Construct differentiation and reliability of a novel curvilinear sprint test in female handball players

Abstract

Background

The curved 6-m goal-area line offers a practical basis for assessing curvilinear sprinting in handball, yet sprint testing in the sport remains predominantly linear. This study evaluated the reliability and construct differentiation of a curvilinear sprint (CS) test based on this existing court marking.

Methods

Twenty-five female handball players completed one familiarisation and two test sessions separated by 48 h. In each session, players performed three linear sprint (LS) trials and three CS trials in each direction over the same 21.85-m distance. Performance differences and session effects were examined using repeated-measures analysis of variance, while reliability was assessed using intraclass correlation coefficients, coefficient of variation, typical error, minimal detectable change, and Bland–Altman analysis.

Results

LS was faster than the right and left CS conditions by 0.80 and 0.83 s, respectively (both p < 0.001), while LS and CS shared 58% to 68% of their variance. Between-session ICCs for the CS conditions ranged from 0.88 to 0.89, with CVs of 1.66% to 1.83%, while three-trial within-session ICCs ranged from 0.95 to 0.98. Performance was faster in the second session (p = 0.003), although the changes were smaller than the corresponding minimal detectable change values.

Conclusıons

Curvilinear sprint performance is strongly related to, but not fully represented by, linear sprint performance. The test showed good repeatability and provides a practical option for directly assessing curvilinear sprint ability in female handball players.

Introduction

Handball is an intermittent team sport. Play alternates between short bursts of high-intensity work and longer periods of low-intensity activity, and players perform explosive actions such as accelerating, decelerating, jumping, blocking, changing direction and shooting at the goal [26, 29]. Over a match, players cover approximately 2,000 to 5,000 m, of which about 30% is accumulated at high intensity [18, 25], and they change activity every 4.6 to 5.9 s, corresponding to 501 to 663 transitions per match [7]. High-intensity running and sprinting account for less than 10% of total playing time [13], yet the sprints performed during ball recovery and counterattacks are among the actions that decide the outcome of a match [18]. The curved geometry of the 6-m goal-area line provides a practical context for examining curvilinear sprint performance in handball. This raises the question of whether conventional linear sprint tests fully represent performance on a curved trajectory.

In team sports such as football, basketball and handball, high-intensity running and multidirectional movement account for a substantial part of competition activity [4, 32, 33]. Sprinting in these sports is expressed in several forms, including linear sprints (LS), curved sprints (CS) and change-of-direction (CoD) sprints [12]. Research has nevertheless concentrated on LS and CoD rather than on CS, and no test specific to curvilinear sprinting existed until recently [11]. The studies that have examined CS have been carried out mainly in football [11, 12, 14, 22] and basketball [3]. Linear and curvilinear sprint times are consistently related, but the strength of the relationship varies widely between studies, from r = 0.64 between linear and curve sprint times in male soccer players [12] to r = 0.93 to 0.95 between the corresponding velocities in elite female soccer players [19]. The overlap is nevertheless incomplete in every case, and in male soccer players it accounts for 41% of the variance [12]. The two tasks also impose different kinematic and neuromuscular demands [10], and in basketball a three-point line curve sprint test produced lower acceleration and velocity outputs than a linear sprint over the same distance [3]. In male youth soccer players, curve sprint performance improves with age and is related to vertical jump height and to agility test performance [28]. Taken together, these findings indicate that curvilinear sprinting is related to but not interchangeable with linear sprinting, and that it should be assessed as a capacity in its own right [3]. What remains unclear is whether this conclusion extends to handball, where the court geometry and the movement patterns differ from those of football and basketball.

The handball court itself provides a standardised curved trajectory through the 6-m goal-area line. CS can be defined as a sprint performed with a gradual and continuous change of direction [11], and it differs from LS not only in technical execution but also in its kinematic and neuromuscular characteristics [3, 10]. Unlike LS, CS requires the generation of centripetal force to negotiate the curve. The peak resultant ground reaction force available is reduced as a result, and the inside and outside legs assume different roles, which lowers the running speed that can be sustained [6, 8]. These mechanical differences provide a rationale for assessing CS separately from LS when curvilinear sprint performance is of specific interest. Although the relationship between CS and performance has been well documented in soccer and basketball, no study specific to handball has been identified, and the reliability of athletic tests in female handball players has itself been examined only to a limited extent [27]. A curvilinear sprint test referenced to handball court markings, and evidence on how it behaves on repeated administration, are therefore both lacking.

