The paramount importance of regular physical activity (PA) to enhance children’s health has been extensively documented (Elhakeem et al., 2018; Janssen and Leblanc, 2010; Jose et al., 2011; Kaplan et al., 1996; Leskinen et al., 2009; Wendel-Vos et al., 2004). The well-established effects of physical activity have led to the development of physical activity guidelines for youth, widely adopted across the globe (World Health Organization, 2020). Despite this, and the many active policies supporting an increase in physical activity in various settings, the majority of children and adolescents does not meet recommended physical activity levels (Guthold et al., 2020). Furthermore, as children move through childhood and adolescence towards adulthood, physical activity participation rates tend to further decline (Conger et al., 2022).

Of note, these general epidemiological trends describe what happens to the average child but fail to address the large individual differences in physical activity behaviors. These individual differences have been shown to be remarkably stable throughout the lifespan (Breau et al., 2022; Telama et al., 2005; van der Zee et al., 2019) such that children who start out to be more physically active in childhood tend to remain more active later in life. This ‘tracking’ of physical activity suggests that it would pay off to increase the number of active children to arrive at larger numbers of adolescents and adults meeting the recommended physical activity levels. Not surprisingly therefore, much effort has been spent on identifying modifiable determinants of childhood physical activity. One of the more promising traits investigated is motor competence (Øglund et al., 2015; Øglund et al., 2014). Globally, children (3–10 years) demonstrate “below average” to “average” motor competence levels (Bolger et al., 2021), suggesting that there is room for improvement of this trait by targeted intervention. However, such intervention is only meaningful to increase youth physical activity levels to the extent that motor competence has a causal effect on physical activity.

Motor competence can be defined as the full complement of a person’s motor abilities needed to execute all forms of goal-directed motor acts necessary to manage everyday tasks (Bolger et al., 2021; Henderson and Sugden, 1992). The potential role of motor competence for physical activity received a large boost with the development of the “Stodden model” by Stodden et al. (2008). The Stodden model identifies motor competence as a main determinant of youth and adolescent physical activity, a basic idea foreshadowed by the earlier work of Hands and Larkin (2002). To be physically active as they grow older, children need fundamental motor skills like running, jumping, catching, and throwing. Children that start out with low actual and perceived motor competence may not engage in sufficient physical activity to develop the motor competence and physical fitness needed to engage in the required level of physical activity during middle and late childhood. This will draw them into a negative spiral of disengagement in which the lower levels of physical activity in turn will amplify their motor skill deficits compared to their more active peers. “This will ultimately result in high levels of physical inactivity and will place these individuals at risk for being obese during later childhood, adolescence, and adulthood. “(Stodden et al., 2008, p. 297).

Three characteristics of the Stodden model turn it into a dynamic but complex model that make it difficult to predict the development of stable physical activity habits as well as moments that would be optimal for change by intervention. First, it adds two mediational pathways, acting through physical fitness and through perceived motor competence, to the direct pathways between motor competence and physical activity. Second, it suggests non-recursive, reciprocal effects in the direct and mediated pathways. Third, the model allows changes in the direction of effects in the pathways as a function of age. While being very complete by incorporating age-moderation, bidirectionality, and mediation, the Stodden model at the same time is undercomplete by solely focusing on the possibility that these pathways reflect causal effects.

In its essence, the Stodden model revolves around a set of five associations between motor competence and physical activity, between motor competence and perceived motor competence, between perceived motor competence and physical activity, between motor competence and physical fitness, and between physical fitness and physical activity (MC-PA, MC-PMC, PMC-PA, MC-Fitness, Fitness-PA). These associations can arise through fundamentally different mechanisms governing the development and the ensuing stability of the associations between the traits as well as the changes in these associations over time. Figure 1 depicts potential sources of the association between motor competence and physical activity at each of three different ages, and how these sources can impact on the stability of these associations across developmental time. To maintain intelligibility, the Figure greatly simplifies the continuous nature of development by using discrete ages 2, 7, and 13, rather than a more fine-grained model that uses steps of, e.g., 2 months. It’s aim is merely to provide an illustration of the complexity of interpreting (longitudinal) associations.

