Global landscape of sports statistics: Initial census of researchers and academics in the field

Background, history and growth of sports statistics

The field of sports statistics has evolved substantially from its origins to its current recognition and importance.^1,2 Reading through the biographies on the “History of Statistics” section of the American Statistical Association (ASA) Amstat News (https://magazine.amstat.org/statisticians-in-history/), a common, often overlooked element emerges: many prominent statisticians had a notable interest and involvement in sports. For instance, Richard L. Anderson (1915–2003), revered for his forward-thinking contributions to statistics included in his book “Statistical Theory Research”, harbored an unacknowledged enthusiasm for sports. He actively engaged in tennis and basketball, even recording basketball games in his spare time. Similarly, Samuel W. Greenhouse (1918–2000), renowned for his groundbreaking methods in clinical trials and epidemiology, was also an avid sports enthusiast. Monroe Sirken (1921–2017) and Joe Ward (1926–2011), known for their seminal work in survey methods and computational modeling, respectively, also held intriguing ties to sports. Sirken once shared an anecdote about his pursuit of UCLA membership to attend popular sporting events held there, highlighting his interest in sports beyond his statistical commitments. Ward grew up in an environment where sports were omnipresent, with his parents being coaches and educators in team sports. This upbringing perhaps instilled a similar passion in him, leading him to a career as a mathematics professor and basketball coach. Arnold Zellner (1927–2010), known primarily for his contributions to Bayesian statistics, had a deep passion for golf and participated in various other sports like baseball and soccer. These nuanced aspects of their lives underscore the lesser-known connection between their statistical eminence and their avid interest in sports.

This confluence raises an intriguing question: would these statisticians have explored the field of sports statistics more extensively if today's opportunities, such as access to vast datasets and advanced computational capabilities, had been available to them? It is likely that the combination of both factors, an abundance of data and enhanced computational tools, alongside the growing interest in sports analytics since the events popularized by Moneyball has driven the rapid expansion of this field. As Markus Stauff points out, sports have long contributed to the quantification and analysis of performance data, yet it was not until recent advances in technology that the field truly began to flourish. ³ This evolution has allowed sports statistics to move beyond simple data collection and to embrace more sophisticated analyses, giving rise to insights that were previously unattainable.

Additionally, several recent works illustrate the breadth of the field: from novel data visualizations for sports analytics ⁴ and speed calculations from tracking data, ⁵ to the analysis of tournament fairness and design^6,7 and retrospective studies of key match events such as penalty kicks. ⁸ By offering advanced statistical procedures alongside user-friendly data visualizations, sports analytics can provide entirely new insights into popular sports like baseball, basketball, football, and tennis.^9,10 This unique combination of scientific rigor and accessibility has helped solidify the role of statistics in modern sports, enhancing both professional performance evaluation and fan engagement. ¹⁰

Evolution and recent contributions to sports statistics

In recent years, the field of statistics has experienced substantial growth and evolution across various domains, including education, research, and industry applications.^11–13 This evolution is closely tied to the rise of data science, a field that heavily relies on advanced statistical methods and techniques, ¹¹ and in October 2012 the Harvard Business Review labelled data scientist as the “sexiest job of the twenty-first century” due to its high career prospects. ¹⁴ The field of statistics encompasses a wide range of specialties, including biostatistics, bioinformatics, socio-economic statistics, official statistics, ecological statistics, mathematical statistics, among many others. Recent advancements in statistical techniques have transformed decision-making processes across various fields, with data science studies gaining immense importance in diverse domains. ¹³ One of the burgeoning specializations is sports statistics, commonly known as Sports Analytics. ¹²

A key branch of sports analytics is sabermetrics, which has been defined as “the search for objective knowledge about baseball”.^15,16 Although originally focused on baseball, sabermetrics marked the beginning of a broader trend towards the use of statistical and data-driven approaches across sports. To better understand the role of sports analytics today, it is essential to frame it as a multidisciplinary field that includes tactical analysis, performance optimization, and game outcome predictions. Sports analytics can be broadly described as “the process of data management, predictive model implementation, and the use of information systems for decision-making to gain a competitive advantage on the field of play”. ¹⁷ This field encompasses a wide range of applications, such as evaluating players, optimizing game strategies, and developing player and team rankings. Beyond performance analysis, sports analytics is increasingly applied in areas like health monitoring, injury prevention, and mental condition assessments, which highlight its growing role in sports medicine and athlete welfare.

