The Superstar Effect in Tennis—A Within-Match Analysis

Abstract

This study examines how superstars affect players’ situational performance in professional men's tennis. We use match-level data from 27,366 main draw Association of Tennis Professionals (ATP) singles matches between the 2004 and 2019 seasons. Estimating a high-dimensional fixed effects model, we use ATP serve and return ratings to proxy performance in dominant (serve) and nondominant (return) situations of both higher-ranked (HR) and lower-ranked (LR) players. We provide evidence that HR players deliberately increase/decrease their performance in (non)dominant match situations based on their rank and the timing of facing a superstar in subsequent matches. Similarly, there are differences in the extent of performance shifts induced by superstars among different rank groups for LR players; however, the differences do not extend to different within-match situations.

Introduction

Tournaments typically refer to competitive events in sports where a considerable number of players participate in a series of matches to determine a winner. In some tournaments, the participation of individuals who stand out because of their exceptional abilities may influence the performance of their peers (MacDonald, 1988; Rosen, 1981). This phenomenon is commonly referred to as the “superstar effect” in the economic literature.

The sport of tennis provides an ideal context for studying this phenomenon. Most professional tournaments adopt a knockout format in which the winner advances to the next round while the loser is eliminated. In the related literature, a key reference is Deutscher et al. (2023), who examine this tournament setting and document that superstar presence is associated with changes in match-winning probabilities. However, relying solely on a binary match outcome provides only a limited indication of player performance. A win or loss captures the final result but fails to reflect the nuances of how players perform across different situations within a match. For example, a player may lose but excel in specific aspects of play, or win despite weaker performance in certain situations.

To gain a more nuanced understanding of how the superstar effect shapes player behavior, alternative performance measures are required. The organizational structure of tennis inherently divides players into two alternating roles: serving and returning. These roles represent distinct match situations, the dominant (serving) and nondominant (returning) phases of play. Evaluating performance across these situations provides a natural way to examine how the superstar effect manifests beyond the simple probability of winning.

Understanding the superstar effect in situational performance also carries practical relevance for various stakeholders. Coaches and sports psychologists can tailor training and mental preparation strategies to address the differing demands of serving and returning. Tournament officials may use such insights to structure draws and scheduling in ways that preserve competitive balance and spectator engagement. Players themselves can benefit from recognizing how their performance dynamics change when facing or anticipating a superstar opponent, thereby improving strategic and psychological preparation.

Building on these practical considerations, this study investigates the superstar effect on professional men's tennis performance across within-match situations. Using data from 2004 to 2019, we extend the tournament-based superstar framework of Deutscher et al. (2023) along two dimensions. First, our study moves beyond match-level win/loss outcomes by examining within-match situational performance, distinguishing dominant and nondominant situations using Association of Tennis Professionals (ATP) serve and return ratings. Second, we allow the superstar effect to vary by pre-match status by estimating heterogeneous effects for higher-ranked (HR) and lower-ranked (LR) players.

Literature Review

The concept of superstars was first introduced by Rosen (1981) and extended by MacDonald (1988). The authors describe how a small number of exceptional individuals can capture a disproportionate share of rewards and recognition, thereby shaping incentives and outcomes within competitive environments. In the sports economics literature, the superstar effect has been analyzed in two distinct contexts. (1) In market demand context, the presence of a superstar may influence, for example, attendance (Brandes et al., 2008; Humphreys & Johnson, 2020; Mullin & Dunn, 2002; Ormiston, 2014) and media attention (Buraimo & Simmons, 2015; Hausman & Leonard, 1997; Konjer et al., 2017; Schreyer & Torgler, 2018). (2) In sport performance contexts, the superstar effect refers to changes in competitors’ performance (Bilen & Matros, 2023; Bryson et al., 2014; Coates et al., 2016) or risk-taking behavior induced by the presence of a dominant player (Lackner, 2023; McFall & Rotthoff, 2020; Meissner et al., 2021). We analyze the superstar effect on the performance side and define a superstar as an athlete whose sustained superiority, visibility, and reputation make them a salient and psychologically influential opponent. Accordingly, the superstar effect in the following refers to how the presence of such a player influences the performance of their opponents.

