*2.2. Procedure*

The matches were downloaded from the official channel of the World Padel Tour [26]. Lince video analysis software was used to collect and register the data [27]. The Kinovea software [28] was used to place a visual grid over the video image for the serve and return directions data recording (Figure 1) and to register the feet of the serving player when impacting the ball, the place where the ball bounced after a successful serve and the direction of the ball after the returner hit it. Four observers specializing in padel (over 5 years of experience) were specifically trained to perform the recording. Observers were specifically trained in the use of the observational instrument for two weeks. The training focused on the clear identification of the variables (serve and return statistics, point outcome and total shots) and the use of the observational instrument software (Lince and Kinovea). Having completed the training process, each observer registered a training set not included in the final sample (*n* = 72 serves; *n* = 68 returns), to calculate inter-rater reliability. Consistency of records was analyzed using the free-marginal multirater Kappa [29] and the weighted Kappa [30]. The minimum score obtained was k = 0.87. Finally, intra-observer evaluation was done at the end of the observation process by Cohen's Kappa calculation, yielding a very good strength of agreemen<sup>t</sup> with scores over 0.92 [31].

### *2.3. Data Analysis*

Descriptive analysis included means, standard deviations and frequencies. Assumptions of normality and homogeneity of variances were verified using the Kolmogorov–Smirnov test and Levene's test. Due to data not following a normal distribution, non-parametric tests were implemented [32]. Chi square analysis was performed to identify differences in serve and return statistics and point outcome between gender and court side. Column proportions were compared using Z tests on serve and return statistics according to the gender of the players and court side. A significance level of *p* < 0.05 was established which was adjusted according to Bonferroni in the Z tests. The associations among the categories of the variables were performed with corrected standardized residuals (CSR). The effect size was calculated using Cramer's V [33]. Rho Spearmen was used to know the relationship between serving point won and the number of strokes per point. IBM SPSS 25.0 Statistics for Macintosh (Armonk, NY: IBM Corp.) was used to process the data.
