Sample and design

This three-year longitudinal and eight-year prospective study analyzed the data of N = 1134 male players (birth cohorts 1993, 1994, 1995) who participated in the German TID program from U12 to U15 at one of the DFB’s competence centers between 2004 and 2009. Every year in the fall, each player participated in motor diagnostics, which resulted in four measurement points (i.e., first assessment in U12, following assessments in U13, U14, and U15) for each athlete (4536 total data points from 1134 athletes). Due to the nationwide assessment at the 366 competence centers, diagnostics were conducted within a time window of about seven to eight weeks. The measurement point within this time window was independent of other parameters of the current research, such as age class or performance level. In order to enhance the objectivity and reliability of the assessments across all sites all staff members conducting the tests were provided with a detailed test manual. Additionally, individual sites received practical support through trained research assistants if needed. Nevertheless, the high number of different staff members needed for the nationwide assessments may have contributed to a decrease in reliability. In order to focus on those players who progressed through the full promotion program at the competence centers during adolescence, only individuals who completed the diagnostics’ individual motor test at each of the four measurement points (i.e., U12-U15) were included in this study. Participants were 11.42 ± 0.28 years old at the time of the first measurement point in U12.

Before entering the TID program, players’ parents provided written informed consent for the recording and scientific use of the data collected in the motor diagnostics. In addition, players provided verbal assent before participating in the motor diagnostics. DFB staff members (i.e., coaches with at least the UEFA B-License) conducted the motor tests and, therefore, measured the predictor variables of the present study. The DFB provided the authors with the data for players from the three birth cohorts. The ethics department of the Faculty of Economics and Social Sciences at the first author’s institution and the scientific board of the DFB approved the implementation of this study.

Based on an examination of the rosters of the five highest German leagues [32, 33], the assessed players were divided into two groups according to their adult performance level (APL) in the 2014/2015 season. At this time, the investigated athletes were between 19 and 22 years old. This resulted in a long-term prognostic period of six, seven, or eight years, respectively (depending on the birth cohort). Players who appeared in at least one match in one of the five highest German soccer divisions (Bundesliga, 2. Bundesliga, 3. Liga, Regionalliga, Oberliga) constitute the first group. Currently, almost 9,800 athletes are under contract by a club that is part of one of these divisions [34]. Thus, with approximately 3.5 million active soccer players in Germany [35], individuals who compete in one of the upper five German soccer divisions belong to the top 1% of all soccer players in Germany. Therefore, athletes in this group were categorized as elite players (N = 145, 12.8%). Non-elite players (N = 989, 87.2%) did not appear in one of the five highest German soccer divisions. Divisions of other countries were not analyzed.

Measures

The motor test battery consisted of five individual tests. The players were assessed in sprint (time for a 20m linear sprint), agility (time in a slalom course without a ball) and dribbling (time in a slalom course with a ball), ball control (time needed to play six passes alternately against two opposing impact walls with at least two ball contacts), and shooting. Times for sprint, agility, and dribbling were measured utilizing light barrier systems, whereas times for ball control were assessed with chronographs. Two attempts were conducted for each of these tests, but only the better one was counted. Players were provided with ample time to recover between the attempts. Whereas all these tests are measured by execution time, the shooting test comprises eight shots at two different target fields within the goal (left and right). The shots are rated by a coach with regard to precision (shot on target field represents a hit) and speed (ranked on a three-stage scale). All five individual tests were negatively coded (i.e., a lower value indicated a better performance). Based on the individual test results, a positively coded motor score (i.e., more points indicate better overall performance) was calculated using the following formula (for a detailed description see [21]): (1) Additionally, each player’s relative age (RA; measured by the day of birth within a year, i.e., the number of days after January 1^st) was registered.

