The reciprocity between various motivation constructs and academic achievement: a systematic review and multilevel meta-analysis of longitudinal studies

Abstract

The current multilevel meta-analyses and systematic review encompass 47 longitudinal studies of motivation and academic achievement in primary and secondary education. Our inclusion criteria extended beyond studies centred on self-belief type of motivation constructs, as we also selected studies using other motivation constructs (e.g. achievement goals). The findings offered mixed support for the reciprocity. The pooled effect of achievement on motivation was nearly twice as strong (β = 0.176) as that of motivation on achievement (β = 0.096). The effect of achievement on motivation held across different motivation constructs, albeit exhibiting more robust reciprocity for self-belief constructs (β = 0.227) than for other constructs (β = 0.091). The effect of motivation on achievement was exclusively found for self-belief (β = 0.157) and not for non-self-belief motivation. Overall, the medium-to-high heterogeneity between studies and within studies suggests that pivotal factors in the motivation–achievement dynamics may not have been captured yet.

Keywords:

• Academic achievement
• academic motivation
• reciprocity
• longitudinal
• meta-analysis
• systematic review

Introduction

The relationship of academic achievement and motivation has been the focus of research for decades, which yielded many influential theories (Eccles & Wigfield, 2002; Hattie et al., 2020; Murphy & Alexander, 2000) and a large body of empirical evidence (Kriegbaum et al., 2018; Robbins et al., 2004). Earlier work predominantly concentrated on the unidirectional effects of motivation on academic achievement (Calsyn & Kenny, 1977; Dweck, 1986), and a recent meta-analysis showed that motivation is an incremental predictor of academic achievement beyond intelligence (Kriegbaum et al., 2018). However, research in the past 10–20 years has been increasingly turning to reciprocal relationships: motivation boosts achievement and achievement in turn fuels motivation (Huang, 2011; Möller et al., 2011; Retelsdorf et al., 2014; Scherrer et al., 2020; Valentine & DuBois, 2005). To date, to the best of our knowledge, there exist three meta-analyses (Huang, 2011; Valentine & DuBois, 2005; Wu et al., 2021) which support the existence of a reciprocal relationship, although with varying conclusions about the relative strength of each link. Both Valentine and DuBois (2005) and Wu et al. (2021) found a stronger effect of initial academic achievement on subsequent motivation (with correlation coefficients β = 0.15/0.16 in the two meta-analyses) compared to the effect of initial motivation on subsequent academic achievement (β = 0.08 in both cases). Huang (2011) also identified the reciprocity but noted nearly equal effects of initial motivation on subsequent academic achievement (β ranging from 0.20 to 0.27) and of initial academic achievement on subsequent motivation (β between 0.19 to 0.25). However, these meta-analyses all had some limitations: (1) they encompassed empirical studies from mixed age and populations and (2) primarily focused on only one type of motivation construct, namely academic self-beliefs (Huang, 2011; Möller et al., 2014; Valentine & DuBois, 2005; Wu et al., 2021). Hence, there is a clear need for a systematic review and meta-analysis that covers a broader range of motivational constructs while centring on primary and secondary school populations.

Understanding the strength of each pathway holds significant importance both theoretically and practically. Theoretically, this knowledge aids in constructing more refined models of academic achievement. On a practical level, grasping the link between motivation and achievement can guide decisions on interventions when both student motivation and performance lag. For example, a school may vacillate between channelling its limited resources into a new pedagogical method to bolster math performance, or investing in a motivation-boosting intervention in the hope that better student motivation will in the long run elevate math performance more than the new teaching approach. Furthermore, discerning between self-concept and other motivation constructs could facilitate more precisely targeted interventions.

Over the past decade, studies on the reciprocal relation of academic achievement and motivation have grown not only in number but also in quality of data and analyses. More than two waves of data are now often included, which was previously rare; sample sizes have increased; and new powerful statistical models were introduced that are well-suited to investigate within- and between-individual changes and their mutual influences over time (Burns et al., 2020; Ehm et al., 2019; Hamaker et al., 2015). However, these recent longitudinal studies present mixed findings. Some support the reciprocal effect (Arens et al., 2017; Dicke et al., 2018; Grygiel et al., 2017; Guo et al., 2015; Marsh et al., 2018; Weidinger et al., 2017), while others do not (Grigg et al., 2018; Hirvonen et al., 2012; Möller et al., 2014; Nuutila et al., 2018; Viljaranta et al., 2014).

The reciprocal relationship of academic achievement and motivation may play out differently between younger and older learners (Chen et al., 2013; Weidinger et al., 2019). For instance, Chen et al. (2013) found that the effect of academic achievement on motivation (specifically, academic self-concept (ASC)) declined with advancing school age, while the effect of motivation on academic achievement increased over time. This signals the importance of considering students’ school age when investigating the reciprocal relationship of academic achievement and motivation. Although school grade does not always coincide with age, age is often indicative of school grade with the latter being one factor that distinguishes younger from older learners and vice versa. Moreover, from a practical point of view, to inform interventions at school, it is logical to review empirical evidence specifically in the compulsory, formal educational settings. Motivation tends to decline throughout primary and secondary education (Scherrer & Preckel, 2019). Consequently, for the current review and meta-analyses, only studies involving students in primary and secondary education were considered.

Our systematic review and meta-analyses (one for the motivation → achievement link, the other for the achievement → motivation pathway) aim to contribute to the literature by testing the generalisability of the reciprocity across a wide spectrum of motivation constructs rather than restricting the analyses to self-belief constructs. Previous empirical research on the reciprocal relationship between motivation and academic achievement predominantly focused on one type of motivation construct: ASC or closely related self-belief constructs such as self-efficacy (Huang, 2011; Möller et al., 2020; Valentine & DuBois, 2005; Wu et al., 2021). However, established theories of motivation implicitly or explicitly conceptualise the motivation and achievement reciprocity using other motivation constructs such as interest, intrinsic motivation, and achievement goal orientation (Vu et al., 2022). For example, in self-determination theory, intrinsic motivation contributes to achievement, which, in turn, fosters increased intrinsic motivation via feelings of competence (Ryan & Deci, 2000). Thus, it is a significant gap in the literature that the reciprocal relations between academic achievement and motivation constructs other than ASC have not undergone meta-analysis. This gap was also noted by other authors, acknowledging the challenge of conducting such a meta-analysis due to the scarcity of studies using motivational measures other than ASC (Möller et al., 2020). In these current meta-analyses, we capitalise on recently available studies to scrutinise the reciprocity of motivation–achievement relations across a range of major motivation constructs identified in seminal theoretical syntheses of Eccles and Wigfield (2002), Hattie et al. (2020), and Murphy and Alexander (2000).

Our search results suggest that about half of the relevant studies focus on self-belief motivation constructs, which delineates students’ perception of themselves and their abilities in academic domains. Among the extensively studied self-belief constructs are self-efficacy and ASC, both are undergirded by the idea that motivation stems from the individuals’ belief in their competence (Bandura & Walters, 1977; Bong & Clark, 1999; Eccles & Wigfield, 2002). However, these constructs can be theoretically and empirically distinguished from each other (Bong & Skaalvik, 2003; Ferla et al., 2010; Scherer, 2013; though differing views exist, such as in Marsh et al., 2019). ASC resides as the academic component within a hierarchical model of self-concept where global self-concept, equivalent to self-esteem, holds the top position (Marsh & Craven, 2006; Marsh & Martin, 2011). On the other hand, self-efficacy revolves around the expectation to succeed at a given task while ASC is the aggregated sense of being academically competent derived from past success and failure experiences (Bong & Skaalvik, 2003). Despite ASC being a motivational concept, there is some overlap between ASC measures and achievement. ASC questionnaires often prompt respondents to reflect on their achievement (e.g. ‘I get good marks in most academic subjects’, ‘I learn quickly in most academic subjects’ (Marsh & O’Neill, 1984), which closely resembles inquiries about grades – a typical measure of achievement). This again underscores the need to assess the reciprocity between academic achievement and motivation constructs beyond ASC and its counterparts.

