Less Information, More Comparison, and Better Performance: Evidence from a Field Experiment

We use a field experiment in professional sports to compare effects of providing absolute, relative, or both absolute and relative measures in performance reports for employees. Although studies have documented that the provision of these types of measures can benefit performance, theory from economic and accounting literature suggests that it may be optimal for firms to direct ∗The London School of Economics and Political Science; †Harvard Business School, Harvard University and Laboratory for Innovation Science at Harvard; ‡WHU—Otto Beisheim School of Management and Laboratory for Innovation Science at Harvard Accepted by Christian Leuz. We thank an anonymous reviewer, Dennis Campbell, Wim Van der Stede, Margaret Abernethy, Karim Lakhani, Benno Torgler, and participants at the 2020 Journal of Accounting Research Conference, Harvard Business School, the Laboratory for Innovation Science at Harvard (LISH), WHU—Otto Beisheim School of Management, the London School of Economics and Political Science, and the University of Melbourne for their very helpful comments. We thank Sascha L. Schmidt for his support and for introducing us to the field site, Anton Paar SportsTec. We also thank Dominik Schreyer for his guidance. We thank Anton Paar SportsTec personnel for helpful comments and providing data access as the field site for our study. This paper did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. All errors are our own. This paper is based on a registered experiment with the AEA RCT Registry that is available on: https://doi.org/10.1257/rct.4141-3.0. [The copyright line for this article was changed on 30 June 2021 after original online publication.] 657 © 2021 The Authors. Journal of Accounting Research published by Wiley Periodicals LLC on behalf of The Accounting Research Center at the University of Chicago Booth School of Business This is an open access article under the terms of the CreativeCommonsAttribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. 658 h. eyring, p. j. ferguson, and s. koppers employees’ attention to some types of measures by omitting others. In line with this theory, we find that relative performance information alone yields the best performance effects in our setting—that is, that a subset of information (relative performance information) dominates the full information set (absolute and relative performance information together) in boosting performance. In cross-sectional and survey-data analyses, we do not find that restricting the number of measures shown per se benefits performance. Rather, we find that restricting the type of measures shown to convey only relative information increases involvement in peer-performance comparison, benefitting performance. Our findings extend research on weighting of and responses to measures in performance reports. JEL codes: C93, D83, D90, M40, M51, M54


Introduction
To guide the use of performance evaluation and reporting, a growing literature in accounting examines how internal performance reports affect employee performance (Hannan, Krishnan, and Newman [2008], Tafkov [2013]). Research on effects of performance reports has explored how information users place weight on and respond to different types of measures, and has found that these weights depend on the combination of measures provided-such as nonfinancial and financial measures, or transitory and persistent measures Salterio [2000, 2002], Elliott, Hobson, and White [2015]). Although the popular press has raised the issue of whether reports that show absolute measures, relative measures, or a combination of these measure types to employees will yield the best performance effects (The Atlantic [2015], The Economist [2016]), empirical research has yet to examine this question (Schnieder [2018]). In settings where employees compete against one another on a range of objective performance measures, this question has received substantial attention (McKinsey [2016], Mercer [2019]). We contribute to this issue by conducting a field experiment in a highly competitive setting-professional European football-to compare effects of providing employees with absolute, relative, or both absolute and relative information on objective performance measures. 1 In collaboration with Anton Paar SportsTec-the producers of skills.lab, a state-of-the-art football training simulator used by elite athletes in Germany and Austria-we manipulate performance information reported 1 We use the word "football" as shorthand for European football, alternatively known as "soccer" in some countries including the United States. We use the term "relative performance information" to refer to performance information that compares the focal agent to his or her peers (e.g., Hannan, Krishnan, and Newman [2008], Blanes i Vidal and Nossol [2011], Tafkov [2013]). We use the term "absolute performance information" to refer to performance information that does not provide a comparison to peers (e.g., Ashton [1990], Kluger and DeNisi [1996]). during training to 117 professional and semiprofessional players. We vary the attributes of this information on two key performance measures: passing accuracy and speed. These skills are highly valued in this industry and affect the career advancement of players (Liu et al. [2015]), who compete in training for selection in upcoming games and to secure ongoing contracts. We also exploit cross-sectional variation in players' task commitment, immediate career concerns, and level of experience to test for and demonstrate individual characteristics that lead to heterogeneous responses to performance information.
In our tests on the full sample, we find that relative information boosts performance more than absolute information. This result is particularly strong for players who have more direct competitors for their position, and so our study demonstrates the benefits of providing employees with ranks on objective measures in settings where career concerns are strongly tied to relative performance. We then show, across a range of performance outcomes, that relative information is as or more effective at boosting performance than the combination of both relative and absolute information. Players who receive relative information alone, rather than receiving the combination of relative and absolute information, achieve an incremental performance improvement that is equivalent to rising from the 50th percentile to roughly the 70th percentile in terms of both proximity to target and speed of completion. Given that these positive effects occur across both speed and accuracy, our study documents broad performance benefits resulting from performance report design.
The performance benefit of providing relative information in isolation is consistent with studies that account for models of human cognition, which suggest that added information can lead to worse performance (Birnbaum [1976], Kahneman [1974, 1983], Simon [1979], Gigerenzer and Goldstein [1996], Casas-Arce, Lourenço, and Martínez-Jerez [2017]). We draw on this literature to understand why relative information outperforms the combination of relative and absolute information in our setting.
First, we exploit cross-sectional variation to examine whether adding absolute information alongside relative information harms performance by providing players with "too many" measures per se. When forced to process quantities of information that exceed their cognitive capacity, information users can become stressed and disengage from the source of the information (Ettema and Zielhuis [1971], Malhotra [1982], Bawden [2001]). Supplemental analysis suggests that this mechanism does not explain our results. Specifically, when we hold constant whether performance information is absolute or relative, we find that adding additional measures does not affect performance. Instead, performance declines only when we provide additional measures by adding absolute information alongside relative information.
We then explore the concept from management accounting literature that employees change their weighting of a measure in the presence of other measures, and that this can lead to worse performance (Kaplan and Norton [1996], Lipe and Salterio [2000], Casas-Arce, Lourenço, and Martínez-Jerez [2017]). In particular, we draw on research from psychology and behavioral economics that shows that individuals place greater weight on and respond more strongly to a cue when it is presented alone rather than in combination with another cue (Lichtenstein, Earle, and Slovic [1975], Birnbaum [1976], Kruschke and Johansen [1999], Dellavigna [2009]). In line with this insight, we find that the average player places less weight on relative performance information-engaging less in peer comparison and performing worse-when we include absolute information alongside relative information in their performance report.
This reduced weighting of relative information does not harm performance for all players, and we examine this cross-sectional variation in our results. Less-experienced players, who are in developmental stages of their careers, benefit from this shift in cognitive weight away from peer comparison and toward their own performance in absolute terms. For the average professional in our study, who is experienced and highly skilled, we find a negative effect of adding absolute information alongside relative information.
Our study makes three primary contributions. First, we extend the literature on the design of performance measurement and evaluation systems within firms. Research in management accounting has demonstrated that internal reports and feedback are effective tools for heightening motivation, communicating strategic priorities, and improving performance (Simons [1995], Kaplan and Norton [1996], Hannan, Krishnan, and Newman [2008], Casas-Arce, Lourenço, and Martínez-Jerez [2017]). To extend this literature, we offer evidence that managers can boost employee performance by reporting the subset of measures that best motivates and facilitates development in a given setting.
Second, our study has implications for the broader accounting literature on the marginal costs and benefits of providing information to decision makers. Studies of financial reporting have shown that, due to investors' limited attention and processing power, disclosing redundant measures or measures that distract from the salience of key measures can lead to inefficient investment (Hirshleifer and Teoh [2003], Elliott, Hobson, and White [2015], Cardinaels, Hollander, and White [2019]). There is also evidence that firms consider these information processing costs when making disclosure decisions (Blankespoor [2019], Basu, Pierce, and Stephan [2019], Blankespoor, deHaan, and Marinovic [2020]). In our study, we highlight that managers should also consider the trade-off between the marginal costs and benefits of providing information within the firm. Managers can draw on our findings as evidence that the marginal costs of adding information to a report can be sufficient to outweigh the marginal benefits.
Third, we extend research that examines heterogeneity in responses to performance information. This literature has explored how effects of performance information depend on moderators including ability, gender, and education (Azmat and Iriberri [2010], Barankay [2012], Eyring and Narayanan [2018]). We extend this literature by examining how effects of performance reports depend on task commitment, career concerns, and professional experience. First, we find that task commitment, which we measure using data on how frequently a player normally trains, enhances the performance benefits of relative performance information. Second, we find that career concerns, which we measure in terms of intensity of competition for immediate career outcomes, also strengthen the performance benefits of relative performance information. 2 This is consistent with theory that relative performance information should have stronger effects on motivation when there are stronger incentives to achieve a high rank (Tafkov [2013]). Third, we also find that less-experienced employees, who are more likely to be in a developmental stage of their careers (Podsakoff and Farh [1989], Goodman, Wood, and Hendrickx [2004]), perform best when they receive both absolute and relative performance measures. This result among players in an early stage of their careers is consistent with theory that a broader set of information is useful for learning and skill development (Song et al. [2018]).
The remainder of this paper is organized as follows. Section 2 describes our field site, skills.lab, and the industry in which it operates. Section 3 outlines the related literature and develops our hypotheses. Section 4 outlines our field experiment and data. Section 5 reports the results of our empirical analysis. Section 6 concludes.

