Longitudinal evidence for the principle of allocation in Japanese university decathletes Authors: Yuki Ashino, Noriyuki Kida Published: 2026 ABSTRACT An ideal decathlete is often expected to perform well across all disciplines, reflecting versatility and balanced capabilities. At the same time, the principle of allocation suggests that excellence in some abilities may come at the expense of others. However, previous studies have rarely examined how such patterns appear over time. We longitudinally examined 179 Japanese university decathletes who competed in local and national intercollegiate competitions between 2000 and 2024 and had records for at least three academic years. Using group-based trajectory modeling of total points, we identified four classes (High, Upper-middle, Lower-middle, and Low). Academic-year-specific factor-score profiles indicated that, in the High class, running speed and body projection, rotational throwing power, translational throwing power, and vertical leaping power tended to remain higher or increase across academic years, whereas running endurance showed little synchronous similarity with total-point development. Similar tendencies were observed in the Upper-middle and Lower-middle classes, in which total points also increased. Sensitivity analyses based on a reduced-missingness observed-only subset yielded qualitatively similar patterns. Taken together, these observational findings are broadly consistent with the principle of allocation in decathletes, while also suggesting that endurance-related performance may be less closely aligned with total-point development than several speed-, power-, and technique-related characteristics.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1038/s41598-026-56830-w. Introduction The decathlon is a combined athletics competition in which athletes aim to obtain the highest total score across 10 disciplines contested over two days. It comprises 100 m, long jump (LJ), shot put (SP), high jump (HJ), 400 m, 110 m hurdles (110mH), discus throw (DT), pole vault (PV), javelin throw (JT), and 1500 m. Research on the decathlon has classified these 10 disciplines into structures of competitive characteristics from physiological and kinematic perspectives. For example, the three categories for the 30 athletes who participated in the 2004 Athens Olympics were “speed and athletic,” “strength and technique,” and “endurance”1. For 3,103 results from 1998–2009, four categories were identified: “sprint abilities,” “jumping abilities,” “throwing abilities,” “mental strength”,2 and for Japanese university athletes, the five characteristics of “running speed and body projection,” “running endurance,” “rotational throwing power,” “translational throwing power,” and “vertical leaping power”3 have been reported as having competitive characteristics (Table 1).Table 1Examples of the competitive characteristics structures of decathletes reported.FactorSchomaker and Christian1Wimmer et. al.2Ashino et al.31Speed and athletic(100 m, LJ, 400 m, 110mH)Sprint abilities(100 m, LJ, 400 m)Running speed andbody projection(100 m, LJ, 400 m, 110mH)2Strength and technique(SP, HJ, DT)Jumping abilities(LJ, HJ, PV)Running endurance(1500 m)3Endurance(1500 m)Throwing abilities(SP, DT, JT)Rotational throwing power(DT)4–Mental strength(400 m, 1500 m)Translational throwing power(SP, JT)5––Vertical leaping power(HJ, PV)LJ, Long Jump; SP, Shot Put; HJ, High Jump; 110mH, 110 m hurdles; DT, Discus Throw; PV, Pole Vault; JT, Javelin Throw. Table 1Examples of the competitive characteristics structures of decathletes reported.FactorSchomaker and Christian1Wimmer et. al.2Ashino et al.31Speed and athletic(100 m, LJ, 400 m, 110mH)Sprint abilities(100 m, LJ, 400 m)Running speed andbody projection(100 m, LJ, 400 m, 110mH)2Strength and technique(SP, HJ, DT)Jumping abilities(LJ, HJ, PV)Running endurance(1500 m)3Endurance(1500 m)Throwing abilities(SP, DT, JT)Rotational throwing power(DT)4–Mental strength(400 m, 1500 m)Translational throwing power(SP, JT)5––Vertical leaping power(HJ, PV)LJ, Long Jump; SP, Shot Put; HJ, High Jump; 110mH, 110 m hurdles; DT, Discus Throw; PV, Pole Vault; JT, Javelin Throw. Examples of the competitive characteristics structures of decathletes reported. Speed and athletic (100 m, LJ, 400 m, 110mH) Sprint abilities (100 m, LJ, 400 m) Running speed and body projection (100 m, LJ, 400 m, 110mH) Strength and technique (SP, HJ, DT) Jumping abilities (LJ, HJ, PV) Running endurance (1500 m) Endurance (1500 m) Throwing abilities (SP, DT, JT) Rotational throwing power (DT) Mental strength (400 m, 1500 m) Translational throwing power (SP, JT) Vertical leaping power (HJ, PV) LJ, Long Jump; SP, Shot Put; HJ, High Jump; 110mH, 110 m hurdles; DT, Discus Throw; PV, Pole Vault; JT, Javelin Throw. Definitions of competitive characteristics (Ashino et al.,3):Running speed and body projection involves acceleration and take-off ability in disciplines such as 100 m, 110mH and LJ.Running endurance refers to aerobic capacity, primarily assessed by 1500 m performance.Rotational throwing power is the ability to throw using rotational force in DT.Translational throwing power represents linear propulsion-based throwing ability such as SP and JT.Vertical leaping power refers to vertical jumping performance seen in HJ and PV. Running speed and body projection involves acceleration and take-off ability in disciplines such as 100 m, 110mH and LJ. Running speed and body projection involves acceleration and take-off ability in disciplines such as 100 m, 110mH and LJ. Running endurance refers to aerobic capacity, primarily assessed by 1500 m performance. Running endurance refers to aerobic capacity, primarily assessed by 1500 m performance. Rotational throwing power is the ability to throw using rotational force in DT. Rotational throwing power is the ability to throw using rotational force in