Well-studied non-self-belief constructs include, for example, intrinsic motivation (Ryan & Deci, 2000), which denotes the innate inclination to learn something for its inherent enjoyment, and interest, which relates to emotions (or valence) directly tied to a learning activity (rather than external rewards) (Eccles & Wigfield, 2002). Other motivation constructs outside the realm of self-belief include extrinsic motivation, where the drive for academic success stems from external rewards (Ryan & Deci, 2000) and achievement goals, which are the cognitive representations of future states that guide academic behaviours. Students can set achievement goals that range between mastery goals (i.e. motivated to acquire competence and achieve task mastery) and performance goals (i.e. motivated to exhibit competence and surpass others’ performance). The mastery-performance dichotomy interacts with the approach-avoidance dimension, leading to four types of achievement goals: mastery-approach, mastery-avoidant, performance-approach, and performance-avoidant (Elliot, 2005; Elliot & Harackiewicz, 1996; Liem & Senko, 2022; Urdan & Kaplan, 2020). In practical settings, such non-self-belief motivation constructs prove effective targets for interventions. Many successful manipulations to change non-self-belief motivation constructs have been reported (Lazowski & Hulleman, 2016). Hence, the present study will encompass both self-belief as well as non-self-belief motivation constructs (see Table 1) for comprehensive analysis.

Table 1. The categories that studies were assigned to, based on whether self-belief or non-self-belief motivation constructs were used.

	Study	Motivation construct	Category
1	Bakadorova and Raufelder (2020) (see Appendix G)	Academic self-concept	Self-belief
2	Chen et al. (2013)	Academic self-concept	Self-belief
3	Dicke et al. (2018)	Academic self-concept	Self-belief
4	Ehm et al. (2019)	Reading self-concept	Self-belief
5	Fu et al. (2020)	Academic self-competence	Self-belief
6	Grigg et al. (2018)	Math self efficacy	Self-belief
7	Grygiel et al. (2017)	Academic self-concept	Self-belief
8	Guo et al. (2015)	Academic self-concept, intrinsic motivation and attainment task values	Self-belief and non-self-belief
9	Hebbecker et al. (2019) (see Appendix G)	Extrinsic and intrinsic reading motivation	Non-self-belief
10	Li et al. (2021)	Attainment task values	Non-self-belief
11	Liu and Hou (2017) (see Appendix G)	Intrinsic motivation and extrinsic motivation	Non-self-belief
12	Marsh et al. (2018)	Academic self-concept	Self-belief
13	Möller et al. (2011)	Self-concept	Self-belief
14	Möller et al. (2014)	Self-concept	Self-belief
15	Niepel et al. (2014)	Academic self-concept and achievement goal	Self-belief and non-self-belief
16	Pinxten et al. (2014)	Math self-concept, enjoyment, and competency beliefs	Self-belief and non-self-belief
17	Potvin et al. (2020)	Science and technology self-concept	Self-belief
18	Retelsdorf et al. (2014)	Reading self-concept	Self-belief
19	Scherrer et al. (2020)	Achievement goals	Non-self-belief
20	Schiefele et al. (2016)	Extrinsic and intrinsic reading motivation	Non-self-belief
21	Schöber et al. (2018)	Self-efficacy	Self-belief
22	Seaton et al. (2014)	Math self-concept and achievement goals	Self-belief and non-self-belief
23	Stäbler et al. (2017)	Math self-concept	Self-belief
24	Sticca et al. (2017) (see Appendix G)	Academic self-concept	Self-belief
25	Walgermo et al. (2018)	Reading self-concept	Self-belief
26	Weidinger et al. (2017)	Intrinsic motivation in math	Non-self-belief
27	Weidinger et al. (2019)	Ability self-concept in math and German	Self-belief
28	Xu et al. (2020)	Goal orientation and interest in math homework	Non-self-belief
29	Zee et al. (2021)	Task motivation and self-efficacy	Self-belief and non-self-belief

Notes. The 29 studies provided 33 samples that were included in our quantitative synthesis. Chen et al. (2013), Scherrer et al. (2020), Li et al. (2021), Grygiel et al. (2017), and Schöber et al. (2018) each described two samples while Weidinger et al. (2017) and Weidinger et al. (2019) describe the same sample.

Another factor worth considering is the length of the time intervals in between measurement waves (Hecht & Zitzmann, 2021), which varies across studies. While some studies assess academic achievement and motivation annually (Arens et al., 2017), others do so more frequently, such as three times a year (Weidinger et al., 2019). It is currently unclear to what extent the time interval between measurement waves affects the size and form of the reciprocal relations between motivation and academic achievement (Vu et al., 2022). Therefore, the current systematic review and meta-analysis will investigate the effect of the time intervals between measurement waves on the reciprocity between academic achievement and motivation.

Last but not least, it is worth noting that achievement can be measured using standardised tests and grades in schools (Arens et al., 2017; Marsh et al., 2018). Additionally, assessment may extend to teacher evaluations or self-assessment (Chamorro-Premuzic et al., 2010). Past meta-analyses and studies on the relationship between motivation and academic achievement (beyond reciprocal relations) have also been mainly concerned with teacher-assigned grades and standardised test scores as the key indicators of academic achievement given their relatively objective nature compared to other measures (Möller et al., 2014; Valentine & Dubois, 2005). Therefore, these meta-analyses specifically centre on teacher-assigned grades and standardised test scores as the primary operationalisations of academic achievement.

The present study

For these reasons, the present systematic review and meta-analyses delve into the longitudinal reciprocal relationship between academic achievement and various motivation constructs among primary and secondary education students, focusing on the time span from 2011 to 2022, a period marked by the emergence of studies featuring innovative designs. Given that existing meta-analyses covered publications predating 2011, our focus is on studies published from 2011 onwards (although we did include studies predating 2011 when they included non-self-belief constructs – see section ‘Methods’). While Wu et al. (2021) covered post-2011 studies, they did not include studies with non-self-belief type of motivation constructs, which we specifically target.

Our investigation aims to address several key research questions: (1) Does the reciprocal relationship between motivation and academic achievement persist in studies published since 2011, involving primary and secondary education students? (2) Does this reciprocity generalise to the relations between academic achievement and non-self-belief motivation constructs? (3) Do the magnitudes of reciprocal relations vary across participant age groups within the primary/secondary education period? And (4) is there an effect of the length of time intervals (i.e. whether time between study waves was long or short) on the magnitude and direction of reciprocal relations? For studies where applicable effect measures could be extracted, we performed two meta-analyses (one for the motivation → achievement direction, and the other the achievement → motivation direction). These quantitative analyses were followed up by a qualitative analysis of all 47 studies found in our systematic review (see details in section ‘Methods’). Additionally, we also explored the variations in the sample populations to uncover potential disparities and test ecological validity of the reciprocal relationship.

Methods

In- and exclusion criteria

We first selected published papers for our systematic review. To be included, studies needed to (a) exhibit a longitudinal design with at least two waves of measurement, (b) solely involve students in primary and secondary education (i.e. excluding university students), (c) focus on typically developing students, (d) investigate reciprocal relations between academic achievement and motivation, (e) be written in the English language, and (f) be published in a peer-reviewed academic journal between 2011 and March 2022. Subsequently, for our meta-analyses, we further refined the selection to studies reporting all effect sizes (standardised beta coefficients) and standard errors for pathways from motivation in one wave to academic achievement in a subsequent wave, and vice versa.

Studies were excluded (a) when the students’ motivation or academic achievement was measured under specific circumstances such as following a natural disaster or in foreign language learners, rather than in regular formal educational settings, (b) when it was not clear from the abstract that the study was longitudinal, and (c) when the study investigated achievement in sports instead of academic achievement. The numbers of included and excluded studies for the meta-analyses as well as for the systematic review can be found in the PRISMA flow diagram (Figure 1).

Figure 1. PRISMA flow diagram showing how studies were identified, screened, and selected for the qualitative and quantitative syntheses. n: number of records/studies/samples. Graph was made based on Page et al. (2021).

Database searches

To systematically identify studies for inclusion, the databases ERIC and PsycInfo were searched. The full search string, that was used for the search in both databases, can be found in Appendix A. The only difference between the search in PsycInfo and ERIC was that PsycInfo provided an additional methodological filter, which allowed the automatic exclusion of the studies that were not longitudinal.The reference manager Zotero was used to identify and exclude duplicates from the two searched databases. A summary of the number of studies identified in each database, as well as the number of excluded and included studies can be found in the flow diagram (Figure 1). Next, each of the remaining studies was screened for the inclusion criteria. For a publication to be considered for full text retrieval, it was sufficient if the abstract disclosed that the study was (a) longitudinal (with at least two waves) and (b) investigated a reciprocal relationship of one or more concepts of motivation and academic achievement. The full texts of the remaining studies were then retrieved. If possible, the effect sizes (standardised beta coefficients), standard errors, and study characteristics (Appendix B) were extracted. In case the effect size or the standard error was not provided, or only partially reported, the authors were contacted by e-mail and asked to share the missing information. If they could not, or did not respond in time, their study was excluded from the meta-analyses, but remained in the systematic review.