field site
Our field site is skills.lab, an indoor training simulator used by professional and semiprofessional football players in Austria and Germany. The simulator guides players through practice passing drills in highly controlled, game-like conditions. 3 Motion-tracking technology allows the simulator to measure and report a player's performance in passing drills (see appendix A, exhibit 1). After a player completes a drill, the simulator displays information on two dimensions of performance that are highly valued in football-passing accuracy and speed (Redwood-Brown [2008], Liu et al. [2015]). Managers and coaches can set the simulator to provide their players with absolute, relative (i.e., ranks), or both absolute and relative performance information.
For our experiment, we take advantage of a number of technical features of the simulator. As the simulator is fully programmable and automated, we keep the elements of the passing task perfectly replicable among participants. We randomly assign different performance information treatments at the player level, and precisely measure the performance of each player in our experiment by using skills.lab's tracking technology. We exploit the combination of these features to provide precise, causal estimates of the performance effects of the different types of information.

industry and incentives
The workers in our study are professional and semiprofessional athletes. They range in experience from rookies to veterans. The typical player in our sample attends daily training sessions-including regular drills in the skills.lab simulator-and competes weekly in games. 4 During practice and as part of their post-game evaluations, elite football teams commonly provide players with detailed, quantitative performance information, including ranks (The New York Times [2019]).
Strong incentives motivate players to outperform their teammates while training in the skills.lab simulator. The players in our study are interested in securing a spot in the best league and a starting spot in their current team. Managers reported using players' rank in the simulator when making these selection decisions. Managers also explained that players are aware of this practice and thus have strong incentives to perform well in the simulator. This data-driven approach to talent identification is widespread in the industry (Güllich [2014]). Players face stronger competition for starting spots in upcoming games when teammates who play the same position are also participating in the drill. The number of players competing for the same starting spot on a team is apparent to the participants in our study given that they arrive to the simulator in teams for their training sessions. We are able to measure within each team the depth of players at a given position who are participating in the experiment and competing for the same starting spot on the team. In cross-sectional analysis, we use variation in this measure to explore how our results depend on players' immediate career concerns.
Career advancement produces large financial payoffs in our setting. The average annual salary for an athlete playing first-division football in one of Europe's top-five leagues is $2.4 million (The Guardian [2018]). Within the season, players receive bonuses when they are selected for games and for meeting performance targets (e.g., scoring a certain number of goals; Frick [2011]). At the end of each season, some players who fail to make regular appearances or who perform poorly in games are cut from the roster, traded to rival teams, or relegated to play in one of their club's lower-level teams (Morris, Tod, and Oliver [2015], Gong, Sun, and Wei [2018]).
Managers' use of the simulator for selection decisions is based on their belief, borne out by observations they reported to us, that objective performance in the simulator is associated with subjective evaluations of players and with players' performance in games. Administrators at skills.lab similarly explained, "we observe a strong correlation between skills.lab and in-game performances." Moreover, training is an integral part of players' job responsibilities. Thus, we are studying a task that is both fundamental to a job and one that is related to and influences career outcomes. This helps our study generalize to other settings with these features.
Our study speaks most directly to settings where highly skilled workers in competitive settings receive performance information in terms of objective measures. Workers in such settings include health care providers, mutual fund managers, school teachers, and airline pilots (Muralidharan and Sundararaman [2011], Berk and van Binsbergen [2015], Song et al. [2018], Gosnell, List, and Metcalfe [2020]). More generally, "star" senior executives-whose relative performance is also highly public and commonly covered in the popular press-face career incentives and performance pressures akin to those that professional athletes face (Brickley, Linck, and Coles [1999], Malmendier and Tate [2009], Lucifora and Simmons [2003]). Furthermore, like professional football teams, firms in financial services, consulting, and high-growth industries commonly employ performance ranks to foster internal competition (Cappelli and Tavis [2016], CNBC [2019]).
For these reasons, a number of studies in accounting, economics and finance have similarly used professional sports as a setting to examine fundamental economic questions concerning the role of incentives and information within firms and markets (Brown [2011], Cadman and Cassar [2015], Black and Vance [2017], Allen et al. [2017]). As in those papers' settings, our study's field site involves objective performance measurement. McKinsey [2016] and Mercer [2019] discuss how a number of firms in settings that include heath care, consumer technology, and logistics have switched among the alternatives of providing relative, absolute, or relative and absolute information on objectively measured performance. In section 3, we discuss theory specific to subjective performance measurement. We acknowledge that our findings do not address issues of actual or perceived favoritism and unfairness that can result from subjective performance evaluation systems (Bol et al. [2010], Bol [2011]).

