggested that students' self-concept of ability provides a basis for their interest in particular school subjects (Eccles et al., 1983;: individuals are more likely to be motivated in activities in which they believe themselves to have good capabilities and expect success. In fact, many studies have shown that interest and self-concept of ability are positively correlated (Deci & Ryan, 1985;Eccles & Wigfield, 1995;Gottfried, 1990;1981) and that this positive association strengthens with age (Fredricks & Eccles, 2002;Jacobs, Lanza, Osgood, Eccles, & Wigfield, 2002;Wigfield et al., 1997). However, only a few studies have examined the relations between interest and self-concept of ability by using longitudinal data. Moreover, the findings of these studies are contradictory. Jacobs et al. (2002), for example, found that decrease in competence beliefs accounted for a decline in task value from Grade 1 to Grade 12; however, they did not study cross-lagged associations between interest and self-concept of ability. Marsh et al. (2005) and Spinath and Steinmayr (2008) in turn examined the cross-lagged effects between interest and self-concept of ability in longitudinal studies lasting approximately one year. Marsh et al. (2005) found evidence for reciprocal relations between math-related interest and self-concept of ability, whereas Spinath and Steinmayr (2008) found a hardly observable relationship. Neither of the other longitudinal studies have found cross-lagged associations between interest and self-concept of ability over time (Skaalvik & Valas, 1999;Spinath & Spinath, 2005).
The present study aims to extend these prior studies by examining the lagged relations between interest, self-concept of ability, and performance in both mathematics and reading by using longitudinal data over seven years, from the very beginning of comprehensive school to the seventh grade of comprehensive school.

Aims
The present study examined the following research questions: (1) Does students' academic performance in mathematics and reading predict their selfconcept of ability and interest? Does the self-concept of ability mediate the impact of academic performance on interest? We hypothesized that academic performance would predict later self-concept of ability (Hypothesis 1a) (Marsh et al., 2005) and interest (Hypothesis 1b) (Deci et al., 1991;Gottfried, 1990;Skaalvik & Valas, 1999), and that the self-concept of ability would mediate the impact of academic performance on interest (Hypothesis 1c) (Eccles et al., 1983). (2) Does students' self-concept of ability predict their subsequent academic performance? We expected that a high level of self-concept of ability would predict later academic performance, as students with positive self-perceptions of their abilities are more likely to show more effort, persistence, and cognitive engagement in academic tasks (Hypothesis 2) . (3) Does students' interest predict their subsequent academic performance? We expected that a high level of interest would predict subsequent academic performance, as students who value and are interested in specific subjects are more likely to enjoy tasks in those subjects, to be more adaptively engaged in them and to choose similar tasks in the future more (Hypothesis 3) (Gottfried et al., 1994;Köller et al., 2001). As previous findings are somewhat inconsistent, we expected that the crosslagged associations between academic performance, on the one hand, and selfconcept of ability and interest, on the other, may go either way.

Method
The Finnish School System The Finnish school system differs from its counterparts in many European countries and the United States. At age seven, Finnish children start comprehensive school, which divides into a lower (grades 1-6) and an upper level (grades 7-9). After comprehensive school, the large majority of adolescents continue their secondary education for a further three years, either in upper secondary school or in vocational school. Those who complete upper secondary school typically go on to universities or polytechnics. Those who complete vocational school either enter the labor market or go on to a polytechnic.

Participants
The present study is a part of the ongoing Jyväskylä Entrance into Primary School (JEPS) study (Nurmi & Aunola, 1999a), the aim of which is to examine children's academic and motivational development from the very beginning of their school career until the end of comprehensive school. The sample consisted of students from two medium-sized districts in central Finland. The two school districts were chosen because they included children growing up in both urban and semi-urban areas. The students were examined four times during their comprehensive school years. The sample size changed at each measurement point, because the students participating in the study had relocated, refused to continue participation in the study or did not participate at certain measurement points, and also because some new students entered the examined classrooms. The changes in the sample size across measurement points were found to be random rather than systematic. In Grade 1 the sample size was 216 (104 girls, 112 boys; 7-8 years old), in Grade 2 it was 224 (8-9 years old), in Grade 4 it was 228 (10-11 years old), and in Grade 7 it was 231 (13-14 years old). However, of the original 216 students, 215 were also examined in Grade 2, 197 in Grade 4, and 187 in Grade 7. A total of 178 students participated at each measurement point. Those students who dropped out of the study did not differ statistically significantly from other participants. At all the measurement points, students were interviewed or asked to fill in a questionnaire on their interest and self-concept of abilities in the spring term of the school year. Students' academic performance was assessed by questionnaires filled in by the students' teachers in grades 1, 2, and 4. In Grade 7, students' academic performance was measured by self-report.
