Abstract
This article examines whether the online context affects tournament theory incentive mechanisms. It investigates the impact of prize structure on player performance in eSports under online and offline conditions. Using a quasi-experimental approach with data from the Counter-Strike Global Offensive eSports scene, the study finds that high prize distributions increase individual output in both settings, but the incentive effect is weaker in online events.
Introduction
Tournament theory suggests a mechanism of incentive for contestants to increase their efforts (Connelly et al., 2014; Lazear & Rosen, 1981). According to Lazear and Rosen (1981) the effective reward structure is based on relative rank rather than absolute levels of output. They showed that the tournament-like incentive structure increases individual performance and may induce non-monetary incentives in terms of intrinsic motives to be highly ranked. Previous studies supported tournament theory empirically in different business and sports settings (see e.g., Cai et al., 2021; Chen et al., 2011; Mao, 2023). Moreover, the last few decades showed that tournament-based contracts can be more effective in eliciting high efforts. They gained popularity and outperformed other types of contracts such as piece-rate and fixed wage contracts because of their flexibility (Sheremeta, 2016). Tournament principles are used in promotion, assigning bonuses and stimulating employee development. At the same time, the efficiency of the application of tournament principles depends on context, for example, on the environmental uncertainty (Nalebuff & Stiglitz, 1983). Consequently, ongoing research interest is driven by insights that appear in different contexts by tournament theory application.
The matter of context is proved by the evidence that different settings determine nuances in tournament theory application, reflected in the presence and the magnitude of spread effect (Connelly et al., 2014). We suppose that the work arrangement may also contribute to the incentive effect of the tournament-like reward system. To the best of our knowledge, empirical evidence for online conditions, with regard to tournament theory, is still missing. Meanwhile, remote work or online operations have become a common routine due to the pandemic period and continue to coexist with offline conditions even though the pandemic is over (Aksoy et al. 2022). Remote work differs from offline work in several key areas (George et al., 2022). Firstly, the issue of location is significant, as offline work requires employees to be physically present in a specific location, whereas remote work allows individuals to work from anywhere with an internet connection. Secondly, flexibility in terms of hours and work schedule is a distinguishing factor between the two. Thirdly, the reliance on digital tools for communication and collaboration sets remote work apart from offline work, which often involves in-person interactions. Finally, the level of technical equipment required for task fulfillment may be lower in the context of remote work. These differences have a direct impact on employee perceptions, engagement and work productivity. George et al. (2022) found that the shift to remote work has led to a decrease in perceived stress levels, but has also resulted in an overall increase in health challenges. Considering these findings, it can be inferred that remote work conditions pose challenges to incentive mechanisms in terms of employee productivity (Barrero et al., 2021; Rietveld et al., 2021). Therefore, it appears relevant to explore tournament theory in the new conditions and reveal the difference if any exists with the offline conditions. This study fulfills this knowledge gap contrasting online and offline eSports tournaments. We compare the relationship between prize spread and player performance in offline and online competition. The research question put forward in this study is as follows: “Does the tournament-based reward system incentivize individual performance differently based on the offline or online competition format?”
We address this research question using the sports setting. Considering that professional sport inherently is oriented to high achievements and strong competition, empirical evidence of tournament theory is rich and profound. The incentive effect of convex price structure was revealed in professional sports such as golf (Ehrenberg & Bognanno, 1990), tennis (Gilsdorf & Sukhatme, 2008), auto racing (Humphreys & Frick, 2019), and eSports (Mao, 2023; Shenkman et al., 2022). Being a nascent professional sport, eSport is getting scholars’ attention (Coates et al., 2018; Mao, 2021; Novak et al., 2020; Parshakov et al., 2019) including prize design of tournaments (Mao, 2023; Shenkman et al., 2022).