Literature Review and Hypotheses
Our predictions draw on literature that examines absolute and relative information as determinants of performance. Absolute performance information describes performance without offering a comparison to others' performance. In contrast, relative performance information describes performance in relation to a benchmark group, such as peers or coworkers (Ilgen, Fisher, and Taylor [1979], Kluger and DeNisi [1996]). Accounting research has shown that both absolute and relative performance information serve as developmental and motivational tools for improving performance (Hannan, Krishnan, and Newman [2008], Tafkov [2013], Casas-Arce, Lourenço, and Martínez-Jerez [2017]).
A number of studies demonstrate that each of these types of performance information can facilitate learning about the relationship between an action and an outcome (Adams [1987], Hirst and Luckett [1992], Kluger and DeNisi [1996], Song et al. [2018]). In our setting, performance information should assist learning by allowing players-in particular, lessexperienced players who are in a developmental stage of their career-to see how ball-handling and striking techniques affect passing accuracy and speed.
In addition to facilitating development, performance information also motivates workers by establishing a feedback loop between effort and reported improvement (Deci [1972], Ilgen, Fisher, and Taylor [1979], Locke and Latham [1990]). Studies in psychology and economics show that reported success generates positive "affect"-or sensations-and this motivates individuals to exert effort in order to perform well (Ilies and Judge [2005], Barankay [2012]). As an example in accounting, auditors were more motivated and improved their performance on a classification task when given absolute performance information (Ashton [1990]). Performance information in our setting should similarly motivate improvement by allowing players to see changes in their performance on the passing drill.
Studies on performance information have examined the incremental benefits of providing relative performance information as compared to providing absolute performance information. 5 General evaluability theory suggests that these benefits derive partly from the added evaluability of relative information, or its ability to aid in interpreting one's own performance (Hsee and Zhang [2010]). Specifically, the benchmark of peer performance helps in identifying whether performance is strong and where improvement is necessary (Song et al. [2018]). Lacking this inherent benchmark, absolute performance information leads the information user to search for a benchmark, which can be an arbitrary standard like position relative to a round number (Abeler et al. [2011], Allen et al. [2017], Eyring and Narayanan [2018]). 6 Especially in competitive settings like ours, where standing relative to peers is relevant for career outcomes, relative information is likely to facilitate learning about the favorability of performance.
In addition to promoting evaluability, relative measures provide sources of motivation that absolute measures alone do not. Specifically, relative measures allow peer-performance comparison, which triggers the inherent desire within individuals to compare favorably to peers (Festinger [1954], Suls and Wheeler [2000], Roels and Su [2014]). As a result, even when payoffs are not tied to relative performance, relative information motivates workers (Frederickson [1992], Azmat and Iriberri [2010], Blanes i Vidal and Nossol [2011], Barankay [2012]). Furthermore, when peers are competing for payoffs, peer-performance comparison operates through financial incentives to motivate workers (Bull, Schotter, and Weigelt [1987], Casas-Arce and Martínez-Jerez [2009], Eriksson, Poulson, and Villeval [2009]). These incentives arise when contracts explicitly account for relative performance, and also when relative performance and career advancement are implicitly connected through hiring and promotion practices (Lazear and Rosen [1981], Campbell [2008], Newman and Tafkov [2014]). The latter scenario describes our setting, in that managers use performance in the simulator as a factor when selecting players for divisions, teams, and starting positions.
We also note that some studies have demonstrated, at least in certain cross-sections, that the provision of relative information in performance reports harms performance (Hannan, Krishnan, and Newman [2008], Ashraf [2018]). The business press has pointed to these and other issues when discussing reasons why some companies have stopped providing relative measures to employees (The Atlantic [2015], Mercer [2019]). For example, relative information can discourage employees who are at developmental stages of their careers (Goodman, Wood, and Hendrickx [2004], Forbes 6 In accounting research that uses experiments to study performance information, participants often receive absolute performance information on either a task that they are unlikely to have completed before or an entirely novel task designed explicitly for the study (Ashton [1990], Hannan, Krishnan, and Newman [2008], Hecht, Tafkov, and Towry [2012], Tafkov [2013]). In our study, the task is different on some parameters-the particular speed and sequence of balls passed to the player from different directions-but generally similar to those that players have experience with. Experience with similar situations should aid evaluability of absolute information in our setting as compared to one in which the task is completely novel. Nonetheless, relative information offers the advantage of an explicit and inherent benchmark suited to the particular task to promote evaluability.
[2018]). Also, some employees complain that relative information in the form of a ranking, as we use in our study, does not allow managers to provide high marks to all employees even when management is highly satisfied with the entire group's performance (The Atlantic [2015], Forbes [2018]).
These downsides of relative information plausibly apply in our setting and in others, ranging from financial service to consumer technology companies, where managers provide relative information on objective measures to employees in a competitive workforce (The New York Times [2015], McKinsey [2016]). There are other issues that arise in particular contexts, such as feelings of unfairness when relative measures are determined subjectively, that are not at play in our setting. Our results regarding the effects of relative information thus speak to settings in which a subset of problems that can result from relative information are relevant. 7 Collectively, this theory and prior empirical work suggests that the informational and motivational advantages of relative information should be especially pronounced in our setting. In particular, the players in our study face substantial competition for career advancement, and theory suggests that this will enhance the motivational effect of relative information (Lazear and Rosen [1981], Newman and Tafkov [2014]). Also, in competitive contexts, relative information is more likely to enable learning about the favorability of performance (Hsee and Zhang [2010], Song et al. [2018]). Moreover, the average player is highly skilled and experienced and so is less likely to experience feelings of discouragement from relative information that occur in developmental stages (Bandiera, Barankay, and Rasul [2013]). 8 Taking these features of our setting into account, we predict that relative information will boost performance more than absolute information for the average player in our study. It is worth noting that this prediction describes an average effect, and that our analyses account for and explore variation in cross-sections of the sample. 7 Some concerns regarding relative information arise in particular contexts and do not, or are less likely to, apply to our setting. For example, the business press has suggested that subjectively measured ranks can be demotivating if employees feel that the ranking system is biased (CNBC [2019]). As in many performance reports (The New York Times [2015], McKinsey [2016]), the measures in our setting are objective, and so our study does not speak to dynamics that apply in the case of subjective measures. Another potential downside of relative information is that some employees feel that rankings are overly harsh or "cruel" (The Atlantic [2015]). The participants in our setting are accustomed to selection that depends on relative standing by objective measures (Güllich [2014]), and so are less likely to view rankings as overly harsh. 8 The average player in our sample is highly experienced and has immediate career concerns to achieve a high rank in the training. In terms of players' generally advanced level in their field, we show in table 1 that the average player began playing organized football about 15 years prior to the experiment. We also note that each player has ascended to a paid level of play and so has career concerns to secure a starting spot as aided by obtaining a high rank in the simulator.