The number of classes varied from year to year. In Grade 1 the students came from 17 classes, in Grade 2 students came from 19 classes, in Grade 4 they came from 32 classes (from 14 classes, only one student participated), and in Grade 7 from 27 classes (from 16 classes, only one student participated). The increase of classes over time was due to the fact that some of the original participants moved to another district and, thus, to another school. The class composition also changed from one year to another. In order to check whether the students in the same class in a certain year would be more similar to each other than to students coming from other classes, we calculated intraclass correlations for each study variable, separately for each year. The results showed that the intraclass correlations varied in interest between .003 and .095, in self-concept of ability between .03 and .068, and in academic performance between .011 and .095. In addition, we calculated the design effects for each variable. According to Muthén (1999; see also Muthén & Satorra, 1995), It is really not the size of the intraclass correlation that is the issue. It is the size of the design effect, which is a function of the intraclass correlation and the average cluster size. A design effect greater than 2 indicates that the clustering in the data needs to be taken into account during estimation. The design effect is approximately equal to 1 + (average cluster size -1)*intraclass correlation.
The design effects we calculated were all below 2. Therefore, no multilevel modeling was used in our study.

Measures Interest
Students' interest in reading and mathematics was assessed in the first, second, and fourth grade in an interview using the Task Value Scale for Children (TVS-C; Nurmi & Aunola, 1999boriginally Eccles et al., 1983). This particular scale measures only the interest value component. The scale consisted of items measuring students' interest (i.e. interest in or liking for a particular task) in reading tasks (three items, e.g. "How much do you like reading?") and in mathematical tasks (three items, e.g. "How much do you like mathematics?"). In the measurement procedure, the students were first read the question, after which they were then shown a set of five faces drawn to depict an evaluative scale running from very positive to very negative. The students were then asked to point out the picture which most describes their liking for a particular subject (unhappy face/1 = "I do not like it at all/ I dislike doing those tasks"; happy face/5 = "I like it very much/I really enjoy doing those tasks").
In the seventh grade, students' interest was assessed with a questionnaire asking (1) how much they like (a) mathematics and (b) Finnish language and (2) how much they like doing tasks related to (a) mathematics and (b) Finnish language (Eccles et al., 1983). The Cronbach's alpha reliabilities for the interest assigned to mathematics were .88, .91, .93, and .84, respectively. The Cronbach's alpha reliabilities for reading-related interest were .83, .81, .84, and .77, respectively.

Self-concept of ability
Students' self-concept of ability in mathematics and reading was assessed in the first, second, and fourth grades at interviews using the Self-Concept of Ability Scale (Nicholls, 1978; see also Aunola, Leskinen, Onatsu-Arvilommi, & Nurmi, 2002;Nurmi & Aunola, 2005). In the test, students were presented with a sheet of paper showing 20 faces in a line from the top to the bottom of the page. They were told that the faces represented students of the same age as themselves and that the one at the top of the page represented the student who was best at mathematics, and so on, down to the poorest performer. In the first and second grades, students were asked about their self-concept of ability in reading by using one question and about their self-concept of ability concerning math by using two questions: "Now, can you show me how good you are at reading/at maths/at writing numbers? Which one are you?" The participants responded by pointing to one of the faces. The faces at the top of the page scored lower values. In Grade 4 the measurement procedure was the same, except that self-concept of ability in both reading and math was assessed by using one question ("Can you show me how good you are at maths/reading?"). When the results for self-concept of ability are presented, the scale is reversed (i.e. the better a student thought he/she was, the bigger the value he/she scored for self-concept of ability).
In Grade 7, students filled in a questionnaire (based on the ideas presented by Eccles & Wigfield, 1995) in which they were asked: (1) "How good are you at mathematics/the Finnish language?"; (2) "How good do you think you are at mathematics/Finnish language compared to the other students in your class?"; and (3) "How hard are tasks related to mathematics/ Finnish language-related tasks for you?" The response scale ran from 1 to 5. The Cronbach's alpha reliabilities for self-concept in mathematics were .56 in Grade 1, .60 in Grade 2, and .87 in Grade 7. The reliability for self-concept in reading was .91 in Grade 7. If there was only one item for construct, this was evaluated when estimating SEM models by setting the measurement errors to equal wherever possible. The reliabilities for self-concept in reading were .74 in Grade 1 and .74 in Grade 2.