Therefore, the objectives of the present study are: (i) to test the construct differentiation of the CS test designed for handball players, (ii) to evaluate the inter-session reliability of the same test, and (iii) to examine the intra-session reliability of the tests. We hypothesised that the CS test would show high intra-session and inter-session reliability, and that CS performance would be systematically slower than, and only partly explained by, LS performance, indicating that the two tests capture partly different aspects of sprint ability. To our knowledge, limited research has examined construct differentiation and both inter-session and intra-session reliability of a curvilinear sprint test based on the existing 6-m handball goal-area line. The findings are intended to inform test batteries used to evaluate sprint performance in handball players.

Methods

Participants

This study included 25 female handball players aged between 16 and 22 years (age: 19.16 ± 2.51 years; height: 169.32 ± 3.87 cm; body weight: 60.11 ± 6.16 kg; BMI: 20.95 ± 1.79). A convenience sampling strategy was used. The players were recruited from the two handball teams in the Turkish 2nd Handball League to which the research team had access, and all athletes at these clubs who met the inclusion criteria and volunteered were enrolled. This approach was adopted because the number of female players competing at this level within reach of the research team was limited; therefore, all eligible players who volunteered from the two accessible clubs were included. The consequences of this strategy for the generalisability of the findings are addressed in the Limitations and Future Directions section. Although the age range (16 to 22 years) spans more than one stage of the Long-Term Athlete Development framework, all participants were competing in the same senior league, and they were therefore treated as a single group. During the season, all participants took part in handball training 4 days per week (each session lasting at least 90 min) and played one competitive match per week; the present study, however, was conducted during a non-competitive period, so no matches were played during the testing period. Menstrual cycle phase and hormonal contraceptive use were not recorded and were not controlled for in the scheduling of the test sessions; this is acknowledged in the Limitations and Future Directions section. The inclusion criteria for the study were as follows: (i) having at least 4 years of handball training experience, (ii) not having suffered an injury in the last three months or having no medical contraindications for the tests, and (iii) participating in all tests. All participants were informed about the risks and benefits of the study, and both the players and their parents or guardians (for those under 18 years of age) signed an informed consent form before starting the study. Approval was obtained from the Scientific Research and Publication Ethics Committee of Iğdır University prior to the start of the study (Date: 02.10.2025; Approval no: 40). All procedures were conducted in accordance with the Helsinki Declaration.

Experimental design

This study was designed to examine the construct differentiation of a 6-m line-referenced CS test relative to the LS test in handball players, and the reliability of that test within and between sessions. The athletes participated in a total of three sessions: a familiarisation session and two separate test sessions (session 1 and session 2). A 48-h rest period was provided between sessions to allow for recovery. During the familiarisation session, participants’ body composition measurements were taken, and they were informed about the test procedures and given a practical introduction. In session 1 and session 2, participants first underwent the linear sprint (LS: 21.85 m) test, followed by the curved sprint (CS: 21.85 m; right and left direction) tests. Each test was performed in three trials, with a 3-min full rest period between trials. Before both test sessions, athletes performed a standard 10-min warm-up (5 min of low-intensity running and 5 min of dynamic stretching exercises) with their coach. All tests were conducted on the handball court where the team played its official matches, under fixed environmental conditions (temperature 20–22ºC, relative humidity 45–50%), and athletes were required to wear the same shoes during the tests. The study was conducted during a non-competitive period of the season. Players were asked not to make any changes to their lifestyle (sleep, nutrition, diet, etc.) and not to use ergogenic support products during the test period. Verbal motivation was provided to participants in all performance tests.

The timeline for the experimental design and testing sessions is given in Fig. 1. Participants completed the LS and CS tests 48 h apart, following practice and anthropometric measurements.