Meta-analyses (quantitative synthesis)

For the purpose of conducting the meta-analyses, standardised beta coefficients and standard errors were extracted from each paper. We focused on standardised regression coefficients as they quantify the expected change in the criterion variable, expressed in standardised units, associated with one standardised unit of change in a predictor variable. Since studies differed substantially in what variables were used, using standardised coefficients is preferable to using unstandardised ones (also see Valentine & DuBois, 2005). All extracted data were sub-divided into two datasheets: (1) standardised beta coefficients and standard errors that indicated the effect of motivation on academic achievement and (2) standardised beta coefficients and standard errors that indicated the effect of academic achievement on motivation. To allow for analysis by age, two additional data sheets were used in which the mean age per wave was added. Because one effect size can be linked to two ages (i.e. at T_X and T_X+1), the last available effect size was included twice, linking to both the next-to-last and last available mean age, to take into account all age information available. Since many studies have more than one effect size (e.g. because of multiple study waves, or because more than one measure of motivation was used, etc.) we conducted multi-level (specifically, three-level) meta-analyses in order to take into account effect size dependencies. Level 1 (participant level) reflects the fact that authors of primary studies ‘pooled’ the results of individual participants. Level 2 (effect size level) accounts for the fact that multiple effect sizes are nested within individual studies. That the aggregated effects of these individual studies leads to the overall effect size constitutes Level 3 (study level). The analyses were conducted using the metafor (Viechtbauer, 2010) and meta (Balduzzi et al., 2019) packages in R. All data and the analysis scripts (in R language) necessary to reproduce the findings reported in this manuscript are available at the following Open Science Framework project page: https://osf.io/cexgd/?view_only=e5d5e5cc4dc04dd79a188714308ee322

The heterogeneity between and within studies was evaluated through I², which estimates the percentage of the total variance that is due to heterogeneity (Higgins et al., 2003). Heterogeneity between subgroups is evaluated by 𝜏² which quantifies the variance of true effect size (Harrer, 2022). Influential cases were identified by calculating Cook’s distance (Harrer, 2022) and using the conventional cut-off of 4/sample size.

Different motivation constructs

To assess potential differences in effect sizes between studies that used self-belief vs. non-self-belief motivation constructs, a subgroup analysis was conducted. Sixteen studies fell into the self-belief motivation construct category, while six studies were categorised under the non-self-belief motivation construct category (see Table 1). Additionally, six studies incorporated both self-belief motivation constructs and non-self-belief motivation constructs (Guo et al., 2015; Niepel et al., 2014; Pinxten et al., 2014; Seaton et al., 2014; Weidinger et al., 2017, 2019; Zee et al., 2021). Specifically, Niepel et al. (2014) used both ASC and achievement goals; Seaton et al. (2014) employed both math self-concept and achievement goals, and Marsh et al. (2018) utilised both ASC and effort.

Age, time interval, and population

To investigate whether the age of students (indicated in months) or the length of time intervals between study waves (indicated in months) would moderate the pooled effect size, random effects meta-regressions were conducted. Due to the higher prevalence of studies from Germany, we carried out a subgroup analysis to contrast studies conducted within German populations (hereafter termed German studies) against those conducted in populations elsewhere (hereafter termed non-German studies).

Sensitivity analysis

A sensitivity analysis was conducted to investigate whether the quality of different studies impacted the outcome of the main analyses (see Appendix C and D). Individual and overall study quality was assessed using the following four criteria: (1) Bias of the measure used to assess motivation, (2) bias of the measure used to assess achievement, (3) level of attrition, and (4) bias due to the selection of participants or schools. The quality of these four aspects was ranked according to the format of the Generic Cochrane risk of bias tool (e.g. low, some concern, high, no information) (McGuinness & Higgins, 2021). The sensitivity analysis was conducted in the form of a subgroup analysis made up of the four risk of bias categories (low, some concern, high) that each study was assigned to.

Systematic review (qualitative synthesis)

Besides the information collected to conduct the meta-analyses (standardised beta-coefficients and standard errors), the following additional study characteristics were extracted from the 47 studies for the purpose of the qualitative synthesis: Theoretical framework, number of study waves, school subject (e.g. mathematics, reading, etc.), number of participants, attrition rate, gender ratio, participant age, concept/construct of motivation (self-beliefs vs. non-self-beliefs), concept of achievement, measure of motivation, and measure of achievement (see Appendix B) as well as number of months between measurements and participant age for the qualitative synthesis. Publication bias was assessed by a visual inspection of the funnel plot and Egger’s regression test (Egger et al., 1997).

Search results

Out of the 1380 studies screened, 47 studies (51 samples, N = 108,162) met the final inclusion criteria for the systematic review (i.e. the qualitative analyses) and the meta-analyses (i.e. the quantitative analyses). The excluded studies were excluded because their abstract indicated that at least one of the inclusion criteria (mentioned in the section ‘Methods’) was not met. Then we screened an additional 17 studies (18 samples) from Wu et al. (2021) that could meet our inclusion criteria (two were eventually added to our quantitative analysis, one to our qualitative analysis, nine did not eventually satisfy the criteria, and five were also found via our own search strategy, see Appendix G). (We note that our selected studies also found 13 studies that were not found by Wu et al., 2021). A full display of the inclusion process, containing reasons for exclusion of studies after the full texts were read, can be found in the PRISMA flow diagram (Figure 1). Characteristics of the 47 included studies and a short summary of their results can be found in Appendix B. One striking finding is that virtually all studies were performed with so-called WEIRD (Western, Educated, Industrialised, Rich, and Democratic) samples, with almost half (23 samples) coming from one nation (Germany).

Out of the 47 studies, 29 studies with 34 samples (N = 73,018) provided sufficient information for the meta-analyses (i.e. the quantitative analyses). The analyses of the 34 samples yielded 162 effect sizes that could be meta-analysed for prior motivation on subsequent achievement, and 164 for prior achievement on subsequent motivation.

While the 2011 cut-off aligns logically with existing meta-analyses (e.g. Huang, 2011), it is essential to note that preceding 2011, studies exploring non-self-belief constructs might exist and warrant inclusion in the current analyses. To this aim, we extended the search for studies with non-self-belief constructs to the time period between 1999 and 2010. Just three studies were found in this additional search, of which only one was suitable for quantitative analyses (with four effect sizes). Further details about the additional search and analyses including the studies from this search can be found in the Supplemental Materials. We thank our reviewer for the suggestion to expand the search.

Results

Meta-analyses

Effect of motivation on academic achievement

In the three-level meta-analytic model, the pooled standardised beta coefficients of the effect of previous motivation on subsequent academic achievement was β = 0.096 with a 95% confidence interval (CI) between 0.063 and 0.129, SE = 0.017, p < .001 (see the forest plot in Figure 2). Random-effect variances for level three, i.e. the variance between studies is σ²_{Level 3} = 0.007, and for level two, i.e. within studies σ²_{Level 2} = 0.005. The I²_Level3 = 53.61 (or 53.61%) of the total variation can be attributed to the between-studies and I²_Level2 = 42.41 (or 42.41%) to within-studies heterogeneity. We found that the three-level model provided a significantly better fit compared to a model with level three heterogeneity constrained to zero (χ²₁ = 40.07, p < .001), indicating that the three-level structure is supported and the within-studies heterogeneity is non-trivial.

Figure 2. Forest plot showing the single and pooled effect size(s) of studies investigating the effect of prior motivation on subsequent academic achievement. T: time (indicative of the study wave). Each study wave was treated separately, resulting in one effect size per study wave. Each horizontal line represents the confidence intervals of the effect size of each study wave. The dotted vertical line indicates the pooled effect size from the three-level meta-analysis.

We identified three influential cases on the level of individual effect sizes. Removing those cases (see Supplemental Materials for the plot) lowered the between-studies heterogeneity to 42.80% (I² = 42.80) and increased the within-studies heterogeneity to 51.94% (I² = 51.94). After removing the influential cases, the pooled beta-coefficient was β = 0.083 with a 95% CI between 0.056 and 0.111, SE = 0.014, p < .001, highly similar to the pooled beta-coefficient for all effect sizes.