The rapid development of video gaming from entertainment leisure activity to professional eSports is reflected in eSports earnings and the growing expertise of eSports players (Reitman et al., 2019). Projections indicate that global revenues in the eSports market are set to surpass $3.8 billion by 2023 (Statista Market Insights, 2023), with tournament prize pools expected to reach $40 million (Esports Charts, 2023). This expansion has pushed competition among teams and attracted attention from investors, sponsors, and scholars alike. Recent papers such as Shenkman et al. (2022) and Mao (2023) studied eSports settings with regard to tournament theory. In particular, Shenkman et al. (2022) revealed the presence of convex prices in individual and team eSports tournaments. Meanwhile, Mao (2023) provided empirical evidence that larger prize differentials stimulate greater effort. We seek to contribute to the ongoing discussion by extending empirical knowledge on the link between prize dispersion and eSport individual performance using the specific setting of Counter-Strike:Global Offensive (CS:GO) videogame. Considering that eSports has a competitive nature due to the embedded aspiration to win (Mao, 2021) and previous empirical results (Mao, 2023), we put forward the first hypothesis:
H1: The prize spread has a positive and significant impact on individual performance in eSport.
The second pool of literature addressed in this study concerns online conditions or remote workplace arrangements. Since Jack Niles (1973) proposed the term “telecommuting,” almost three decades past before IBM permitted 5 employees to work from home. The first studies on the pros and cons of remote work are dated after 2010. For example, Bloom et al. (2013) presented a 9-month study of 16,000 Chinese employees, which demonstrated a 13% increase in productivity due to remote working. Everything changed with the COVID pandemic, which forced online work and dramatically challenged human resource management. In 2021, Mckinsey & Company reported that 52% of employees would prefer a more flexible working model after the pandemic was over. Simultaneously, the empirical evidence regarding the impact of remote work on employee productivity is inconclusive (Aksoy et al. 2022; George et al., 2022; Rietveld et al., 2021). These inconsistencies may be attributed to the specific characteristics of tasks, individual perceptions, and levels of engagement within the remote work setting. Differences between remote and in-office work settings are shaped by a mix of psychological, economic, technological, and behavioral factors. George et al. (2022) highlight potential sources of both support and challenges when working from home, including job characteristics, personal resilience, social support, and organizational values. As a result, the question of whether a remote or in-office work environment is more conducive to employee productivity remains open-ended.
Different sport settings were also investigated, considering the impact of on-line format of competition on players’ performance. Thus, the study by Künn et al. (2020) investigating professional chess tournaments, found the adverse effect of teleworking on workers performing cognitive tasks. In the case of eSports events, the format of competition matters as well. This distinction primarily revolves around two key aspects: location and technical equipment considerations. Online tournaments are commonly conducted over the internet, allowing players from diverse locations to participate. While these tournaments may feature smaller prize pools, they are susceptible to issues such as internet connectivity and potential cheating. Conversely, offline tournaments or LAN (local area network) events take place in a physical setting where all participants are physically present. These events offer a controlled environment, minimizing the risk of cheating. Note that the in game communication is the same in both type of events: speed and clarity of communication are indeed critical for a team's success. Players typically wear specialized earphones that effectively block out background noise from the crowd during LAN events. This ensures that communication remains clear and uninterrupted, regardless of the environment.
LAN events often boast larger prize pools and attract a substantial viewership. According to the chat of the CS:GO community “…when you're in LAN the environment plays a lot: the sound, the feeling of being watched, the stress of seeing your teammates and your opponents right next to you” (reddit.com). Off-line conditions generate more pressure of having to perform, both for team and individual players”. The players highlighted various technical challenges inherent in the online format that impact their performance, such as disparities in hit frequency caused by connectivity issues. Therefore, we expect lower productivity in online conditions and put forward the second hypothesis:
H2: Online tournament conditions decrease the individual performance of eSports players.
