Technical determinants of success in professional women's soccer: A wider range of variables reveals new insights.

Knowledge of optimal technical performance is used to determine match strategy and the design of training programs. Previous studies in men's soccer have identified certain technical characteristics that are related to success. These studies however, have relative limited sample sizes or limited ranges of performance indicators, which may have limited the analytical approaches that were used. Research in women's soccer and our understanding of optimal technical performance, is even more limited (n = 3). Therefore, the aim of this study was to identify technical determinants of match outcome in the women's game and to compare analytical approaches using a large sample size (n = 1390 team performances) and range of variables (n = 450). Three different analytical approaches (i.e. combinations of technical performance variables) were used, a data-driven approach, a rational approach and an approach based on the literature in men's soccer. Match outcome was modelled using variables from each analytical approach, using generalised linear modelling and decision trees. It was found that the rational and data-driven approaches outperformed the literature-driven approach in predicting match outcome. The strongest determinants of match outcome were; scoring first, intentional assists relative to the opponent, the percentage of shots on goal saved by the goalkeeper relative to the opponent, shots on goal relative to the opponent and the percentage of duels that are successful. Moreover the rational and data-driven approach achieved higher prediction accuracies than comparable studies about men's soccer.

Introduction

Professional women’s soccer is a relatively young sport compared to its men’s counterpart, with the first official World Cup held in 1991 [1]. Currently, it is experiencing rapid growth, both in numbers of participants [2-4] and professionalism [5]. Likewise, research attention is growing; however, this has mainly focused on the physical aspects of the game [6-8] and only a limited number of studies have investigated the technical side [9-11]. Understanding optimal technical performance enhances training programming and match strategy planning [12]. Of those that have examined technical performance, it has been shown that a quarter of goals are made from a cross [9], scoring first increases the likelihood of winning [10] and that free kicks within 7 m of the penalty circle should be taken as a direct free kick (shot on goal) to increase goal scoring [11]. Another study included some technical variables and found that duels both in offense and defence, pass accuracy and effective execution of dead-ball moments were significant contributors to victory [13]. These few studies give us the first insights into technical actions required to increase the likelihood of scoring or winning in women’s soccer. In men’s soccer, several studies have investigated technical performance related to success. It has been found that technical actions such as shots on goal [14-17], passes [18, 19] and assists [15, 17] were related to winning and these variables could also be important contributors to success in women’s soccer.

The identification of optimal technical performance in women’s soccer should use the best available analytical methods. This can be informed by an exploration of the strengths and limitations of analytical methods that have been used previously in soccer. Some previous studies may have been limited in sample size, as they were based on only one season or one tournament [9–11, 13, 14, 16, 17, 19–21], or a few seasons of a single tournament or league [15, 22]. While these sample sizes may not violate the statistical assumptions of the analytical methods that were used, they may limit the generalisability of their conclusions.

Some studies may also have been limited in the range of technical variables used in their analyses. The number of technical variables used in the limited studies in women’s soccer was relatively small (n = 1–9), even when combining the studies together [9-11]. Most previous studies in men’s soccer used 15–25 technical variables [14–17, 19–22] in their analysis and all except one [17] did not provide a clear rationale for the use of those particular variables. Previous research in Australian Rules football has shown that utilising a wide variety of variables gives more accurate match outcome models [23] and this could apply to modelling of match outcome in soccer as well. For some studies, the limited number of technical variables used may be ascribed to the number of variables that were available. For other studies, the number of variables used may also be a result of decisions based on the assumptions of statistical models chosen for analysis. The classification accuracy or predictive accuracy of models increases when the number of variables increases but so does the likelihood of overfitting [24]. Other than the variety of variables, it was found in Australian Rules football that variables in their relative form (i.e. difference in value to the opposing team in the same game) are better predictors than absolute values [23].

A further limitation of some studies of technical performance relates to the statistical models used. Some studies use statistical models that assume independence of observations [11, 13, 14, 20], which may not be met when observations from all teams appear repeatedly within a dataset. Moreover, to the best of our knowledge, none of the studies in the literature validated the performance of their modelling on a new data set or a part of the data set that is kept apart from the model development phase. This is a common practice in data mining and has also been suggested as an appropriate method to reduce overfitting [24]. Examples are splitting the dataset into a model training and test-set or using a 10-fold cross-validation.