A subgroup analysis within the three-level model indicated that the effect was significantly larger (F_1,145 = 92.533, p < . 001) for studies using self-belief motivation constructs (β = 0.157, 95% CI [0.124, 0.189]) than those using non-self-belief constructs (β = −0.002, 95% CI [–0.037, 0.034]). A random-effect model that allowed separating the heterogeneity per group showed that it was higher in the self-belief subgroup (𝜏² = 0.023) than in the non-self-belief subgroup (𝜏² = 0.002). The results of the test for residual heterogeneity showed that there was significant unexplained variance left between all effect sizes in the dataset (QE(145) = 1100.159, p < .001) even after taking into account the moderating effect of self-belief vs. non-self-belief motivation construct.

A random effects three-level meta-regression conducted to test a possible moderator effect of time interval on the strength of the motivation → achievement relationship failed to find a significant effect (F_1,145 = 3.131, β = −0.003, p = .079, 95% CI [–0.006, 0.000]). Residual heterogeneity remained significant (QE(145) = 1248.91, p < .001). A further meta-regression showed that participants’ mean age (measured at each interval) did not moderate the effect sizes either: (β = 0.003; p = .576, 95% CI [–.006, 0.011]) with significant residual heterogeneity (QE(248) = 2156.906, p < .001). A further subgroup analysis within the three-level model comparing German with non-German studies indicated that the effect sizes in German studies (β = 0.069, 95% CI [0.020; 0.118]) were not significantly different (F_1,145 = 2.124, p = .147) from those in non-German studies (β = 0.118, 95% CI [0.073; 0.163]) with significant residual heterogeneity (QE(134) = 1240.748, p < .001). A random-effect model that allowed separating the heterogeneity per group showed that the heterogeneity was smaller in the German studies (𝜏² = 0.003) than in the non-German studies (𝜏² = 0.015).

A multivariate model taking into account the possible moderating effects of types of motivation constructs, time interval, and country still yielded significant residual heterogeneity (QE(143) = 1072.276, p < .001). Therefore, it can be concluded that the moderators and covariate combined still did not substantially explain the heterogeneity between studies. However, in this multivariate model (i.e. after partialling out the effects of motivation constructs and country), the effect of time interval was significant, (β = −0.003; p = .011, 95% CI [–0.006, −0.001]), suggesting that the effect that prior motivation had on subsequent achievement attenuated over time if the time in between measurements was larger.

The funnel plot for assessing risk of bias (for the motivation → academic achievement effects) is provided in Appendix E. Egger’s test result indicates that the data in the funnel plot are asymmetrical (β₀ = 0.913, t(145) = 2.74, p = .007). While this suggests a publication bias in favour of studies with larger effect sizes, such a conclusion has to be treated with caution due to publication bias being only one out of many reasons for asymmetry and due to the high heterogeneity between studies (Harrer, 2022). Since we contacted authors to ask for their data, we also included some unpublished (non-significant) relationships in the current meta-analyses. This, however, does not guarantee an exhaustive treatment of unpublished effects as we had no access to unpublished studies.

Effect of academic achievement on motivation

In the three-level meta-analytic model, the pooled standardised beta-coefficients for the effect of previous academic achievement on subsequent motivation was β = 0.176 with a 95% CI between 0.124 and 0.227 (see forest plot in Figure 3). Random-effect variances for level three, i.e. the variance between studies is σ²_{Level 3} = 0.020, and for level two, i.e. within studies σ²_{Level 2} = 0.011. I²_Level3 = 64.34 (or 64.34%) of the total variation can be attributed to the between-studies and I²_Level2 = 34.79 (or 34.79%) to within-studies heterogeneity. The three-level model provided a significantly better fit compared to a model with level-three heterogeneity constrained to zero (χ²₁ = 123.559, p < .001), indicating that the three-level structure is supported and the within-studies heterogeneity is non-trivial.

Figure 3. Forest plot showing the single and pooled effect size(s) of studies investigating the effect of prior academic achievement on subsequent motivation. T: time (indicative of the study wave). Each study wave was treated separately, resulting in one effect size per study wave. Each horizontal line represents the confidence intervals of the effect size of each study wave. The dotted vertical line indicates the pooled effect size from the three-level meta-analysis.

After removing two influential cases (see Supplementary Material for the plot), heterogeneity remained practically unchanged (I²_{Level 3} = 64.99 and I²_{Level 2} = 34.11%). The overall pooled beta-coefficient after removing these cases was β = 0.173 (95% CI [0.121, 0.224]), highly similar from that of all effect sizes.

A subgroup analysis that compared self-belief motivation constructs to non-self-belief motivation constructs indicated that the effects were larger (F_1,162 = 31.305, p < . 001) in the subgroup of self-belief construct studies (β = 0.227, 95% CI [0.178, 0.277]) than in those using non-self-belief motivation constructs (β = 0.091, 95% CI [0.036, 0.146]). Residual heterogeneity test pointed to significant unexplained variance left between all effect sizes (QE(162) = 9042.523, p < .001). A random-effect model that allowed separating the heterogeneity per group showed that it was higher in the self-belief subgroup (𝜏² = 0.028) than in the non-self-belief subgroup (𝜏² = 0.007).

A random effects meta-regression analysis was conducted to investigate whether the difference in time intervals between study waves moderated the effect size. The result yielded a non-significant moderating effect of time interval (β = 0.002; p = .378; 95% CI [–0.003, 0.007]) with significant residual heterogeneity (QE(144) = 8673.582, p < .001).

A further meta-regression showed that participants’ mean age did significantly moderate the effect size (β = −0.019, p = .001; 95% CI [–0.030, −0.007]), suggesting smaller effect sizes in older samples (but with significant residual heterogeneity, QE(250) = 12354.501, p < .001). A further subgroup analysis indicated German studies (β = 0.168, 95% CI [0.091; 0.245]) had similar effect sizes (F_1,162 = 0.074, p = .787) to non-German studies (β = 0.182, 95% CI [0.112; 0.253]) and there was significant residual heterogeneity (QE(162) = 9731.591, p < .001).

A multivariate model taking into account the possible moderating effects of types of motivation constructs, time interval, and country still yielded significant residual heterogeneity (QE(142) = 6736.188, p < .001) (and significant effect of type of motivation constructs). Therefore, it can be concluded that only the type of motivation constructs (self-belief vs. non-self-belief) and participant age (but neither the intervals between study waves nor the study population) can account for some heterogeneity between studies.

The funnel plot for assessing risk of bias (for the academic achievement → motivation effects) is provided in Appendix F. Egger’s test result indicates that the data in the funnel plot are asymmetrical (β₀ = 4.559, t(162) = 6.19, p < .001). While this again suggests a publication bias in favour of studies with larger effect sizes, this conclusion has to be treated with caution due to the aforementioned reasons.

Quality assessment

The risk of bias of studies included in these meta-analyses was predominantly low, as can be seen in Appendix C and D, which provides a visualisation of the risk of bias rating of each individual study. Overall, the risk of bias of the measurement of motivation was rated as low for all studies, as all of them used questionnaires. Risk of bias of the measurement of achievement was rated as low in 19 studies and as somewhat concerning in 12 studies that used school grades instead of standardised tests. One academic achievement measure (Grygiel et al., 2017) was rated as ‘high’ risk of bias because it used teachers’ perception of students’ achievement as the achievement measure. Risk of bias due to attrition was rated as low in 18 studies which reported attrition below 30%, four studies were rated with ‘some concern’ as the reported attrition level was above 30% but below 50%, attrition exceeded 50% in three studies which were rated with high risk of bias. The level of attrition remained unknown in seven studies. Risk of bias due to selection of schools or participants was rated as low when the selection of schools or individual students was random, this was the case in seven of the included studies. Thirteen studies were rated with ‘some concern’ because they selected schools from only one specific district or region of the country. Two studies were labelled with a ‘high’ risk of bias because schools were selected according to a criterion such as better performing students. Ten studies did not provide information on how the samples were selected; these studies were labelled with ‘no information’. A sensitivity analysis was conducted, which showed no association between the risk of bias of a study (low, some concern, high) and the study outcome.