As a major innovation to the Stodden model, Figure 1 adds latent determinants that may act as confounders of the associations between motor competence and physical activity, or its putative mediators, perceived motor competence and physical fitness. Two sets of latent determinants have been repeatedly nominated by the field of behavior genetics to play a role in many developmental traits. The first set of determinants consists of the common or shared environment that contains all factors shared by family members living in the same household, including the physical home environment, family warmth and mutual support, parenting style and example setting, neighborhood characteristics, and socioeconomic status (including education level of the parents). The second set consists of the genetic variance shared by family members which may reflect additive trait effects of the two parental alleles in a gene, or non-additive trait effects due to allelic dominance or allelic interaction (epistasis).

Starting at the top of the model shown in Figure 1, we see that motor competence at age 2 (“MC age 2”) is considered to be influenced by latent determinants (“Det MC2”). These may involve genetic variants that influence sensorimotor brain functioning and neuromuscular control, or differences in motor skill challenges related to the family that children grow up in or other environmental factors such as climate or exposure to structured physical education in school or childcare settings. The influence of these genetic and environmental determinants of motor competence can show substantial stability over time (reflected in the purple arrows emanating from the latent “Det MC” factor that influences motor competence at all ages) because the genetic code does not noticeably change after conception, and influences related to parental rearing styles, and neighborhood or household characteristics can also be stable. However, the influence of some of the latent determinants may be confined to specific ages (reflected in the latent “Det MC2 … Det MC13” factors). For example, parental social support effects may be strong at ages 2 and 7 but become more diluted when children enter secondary school. While some genetic variants may be expressed at all ages, other variants may show age-specific (suppression of) gene-expression as part of maturation. At the bottom of the figure, we see a parallel situation for physical activity, again with both age-invariant and age-specific latent factors influencing physical activity behaviors at the three ages depicted.

At each age, an association between motor competence and physical activity may arise entirely through the correlation of the age-invariant and/or age-specific factors, without the need for a direct causal path between the two traits. For example, part of the many genetic variants that influence motor competence may overlap with those influencing physical activity, creating horizontal genetic pleiotropy when they influence these traits through independent routes (Minica et al., 2020; Minica et al., 2018; Solovieff et al., 2013; Verbanck et al., 2018). Likewise, environmental risk factors like household poverty and parental rearing styles may independently restrict motor competence development and reduce opportunities for regular physical activity. If the effects of genetic or environmental factors change in strength from childhood to adolescence, they may also cause a strengthening or weakening of the associations over time. Such a confounder-induced age-related change in the association would not be discriminable from an age-moderation effect on the putative causal pathway between motor competence and physical activity.

The confounding genetic or environmental effects may work directly on the two traits themselves, but also make use of an intermediate trait that itself exerts a causal effect on both traits. A first example would be that the same genetic variants that influence motor competence also influence physical activity through their effects on the dopaminergic brain systems that influence motor control as well as exercise reward pathways. A second example of such confounding would be that an obesogenic family environment would increase body mass index (BMI), with BMI having effects on both motor competence and physical activity. In short, cross-sectional associations between motor competence and physical activity at each age can reflect confounding by correlated determinants, which may be genetic or environmental in nature.

However, the effect of the latent underlying factors does not rule out the additional existence of causal effects of motor competence at an earlier age on current physical activity. These causal effects are reflected in the cross-lagged paths of Figure 1. For example, physical activity at age 7 may, in part, depend on the ability to perform basic motor actions at a sufficient level to engage in active play with parents, siblings, or peers at school from age 2 to 7. Conversely, the lagged causal effect may also work in the other direction. Daily engagement in physical activity from age 2 to 7, i.e., playing regular ball games in preschool, may actively contribute to building up motor competence, i.e., lead to increased kicking/throwing skills, at age 7. Such bidirectional causal mechanisms are suggested by the Stodden model as the main cause of the association between motor competence and physical activity in middle and late childhood.

