Classification of four pitching styles in Japanese baseball players

Abstract

This study aimed to provide normative biomechanical data of baseball delivery styles and to verify an appropriate criterion for quantitatively classifying baseball pitches into four delivery styles. In total, 74 pitches were visually identified by seven coaches into 24 overhand (OS), 17 three-quarter (TS), 21 sidearm (SS) and 12 underhand (US) deliveries. The accuracies of the classifications of the pitches using the arm slot angle (θ_AS) and using the combination of trunk lateral tilt (θ_T) and upper arm elevation (θ_A) angles at release were compared. Average values for θ_AS were 53.9°, 31.5°, 5.8°, and −34.8° in the OS, TS, SS and US; corresponding values for θ_T were 31.9°, 15.4°, −0.9°, and −40.1°; and for θ_A were 39.8°, 16.1°, −6.3°, and−43.0°. Both variables (θ_AS and the average of θ'_T and θ'_A, which corresponded to the projection of each point on the regression line) correctly classified the delivery styles, but the accuracy of the classification using θ_AS (96%) was higher than that using the average of θ'_T and θ'_A (89%). There were fewer and smaller overlaps of the ranges of pitches classified using θ_AS (only 6 pitches between TS and SS (14.9° to 23.0°), which can be considered a buffer zone) compared to the average of θ'_T and θ'_A. Consequently, θ_AS seems the most appropriate for classifying baseball pitches into the four delivery styles. The separation points between the four styles were marked by θ_AS values of 45° (between OS and TS), 19° (between TS and SS) and −13° (between SS and US).

Keywords

Kinematics motion analysis overhand sidearm throwing underhand sport biomechanics

Introduction

The baseball pitching motion is generally classified into four delivery styles based on the visually identified release position of the throwing arm in a global reference system (GRS): overhand (OS), three-quarter (TS), sidearm (SS) and underhand (US) (or “submarine”) styles.^1,2 However, as the classification based on coaches’ intuition can be individually different, it is necessary to find quantitative criteria for the classification of delivery styles in order to reveal the characteristics of each delivery style and to improve specific (strength) training programs for each delivery style based on body segment positions and angles. Moreover, quantitative criteria will be important for pitchers who are not coached by highly experienced coaches, as it is difficult for less experienced coaches to classify pitches.

Escamilla et al.³ investigated professional major league baseball pitchers, and classified pitches into three delivery style groups (OS, TS, and SS) based on coaches’ visual identification. They introduced the arm slot angle (θ_AS), defined as the angle of the vector from shoulder to hand relative to the horizontal at release, and used it as a criterion to classify pitches into three delivery style groups (OS ≥ 50°, TS 30° − 40°, SS ≤ 20°). This left large buffer ranges in between styles (between 40° and 50° and between 20° and 30°), which served the authors’ purposes by ensuring that none of the pitches that they analyzed would be misclassified. However, it is not good for general purposes, as it leaves many pitches unclassified. Another possible idea for classifying pitches into delivery styles could be to use the combination of the positions of the upper trunk and upper arm relative to the GRS as a criterion. Atwater⁴ suggested that, regardless of the delivery style used, in most throwing skills the upper arm is abducted approximately 90° from the upper trunk at or near release, indicating that the trunk lateral tilt (θ_T) and upper arm elevation (θ_A) angles at release have a linear relationship. This association has never been quantitatively verified, but it presents the possibility that a combination of the upper trunk lateral tilt and upper arm elevation angles might serve as a good alternative to the slot angle as a criterion for the classification of baseball pitching styles.

Therefore, our purpose was to investigate what criterion could be the best for classifying pitches into the delivery styles. Escamilla et al.³ did not include a US style in their study. As the US style is not too rare in professional baseball, especially in Japan, we decided to include the US style in our project. In summary, this study aimed to find an appropriate criterion for quantitatively classifying baseball pitches into four delivery style groups, and to establish the values of this criterion that separated one style from another.

Methods

Participants

In total 70 male Japanese baseball pitchers (mean ± SD: age 19.3 ± 2.2 years; standing height 1.75 ± 0.05 m; body mass 72.0 ± 7.0 kg; throwing experience 11.4 ± 3.0 years), of whom 61 were right-handed and 9 were left-handed, and including 3 post-collegiate, 55 collegiate and 12 high school pitchers, participated in this study. All of them were healthy and had no history of arm surgery or present arm pain. Written informed consent was obtained from all participants after explaining the aims, risks of involvement, and experimental protocols of the study. Informed consent for underage players (less than 20 years of age) was obtained from their parents prior to the experiments. This study was approved by the institutional research ethics committee.