Systematic review

A full list of study characteristics of all studies (in total 47 studies plus two studies found in the additional search) included in the systematic review can be found in Appendix B. The 18 studies (N = 36,320) from the original search that, due to missing statistical information, could not be included in the meta-analyses, largely reported results in line with those of the meta-analyses. We will summarise qualitative results of 47 included studies below (and those of the two studies found in the additional search is discussed in the Supplementary Materials).

First, we examined whether the conclusions drawn in the studies match those in our meta-analyses, with particular attention to the 18 studies not included in the meta-analyses. A majority of both the samples included (28 of 34) and excluded (12 of 18) in the meta-analyses reported evidence in favour of a reciprocal relation between academic achievement and motivation. In line with the findings of the meta-analyses, several of those studies reported a stronger relation from achievement to motivation than vice versa (e.g. Hwang et al., 2016; Liu et al., 2018; Möller et al., 2014; Sewasew & Schroeders, 2019). In this regard, there were no obvious differences between the studies that were included or not included in the meta-analyses. This suggests that regarding reciprocity, the studies included in the meta-analyses were representative for all studies that met the inclusion criteria.

Eight studies used more than one motivation constructs and found that one or more motivation constructs had a reciprocal relation with achievement while the other motivation constructs did not. Three studies found that ASC (Niepel et al., 2014; Preckel et al., 2013) and specifically, reading self-concept (Walgermo et al., 2018) were reciprocally related to academic achievement. In the same three studies, neither literacy interest (Walgermo et al., 2018) nor social self-concept (Preckel et al., 2013) nor achievement goal orientation (Niepel et al., 2014), all three of which are non-self-belief motivation constructs, were found to have significant reciprocal relationships with academic achievement. Three studies included achievement goal orientation as motivation constructs and found conflicting results. One study (out of these three) found that performance goals were related to academic achievement, while other achievement goals were not (Gunderson et al., 2018). The two other studies (out of the three) found that only mastery goals had a reciprocal relation to academic achievement, while other achievement goals did not (Preckel & Brunner, 2015; Seaton et al., 2014). Seaton et al. (2014) also included ASC as motivation construct, which was also reciprocally related to achievement. Two studies reported that interest was reciprocally related to academic achievement while self-belief (Jensen & McHale, 2015) and effortful engagement (Steinhoff & Buchmann, 2017) were not. In sum, these eight studies reinforce the notion that self-belief motivation constructs have stronger reciprocal relations with achievement than do non-self-belief motivation constructs.

One study found that reciprocal relations were stronger for grades than for standardised test scores (Marsh et al., 2018). Another study found only one significant effect in relation to standardised test scores but the effect was larger than the two significant effects in relation to grades (Scherrer et al., 2020).

Four studies found that either study samples or participant age impacted whether a reciprocal effect was found or not. Miyamoto et al. (2018) found reciprocal relations for students without migration background but only a unilateral effect of reading competence on intrinsic reading motivation (i.e. from achievement to motivation) for immigrant students. Similarly, Schaffner et al. (2016) found reciprocal relations between academic achievement and motivation for students in an academic track while no significant relations were found for students in a vocational track. Retelsdorf et al. (2014) found reciprocal relations between reading achievement and reading self-concept in the beginning of secondary education. During measurements later in secondary school, however, only the unilateral effect of achievement on motivation was found. Scherrer et al. (2020) found the opposite: unilateral effects of mathematics achievement on motivation (achievement goal orientation) for the beginning of secondary education, and reciprocal relations later in students’ career. These four studies do not suggest a clear pattern in how motivation–achievement interactions vary with age or population, again reinforcing the conclusions of the meta-analyses.

Four studies found reciprocal relations for some school subjects, but not others. While bidirectional relations were found for student’s interest in social, investigative and networking activities and achievement, no relations or only unilateral relations were found for other types of interests (Höft & Bernholt, 2019). Schiefele et al. (2016) found reciprocal effects when primary school students were tested on reading single words and sentences; however, when they were tested on comprehension of text passages, a unilateral effect of achievement on motivation was found. Chen et al. (2013) found differences in reciprocity between achievement and motivation to be moderated by the participants’ age and school subject. For high school students, reciprocal effects were found in both mathematics and Chinese. However, for preadolescent participants, the reciprocal effects were only found in mathematics while unilateral effects (motivation → achievement) were found for Chinese. Weidinger et al. (2019) found a significant reciprocal effect for primary school students when the school subject was German. When the subject was mathematics, no effect was found for students below grade 3, while a unilateral positive effect of achievement on motivation (ASC) was found in students above grade 3. In the meta-analyses, we did not find a clear dependence of relations on the school subject, and these four studies do not point to a clear pattern either.

For one study, the conclusions depend on the statistical model that was used. Reciprocal relations were found for the full-forward cross-lagged panel model (FF-CLPM) but no relations when other models were used (Ehm et al., 2019). FF-CLPMs are those where not only auto-regressive and cross-lagged paths are tested, but the effects across non-adjacent waves and within-time correlations are also included.

Nine studies did not find a reciprocal effect of academic achievement and motivation. Instead, six found a unilateral effect of academic achievement on motivation (Gunderson et al., 2018; Jensen & McHale, 2015; Möller et al., 2014; Nuutila et al., 2018; Schöber et al., 2018; Viljaranta et al., 2014). These studies had little in common: different age ranges, school subject, and motivation constructs that were used (although four used self-efficacy, one of these also included other measures of motivation). Three found a unilateral effect of motivation on academic achievement (Grigg et al., 2018; Hirvonen et al., 2012; Retelsdorf et al., 2014). Here, the measure of motivation was self-efficacy, self-concept and task-avoidant behaviour, which affected performance of primary school students but was not affected by it. Again, no clear commonality between the studies finding just a one-way effect could be found.

Discussion

In the current research, we undertook a systematic review and meta-analyses of longitudinal studies on the reciprocity between motivation and academic achievement among primary and secondary education students. Our study sought to achieve three main objectives: (1) provide the first overview of the longitudinal empirical studies conducted since 2011 (and extended back to 1990 with the additional search), (2) give insights into how the reciprocal relationship between academic achievement and motivation unfolds specifically within the context of primary and secondary education, and (3) investigate whether the reciprocity also generalises to the relationships between academic achievement and motivation constructs beyond ASC and its related self-belief constructs.

Overall, our meta-analytic results were in support of a reciprocal relationship between academic achievement and motivation. Notably, the reciprocity was non-symmetrical: the influence of previous academic achievement wielded twice as strong influence on subsequent motivation compared to the reverse direction. Our qualitative review corroborated these findings. Among the 18 studies lacking sufficient data for the meta-analyses, the majority also found reciprocal relations between motivation and achievement, indicating a stronger pathway from achievement to motivation than vice versa. Consistent evidence of this reciprocity was also present in three previous meta-analyses (Huang, 2011; Valentine & DuBois, 2005; Wu et al., 2021). In line with our results, Valentine and DuBois (2005) and Wu et al. (2021) exhibited closer alignment regarding the effect sizes (achievement on self-belief: β = 0.15 and β = 0.16 and; self-belief on achievement: β = 0.08 and β = 0.08, respectively), compared to Huang’s (2011) study which showed considerably larger effect ranges overall (achievement on self-concept: β ranged between 0.19 and 0.25; self-concept on achievement: β = 0.20–0.27).

This asymmetry holds significant implications for the underlying mechanisms. Motivation has been theorised to affect achievement by altering either the quantity of academic behaviours (e.g. effort, persistence, etc.) or the quality of such behaviours (e.g. learning strategies, meta-cognition, etc.) (Vu et al., 2022). Conversely, how academic achievement can impact motivation also likely follows different routes: either it increases a sense of competence and thereby modifies the values attached to learning activities or it induces a feeling of flow which makes academic behaviours rewarding in their own right (Vu et al., 2022). This asymmetry that we found suggests that it takes more to increase quantity and quality of academic behaviours than to boost motivation by positive perceived achievement, such as instances of success. Interestingly, this finding is also broadly in line with large twin studies which revealed that literacy achievement begets literacy achievement, while motivation (specifically, literacy enjoyment) does not (van Bergen et al., 2018, 2023).