DT. Translational throwing power represents linear propulsion-based throwing ability such as SP and JT. Translational throwing power represents linear propulsion-based throwing ability such as SP and JT. Vertical leaping power refers to vertical jumping performance seen in HJ and PV. Vertical leaping power refers to vertical jumping performance seen in HJ and PV. Specialization is generally considered suboptimal in the decathlon, where the ideal is to maintain a competitive performance across all 10 disciplines. Kenny et al.4, for example, argued that consistent performance across disciplines is important, whereas Ramadani et al.5 emphasized that weak disciplines can substantially affect the final result. At the same time, however, recent work has suggested that improvement in one domain may occur at the expense of another6. In this study, we distinguish three related but non-identical concepts. Specialization refers to a relative emphasis or superiority in a subset of disciplines or underlying abilities rather than a balanced development across all disciplines. A trade-off refers to an empirical pattern in which an improvement in one characteristic is accompanied by stagnation or deterioration in another. In contrast, the principle of allocation is used here as a broader explanatory framework, suggesting that such trade-offs may arise because finite adaptive resources and competing neuromuscular or energetic demands can constrain the simultaneous development of all characteristics. Under this framework, specialization may be viewed as a possible outcome of uneven development across characteristics, whereas a trade-off is an observable pattern consistent with allocation. This perspective is particularly relevant in late adolescence (approximately 16–20 years old), corresponding to the university years, because this developmental stage is characterized by substantial year-to-year changes in muscle strength and power associated with biological maturation7. Van Damme et al.8 reported signs of the principle of allocation in top-class decathletes, but their analysis was based on cross-sectional data. In contrast, decathletes train continuously while adjusting to their strengths and weaknesses, making year-to-year changes in competitive characteristics especially relevant. Therefore, this study aimed to examine whether the longitudinal patterns of performance development in decathletes are broadly consistent with the principle of allocation. Additionally, world-championship and Olympic athletes represent a highly selected upper tier within the population of decathletes. Because such sample characteristics may affect competitive-characteristic structures, we focused on Japanese university decathletes who are not necessarily elite but have competed actively during their university years. Although detailed case studies on the training structure of Japanese university decathletes are limited, institutional and practitioner-oriented sources provide useful contextual background. The Japan Association of Athletics Federations (JAAF) guidelines describe annual track-and-field training as progressing through general preparation, specialized preparation, competition, and transition phases, with training volume and intensity adjusted across phases9. In addition, a report by Keisuke Ushiro, the first Japanese decathlete to exceed 8,000 points, emphasized general strength development, basic physical conditioning, and event-specific work in pole vault and sprint-related disciplines as important contributors to performance improvement10. These sources do not provide athlete-level evidence for the present sample, but they offer useful contextual background suggesting that strength-, power-, and technique-related development may be emphasized in Japanese decathlon practice. Explicit discussion of endurance-oriented training was comparatively limited in these sources, making the longitudinal behavior of endurance-related performance in university decathletes a relevant empirical question. In this study, we used a two-stage approach to examine whether longitudinal patterns of performance development in Japanese university decathletes are broadly consistent with the principle of allocation. First, we identified classes of total-point development across academic years. Second, we descriptively compared these total-point trajectories with academic-year-specific profiles of competitive characteristics. We expected that increases in total points would be more closely aligned with speed-, power-, and technique-related characteristics than with running endurance. Materials and methods Subjects and setting In this study, we compiled the results of local and national inter-collegiate athletics competitions in which Japanese university decathletes participated from publicly available online records. The study included 179 athletes who had competition records in at least three academic years. Because the public sources used for data collection provided academic year information but not exact age, participants would generally correspond to the typical Japanese university age range of approximately 18–22 years. As the dataset was derived from publicly available competition records, detailed personal information such as height, body mass, and training history was not available. Although year-of-birth information may be retrievable for some high-profile athletes through external databases such as World Athletics, the present sample consisted primarily of Japanese university decathletes competing in domestic intercollegiate competitions, for whom such information was not consistently available in public sources. Therefore, academic year, rather than exact age, was used as the common indicator of developmental stage across the sample. Variables and missing-data handling We recorded 23 variables, including athlete name, academic year, raw discipline