Experiment

The experiment was conducted on the pitching mound at a baseball stadium. After a warm-up that included throwing exercises, each pitcher was asked to throw 10 fastballs at maximum effort toward the catcher. Sufficient rest for full recovery, ranging from 1 to 2 minutes, was allowed between trials. Two participants each pitched with two styles, and one participant pitched with three styles, resulting in 74 analyzed pitches in total. All pitches were videotaped using two high-speed complementary metal oxide semiconductor (CMOS) video cameras (GC-LJ20B, JVC, Tokyo, Japan) at 240 Hz. An LED light (PH-145, DKH, Tokyo, Japan) was projected on the camera images for synchronization of the two cameras. Ball speed was recorded with a radar gun (PSK-DSP, Decatur Electronics, CA, USA). The fastest trial in which the ball was judged as a strike was selected for subsequent data processing and analysis.

Data processing

Two-dimensional coordinates of 21 body landmarks (vertex, gonion, suprasternale, both shoulders, elbows, wrists, knuckles, hips, knees, ankles, heels, and toes) and the ball center were manually digitized using a video motion analysis system (Frame-DIAS V, DKH, Tokyo, Japan) by five experienced researchers, and the digitization was finally checked and corrected by one highly experienced researcher. Image distortion due to the progressive downward scan of the CMOS cameras was corrected by taking into account the cameras’ blanking period.⁵ The three-dimensional (3D) body landmark and ball center coordinates were calculated using the direct linear transformation method,⁶ and then smoothed using quintic spline functions⁷ with optimal cutoff frequencies (range 4–24 Hz). The optimal cutoff frequencies were determined by residual analysis.^8,9 For the analysis, the 3D coordinate data of left-handed pitchers were subjected to a symmetrical transformation to make them appear as right-handed pitchers.

To classify the style of each pitch (OS, TS, SS, or US), the pitch was visually evaluated by four active and three former coaches (including four head coaches) from four university baseball teams using wireframe 3D graphics showing the view from the rear (second base). The coaches’ career experience was 3–21 years. Each pitch was classified separately by each coach; final classification was decided based on the highest number of votes. Out of the 74 trials, classifications by all seven coaches were matched for 33 pitches, classifications by six coaches were matched for 16 pitches, classifications by five coaches were matched for 14 pitches, and classifications by four coaches were matched for 11 pitches.

Data analysis

The arm slot of the throwing arm was defined as a vector pointing from the shoulder to the head of the third metacarpal; the trunk segment was defined as a vector pointing from the mid-point of the two hip joints to the suprasternale; and the upper arm segment of the throwing arm was defined as a vector pointing from the shoulder to the elbow, as shown in Figure 1. The GRS was set up. It was defined as a right-handed Cartesian coordinate system. The Y- and Z-axes of the GRS pointed in the forward (direction of throw) and vertical directions, and the X-axis was defined as the cross product of the Y- and Z-axes (Figure 1). Angle θ_AS was defined as the angle between the arm slot of the throwing arm and the horizontal in the XZ plane at ball release in accordance with Escamilla et al.’s study.³ Angle θ_T was defined as the angle between the trunk segment vector and the vertical in the XZ plane, and θ_A as the angle between the upper arm segment vector and the horizontal in the XZ plane (Figure 1).

Figure 1.

Definitions of the angles for arm slot (θ_AS), trunk lateral tilt (θ_T) and upper arm elevation (θ_A). The six indicated body landmarks were used as measurement points to calculate the three angles. A global reference system (GRS) was set up. It was defined as a right-handed Cartesian coordinate system. The Y- and Z-axes of the GRS pointed in the forward (direction of throw) and vertical directions, and the X-axis was defined as the cross product of the Y- and Z-axes. See the main text for details.

The values of the slot angle θ_AS were compared between delivery styles. When there was no overlap in the values of θ_AS between adjacent styles, the point of separation between the two styles was assigned to the average of the nearest values of θ_AS on either side of the gap. When there was an overlap, the point of separation was assigned to the mean of the largest and smallest values of θ_AS in the overlapping trials. Due to the overlaps, some of the points classified by the coaches as belonging to one style fell outside the range of that style as determined earlier. The number of these misclassified trials was counted.