Apart from the mechanisms inducing cross-sectional associations at each age, Figure 1 also depicts the mechanisms that lead to stability of the association of motor competence and physical activity over developmental time. A first mechanism causing stability of the associations between motor competence and physical activity is the autoregression of each of the traits separately. Substantial evidence shows that, even if absolute levels show large maturational changes, the individual differences in both motor competence and physical activity are stable across time (Barnett et al., 2010; Branta et al., 1984; Farooq et al., 2020; Malina, 1990; McKenzie et al., 2002; Pereira et al., 2022; Schmutz et al., 2020). This ‘tracking’ of motor competence and physical activity may arise from direct causal influences of the trait level at a starting age on the trait level at a later age. For example, once a neuromotor skill has been mastered (running, balancing on a beam) it will not be easily lost, and habit formation may solidify physical activity behaviors once these have been taken up in an initial period (Rebar et al., 2024).

Autoregression can be a first source of stability of the association of motor competence and physical activity over time. Once an association has come into existence, e.g., is bootstrapped at age 2 by correlated underlying determinants, it will be propagated across time sheerly by the stability in each of the two traits. A second mechanism that leads to stability of the association of motor competence and physical activity over time are the causal effects of motor competence on physical activity and the reverse causal effects of physical activity on motor competence. Finally, a third mechanism causing stability of the association is a correlation of the age-invariant genetic or environmental determinants of motor competence and physical activity (“Det PA” and “Det MC” in Figure 1). These will not just induce cross-sectional association but also longitudinal associations between motor competence and physical activity.

The Stodden model was created when the associations between the traits in the five pathways (MC-PA, MC-PMC, PMC-PA, MC-Fitness, Fitness-PA) were mostly observed in cross-sectional studies. Cross-sectional studies cannot discriminate between the mechanisms outlined in Figure 1 and outlined above. Nonetheless, if one of the hypothesized associations in the Stodden model is found to be absent, it would at once tell us that no causal effect is likely to exist. In that sense, cross-sectional studies are vital in first demonstrating the primary possibility of a causal association. Longitudinal studies are a step up from cross-sectional studies in that they establish the presence of cross-time (lagged) associations between the traits and can rule out reversed causation. If the assumed causal trait (e.g., motor competence at an early age) is seen to predict the assumed caused trait (e.g., physical activity in adolescence) in the future but in parallel, the association is not seen to hold in the opposite direction this would falsify reverse causation of motor competence by physical activity.

The strongest design to show true causality in the pathways of the Stodden model is the intervention design, where either motor competence or physical activity are manipulated, and it is tested whether the induced changes in one trait led to changes in the other trait. Well-conducted RCTs remain the highest level of evidence for a true causal effect. However, large individual differences can be seen in the response to intervention and it is not always clear what is driving these differences (Kennedy et al., 2021; Liu et al., 2024; Ma et al., 2021; Prochaska et al., 2008). Again familial factors are a potential source of the heterogeneity in responding to intervention. Attempts to increase motor competence and physical activity may fall on more fertile ground in some children compared to others, simply based on their genetic abilities and/or more supportive family environment. So, to further add to complexity, the underlying determinants of motor competence and physical activity in Figure 1 may partly act through their moderating effects of (parental or school-based) attempts to change these traits.