Building on the foundational work of sabermetrics, sports analytics has expanded to include various approaches that combine statistical methods, machine learning, and data visualization. Jeremy Abramson ¹⁸ described sports analytics as “the discovery and communication of meaningful patterns in data”, emphasizing its role in tactical decision-making. Similarly, Bill Gerard ¹⁸ highlighted its application in analyzing in-game patterns, further underscoring the importance of this field for both team and individual sports contexts.

In the field of health, particularly in sports and exercise medicine research, additional specialized domains are gaining prominence. Sports biostatistics focuses on the use of statistical methods to understand issues such as performance and injury prevention.^19,20 Sports epidemiologists play a critical role in studying the distribution and determinants of health outcomes in athletes. ²¹ These professionals, along with meta-researchers, are driving the push for more rigorous and transparent research practices in sports medicine. Recent studies^22,23 have underscored the need for open science practices to ensure the robustness and reproducibility of findings in sports medicine and orthopaedic research. These advancements reflect the broader trend toward greater scientific rigor and transparency in the study of sports and health, where evidence-based approaches are increasingly essential for improving both performance and injury prevention strategies.

An early support for this new field of sports statistics has come from the American Statistical Association (ASA). In the 1992 Joint Statistical Meetings, the Statistics in Sport Section (SiS) was established under the ASA's umbrella, aiming to foster statistical developments and their applications within sports. Over time, various contributions have elevated the significance of sports statistics, including portrayals of statistical concepts in media. Notable examples are the well-known movie “Moneyball” ²⁴ and book publications like “Analyzing Baseball Data with R”.^1,25 Recent developments have seen the emergence of specialized consultancies focused on statistical and computational analysis dedicated specifically to the sports industry, exemplified by Zelus Analytics, founded by Luke Bornn and Doug Fearing in 2019 and later integrated into Teamworks Intelligence (https://teamworks.com/intelligence/). Monthly webinars organized by the S-Training group (Sports – Training and Research in DAta ScIence Methods for ANalytics and INjury Prevention, [https://s-training.eu/Homepage.html]) since September 2020 further solidify the importance of sports statistics within the scientific community, addressing key challenges such as injury prevention and performance optimization.

Building a community: Publications, networks, and the case for a census

Census creation process

Survey as an information retrieval tool

Building upon the core census variables previously defined (Table S1), a structured survey was designed to enrich the dataset with additional variables not publicly available (Table S2) (see Section 1.4 of the Supplementary Material). The survey consisted of seven closed and open-ended questions and was distributed via Qualtrics to all researchers identified in the census process.

The survey was administered in successive contact phases between March 2023 and February 2026, including follow-up reminders to maximize participation. In total, 274 researchers were invited to participate, yielding 134 responses (overall response rate: 48.91%).

The survey was intended to enrich the dataset with additional contextual and career-related variables; inclusion in the census itself did not depend on survey participation.

The survey design and implementation were conducted by three of the authors (MO, MC, and DF), with personalized contact initiated by MC to encourage participation. Participants were assured confidentiality, and all results are reported in aggregate form. Data were stored securely on password-protected and encrypted devices accessible only to the research team.

Data processing and final census selection

To comprehensively depict the entire process, we followed the PRISMA statement (Preferred Reporting Items for Systematic Reviews and Meta-Analyses),^28,29 which includes a flow chart that describes the workflow of the whole review process of the census creation (Figure 1).

OPEN IN VIEWER

Figure 1.

PRISMA flow diagram for obtaining the final census and statistical analysis dataset for clustering.

The PRISMA flowchart in Figure 1 outlines the selection process, illustrating the refinement of the census. Initially, we identified 327 researchers, but 53 were excluded for not meeting the criteria, leaving 274.