H1:
Players who receive relative performance information outperform players who receive absolute performance information.
In cross-sectional analyses, we explore moderators of the effect of relative information. We first consider the role of task commitment. Relative performance information presents a descriptive norm, or implicit goal, to outperform peers (Azmat and Iriberri [2010], Blanes i Vidal and Nossol [2011]). Research in psychology has shown that task commitment is a determinant of willingness to exert effort toward a goal (Locke and Latham [2002]). We expect that players who are committed to outperforming peers, as indicated by the frequency with which they train, will benefit more from relative performance information and the implicit goal to outperform peers.
We also expect career concerns to positively moderate the effects of relative performance information on performance. Research suggests that performance effects of relative information will be greater in situations where it is more important to achieve a high rank in order to obtain payoffs (Tafkov [2013]). In our setting, these conditions plausibly apply when a player is competing with more teammates for the same starting lineup spot. 9 In such cases, managers can compare more players based on their rank in the simulator and be more selective when deciding on the starter. We predict a stronger response to relative information under these conditions.
We then consider how adding absolute information to relative will affect performance. On one hand, theories regarding motivation and learning suggest that there are distinctive benefits of relative and absolute information. Thus, it is plausible that providing both measure types will offer the performance benefits of each. 10 For instance, players who have access to absolute measures alongside relative measures should gain the motivational effects of peer-performance comparison and benefit from being able to cleanly observe how small changes in effort affect performance-changes that are unlikely to show up in "chunky" relative measures like ranks (Bonner and Sprinkle [2002]). In theory, this fea-9 As mentioned in the setting and institutional background section, the number of players competing for the same spot on a team is apparent to the participants in our study given that they arrive in teams to the simulator for their training sessions. 10 Hannan et al.
[2019] study feedback that varies in its temporal aggregation and detail. They test some conditions that contain absolute information alone and others that include absolute and relative information. They do not compare these to relative information alone. Their lab setting does not involve career concerns or explicit incentives for performance. The authors do not predict or report effects on performance. Instead they study how feedback directs effort toward or away from firm-desired allocations. We assess performance effects in the presence of strong implicit incentives where firm-desired allocation of effort across tasks is not a salient feature (i.e., managers express a strong interest in both passing speed and accuracy). Our study explores whether the provision of less information in a report is able to generate positive, multidimensional performance effects. ture of absolute-and-relative performance information allows for greater learning and heightened motivation. 11 On the other hand, two bodies of research on human cognition suggest reasons why players would perform worse when we add absolute measures alongside relative measures (Simon [1979], Einhorn and Hogarth [1981]). First, the literature on information overload suggests that performance reports can present enough measures to exceed an employee's cognitive limits. Studies have shown that difficulty processing large amounts of information can lead to stress and to disengagement with the content (Bawden [2001]). The addition of absolute measures alongside relative measures could provide a sufficient number of information cues to cause information overload. Empirical evidence of information overload, though, has generally found these effects in settings with a larger number of measures than we include in many of our study's performance reports (Miller [1956], Eppler and Mengis [2004]). Moreover, accounting research on the balanced scorecard suggests that managers can learn from large numbers of measures (Kaplan and Norton [1996], Campbell et al. [2015]).
A second body of research explains why the addition of absolute measures alongside relative measures would harm performance by inducing a shift in the weights that players place on information. This literature documents that individuals change the weights that they place on a set of cues if additional information or cues are included in a report (Birnbaum [1976], Gigerenzer and Goldstein [1996], Kruschke and Johansen [1999]). Prior research has shown evidence of this effect whereby information factors less in judgement and decision-making in the presence of other, less useful, cues (Kahneman and Tversky [1972], Kahneman [1974, 1983], Gigerenzer and Goldstein [1996]).
In accounting literature in particular, a number of studies have explained how employees reduce their weight on, and response to, performance measures due to their context alongside other measures. 12 In our setting, this suggests that players will reduce the cognitive weight they place on relative information when absolute measures are included alongside relative measures. We expect that this shift in weight away from relative information-and the associated decrease in peer-performance 11 Behavioral decision research has also shown that individuals who are inexperienced with feedback-based decisions are also more risk-seeking in the loss domain (Barron and Erev [2003]). In our setting, this finding suggests less-experienced players may be more likely to experiment with different techniques and explore "risky" passing strategies. Performance information that contains both absolute and relative performance measures could help this process of experimentation and learning.
12 For example, Lipe and Salterio [2000] demonstrate that information users ignore unique measures when they are provided with reports that also contain measures that are shared across business units. Lipe and Salterio [2002] and Cardinaels and van Veen-Dirks [2010] document that the groupings and categorizations of measures in a report can cause managers to weight nonfinancial numbers less than financial numbers. comparison-will, on average, inhibit the learning and motivational benefits of relative measures and harm performance. 13 Our second set of hypotheses introduces a test of whether adding absolute measures alongside relative measures harms performance and whether this is consistent with information overload. H2a predicts that the addition of absolute measures alongside relative measures will lead to worse performance than providing relative measures alone. H2b predicts that, holding constant whether a report conveys performance in absolute or relative terms, increasing the number of measures by a similar amount through the addition of detail will similarly harm performance. If we find support for H2a and H2b in our setting, our results would be consistent with the idea that this increase in the number of measures in a report induces information overload.
H2a: Players who receive relative performance information alone outperform players who receive absolute and relative performance information. H2b: Holding constant whether performance information is absolute or relative, players who receive aggregated performance information outperform players who receive detailed performance information.
We then examine whether the weight that players place on information shifts across measures when we add absolute measures alongside relative measures. As we discussed earlier in this section, a shift in cognitive weight away from relative information and associated peer-performance comparison would plausibly reduce performance on average in our setting. To test for this mechanism whereby adding absolute to relative information can reduce performance, we first predict with H3a that peer-performance comparison is positively related with performance in our setting (Tafkov [2013]). H3b then predicts that this involvement in peer-performance comparison will decrease when we add absolute measures alongside relative measures.
H3a: Players who are more involved in peer-performance comparison perform better. H3b: Compared to players who receive only relative performance information, players who receive both absolute and relative performance information are less involved in peer-performance comparison.
Due to variation in player types across our sample, we also allow for heterogeneity in performance effects. Theory suggests that workers at a developmental stage of their careers are focused on trying different strategies to improve their performance in absolute terms and find relative information less motivating (Vaughn [1936], Podsakoff and Farh [1989], Locke and Latham [2002], Goodman, Wood, and Hendrickx [2004]). For these players, the shift in weight away from relative and toward absolute information is likely to benefit performance. We address these portions in cross-sectional tests of our estimated performance effects.

subjects
We sourced the players in our study from 11 different professional clubs, with four clubs providing 79% of all participants. On average, the athletes in our sample started playing football at five years of age and have approximately 15 years of experience competing in the sport. Players in the sample frequently practice their skills in training and matches, with the average player training just over five times a week and competing in matches most weeks during the season. Seventy-two percent of the individuals in our sample were either starting or substitute players for a professional team in the 2018-2019 football season. Around 10% of players in our sample are goalkeepers, 30% are defenders, 44% are midfielders, and 16% are forwards. This reflects the standard composition by position of players on a team's roster (Bundesliga [2020]). Almost 85% of the sample had trained in the simulator prior to the experiment.

real-effort task
As part of their regular training, players had to complete a passing drill two times at the skills.lab simulator (see appendix A, exhibit 2). The training drill required a player to complete 12 passes. To complete a pass, the player received a ball fired sequentially from one of two ball machines located on the left and right sides of the simulator. After receiving the ball, the player then passed the ball to a dynamic target projected onto one of the walls of the simulator. The target consisted of a semicircle with a diameter of 1.85 m at the feet of a projected teammate (see appendix A, exhibit 3).
The specifications of the drill, including the speed of the passes and the target for passes, were adapted for our study. None of the players had practiced this specific version of the drill before. To prevent players from anticipating the location and movement of the target, a sequence of 12 starting and end positions for the target (one for each ball) was randomly drawn for each round. Players had a maximum of about 11 seconds to complete a pass before the next ball was fired from the simulator's ball machines. 14 Players completed the task once and then received performance information. Players then completed the drill a second time and again received information on their performance. This information showed only the player's performance on the most immediate iteration of the drill. Players did not exit the simulator between drills and were not exposed to other participants during treatment.