In individual sports, Brown (2011) finds that Tiger Woods’ participation reduced other top golfers’ performance, while Tanaka and Ishino (2012) and Meissner et al. (2021) document similar negative effects in golf and gymnastics, respectively. Conversely, Hill (2014) reports a positive influence of Usain Bolt on fellow sprinters, and Babington et al. (2020) find that underdog skiers perform better when a superstar is present. Lackner (2023) provides complementary evidence from professional basketball, showing that effort and risk-taking vary with the degree of superstar dominance. Effort declines when dominance is overwhelming but improves when competition is more balanced, underscoring that the strength of dominance shapes behavioral responses, a mechanism likely to extend to individual tournaments.

In the context of professional tennis, Deutscher et al. (2023) present evidence that superstar presence is associated with lower winning probabilities for top players. However, McFall and Rotthoff (2020) note that analyses relying mainly on match-level outcomes may miss important within-match variation and therefore encourage the use of more detailed performance measures. Their discussion highlights the value of indicators that capture in-game dynamics and potential behavioral adjustments when a superstar is present.

In response, we move beyond match outcomes to within-match analysis to examine whether the estimated superstar effects differ between HR and LR players. More broadly, our study connects to the literature on dominance in professional sports (Brown, 2011; Lackner, 2023) and to theories of strategic effort in tournaments (Ely et al., 2017; Jane, 2015; Lallemand et al., 2008; Lazear & Rosen, 1981; Rosen, 1986; Sunde, 2009; Walker & Wooders, 2001). In addition, the paper contributes to the tennis economics literature on player performance and decision-making (Brown & Minor, 2014; Klaassen & Magnus, 2001).

Data

To collect data pertaining to tournaments, matches, and individual players, we use a web-scraping algorithm. First, our algorithm retrieves basic information on ATP singles tournaments for the 2004–2019 seasons from www.atptour.com. Second, by looping over each tournament in the provided seasons, we systematically obtain match information such as a unique player identifier, the match-level performance ratings of both contestants (HR and LR), the match round, and the final score. Third, we utilize the player identifiers to collect contestant-specific information, for example, a player's current world ranking position and country of origin. We link each match with both participants using the individual player identifiers. This way, we retrieve detailed information on 48,882 matches.

The data cleaning process follows a two-step procedure. First, we eliminate matches with unsuitable tournament structure. (1) Team and doubles competitions (World Team Championship and ATP Cup) are excluded because it is not possible to identify HR and LR players at the match level. (2) Events that do not generate world ranking points (Olympics, Laver Cup, and Next Gen ATP Finals) are not considered, as such tournaments are only accessible to a selected group of players. (3) We discard 4,384 matches in which a player received a bye and exclude 1,728 matches due to irregularities in the outcome (e.g., injury, disqualification, or walkover). (4) Detailed match information or player ranks are unavailable for 1,711 matches. (5) As they deviate from the conventional elimination structure of tennis tournaments (Gilsdorf & Sukhatme, 2008), we exclude round-robin competitions (216 matches), such as the ATP World Tour Finals.

Second, we follow the data cleaning instruction by Deutscher et al. (2023). (1) 3,295 matches involving tennis superstars during their dominant seasons are omitted (Novak Djokovic, 2007–2019; Roger Federer, 2004–2019; Rafael Nadal, 2005–2019; Andy Murray, 2008–2016). (2) The Grand Slams (Australian Open, French Open, Wimbledon, and U.S. Open) and the two annual ATP Masters 1000 events in Indian Wells and Miami (9,236 matches) are excluded due to concerns regarding self-selection and structural differences in the competitions. (3) Since it is not possible to examine whether a superstar impacts a player's performance in the next round if the match is a final or if all the superstars are eliminated from the tournament, we discard 943 matches. The final dataset features 936 tournaments on the ATP World Tour, resulting in a total of 27,366 observations identified at the match level (2004–2019). Each observation contains information on competing HR and LR players. A total of 920 players are included in the dataset. Table 1 presents a comprehensive summary of the selected variables in our dataset.

Table 1.

Variable Description and Summary Statistics.