For a classification method using θ_T and θ_A, θ_A was first plotted against θ_T for all pitches, and a regression line was calculated (Figure 2). Since a high positive correlation was obtained between the two values, each point was projected onto the regression line: y = 1.058 x − 0.164. Specifically, θʹ_T and θʹ_A that correspond to the coordinate values on the regression line were respectively obtained using the following equations¹⁰:

θ_{T}^{'} = \frac{b (b θ_{T} - a θ_{A}) - a c_{}}{a^{2} + b^{2}}

(1)

θ_{A}^{'} = \frac{a (- b θ_{T} + a θ_{A}) - b c_{}}{a^{2} + b^{2}}

(2)

where, a = 1.058, b = − 1.0 and c = − 0.164.

Figure 2.

Scattergram of the relationship between the trunk lateral tilt (θ_T) and upper arm elevation (θ_A) angles in the four delivery styles of pitching motion. The linear regression is indicated by the solid straight line. The broken straight line indicates theoretical positions corresponding to a right angle (90°) between the trunk and upper arm segments. The circles, plus, triangles and crosses indicate the overhand (OS), three-quarter (TS), sidearm (SS) and underhand (US) groups, respectively.

The θʹ_T and θʹ_A values of the projected points were then averaged to represent the two values as a single variable, and used for the classification. The points of separation between adjacent styles and the number of points of each style that fell outside their correct range were then computed using the same process followed for the slot angle calculations described above.

Statistical analysis

Means and standard deviations were calculated for all variables. Fleiss’ kappa coefficient (κ) was calculated to test the inter-rater reliability for the consistency of the seven coaches.¹¹ The ranges for the strength of agreement of the κ values were < 0.00 (poor); 0.00 − 0.20 (slight); 0.21 − 0.40 (fair); 0.41 − 0.60 (moderate); 0.61 − 0.80 (substantial) and 0.81 − 1.00 (almost perfect).¹² The relationship between θ_T and θ_A at ball release was quantified using linear regression analysis. A one-way analysis of variance (ANOVA, unpaired) was performed to assess differences between the four style groups for physical characteristics and kinematic variables. A post hoc Bonferroni correction was performed to examine the difference between each pair of delivery styles. Eta square (η²) was computed as a measure of effect size (ES) of ANOVA.¹³ Threshold values for the interpretation of the ES were 0.01 (small), 0.06 (medium), and 0.14 (large).¹³ Statistical significance level was set at p < 0.05 for each test. All statistical analyses were performed using IBM SPSS Statistics 28 (IBM, Armonk, NY, USA).

Results

The pitches were classified by the seven coaches as follows: 24 OS, 17 TS, 21 SS and 12 US. The κ value was 0.664 (substantial: p < 0.001, 95% confidence interval (CI95) = 0.634 − 0.694). No differences were found in age, standing height, body mass and years of throwing experience between the four groups. Ball velocities were larger in the OS, TS and SS groups than in the US group (Table 1). Significant differences were found in θ_AS, θ_T, and θ_A for every pair of the four groups. The θ_AS, θ_T, and θ_A values of the OS, TS, SS and US groups were the largest, second largest, third largest, and smallest, respectively (Table 1).

Table 1.

Comparison of the ball velocity, angles for arm slot (θ_AS), trunk lateral tilt (θ_T), and upper arm elevation (θ_A) among the four delivery styles of pitching motion.

Variables (units)	OS(n = 24)	TS(n = 17)	SS(n = 21)	US(n = 12)	F	η ²	CI95	Significant differences^†
Ball velocity (m/s)	35.4 (1.7)	35.5 (1.8)	34.2 (1.3)	32.0 (1.9)	13.87	0.373	0.177–0.496	*(c)(e), (f)
Arm slot angle (θ_AS) [deg]	53.9 (5.3)	31.5 (7.9)	5.8 (10.7)	−34.8 (16.3)	237.13	0.910	0.865–0.930	***(a)(b)(c)(d)(e)(f)
Trunk lateral tilt angle (θ_T) [deg]	31.9 (5.7)	15.4 (7.1)	−0.9 (11.5)	−40.1 (17.9)	130.23	0.848	0.773–0.882	***(a)(b)(c)(d)(e)(f)
Upper arm elevation angle (θ_A) [deg]	39.8 (7.3)	16.1 (9.5)	−6.3 (12.6)	−43.0 (15.1)	165.63	0.877	0.815–0.904	***(a)(b)(c)(d)(e)(f)

OS, overhand; SS, sidearm; TS, three-quarter; US, underhand delivery groups.

CI95 for η² indicates a 95% confidence interval (upper to lower limits).