Our finding of the asymmetric reciprocity also has practical significance for how to build effective interventions to help students to succeed academically. Previous findings that self-concept and academic performance had a reciprocal relation led to the advice that interventions should equally prioritise enhancing both self-concept and building skills (as seen in the meta-analysis of self-concept interventions of O'Mara et al., 2006) because enhancing one without the other would lead to short-lived effects. Our findings are consistent with the idea of targeting both sides of the relations; yet they suggest a greater emphasis on achievement. Sole focus on skills has been said to possibly lead to undermined self-concept (O'Mara et al., 2006), yet the emphasis on enhancing self-concept could lead to erroneous estimate of necessary effort to invest in learning tasks (Feldon et al., 2019), eventually resulting in underachievement. A revised recommendation for practitioners would be to first focus on direct assistance to students in mastering learning materials and building academic skills. Subsequently, it becomes crucial to help students integrate these achievements into their ASC instead of providing positive feedback solely to boost self-belief. Previous research has also noted that interventions aiming to improve students’ academic achievement seem to be most effective when administered to students with diagnosed underachieving issues (O'Mara et al., 2006; Sisk et al., 2018).

We discovered that studies employing motivation constructs other than self-concept/self-belief only demonstrated a unilateral relationship between academic achievement and motivation. Here, prior academic achievement boosted subsequent motivation but prior motivation did not necessarily affect subsequent achievement. This is surprising to say the least, since most theories of academic motivation, even those conceptualising motivation differently than self-beliefs, imply reciprocal relations between achievement and various motivation constructs (Vu et al., 2022). Our meta-analyses uncovered that studies utilising self-concept or self-belief as a measure of motivation yielded larger pooled effect sizes than studies utilising other motivation constructs. There are at least three plausible explanations for this stronger effect of self-beliefs than other motivation constructs. First, self-concept or self-belief could be more central in academic motivation than other constructs, and therefore theoretically the most powerful construct to use. Second, these constructs might be more easily and reliably measured through self-reports than non-self-belief motivation constructs. Since unreliable measurement attenuates covariance, this could result in larger betas for self-belief motivation constructs. However, to our knowledge, there is currently little independent evidence supporting both explanations (see some of the discussion related to the second point in Fulmer & Frijters, 2009).

A third explanation is that self-beliefs are not merely motivation constructs. Self-beliefs stem from assessments of one’s own past achievements, essentially serving as self-evaluations. To the extent that these self-beliefs are accurate as self-assessment, they would naturally correlate with achievement. This idea, that self-beliefs correlate with achievements simply because they are accurate assessments of achievement, forms the core of the so-called skill improvement hypothesis (e.g. Möller et al., 2011; Retelsdorf et al., 2014). This hypothesis posits that improved skills lead to achievement, which subsequently leads to an enhanced self-concept. However, the hypothesis does not elucidate why current self-concept predicts future achievement. This requires an additional hypothesis: Self-beliefs may integrate skill improvements that are not directly assessed in achievement tests. Should these skills manifest in later standardised test scores, this could produce a spurious current self-belief → future achievement link (Figure 4). Eventually, this is not an issue that could be solved through statistical analyses as these possibly entangled effects are due to conceptual fuzziness. Note that such explanations of motivation–achievement relations pertain to self-belief motivation constructs, rather than non-self-belief ones. For example, an achievement goal does not involve self-assessment; hence, correlations between goal orientation and achievement cannot be explained by the skill improvement hypothesis.

Figure 4. Possible explanation for why the self-concept and achievement reciprocal relationship was stronger than non-self-concept motivation constructs and achievement reciprocal relationship. Self-concepts are possibly more than motivation constructs as they also incorporate a self-assessment of skills. The skills obtained at a previous time point can exert an influence on self-concept in the time point n. However, the reverse pathway is less straightforward as the effects on the subsequent achievement could come from both skills and self-concept enhancement. For the sake of clarity, the focal time point here is time point n. The figure only includes the arrows representing the effects on this time point while those on other time points are not shown. T: time point; M: motivation; S: skill; A: academic achievement.

Strikingly, our categorisation of motivation constructs (self-belief vs. non-self-belief) and age emerged as the sole potential moderators of the motivation–achievement relations (albeit cautiously, considering the moderate levels of remaining heterogeneity). Conversely, we did not find an effect of study population (German vs. non-German studies). The absence of age effects for the motivation → achievement pathway suggests a relatively consistent influence of motivation on achievement across primary and secondary education – a surprising finding given the substantial developmental changes between the ages 6 and 18. However, concurrently, we found that age mattered for the achievement → motivation pathway, with smaller effect sizes in older samples. This suggests that as students progress through their school years, they integrate less of their previous achievement in their subsequent motivation. This juncture appears pivotal for school interventions. Wu et al. (2021) did conclude in their meta-analyses that age mattered for both pathways, possibly because they focused on a wider age range.

We found a lack of effect of time interval in the achievement → motivation pathway and detected a very small (and only statistically significant in the multivariate model) effect of time interval in the motivation → achievement pathway. These two findings suggest that motivation and achievement either are stable over extended periods or have long-range effects. In the first scenario, motivation could impact achievement within a short timeframe, after which motivation remains relatively constant for the rest of the semester or the year, accumulating an effect on achievement in the following measurement. Alternatively, in the second scenario, heightened motivation today may prompt students to engage in learning behaviours that affect achievement only in a semester or a year. From a practical standpoint, this implies that the findings on motivation and achievement might hold general applicability in both primary and secondary schools.

It is worth noting that the majority of the effects in the studies included in the meta-analyses would be classified as long-range due to data collection occurring at extensive time intervals (e.g. months, semester, and years). However, theoretical accounts propose that the effects of motivation on achievement, and vice versa, can take place in shorter time frames, such as weeks, days, or even minutes (Vu et al., 2022). In addition, academic achievement and setbacks likely occur regularly across days and weeks during schooling. To address the question whether the reciprocal effects were stable across time or long-range in its nature, more data collected at shorter time intervals measuring motivation and achievement are needed. This would enable capturing the possible corresponding short-range dynamics.

Even though we could find variables that potentially affect the size of effects, heterogeneity between studies remained high. This large heterogeneity is theoretically unsatisfactory. It suggests the presence of various unexplored factors influencing the relationship between motivation and achievement within existing empirical studies. There might exist factors, like intrinsic rewards, elucidated in theoretical models of reciprocal relations between motivation and achievement (refer to Figure 1 in Vu et al., 2022), that modify or impact the achievement-motivation pathway or its behavioural mediators. Broadly speaking, external influences on the reciprocity such as quality of instructions, verbal persuasion, social and cultural pressures are seldom considered. In essence, it seems that we have yet to fully comprehend the nuanced interplay between motivation and achievement in this field.

Limitations

Despite its contribution, the present study also has several limitations that should be taken into consideration when interpreting the results. First, in the course of the subgroup analyses on concepts of motivation, we were restricted to comparing self-concept and self-belief against a broad category labelled as 'non-self-belief motivation constructs.' Yet, this category encompasses highly heterogeneous constructs from various motivation theories. For instance, in a comprehensive study, Scherrer and Preckel (2019) identified heterogeneous developmental patterns of various motivation constructs such as achievement goals and intrinsic motivation. This suggests the likelihood of varying reciprocity based on specific non-self-belief motivation constructs. However, none of such constructs occurred often enough to group them by, for example, theory. Hence, to gain a more detailed understanding of the differences between motivation constructs necessitates the inclusion of additional empirical studies with similar or theory-specific conceptualisations of motivation. The current research will hopefully inspire future studies including non-self-concept constructs which will enable meta-analyses where the effects can be split up per construct or theory.

Second, potential limitations within the statistical models most commonly used in most reviewed studies could undermine the conclusions drawn from our meta-analyses (Hamaker et al., 2015). In particular, Núñez-Regueiro et al. (2022) recently argued that the cross-lagged panel model (CLPM), which supports the reciprocity between academic motivation and achievement, assumed no individual differences in students’ growth trajectories in academic motivation and achievement. However, students not only differ in motivation and achievement at the outset, but also in the trend over time. Failure to account for these differences might result in conflated estimates of intra- and interindividual differences of change. In their analysis, Núñez-Regueiro et al. (2022) demonstrated that while CLPMs supported the reciprocal relations, models that accounted for random effects and growth factors did not. Virtually, all studies reviewed by us analysed their data without integrating such controls. The only included study that did incorporate the controls (Ehm et al., 2019) reached a conclusion akin to that of Núñez-Regueiro et al. (2022) (see also Liu et al., 2023; van Bergen et al., 2021). However, Hübner et al. (2023), focusing on causal inferences using weighting methods, suggested that either type of model could be appropriate depending on whether developmental processes or causal effects were the primary interest. The on-going debate highlights that modelling choices and assumptions play a crucial role in the identification of reciprocal relations between motivation and academic achievement. This underscores the need for a series of simulation studies and comprehensive reanalyses of existing studies to investigate which modelling strategy is most appropriate and to draw firm conclusions regarding the robustness of the reciprocity.