If the genetic and shared environmental determinants independently influence motor competence and physical activity behavior, the size of the causal effects hypothesized to underlie the observed association between motor competence and physical activity behavior would be incorrectly estimated from the size of the association when this confounding is not taken into account. Since the publication of the Stodden model, a very large amount of systematic reviews with or without meta-analyses have been published on one or more of the Stodden pathways. A reasonable expectation, therefore, is that this large volume of work has duly taken the potential of familial confounding into account. Cursory inspection of some highly cited reviews (Barnett et al., 2022; De Meester et al., 2020; Engel et al., 2018; Figueroa and An, 2017) suggested that this might not be the case, but more systematic inspection of the large volume of systematic reviews is needed. In addition, for familial confounding to be a potential issue, it is required that the traits in the Stodden model show substantial variance caused by shared environmental or genetic factors. This requires a review of studies on these traits in the behavioral genetics literature.

The first aim of this narrative review is to examine whether and how shared environmental or genetic confounding had been considered, and possibly ruled out, in the large body of literature on the five pathways in the Stodden model. To do so, we inspected all systematic reviews and meta-analysis of primary studies published after 2008 and searched for discussions on potential confounding by genetic and shared environmental factors.

As our second aim, we compare the strength of the evidence and effect sizes obtained for the effect of motor competence on physical activity in cross-sectional and longitudinal designs. This path of the Stodden model is important for intervention studies aiming to increase middle and late childhood physical activity. If familial confounding is present in this main Stodden pathway, we expect cross-sectional associations to be stronger than longitudinal associations. In addition, we expect that interventions on motor competence would not increase physical activity to the degree predicted from the cross-sectional effect sizes.

As a third aim, we explicitly test the potential for familial confounding in the five bidirectional pathways of the Stodden model. This requires that the variance in the traits in the Stodden model in childhood and adolescence are caused by genetic and shared environmental factors. This can be tested in a nuclear family design (e.g., parental, spousal and sibling correlations) or in a wider pedigrees (correlations between, e.g., self-aunt/uncle, self-niece/nephew, etc.), but the strongest design focuses on the comparison of MZ and DZ twin correlations (Knopik et al., 2017; Polderman et al., 2015). We, therefore, review the existing twin studies on the contribution of genetic and shared environmental factors to each of the traits in the Stodden model. Furthermore, we review direct tests of familial confounding that estimate the overlap in genetic and shared environmental factors influencing multiple traits, e.g., between motor competence and physical activity.

In short, our research questions are:

To address research question 1, a literature search was performed for systematic reviews and/or meta-analyses of the relationships between motor competence, physical activity, perceived motor competence, and physical fitness. Definitions and assessment strategies for these traits can be found in the Supplementary Sections 1, 2. We searched the Pubmed, Web of Science, and EMBASE databases for reviews published after January 1, 2009 (i.e., after the publication of the Stodden model) and before January 15, 2025 (date of final search). The detailed search strategy is shown in Supplementary Methods Section 3’. Of note, for physical activity traits we only extracted results on total physical activity (TPA), moderate-to-vigorous physical activity (MVPA) or leisure time physical activities (LTPA, including exercise and sports) but discarded light physical activity and sedentary behavior.

The extracted titles and abstracts were initially screened by YZ to identify reports fulfilling the inclusion criteria. Articles were stored in the Endnote citation manager. Full-text reading of selected systematic reviews and meta-analyses was performed independently by YZ and EdG. Discrepancies in article selection were discussed and resolved. References were checked to identify additional systematic reviews and meta-analyses on the traits in the Stodden model.

To address research question 3 on the potential for familial confounding, we retrieved all twin studies on the four traits of the Stodden model. The twin design compares the intra-pair resemblance between two types of sibling relationships; genetically identical twins or monozygotic (MZ) twins, a result of division of a single fertilized egg during an early stage in embryonic development, and non-identical twins or dizygotic (DZ) twins, resulting from two separate fertilized eggs (de Geus, 2023). Consequently, MZ twins are genetically identical and the difference between the twins is due to person-specific environmental factors, i.e., experiences that one of the twins has and the co-twin does not. Dizygotic twins shared on average 50% of their genetic make-up. In contrast to familial aggregation studies, that cannot separate genetic and familial environmental sources of covariance, twin studies can decompose all phenotypic variance of the trait of interest into sources of additive (‘A’) and non-additive (‘D’) genetic influences shared environmental influences (influences shared with other family members, e.g., upbringing; referred to as ‘C’) and person-specific environmental influences (influences unique to the individual; referred to as ‘E’).