treatment conditions
We randomly assigned each player to receive one of six performance information treatments. On the measure-type dimension of treatment, we randomly assigned each player to receive either absolute, relative, or both absolute and relative performance information. On the information detail dimension of treatment, we randomly assigned each player to receive either a report of average performance for all passes during the round (i.e., aggregate information) or to receive a report of average performance for passes that required a right turn, average performance for passes that required a left turn, and average performance for all passes during the round (i.e., detailed information; see appendix A, exhibit 4). The breakdown by turn direction is relevant given that it requires the use of either the dominant or nondominant foot in a different combination, which can lead to slower times and less accuracy for the player's less preferred side. Skill at both types of passes is key to performance at the highest levels of football (Rein, Raabe, and Memmert [2017]). The procedure of the experiment is outlined in appendix A, exhibits 5 and 6.
Players in the Absolute information treatment arm received the following performance measures on passing accuracy and speed of execution: percent of passes on target, average distance of pass from the center of the target (in meters), fastest time handling ball (in seconds), and average time handling ball (in seconds). We describe these measures more fully in subsection 4.4 and in appendix B. Players in the Relative information treatment arm received the same performance measures but presented in the form of a performance rank for each of the measures (e.g., 19 of 32 for average time handling the ball, etc.). To arrive at this rank, each player's performance was compared to the performances of players from a common reference group. The reference group consisted of 31 professional or semiprofessional players who completed the same training drill in the simulator at skills.lab during the pre-experiment period. No player from the reference group participated in the experiment. The players in the reference group were of a similar age, ability, and level of experience as the subjects who participated in the experiment. Finally, players in the Absolute&Relative information treatment arm also received the same performance measures but presented in both absolute and relative terms.
In addition to receiving the same measures as the players in the Aggregate information treatment arm, players in the Detail information treatment received a breakdown showing the measures for passes that required the player to turn to their right and for passes that required the player to turn to their left. As such, players in the Detail information treatment arm received three times the number of performance measures as players in the Aggregate information treatment arm.
The interaction of these two treatment arms led to six treatments: Absolute x Aggregate, Absolute x Detail, Relative x Aggregate, Relative x Detail, Absolute&Relative x Aggregate, Absolute&Relative x Detail (see appendix A, exhibit 7). In line with prior field-based research on performance information, we do not have an experimental group that received no performance information (Casas-Arce, Lourenço, and Martínez-Jerez [2017], Song et al. [2018]). We take this approach because feedback is a strong organizational norm in our setting and players who train at skills.lab expect to receive at least some performance information after each drill. In terms of the total number of performance measures contained within the different treatments, Absolute

measures
Our study employs player performance data at the drill level. Our measures of performance on the drill are the main dependent variables. These measures capture the accuracy and speed of the passes completed by each player. To describe the sample and address effect moderators and mechanisms, we gathered demographic data and self-reported measures of responses to the performance information.
In our analysis, we use the two measures of passing accuracy reported to players while in the training simulator. The first, Hitrate, is defined as the proportion of passes on target that a player completed during the drill. Distance is a continuous measure of accuracy and is defined as the average distance (in meters) from the center of the target for the 12 passes that the player completed during the drill. The lower the score on this measure, the more accurate the player's passing.
We also use the two measures of ball-handling and passing speed recorded by the training simulator and included in the performance information. Fast_Time is the quickest time (in seconds) it took the player to complete a pass during the drill. Avg_Time is the average time (in seconds) it took the player to complete a pass over the 12 passes completed during the drill. For both of these measures, lower times reflect quicker ball-handling and passing.
Our analysis also uses player demographics and responses to surveys. We observe the age, gender, height and weight, playing attributes, experience, and language of all subjects in our sample. Using a post-survey questionnaire, we collected information on peer-performance comparison and on perceptions of the quantity of information provided.
Appendix B contains a full list of variable definitions. We use Hitrate, Distance, Fast_Time, and Avg_Time as the main dependent variables in our experiment. This table presents the descriptive statistics for measures of performance, demographic variables, and post-experiment survey questions. The performance measures are reported as in Round 2, after the delivery of performance information. The factor analysis used to compute Comparison yields a score that is standardized to mean zero. An analysis of covariate balance shows that there are no statistically significant differences among the six assigned treatment groups on performance prior to the delivery of performance information or on demographic variables at the 0.05 level. Appendix B contains a full list of variable definitions. formance measures, player demographics, and post-experiment survey responses. 15

Analysis
We employ ordinary least squares (OLS) regression, with full interaction of treatments, to test for effects. We include controls for initial performance, league level, and player demographics in line with the guidance from econometric research that, in field experiments where randomization is used, "it is customary to control for covariates to correct for chance associations between treatment status and applicant characteristics and to increase precision" (Angrist and Pischke [2008]). Field-experimental re- 15 An analysis of covariate balance shows that there are no statistically significant differences among the six assigned treatment groups on pre-treatment performance or on demographic variables at the 0.05 level. We note that players in all treatments report considering comparison at multiple times during the experiment. Those players in the relative treatment arm report considering comparison significantly more often than those in the other treatments. This effect represents a 0.98-point increase (p < 0.01) in comparison on the construct's one-to seven-point scale.