Academic performance
Students' academic performance was assessed in grades 1 and 2 by asking the students' teachers to evaluate how good the students were at mathematics and reading on a scale from 1 to 5. In Grade 4 teachers were asked to rate, on a scale from 1 to 5, students' (1) comprehension and (2) technical reading skills, and students' (1) mechanical and (2) applied skills in mathematics. Teacher ratings of students' academic performance have been shown to correlate highly with students' test performance (Aunola & Nurmi, 2008). In Grade 7 students were asked to report their grade in mathematics and the Finnish language from the preceding term (i.e. from the preceding autumn). Self-reported grade point average has been shown to correlate by .96 with actual grade point average (Holopainen & Savolainen, 2005). The reliabilities for performance in mathematics were .72 in Grade 1, .77 in Grade 2, and .93 in Grade 4; for performance in reading they were .77 in Grade 1, .77 in Grade 2, and .92 in Grade 4.

Analysis Strategy
All analyses were done in two steps. First, in order to investigate the stability of the constructs before entering concurrent and cross-lagged associations to the models, a stability model with four factors (measurement points) was built separately for each of the measured constructs-academic performance, self-concept of ability, and interest-separately for math and reading. The gender differences in the variable means were controlled in each model by entering it as a control variable. Second, the cross-lagged relations between performance, self-concept of ability, and interest were examined according to the hypotheses, separately in math and reading, with the following procedure: (1) a model including performance and self-concept was estimated without any cross-lagged paths between these two constructs; (2) paths from performance at each measurement point to self-concept at following measurement point were estimated to test Hypothesis 1a; (3) paths from self-concept at each measurement point to performance at following measurement point were estimated to test Hypothesis 2; (4) cross-lagged paths from performance to later self-concept and from self-concept to later performance were estimated in the same model; (5) a model including performance and interest was estimated without any cross-lagged paths between these two constructs; (6) paths from performance at each measurement point to interest at following measurement point were estimated to test Hypothesis 1b; (7) paths from interest at each measurement point to performance at following measurement point were estimated to test Hypothesis 3; (8) cross-lagged paths from performance to later interest and from interest to later performance were estimated in the same model; (9) a model including self-concept and interest was estimated without any cross-lagged paths between these two constructs; (10) paths from self-concept at each measurement point to interest at following measurement point were estimated; (11) paths from interest at each measurement point to selfconcept at following measurement point were estimated; (12) cross-lagged paths from self-concept to later interest and from interest to later self-concept were estimated in the same model.
After each model was estimated, χ 2 difference test for nested models testing the needs of cross-lagged paths was used. As a last step of the analyses (13), the statistical significance of the each single mediator effect from academic performance to interest via self-concept of ability was tested.
When building the stability model for reading performance (χ 2 (4) = 0.66, p = .96, CFI = 1.00, TLI = 1.00, RMSEA = .00, SRMR = .00; see Figure 2), the measurement errors of first and second measurement were again set to equal in order to estimate the error variance. With this we were able to evaluate the reliabilities of the single-item measures of performance. The model showed high stabilities between successive measurements, especially between Grade 1 and Grade 2, as well as between Grade 2 and Grade 4 ( Figure 2). In case of reading-related self-concept of ability (χ 2 (7) = 14.76, p < .05, CFI = .98, TLI = .93, RMSEA = .06, SRMR = .04; see Figure 2), measurement errors of first and second measurement were set to equal. In addition, the modification indices suggested that the path between grades 1 and 7 should be added to the model. The model showed moderate stabilities between successive measurements, especially between Grade 1 and Grade 2, as well as between Grade 2 and Grade 4. Since modification indices suggested that the path between grades 1 and 7 should also be estimated, the stability between grades 4 and 7 had to be calculated separately, and it appeared to be .27 ( Figure 2). Finally, the stability model for reading-related interest (χ 2 Figure 1. Stability models for math-related skills, self-concept of ability, and interest. Standardized solution. *p < .05; ** p < .01; *** p < .001. (40) = 45.37, p = .26, CFI = .99, TLI = .99, RMSEA = .02, SRMR = .04; see Figure 2) showed moderate or low stabilities between successive measurements.