results, discipline scores, and total points. After data verification, names were replaced with anonymized athlete IDs. Missingness was assessed at the cell level (academic year × discipline), and the full breakdown is provided in Table S1; percentages refer to the proportion of missing values relative to the number of possible observations in each cell. For athletes with missing records in any academic year, multiple imputation was conducted separately by academic year using Bayesian linear regression. Imputation targeted the 10 raw discipline results (100 m, LJ, SP, HJ, 400 m, 110mH, DT, PV, JT, and 1500 m). Missing data were defined as the absence of a participation record; zero values were not treated as missing. At the athlete level, missing data requiring imputation were present in 140 of 179 participants (78.2%). By academic year, missing records were observed for 105 athletes (58.7%) in the first academic year, 13 (7.3%) in the second, 10 (5.6%) in the third, and 12 (6.7%) in the fourth. These missing values likely reflect differences in record availability across academic years. To preserve longitudinal information, multiple imputation by chained equations was implemented using the mice package11 with Bayesian linear regression (method = “norm”), m = 20 imputations, a maximum of 50 iterations per imputation, and seed = 12345 for reproducibility. The predictor matrix was manually specified with reference to prior research3 so that variables with correlations below 0.29 were not used as predictors for one another, thereby restricting the imputation model to sufficiently related predictors. Confirmatory factor analysis and derivation of year-specific factor scores To derive factor scores summarizing competitive characteristics within each academic year, confirmatory factor analysis (CFA) was performed separately for each academic year using the 10 raw discipline results as observed variables. The initial factor structure was based on Ashino et al.3, which reported the factor structure of Japanese collegiate decathletes. However, our preliminary analyses indicated that imposing the same factor structure across all academic years resulted in inadequate fit and did not sufficiently capture academic-year-specific differences in performance structure. Therefore, strict longitudinal measurement invariance was not imposed in the present study. Accordingly, academic-year-specific CFA models were estimated for each academic year. When model fit was insufficient but theoretical interpretability could be maintained, the observed variable with the lowest standardized factor loading was sequentially removed and the model was re-estimated. If the initial 10-variable model (see Table 1 and Ashino et al.3) failed to meet the fit criteria, CFA models with 9 variables and, if necessary, 8 variables were subsequently examined. Model adequacy was evaluated using the chi-square test, comparative fit index (CFI), Tucker–Lewis index (TLI), root mean square error of approximation (RMSEA), and standardized root mean square residual (SRMR). For the chi-square test of exact fit, the significance level was set at α = 0.05, although model selection was based primarily on overall fit indices and theoretical coherence rather than on the chi-square test alone. Following Hu and Bentler12, good fit was defined as TLI or CFI ≥ 0.96 and SRMR ≤ 0.09, or RMSEA ≤ 0.06. When no model met these criteria, the academic-year-specific model with the most acceptable relative fit and theoretical coherence was retained for descriptive use, with CFI or TLI ≥ 0.90 considered minimally acceptable. Because the final indicator composition differed slightly across academic years, the resulting factor scores were treated as academic-year-specific summaries of competitive characteristics rather than as strictly invariant latent variables on a common scale across years. Accordingly, across-year patterns in these factor scores were interpreted descriptively and were not treated as exact longitudinal changes in identical constructs. The factor labels were assigned with reference to previous research3 as follows. “Running speed and body projection” was intended to represent acceleration ability and body-projection ability during take-off in disciplines such as the 100 m, LJ, 400 m, and 110 mH. “Running endurance” was intended to represent sustained running ability reflected primarily in the 1500 m. “Rotational throwing power” referred to throwing performance involving rotational body movement, reflected mainly in the DT. “Translational throwing power” referred to throwing performance based on a linear approach and upper-limb propulsion, reflected in the SP and JT. “Vertical leaping power” referred to vertical jumping performance in the HJ and PV, based on lower-limb extensor function and event-specific technique. Group-based trajectory modeling of total points To classify longitudinal patterns of overall competitive performance, group-based trajectory modeling (GBTM) was conducted using total points across the four academic years as the longitudinal outcome. This approach probabilistically clusters individual change patterns and assigns each athlete to the class with the highest posterior membership probability. Model selection followed the 8-step framework of Lennon et al.13. First, models with 1 to 7 classes were compared using the Bayesian information criterion (BIC) as the primary criterion. Next, within the selected class solution, polynomial degrees for the trajectory shape were compared. When improvements in BIC were small or resulted in excessive fragmentation or very small classes, average posterior probability of assignment (APPA ≥ 0.70), odds of correct classification (OCC > 5.00), relative entropy (≥ 0.50), and practical interpretability were also considered. Descriptive comparison between total-point trajectories and factor-score profiles After