Höner and colleagues [21] provided a detailed description of the individual tests and analyzed the test battery’s psychometric properties for a sample comprising nearly 70,000 competence center players (U12-U15). They found excellent internal consistencies in terms of Cronbach’s Alpha (α = .89) as well as satisfying test-retest reliabilities (r = .74) for the composite motor score. With regard to the individual tests, they discovered extraordinarily good internal consistencies for sprint (α = .95) and agility (α = .91), whereas these values for ball control (α = .68), dribbling (α = .61), and especially shooting (α = .41) were lower. With the exception of shooting (r = .30), test-retest reliabilities ranged between r = .50 for ball control and r = .76 for sprint.

Statistical analysis

Data were analyzed using SPSS (Version 24) and R (Version 3.2.2). Pre-analyses of Random-Intercept-Only models revealed an Intra-Class-Correlation (ICC) of at least 0.10 for the total motor score (ICC_Score = 0.14) and for each single test (ICC_Sprint = 0.27, ICC_Agility = 0.32, ICC_Dribbling = 0.23, ICC_BallControl = 0.10, ICC_Shooting = 0.14). Consequently, in accordance with the recommendations of Kreft and De Leeuw [36] and Teachman and Crowder [37], the longitudinal development of players’ performances in the motor diagnostics was analyzed using multilevel regression analyses.

Two-level regression analyses (Random-Intercept-and-Random-Slope models) were performed for the overall score and each single test. To meet the requirements of the longitudinal data set’s hierarchical structure (i.e., different measurement points nested within different individuals), the repeated measurements were analyzed within (level 1) and between individuals (level 2). In order to describe the developmental changes in each motor performance parameter (MP_ij) for the player j at the measurement point i a Random-Intercept-and-Random-Slope model was built: (2) This model included fixed and random effects to account for both players’ overall motor performance development (regarding mean values; i.e., fixed effects) and the unexplained (inter-individual) variation around these means for the intercept and slopes (i.e., random effects; see [38]). More specifically, γ₀₀ (intercept, average performance of all individuals at T0), γ₁₀ (fixed time slope, mean change in motor performance per year), and γ₂₀ (fixed time² slope, mean change in motor performance per year²) contributed to the fixed part of the regression formula. The time² slope was added to the model to investigate whether non-linear changes occur in motor performance development. The parameters u_0j (random intercept, individual deviation from mean performance γ₀₀ at T0), u_1j (random time slope, individual deviation from mean time slope γ₁₀ per year) and u_2j (random time² slope, individual deviation from mean time² slope γ₂₀ per year²) comprised the random part.

In addition to the time variables (in years after the first assessment in U12, level I predictors), RA (in number of days after January 1^st) was inserted in the model as level 2 covariate to check whether this variable had an influence on the respective motor performance parameter. Similarly, although APL semantically represents an outcome and not a predictor, this variable was included as level 2 covariate in the model to analyze the differences in developmental changes between the two groups (elite vs. non-elite). Interactions between the time variables and these covariates (e.g., Time x APL) were also investigated to test for any potential relationship between developmental change over time and players’ RA or APL. For each regression analysis (i.e., for each motor performance parameter) all given variables (i.e., random and fixed parts for intercept, Time and Time² as well as RA, APL, and interaction effects) were added to the model and systematically excluded based on model fit changes indicated by -2 log likelihood (deviance). Only significant predictors (α = .05) of the respective motor performance variable were included in the final regression model. Additionally, deviances from the respective final and null model were used to compute Maddala’s R² to give insight into the explained variance of the regression models [39, 40].

Finally, to also control for cohort effects that may confound the analysis of motor performances’ differences between selection levels [41], two-level regression analyses were performed for each birth cohort separately. These models revealed the same structure of significant predictors for motor performance development as the models based on the data of all three birth cohorts (1993, 1994, and 1995) combined. Therefore, the birth cohorts were accumulated in order to achieve a sufficient number of elite players within the sample and, thus, provide robust results for the developmental changes of the considered motor variables. Only the results for the accumulated data are displayed in the following section.