^†

Significant differences for Bonferroni correction between (a) OS and TS, (b) OS and SS, (c) OS and US, (d) TS and SS, (e) TS and US, and (f) SS and US.

Significant differences for Bonferroni correction: *p < 0.05; **p < 0.01; ***p < 0.001.

The relationship between θ_A and θ_T showed a high coefficient of determination (R² = 0.839), and the slope of the regression line was 1.058 (Figure 2). The use of θ_AS as a variable for classifying delivery styles resulted in an overlap between TS and SS (14.9°–23.0°, 6 pitches). The boundaries between the four delivery style groups were 45°, 19°, and − 13°. This method correctly classified 24 out of 24 pitches as OS, 16 out of 17 pitches as TS, 19 out of 21 pitches as SS, and 12 out of 12 pitches as US (Figure 3(a)). The average of θʹ_T and θʹ_A as a criterion for classifying delivery styles resulted in overlaps of the ranges between OS and TS (21.0°–26.5°, 6 pitches), TS and SS (3.1°–10.5°, 7 pitches), and SS and US ( − 20.4° −19.1°, 2 pitches). The boundaries between the four delivery style groups were 24°, 7°, and − 20°. This method correctly classified 23 out of 24 pitches as OS, 14 out of 17 pitches as TS, 18 out of 21 pitches as SS and 11 out of 12 pitches as US (Figure 3(b)).

Figure 3.

Use of (a) the arm slot angle (θ_AS), and of (b) the combination of trunk lateral tilt (θ_T) and upper arm elevation (θ_A) angles as criteria for classifying delivery styles. The vertical lines indicate the separations between the four styles.

Discussion

This study defined four baseball pitching styles (OS, TS, SS, and US styles) based on observation of the players’ motions by seven highly experienced coaches, and provided normative biomechanical data for each one of the pitching styles. Although the two variables tested (θ_AS and the average of θʹ_T and θʹ_A) correctly classified delivery styles in general, the accuracy rate of the classification using θ_AS (96%) was higher than that using the average of θʹ_T and θʹ_A (89%).

As the inter-rater reliability examined using κ value was substantial, the pitches could be classified appropriately by the coaches’ judgments. That said, although all the pitches were classified into four delivery styles based on the coaches’ evaluations, the classifications for 11 pitches were matched by only four out of seven coaches. The reasons for the inconsistency are not clear, but these pitches that were matched by only four of the coaches were located between OS and TS and between TS and SS, and therefore pitches allocated into these boundary areas should be treated carefully. The average values of θ_AS, θ_T, and θ_A differed between every pair of groups (Table 1). Thus, all these variables have the potential to be biomechanical criteria. The value of θ_A was strongly associated with the value of θ_T, and the slope of the regression line had a value near 1 (slope = 1.058) (Figure 2). This indicates that θ_A changes essentially in unison with θ_T, and therefore baseball pitchers mainly determine their delivery styles through the lateral tilt of the trunk, confirming quantitatively what was suggested by Atwater.⁴ Thus, the average of θʹ_T and θʹ_A values that correspond to the projection of each point on the regression line could be used for distinguishing the delivery styles of baseball pitches.

Using the coaches’ judgments and θ_AS, the biomechanical criteria distinguishing the delivery styles were 45° (between OS and TS), 19° (between TS and SS) and −13° (between SS and US). Escamilla et al.³ reported ranges of θ_AS for OS (≥ 50°), TS (30° − 40°), and SS (≤ 20°). (They excluded pitches near the style boundaries, i.e. between 40° and 50° and between 20° and 30°.) Even though the participants in the two studies were different (professional major league baseball pitchers in the previous study versus Japanese amateur pitchers in the present study), the range of each delivery style was comparable between the studies (keeping in mind that Escamilla et al.³ did not include the US style). Using the coaches’ judgments and the combination of θ_T and θ_A, the biomechanical criteria distinguishing the delivery styles were 24° (between OS and TS), 7° (between TS and SS) and − 20° (between SS and US). Although both distinguishing methods correctly classified delivery styles in general, the accuracy rate of the classification using θ_AS (96%, 71/74 pitches) was better than that using the average of θʹ_T and θʹ_A (89%, 66/74 pitches). Moreover, the use of θ_AS produced fewer and smaller overlaps (only one overlap, of TS and SS, between 14.9° and 23.0°, involving 6 pitches) than the average of θʹ_T and θʹ_A. Another disadvantage of the use of the average of θʹ_T and θʹ_A was that different combinations of θ_T and θ_A could produce the same average value of θʹ_T and θʹ_A.