Variable	Description	Mean	SD	Min	Max
HR: won	Higher-ranked player won the match (1 = yes; 0 = no)	0.63		0	1
HR: serve	Higher-ranked player serve rating	272.23	36.71	90	392
HR: return	Higher-ranked player return rating	149.42	51.93	6	343
LR: serve	Lower-ranked player serve rating	255.17	39.24	86	384
LR: return	Lower-ranked player return rating	128.56	52.34	5	356
HR: 1–20	Higher-ranked player ranked in positions 1–20 (1 = yes; 0 = no)	0.29		0	1
HR: 21–50	Higher-ranked player ranked in positions 21–50 (1 = yes; 0 = no)	0.38		0	1
HR: > 50	Higher-ranked player ranked in positions > 50 (1 = yes; 0 = no)	0.34		0	1
LR: 1–20	Lower-ranked player ranked in positions 1–20 (1 = yes; 0 = no)	0.02		0	1
LR: 21–50	Lower-ranked player ranked in positions 21–50 (1 = yes; 0 = no)	0.19		0	1
LR: > 50	Lower-ranked player ranked in positions > 50 (1 = yes; 0 = no)	0.79		0	1
Star	Superstar participates in current tournament round (1 = yes; 0 = no)	0.33		0	1
Next	Superstar is potential opponent in the next round (1 = yes; 0 = no)	0.06		0	1
HR: HA	Higher-ranked player has a home advantage (1 = yes; 0 = no)	0.13		0	1
LR: HA	Lower-ranked player has a home advantage (1 = yes; 0 = no)	0.19		0	1
HR: time zones	Number of time zones traversed by higher-ranked player prior to tournament	3.16	4.07	0	19
HR: surface match	Tournament surface matches most familiar surface of higher-ranked player (1 = yes; 0 = no)	0.51		0	1
LR: time zones	Number of time zones traversed by lower-ranked player prior to tournament	2.72	3.78	0	19
LR: surface match	Tournament surface matches most familiar surface of lower-ranked player (1 = yes; 0 = no)	0.54		0	1
Duration	Match duration in minutes	97.56	29.93	29	290

Note: Data include 27,366 main draw ATP singles matches played in 936 tournaments between the 2004 and 2019 seasons. The dataset features a total of 920 players.

Empirical Strategy

Superstar Definition and Identification

We define superstar status in men's professional tennis by focusing on Novak Djokovic, Roger Federer, Andy Murray, and Rafael Nadal, who are commonly referred to as the “Big Four” in the literature (Brown & Yang, 2019; Deutscher et al., 2023). From 2004 to 2019, they won 89% of Grand Slam tournaments and 69% of ATP Finals titles, and a Big Four player finished year-end world No. 1 in every season from 2004 through 2019. At least two Big Four players finished year-end in the top four in 15 seasons, at least three did so in 11 seasons, and all four occupied the year-end top-four positions in five seasons (see Table A1).

However, the Big Four are not superstars from the start of their professional careers. We identify the superstar effect in our model using player-specific dominance windows to distinguish periods of superstar dominance from earlier career years. As a baseline, we date the onset of a player's dominant phase to the first season in which he finishes the year in the top four in our data set. We define the end of a player's dominant phase as the last season before the player finishes outside the year-end top four for two consecutive seasons; for players whose careers extend beyond our sample, the dominance window runs through the end of the sample period (2019). The two-year rule accounts for temporary declines in year-end ranking that can arise from short-run disruptions such as injuries or limited participation, and it avoids treating a single off-year as evidence that a player is no longer a dominant superstar. Accordingly, we classify Federer's 2013 (back) and 2016 (knee) seasons, Nadal's 2016 (wrist) season, and Djokovic's 2017 (elbow) season as dominant seasons for the respective player, despite injury-related absences. In addition, players’ win percentages remain high in those seasons (Deutscher et al., 2023). In contrast, Murray had persistent difficulty recovering from a hip injury and was unable to play enough matches or return to his prior performance level after 2016, which is why his dominance window ends in 2016.

This procedure yields the following dominance windows used in the baseline analysis: Federer (2004–2019), Nadal (2005–2019), Djokovic (2007–2019), and Murray (2008–2016). These dominance windows coincide with those established in the literature (Deutscher et al., 2023). In the robustness section, we test a more conservative dominance-window definition to assess the sensitivity of our results to the baseline coding. Moreover, we test whether our findings are sensitive to considering other elite players such as Stan Wawrinka, Juan Martin del Potro, and Marin Cilic.