classifying athletes according to total-point trajectories, academic-year-specific factor scores were descriptively summarized within each total-point trajectory class. For each class, class-level mean factor scores and class-level mean total points across the four academic years were used to characterize longitudinal profiles. To evaluate the similarity between class-level total-point trajectories and class-level factor-score profiles across academic years, lag-0 cross-correlation coefficients (CCs) were calculated using the four-point series from the first to the fourth academic year. CC was calculated as follows:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$CC (0)=\frac{{\sum}_{t=1}^{T}\left({X}_{t}-\overline{X }\right)\left({Y}_{t}-\overline{Y }\right)}{\sqrt{{\sum}_{t=1}^{T}{\left({X}_{t}-\overline{X }\right)}^{2}}\hspace{0.17em}\sqrt{{\sum}_{t=1}^{T}{\left({Y}_{t}-\overline{Y }\right)}^{2}}},T=4$$\end{document}where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${X}_{t}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${Y}_{t}$$\end{document} represent the class-level mean total points and class-level mean factor scores in academic year t, T = 4, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overline{X }$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overline{Y }$$\end{document} are the mean values of each series. CC values closer to + 1 indicate greater synchronous similarity in the observed patterns, whereas values closer to 0 indicate greater dissimilarity. Because only four academic years were available and the factor scores were derived from academic-year-specific CFA models, these CCs were used solely as descriptive indicators of synchronous pattern similarity. No inferential testing was performed, and the coefficients were not interpreted as evidence of causality or strict longitudinal equivalence. Sensitivity analyses Sensitivity analyses were conducted to assess the robustness of the main descriptive findings to the handling of missing data. Specifically, key analyses were repeated in a reduced-missingness observed-only dataset restricted to athletes with complete records for total points and raw discipline results in the second to fourth academic years. No imputation was applied in this subset. Within this reduced-missingness dataset, group-based trajectory modeling was repeated using total points from the second to fourth academic years only. Because only three time points were available, models with 1 to 4 classes and linear trajectory shapes were primarily considered, with model selection based on the Bayesian information criterion (BIC), average posterior probability of assignment (APPA), odds of correct classification (OCC), relative entropy, and practical interpretability. To evaluate whether the main descriptive conclusions were preserved without imputation, class-level mean trajectories of representative observed disciplines were additionally examined within each trajectory class. The representative disciplines were 100 m, 1500 m, DT, JT, and PV, corresponding broadly to running speed and body projection, running endurance, rotational throwing power, translational throwing power, and vertical leaping power, respectively. These sensitivity analyses were intended to assess qualitative consistency in the direction of the main findings rather than to reproduce all model estimates exactly. Software All analyses were performed using R version 4.1.2 (2021–11-01)14 and RStudio version 2023.3.0.38615. Multiple imputation was conducted using the mice package11, confirmatory factor analysis using lavaan16, and group-based trajectory modeling using gbmt17. For GBTM, the number of classes and polynomial degree were varied stepwise during model selection, whereas other settings were left at package defaults unless otherwise specified. To avoid variable standardization in GBTM, scaling = 0 was explicitly specified. Cross-correlation coefficients were calculated using ccf(). Results The summary statistics after multiple imputations using Bayesian linear regression are shown in Table 2. As shown descriptively in Table 2, mean performance improved across academic years in several disciplines, including the 100 m (11.57 ± 0.35 s in the first academic year to 11.30 ± 0.42 s in the fourth), SP (9.19 ± 1.53 m to 10.41 ± 1.23 m), and DT (24.70 ± 4.14 m to 29.87 ± 5.47 m). Mean total points also increased from 5676.95 ± 693.38 pts in the first academic year to 6287.63 ± 863.86 pts in the fourth academic year.Table 2Summary statistics after multiple imputation.VariablesAcademic year1234100 m11.57 ± 0.3511.47 ± 0.3611.35 ± 0.3511.30 ± 0.42LJ6.40 ± 0.426.38 ± 0.686.51 ± 0.636.59 ± 0.45SP9.19 ± 1.539.84 ± 1.3010.16 ± 1.2510.41 ± 1.23HJ1.77 ± 0.131.77 ± 0.191.77 ± 0.181.77 ± 0.29400 m52.96 ± 2.1352.28 ± 2.1351.83 ± 2.0251.76 ± 2.87110mH16.31 ± 1.4115.92 ± 2.2315.66 ± 2.3715.42 ± 2.84DT24.70 ± 4.1427.32 ± 6.1628.74 ± 5.6929.87 ± 5.47PV2.83 ± 1.093.16 ± 1.143.68 ± 0.713.59 ± 1.04JT44.46 ± 6.5646.30 ± 7.4247.88 ± 7.6148.52 ± 7.461500 m292.22 ± 16.81292.87 ± 16.28290.87 ± 16.84285.58 ± 48.23Total points5676.95 ± 693.385952.24 ± 784.966229.76 ± 720.646287.63 ± 863.86LJ, Long Jump; SP, Shot Put; HJ, High Jump; 110mH, 110 m hurdles; DT, Discus Throw; PV, Pole Vault; JT, Javelin Throw.Units: Running disciplines (e.g., 100 m, 110mH, 400 m, 1500 m) are shown in seconds (s), field disciplines (e.g., LJ, SP, HJ, DT, PV, JT) are shown in meters (m), and total points in points (pts). Table 2Summary statistics after multiple imputation.VariablesAcademic year1234100 m11.57 ± 0.3511.47 ± 0.3611.35 ± 0.3511.30 ± 0.42LJ6.40 ± 0.426.38 ± 0.686.51 ± 0.636.59 ± 0.45SP9.19 ± 1.539.84 ± 1.3010.16 ± 1.2510.41 ± 1.23HJ1.77 ± 0.131.77 ± 0.191.77 ± 0.181.77 ± 0.29400 m52.96 ± 2.1352.28 ± 2.1351.83 ± 2.0251.76 ± 2.87110mH16.31 ± 1.4115.92 ± 2.2315.66 ± 2.3715.42 ± 2.84DT24.70 ± 4.1427.32 ± 6.1628.74 ± 5.6929.87 ± 