Model for Mathematics
In the second step of the analyses, separate models that combined the previous models for academic performance, self-concept of ability, and interest were created for (a) math ( Figure 3) and (b) reading ( Figure 4). The model included stability coefficients as well as cross-lagged paths. The results for this final math model (χ 2 (228) = 313.449, p < .01, CFI = .97, TLI = .96, RMSEA = .04, SRMR = .07; see Figure 3) showed that math performance at each time point predicted self-concept of ability at the next time point: the better math performance was, the more positive was self-concept of ability later on. The results showed further that self-concept of ability also predicted interest later on: the more positive the child's self-concept of ability was in Grade 2 and Grade 4, the higher was his/her interest in the next measurement point. In no case did self-concept of ability predict later academic performance or interest predict self-concept of ability or academic performance.
To test the statistical significance of the mediator effects from academic performance to later interest via self-concept of ability, the indirect effects were calculated. For the mediator effect from academic performance at Grade 2 to interest at Grade 7 via self-concept of ability at Grade 4, the standardized value was 0.06 (p < .05) Therefore, the mediator effect was statistically significant. The mediator effect from academic performance at Grade 1 to interest at Grade 4 via self-concept of ability at Grade 2 (standardized value 0.06), in turn, was statistically marginally significant (p < .10).

The model for reading
The results for reading (χ 2 (184) = 196.188, p = .26, CFI = 1.00, TLI = .99, RMSEA = .02, SRMR = .05; Figure 4) showed that reading performance at Grade 1 predicted self-concept of ability at Grade 2 and performance at Grade 4 predicted self-concept of ability at Grade 7: the better reading performance was, the more positive was self-concept of ability later on. Selfconcept of ability predicted interest in two cases: the more positive a student's self-concept of ability was in Grade 2, the higher also was also his/her interest in Grade 4. Moreover, the more positive a student's self-concept of ability was in Grade 4, the higher was his/her interest in Grade 7.
In no case did self-concept of ability predict later academic performance or interest predict self-concept of ability or academic performance. The mediator effect from academic performance at Grade 1 to interest at Grade 4 via selfconcept of ability at Grade 2 was found to be marginally significant (standardized estimate of the indirect effect 0.09, p < .10).

Discussion
The aim of the present study was to examine the developmental dynamics-that is, crosslagged associations-between students' interest, self-concept of ability, and academic performance in mathematics and reading as students moved from the first school year to the seventh grade of comprehensive school. The results showed, in particular, that both reading and math performance predicted students' self-concept of ability later on. Evidence was also found that self-concept of ability mediated the association between performance and interest. Moreover, no evidence was found for the assumption that self-concept of ability or interest would predict subsequent math and reading performance. The first research question of the present study examined whether the students' academic performance in mathematics and reading would predict their related self-concept of ability and interest, and whether self-concept of ability would mediate the impact of academic performance on interest. As expected (Hypothesis 1a), the results showed that students' academic performance in both math and reading predicted their subsequent self-concept of ability. In the case of mathematics, the higher the performance at a certain measurement, the more positive was students' self-concept of ability at the next measurement point. In relation to reading this was true from first to second grade, as well as from Grade 4 to Grade 7: the higher the performance in reading, the more positive was students' readingrelated self-concept of ability later on. This result is in line with Eccles' expectancy-value model, according to which earlier academic experiences and related feedback provide the basis for the development of students' self-concept of ability (Eccles et al., 1983;Eccles et al., 1998). This relation has also been found in many previous studies carried out among both younger Chapman & Tunmer, 1997) and older (Marsh et al., 2005) students. Students' performance may be associated with their self-concept of ability via many mechanisms. For example, teachers communicate their conceptions of the level of their students' performance in many ways in the daily classroom situations, which provides students with a salient source of feedback about their performance. This, in turn, affects students' beliefs in their abilities (Gottfried, 1990). In addition, other processes, such as social comparisons with other students, may play a role in how students' performance contributes to their self-concept of ability.