Little is known about optimal technical performance in women’s soccer and the findings of research in men’s soccer probably has restricted transferability to women’s soccer. It is known that there is a difference in physical performance between genders [25, 26] which could also be the case for technical performance. The limitations of research regarding technical performance in soccer, taken with the limited transferability of findings from the men’s game to women’s game, indicate the scope for further research. Given the apparent increase in the professionalism of women’s soccer, there is a need to improve our knowledge of optimal technical performance and how it may differ from men’s soccer. Therefore, our aim was two-fold: understand the technical determinants of success in women’s soccer using a large sample size and a large number of variables, and identify analytical approaches that provide the most accurate and reliable information based on the prediction accuracy of the models.

Materials and methods

Data collection

Data was acquired from Opta Sports (London, United Kingdom) and covered the British Football Association Women's Super League (FAWSL) 2015, 2016 and 2017/18 seasons, the American National Women’s Soccer League (NWSL) 2016–2018 seasons, the 2013 and 2017 Union of European Football Associations (UEFA) Women’s Euros Championships (EC) and the 2011 and 2015 Fédération Internationale de Football Association (FIFA) Women’s World Cups (WC). This resulted in 695 matches and 1390 samples (sets of technical variable values per team per match) (see Table 1). Data consisted of team aggregates per match of all events that Opta Sports collects (i.e. 266 variables per team in each game). A detailed description of the events can be found on the Opta Sports website [27]. Empty variables were removed (n = 3), and match outcome (Win/Loss/Draw) was calculated based on the goals scored per team relative to the opponent. All analyses were done in R [28] and ethical exemption was received from the Deakin University Human Research Ethics Committee (2018–392) because of the use of pre-existing non-identifiable data.

Table 1

Distribution of matches per league or tournament.

League/Tournament	Team performances
EC 2013	50
EC 2017	62
FAWSL 2015	112
FAWSL 2016	144
FAWSL 2017–18	180
NWSL 2016	206
NWSL 2017	246
NWSL 2018	222
WC 2011	64
WC 2015	104
Total	1390

Analysis

Outlier detection

A search for outlying samples was done using a K-nearest neighbour outlier detection method [29] by adjustment of the do_knno()-function of the adamethods-package [30]. This identified 14 samples (7 matches) standing out from the rest of the data. Further investigation identified that all those matches were valid data points and it was decided to retain them in the dataset. Numerical variables were checked for normality, which was violated. An analysis for typographical errors within variables was done with the Adjusted Outlyingness, a method suitable for skewed data [31], using the adjbox()-function of the robustbase-package [32]. No errors were identified. In order to identify the relationship between technical performance and match outcome, 31 variables that were a function of score were removed, examples are: winning goal, goals from corners and goals conceded. All remaining variables that were not considered sample identifiers were expressed in absolute and relative forms (value relative to the opposing team in each match).

Outline of analysis process

Initially, a feature selection process was executed, followed by two phases of modelling and then a comparison of the results for the different models. During the modelling in each phase, three different approaches (i.e. sets of technical variables) were used to identify the relationship between technical performance and match outcome. This was to allow comparability with the previous literature, but also to determine whether newer approaches may provide better outcomes. The first approach was data-driven (DDA) and for the first phase, it was decided that all 450 variables were used for feature selection. For the second phase, 152 variables were included. The second approach was a rational approach (RA) in which two authors selected a broad range of variables (n = 156 in phase one and n = 74 in phase two) that were considered relevant to coaches. This was done to reduce overfitting and increase the practical application. In the third approach, variables were selected that had been used previously in the literature on men’s soccer (n = 43 in phase one and n = 16 in phase two) [14–17, 22] in order to make some form of comparability possible. Henceforth, this will be called the literature-driven approach (LDA). After a multicollinearity check and the first phase of modelling, a second phase of modelling was executed. For this second phase, all variables that were considered pseudo-score related were removed. Pseudo-score related in this context meant variables that will almost certainly precede a goal scored or by their definition mean that a goal was scored, such as ‘shots on goal’ and ‘scoring the first goal’. This step was intended to allow us to determine whether these variables are/not important in match outcome and then exclude them from further modelling. This allowed the identification of aspects of technical performance that might not have been identified if pseudo-score related were retained. This resulted in 152 variables being used for feature selection in the DDA, 74 in the RA and 16 in the LDA. A flowchart of the full analysis process can be found in Fig 1.