5.47PV2.83 ± 1.093.16 ± 1.143.68 ± 0.713.59 ± 1.04JT44.46 ± 6.5646.30 ± 7.4247.88 ± 7.6148.52 ± 7.461500 m292.22 ± 16.81292.87 ± 16.28290.87 ± 16.84285.58 ± 48.23Total points5676.95 ± 693.385952.24 ± 784.966229.76 ± 720.646287.63 ± 863.86LJ, Long Jump; SP, Shot Put; HJ, High Jump; 110mH, 110 m hurdles; DT, Discus Throw; PV, Pole Vault; JT, Javelin Throw.Units: Running disciplines (e.g., 100 m, 110mH, 400 m, 1500 m) are shown in seconds (s), field disciplines (e.g., LJ, SP, HJ, DT, PV, JT) are shown in meters (m), and total points in points (pts). Summary statistics after multiple imputation. LJ, Long Jump; SP, Shot Put; HJ, High Jump; 110mH, 110 m hurdles; DT, Discus Throw; PV, Pole Vault; JT, Javelin Throw. Units: Running disciplines (e.g., 100 m, 110mH, 400 m, 1500 m) are shown in seconds (s), field disciplines (e.g., LJ, SP, HJ, DT, PV, JT) are shown in meters (m), and total points in points (pts). Table 3 summarizes the results of the confirmatory factor analysis, based on the factor structures proposed in previous studies2,3. In the second and third academic years, model fit indices (CFI ≥ 0.96, SRMR ≤ 0.09, RMSEA ≤ 0.06) indicated good fit. In the first and fourth academic years, although CFI values were slightly lower, SRMR remained within acceptable limits; thus, these academic-year-specific models were retained for descriptive use, taking both fit indices and theoretical coherence into account.Table 3Model used for confirmatory factor analysis.FactorAcademic year1234Running speed and body projection100m, LJ, 110mH100m, LJ, 110mH100m, LJ, 110mH100m, LJ, 400mRunning endurance1500m1500m1500m1500mRotational throwing powerDTDTDTDTTranslational throwing powerSP, JTSP, JTSP, JTJTVertical leaping powerPV, HJPV, HJPV, HJPV, HJIndex p-value (chi-square) < 0.001 < 0.001 < 0.001 < 0.001 CFI0.9130.9670.9650.938 TLI0.8350.9370.9350.867 RMSEA0.1220.0590.0500.105 SRMR0.0610.0470.0470.055LJ, Long Jump; SP, Shot Put; HJ, High Jump; 110mH, 110 m hurdles; DT, Discus Throw; PV, Pole Vault; JT, Javelin Throw. CFI, comparative fit index; TLI, Tucker-Lewis index; RMSEA, root mean square error of approximation; SRMR, standardized root mean square residual. Table 3Model used for confirmatory factor analysis.FactorAcademic year1234Running speed and body projection100m, LJ, 110mH100m, LJ, 110mH100m, LJ, 110mH100m, LJ, 400mRunning endurance1500m1500m1500m1500mRotational throwing powerDTDTDTDTTranslational throwing powerSP, JTSP, JTSP, JTJTVertical leaping powerPV, HJPV, HJPV, HJPV, HJIndex p-value (chi-square) < 0.001 < 0.001 < 0.001 < 0.001 CFI0.9130.9670.9650.938 TLI0.8350.9370.9350.867 RMSEA0.1220.0590.0500.105 SRMR0.0610.0470.0470.055LJ, Long Jump; SP, Shot Put; HJ, High Jump; 110mH, 110 m hurdles; DT, Discus Throw; PV, Pole Vault; JT, Javelin Throw. CFI, comparative fit index; TLI, Tucker-Lewis index; RMSEA, root mean square error of approximation; SRMR, standardized root mean square residual. Model used for confirmatory factor analysis. LJ, Long Jump; SP, Shot Put; HJ, High Jump; 110mH, 110 m hurdles; DT, Discus Throw; PV, Pole Vault; JT, Javelin Throw. CFI, comparative fit index; TLI, Tucker-Lewis index; RMSEA, root mean square error of approximation; SRMR, standardized root mean square residual. In conducting group-based trajectory modeling, we first compared models with one to seven classes using the BIC (Table 4). The four-class model with the lowest BIC was selected. Next, we sequentially estimated polynomial degrees (number of terms) for the trajectory shape from two to four (Table 5). While BIC showed minimal improvement for polynomials of degree three or higher, consideration of classification accuracy (APPA and OCC) and consistency (relative entropy) indicated that the four-class, three-term model was optimal. In this solution, APPA was ≥ 0.97, OCC ≥ 68.1, and entropy > 0.96 for all classes. These values greatly exceed the thresholds proposed by Lennon13 et al. (APPA ≥ 0.7, OCC ≥ 5.0, entropy ≥ 0.5), demonstrating high classification reliability.Table 4Comparison of the number of classes using BIC in group-based trajectory modeling.ClassesBIC119532.28219161.35319050.78418930.25518947.56619091.43719331.74BIC, Bayesian information criterion.Lower values indicate better balance between model fit and parsimony. The model with the lowest BIC (in this case, 4-class model) was selected as optimal.Table 5Comparison of degrees using APPA, OCC, and relative entropy.ClassesDegreeAPPA(> 0.70)OCC(> 5.00)Relative entropy (> 0.50)420.97, 0.96, 0.98, 0.9982.8, 59.6, 153.8, 1328.20.95430.97, 0.98, 0.98, 1.0068.1, 119.8, 134.5, 1641.60.96APPA, average posterior probability of assignment; OCC, odds of correct classification. Thresholds: APPA ≥ 0.70, OCC ≥ 5.00, Entropy ≥ 0.50 (Lennon et al.,13). Table 4Comparison of the number of classes using BIC in group-based trajectory modeling.ClassesBIC119532.28219161.35319050.78418930.25518947.56619091.43719331.74BIC, Bayesian information criterion.Lower values indicate better balance between model fit and parsimony. The model with the lowest BIC (in this case, 4-class model) was selected as optimal. Comparison of the number of classes using BIC in group-based trajectory modeling. BIC, Bayesian information criterion. Lower values indicate better balance between model fit and parsimony. The model with the lowest BIC (in this case, 4-class model) was selected as optimal. Table 5Comparison of degrees using APPA, OCC, and relative entropy.ClassesDegreeAPPA(> 0.70)OCC(> 5.00)Relative entropy (> 0.50)420.97, 0.96, 0.98, 0.9982.8, 59.6, 153.8, 1328.20.95430.97, 0.98, 0.98, 1.0068.1, 119.8, 134.5, 1641.60.96APPA, average posterior probability of assignment; OCC, odds of correct classification. Thresholds: APPA ≥ 0.70, OCC ≥ 5.00, Entropy ≥ 0.50 (Lennon et al.,13). Comparison of degrees using APPA, OCC, and relative entropy. APPA, average posterior probability of assignment; OCC, odds of correct