After sending the survey to 274 individuals, we received 134 responses (48.91%). Among the 17 doubtful cases, 11 responded. Based on their answers, we confirmed that 10 of these 11 should be included in the census, while 1 was excluded for not considering themselves an academic. The remaining 6 individuals, who did not respond, were excluded from the census. Additionally, 7 respondents, initially confirmed as part of the census, stated they did not consider themselves academics, and 1 indicated they did not specialize in sports statistics. Consequently, a total of 15 researchers were excluded from the final census: 1 doubtful case who did not consider themselves an academic, 6 doubtful cases who did not respond, 7 profiles initially included but not considering themselves academics, and 1 profile not specializing in sports statistics. The final census was reduced from 274 to 259 researchers.

The analysis database consisted only of researchers who responded to the initial survey (n = 134). To avoid missing values, researchers without a Google Scholar profile were excluded, totaling 15 individuals. It is important to note that these 15 researchers were not removed from the census, as they met the initial criteria, but were excluded only from the analysis database. Additionally, 1 individual reported not being involved in sports statistics, and 8 indicated they did not consider themselves academics. As a result, 24 researchers (15 without Google Scholar profiles, 1 not involved in sports statistics, and 8 not considering themselves academics) were excluded from the analysis database, reducing the database from 134 entries to 110.

Statistical analysis

Absolute (n) and relative (%) frequencies were computed for categorical variables, while measures of central tendency and dispersion were calculated for the continuous variables. At multivariate level, the semiparametric clustering method KAMILA (KAy-means for MIxed LArge data) ³⁰ was used because of the presence of mixed-type variables. Clustering mixed-type data, including datasets containing both continuous and categorical variables, is a non-trivial problem because most classical clustering algorithms, such as k-means, are designed exclusively for continuous data and rely on Euclidean distance, which is not meaningful for categorical variables. A common but suboptimal approach is to preprocess categorical variables by dummy coding and then apply standard methods, which distorts the geometry of the feature space and can bias the clustering solution.

A more principled alternative is KAMILA, the method adopted in this work. KAMILA handles mixed-type data natively by modelling continuous variables using a kernel density estimator evaluated at cluster centroids, and categorical variables using a multinomial distribution, combining both within an Expectation-Maximization (EM)-type iterative algorithm. This semiparametric approach does not require dummy coding, avoids the scale-sensitivity issues of purely distance-based methods, and has been shown to scale well to large datasets. The number of clusters is selected by maximising a penalised likelihood criterion computed over multiple random initialisations, which reduces sensitivity to local optima. This technique was applied through the kamila R package. ³¹ The cluster tendency was assessed using the Hopkins statistic. ³² The optimal number of clusters was determined according to the prediction strength method. ³³ See Supplementary material for more details.

The compareGroups package ³⁴ was used to describe cluster profiling. For categorical variables, Pearson's Chi-squared tests were employed, with two-sided Fisher's exact tests when expected frequencies per category were less than five. The Shapiro-Wilks test had revealed that the continuous numerical variables do not follow a normal distribution, and hence these variables were analyzed using the nonparametric Mann-Whitney U-test. A two-sided p-value < 0.05 was considered statistically significant in the between-cluster comparisons.

All analyses were conducted using R statistical software version 4.1.3. ³⁵ The dataset and R code used for the statistical analyses are available at https://github.com/marticasals/Census_Sports_Statisticians.

Demographic and research characteristics of selected sports statisticians

Figure 2 shows a world map with the initial geographical distribution of the 110 survey responders. Countries with presence of researchers are highlighted in blue, including Canada, the United States, parts of Western Europe, South Africa, Argentina, China, India, Japan, and Australia, while large portions of Africa, South America, Southeast Asia, and Eastern Europe are not represented (gray).

OPEN IN VIEWER

Figure 2.

Global map representing researchers that completed the survey.