Motor performance factors’ prognostic validity

The current study confirmed the prognostic validity of the presented motor test battery for adult success (6–8 years later). A significant difference was found between future elite and non-elite players’ overall motor score assessed within early to middle adolescence as well their respective speed abilities (agility) and technical skills (dribbling, ball control, and shooting). These results support the current state of research concerning the predictive power of motor performance (e.g., [4, 12, 13, 42]) and can likely be explained by the essential impact speed abilities and technical skills have on match-winning situations in soccer [9, 13]. For instance, a better agility or dribbling performance (indicated by a better time in the diagnostics) may result in a more ideal position on the pitch for winning or defending the ball against an opponent. This advantage may exist with regard to players’ current performance within early adolescence, and also manifest itself in their likelihood to be further promoted in middle to late adolescence. Furthermore, due to the sufficient long-term stabilities for the motor score investigated in this study (r_U12/U15 ≈ .51; [43]), a better performance in early adolescence may also be linked to an increased probability of superior performance in adulthood. However, despite the prognostic relevance of motor performance, there was no statistically significant relationship of motor performance development and participants’ future playing level in adulthood. These findings are consistent with the conclusions of [26], suggesting that players who reach an elite level in adulthood, on average, already possess advanced motor abilities upon entry into the TID program. In other words, they were able to maintain their advantage over future non-elite players throughout the stages of promotion.

Motor development

Nevertheless, a substantial improvement from U12 to U15 was detected in participants’ performance across all investigated motor performance factors. This progression was non-linear, as players’ motor performance increased considerably more in the first year of their promotion (i.e., U12 to U13) than in subsequent years (i.e., U13 to U14 or U14 to U15). These findings are mostly in line with previous research (e.g., [42, 44]). Most recently, Fransen and colleagues [45] found that participants’ motor competence developed faster before they reached the age of peak height velocity. Similarly, Huijgen et al. [44] indicated a meaningful improvement of players’ dribbling performance between the ages of 12 and 14, followed by a phase where this development plateaued. It is likely that this non-linear development is due to the rapid biological, maturational changes in individuals during early adolescence. An alteration in body composition (e.g., changing hormones, metabolic rhythms) can be accompanied by shifts in motor dispositions. In fact, athletes’ physical or physiological dispositions are expected to suddenly change when they go through puberty during early adolescence [5, 46].

The effect of RA, in general, was found to be rather low for most of the motor performance factors in the present study. Except for sprinting and the total motor score, this variable did not show a significant influence on players’ performance. Nevertheless, in contrast to Votteler and Höner’s suggestion [47], findings of the present study could not confirm an effect of RA on motor performance development. Therefore, additional focus should be given to the influence of relative age characteristics in future research. Although RA seemed to play a limited role in the motor development of elite and non-elite athletes (which is in line with Höner et al. [4]), it seems likely that some future elite players were already advanced in maturation (independent of their relative age). This advantage, among other implications, may have resulted in possessing better speed abilities and technical skills compared to those individuals who mature later [48]. However, these maturational aspects were not part of the present study and, therefore, other age-related considerations such as biological maturity (e.g., skeletal age or age at peak high velocity; see [49]), which may provide additional insight into inter-individual developmental differences, should be explored in future research.

Further attention should also be given to the individual differences that was found with respect to athletes’ motor performance development over time. The random effects in multilevel modelling indicated significant inter-individual variations between players with regard to their baseline performance at the beginning of the TID program (i.e., random intercept) as well as with regard to their motor performance development over time (i.e., random slope). This finding is particularly meaningful given the preselected, homogenous subsample of athletes who belong to the top 4% of youth soccer players in Germany. Similar inter-individual differences in young soccer players’ motor development were found in longitudinal studies about dribbling [26] and match-running performance [25].