Effect of Relative Information and Detail in Performance Report
(1)
• denotes that the estimated effect of Relative is of greater magnitude than the estimated effect of Absolute&Relative at at least the 0.1 level. Fig 1.-Hitrate by type of measures in performance report. This figure displays performance on the measure Hitrate, or proportion of passes that were on target, from the second round of the passing drill after the provision of performance information. Performance is presented by the type of measures in the report-absolute only, relative only, or combined absolute and relative. The symbols represent the mean for each treatment type and the lines represent 90% confidence intervals.
search in accounting and economics has applied OLS in the same manner (Kreuger and Whitmore [2001], Angrist and Pischke [2008], Casas-Arce, Lourenço, and Martínez-Jerez [2017]). 16 Table 2 contains the main performance effects and the fully interacted performance effects of our 3 × 2 treatment design for each of our four 16 In our analysis, we follow Kreuger and Whitmore [2001] and Banerjee et al. [2010] by comparing levels after treatment. Our results also include initial performance as a control when regressing performance on treatment conditions. This approach is widely employed in studies in economics that use field experiments (Duflo, Kremer, andRobinson [2011], Fryer et al. [2020]). In untabulated robustness tests, we also conducted each of our main analyses by comparing residualized gain scores (Cronbach and Furby [1970], Hendrix, Carter, and Hintze [1978], Knapp and Schafer [2009]). We find that our estimates are of similar magnitude and statistical significance with this approach as in our main specification. -Distance by type of measures in performance report. This figure displays performance on the measure Distance, or average distance in meters between passes and the target, from the second round of the passing drill after the provision of performance information. Performance is presented by the type of measures in the report-absolute only, relative only, or combined absolute and relative. The symbols represent the mean for each treatment type and the lines represent 90% confidence intervals. performance measures. The omitted treatment condition is Absolute x Aggregate information. Thus, the coefficients on Relative information and on Absolute&Relative information represent the estimated effect of those treatments as compared to Absolute information. By the same principle, the coefficients on Detail information represent the estimated effect of that type of information as compared to Aggregate information. For each dependent variable, we run the model with main effects-Relative, Abso-lute&Relative, and Detail-in the odd-numbered columns. We run the fully interacted model in the even-numbered columns. 17 Fig 3.-Avg_Time by type of measures in performance report. This figure displays performance on the measure Avg_Time, or average time in seconds taken to complete a pass, from the second round of the passing drill after the provision of performance information. Performance is presented by the type of measures in the report-absolute only, relative only, or combined absolute and relative. The symbols represent the mean for each treatment type and the lines represent 90% confidence intervals.
For all of our measures of performance, and consistent with H1, there is a statistically significant performance benefit of providing relative rather than absolute information. The statistically significant coefficients on Relative in table 2, columns 1, 3, 5, and 7, document this result. In reading these coefficients, note that a decrease in Avg_Time or Fast_Time, represented by a negative coefficient, reflects an improvement in time-related performance. Similarly, a decrease in Distance, also represented by a negative coefficient, reflects an improvement in passing accuracy. The magnitudes of these performance effects are substantial. Providing relative information alone leads to at least a 0.4-standard-deviation increase in performance, depending on the measure, as compared to the performance of players who receive absolute information alone. Performance is presented by the type of measures in the report-absolute only, relative only, or combined absolute and relative. The symbols represent the mean for each treatment type and the lines represent 90% confidence intervals.
H2a predicted that performance would be better when the performance report contained only relative information. In table 2, we use a circle symbol to denote that the effect of Relative is statistically significantly stronger than the effect of Absolute&Relative for Hitrate, Distance, and Avg_Time. These results support H2a. In subsequent analyses, we explore why there is not a statistically significant difference between the coefficient on Relative and the coefficient on Absolute&Relative for the measure Fast_Time. Figures 1-4 show the mean and a 90% confidence interval for each performance measure (Hitrate, Distance, Avg_Time, and Fast_Time) after the provision of performance information, by assignment to Absolute, Relative, or Absolute&Relative information. These figures show that the Relative treatment led to higher performance than the Absolute&Relative treatment, which in turn generally led to better performance than the Fig 5.-Hitrate by type and number of measures in performance report. This figure displays performance on the measure Hitrate, or proportion of passes that were on target, from the second round of the passing drill after the provision of performance information. Performance is presented by the type of measures in the report-absolute only, relative only, or combined absolute and relative-and the number of measures in the report. The symbols represent the mean for each treatment type and the lines represent 90% confidence intervals. Absolute treatment. 18 As shown across figures 1-4, this pattern in the results holds for measures of both passing accuracy (Hitrate, Distance) and passing speed (Avg_Time). Research on multidimensional performance has documented that speed and accuracy are typically competing dimensions of performance and that it is challenging to generate improvement on both at the same time (Bonner and Sprinkle [2002]). 19 Our study documents a means of improving overall performance across these dimensions.
With regard to the Detail condition, the coefficients on this term in all columns of table 2 document that we find no statistically significant evidence that increasing the number of measures per se while holding con- Fig 6.-Distance by type and number of measures in performance report. This figure displays performance on the measure Distance, or average distance in meters between passes and the target, from the second round of the passing drill after the provision of performance information. Performance is presented by the type of measures in the report-absolute only, relative only, or combined absolute and relative-and the number of measures in the report. The symbols represent the mean for each treatment type and the lines represent 90% confidence intervals.
stant the scope of performance information-that is, as absolute, relative, or combined absolute and relative-leads to worse performance. This fails to support H2b, and suggests that the benefit of restricting a report to contain relative information is not a function of the quantity of measures displayed per se. Furthermore, if the negative effect of Absolute&Relative, as compared to Relative, were driven by information quantity, we would expect the negative effect to be moderated by Detail, which represents an increase in information quantity. The coefficients on the interaction terms in table 2 generally fail to provide evidence to support this.
We use figures 5-8 to illustrate that, across measures of passing speed and accuracy, we observe that there is no pattern whereby an increase in the number of measures in a report harms performance. Specifically, for the given dimension of performance shown in each of figures 5-8, we do not see a negative trend in performance as we move further along the X-axis by increasing the number of measures in the report. Rather, we see differences by symbol color, which represents whether the player was assigned to the Absolute, Relative, or Absolute&Relative condition. This demonstrates that the pairing of absolute and relative performance information in particular, Fig 7.-Avg_Time by type and number of measures in performance report. This figure displays performance on the measure Avg_Time, or average time in seconds taken to complete a pass, from the second round of the passing drill after the provision of performance information. Performance is presented by the type of measures in the report-absolute only, relative only, or combined absolute and relative-and the number of measures in the report. The symbols represent the mean for each treatment type and the lines represent 90% confidence intervals. rather than simply any increase in the number of measures included in the report, drives the performance effects that we observe. Table 3 uses survey data to further document the lack of evidence to support the idea that providing both absolute and relative information provides too many measures per se. In the post-experiment survey, reports that the number of measures was "too many" are not statistically significantly affected by whether players received the combination of absolute and relative information instead of either absolute or relative information alone. Column 1 shows this result for the full sample. Columns 2 and 3 show that this result is similar whether the information displayed was aggregated or detailed.
To address another potential mechanism for the benefit of restricting performance reports to show relative measures alone, we test whether providing only relative measures increases the intensity of peer-performance comparison. We follow prior research on relative performance information and peer-performance comparison by conducting a factor analysis of questions regarding self-reported involvement in peer-performance comparison (Tafkov [2013]). We use the resulting factor, Comparison, as our Fig 8.-Fast_Time by type and number of measures in performance report. This figure displays performance on the measure Fast_Time, or minimum time in seconds taken to complete a pass, from the second round of the passing drill after the provision of performance information. Performance is presented by the type of measures in the report-absolute only, relative only, or combined absolute and relative-and the number of measures in the report. The symbols represent the mean for each treatment type and the lines represent 90% confidence intervals. measure of peer-performance-comparison involvement. Table 4 shows that there is a statistically significant effect of Relative, as compared to both Ab-solute&Relative and Absolute. Table 5 documents that our measure of peerperformance comparison is positively and statistically significantly related to performance for all four of our measures. These results support H3a and H3b, suggesting that restricting information to relative terms alone yields a performance benefit through increased engagement in peer-performance comparison.
To further examine our results, we use cross-sectional and supplemental analyses. The first of these analyses examines why we find statistically significant improvement resulting from the Absolute&Relative treatment for the measure Fast_Time. Although the other three performance measures are averages, and so would require consistent effort for a player to improve, Fast_Time is an extremum, and could be boosted with a trade-off that involves making a few passes with low accuracy but high speed. This strategy would dramatically improve reported performance on Fast_Time while having a relatively small effect on reported performance by accuracy, which is an average of 12 passes. Studies of trade-offs among measures have es-