The main focus of this article concerns the interaction effect of online conditions and prize spread on individual performance, assuming the workplace transformation challenges incentive mechanism for individual productivity. Firstly, we expect that the online context matters for the tournament-like reward structure. It seems that spreads and prizes should be different depending on the conditions. Reviewing literature (Shenkman et al., 2022) and gamers’ chats, we conclude that on-line events are less risky and less prestigious. Therefore, we might expect that for on-line tournaments, lower spreads and lower total prizes are expected. Partly it corresponds with Nalebuff and Stiglitz (1983) and Lazear and Oyer (2007), who claim that risky industries should have higher spreads. Secondly, we suppose that online tournaments’ conditions decrease the incentive effect of prize dispersion. Our intuition is based on the environmental characteristics determined by the online format and generational perceptions with regard to remote work. Taking into account that online format is less prestigious, the players will demonstrate a lower performance (in comparison with offline performance) for the same spread. Considering the generational perceptions of online work arrangements, Report Global Workplace Analytics (2022) showed that generation Z expect remote work should pay more. Therefore, eSports players, for the most part representing digital natives, will demand a higher spread in online tournaments for the same efforts. In other words, the same spread induces less individual performance in online conditions. The third hypothesis is as follows:
H3: Online tournament conditions decrease the incentive effect of a convex prize structure on the individual performance of eSports players.
We seek to contribute to the literature surrounding sports economics, revealing empirical knowledge on tournament theory in online conditions. Moreover, eSports offers massive opportunities for collecting and analyzing data on digital natives, their behavioral patterns and motivation sources (Reitman et al., 2019). We aim to derive practical implications for the personnel economics field. Taking into account that digital natives are becoming a notable labor force, it seems relevant to study the challenges faced by personnel economics in terms of employee incentives, compensation and turnover (Lazear & Oyer, 2007). Should these human resource management practices be different or not? This study addresses the aspect of incentives and tests tournament theory predictions. Further, we present the data and empirical method used in the study. Our findings, discussion and conclusion finalize this article.
Data
CS:GO is a first-person shooter video game that has gained significant popularity in the gaming and esports communities (Bednárek et al., 2018). The game pits two teams against each other: the Terrorists and the Counter-Terrorists. Both teams are tasked with completing objectives or eliminating the enemy team. The game features a wide variety of weapons and equipment that can be purchased with in-game currency. The popularity of this game has led to the development of various tournaments and leagues, both online and offline.
We gather information about all major CS:GO tournaments from HLTV.org, as the leading website with statistics on this particular game. For the purpose of the analysis, we restrict the dataset only to single-elimination tournaments. In this tournament format, all contestants are divided into pairs, after each game, the loser leaves the tournament, while the winner goes to the next round until the end of the tournament. At the beginning of each match, players understand how much money they can win (or win nothing) compared with the first-placed prize, a critical assumption for testing tournament theory.
CS:GO tournaments vary in size, typically featuring anywhere from a few dozen to several hundred teams. Major tournaments, can attract the top teams from around the world and involve extensive qualification processes. Prize money in CS:GO tournaments is usually shared equally among team members. Teams often consist of five players, and the total prize money won by a team is divided among them. However, the exact distribution can vary depending on team agreements and contracts.
The dataset covers the time period from June 2017 till March 2022 and includes 1,272 tournaments involving 1533 players across 1382 teams. The majority of these tournaments, 57%, were conducted online, while the remainder took place offline in a LAN setting. Only 54 tournaments were structured as “winner-take-all,” where a single winner claims all the prize money, while the remaining tournaments featured a more complex prize distribution. Among the “winner-take-all tournaments,” 13 were held in LAN format.
To empirically test tournament theory, we need to construct variables that reflect the different prize structures of tournaments. For this purpose, first, we constructed the spread as a classical prize spread following Rosen (1986). The spread is calculated as the difference between the prize for first place and the prize that the team will get if it loses a particular match. The average spread in the sample is $0.03mln with 0.09 as the standard deviation (Table 1). But the sample consists of tournaments with a wide range of prize pools from few hundred dollars to $1.5 mln. Intuitively, the spreads in tournaments with highly different prize pools should affect players’ performance differently. To solve the issue, we construct a normalized spread as the ratio of spread to the prize pool. Notice, the normalized spread varies from 0.03 to 1 as in the situation of the winner-take-all tournament. On average, the normalized spread is equal to 0.48, so, in general, players expect the prize for winning will be approximately equal to half of the prize pool.