classification. Thresholds: APPA ≥ 0.70, OCC ≥ 5.00, Entropy ≥ 0.50 (Lennon et al.,13). The trajectory of the estimated four-class total points is shown in Fig. 1. The four classes were labeled High (N = 55, 30.7%), Upper-middle (N = 55, 30.7%), Lower-middle (N = 46, 25.7%) and Low (N = 23, 12.8%) based on the total points. Although the overall scores increased steadily in the High, Upper-middle, and Lower-middle class, the trajectories of the two middle classes were similar, both showing moderate increases over the four years. On the other hand, the Low class showed a gradual decrease, unlike the other three classes.Fig. 1Trajectories of total points among university decathletes across academic years. (a) Class means trajectories with 95% confidence intervals (CIs) calculated using Student’s t-distribution based on standard errors. (b) Individual trajectories for all athletes, colored by classification based on trajectory modeling. Trajectories of total points among university decathletes across academic years. (a) Class means trajectories with 95% confidence intervals (CIs) calculated using Student’s t-distribution based on standard errors. (b) Individual trajectories for all athletes, colored by classification based on trajectory modeling. The year-specific factor-score profiles for each total-point trajectory class are shown in Fig. 2. Descriptively, running speed and body projection, rotational throwing power, translational throwing power, and vertical leaping power tended to remain higher or increase across academic years in the High class, whereas the opposite tendency was observed in the Low class. The Upper-middle and Lower-middle classes showed patterns broadly corresponding to their total-point profiles. Running endurance appeared broadly stable in the High, Upper-middle, and Lower-middle classes, whereas a different pattern was observed in the Low class. Vertical leaping power appeared lower in the first academic year and increased from the second academic year onward, whereas running speed and body projection appeared more stable in the Upper-middle class than in the Lower-middle class. Because these factor scores were derived from academic-year-specific CFA models and standardized within each academic year, these patterns should be interpreted as descriptive profiles rather than as exact longitudinal changes in identical latent constructs.Fig. 2Trajectories of factor scores across four academic years for each competitive characteristic among Japanese university decathletes. (a) Running speed and body projection. (b) Running endurance. (c) Rotational throwing power. (d) Translational throwing power. (e) Vertical leaping power. Classes (High, Upper-middle, Lower-middle, Low) were estimated using group-based trajectory modeling (GBTM) based on total points. Factor scores were standardized within each academic year. Trajectories of factor scores across four academic years for each competitive characteristic among Japanese university decathletes. (a) Running speed and body projection. (b) Running endurance. (c) Rotational throwing power. (d) Translational throwing power. (e) Vertical leaping power. Classes (High, Upper-middle, Lower-middle, Low) were estimated using group-based trajectory modeling (GBTM) based on total points. Factor scores were standardized within each academic year. The lag-0 cross-correlation coefficients shown in Table 6 descriptively summarize the synchronous similarity between total-point trajectories and academic-year-specific factor-score profiles. In the High class, positive similarity was observed for running speed and body projection, rotational throwing power, translational throwing power, and vertical leaping power, whereas running endurance showed little synchronous similarity. The Upper-middle and Lower-middle classes showed more mixed patterns: running speed and body projection and rotational throwing power showed positive similarity in both classes, whereas translational throwing power and vertical leaping power showed negative similarity in the Upper-middle class. In the Low class, running speed and body projection, rotational throwing power, and vertical leaping power showed positive similarity, whereas translational throwing power showed negative similarity. Because these coefficients were calculated from four-point class-level series, they should be interpreted as descriptive summaries of synchronous pattern similarity rather than as inferential evidence.Table 6Cross-correlation coefficients between trajectories of factor scores and that of total points for each class (lag = 0).ClassesStructure of the competitive characteristicsRunning speed andbody projectionRunning enduranceRotational throwing powerTranslational throwing powerVertical leaping powerHigh0.95− 0.160.760.850.83Upper-middle0.530.390.85−0.85−0.84Lower-middle0.720.280.720.090.71Low0.770.290.70 − 0.370.61High, Upper-middle, Lower-middle, and Low refer to the four classes derived from group-based trajectory modeling (GBTM) based on total points.Values represent cross-correlation coefficients (lag = 0) between the trajectory of total points and each factor score.Lag 0 indicates synchrony within the same academic year; values close to 1 suggest stronger similarity in trajectories.Cross-correlations were calculated across all four academic years, using within-year pairings between factor scores and total points. Table 6Cross-correlation coefficients between trajectories of factor scores and that of total points for each class (lag = 0).ClassesStructure of the competitive characteristicsRunning speed andbody projectionRunning enduranceRotational throwing powerTranslational throwing powerVertical leaping powerHigh0.95− 0.160.760.850.83Upper-middle0.530.390.85−0.85−0.84Lower-middle0.720.280.720.090.71Low0.770.290.70 − 0.370.61High, Upper-middle, Lower-middle, and Low refer to the four classes derived from