Fig 1

Flowchart of the analysis process.

Details of the techniques are explained in the text.

Feature selection

For all three approaches (DDA, RA and LDA), variables were checked for multicollinearity and selected/excluded using correlation-based feature selection [33] applying the findCorrelation-function [34] and a cut-off Pearson correlation coefficient of >0.90. For phase one, this left 367 variables in the DDA, 140 in the RA and 41 in the LDA and for phase two, 52 variables in the DDA, 74 in the RA and 16 in the LDA for further analysis. Thereafter, drawn matches (n = 292) were removed from further analysis since they can neither be considered clearly successful nor unsuccessful.

Following a wrapper method, recursive feature selection (RFS) using a random forest [35], was applied using the rfe-function and a 10-fold cross-validation (CV) repeated 5 times in the caret-package [34] to reduce the number of variables for the DDA and RA in both phases. This was done to avoid reduced model accuracy caused by a high number of variables in the modelling phase (i.e. too many variables risks adding noise to the model and or overfitting). In the first phase for both approaches, the RFS showed that the optimal number of variables was around 50 and therefore it was decided for each approach, to select the 50 variables with the highest score on the variable importance measure. The same analysis was done in the second phase, but this gave an optimal number of 30 for the DDA and 60 variables for the RA. Following this, the same procedure was applied to determine a variable importance measure for all variables in the LDA for both phases. The first 20 variables after feature selection for the three different approaches can be found in Table 2.

Table 2

The most valuable 20 technical variables identified in feature selection for the three analytical approaches.

Rank	Data-Driven Approach	Rational Approach	Literature-Driven Approach
1	First Goal	Relative Percentage Shots On Saved	Relative Shots On
2	Relative Intentional Assist	Relative Intentional Assist	Shots On
3	Assists	First Goal	Relative Total Shots
4	Relative Big Chances On Target	Percentage Shots On Saved	Relative Aerial Duels Won
5	Relative Shots On Conceded Inside Box	Relative Shots On From Inside Box	Total Shots
6	Intentional Assist	Intentional Assist	Relative Successful Crosses Corners
7	Relative Attempts Open Play On Target	Shots On	Total Clearances
8	Relative Big Chance Created	Relative Total Shots	Relative Total Passes
9	Relative Right Foot Shots On Target	Shots On From Inside Box	Relative Total Clearances
10	Shots On Conceded	Relative Saves Made	Relative Offsides
11	Shots On Conceded Inside Box	Relative Recoveries	Relative Corners Conceded
12	Big Chances Faced	Total Shots	Relative Unsuccessful Crosses Corners
13	Big Chances On Target	Relative Shots On Target Outside Box	Successful Crosses Corners
14	Shots On From Inside Box	Relative Second Assists	Corners Conceded
15	Relative Touches Open Play Opponent Box	Percentage Successful Duels	League Or Tournament
16	Relative Shooting Accuracy Right Foot	Relative Successful Dribbles	Total Passes
17	Relative Recoveries	Saves Made	Unsuccessful Crosses Corners
18	Relative Defensive Aerial Duels Won	Relative Percentage Total Successful Passes	Corners Taken
19	Total Shots Conceded	Relative Successful Crosses Corners	Aerial Duels Won
20	Relative Second Assists	Relative Aerial Duels Lost	Offsides

All variables were used in the first phase of analysis, which included variables that have a pseudo-relationship with scoring. The pseudo-score related variables (underlined) were removed in the second phase of the analysis.

Modelling

For the modelling phase, the samples were split into a training and testing set, with an 80:20 ratio. For each approach a binominal generalized linear model (GLM) with lasso penalized maximum likelihood [36] using the glmnet()-function of the caret-package [34] was utilised. In order to compare a partially interpretable model with an interpretable model, a decision tree (DT) was applied for each approach using the rpart()-function of the rpart-package [37] within caret. All modelling was done with a 10-fold CV repeated 10 times. Model performance was evaluated by determining their prediction accuracy using the testing data. Accuracies are presented on scale 0–1 with a 95% Confidence Interval (CI).