Furthermore, while no interaction between motor performance development and a player’s APL materialized, these results were derived from a group-based perspective (i.e., based on the average player within the investigated sample). The detected variation between players as well as the different measurement points potentially allows for such a relationship to emerge on an individual level. Thus, it seems likely that single case analyses of developmental pathways—in addition to group based considerations—will provide a more in-depth understanding of youth soccer players’ motor development in early adolescence. For instance, over 20% of the investigated elite players in the current study had a lower motor score performance in U12 than the average non-elite player. The consideration of such specific types of players (e.g., athletes who reach an elite adult performance level although they demonstrated a below-average performance at the outset of promotion) might be useful to gain more knowledge about the individual development of talented young soccer players.

Limitations

The current study demonstrated the general prognostic relevance of motor performance factors for adult success. Yet there were no significant interactions between the non-linear motor development and future performance level. Additionally, RA generally showed a rather small impact on motor performance development. Despite the usefulness of these findings, there are various limitations that should be addressed in future research.

First, as previously described, selection effects (i.e., the sample consisted of individuals who belong to the top 4% of youth soccer players of the U12 age class in Germany) could have affected the current sample’s constitution regarding its homogeneity [50]. In addition, only players who performed the motor tests regularly (once a year) during their three-year promotion from U12 to U15 were included in the study. Accordingly, players who transferred to a youth academy within the investigated time span or dropped out of the TID program were not considered. It seems likely that players who reached the youth academy level belonged to the better motor performers (e.g., see [21]), whereas players who dropped out were presumably athletes with lower motor performances. This possibly resulted in a more homogenous subgroup of players with similar motor development within the promotion program at the DFB’s competence centers. This study focused on the development of competence center players who were part of the complete promotion program (U12-U15). Based on this investigation, future research should address issues of homogeneity by, for instance, analyzing the motor development of players who dropped out to gain insight into potential similarities and differences with those individuals who stayed in the TID program.

Second, while the current study longitudinally assessed young athletes’ motor performance, it should be noted that environmental factors might play an important role during an athlete’s development in early adolescence. For instance, coaches have a meaningful influence on physical and psychological development [51]. With respect to the current findings, coaches likely differ in the training programs they provide or the contextual environment they nurture. Thus, such social facets should be included in future research addressing inter-individual variation with respect to motor development.

Third, the developmental periods in which talent factors are assessed can have a meaningful impact on their predictive value [52]. Thus, the different motor dimensions that have been addressed within this study (speed and technique) might also show a specific (e.g., quick, deaccelerated, or abrupt) development in earlier and later age classes which may, in turn, affect their predictive value (e.g., [4, 6]). For instance, as mentioned before, Philippaerts et al. [53] stated that phases of accelerated and deaccelerated development could occur just before and after the age of peak height velocity (see also [45]) and might influence talent factors’ discriminating power. The current results are based on data from a sample of 11 to 14-year-old individuals, and, therefore, similar investigations should be conducted with youth players from other developmental phases in adolescence.

Last, the current study only focused on two dimensions of talent factors (i.e., speed abilities and technical skills) and analyzed their influence on individuals’ future success. Development of these motor performance factors was found to be quite similar for elite and non-elite players. It is possible that the investigated non-elite players spend additional time and effort on improving their technical skills and speed abilities during their promotion in the TID program in order to continue to be competitive with their elite counterparts. However, Huijgen et al. [26] argued that due to this additional focus on improving their motor performance, these players’ may not be able to sufficiently develop other skills that would be necessary to reach the top level in adulthood. Given this concern, it seems necessary to “leave the one-dimensional arena and focus on a multidimensional playing field” (see p. 7 in [18]) by conducting research that investigates a more complex and multidimensional array of variables. For example, recent studies illustrated the predictive power of various potential relevant talent facets, including psychological attributes such as motivational and volitional components (e.g., [54]) or tactical competences such as positioning and deciding (e.g., [55]). By pursuing a multidimensional pathway, it may be possible to investigate whether specific weaknesses in one of the different attributes may be compensated by exceptional strengths in others [18].