Model Set Up

To proxy within-match performance in men's professional tennis, we use ATP serve and return ratings as the dependent variables in our model, capturing performance in dominant and nondominant situations. We control for standard determinants of match outcomes, following Deutscher et al. (2023). To account for additional dynamics in our within-match indicator, we add jet-lag and surface-familiarity controls (Creutzburg et al., 2026). This yields the following linear fixed-effects model:

$$\begin{gathered} \begin{array}{rl}& HR:serve[HR:return{]}_{hlm}=\sum _k^K({\alpha }_{k}Sta{r}_m\times Ca{t}_{khm} \\ +{\beta }_{k}Nex{t}_m\times Ca{t}_{khm}+{\gamma }_k\mathrm{\Delta }Ran{k}_{hlm}\times Ca{t}_{khm}\\ & +{\delta }_k{\mathrm{\Delta }}^{2}Ran{k}_{hlm}\times Ca{t}_{khm})+{ϵ}_{1}H{A}_{hm}\\ & +{ϵ}_{2}H{A}_{lm}+\zeta {\mathit{F}}_{hlm}+H{R}_{hm}+R_m+T_m+u_{hlm}\end{array} \end{gathered}$$

(1)

where $HR:serve[HR:return{]}_{hlm}$ denotes the HR player's serve [return] performance in match m against LR opponent l.

$Sta{r}_m$ equals one if at least one superstar competes in the tournament during a defined period of dominance; Novak Djokovic (2007–2019), Roger Federer (2004–2019), Rafael Nadal (2005–2019), or Andy Murray (2008–2016). $Nex{t}_m$ equals one if the winner of the match could face a superstar in the next round; by construction, $Sta{r}_m$ (=1) is a necessary condition for $Nex{t}_m$ (=1).

Equation (1) features a set of covariates: $Ca{t}_{khm}$ indicates whether the HR player is ranked in one of $K=3$ distinct rank categories ( $k=1$ : ranks 1–20; $k=2$ : ranks 21–50; $k=3$ : ranks > 50). The difference in ranking positions between players ( $\mathrm{\Delta }Ran{k}_{hlm}$ ) accounts for their heterogeneous abilities. Including a squared term ( ${\mathrm{\Delta }}^{2}Ran{k}_{hlm}$ ) accounts for a nonlinear relationship. If a player has a home advantage (HA) in match m, the binary variable $HA$ equals one; a player is considered to have a HA if a match is hosted in the player's home country (Creutzburg et al., 2026). While player-season fixed effects ( $H{R}_{hm}$ ) capture players’ season-specific forms, $R_m$ and $T_m$ are round and tournament fixed effects. In addition, ${\mathit{F}}_{hlm}$ captures match-specific characteristics relevant for players’ performance beyond the HA effect. In particular, players’ jet lag and surface-familiarity effects mitigate against an omitted variable bias regarding players’ HA effects. Two variables are included in the analysis to control for the number of time zones traversed by both players between tournaments; two dichotomous variables indicate whether the tournament surface matches the surface utilized predominantly in the HR and LR player's home country (Creutzburg et al., 2026). Similarly, the match's duration (in minutes) decreases with the rank difference between HR and LR players (Jane, 2020) and most likely affects players’ serve and return performance. Hence, including the duration rules out a potential omitted variable bias concerning the effects of $\mathrm{\Delta }Ran{k}_{hlm}$ and ${\mathrm{\Delta }}^{2}Ran{k}_{hlm}$ .

The idiosyncratic error term is given by $u_{hlm}$ . We present cluster-robust standard errors at the level of the HR player-season and tournament-season. In all the following specifications, we employ the estimator suggested by Correia (2017).

Equation (1) is estimated separately for serving and returning performance. The dependent variables are the ATP performance ratings of the high-ranked player: the serve rating $(HR:serve{)}_{hlm}$ and the return rating $(HR:return{)}_{hlm}$ .

The server generally has a higher probability of winning a point, making the serving position the dominant situation in a match (Albert & Kovalchik, 2017). The ATP's serve rating summarizes serving efficiency as:

$$\begin{array}{rl}\mathrm{Serve}\mathrm{rating}=& \text{\%}1\mathrm{st}\mathrm{serve}\mathrm{in}+\text{\%}1\mathrm{st}\mathrm{serve}\mathrm{points}\mathrm{won}\\ & +\text{\%}2\mathrm{nd}\mathrm{serve}\mathrm{points}\mathrm{won}\\ & +\text{\%}\mathrm{service}\mathrm{games}\mathrm{won}+\mathrm{Aces}\\ & -\mathrm{Double}\mathrm{faults}\end{array}$$

(2)

A higher $(HR:serve{)}_{hlm}$ indicates more consistent and effective serving performance.