Results

The dataset provided by the RA in the first phase (i.e. inclusive of pseudo-score related variables) allowed the creation of the most accurate models from both the GLMs (0.99 CI 0.97–1.00) and the DTs (0.89 CI 0.84–0.92) (see Table 3). Moreover, of all 12 models, the GLM using the RA in the first phase is the most accurate. For all approaches, the models that retained pseudo-score related variables, outperformed models without. In both phases, the dataset provided by the LDA gives the lowest prediction accuracy, both for the GLM (phase 1: 0.78 CI 0.72–0.84; phase 2: 0.65 CI 0.58–0.71) and the DT (phase 1: 0.75 CI 0.68–0.80; phase 2 0.66 CI 0.59–0.72). The accuracies of the models based upon data from the LDA in the second phase, perform just slightly higher than a 50% chance of win/loss.

Table 3

The structure and performance of the generalised linear (GLM) and Decision Tree (DT) models in both phases of the analysis.

Phase 1	Data Driven Approach (DDA)		Rational Approach (RA)		Literature Driven Approach (LDA)
	Variable	Coefficient	Variable	Coefficient	Variable	Coefficient
GLM Predictors	First Goal	-1.47	First Goal	-1.24	Relative Shots On Goal	-0.20
	Relative Intentional Assist	-0.58	Relative Percentage Shots On Goal Saved	-0.95
	Relative Big Chances On Target	-0.07	Relative Intentional Assist	-0.49
	Relative Right Foot Shots on Target	-0.01	Relative Shots On From Inside Box	-0.06
	Relative Shots On Conceded Inside Box	0.01
GLM Accuracy (95% CI)	0.96 (0.93, 0.98)		0.99 (0.97, 1.00)		0.78 (0.72, 0.84)
DT Accuracy (95% CI)	0.88 (0.83, 0.92)		0.89 (0.84, 0.92)		0.75 (0.68, 0.80)
Phase 2
GLM Predictors	Relative Intentional Assist	0.75	Percentage Successful Duels	-2.01	Relative Aerial Duels Won	-0.05
	Assists	0.17	Relative Intentional Assist	-0.69	Relative Offsides	-0.04
	Relative Touches Open Play Opponent Box	-0.002	Intentional Assist	-0.24	Relative Total Passes	-0.001
			Relative Successful Dribbles	-0.02
			Relative Recoveries	-0.01
			Relative Aerial Duels Lost	0.003
GLM Accuracy (95% CI)	0.87 (0.82, 0.91)		0.78 (0.72, 0.83)		0.65 (0.58, 0.71)
DT Accuracy (95% CI)	0.76 (0.70, 0.82)		0.77 (0.70, 0.82)		0.66 (0.59, 0.72)

Phase 1 of the analysis included pseudo-score related variables, whereas in phase 2 they were excluded. There were also three analytical approaches that reflect different sets of variables that were used for modelling. Accuracies presented on scale 0–1 with 95% Confidence Interval (CI).

In the second analysis phase (i.e. where pseudo-score related variables were excluded), the DDA allowed the creation of a more accurate model GLM (0.87 CI 0.82–0.91), compared to both the RA (0.78 CI 0.72–0.83) and the LDA (0.65 CI 0.58–0.71) (Table 3). In general, the DTs did not perform as well as the GLMs, although there were combinations of approach and phase where they performed almost equally.

When looking at the specific variables that were included in the models in the first analysis phase, the most important technical action for winning a match in women’s soccer is scoring the first goal. With 20.3% of the matches won with 1–0 (141 out of 695 matches) and 55.3% of teams who scored a goal also winning their game (992 teams scored one or more goal, and 549 of those teams won their game). Following this, the next most important technical actions are intentional assists relative to the opponent and percentage of shots on goal saved by the goalkeeper relative to the opponent. The different models relied on different combinations of technical actions (e.g. relative big chances on target, relative right foot shots on target, relative shots on conceded inside box, relative percentage shots on goal saved and relative shots on from inside box) to explain match outcome. The LDA only had one significant contributor, namely shots on goal relative to the opponent, both in the GLM and the DT. For all models, variables that are expressed relative to their opponent are stronger contributors to match outcome than variables in their absolute form.

For the second phase of the analysis, in both the DDA and the RA, relative intentional assists are still strong contributors to the model. They are the first contributor in the data-driven approach GLM and DT, rational approach DT and the second in the rational approach GLM. Other important contributors to success in this second phase are intentional assists in their absolute form, normal assists and the percentage of successful duels. The coefficients of the literature-driven approach GLM are relatively small. Finally, for both phases in the LDA, intentional assists do not appear in the models since this variable has not been identified in the existing literature in men’s soccer and therefore was not included in the LDA. The DT for the DDA of the second phase is shown in Fig 2, and it can be seen that a value increment of one can make a difference.