One interesting finding of the present study was that in mathematics performance predicted self-concept of ability at each consecutive measurement point, whereas in reading the same was true from Grade 1 to Grade 2 and from Grade 4 to Grade 7. This result may be due to the fact that the processes of learning literacy and mathematics differ from each other. The nature of the Finnish language makes it relatively easy for children to learn to read, and the large majority of them achieve accurate and fluent word-reading skill before the end of the first school year (Aunola, Nurmi, Niemi, Lerkkanen, & Rasku-Puttonen, 2002;Seymour, Aro, & Erskine, 2003). This might explain why students' skills in reading predicted their reading-related self-concept of ability at the very beginning of their school career: the beliefs concerning one's abilities are established while learning to read, which typically takes place before the end of the second grade. During later schooling different kinds of skills in reading, such as reading comprehension, are learned. This might be another important point of schooling during which reading skills might play an especially important role on students' ability beliefs. By contrast, learning mathematics happens in a hierarchical manner: learning basic skills is a necessary foundation for mastering more complex skills and procedures (Entwisle & Alexander, 1990). Therefore, it is a process during which children's individual differences tend to increase continuously during the school years (Aunola, Leskinen, Lerkkanen & Nurmi, 2004). Hence, these increasing individual differences are also likely to be reflected in children's self-concept of abilities.
Our results provided no support for our Hypothesis 1b suggesting that academic performance predicts subsequent interest. This is an interesting finding, since previous research has shown relatively strong evidence for the relation between performance and interest (see e.g. Deci et al., 1991;Gottfried, 1990;Köller et al., 2001;Viljaranta et al., 2009): students who perform well at school continue to enjoy academic tasks more than students who have difficulties in academic topics. However, the most important finding in the present study was the result concerning the mediator effect of self-concept of ability in the relation between academic performance and interest (Hypothesis 1c). According to our results, the effect from academic performance to later interest goes via self-concept of ability. The results showed that students who performed better in mathematics showed more positive self-concept of ability which then predicted their later interest in mathematics. Some evidence was also found for this in the case of reading during the beginning of the school career. Overall, these results suggest that students who are performing well in mathematics and reading also feel that they are competent in it, which then leads to higher interest in and enjoyment of these subjects. This result adds in an important way to several earlier results showing that students' self-concept of ability predicts their later interest (Eccles et al., 1983, Eccles et al., 1993Jacobs et al., 2002;Marsh et al., 2005;Wigfield & Eccles, 2000), by indicating it is actually students' actual performance that contributes to their later interest via these ability beliefs. Eccles' expectancy-value model suggested this relation existed (Eccles et al., 1983;Eccles et al., 1998) but it had not been empirically tested with longitudinal crosslagged data spanning over several years. Although some of the mediating effects were statistically only marginally significant, our results indicate that students' earlier academic experiences and related feedback provide the basis for their later ability beliefs which further predicted interest in different subjects.
Our second and third research questions were whether students' self-concept of ability or interest would predict their subsequent academic performance. Surprisingly, no support was found for our Hypothesis 2 or Hypothesis 3 suggesting that a high level of self-concept of ability or interest would contribute to later academic performance. Our findings are surprising, since there are several previous studies showing strong associations between self-concept of ability and later performance (e.g. Marsh et al., 2005; see also Marsh & Craven, 2006;Valentine et al., 2004), as well as between interest and later performance (e.g. Aunola et al., 2006;Ecalle, Magnan, & Gibert, 2006;Gottfried, 1990;Viljaranta et al., 2009;Wigfield, 1997). There are several possible reasons why our results were different from many previous ones. First, our study spanned seven years of schooling starting from grade 1, whereas several other studies, especially those concerning self-concept of ability, have concentrated on older students (e.g. Marsh et al., 2005). Thus, it is possible that the impact of selfconcept of ability on later academic performance is particularly evident during later school years, when the self-concept of ability has become more stable. During the comprehensive school years students receive a growing amount of feedback on their performance and skill development. As they get older they may also become better able to integrate this feedback with the beliefs they have about their skills, strengths and weaknesses (Wigfield & Eccles, 2000;Stipek & Mac Iver, 1989). This development finally leads to a more stabilized self-concept of ability, which then could have a more powerful effect on later academic performance. In the case of interest it is possible, as suggested by Köller et al. (2001) and Wigfield and Eccles (1992Eccles ( , 2000, that the role of interest as a predictor of academic outcomes may become more important in the later school years, when students have to make active decisions concerning the future course of their education. Second, our study investigated the cross-lagged associations across one year (from Grade 1 to Grade 2), across two years (Grade 2 to Grade 4), and across three years (Grade 4 to Grade 7). Our results may have been different if the measurements had been conducted at shorter intervals. For example, it is possible that the impact of self-concept of ability and interest does not span more than one year of a student's school career. Therefore, it is possible that in order to capture the developmental dynamics and all the possibly existing associations between selfconcept of ability or interest and academic performance, the measurements should have been conducted at shorter intervals, which was generally the case in the previous studies finding a cross-lagged association. Third, academic skills show substantial stability over time. Therefore, it may be difficult to find constructs that can statistically significantly predict such a stable construct over a long period of time by using the kind of method used in the present study. Due to the relatively low stability of self-concept of ability and interest, it is more plausible that performance predicts this variable, rather than vice versa (i.e. because self-concept is not as stable as performance, it is more prone to external impacts). One additional reason for the differences between our results and those which show strong predictive power from self-concept or interest to later performance is differences in the data analysis. As Marsh et al. (2005) suggest, the multicollinearity of the different constructs might lead to weaker associations between different constructs in models including more than two constructs, compared to models where associations between only two different constructs are examined.