Among these 110 sports statisticians, the majority are male (87.3%) and in an age range from 30 to 44 years (48.2%). Researchers from the United States represent the highest percentage (20.9%), followed by those from Italy (15.5%). Most researchers are affiliated with public universities (77.3%). In terms of educational studies, nearly 80% have more than five years of teaching experience, and 90% are fully dedicated to academia. Additionally, a non-negligible number of researchers are not affiliated with any specific research group (58.2%). Several sports could be studied by a single researcher. Individually, soccer remains the most researched sport (18.2%), followed by basketball (9.1%). For the number of citations from Google Scholar, there is a notable disparity between the first (P25) and the third quartile (P75), indicating a high asymmetry in this variable. Table 1 shows full details about these variables.

OPEN IN VIEWER

Table 1.

Summary of the variables included in the analysis dataset.

	Category			n (%)
Gender	Male			96 (87.27%)
	Female			13 (11.82%)
	Prefer not to say			1 (0.91%)
Age group	18–29			7 (6.36%)
	30–44			53 (48.18%)
	45–59			34 (30.91%)
	60 +			16 (14.55%)
Country	United States			23 (20.91%)
	Italy			17 (15.45%)
	Spain			13 (11.82%)
	Australia			7 (6.36%)
	Other*			50 (45.45%)
Type of university/center	Public			85 (77.27%)
	Private			25 (22.73%)
Highest level of education	Doctorate or PhD			105 (95.45%)
	Master's degree			5 (4.55%)
Years of experience in sports statistics	5 years or more			84 (76.36%)
	3–4 years			20 (18.18%)
	1–2 years			4 (3.64%)
	Less than 1 year			2 (1.82%)
Profile	Full-time academic of sports statistics/analytics field.			99 (90.00%)
	Part-time sports statistician in academia and part-time in the sports industry.			6 (5.45%)
	Academic sports statistician with years of expertise and contributions, but currently working in the sports industry.			5 (4.55%)
Belonging to a sports statistics research group?	No			64 (58.18%)
	Yes			46 (41.82%)
Main Sport(s) of Research	Soccer			20 (18.18%)
	Basketball			10 (9.09%)
	Basketball, Soccer			5 (4.55%)
	Running			5 (4.55%)
	Baseball			4 (3.64%)
	Soccer, Tennis			4 (3.64%)
	American Football			3 (2.72%)
	Basketball, Baseball, American Football			3 (2.72%)
	Cricket			3 (2.72%)
	Basketball, Soccer, American Football			2 (1.82%)
	Tennis			2 (1.82%)
	Other**			49 (44.55%)
Numeric variables (n = 110)		P25	Median (P50)	P75
Citations in GS	Overall	577	1407	3544
	Since 2021	353	794	1733
H-index	Overall	11.25	18	29
	Since 2021	9.25	14	19
Number of citations of the most cited article		103.5	209.5	472.8
Number of sports studied		1	2	3

* Belgium, Canada, Czech Republic, Denmark, France, Germany, Greece, Hungary, India, Ireland, Japan, Luxembourg, Netherlands, Norway, Sweden, Switzerland, United Kingdom.

** Frequency equal to 1 among many other sports.

P25: 25th percentile; P50: 50th percentile (median); P75: 75th percentile.

Clusters of census participants

After conducting an exploratory analysis, we applied the KAMILA algorithm for clustering. Initially, the prediction strength method suggested an optimal clustering solution for two clusters (Figure S1 in Supplementary material). However, during the execution of the algorithm, we identified three outlier records, i.e., researchers with extremely high Google Scholar citation counts, that at the end formed a third distinct cluster: while the prediction strength method recommended two clusters, we temporarily assigned the three outlier researchers to a separate cluster. This allowed us to re-run the KAMILA algorithm on the remaining dataset.

The results of the initial clustering, including the outlier cluster, are shown on the left panel of Figure 3. The refined clustering results, obtained after removing the outlier cluster and re-running the KAMILA algorithm, are displayed on the right panel of Figure 3. Note that the overlap between clusters in the right panel may be due to clustering after dimensionality reduction via principal component analysis (PCA). To visually assess the robustness of this clustering solution, the KAMILA algorithm was additionally applied after outlier removal, forcing k = 3 groups on the dataset. The results confirmed a high degree of stability of the two main clusters: the clusters obtained with k = 3 (clusters 2 and 3) closely correspond to the two clusters reported here, while the third cluster represents an approximate further partition within one of them.