Fig 2

Decision tree for the data-driven approach in the second phase.

The tree indicates the association between the values of the key performance indicators and the likely match outcome. This model predicted match outcome with a classification accuracy of 76%. Percentages mentioned are the proportion of the total dataset. Probability shown is the likelihood of a match with those conditions falling into the final category. The number on the branches of the tree indicates the cut off value for the technical action.

Discussion

This study aimed to identify technical determinants of match outcome in women’s soccer and to compare different analytical approaches. The main findings are that scoring first and having more assists than an opponent are very important in women’s soccer. When pseudo-score related variables are removed, technical actions such as successful duels, become important, which have not previously been identified in men’s soccer. In addition, this study highlights the value of using a larger number of games and variables in this type of analysis, as they tend to produce improvements in model accuracy and reveal novel aspects of technical performance.

The most important contributor to success in women’s soccer is scoring the first goal. It is not clear why this occurs but it may relate to the psychological impact on both teams. This aligns with the findings of previous research in women’s [10] and men’s soccer [38, 39]. Even though this might be hard to implement in training and is very dependent on the match situation, awareness of this fact is important to coaches. It could be included in a women-specific match strategy or incorporated in the tactical plan for after the first goal. Another important contributor to success in women’s soccer is intentional assists relative to the opponent. Intentional assist refers to a player making an intentional pass to a teammate who then makes a shot on goal, so there is no deflection of the pass before the shot [40]. This finding is novel and might be women-specific, as our models suggest that assists were more important than shots on goal. However, given that shots on goal precedes assists, it is difficult to definitively separate the importance of these two actions. Shots on goal did not appear in 11 of our 12 models whereas assists or intentional assists were a strong contributor in 8 of our 12 models. The number of goals saved by the goalkeeper relative to the opponent was also an important determinant of match outcome. Furthermore, to our knowledge, no previous study has presented evidence that the performance of the goalkeeper is an important determinant of match outcome.

The second phase of modelling excluded pseudo-score related variables that were considered to be difficult to train in a game-play setting. The analysis revealed that relative intentional assists are still very important. In addition, intentional assists, assists, percentage successful duels, relative aerial duels won, relative defensive aerial duels won, relative total passes and relative offsides were important contributors to match outcome. Interestingly most of those variables could be considered typical of an aggressive style of offensive and defensive play. These findings could imply that a more aggressive style of play is associated with winning matches in women’s soccer, which could be incorporated in the women-specific match strategy.

The literature-driven approach in the present study was designed to permit the comparison of men’s and women’s soccer. Previous studies in men’s soccer showed that passes [16-18], shots on goal [14–19, 41], and assists [15, 17] were important determinants of match outcome. Our literature driven approach also identified these variables as being important in women’s soccer, with the exception of assists. However, comparisons should be made with caution considering the lower prediction accuracy of the LDA models compared to the DDA and the RA models.

Similar studies have presented models of match outcome in soccer with classification accuracies of 69.5 to 76.9%. This includes models that work as a three-class problem (win, draw, loss) and models that use player-ratings [42]. The relatively high performance of the models (65–99% classification accuracy) in the present work, demonstrates the value of utilising a broad variety of variables when trying to understand the optimal technical characteristics of performance in women’s or men’s soccer. The present work may also demonstrate the value of training models on relatively large datasets and the use of variable values expressed as relative to the opponent.

This study has some limitations. We used a wide variety of data covering several seasons of two major leagues and two tournaments to increase generalisability to the larger population. However, it is possible that different leagues have different playing styles and that there is a difference in playing style between leagues and tournaments (although we have not assessed this). This could have increased the noise within the data and specific variables important for one league might not be important for another and therefore stay hidden. The initial selection of variables in the rational approach was done by the authors and one could argue that this should have been done by coaches. The feature selection method we used, may retain and remove different variables if used in the future on different data. Furthermore the collinear variables that were excluded from analysis should not be considered unimportant. Finally, our focus on technical performance should not eclipse the importance of other types of factors that affect match outcome (e.g. physical, tactical, etc.).