Our search detected 106 systematic reviews, of which 67 added a meta-analysis, on a total of 1,344 unique primary studies that examined one or more of the associations implied by the Stodden model. Table 1 lists all systematic reviews and meta-analyses and Supplementary Table 1 reports on the primary studies, their study design, sample size, and in which systematic reviews and meta-analyses they had been included.

There were 44 systematic reviews addressing the bidirectional MC-PA pathway, 4 systematic reviews on the bidirectional MC-PMC pathway, 5 systematic reviews on the bidirectional PMC-PA pathway and 10 systematic reviews on the path from MC to Fitness. By far the largest amounts of systematic reviews (n = 54) were done on the path from physical activity to muscular or cardiorespiratory fitness (PA-Fitness) which is a major area of interest in pediatric exercise science. No reviews included primary studies on the path from fitness to MC or fitness to PA, as manipulation of fitness without also manipulating PA is not feasible in children. Many of the systematic reviews on a specific pathway used overlapping sets of the primary studies, and the more recent reviews generally used the largest number of primary studies.

The patterns in the numbers of the primary studies largely followed that of the systematic reviews (see Supplementary Table 1). There were 426 primary studies reporting on the MC-PA pathway, 102 primary studies on the MC-PMC pathway, 74 primary studies reporting on the PMC-PA pathway, and 100 studies reporting on the MC-fitness pathway. Again, the vast majority of primary studies (n = 937) reported on the PA-fitness pathway. As expected, the PA-fitness and MC-fitness studies exclusively tested the effect of MC/PA on fitness traits (not the reverse path). The total number of children and adolescents that have participate in the primary studies testing the Stodden design is over 1.2 million participants.

The very large amount of primary studies described by the systematic reviews should allow us to uncover whether and how past studies on the Stodden model have taken confounding by genetic and shared environmental factors into account. We rely on the authors of the systematic reviews and meta-analyses to have explicitly reported on this. Table 1 reports on our inspection of the discussion sections of the systematic reviews and meta-analyses on text related to potential familial confounding. Our inspection revealed that almost all reviews have taken the potential of unmeasured confounding into account as part of the quality rating the primary studies as prescribed by various guidelines for systematic reviews and meta-analyses, e.g., STROBE, PRISMA, and GRADE (Balshem et al., 2011; Page et al., 2021; von Elm et al., 2007). The quality of the primary studies was often judged to be low to moderate, and it was rarely rated good (see Supplementary Methods Table 4.1). These low scores might in principle have been caused by the majority of primary studies not taking into account potential confounder effects on the associations obtained in the study. We, therefore, considered studies reporting risk of bias as having mentioned ‘general confounding’ (marked by √ in the column of Table 1).

Despite this guideline-enforced attention to the risk of bias, as expressed in study quality scores, explicit mention and discussion of (sources of) familial confounding was very rare. For example, in the main MC-PA pathway, aspects of the family environment were explicitly mentioned as a potential source of confounding in only 5 out of the 44 systematic reviews on this pathway. Most cited shared environmental confounders were parental socioeconomic status and parental social support which may influence both motor competence and physical activity (Barnett et al., 2016; Øglund et al., 2015; Xin et al., 2020). Strikingly, clear mention or discussion of genetic confounding was absent. Only Barnett and colleagues, in their authoritative review of 2022, wrote: “The broad scope of this review meant that we could not assess how other relevant variables (e.g., diet, genetics, cultural settings, growth and maturation, cognition, motivation) related to the core variables in the model.” This indirect allusion is the only reference to genetics in the context of confounding in 106 systematic reviews and meta-analyses.