Finally, in the model concerning reading, the modification indexes suggested there should be a path from reading-related self-concept of ability from Grade 1 to Grade 7. This result indicates that reading-related self-concept of ability in Grade 1 includes some information valuable in predicting the same variable in Grade 7 that is not present in Grade 2 and Grade 4 measurements. For example, it is possible that self-concept of ability is impacted not only by the level of reading skills but also some attitudinal variables, such as optimism, which are independent of the development of reading skills.
The results of the present study have important practical implications for teachers and other adults working in important roles with students: students' academic performance and related feedback can have long-term effects on their beliefs about themselves as learners and, through these beliefs, on their task motivation; our results suggest that in order to provide every student with positive learning experiences, teachers should pay special attention to the ways in which they encourage and give feedback to students who differ in their level of performance.
There are also some limitations that should be taken into account in any attempts to generalize the findings of the present study. First, some of the constructs were measured by only one item. In the case of these constructs, we made an effort to evaluate the reliability of these items when estimating SEM models by setting the measurement errors to equal every time it was possible, although this is not an ideal solution for this problem. However, in the third measurement of self-concept of ability it was not possible to estimate the error variance, the associations to the other factors in the model are under-estimated and the statistical power to find statistically significant associations is lowered. These effects are, however, small. Second, academic performance was measured by teacher ratings, not by actual test performance. Although teacher ratings have been shown to correlate highly with students' performance (Hecht & Greenfield, 2002), it is possible that test results may have given somewhat different result. Third, in grade 7, self-report of grades in math and reading were used. The reason why the measure concerning the performance was changed was that at the first three measurement points, teachers filled in a questionnaire concerning students' skills in reading and math; however, at Grade 7 students are no longer taught by only one teacher, and, therefore, teachers were no longer asked to participate in the study. Although in another Finnish study self-reported grade point average was shown to correlate highly with actual grade point average (.96;Holopainen & Savolainen, 2005), results from self-reports should be interpreted with caution (Kuncel, Credé, & Thomas, 2005).
Fourth, the measures were not identical across measurement points. There are several reasons for this. Most importantly, our study spanned over several years and it would have been difficult to use identical measures with children from grades 1 to 7. The contents of school subjects change and, consequently, it is not reasonable to evaluate the self-concept, for example, in the same way throughout all grades. Also, as they grow older, pupils are more capable of answering more detailed questions concerning their self-concept. Even though our measures were not equal at different measurement points, the reliabilities of our measures were relatively good across time. In addition, there is no reason to expect the differences in measures to have a notable effect on the cross-lagged paths that were the main focus of the present study. It is possible, however, that the differences could have an effect on the stabilities of each measured construct over time, so that stabilities might be somewhat lower when different measures have been used, compared to identical measures over time.
The fifth limitation of the study is that the measurements were conducted at intervals of 1-3 years. It is possible that in order to capture the developmental dynamics between selfconcept, interest, and performance the measurements should have been conducted at shorter intervals. Finally, the present study does not focus on the associations between different school subjects. In the future it would be interesting to examine, whether, for example, good skills and high self-concept of ability in one subject are related to lower level of skills and self-concept in another subject (the internal/external frame of reference model; see e.g. Marsh, 1986;Marsh & Hau, 2004).
The results of the present study increase our understanding of the developmental dynamics between students' academic performance, self-concept of ability, and interest by showing that during comprehensive school, students' academic performance in mathematics and reading is particularly associated with their subsequent ability beliefs and, in the case of mathematics, also through that to their interest. The findings have important practical implications for teachers and other important adults working with students: students' academic performance and related feedback can have long-term effects on their beliefs about themselves as learners and, through these beliefs, on their interest. Our results suggest that to provide every student with positive learning experiences, teachers should give special attention to the ways in which they encourage and give feedback to students who differ in their level of performance.