To conclude we found that the technical actions; scoring first, intentional assists relative to the opponent, the percentage of shots on goal saved by the goalkeeper relative to the opponent, shots on goal relative to the opponent and the percentage of duels that are successful, are the strongest determinants of success in professional women’s soccer. These findings will increase the ability for coaches to plan more women-specific match strategies and training plans. We also found that a wider range of variables selected for analysis increases the prediction accuracy of the models. It is important for analyst to collect enough data to create accurate models, but not too much to avoid overfitting.

Recommendations for future research in modelling match outcome based on technical performance in men’s soccer are to increase the number and variety of variables used, to include the use of variables in their relative form, to use a relatively large sample of matches, to apply more accurate modelling techniques and to report the accuracy of the method used. Future research should also make a direct comparison of the modelling of match outcome for both genders within the same study using data mining approaches. Currently, only comparisons between genders on technical performance exists using methods that look at the differences in absolute values of technical variables. The results of the two studies using this type of analysis could indicate a different playing style between the genders. It was found that women players were less accurate in their passes and had higher numbers of lost balls [43, 44], but also have a more attacking style of play. Which is a results of higher number of attacks, interceptions, recoveries and successful challenges and tackles [44]. The more attacking playing style is in line with the results of this study.

Our models included novel variables that provide new insights about optimal technical performance in women’s and possibly men’s soccer. While a wide range of variables is beneficial, it is also important to avoid redundant data collection in the applied sport science setting [45] and one way to achieve this is to evaluate the importance of variables collected. The feature selection provided an indication of what is and is not important, and the models themselves revealed the most important variables that should be the focus of coaching and feedback. Finally, our findings reinforce the idea that variables that are expressed in their relative form are more important than in their absolute form [23]. Performance analysts and coaches should evaluate the difficulties associated with applying technical performance indicators in their relative form (e.g. not seek to achieve 6 assists in a game, but to seek 2 more assists than their opponent).

17 Jul 2020

PONE-D-20-13014

Technical determinants of success in professional women’s soccer: A wider range of variables reveals new insights

PLOS ONE

Dear Dr. de Jong,

Thank you for your nice submision to PLOS ONE! After careful consideration, we feel that your manuscript has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

Both reviewers looked very favourably on your manuscript. They recommended some clarifications and some rewording. Please consider these comments and change the manuscript accordingly.

Both reviewers also indicated that te raw data for this paper was not available. Please make the raw data accessible, since this is one of the lublication criteria of PlosOne.

I'm looking forward to receiving your revisions.

Please submit your revised manuscript by Aug 31 2020 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols

We look forward to receiving your revised manuscript.

Kind regards,

Peter Andreas Federolf

Academic Editor

PLOS ONE

Journal Requirements:

When submitting your revision, we need you to address these additional requirements.

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at

https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and

https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

2.  We note that you have indicated that data from this study are available upon request. PLOS only allows data to be available upon request if there are legal or ethical restrictions on sharing data publicly. For information on unacceptable data access restrictions, please see http://journals.plos.org/plosone/s/data-availability#loc-unacceptable-data-access-restrictions.

In your revised cover letter, please address the following prompts:

a) If there are ethical or legal restrictions on sharing a de-identified data set, please explain them in detail (e.g., data contain potentially identifying or sensitive patient information) and who has imposed them (e.g., an ethics committee). Please also provide contact information for a data access committee, ethics committee, or other institutional body to which data requests may be sent.

b) If there are no restrictions, please upload the minimal anonymized data set necessary to replicate your study findings as either Supporting Information files or to a stable, public repository and provide us with the relevant URLs, DOIs, or accession numbers. Please see http://www.bmj.com/content/340/bmj.c181.long for guidelines on how to de-identify and prepare clinical data for publication. For a list of acceptable repositories, please see http://journals.plos.org/plosone/s/data-availability#loc-recommended-repositories.

We will update your Data Availability statement on your behalf to reflect the information you provide.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

Reviewer #2: Yes

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: Yes

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: No

Reviewer #2: No

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: This is an interesting paper assessing the determinants of match outcome in élite women football. The paper is fairly well structured and conceived.

The methodological approach is sound and nicely explained through informative pictures.

The discussion is relevant and provides interesting insights.

I’ve noticed just a list of minor issues / clarifications that should be fixed before acceptance:

ABSTRACT

rows 16-21: please summarise previous research more succinctly.

Line 21: tautology, “determinants” are “important” by definition.

Line 22: “analytical approached”, there’s something unclear in the sentence. Please define the sample size precisely (not “relatively large sample”).