Forty-four systematic reviews (n = 22) and meta-analyses (n = 22) specifically investigated the direct or mediated paths between motor competence and physical activity in both directions (see Table 2 for details). Of the reviews, 25% reported on a cross-sectional design, 12% on a longitudinal, 32% on an intervention design and 31% were (cluster) RCTs. Many reviews reported on multiple analyses of the MC-PA association from the same primary studies. Separate results were given for the total score for motor competence, and/or scores for all or one of the specific domains of locomotor, object, and stability skills. Separate results were also given for different types of physical activity, including either TPA or MVPA and cardiorespiratory as well as strength training activities.

Twenty-nine systematic reviews (n = 10) or meta-analyses (n = 19) (Behringer et al., 2011; Carson et al., 2017; Chen et al., 2024; Collins et al., 2019b; Comeras-Chueca et al., 2021; Dudley et al., 2011; Engel et al., 2018; García-Hermoso et al., 2020; Grady et al., 2025; Hassan et al., 2022; Hesketh et al., 2017; Johnstone et al., 2018; Li et al., 2022; Liu et al., 2020; Lorås, 2020; McDonough et al., 2020; Moon et al., 2024; Norris et al., 2016; Oppici et al., 2022; Poitras et al., 2016; Rico-González, 2023; Sinclair and Roscoe, 2023; Sun and Chen, 2024; Timmons et al., 2012; Van Capelle et al., 2017; Veldman et al., 2021; Wang and Zhou, 2024; Zeng et al., 2017; Zhang et al., 2024) directly addressed the causality in the association between motor competence and physical activity by including intervention studies only. Of these, only two reviews focused on the crucial path of the Stodden model where an increase in motor development would lead to an increase in physical activity levels (Engel et al., 2018; Hesketh et al., 2017). Indeterminate evidence at best was found that intervention on motor competence increases physical activity levels, either in early or in middle childhood samples.

All other 27 reviews included primary studies that tested the reverse effects of physical activity interventions on motor competence, the majority using RCTs. Fifteen of these reviews focused on early childhood. Thirteen of these (12 out of 16 analyses) reported strong evidence for a causal effect of physical activity on motor competence, whether expressed in a total score or in separate scores for locomotor and object control skills. Three reviews (4 out of 16 analyses) reported indeterminate evidence (Johnstone et al., 2018; Liu et al., 2020; Oppici et al., 2022). When reported, the average effect sizes of the PA interventions in early childhood varied from small to large, with larger effect sizes seen when PA was mixed with a deliberate motor skills training component (Chen et al., 2024; Wang and Zhou, 2024). Twelve reviews focused on middle-childhood reporting on RCTs using physical activity intervention to improve motor competence in. Strong evidence (13 out of the 17 analyses) was found that increasing physical activity improved a trait in the MC domain, in all but one review. No, or indeterminate, evidence was found in only 4 of the analyses (see Table 2).

The systematic search for twin studies on the four Stodden traits uncovered only five studies for motor competence (Goetghebuer et al., 2003; Peter et al., 1999; Smith et al., 2017; Zi et al., 2024; Zi et al., 2023b) and none for perceived motor competence. In contrast, very many twin studies were found that reported on cardiorespiratory or muscular fitness and physical activity traits (see Table 3).

To examine whether the potential for familial confounding was real, we tested the core requirements for such confounding in the motor development - physical activity pathway: (1) the individual differences in motor competence are heritable and/or caused by shared environmental factors, (2) the individual differences in physical activity are heritable and/or caused by shared environmental factors, and (3) there is significant overlap in either the genetic or the shared environmental factors influencing both motor competence and physical activity. We used twin studies on the traits in the Stodden model as our main vehicle.