The returning player occupies a nondominant position. The ATP's return rating measures a player's ability to anticipate and respond to the opponent's serve:

$$\begin{array}{rl}\mathrm{Return}\mathrm{rating}=& \text{\%}1\mathrm{st}\mathrm{serve}\mathrm{return}\mathrm{points}\mathrm{won}\\ & +\text{\%}2\mathrm{nd}\mathrm{serve}\mathrm{return}\mathrm{points}\mathrm{won}\\ & +\text{\%}\mathrm{Return}\mathrm{games}\mathrm{won}\\ & +\text{\%}\mathrm{Break}\mathrm{points}\mathrm{converted}\end{array}$$

(3)

A higher $(HR:return{)}_{hlm}$ reflects stronger anticipation and effectiveness in countering the opponent's serve. Estimating equation (1) with either $(HR:serve{)}_{hlm}$ or $(HR:return{)}_{hlm}$ as the dependent variable enables an assessment of how superstar presence influences player performance across dominant and nondominant match situations.

Results

Baseline Results

We report the results obtained from estimating our main equation (1) in Table 2. The dependent variables are HR players’ serve (HR: serve) and return (HR: return) ratings.

Table 2.

Superstar Effect on Higher-Ranked Players’ Serve (HR: Serve) and Return (HR: Return) Performance.

	(1)		(2)
	HR: serve		HR: return
Star × HR: 1–20	−2.358**	(1.118)	−1.904	(1.571)
Next × HR: 1–20	−2.124*	(1.160)	−3.655*	(2.128)
Star × HR: 21–50	−2.913***	(1.084)	−1.883	(1.389)
Next × HR: 21–50	0.857	(1.457)	−0.406	(2.156)
Star × HR: > 50	2.577**	(1.263)	0.228	(1.803)
Next × HR: > 50	0.068	(2.391)	7.011**	(3.457)
HR: HA	2.742***	(0.727)	4.980***	(1.085)
LR: HA	1.007*	(0.603)	−0.529	(0.891)
HR: time zones	−0.353***	(0.082)	−0.167	(0.112)
LR: time zones	0.143*	(0.077)	0.204*	(0.106)
HR: surface match	2.930***	(0.578)	6.228***	(0.810)
LR: surface match	−1.943***	(0.495)	−1.354*	(0.716)
Duration	−0.096***	(0.008)	−0.358***	(0.012)
N	26,901		26,901
R 2	0.266		0.221
Within R2	0.023		0.059

Note: The standard errors in parentheses are clustered at the HR player-season and tournament-season levels. All estimations include rank difference interactions, round FEs, tournament FEs, and HR player-season FEs. *p < 0.1, **p < 0.05, ***p < 0.01.

Our analysis reveals several superstar effects. For the group of top 20 players, there is an adverse main effect of superstar presence (Star × HR: 1–20) on players’ performance in dominant match situations; in column (1), a (top 20) HR player's serve rating is approximately 2 points lower than when no superstar is present. This effect is further accentuated when the player is likely to face a superstar in the next round (Next × HR: 1–20). In such constellations, a top 20 HR player's performance in dominant match situations is reduced by approximately two points. In contrast, in nondominant match situations (Column 2), the estimated main effect (Star × HR: 1–20) is not significantly different from zero. However, the moderator effect (Next × HR: 1–20) is marginally significant, suggesting that the return performance (nondominant match situation) of the top 20 HR players decreases by approximately four points only if the next match may be against a superstar.

HR players ranked between positions 21–50 experience a three-point decline in their serve rating when a superstar is present (Star × HR: 21–50). The main effect is not significantly contingent on the possibility of a match against a superstar in the subsequent round (Next × HR: 21–50). Proxied by the player's return rating, the player's performance in nondominant match situations is not significantly affected by the superstar effect.

HR players ranked outside of the top 50 enjoy an increase of three points in their serve rating in the case of superstar presence, the main effect (Star × HR: > 50). It is estimated that competing against a superstar in the subsequent round (Next × HR: > 50) will not result in a notable alteration in the player's serve performance. However, it does increase performance in nondominant positions, proxied by the player's return rating.

Finally, we find that an HR player increases his performance in both dominant and nondominant match situations if he competes on his most familiar surface (HR: surface match). In addition, the estimated coefficients regarding surface familiarity are (in absolute terms) greater than the main effect of superstar presence.