OPEN IN VIEWER

Figure 3.

KAMILA algorithm results for the full dataset (left panel) and without outliers (right panel). Clustering was performed after dimensionality reduction using PCA.

The refined analysis identifies three clusters: one consisting of three researchers with exceptionally high citation counts, and the two additional clusters depicted in Figure 3. The next section will detail the profiling of researchers within each cluster.

Cluster profiling

Table 2 shows the comparison between the two main clusters resulting from the analysis.

OPEN IN VIEWER

Table 2.

Comparison of profiles across the two larger clusters.

Variable		Cluster 1 (N = 48)	Cluster 2 (N = 59)	OverallP-value
Age group				< 0.001
	18–29	7 (14.6%)	-
	30–44	29 (60.4%)	24 (40.7%)
	45–59	9 (18.8%)	25 (42.4%)
	60+	3 (6.25%)	10 (16.9%)
Country categorized				< 0.001
	Australia	1 (2.08%)	4 (6.78%)
	Italy	5 (10.4%)	12 (20.3%)
	Other	12 (25.0%)	38 (64.4%)
	Spain	11 (22.9%)	2 (3.39%)
	United States	19 (39.6%)	3 (5.08%)
Type				< 0.001
	Private	22 (45.8%)	3 (5.08%)
	Public	26 (54.2%)	56 (94.9%)
Years of experience in sports statistics				< 0.001
	Less than 1 year	2 (3.39%)	-
	1–2	4 (1.68%)	-
	3–4	20 (28.8%)	-
	5 years or more	22 (66.1%)	59 (100%)
Citations in GSMedian [P25 ; P75]	Overall	870 [349 ; 1536]	2203 [834 ; 6263]	< 0.001
	Since 2021	575 [282 ; 968]	1129 [522 ; 2182]	< 0.001
H-indexMedian [P25 ; P75]	Overall	14 [10 ; 18]	24 [14 ; 34.5]	< 0.001
	Since 2021	11.5 [8 ; 16]	17 [12 ; 24]	< 0.001
Number of citations of the most cited articleMedian [P25 ; P75]		145 [77.8 ; 338]	254 [124 ; 621]	0.009

Data are presented as mean (SD) or n (%). P-values were calculated using the Mann-Whitney U test for numeric variables and the chi-square test for categorical variables.

P25: 25th percentile; P50: 50th percentile (median); P75: 75th percentile.

Regarding the country, we analyzed the distribution of the researchers ( Table S3 in Supplementary material) stratified by their experience in sports statistics. Most countries show a high percentage in the category of ‘5 years or more’; however, Spain presents a remarkable lower percentage. Table 3 summarizes the key characteristics of each cluster by means of representative profiles to highlight their distinct attributes.

OPEN IN VIEWER

Table 3.

Summary of key characteristics for each cluster. Qualitative characteristics are described according to the more prevalent categories in each cluster.

Characteristics	Cluster 1	Cluster 2	Cluster 3
Age range	18–29 years and 30–44 years	45–59 years and 60 + years	60 + years
Country	Spain and United States	Countries of low frequency and Italy	Australia
Type of University	Private	Public	Public
Years of experience in sports statistics	Less than 5 years	5 years or more	5 years or more
Citations according to Google Scholar	Low number of citations (median citations by researcher: 870)	High number of citations (median citations by researcher: 2203)	Extremely high number of citations (median citations by researcher: 63,788)
Number of researchers	48	59	3
Examples	Michael Lopez (https://statsbylopez.netlify.app/)	Jim Albert (https://bayesball.github.io/)	William Hopkins (https://www.sportsci.org/team/will.html)
	Benjamin Baumer (https://www.smith.edu/people/ben-baumer)	Marica Manisera(https://www.unibs.it/it/ugov/person/3750)	Caroline Finch (https://orcid.org/0000-0003-1711-1930)