T A B L E 3 Effect of Providing both Absolute and Relative Information on Reports that the Number of Measures is Too Many
(   tablished widespread evidence that employees use such strategies, sacrificing performance to a small degree by one measure when this leads to a larger or preferred effect on another measure (Asch [1990], Courty and Marschke [2004]). Although less effective than Relative in our setting, Abso-lute&Relative may be sufficiently motivating to encourage players to opportunistically trade off passing speed and accuracy, a strategy that does not require sustained effort to generate reported improvement.
To examine this idea empirically, we test whether there is a change in the relationship between Hitrate and Fast_Time after we provide players with relative information. If players trade off performance outcomes by making a small trade in terms of a slightly reduced Hitrate to achieve a significantly reduced Fast_Time, we would expect to see a stronger (positive) relationship between these two variables in Round 2 than in Round 1.   *, **, and *** denote significance at the 0.1, 0.05, and 0.01 levels, respectively. • denotes that the estimated effect of Relative is of greater magnitude than the estimated effect of Abso-lute&Relative at at least the 0.1 level.
Round 2. This interaction only occurs for the measure of speed that can be strategically manipulated-Fast_Time-and not for the measure of speed that requires sustained performance to improve-Avg_Time. Moreover, the interaction does not occur unless the report includes relative information. Collectively, these results suggest the effect of relative information on Fast_Time works partly through a trade-off between accuracy and Fast_Time. Table 7 provides a test of how motivation to achieve immediate career advancement affects the relevance of information. When more teammates who are competing to start at a given position are participating in the experiment, the team's managers will have more players to compare on simulator performance when selecting the starters. We use the number of teammates who are competing for the same starting spot as the given player, scaled by the number of places in the starting lineup at that position, to proxy for immediate career concerns. In This table presents estimates of the relationships between Comparison and the performance measures Hitrate, Distance, Avg_Time, and Fast_Time measured in rounds 1 and 2 of the passing drill. Comparison is the result of a factor analysis of questions regarding the degree of peer-performance comparison that the participant felt involved in during the experiment. t-Statistics are based on standard errors clustered at the participant level and are reported in brackets below each coefficient.
*, **, and *** denote significance at the 0.1, 0.05, and 0.01 levels, respectively. Position Depth and on Absolute&Relative x Position Depth offer evidence that the effects of relative information are stronger when career concerns, as proxied by Position Depth, are greater. 20 This suggests that much of the benefit of Relative information in our setting is due to the presence of career concerns-that is, competition for starting positions, rather than self-image concerns alone. Table 8 provides a test of how task commitment influences responses to relative information. To proxy for a player's task commitment, we use the measure Training, or the number of days per week that the player reports playing football. We find that both the Relative and Absolute&Relative treatments boost performance more on the dimensions Hitrate, Distance, and Avg_Time for players who train as or more frequently than the median player in our sample than for players who train less than the median. We do not find these differences in our results across partitions when, in unt- This table presents results of a test for whether there was a change in the relationship between Hitrate and Fast_Time or between Hitrate and Avg_Time that occurs between rounds 1 and 2 of the passing drill, after the provision of performance information. The coefficient on the interaction terms represents whether there was a change in the relationship after the performance information was provided. t-Statistics are based on standard errors clustered at the player level and are reported in brackets below each coefficient.
*, **, and *** denote significance at the 0.1, 0.05, and 0.01 levels, respectively. abulated analyses, we partition the sample and examine players by whether they had above-or below-median initial performance. This suggests that task commitment, rather than associated stronger initial performance, is the source of variation in our results. This analysis provides evidence that relative information is particularly beneficial in the presence of greater task commitment. Table 9 examines how our effects depend on whether a player is in a developmental stage of his or her career. Specifically, we test our effects among players who are within two years of the minimum age for their league and who are in the lower half of experience with their parent club within that group. Among these players, the Absolute&Relative condition outperforms the Relative condition-that is, the coefficient on Ab-solute&Relative is statistically significant for all four dimensions of performance while the coefficient on Relative is not. Despite the reduced power we have to compare treatment effects in this sample of 40 trainees, we find that the difference in estimated effects of Relative and Absolute&Relative are statistically significant for two of those dimensions of performance, spanning both speed and accuracy. These results demonstrate cross-sectional variation in information demands and effects. In particular, the typical player in our sample benefits when we isolate relative information, but developing players benefit from the addition of absolute information alongside relative information.
• denotes that the estimated effect of Relative is of greater magnitude than the estimated effect of Absolute&Relative at at least the 0.1 level. † denotes that the estimated effect of the given treatment type (Relative or Absolute&Relative) is different in the median-or above-median training frequency sample than in the below-median training frequency sample at at least the 0.1 level.

T A B L E 9
Effect of Relative Information in Performance Report, Partitioned by Level of Experience (1) (3) based on heteroskedastic-robust standard errors and are reported in brackets below each coefficient. *, **, and *** denote significance at the 0.1, 0.05, and 0.01 levels, respectively.
• denotes that the estimated effect of Absolute&Relative is of greater magnitude than the estimated effect of Relative at at least the 0.1 level. † denotes that the estimated effect of Absolute&Relative is different in the Trainee sample than in the Advanced sample at at least the 0.1 level.

Conclusion
We use a field experiment in a highly competitive setting-professional European football-to compare the performance effects of providing absolute, relative, or both absolute and relative performance information. Although performance information interventions typically use relative and absolute information either alone or in some combination, ours is the first study to assign this set of conditions in a randomized controlled trial to assess performance effects. Our analysis extends accounting research on how firms can deliver internal performance reports in a way that best affects performance. We also contribute to management accounting literature that explores how employees place weight on and respond to different types of performance measures. Contrary to the notion that increased information is more advantageous for performance, we find, on average, that restricting performance information to relative measures alone yields the best effects.
Our survey data shed light on the mechanism for this result. We show that adding absolute information alongside relative information causes workers to place less weight on the relative measures and more weight on the absolute measures, leading to less peer-performance comparison. We also use demographic data to understand heterogeneity in responses to the performance information we provide. We find evidence that there are greater benefits of relative performance information in the presence of greater task commitment and stronger career concerns. For substantially lessexperienced players, who are engaged in learning and skill-development, we show that providing the broader information set-both absolute and relative measures-benefits performance. Thus, in addition to documenting an average performance benefit of limiting the information in a report in our setting, we demonstrate the potential to customize the provision of absolute and relative information based on a worker's characteristics and context in order to achieve the strongest performance effects. appendix a: field site and treatment Exhibit 1.-Layout of the football training simulator skills.lab This exhibit shows the layout of the skills.lab football training simulator. The integrated projectors display the passing target onto the walls of the simulator. One of the four ball machines is shown on the right side of the simulator. For scale, a football player is shown in the right corner of the simulator.
Exhibit 2.-The passing drill: "Passes after controlling the ball with a 180°turn" This exhibit describes in detail the passing drill. In the first step of the task, the football player waits for a ball to be passed from one of the simulator's ball machines. In the second step of the task, the player must control the ball and locate the passing target. In the final step of the task, the player passes the ball to the target.  This exhibit shows the fully interacted 3 × 2 design employed in the experiment. The measure-type treatment arm contains three treatments: Absolute, Relative, or Absolute&Relative. The information-detail treatment arm contains two treatments: Aggregate or Detail.
Exhibit 5.-On-site procedure of the field experiment This exhibit shows in detail the experimental procedure. In the first step of the procedure, participants registered for study participation with Anton Paar SportsTec. In the second step, participants read and signed the terms of use of skills.lab and provided consent. At this step, participants were randomly assigned to one of the treatment groups. Participants were not informed of the details of the experiment nor the specific treatment to which they were assigned. Participants next watched a standardized video briefing that explained the experimental task and procedure. In the fourth step, participants took part in a demographic survey. This survey was consistent with the regular data entry process employed by the site when collecting information from users. In the fifth step, participants performed a warm-up drill. After this, each participant completed the competitive passing drill, where each player faced a total of 12 balls. This was followed by a short cool-down period after which the player received performance information. This report was displayed on to the walls of the simulator. After receiving this information, participants again completed the passing drill and received a second round of performance information. Following the experimental drills, participants exited the simulator and individually completed the post-experimental survey. Each participant was then debriefed. As compensation for taking part in the study, participants could play five trainings free of charge at skills.lab. This exhibit shows the flow of participants through the training facility during the field experiment. The experiment procedure was designed such that participants could not interact with each other between steps 5 and 8 of the experiment when we collected demographic information, delivered the intervention, measured performance, and collected post-survey responses.