Line 28: it is unclear what is being predicted.

The last sentence seems a repetition of what stated before.

INTRODUCTION

Line 41: in my opinion, as you’re talking about women football, it’s not necessary to highlight the drawbacks of studies on male players. Stating that such studies on women don’t exist could be a strong enough reason to justify the research.

Line 55: the classification “accuracy”, “power”, etc. A word should follow “classification”.

Lines 60-61 are unnecessary and can be removed.

Line 62: an “s” is missing after “relate”

Line 7: please add a space before “m” (meters)

Line 92: why “understand” has capital letter?

METHODS

I would call the first paragraph “data collection”

An explanation of the variables domain (it’s not feasible to describe every one of the 450 variables) would be strongly advised.

Line 105: so, if I get it right, you should have 266 variables per each team per each game, right? If so, this should be explained more clearly.

Line 105: “variables without values in them” awkward expression.

Line 107: data normality testing should be explained in the data analysis paragraph.

Line 109: there’s no need to explain what R is.

Please remove the “tables” heading.

Line 138: why did you keep 152 variables? Which was the criterion?

Line 159: how did you choose the correlated variable to remove?

Line 188: please remove “was utilised”

DISCUSSION

Line 255: is there anything related to women’s specific psychology in that? I mean, could it be that women when trailing by one goal, could more easily demoralised and struggle to score? Try to clarify / comment this point.

Line 266: just a comment, GK proficiency is probably even more important in women soccer for physical (stature) reasons. Thus, disposing of a tall, athletic and skilled keeper (i.e. one able to save a lot of shots) is a competitive advantage for women (relatively more than in men).

Line 321: that’s true. Why wasn’t it done asking coaches to select variables? Have the Authors themselves first-person experience in women’s football.

Line 328: please remove the comma after “analysis”.

The last sentence could be removed painless.

REFERENCES

Please double check references for consistency (ie. n. 6 but not limited to)

Reviewer #2: General Comments to Author

The authors used three different approaches to identify important technical features/variables in women’s soccer. The authors investigated a high number of variables and a large sample size, which is a major strength of the present manuscript.

Specific comments

Introduction

In the introduction authors focus primary on studies in men’s soccer and the corresponding limitations (lines 39-69). While the comparison of men’s soccer and women’s soccer is of additional interest, the primary focus of the present manuscript should be on studies in women’s soccer and the corresponding limitations (as stated in the aims of the study in lines 92-95). The reviewer suggests shifting the focus on studies in women’s soccer und to reformulate the corresponding sections.

Materials and Methods

Line 105: Please insert the number of variables that were removed (n=10?).

Line 140: At this point it is unclear to the reviewer how the variables relevant to coaches were chosen. As mentioned in the discussion section, variables were not chosen based on coaches feedback. Why?

Line 147: Please be more specific what a “nearly direct relationship” to goal scoring mean. For example, ‘shots on goal’ were excluded while ‘intentional assists’, which result in shots on goal, were excluded.

Lines 150: Please reformulate “more trainable aspects”, because it is not consistent with ‘shots on goal’ (line 148) for example.

Line 159: In the reviewer’s opinion, authors should check these excluded variables. For example, if the features selection process identifies an important variable, which has a high correlation (r>0.9) with one or more of the excluded variables, these excluded variables might also be important.

Line 168: Please be more specific with respect to “reduced model accuracy” because a higher number does not necessarily reduce model accuracy. For example, phase 1 resulted in higher accuracy than phase 2 where lower variables were included.

Line 174: Please specify “MA”, because it was not specified before.

Line 177: Table 2: A reader of the manuscript might benefit from a more detailed description of the listed variables. If possible, the reviewer would suggest providing a more detailed description as supplementary information.

Results

Line 197: Table 3: In Table 3 the listed variables correspond to the GLM model only. This should be indicated somewhere.

Line 203: Please reformulate “slightly higher than random choice”.

Line 206: Table 3 might be more appropriate than Table 2.

Line 208: Please insert almost or another alternative before equally since the models did not perform equally.

Lines 215-227: As noted above it should be noted that the specific variables correspond to the GLM models. Only in lines 235 to 237 specific variables corresponding to the DT approach are presented. Why did the authors decide to exclude a more detailed description of the specific variables of the DT approach in the manuscript?