group-based trajectory modeling (GBTM) based on total points.Values represent cross-correlation coefficients (lag = 0) between the trajectory of total points and each factor score.Lag 0 indicates synchrony within the same academic year; values close to 1 suggest stronger similarity in trajectories.Cross-correlations were calculated across all four academic years, using within-year pairings between factor scores and total points. Cross-correlation coefficients between trajectories of factor scores and that of total points for each class (lag = 0). High, Upper-middle, Lower-middle, and Low refer to the four classes derived from group-based trajectory modeling (GBTM) based on total points. Values represent cross-correlation coefficients (lag = 0) between the trajectory of total points and each factor score. Lag 0 indicates synchrony within the same academic year; values close to 1 suggest stronger similarity in trajectories. Cross-correlations were calculated across all four academic years, using within-year pairings between factor scores and total points. In the reduced-missingness observed-only subset restricted to athletes with complete records for the second to fourth academic years, a 3-class linear trajectory solution was selected in the sensitivity analysis. Although the number of classes differed from that in the primary analysis, the broad pattern of total-point development was preserved, with higher- and middle-performing classes showing improvement and the lower-performing class showing less stable change. Representative observed disciplines showed qualitatively similar patterns to those in the primary analysis: 100 m, DT, JT, and PV tended to improve or remain stable in the higher-performing classes, whereas 1500 m performance did not show a comparably pronounced improvement pattern. These findings support the robustness of the main descriptive conclusions to the handling of missing data. Discussion This observational study examined whether the longitudinal performance patterns of decathletes were consistent with the principle of allocation. Descriptively, in the High class, total-point trajectories showed greater synchronous similarity with running speed, body projection, rotational throwing power, translational throwing power, and vertical leaping power than with running endurance. In the Upper-middle and Lower-middle classes, which showed more gradual increases in total points, running endurance showed only a small-to-moderate positive similarity, and the overall patterns were more mixed than in the high class. Taken together, these findings are broadly consistent with the principle of allocation, although they should be interpreted as observational and descriptive rather than definitive. In the high class, total-point development appeared more closely aligned with characteristics other than running endurance In the High class, total-point trajectories showed greater synchronous similarity with running speed, body projection, rotational throwing power, translational throwing power, and vertical leaping power than with running endurance. Previous studies have identified sprint- and throwing-related disciplines as important contributors to decathlon performance, particularly among higher-level athletes5,18–21. Consistent with the literature, the high class in the present study showed descriptive synchronous similarity between total-point development and running speed, body projection, rotational throwing power, and translational throwing power. Vertical leaping power also showed positive synchronous similarity, suggesting that jumping-related performance may remain relevant, even though its direct contribution to total points is generally smaller than that of several sprinting and throwing disciplines22–24. Overall, these observations are broadly consistent with the view that speed-, power-, and technique-related characteristics are closely associated with higher decathlon performances. These characteristics may also share similar biomechanical and neuromuscular demands. Sprinting, jumping, and throwing performance are often associated with the ability to generate force rapidly within a short time, which is commonly described as the rate of force development25. Previous research has linked such performance to muscle tissue characteristics, including muscle fiber architecture and a greater contribution of type II fibers25. Accordingly, the descriptive similarity observed among these characteristics in the High class may be broadly consistent with a shared dependence on rapid force production capacity. The present sample was restricted to university-aged athletes and therefore reflects a relatively early stage of the decathlon career20 rather than the age range more commonly associated with peak performance26. Therefore, the descriptive patterns observed here may partly reflect the developmental characteristics specific to this stage. Long-term longitudinal observations extending beyond university years may help clarify how these performance-allocation patterns change as athletes approach their competitive peak. In classes with increasing total points, running endurance did not show a comparably clear improvement pattern An important observational finding was that running endurance did not show a synchronous similarity with total-point development as clearly as several other competitive characteristics in the classes with increasing total points. Previous studies based on competition results have likewise suggested that 1500 m performance contributes less to overall decathlon performance than several sprinting, throwing, and jumping disciplines5,19,23,24. In addition, aerobic endurance has been described as a relatively weaker component of decathlon performance27, and endurance-related performance may gradually decline across development21. Taken together, the present descriptive results are