Fig. 1.

Study design. LS = Linear Sprint; CS = Curvilinear Sprint

Data collection

Body composition measurements

The athletes’ heights were measured using a wall-mounted stadiometer (SECA, Germany), and their body weights were measured using an electronic scale (Tanita BC-418, Japan) in accordance with the protocols.

Linear sprint test

Participants’ 21.85 m sprint performances were recorded using photocells (Fusion Sport Smart Speed, Australia) placed at the start and finish points. Each athlete completed the test three times, and the mean of the three trials was used for analysis (Fig. 2). All sprints were performed from a standing start. A line was marked on the court 1 m behind the first timing gate, and participants were instructed to position their leading foot behind this line and not to cross it before setting off, so that no run-up was possible and every trial began from a stationary position. Because starting distance is known to affect absolute sprint times [1], the same 1-m line was used for every trial and every session, so that the procedure was identical throughout. A full 3-min rest period was allowed between trials.

Fig. 2.

Linear sprint test setup (21.85 m)

Curvilinear sprint test

The curved trajectory was standardised using the court markings. Participants followed the 6-m goal-area line (solid arc; radius = 6 m), which defined the running path, with the 9-m free-throw line (dashed arc) marking the outer limit of the running lane. The sprint began from a standing start behind the same 1-m line described above, followed the goal-area line around the goal area over an arc of approximately 180°, and covered a total distance of 21.85 m to the finish line. This distance is the length of the 6-m goal-area line, which is formed by two quarter-circles of 6 m radius joined by a 3 m straight section. Timing gates were positioned at the start and finish to record sprint time (Fig. 3).

Fig. 3.

Curvilinear sprint test setup (21.85 m)

Statistical analysis

All data are presented as mean ± SD. For the construct differentiation, correlation and inter-session reliability analyses, the mean of the three trials within each condition and session was used as the unit of observation, giving one value per participant, condition and session (n = 25). The intra-session reliability analyses and the test of the trial effect were necessarily conducted on the individual trials. Trials were averaged rather than entered as separate observations because repeated trials performed by the same athlete within a session are not independent; treating them as independent would inflate the effective sample size and narrow the confidence intervals artificially.

Construct differentiation across sprint conditions (LS, CLSR, CLSL) was examined with a two-way repeated-measures analysis of variance, with Condition (LS, CLSR, CLSL) and Session (1, 2) as within-participant factors and their interaction. Sphericity was assessed with Mauchly’s test and, where required, degrees of freedom were corrected using the Greenhouse–Geisser estimate; partial eta squared is reported as the effect size. Pairwise comparisons (LS vs CLSR; LS vs CLSL; CLSR vs CLSL) were computed as paired-samples comparisons and reported as mean differences (Δ, s) with 95% confidence intervals and Cohen’s dz (mean difference divided by the standard deviation of the difference scores); p values for these comparisons were adjusted using the Holm procedure.

Relationships among sprint outcomes (LS, CLSR, CLSL), as well as the good side, weak side and asymmetry variables, were examined with Pearson correlations (r), alongside R² and 95% confidence intervals for r. For these analyses each participant contributed a single value per condition, taken as the mean of the two sessions to reduce the influence of session-to-session measurement error. The good side was defined as the direction in which a participant recorded the faster curve sprint time and the weak side as the direction in which the slower time was recorded; sides were therefore assigned on the basis of measured performance rather than self-reported preference, throwing arm or limb dominance. Asymmetry was calculated as the difference in sprint time between the weak and the good side (s). Correlation magnitudes were interpreted as trivial (< 0.10), small (0.10–0.29), moderate (0.30–0.49), large (0.50–0.69), very large (0.70–0.89), and nearly perfect (≥ 0.90) [16, 17].

Session-to-session differences within each condition were tested with paired-samples t tests, with Cohen’s d reported as the effect size. The normality of the difference scores was examined with the Shapiro–Wilk test, and the Wilcoxon signed-rank test was used as a sensitivity analysis for any condition in which the assumption was not met.