Lines 231-232: A reader might be interested in the difference between “intentional assists” and “normal assists”. As mentioned above a more detailed description would be beneficial and should be provide as supplementary information, if possible.

Discussion

In the first paragraph should refer to the two objectives (aims) of the study stated at the end of the introduction (i.e., understanding the technical determinants of success in women’ soccer (a) and analytical approaches (b)). Although the comparison to men’s soccer is of interest, it should be discussed in a later paragraph.

Lines 261-264: Because an intentional assists results in a shot on goal, one might also infer that shots on goal preceding an intentional assist are more important than shots on goal in general. Please consider this.

Line 274: Please reformulate “offensive style of play”, because passes refer more to ball possession and (aerial) duels might refer more to the physical condition (i.e., sprinting and jumping abilities). These characteristic might be well addressed in training as well.

Line 290: Please consider that incorporating relative scores might also be important.

Lines 291-303: Please move this paragraph, because recommendations for future research are typically given at the end of the manuscript.

Lines 313-314: Please reformulate the example, because “seeking 2 more assists than their opponent” is not applicable.

Lines 317-319: Did the authors look at the results for a subset of the data to support this hypothesis?

Lines 320-322: See comment above (Line 140).

Lines 315-322: Please consider in the limitations also the following points. First, technical indicators are not the only determinants of success. For example, physical aspect, tactical decisions or psychological factors are also import. Second, please consider also the problem of stability of feature selection (i.e., in the presence of correlations).

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: Yes: Matteo Zago

Reviewer #2: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

2 Sep 2020

We would like to thank the reviewers for their time and effort, it has certainly strengthened the manuscript.

Submitted filename: Response to Reviewers.docx

Click here for additional data file.

22 Sep 2020

PONE-D-20-13014R1

Technical determinants of success in professional women’s soccer: A wider range of variables reveals new insights

PLOS ONE

Dear Dr. de Jong,

Thank you for submitting your manuscript to PLOS ONE.

The reviewers were satisfied with the changes to the manuscript, but one reviewer has some additional recommendations. I invite you to consider these comments.

Please submit your revised manuscript by Nov 06 2020 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols

We look forward to receiving your revised manuscript.

Kind regards,

Peter Andreas Federolf

Academic Editor

PLOS ONE

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #2: (No Response)

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #2: Yes

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #2: Yes

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #2: No

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #2: Yes

**********

6. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #2: Dear authors,

Thanks very much for addressing my comments (in detail). I only have one remaining remarks:

Introduction:

The authors still focus primary on studies in men's soccer and the corresponding limitations. The reviewer suggests again shifting the focus on studies in women's soccer and to reformulate the corresponding sections.

In particular, the second paragraph (lines 54-60) does not add important information because it focuses on men's soccer and the problem of limited sample size. The problem of limited sample size is already mentioned in the first paragraph (lines 51-53).

In the third paragraph, the authors focus on the technical variables used in previous studies in men’s soccer (i.e., lines 61-64). Why? The present reviewer recommends to focus on previous studies in women’s soccer and the corresponding technical variables.

In the fourth paragraph, the authors focus on statistical models in men's soccer (i.e., lines 74-75). As above, it is recommended to focus on previous studies in women’s soccer and the corresponding statistical models.

In the fifth paragraph, the restricted transferability of the results of studies in men's soccer (to women's soccer) is addressed. Here, further limitations of studies on men's soccer might be listed by the authors, if these are classified as important.

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #2: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

2 Oct 2020

We thank the reviewer for their final comments about the introduction.

Submitted filename: Response to Reviewer #2.docx

Click here for additional data file.

7 Oct 2020

Technical determinants of success in professional women’s soccer: A wider range of variables reveals new insights

PONE-D-20-13014R2

Dear Dr. de Jong,

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

Kind regards,

Peter Andreas Federolf

Academic Editor

PLOS ONE

Additional Editor Comments (optional):

Thank you for your submission. I accept your arguments for not changing some of the points that were suggested in the second round of review. With this email I have accepted your manuscript and forwarded it to for production.

Kind regards,

PF

13 Oct 2020

PONE-D-20-13014R2

Technical determinants of success in professional women’s soccer: A wider range of variables reveals new insights

Dear Dr. de Jong:

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

If we can help with anything else, please email us at plosone@plos.org.

Thank you for submitting your work to PLOS ONE and supporting open access.

Kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Dr. Peter Andreas Federolf

Academic Editor

PLOS ONE