In our past work, we reviewed family and twin studies on physical activity and fitness traits in childhood and adolescence (de Geus, 2023) but Table 3 provides an update incorporating a number of recent large twin studies on motor competence (Zi et al., 2023a; Zi et al., 2024; Zi et al., 2023b). Across different types of physical activity, across different countries, and across assessment methods (device-based or self-report), we show heritability of physical activity to be around 29% for girls and 37% for boys. Shared environment contributes 49 and 33% to the variance in motor competence in girls and boys, respectively. These findings signal that the first two of the requirements for potential familial confounding are met. The remaining condition for confounding is a substantial overlap in either the genetic or the shared environmental factors influencing both motor competence and physical activity. This question has only been directly addressed in a single paper so far (Zi et al., 2024). Using longitudinal data across a 12-year time span in a large population-based sample of MZ and DZ twins, they investigated the prediction of future exercise behavior by early motor development. Early motor development explained 4.3% of the variance in future exercise behavior in boys but only 1.9% in girls. In boys, there was evidence for a significant overlap in the genetic factors influencing early motor development and future exercise behavior, while the regressions between the shared and unique environmental factors were not significant. In girls, neither genetic nor shared and unique environmental regressions were significant, possibly reflecting low power to detect such effects at this very low amount (1.9%) of explained variance.

Three systematic reviews on the Stodden model explicitly examined the evidence for mediation of the association between motor competence and physical activity by perceived motor competence or physical fitness (Barnett et al., 2022; Jones et al., 2020; Robinson et al., 2015). Most comprehensive testing was done by Barnett et al. (2022). They conclude that there is indeterminate evidence for mediation by perceived motor competence across the age range 3 to 18 years. In contrast, mediation by physical fitness of the path between physical activity and (future) motor competence, and the reverse path were supported by strong evidence. However, just as is true for the direct associations, associations mediated by fitness could be due to genetic and shared environmental factors independently acting on the traits in the mediating paths, i.e., genetic or environmental confounding of the MC-fitness and fitness-PA pathways.

Past systematic reviews and meta-analyses had shown that, throughout childhood and adolescence, individual differences in cardiorespiratory and muscular fitness are dominated by genetic factors with only a minimal role for shared environmental factors (Miyamoto-Mikami et al., 2018; Schutte et al., 2016b; Zi et al., 2023a). The update on this literature shown in Table 3 confirms this with, on average, 65–67% of the variance in physical fitness traits in both girls and boys being explained by genetic factors, with negligible shared environmental effects. Given the heritability of motor competence and physical activity, this additional heritability of fitness compromises testing of the mediation of the association between motor competence and physical activity by physical fitness. By correcting the direct path for fitness, we correct for the genetic factors influencing fitness. If these genetic factors partly overlap with either those of motor competence or those of physical activity, this correction would attenuate the association between motor competence and physical activity, mimicking ‘mediation by fitness.’

Direct support for a genetic overlap between fitness traits and physical activity comes from bivariate modeling in adolescent and young adult twin studies that assessed cardiorespiratory and muscular fitness phenotypes and daily regular exercise levels, both cross-sectionally and longitudinally (Schutte et al., 2019). This confirmed that physical fitness and physical activity behaviors are genetically overlapping, with genetic correlations between endurance capacity (VO₂max) and regular exercise and sports activities in leisure time as high as 0.43. In short, the substantial contributions of genetic factors to motor competence, physical fitness, and physical activity provide a clear potential for genetic confounding in the ‘mediating’ paths using physical fitness.

We conclude that to more completely test the pathways in the model by Stodden et al. (2008), studies are needed with designs that can detect and account for potential genetic and shared environmental confounding. In the past we were dependent on the use of behavioral genetics studies in (extended) twin families to achieve this. However, despite the elegance of the twin design, not all researchers will have readily access to twin family data. Fortunately, nowadays we can also rely on DNA sampling and genome-wide genotyping in the participants of any cohort, the added costs of which have gone down to ~50 Euro per participant. From these genome-wide genotype data we can create genetic instruments for Mendelian Randomization and extract polygenetic indexes (PGI’s) using the publicly available current (and upcoming) summary statistics of genome-wide association studies (GWAS) on motor competence, physical fitness, and physical activity.