Inter-session reliability was evaluated using intraclass correlation coefficients with 95% confidence intervals. ICC(3,1) (two-way mixed effects, consistency, single measures) was calculated on the session mean values. For intra-session reliability, the same model was applied across the three trials within a session, with ICC(3,1) reported for a single trial and ICC(3,3) for the mean of three trials. Because trials were administered in a fixed order, the trial effect was treated as fixed and was tested separately using a one-way repeated-measures analysis of variance. ICC values were interpreted as poor (< 0.50), moderate (0.50–0.75), good (0.75–0.90), and excellent (> 0.90) [20]. Absolute reliability was quantified using the typical error (TE = SD of the difference scores/√2), the coefficient of variation (CV, %), expressed as the typical error relative to the grand mean of the two sessions, and the minimal detectable change at the 95% confidence level (MDC₉₅ = 1.96 × √2 × TE). For the intra-session analyses, where three trials rather than two sessions are compared, the typical error was taken as the square root of the residual mean square of the participants × trials model and the coefficient of variation was expressed relative to the grand mean of that session,MDC₉₅ was calculated with the same formula. Means and standard deviations in Table 3 are calculated across the 25 session mean values, whereas those in Table 4 are calculated across all individual trial values, so the standard deviations differ slightly between the two tables. CV values below 5% were considered indicative of acceptable reliability [21], and MDC₉₅ was interpreted as the minimum change required to exceed measurement error with 95% confidence [34]. Agreement between the two sessions was additionally examined with Bland–Altman analysis [5]. The bias and the 95% limits of agreement were calculated from the session-to-session difference scores, and proportional error was assessed by correlating the difference between sessions with the mean of the two sessions [2].

Table 3.

Session-to-session reliability

Test	Session 1 (s)	Session 2 (s)	Δ (S2 – S1) (s)	p (paired t)	ICC (3,1) 95% CI	CV (%)	TE (s)	MDC95 (s)
LS	3.38 ± 0.17	3.34 ± 0.17	− 0.03	0.015	0.93 (0.85–0.97)	1.32	0.04	0.12
CLSR	4.18 ± 0.24	4.14 ± 0.20	− 0.05	0.043	0.88 (0.75–0.95)	1.83	0.08	0.21
CLSL	4.22 ± 0.21	4.15 ± 0.21	− 0.06	0.003	0.89 (0.77–0.95)	1.66	0.07	0.19

Table 4.

Intra-session reliability (three trials per session)

Test	Session	Mean ± SD (s)	ICC(3,1) 95% CI	ICC(3,3) 95% CI	CV (%)	TE (s)	MDC95 (s)
LS	1	3.38 ± 0.17	0.92 (0.85–0.96)	0.97 (0.94–0.99)	1.46	0.049	0.14
LS	2	3.34 ± 0.17	0.93 (0.86–0.96)	0.97 (0.95–0.99)	1.39	0.047	0.13
CLSR	1	4.18 ± 0.25	0.95 (0.90–0.98)	0.98 (0.97–0.99)	1.33	0.056	0.15
CLSR	2	4.14 ± 0.22	0.91 (0.83–0.96)	0.97 (0.94–0.98)	1.55	0.064	0.18
CLSL	1	4.22 ± 0.22	0.89 (0.81–0.95)	0.96 (0.93–0.98)	1.67	0.071	0.20
CLSL	2	4.15 ± 0.22	0.85 (0.74–0.93)	0.95 (0.89–0.97)	2.05	0.085	0.24

Because the sample size was determined by the number of players available at the two participating clubs, no a priori sample size calculation was performed. The precision achieved with the present sample was therefore evaluated a posteriori from the width of the ICC confidence intervals. With 25 participants and two sessions, the 95% confidence intervals spanned approximately ± 0.06 for LS, ± 0.10 for CLSR and ± 0.09 for CLSL, with lower bounds of 0.85, 0.75 and 0.77 respectively. The precision of the reliability estimates reported below should be judged against these intervals.

Analyses were conducted in JASP (JASP version 0.18.3, Amsterdam, The Netherlands). Statistical significance was set at p < 0.05.