The second part of our analysis examines the superstar effect on situational LR players’ performance. We rewrite equation (1) such that we model the performance of LR player l competing against an HR player h in match m. In particular, we now use the LR player's serve and return ratings as dependent variables. Moreover, we exchange the LR player season fixed effects ( $L{R}_{lm}$ ) and the player's ranking category ( $Ca{t}_{klm}$ ) accordingly. The results are presented in Table 3.

Table 3.

Superstar Effect on Lower-Ranked Players’ Serve (LR: Serve) and Return (LR: Return) Performance.

	(1)		(2)
	LR: serve		LR: return
Star × LR: 1–20	−0.930	(4.018)	−5.184	(5.757)
Next × LR: 1–20	−1.425	(3.695)	2.355	(5.393)
Star × LR: 21–50	−6.850***	(1.286)	−4.538**	(1.856)
Next × LR: 21–50	6.699***	(1.705)	4.556*	(2.382)
Star × LR: > 50	−0.785	(1.045)	−0.039	(1.372)
Next × LR: > 50	−1.000	(1.491)	1.548	(1.985)
HR: HA	−1.452*	(0.781)	−5.702***	(1.079)
LR: HA	3.868***	(0.841)	1.894	(1.167)
HR: time zones	0.172*	(0.088)	0.020	(0.127)
LR: time zones	−0.257***	(0.091)	−0.051	(0.128)
HR: surface match	−2.864***	(0.586)	−2.336***	(0.770)
LR: surface match	2.677***	(0.674)	4.802***	(0.935)
Duration	0.244***	(0.009)	0.144***	(0.012)
N	26,307		26,307
R 2	0.309		0.209
Within R2	0.054		0.020

Note: The standard errors in parentheses are clustered at the LR player-season and tournament-season levels. All estimations include rank difference interactions, round FEs, and LR player-season FEs. *p < 0.1, **p < 0.05, ***p < 0.01.

The single most surprising finding to emerge from Table 3 relates to an asymmetric effect of the timing of potentially facing a superstar on the performance of LR players ranked between positions 21–50. In particular, there is a significant negative superstar effect on both the player's serve performance and return performance (Star × LR: 21–50). Interestingly, we find that potentially facing a superstar in the subsequent round (Next × LR: 21–50) approximately absorbs the estimated main effect; the player's serve [return] performance increases by 7 [5] points. Like HR players, LR players appear to benefit on both their serve and return if playing on the most familiar surface. In contrast, those effects are smaller than the main effect of superstar participation in a tournament round.

Robustness Checks

We conduct several robustness checks to test the sensitivity of our results across different specifications; for brevity, all results are reported in the Appendix.

First, we include opponent player-season fixed effects to account for unobserved, season-specific heterogeneity on the opponent side (Sung & Pyun, 2023). Specifically, for the HR players’ performance equations, we augment our baseline specification (Table 2 in the main manuscript) by adding LR player-season fixed effects in addition to the existing controls and HR player-season fixed effects. The resulting estimates are reported in Table A3. For the LR players’ performance equations, we augment the baseline specification (Table 3) by adding HR player-season fixed effects in addition to the existing controls and LR player-season fixed effects; these results are reported in Table A4. In both sets of regressions, standard errors are clustered at the HR player-season, LR player-season, and tournament-season levels. Comparing Table 2 with Table A3, the results for HR players are broadly unchanged in the sense that our main conclusion continues to hold: HR players strategically adjust their performance in (non)dominant match situations depending on their rank group and on whether they are (or may soon be) exposed to a superstar in subsequent matches. At the same time, conditioning on LR player-season heterogeneity generally yields smaller point estimates and less precise inference. Several coefficients that are statistically significant in Table 2 remain significant in Table A3, whereas others lose statistical significance. Comparing Table 3 with Table A4, the central results for LR players ranked 21–50 remain intact: the Star effect remains negative and statistically significant for both serve and return performance, and the Next effect remains positive and statistically significant for both outcomes.

Second, we extend our baseline specification to account for potential variation in players’ awareness of future opponents. In particular, we include a binary variable that equals one if all superstar players have been eliminated from a tournament and interact it with the rank categories (Deutscher et al., 2023). In this specification, none of the interaction terms are statistically significant, suggesting that the observed behavioral adjustments depend on the continued presence of a superstar (see Table A5 for HR players and Table A6 for LR players).