Variable Definitions
Variables Description

Performance measures Hitrate
The proportion of the total amount of 12 balls played in one round of the task where a participant has hit the target area.

Distance
The mean distance from a participant's passes to the center of the target for all 12 balls played in one round of the task, measured in meters.

Avg_Time
The mean time that it took a player to complete a pass averaged across all 12 balls played in one round of the task, measured in seconds.

Fast_Time
The fastest time that it took a player to complete a pass among the 12 balls played in one round of the task, measured in seconds.

Treatment variables Absolute
An indicator variable equal to 1 if the participant is assigned to the treatment group that receives absolute performance information on passing accuracy and speed of execution in the experiment.

Relative
An indicator variable equal to 1 if the participant is assigned to the treatment group that receives performance ranks relative to a reference group on passing accuracy and speed of execution in the experiment.

Absolute&Relative
An indicator variable equal to 1 if the participant is assigned to the treatment group that receives absolute performance information and performance ranks relative to a reference group on passing accuracy and speed of execution in the experiment.

Aggregate
An indicator variable equal to 1 if the participant is assigned to the treatment group that receives a summary result for all 12 balls played in one round of the task in the experiment.

Detail
An indicator variable equal to 1 if the participant is assigned to the treatment group that receives a summary result for all 12 balls and two subcategories (turn to pass left and right) for six balls each played in one round of the task in the experiment.

Post-experiment survey Too Many Measures
Response to survey question, "Based on the training you just played, please answer the following question: I would prefer to see less results." 1 = Strongly disagree; 4 = Neither agree or disagree; 7 = Strongly agree.

Variables Description
Prior Performance Performance in the first round of the drill on the dimension of performance (Hitrate, Distance, Avg_Time, or Fast_Time) that is the dependent variable in the given model.

Position Depth
The number of participants from a player's team who are competing for the same position on the team as the player (goalkeeper, defender, midfielder, striker) divided by the number of places in the starting lineup at that position.

Support for Prediction
Prediction Plus Minus Conclusion

H1
Players who receive relative performance information outperform players who receive absolute performance information.
Relative measures enable peer-performance comparison. This increases competition and enhances motivation (Lazear and Rosen [1981], Tafkov [2013]). Relative measures also provide an inherent benchmark of one's peers. This gives the information more evaluability, which aids in learning (Hsee and Zhang [2010], Song et al. [2018]). For developing players, who are less likely to be in a position to outperform their peers, relative measures and peer-performance comparison can be demotivating (Goodman, Wood, and Hendrickx [2004]). Absolute measures can motivate these players to improve over their own prior performance (Podsakoff and Farh [1989], Locke and Latham [2002]).
Advanced players and those facing substantial competition for a starting spot on their team are more likely to be competing for top positions and so benefit from the motivation that relative information provides. Given that these types of players make up the majority of our sample, we expect that, on average, players in our setting who receive relative measures will outperform players who receive absolute measures. We test for different responses among developing players.

Support for Prediction
Prediction Plus Minus Conclusion

H2a
Players who receive relative performance information alone outperform players who receive absolute and relative performance information.
Presenting absolute measures alongside relative measures may harm performance for two reasons. First, adding absolute measures alongside relative measures could induce information overload (Miller [1956], Eppler and Mengis [2004]). Second, adding absolute measures alongside relative measures could cause players to reduce the cognitive weight they place on the relative measures, reducing peer-performance comparison and its associated performance effects (Birnbaum [1976], Gigerenzer and Goldstein [1996]).
Absolute and relative measures each have benefits as described in the motivation for H1. Players who receive both absolute and relative measures would plausibly receive the benefits of each type of information. Studies have generally found that information processing constraints bind in settings with a larger number of measures than we include in many of our study's performance reports (Miller [1956], Eppler and Mengis [2004]). In light of the two avenues whereby adding absolute information alongside relative measures could harm performance-through information overload and changes in the cognitive weighting of information-we expect that adding absolute information alongside relative information will negatively affect performance. We conduct follow-on tests to identify if support for this hypothesis works through either or both of these avenues.

Support for Prediction
Prediction Plus

Conclusion
H2b Holding constant whether performance information is absolute or relative, players who receive aggregated performance information outperform players who receive detailed performance information.
Added detail in a report can present an overwhelming number of measures that exceeds the information user's cognitive limits. This, in turn, can induce information overload and associated disengagement from the source of information (Miller [1956], Bawden [2001], Eppler and Mengis [2004]).
Studies have generally found that information processing constraints bind in settings with a larger number of measures than we include in many of our study's performance reports (Miller [1956], Eppler and Mengis [2004]).
This test helps us to examine whether adding absolute measures alongside relative measures harms performance because the number of measures becomes too great. If so, we would expect a similar negative effect when we increase measure quantity in a report by a similar magnitude through adding detail. H2b proposes this hypothesis, although prior research suggests that the number of measures added may not be large enough to cause information overload.

H3a
Players who are more involved in peer-performance comparison perform better.
Peer-performance comparison increases competition and should relatedly motivate players (Lazear and Rosen [1981], Tafkov [2013]). For developing players, who are less likely to be in a position to outperform their peers, relative measures and peer-performance comparison can be demotivating (Goodman, Wood, and Hendrickx [2004], Bandiera, Barankay, and Rasul [2013]).
Advanced players and those with more competition for a starting spot on their team are more likely to experience a performance benefit associated with peer-performance comparison.
Given that these types of players make up the majority of our sample, we expect a positive association on average between peer-performance comparison and performance in our setting.

Support for Prediction
Prediction Plus

H3b
Compared to players who receive only relative performance information, players who receive both absolute and relative performance information are less involved in peer-performance comparison. Players are likely to place greater cognitive weight on relative measures when these cues are presented alone rather than alongside absolute measures (Birnbaum [1976], Gigerenzer and Goldstein [1996]). Players are, relatedly, likely to be more involved in peer-performance comparison when they place greater cognitive weight on relative measures as these measures enable comparison.
Presenting relative measures alone, rather than alongside absolute measures, keeps players focused on peer-performance comparison.