consistent with the view that running endurance is less closely aligned with overall score development than with several other competitive characteristics. This pattern may reflect the ecological and physiological demands of decathlon performances. In the present study, four of the five competitive characteristics were related primarily to strength, explosive power, and event-specific technique, whereas running endurance was represented mainly by 1500 m performance. Previous research has suggested that sprinting, jumping, and throwing performance are associated with a greater contribution of fast-twitch muscle fibers and rapid force-production capacity25, whereas middle-distance running requires a different balance of speed- and endurance-related characteristics28,29. Muscle fiber composition and performance characteristics may also adapt differently depending on the training emphasis30–35. Accordingly, the comparatively stable pattern of running endurance observed here may partly reflect the possibility that training aimed at improving speed-, power-, and technique-related performance does not necessarily produce equally clear improvements in endurance-related performance. This interpretation is also consistent with practitioner-oriented reports that emphasize strength, sprinting, and technical training as major components of decathlon preparation10,36. Although previous research has suggested that training for sprinting, jumping, and throwing can also improve some aspects of middle- and long-distance performance37, running endurance in the present study remained broadly stable in the High, Upper-middle, and Lower-middle classes rather than showing a comparably clear improvement pattern. This suggests that improvements in speed-, power-, and technique-related performance do not necessarily translate into equally clear improvements in endurance-related performance in decathletes. Overall, the observed patterns were consistent with event-specific demands in the decathlon. In the present study, endurance-related demand was concentrated primarily in the 1500 m and, to a lesser extent, in the 400 m, whereas higher total-point development was more closely aligned with speed-, power-, and technique-related characteristics. These findings suggest that the ideal of an all-round decathlete who improves equally in all domains may be more aspirational than typical, at least in this sample of university-level athletes. These findings may also have practical implications for coaches overseeing athlete development. The identified trajectory classes may provide a descriptive framework for individualized training planning while recognizing that the present results are observational and not prescriptive. In particular, the results are consistent with an approach in which endurance is maintained at a level sufficient to complete the two-day competition, whereas speed-, power-, and technique-related development is prioritized according to the athlete’s trajectory and developmental stage. Limitations This study has several limitations. First, the factor scores were derived from academic-year-specific CFA models, and strict longitudinal measurement invariance was not imposed. Therefore, these scores should be interpreted as descriptive summaries of competitive characteristics rather than as fully comparable latent values on a single common scale across academic years. Second, the dataset was based solely on publicly available competition results, precluding access to individual physical characteristics (e.g., height and body mass), training history, and other contextual factors that may influence performance trajectories and factor structures, including competition-day weather conditions, which may have affected recorded performances in some disciplines. Third, reliance on imputed values for missing discipline records introduces uncertainty, particularly because missingness was concentrated in the first academic year. Although sensitivity analyses based on a reduced-missingness observed-only subset for the second to fourth academic years yielded qualitatively similar patterns, uncertainty remains, particularly for lower-performing classes and endurance-related performance. Fourth, the absence of randomized training assignment and the possibility of cohort effects (e.g., differences in coaching staff or institutional training emphasis across entry years) may have influenced the observed patterns. Fifth, our analysis focused exclusively on male Japanese university decathletes, and generalizability to female athletes, other age groups, or different countries remains to be established. Finally, because of the observational design, the present findings should not be interpreted as causal evidence. Future directions To address these limitations, future research should: (1) conduct longitudinal studies collecting individual-level data on athlete characteristics and training history, enabling finer analysis of the factors underlying ability development; (2) undertake international comparative studies to clarify how cultural or institutional factors shape multi-event performance and growth; (3) compare different periodization models to evaluate their impact on the allocation of training emphases and performance trajectories; (4) extend the analysis to other multi-event disciplines (e.g., heptathlon, indoor combined events) to assess the generalizability of allocation patterns; (5) explore the potential of genetic profiling to predict individual predispositions for specific allocation trade-offs; and (6) track post-university careers to examine how university-period ability development affects long-term athletic sustainability. Such studies would significantly advance understanding of both the mechanisms and durability of ability allocation in decathlon and other multi-event disciplines. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1