Third, we test whether our findings could be influenced by pre-Grand Slam preparation rather than genuine superstar effects. Players sometimes enter smaller tournaments shortly before Grand Slams to train and adjust to surfaces or regain form, which could lead to reduced effort. If unaccounted for, this behavior might bias the baseline estimates. Although our design partially mitigates such concerns by using within-player × season variation and including tournament and round fixed effects, we further exclude all tournaments starting within two weeks and, separately, one month before a Grand Slam. The superstar coefficients remain stable under both restrictions, indicating that the observed patterns are not driven by pre-Grand Slam preparation. Results are reported in Table A7 for HR players and Table A8 for LR players.

Fourth, to verify that our findings are not an artifact of the ATP performance rating methodology, we construct alternative continuous performance measures based on the net difference between points won and points lost for both HR and LR players. Specifically, we calculate the net number of points won by each player overall. The results obtained from these alternative measures align closely with those of our main analysis in both direction and statistical significance, indicating that the observed superstar effects are not driven by the construction of the ATP ratings (see Tables A9 and A10). To further explore whether the observed decline in serve and return performance among HR players reflects reduced effort or a loss of technical precision, we consider an alternative mechanism related to unforced errors. If players expect to face a superstar, they may experience increased psychological pressure, leading to more execution mistakes, such as double faults, rather than strategic effort adjustments (Brown, 2011). To test this, we re-estimate the model using the number of double faults per match as the dependent variable. The coefficients for both superstar presence and anticipation are statistically insignificant for HR. These results suggest that the observed decreases in serve and return ratings are not due to higher error rates or loss of precision, but instead reflect intentional, strategic changes in performance intensity. Results are shown in Table A9 for HR players and Table A10 for LR players.

Fifth, to ensure that the definition of each player's dominant period reflects contemporaneous recognition rather than retrospective judgment, we refine the identification of superstar seasons based on observable milestones of established superiority. Specifically, a player's dominant phase begins in the season when sustained success, multiple major titles, and broad public acknowledgment as a leading figure in men's tennis converge. Under this definition, Djokovic's period of dominance begins in 2011, when he won three Grand Slam titles and reached world No. 1, while Murray's begins in 2012, marked by his U.S. Open victory and Olympic gold medal (Table A1). This refinement restricts the classification to seasons in which players had already achieved lasting competitive authority and recognition by peers, media, and fans as part of the sport's elite. Re-estimating the model using these revised boundaries produces results consistent with the baseline specification. This suggests that the observed behavioral responses are not artifacts of how superstar periods are defined, but rather reflect systematic performance adjustments linked to the presence of clearly established superstars. Results are presented in Table A11 for HR players and Table A12 for LR players.

Sixth, players who achieved brief periods of success, such as single Grand Slam victories or surface-specific dominance, are not included in our core superstar definition, as their achievements did not translate into sustained superiority. To examine whether the superstar effect is exclusive to the four players we identify as superstars, we conduct a robustness analysis testing if a similar effect extends to a group of second-tier superstars (Jedelhauser et al., 2023). These players consistently ranked within the Top 10, won at least one Grand Slam title, and occasionally defeated Federer, Nadal, Djokovic, or Murray at major stages, yet never reached comparable long-term dominance. This group includes Stan Wawrinka (2014–2017), Juan Martin del Potro (2009–2013; 2016–2018), and Marin Cilic (2014–2018). Estimating the effect of second-tier superstar presence therefore serves as a test of whether the behavioral responses identified in our main analysis are specific to the exceptional and sustained dominance of the Big Four, or whether they generalize to other elite competitors. The results, presented in Tables A13 and A14, show no significant interactions between second-tier superstar presence and the HR or LR player categories, suggesting that the behavioral adjustments are indeed exclusive to the sustained dominance of the four leading superstars.

Seventh, we test whether differences in player experience influence within-match performance and could confound the estimated superstar effects. More experienced players may exhibit greater composure under pressure, stronger tactical awareness, and more efficient energy management, all of which might shape their response to the presence or anticipation of a superstar opponent. To capture this dimension, we construct a measure of professional experience defined as the number of years since a player's first full ATP season. This variable is included both in levels (for the opponent) and interacted with the key superstar indicators ( $Star$ and $Next$ ) for the focal player. This specification allows experience to moderate the behavioral response to superstar contexts while maintaining the within-player identification of our model. Results presented in Tables A15 and A16 show that the inclusion of these experience interactions does not materially change the estimated superstar coefficients for HR and LR players. This indicates that the observed performance adjustments are not driven by systematic differences in tenure or learning effects. More broadly, the findings suggest that even after accounting for accumulated professional maturity, the behavioral responses to superstar presence and anticipation persist, reinforcing the interpretation that these effects reflect situational effort modulation rather than experience-related composure or fatigue.