These two methods can be used to test causal hypotheses while controlling for confounding. Mendelian Randomization (MR) is the experiment of nature that comes most close to an RCT in which participants are allocated to different exposure levels independently of confounding. Like an RCT, MR also rules out reverse causality, but uses nature’s randomization to lifetime exposure to DNA variants that influence the exposure (Davey-Smith and Hemani, 2014; Pingault et al., 2018; Speed et al., 2019). These are determined at birth and an advantage of MR over RCTs is that it captures effects of prolonged exposure to, e.g., a propensity to develop motor skills, rather than the mere weeks or months of motor skill training within an RCT. If a genetic instrument is also available for the outcome, e.g., a set of single nucleotide polymorphisms (SNPs) significantly associated with physical activity, bidirectional MR can explicitly test a possible reciprocal causal relationship between exposure (motor competence) and outcome (physical activity).

When our basic concern is to adjust for genetic confounding of an association, e.g., between motor competence and physical activity, the computation of a PGI is another useful strategy (Choi et al., 2020). If a DNA sample with genome-wide SNP genotyping is available for the study participants, the PGI for motor competence and physical activity can be computed using the summary statistics of GWAS on these traits. The PGI is the sum of all the (tiny) genetic effects on the trait across all hundreds or thousands of SNPs that the GWAS detected as meaningfully contributing to the trait (Kujala et al., 2020). If there is genetic confounding, such that the genetic risk for low motor competence is causing (future) low levels of physical activity, then stratifying observed cross-sectional or longitudinal associations for deciles of the PGI for motor competence should lead to attenuation/disappearance of the motor competence—physical activity relationship within each decile of the PGI. If the relationship does not attenuate in the genetically stratified analysis, the association is shown to not depend on genetic confounding. This means that even individuals with a high genetic vulnerability for low motor competence would see their physical activity levels increased by interventions on their motor competence. Indeed, computing the polygenetic propensity for the traits of the Stodden model can also be used to strongly enrich intervention trials. As explained above, the heterogeneity in responsivity to motor competence or physical activity interventions may be partly due to genetic factors. Intervention trials in youth targeting motor competence or physical activity could start to account for this genetic moderation of intervention effects, simply by adding sampling of DNA in the trial participants to the research protocol and using their PGIs as moderator variables.

To apply the MR and PGI approaches to evaluate the Stodden model, we need publicly available summary statistics of large international consortia performing meta-analyses of genome-wide association studies on hundreds of thousands of children. Such summary statistics based on child samples are currently only available early motor coordination (Mountford et al., 2021) but the sample sizes of this GWAS was relatively modest. Larger GWAS efforts on motor milestones in much large samples are currently underway (Gui et al., 2024). However, for physical activity (Doherty et al., 2018; Klimentidis et al., 2018; Wang et al., 2022) and muscular (Willems et al., 2017) and cardiorespiratory fitness (Klevjer et al., 2022), we only have GWAS results based on adult samples, and the authors have not seen clear initiatives for GWAS on these traits in youth. This is where future work is direly needed because we cannot simply assume that the same genetic variants operate throughout the lifespan.

In summary, our synthesis of reviews leads us to fully support the viewpoint expressed earlier by Barnett et al. (2022) and Burton et al. (2023): To truly test the model authored by Stodden et al. (2008) the field is in need of robust longitudinal studies across early childhood and into adolescence. We agree with their recommendations that such longitudinal assessment should aim to include multiple traits from the model, use a combined motor competence assessment (i.e., process and product), and account for biological maturation. However, we offer an update to these recommendations by a strong plea for doing such studies in genetically informative samples that can quantify and account for potential confounding in all pathways of the Stodden model.