The remaining imperfections reflect difficulties experienced by the coaches in the classification of pitches. As a result, when θ_AS is between 14.9° and 23.0°, the pitch should be considered to be in a buffer zone, and its classification as TS or SS should be taken with caution.

Consequently, as summarized in Figure 4, the slot angle θ_AS and the aforementioned criterion angles seem to be the most appropriate way to classify baseball pitches into the four delivery styles in terms of biomechanical data. In turn, this should eventually help with the development of specific resistance training programs for each delivery style of the pitching motion.

Figure 4.

The ranges of the criterion angle that define each delivery style of pitching based on the arm slot angle (θ_AS). The numbers shown at the bottom indicate the ranges of θ_AS. The range between 23° and 15° can be considered a buffer zone between the three-quarter (TS) and sidearm (SS) styles.

Conclusion

The present study provided normative data for arm slot, upper trunk lateral tilt, and upper arm elevation angles of four delivery styles, and determined that the arm slot angle seems to be the most appropriate as variable for classifying baseball pitches into the four delivery styles in terms of biomechanical data. Moreover, the biomechanical criteria distinguishing the delivery styles using the arm slot angle are 45° (between OS and TS styles), 19° (between TS and SS styles) and − 13° (between SS and US styles). This information will be helpful to quantitatively separate baseball pitches into four delivery styles and to develop specific resistance training programs for each delivery style of the pitching motion.

Footnotes

Acknowledgments

The authors thank the players of the baseball teams of the Sendai University, University of Tsukuba, Kanazawa Seiryo University, Biwako Seikei Sport College, Keio University, Hitachi Ltd, East Japan Railway Company, Shibata High School, Hitachi First High School and their staffs for their cooperation. The authors also wish to thank Dr Jesús Dapena for his English proofreading and constructive suggestions.

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Ethical approval

This study was approved by the Institutional Research Ethics Committee.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the JSPS KAKENHI (grant number 16H03235).

Informed consent

Written informed consent was obtained from all participants after explaining the aims, risks of involvement, and experimental protocols of the study. Informed consent for underage players (less than 20 years of age) was obtained from their parents prior to the experiments.

ORCID iDs

Tomohisa Miyanishi

Ryu Nagahara

References

Alton

Weiskopf

. The complete baseball handbook: strategies and techniques for winning. Boston, MA: Allyn and Bacon, 1980, pp.87–153.

Nolan

Torre

. Pitching and hitting. Englewood Cliffs, NJ: Prentice-Hall, Inc., 1977, pp.13–36.

Escamilla

Slowik

Diffendaffer

, et al. Differences among overhand, 3-Quarter, and sidearm pitching biomechanics in professional baseball players. J Appl Biomech 2018; 34: 377–385.

Atwater

. Biomechanics of overarm throwing movements and of throwing injuries. In: Hutton

Miller

(eds) Exercise and sport sciences reviews 7. Philadelphia, PA: Franklin Institute Press, 1979, pp.43–85.

Miyanishi

. Measurement of the blanking period and correction of image-time distortion for the sports coaching CAM: problem of motion video analysis for CMOS camcorder using a rolling shutter. Bull Sendai Univ 2017; 48: 53–58.

Abdel-Aziz

Karara

. Direct linear transformation from comparator coordinates into object space coordinates in close-range photogrammetry. In: Proceedings of the American Society of Photogram Symposium on Close-Range Photogrammetric. Falls Church, VA: ASP, 1971, pp.1–18.

Woltring

. A Fortran package for generalized, cross-validatory spline smoothing and differentiation. Adv Eng Soft 1986; 8: 104–113.

Well

Winter

. Assessment of signal and noise in the kinematics of normal, pathological and sporting gait. In: Proceedings of the First Conference of the Canadian Society for Biomechanics, Human Locomotion I, Ontario, London: CSB, 1980, pp.92–93.

Winter

. Biomechanics and motor control of human movement. 2nd ed. New York: John Wiley & Sons, 1990, pp.27–45.

10.

Larson

Hostetler

. Precalculus: a concise course. Boston, MA: Houghton Mifflin, 2007, pp. 450–456.

11.

Fleiss

. Measuring nominal scale agreement among many raters. Psych Bull 1971; 76: 378–382.

12.

Landis

Koch

. The measurement of observer agreement for categorical data. Biometrics 1977; 33: 159–174.

13.

Cohen

. Statistical power analysis for the behavioral sciences. 2nd ed. Hillsdale, NJ: Lawrence Erlbaum Associates, 1988, pp.273–288.