Third, given that almost all studies were conducted in WEIRD countries (with a majority in Germany), the generalisability of these meta-analyses is possibly limited. Currently, Marsh and Martin (2011)’s remark that it is premature to conclude that the reciprocal relationship between motivation and achievement is universal still holds true. The differences across student populations may have qualitative ramifications for the reciprocal relationship between motivation and achievement that we do not know now (Vu et al., 2022). For example, previous research suggested that the same motivation construct might exert different influences on achievement in different groups (Chiu & Klassen, 2010). In addition, there might be population-dependent pathways between achievement and motivation as cultural differences between groups likely have strong influences on how students attribute their successes to their self-concepts (Graham, 2020).

Finally, it is crucial to consider the pooled effect estimates in light of the fact that researchers used a diverse array of control variables across the studies included in our meta-analyses. This wide range of control variables in their statistical model means that the effects that we extracted from the studies also differ from one another in an extensive array of possibilities. Consequently, the effects of these control variables cannot be straightforwardly isolated or partialed out. This diversity in control variables may, to some extent, contribute to the large heterogeneity observed between the studies in our meta-analyses.

Conclusions

In conclusion, the present study revealed mixed support for reciprocal relations between academic achievement and motivation among primary and secondary education students. Self-beliefs exhibit reciprocal associations with academic achievement. Notably, the reciprocity is not symmetrical: previous academic achievement had a twofold greater impact on subsequent self-beliefs than vice versa. Although no causality can be assumed from non-experimental studies, our findings suggest that efforts to improve academic achievement in schools might be more effective by focusing on honing academic skills rather than solely bolstering self-beliefs. For other motivational constructs, such as achievement goals, interest, and the likes we only found an unidirectional effect of academic achievement on subsequent motivation. Our findings underscore the ongoing necessity for further research in this area, since the major factors in the motivation–achievement interactions may not yet have been fully captured.

Author contributions

Conceptualisation: Tuong Van Vu and Martijn Meeter. Formal analysis and software: Tuong Van Vu, Aurelia Lilly Scharmer, and Elise van Triest. Funding acquisition and supervision: Martijn Meeter and Nienke van Atteveldt. Methodology and visualisation: Tuong Van Vu, Aurelia Lilly Scharmer, Elise van Triest, and Martijn Meeter. Project administration: Aurelia Lilly Scharmer, Elise van Triest, and Tuong Van Vu. Data curation, investigation, resources, and validation: Aurelia Lilly Scharmer, Elise van Triest, and Tuong Van Vu. Writing – original draft preparation: Aurelia Lilly Scharmer. Writing – review and editing: Tuong Van Vu, Martijn Meeter, Aurelia Lilly Scharmer, Elise van Triest, and Nienke van Atteveldt.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

All data and the analysis scripts (in the R language) necessary to reproduce the findings reported in this manuscript are available at the following Open Science Framework project page: https://osf.io/cexgd/?view_only=e5d5e5cc4dc04dd79a188714308ee322

Additional information

Funding

This work was supported by the Jacobs Foundation Science of Learning Pilot Grant [Project Number 2019 1329 00].

Appendix A.

Search string

AB (‘achievement goal*’ OR ‘expectancy-value’ OR ‘attribution theory’ OR ‘locus of control’ OR ‘self belie*’ OR DE ‘self belie*’ OR ‘self determination’ OR DE ‘self determination’ OR flow OR DE flow OR interest OR DE interest OR ‘self concept’ OR DE ‘self concept’ OR ‘self efficacy’ OR DE ‘self efficacy’ OR motivation* OR DE motivation*) AND AB (Achievement OR DE Achievement OR Ability OR DE Ability OR ‘Educational Attainment’ OR DE ‘Educational Attainment’ OR performance OR DE performance) AND AB (Education* OR DE Education OR academic OR DE academic OR school OR DE school) AND AB (‘frame of reference model’ OR DE ‘frame of reference model’ OR dynamic* OR DE dynamic* OR multiple OR longitudinal OR DE longitudinal OR reciprocal or bidirectional OR DE bidirectional OR ‘cross-lagged panel’) NOT AB (sport* OR DE sport OR music* OR DE music* OR athlete* OR DE athlete* OR PE OR ‘physical education’ OR DE ‘physical education’ OR ‘teachers motivation’ OR DE ‘teachers motivation’) AND AB (student* OR DE student* OR child OR DE child OR pupil OR DE pupil OR adolescent OR DE adolescent)

Appendix B

REM: Reciprocal Effects Model; I/E: Internal/External; EVT: expectancy value theory; SDT: self-determination theory; L2: second language learning; SD: standard deviation; T: time (indicative of the study wave); MC: multiple choice; GPA: grade point average; SES: socio-economic status; IRT: item response theory; (CL-/FF-)SEM: (cross-lagged/full forward) structural equation model; (RI-)CL(P)M: (random intercept) cross-lagged (panel) model; CLPA cross-lagged panel analysis; ANCOVA: analysis of covariance; CFA: confirmatory factor analysis; (L)G(C)M: (latent) growth (curve) modelling; PLS: partial least squares; n/a: information was not available.

Appendix C

Overview of the risk of bias assessment for the studies included in the meta-analysis.

Note. Graph was created using the generic Cochrane risk of bias tool (McGuinness & Higgins, 2021).

Appendix D

Risk of bias assessment for individual studies included in the meta-analysis

Note. Graph was created by the generic Cochrane risk of bias tool (McGuinness & Higgins, 2021).

Appendix E

Funnel plot for assessing risk of bias (for the motivation → academic achievement effects)

Note. The middle vertical line indicates the pooled effect size in studies investigating the effect of prior motivation on subsequent academic achievement. Each dot represents a single effect size. Visual inspection of the funnel plot suggested a small publication bias in favour of studies with larger effect sizes, Egger’s test result indicates that the data in the funnel plot are indeed asymmetrical (β₀ = 0.913, t(145) = 2.74, p = .007).

Appendix F

Funnel plot for assessing risk of bias (for the academic achievement → motivation effects)

Note. The middle vertical line indicates the pooled effect size in studies investigating the effect of prior academic achievement on subsequent motivation. Each dot represents an individual effect size. Visual inspection of the funnel plot suggested a small publication bias in favour of studies with larger effect sizes, Egger’s test result indicates that the data in the funnel plot are indeed asymmetrical (β₀ = 4.559, t(162) = 6.19, p < .001).

Appendix G

List of studies found from Wu et al. (2021) and added to the quantitative analyses

Bakadorova, O., & Raufelder, D. (2020). The relationship of school self-concept, goal orientations and achievement during adolescence. Self and Identity, 19(2), 235–249.

Sticca, F., Goetz, T., Nett, U. E., Hubbard, K., & Haag, L. (2017). Short-and long-term effects of over-reporting of grades on academic self-concept and achievement. Journal of Educational Psychology, 109(6), 842.

List of studies found from Wu et al. (2021) and added to the qualitative analyses

Han, F. (2019). Longitudinal relations between school self-concept and academic achievement. Revista de Psicodidáctica (English Ed.), 24(2), 95–102.

List of studies found from Wu et al. (2021), screened and excluded because they did not meet our criteria

Gniewosz, B., Eccles, J. S., & Noack, P. (2012). Secondary school transition and the use of different sources of information for the construction of the academic self‐concept. Social Development, 21(3), 537–557.

Loughlin-Presnal, J., & Bierman, K. L. (2017). How do parent expectations promote child academic achievement in early elementary school? A test of three mediators. Developmental Psychology, 53(9), 1694.