Results

Sprint times differed markedly between conditions, F(2, 48) = 754.39, p < 0.001, ηp² = 0.969. LS was faster than both CLSR and CLSL (Δ = 0.80 s and 0.83 s, respectively; both p < 0.001; dz = 6.46 and 6.20; Table 1), whereas the two curve conditions did not differ from each other (Δ = 0.03 s; 95% CI − 0.02 to 0.07; p = 0.200; dz = 0.26). There was a main effect of Session, F(1, 24) = 10.54, p = 0.003, ηp² = 0.305, reflecting the slightly faster times recorded in the second session, but the Condition × Session interaction was not significant, F(2, 48) = 1.46, p = 0.243, ηp² = 0.057, indicating that the session-to-session change did not differ across sprint conditions. Sphericity was not violated for either effect involving Condition (Greenhouse–Geisser ε = 0.93 in both cases), and applying the correction did not alter any conclusion.

Table 1.

Pairwise contrasts between sprint conditions, pooled across sessions

Contrast	Δ (s)	95% CI	t (24)	p
CLSR − LS	0.80	0.75–0.85	32.32	< 0.001
CLSL − LS	0.83	0.77–0.88	31.01	< 0.001
CLSL − CLSR	0.03	− 0.02–0.07	1.32	0.200

The difference between LS and CLSL was nominally larger than the difference between LS and CLSR (0.83 vs 0.80 s), although the two curve conditions did not differ significantly from one another. The correlations between LS and the curve conditions were very large (LS–CLSR: r = 0.82; LS–CLSL: r = 0.76); however, the proportion of shared variance was higher for CLSR than for CLSL (R² = 0.68 vs 0.58) (Table 2). LS was also very largely associated with both the good side (r = 0.81; R² = 0.66) and the weak side (r = 0.80; R² = 0.64).

Table 2.

Correlations (r) and R² between linear and curvilinear sprint performance

Comparison	r	R2	95% CI for r
LS vs. CLSR	0.82	0.68	0.64–0.92
LS vs. CLSL	0.76	0.58	0.52–0.89
CLSR vs. CLSL	0.88	0.77	0.74–0.95
LS vs. Good side	0.81	0.66	0.61–0.91
LS vs. Weak side	0.80	0.64	0.60–0.91
LS vs. Asymmetry	− 0.22	0.05	− 0.56–0.20

In contrast, the association between LS and asymmetry, defined as the difference in sprint time between the weak and the good side, was small and negative (r = − 0.22; R² = 0.05; mean asymmetry 0.08 ± 0.07 s), and the 95% CI crossed zero (Table 2). Inter-session reliability was good for both CS conditions and excellent for LS, with ICC(3,1) values ranging from 0.88 to 0.93 and CV values from 1.32% to 1.83%. Sprint times were faster in Session 2 in all three conditions, and the mean change was of a similar magnitude throughout (LS Δ = − 0.03 s, p = 0.015, d = 0.52; CLSR Δ = − 0.05 s, p = 0.043, d = 0.43; CLSL Δ = − 0.06 s, p = 0.003, d = 0.66). The Shapiro–Wilk test indicated that the difference scores were normally distributed for LS (W = 0.984, p = 0.947) and CLSL (W = 0.946, p = 0.205) but not for CLSR (W = 0.893, p = 0.013); the Wilcoxon signed-rank test was therefore used as a sensitivity analysis and gave p = 0.023 for LS, p = 0.081 for CLSR and p = 0.006 for CLSL, so the two tests agreed in direction for all three conditions but fell on opposite sides of the significance threshold for CLSR. In every condition the mean change was smaller than the corresponding MDC₉₅, so a shift of this size would not be distinguishable from measurement error in an individual athlete. Absolute error was low (TE = 0.04–0.08 s), and MDC₉₅ ranged from 0.12 s (LS) to 0.21 s (CLSR) (Table 3).

Figure 4, Individual trial times are shown for the linear sprint (LS) and the curvilinear sprint conditions (CLS Left, CLS Right) across Session 1 and Session 2. Circles, triangles and squares represent the 1st, 2nd and 3rd trials respectively. Solid lines indicate the mean value for each condition across trials. Lower times indicate better performance.