Finally, results from estimating our equation using binary win/loss outcomes in the style of Deutscher et al. (2023) are reported in Table A2 in the Appendix and support the robustness of our results.

Discussion

In the context of professional sports, the presence of superstar players has the potential to significantly impact the performance dynamics of other competitors. This phenomenon, often referred to as the superstar effect, has been the subject of extensive research, which has revealed intricate patterns of strategic adaptation among players (see, for example, McFall & Rotthoff, 2020).

We affirm previous findings that superstar-induced shifts in HR players’ performance differ across rank categories and that HR players adjust their performance according to the timing of potentially facing a superstar in tennis tournaments (Deutscher et al., 2023). As central results, we uncover that, unless there is an immediate threat of facing a superstar, HR players only adjust performance in match situations in which they have the upper hand (dominant positions). Our findings suggest that performance in nondominant game situations is affected, if at all, only when an HR player potentially encounters a superstar in the subsequent round. This underscores a strategic dimension in HR players’ competitive play: the players are not only reactive to immediate circumstances but also engaged in strategic planning (Brown & Minor, 2014), such as by adjusting their performance to preserve their physical state. Similar to HR players, our results indicate that there are differences in the extent of performance alterations induced by superstars among different rank groups for LR players. However, these discrepancies do not extend to different match scenarios. Finally, while smaller for HR players in absolute terms, the main effect of superstar participation on LR players’ situational performance is greater than, for example, the impact of surface familiarity.

For LR players, we observe a positive performance shift when a potential matchup with a superstar looms in the next round. This pattern may capture distinct motivational mechanisms. In particular, the prospect of competing against a superstar can generate symbolic and reputational incentives, including greater media attention, commercial exposure, and the intrinsic prestige of sharing the court with an iconic player. Such encounters may also be perceived as learning opportunities, motivating players to invest additional effort to reach that stage. From a behavioral point of view, LR players may experience reduced loss aversion, as expectations of winning are low and the psychological cost of failure diminishes, allowing performance to improve (Diel et al., 2021; Kahneman & Tversky, 1979). These mechanisms provide plausible explanations for the positive effect of facing a superstar in the next round observed among LR players.

In contrast, HR players appear to respond differently to upcoming superstar encounters. The observed asymmetry in performance shifts between HR and LR players aligns with the logic of strategic effort allocation. HR players, aware of their favored status and anticipating tougher matches ahead, may deliberately temper their performance in earlier rounds to conserve energy for a potential clash with a superstar. This interpretation is consistent with theoretical models of multi-stage contests, which show that competitors optimally distribute effort across stages when future stakes vary in intensity (Malueg & Yates, 2010). Empirical analyses of professional tennis further document such forward-looking adjustments, where players strategically manage exertion within and across matches to balance immediate performance with long-term endurance demands (Depken et al., 2022). From this point of view, the performance pattern among HR players reflects calculated effort management aimed at maintaining competitiveness in later, more demanding rounds against stronger opponents.

Conclusion

This study examines how superstar presence relates to performance in professional men's tennis across within-match situations using data from 2004 to 2019. Performance is measured using ATP serve and return ratings, which capture effectiveness in dominant (serve) and nondominant (return) situations for HR and LR players.

We provide evidence that HR players deliberately adjust their performance in specific (non)dominant match situations in response to superstar participation in a tournament, rather than dispersing effort across all match situations indiscriminately. In contrast, while we observe that superstar-induced changes in LR players’ performance vary across rank groups, their shifts in performance do not differ across match situations.

We acknowledge that the use of the ATP serve and return ratings presents inherent limitations that warrant careful consideration. The ratings are tennis-specific indicators, and extrapolating these findings to other sports and tournament settings must be approached with caution to avoid overgeneralization and misinterpretation. In addition, given that the study is based on data from male players only, it would be beneficial to conduct a similar investigation using data from the Women's Tennis Association to gain insights into potential gender differences.