Mcmillian-Robinson, M. M., Frierson, H. T., & Campbell, F. A. (2011). Do gender differences exist in the academic identification of African-American elementary-school aged children? Journal of Black Psychology, 37(1), 78–98. https://doi.org/10.1177/0095798410366709

McMillian, M. M., Carr, M., Hodnett, G., & Campbell, F. A. (2016). A longitudinal study of academic identification among African American males and females. Journal of Black Psychology, 42(6), 508–529. https://doi.org/10.1177/0095798415603845

Roebers, C. M., Cimeli, P., Röthlisberger, M., & Neuenschwander, R. (2012). Executive functioning, metacognition, and self-perceived competence in elementary school children: An explorative study on their interrelationships and their role for school achievement. Metacognition and Learning, 7(3), 151–173. https://doi.org/10.1007/s11409-012-9089-9

Upadyaya, K., & Eccles, J. (2015). Do teachers’ perceptions of children’s mathematics and reading related ability and effort predict children’s self-concept of ability in mathematics and reading? Educational Psychology, 35(1), 110–127. https://doi.org/10.1080/01443410.2014.915927

Weidinger, A. F., Steinmayr, R., & Spinath, B. (2018). Changes in the relationship between competence beliefs and achievement in mathematics across elementary school years. Child Development, 89(2): e138–e156. https://doi.org/10.1111/cdev.12806

Wouters, S., De Fraine, B., Colpin, H., Van Damme, J., & Verschueren, K. (2012). The effect of track changes on the development of academic self-concept in high school: A dynamic test of the big-fish–little-pond effect. Journal of Educational Psychology, 104(3), 793.

Zhu, M., Urhahne, D., & Rubie-Davies, C. M. (2018). The longitudinal effects of teacher judgement and different teacher treatment on students’ academic outcomes. Educational Psychology, 38(5), 648–668. https://doi.org/10.1080/01443410.2017.1412399

List of studies we included in quantitative analysis while Wu et al. (2021) did not

Grigg, S., Perera, H. N., McIlveen, P., & Svetleff, Z. (2018). Relations among math self efficacy, interest, intentions, and achievement: A social cognitive perspective. Contemporary Educational Psychology, 53, 73–86. https://doi.org/10.1016/j.cedpsych.2018.01.007

Grygiel, P., Modzelewski, M., & Pisarek, J. (2017a). Academic self-concept and achievement in Polish primary schools: Cross-lagged modelling and gender-specific effects. European Journal of Psychology of Education, 32(3), 407–429.

Hebbecker, K., Förster, N., & Souvignier, E. (2019). Reciprocal effects between reading achievement and intrinsic and extrinsic reading motivation. Scientific Studies of Reading, 23(5), 419–436.

Liu, Y., & Hou, S. (2018). Potential reciprocal relationship between motivation and achievement: A longitudinal study. School Psychology International, 39(1), 38–55.

Marsh, H. W., Pekrun, R., Lichtenfeld, S., Guo, J., Arens, A. K., & Murayama, K. (2016b). Breaking the double-edged sword of effort/trying hard: Developmental equilibrium and longitudinal relations among effort, achievement, and academic self-concept. Developmental Psychology, 52(8), 1273–1290.

Moller, J., Retelsdorf, J., Koller, O., & Marsh, H. W. (2011). The reciprocal internal/external frame of reference model: An integration of models of relations between academic achievement and self-concept. American Educational Research Journal, 48(6), 1315–1346.

Niepel, C., Brunner, M., & Preckel, F. (2014). Achievement goals, academic self-concept, and school grades in mathematics: Longitudinal reciprocal relations in above average ability secondary school students. Contemporary Educational Psychology, 39(4), 301–313. https://doi.org/10.1016/j.cedpsych.2014.07.002

Niepel, C., Brunner, M., & Preckel, F. (2014). Achievement goals, academic self-concept, and school grades in mathematics: Longitudinal reciprocal relations in above average ability secondary school students. Contemporary Educational Psychology, 39(4), 301–313. https://doi.org/10.1016/j.cedpsych.2014.07.002

Scherrer, V., Preckel, F., Schmidt, I., & Elliot, A. J. (2020). Development of achievement goals and their relation to academic interest and achievement in adolescence: A review of the literature and two longitudinal studies. Developmental Psychology, 56(4), 795–814. https://doi.org/10.1037/dev0000898

Schöber, C., Schütte, K., Köller, O., McElvany, N., & Gebauer, M. M. (2018). Reciprocal effects between self-efficacy and achievement in mathematics and reading. Learning and Individual Differences, 63, 1–11. https://doi.org/10.1016/j.lindif.2018.01.008

Schiefele, U., Stutz, F., & Schaffner, E. (2016). Longitudinal relations between reading motivation and reading comprehension in the early elementary grades. Learning and Individual Differences, 51, 49–58. https://doi.org/10.1016/j.lindif.2016.08.031

Weidinger, A. F., Steinmayr, R., & Spinath, B. (2017b). Math grades and intrinsic motivation in elementary school: A longitudinal investigation of their association. British Journal of Educational Psychology, 87(2), 187–204. https://search.ebscohost.com/login.aspx?direct=true&db=eric&AN=EJ1140918&site=ehost-live

Weidinger, A. F., Steinmayr, R., & Spinath, B. (2019a). Ability self-concept formation in elementary school: No dimensional comparison effects across time. Developmental Psychology, 55(5), 1005–1018.

List of studies we included in the qualitative synthesis while Wu et al. (2021) did not

List of studies we included in the qualitative synthesis while Wu et al. (2021) did not are Arens et al. (2017), Collie et al. (2015), Hirvonen et al. (2012), Höft and Bernholt (2019), Hwang et al. (2016), Jensen and McHale (2015), Laine et al. (2020), Miyamoto et al. (2018), Nuutila et al. (2018), Paulick et al. (2013), and Schaffner et al. (2016).

Arens, A. K., Marsh, H. W., Pekrun, R., Lichtenfeld, S., Murayama, K., & vom Hofe, R. (2017). Math self-concept, grades, and achievement test scores: Long-term reciprocal effects across five waves and three achievement tracks. Journal of Educational Psychology, 109(5), 621–634. https://doi.org/10.1037/edu0000163

Collie, R. J., Martin, A. J., Malmberg, L.-E., Hall, J., & Ginns, P. (2015). Academic buoyancy, student’s achievement, and the linking role of control: A cross-lagged analysis of high school students. British Journal of Educational Psychology, 85(1), 113–130. https://doi.org/10.1111/bjep.12066

Hirvonen, R., Tolvanen, A., Aunola, K., & Nurmi, J.-E. (2012). The developmental dynamics of task-avoidant behaviour and math performance in kindergarten and elementary school. Learning and Individual Differences, 22(6), 715–723. https://doi.org/10.1016/j.lindif.2012.05.014

Höft, L., & Bernholt, S. (2019). Longitudinal couplings between interest and conceptual understanding in secondary school chemistry: An activity-based perspective. International Journal of Science Education, 41(5), 607–627. https://doi.org/10.1080/09500693.2019.1571650

Hwang, M. H., Choi, H. C., Lee, A., Culver, J. D., & Hutchison, B. (2016). The relationship between self-efficacy and academic achievement: A 5-year panel analysis. Asia-Pacific Education Researcher, 25(1), 89–98.

Jensen, A. C., & McHale, S. M. (2015). What makes siblings different? The development of sibling differences in academic achievement and interests. Journal of Family Psychology, 29(3), 469–478. https://doi.org/10.1037/fam0000090

Laine, E., Veermans, M., Gegenfurtner, A., & Veermans, K. (2020). Individual interest and learning in secondary school STEM education. Frontline Learning Research, 8(2), 90–108. https://doi.org/10.14786/flr.v8i2.461

Miyamoto, A., Pfost, M., & Artelt, C. (2018). Reciprocal relations between intrinsic reading motivation and reading competence: A comparison between native and immigrant students in Germany. Journal of Research in Reading, 41(1), 176–196. https://doi.org/10.1111/1467-9817.12113

Nuutila, K., Tuominen, H., Tapola, A., Vainikainen, M.-P., & Niemivirta, M. (2018). Consistency, longitudinal stability, and predictions of elementary school students’ task interest, success expectancy, and performance in mathematics. Learning and Instruction, 56, 73–83. https://doi.org/10.1016/j.learninstruc.2018.04.003

Paulick, I., Watermann, R., & Nückles, M. (2013). Achievement goals and school achievement: The transition to different school tracks in secondary school. Contemporary Educational Psychology, 38(1), 75–86. https://doi.org/10.1016/j.cedpsych.2012.10.003

Schaffner, E., Philipp, M., & Schiefele, U. (2016). Reciprocal effects between intrinsic reading motivation and reading competence? A cross-lagged panel model for academic track and non-academic track students. Journal of Research in Reading, 39(1), 19–36. https://doi.org/10.1111/1467-9817.12027