Fig. 4.

Session-to-session sprint times for linear and curvilinear sprint conditions

Agreement between sessions was additionally examined using Bland–Altman analysis (Fig. 5). The bias was − 0.03 s for LS, − 0.05 s for CLSR and − 0.06 s for CLSL, consistent with the slightly faster times recorded in the second session. The 95% limits of agreement were − 0.16 to + 0.09 s for LS, − 0.26 to + 0.17 s for CLSR and − 0.26 to + 0.13 s for CLSL. There was no clear evidence of proportional error in any condition: the correlation between the difference and the mean of the two sessions was negligible for LS (r = − 0.001, p = 0.998) and CLSL (r = 0.025, p = 0.907) and did not reach significance for CLSR (r = − 0.341, p = 0.095), and the same held when the absolute difference was used (all p > 0.10).

Fig. 5.

Intersession adaptation for linear sprint, left curvilinear sprint, and right curvilinear sprint

Figure 5, Bland–Altman plots for session-to-session agreement in the linear sprint (left), left curvilinear sprint (middle), and right curvilinear sprint (right). Each point represents one participant (n = 25), plotted as the difference between sessions (session 2 − session 1, in seconds) against the mean of the two sessions. The central dashed line indicates the mean bias, and the outer dashed lines indicate the 95% limits of agreement; shaded areas represent the corresponding 95% confidence intervals. Note that the vertical scale differs between panels.

Intra-session reliability ranged from good to excellent for single trials (ICC(3,1) = 0.85 to 0.95) and was excellent for the mean of three trials (ICC(3,3) = 0.95 to 0.98). Absolute variability was low (CV = 1.33–2.05%), with TE values between 0.047 and 0.085 s and MDC₉₅ ranging from 0.13 to 0.24 s (Table 4). Averaging three trials improved reliability in every condition, and LS showed slightly lower absolute error than the curve conditions. Sprint times also changed systematically across the three trials within a session. Times became faster from the first to the third trial in every condition, and this trial effect was significant in five of the six condition-by-session combinations: LS Session 1, F(2, 48) = 4.03, p = 0.024; CLSR Session 1, F(2, 48) = 62.02, p < 0.001; CLSR Session 2, F(2, 48) = 27.39, p < 0.001; CLSL Session 1, F(2, 48) = 8.79, p < 0.001; and CLSL Session 2, F(2, 48) = 11.07, p < 0.001. The effect was not significant for LS in Session 2 (F(2, 48) = 1.04, p = 0.362). It was largest in the right curve condition, where the first trial was on average 0.17 s slower than the third in Session 1. Individual trial times for all conditions and both sessions are shown in Fig. 4.

Dıscussıon

The main finding of this study is that curvilinear sprint performance around the 6-m goal-area line is related to, but not interchangeable with, linear sprint performance. Although LS and CS were strongly associated, linear sprint performance accounted for 58% to 68% of the variance in the curve conditions, while players were approximately 0.8 s slower when covering the same distance on the curved trajectory. Together, these findings suggest that linear sprint ability contributes substantially to curvilinear sprint performance but does not fully represent the additional demands imposed by running on a curve. This supports the construct differentiation of the CS test and indicates that linear sprint performance alone does not fully represent performance on the 6-m curved trajectory. The test also showed good between-session and excellent three-trial within-session reliability, with low absolute measurement error, supporting its use for monitoring performance when changes are interpreted against the corresponding MDC₉₅ values. An important additional finding, however, was that performance improved both between sessions and across trials within a session, showing a pattern consistent with a familiarisation effect that should be considered when the test is implemented in practice.