Descriptive Statistics for the Whole Sample.
Second, for a robustness check, we construct an analogue of the Herfindahl–Hirschman index (HHI) for every tournament. Basically, the HHI reflects market concentration. In the study, we create the HHI index as a sum of squared ratios of prizes for places to the prize pool, so it reflects the prize concentration within the tournament. The lower the index is, the more uniform the distribution of prizes within the tournament is. The HHI ranges from 0.13 to 1 in our sample, with 1 for the “winner-takes-all” tournament.
To capture individual performance, we collect players’ ratings from HLTV.org, known as “rating 2.0.” Introduced in June 2017, this metric integrates multiple aspects of players’ behavior during the game. It incorporates along with the widely used “kill-to-death ratio,” “survival rating,” and “damage rating,” several other popular in-game statistics. 1 Generally, the rating evaluates the effectiveness of a player's behavior within the specific match in a tournament, independent of previous matches’ results. A higher rating indicates better performance. The rating system is normalized such that an average player receives a rating of 1. A rating above 1 suggests that a player is performing above average, while a rating below 1 indicates performance below the average.
We have 3 variables, based on the HLTV “rating 2.0” in the sample. The first one is Rating—an individual player's rating, calculated by HLTV after every match in the tournament. Average player in the sample (Table 1) has a rating, which is more than 1, so he outperforms a typical player in CS:GO. Moreover, the best player outperforms the average one by over 3 times. 42.3% of players in the sample have ratings smaller than 1.
The other two rating metrics are applied as control variables. The average past rating is an average player's rating, calculated on the basis of the previous 6 months’ statistics. This variable is generally applied as a control variable to capture a player's talent (Parshakov et al., 2022). The average past rating is 1.08, which is similar to the mean individual rating. It means that, in general, the average player in the sample is more talented than the average CS:GO player. Notice, the average past rating is less dispersed than the individual rating.
The third rating metric is the Team rating. It is calculated as the average rating of the remaining players in the team. This variable reflects the environment within the team (Coates et al., 2020; Molodchik et al., 2021), in other words, the average talent of the team. We can see that teams in the sample can be called talented, as the mean team rating is higher than 1.
LAN and online tournaments significantly differ from each other, not only from a technical point of view, but from the reward structure as well. Table 2 represents the descriptive statistics on LAN and online tournaments. On average, the prize pool and spread in LAN tournaments are significantly larger than online tournaments. However, the normalized spread and HHI are higher in online tournaments than in offline settings. It seems that lower absolute values of prize and spread in online tournaments are compensated with higher relative values of prize differential and distribution. The difference of all tournaments’ characteristics is significant at a 1% level.
Descriptive Statistics for LAN and Online Tournaments.
Note: Standard deviation is presented in parentheses
The average “rating 2.0” of players participating in LAN and online tournaments are 1.10 and 1.08, respectively, with absolute difference of 0.02. This difference, though small, is statistically significant at the 1% level. Similar patterns are observed for the average past ratings of players and team ratings, with both metrics being statistically higher in LAN tournaments compared to online ones. Consequently, we can assume that LAN tournaments generally feature stronger players and teams than online tournaments, although the absolute difference in ratings is relatively small.
Empirical Method
In order to test the hypotheses (1) and (2) we estimate the relationship between individual performance and prize spread by the following panel regression model:
There are 3 groups of control variables that affect individual performance in the model. The first group is individual player characteristics—
In model (1) coefficient
Empirical Results
Model 1–2 in Table 3 shows the results for the normalized prize spread, which varies from stage to stage. In these models the normalized spread is statistically significant and the coefficient is positive. This suggests that the higher the prize spread, the better the player performance, which is in line with tournament theory and supports our first hypothesis. The dummy variable of online event in model 1 is statistically significant and negative, which supports our second hypothesis.