The construct differentiation observed between linear and curvilinear sprinting is also biomechanically plausible. Running on a curve requires athletes to generate centripetal force while maintaining forward propulsion, resulting in different demands on the inside and outside legs. Chang and Kram [6], using curves with radii of 1 to 6 m, reported lower peak resultant forces during curved compared with straight running and consistently lower peak forces in the inside leg. Differences between the inside and outside steps have also been reported for ground contact time, touchdown distance, and hip motion during bend sprinting, although these findings were obtained on a substantially larger curve radius and should therefore be applied to the present test with caution [8]. More recently, hip abductor strength has shown stronger associations with curvilinear than with linear sprint performance, further suggesting that curve sprinting involves demands that are not fully represented by straight-line speed [30]. These mechanisms were not measured in the present study and should therefore be regarded as plausible explanations rather than direct evidence for the performance differences observed here. The degree of overlap between linear and curvilinear sprint performance also varies considerably across studies. Shared variance has ranged from approximately 35% in semi-professional male soccer players [11] to 86% to 90% in elite female soccer players [19], with the 58% to 68% observed in the present study falling between these values. Differences in course geometry may partly explain this variation. Unlike the continuously curved trajectory used by Fílter et al. [11], the handball 6-m line includes a 3-m straight central section, which may increase the contribution of linear sprint ability to overall performance. Population and competitive level may also contribute, as Kobal et al. [19] suggested that higher-level players may be better able to express their linear sprint capacity while running on a curve. This distinction is also supported by longitudinal evidence. In male under-16 soccer players, Solleiro-Duran et al. [31] found that six weeks of linear sprint training did not improve curvilinear sprint performance, whereas curvilinear sprint training improved performance on the curved course. Although these findings were obtained in a different sport and population, the different responses to the two training approaches provide further evidence that improvements in linear sprint ability do not necessarily transfer directly to curvilinear sprint performance. Taken together, the available evidence suggests that CS performance shares substantial common variance with LS while also involving additional task-specific demands, supporting the direct assessment of curvilinear sprint performance when this ability is of specific interest.

The CS test also showed good repeatability within and between sessions. Between-session ICC values were 0.88 and 0.89 for the right and left curve conditions, respectively, while averaging three trials within a session yielded ICC(3,3) values between 0.95 and 0.98. These findings are comparable with previous studies reporting high reliability for curvilinear sprint tests in soccer (ICC = 0.89 to 0.93; [11], basketball (ICC = 0.93 to 0.94, [3], and performance testing in female handball players (ICC = 0.82 to 0.94, [27]. Absolute measurement error was also low, with within-session CV values of 1.33% to 2.05%, well below the 5% criterion commonly used in sport performance testing [21]. The standardised testing procedure may have contributed to this consistency, as the running path was defined by permanent court markings, the starting position was fixed 1 m behind the timing gate, and all sessions were performed on the same court under the same testing conditions. From a practical perspective, the TE and MDC₉₅ values are particularly useful because they allow practitioners to determine whether a change in an individual player’s sprint time is greater than the expected measurement error [15, 34]. Bland–Altman analysis provided no clear evidence of proportional error across sessions, further supporting the consistency of the measurements. Overall, these findings indicate that the test provides sufficient repeatability for monitoring curvilinear sprint performance in female handball players when the mean of three trials is used and individual changes are interpreted relative to the corresponding MDC₉₅ values.

An important consideration when implementing the test is the influence of familiarisation and trial order. Sprint times improved from the first to the second session across all conditions. The improvement did not differ significantly between conditions, suggesting that the pattern was not specific to linear sprinting or either turning direction. Although the mean changes were smaller than the corresponding MDC₉₅ values, the consistent direction of improvement suggests that a single familiarisation session may not have been sufficient to establish a completely stable baseline. This interpretation is consistent with previous evidence showing that performance in unfamiliar physical tests may continue to improve after an initial familiarisation session [23]. A similar pattern was observed within sessions, where sprint times generally improved across successive trials. The clearest example occurred in the right curve sprint during Session 1, where the first trial was 0.17 s slower than the third, exceeding the corresponding within-session MDC₉₅ of 0.15 s. This finding supports the use of multiple trials and suggests that practitioners should avoid basing decisions on a single early attempt. Small differences between turning directions should also be interpreted cautiously. Right and left curve sprint times did not differ at the group level, the mean asymmetry of 0.08 s was similar in magnitude to the typical error of the curve conditions, and five of the 25 players changed their faster side between sessions. Together with previous evidence that the direction of asymmetry can vary across non-linear tasks [12], these findings suggest that small side-to-side differences may partly reflect normal measurement variability rather than a stable directional characteristic.

Limitations and future directions

Several limitations should be considered when interpreting the present findings. First, although the test was designed around the existing 6-m goal-area line to provide a court-based and easily reproducible trajectory, it remains unknown how closely this course reflects the curved sprints performed during competition. The central 3-m straight section may also increase the contribution of linear sprint ability to overall test performance. Future studies using positional or time-motion data could identify the trajectories most frequently performed during match play and determine how well the present test represents these demands. Second, only one familiarisation session was completed, while performance continued to improve across sessions and trials. Additional familiarisation may therefore be required before establishing an individual baseline. Third, the sample consisted of 25 female players from two clubs competing in the Turkish 2nd Handball League. The findings should consequently be generalised cautiously to male players, other competitive levels, age groups, and playing positions, and larger samples would allow these factors to be examined directly. Fourth, no biomechanical variables were collected. The mechanical explanations discussed above are therefore based on previous research and should not be interpreted as mechanisms demonstrated by the present study. Finally, menstrual cycle phase and hormonal contraceptive use were not recorded or controlled. Although the average effects of menstrual cycle phase on physical performance appear to be small and variable across individuals, reproductive status remains an important methodological consideration in studies involving female athletes [9, 24]. Its potential influence on the present reliability estimates therefore cannot be determined.

Practical ımplications

The proposed CS test can be incorporated into routine handball testing using the existing 6-m goal-area line, providing a standardised way to assess curved sprint performance without constructing an additional running course. Because linear sprint performance accounted for 58% to 68% of the variance in CS performance, coaches and practitioners should assess curvilinear sprint ability directly when this quality is relevant rather than infer it from linear sprint speed. For repeated monitoring, the mean of three trials should be used following adequate familiarisation, and changes greater than 0.21 s for the right and 0.19 s for the left curve sprint can be interpreted as exceeding the between-session measurement error represented by the MDC₉₅.

Conclusion

The curvilinear sprint test based on the 6-m handball goal-area line demonstrated construct differentiation from linear sprinting while showing good repeatability in female handball players. Although linear and curvilinear sprint performances were strongly related, linear sprint ability did not fully represent performance on the curved trajectory, supporting the direct assessment of curvilinear sprinting when this ability is of interest. The low measurement error and acceptable between-session reliability support the use of the test for repeated monitoring, although adequate familiarisation and multiple trials should be considered when establishing an individual baseline. Because the test uses an existing court marking and standard timing-gate equipment, it provides a simple and practical option for incorporating curvilinear sprint assessment into handball performance testing.

Abbreviations

CS

Curvilinear sprint

LS

Linear sprint

CoD

Change of direction

CLSR

Curvilinear sprint, right direction

CLSL

Curvilinear sprint, left direction

ICC

Intraclass correlation coefficient

CV

Coefficient of variation

TE

Typical error

MDC₉₅

Minimal detectable change at the 95% confidence level

Authors’ contributions

Authors Contributions Conceptualization and study design: DŞ, GAT, EG. Data collection, analysis, and interpretation: GAT, NEA, EG, DŞ. Manuscript drafting and revisions: GAT, DŞ. All the authors reviewed and approved the final manuscript.

Funding

The authors declare that no financial support or funding was received for this study.

Data availability

The datasets generated and/or analyzed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

The study was conducted in accordance with the Declaration of Helsinki. All participants were informed in detail about the purpose of the study, the testing procedures, and possible risks. Participation was voluntary, and written informed consent (participation consent statement) was obtained from all participants, with additional written consent from a parent or guardian for participants under 18 years of age. Approval was obtained from the Scientific Research and Publication Ethics Committee of Iğdır University prior to the start of the study (Date: 02.10.2025; Approval no: 40). Clinical trial number: Not applicable. Consent for Publication: Not applicable.

Competing interests

The authors declare no competing interests.