Regression Results.
Note: *p < 0.1; **p < 0.05; ***p < 0.01.
Model 2 in Table 3 includes an interaction term of normalized spread and a binary indicator of an online event, which is statistically significant and negative. The overall effect of spread and the interaction is statistically significant and positive, indicating that motivation caused by the difference in prize spreads is lower for online events than for live events. Therefore, we have found empirical evidence that online tournament conditions decrease the incentive effect of the convex prize structure on individual performance of eSports players supporting H3.
Model 3–4 in Table 3 reports the results for HHI as a metric of prize structure inequality, which is constant across the stages of the tournament. We report these results for two reasons: to show that our results are robust to the metric of prize inequality, and to infer if both prize difference over the stages (prize spread) and overall prize difference between all stages (HHI) are important. The results for HHI are similar to prize spread results: HHI is statistically significant and the coefficient is positive. An interaction term of HHI and the binary indicator of online event is statistically significant and negative.
The coefficients for control variables, as presented in Table 3, indicate the following: previous ratings have a positive effect on performance, while gender and age do not show statistical significance across models 1–4. Notably, across all models (1–4), it is observed that the prize pool has a consistent negative impact on performance, a finding that may initially seem counterintuitive. However, there is a plausible explanation for this phenomenon. Within the models, the prize pool is utilized as a control variable to account for the quality and prestige of the tournament. Typically, more prestigious tournaments have significantly higher prize pools, attracting more skilled and ambitious competitors. As a consequence, these tournaments feature stronger competition. This heightened competitive environment makes it challenging for individual players to earn high ratings. Therefore, an increase in the prize pool—indicative of increased competition within the tournament—leads to a decrease in individual player ratings at the conclusion of the matches, as reflected in the results.
Robustness Check
One might argue that there is a self-selection in particular tournament type. LAN tournaments are generally perceived as more prestigious and tend to attract more talented, ambitious, and higher-performing players and teams than online tournaments. According to Table 3, there is a statistical difference in player ratings between LAN and online tournaments, but the difference in absolute value is rather small. This may be attributed to strong competition in LAN tournaments, where it is notably more challenging to achieve an additional rating point. In contrast, in online tournaments o typically feature less rigorous competition, allowing weaker players more easily gain points of rating.
To address potential sample selection problem, we employ two different robustness checks by restricting the scope of our sample. Firstly, the sample is restricted only to so-called versatile players, who participated in at least one online and one offline tournament within the same year; secondly, the sample is restricted to encompass only the most prestigious tournaments.
Robustness Check for Versatile Players
We conduct estimations of models (1) and (2) on a subset of players who took part in both online and offline tournaments during the same year. Assuming that players’ skill levels remain relatively stable over one year, this approach allows us to compare performance of the same players under online and offline tournaments. This method is akin to a classical, quasi-experimental approach, called “difference-in-differences,” where the control group consists of players in offline tournaments, and the treatment group comprises the same players in online settings. Consequently, model estimates are less biased and reflect the causal relationship between prices structure inequality, measured as normalized spread and HHI, and rating.
The results of the robustness checks are presented in Table 4. Models 1 and 2 in Table 5 show the results for the normalized spread, while Models 3 and 4 demonstrate the results for HHI on performance. Across all models (1–4), the main findings are consistent. Specifically, both the normalized spread and the HHI significantly and positively affect player performance. And under online conditions, the effect remains positive but is of lesser magnitude, which is in line with our main estimations. Notably, within this restricted subsample, the prize pool continues to negatively impact performance, and the effects of other control variables are also maintained.
Regression Results for Robustness Check for Versatile Players.
Note: *p < 0.1; **p < 0.05; ***p < 0.01.
Regression Results for Robustness Check for Best Tournaments.
Note: *p < 0.1; **p < 0.05; ***p < 0.01.
Robustness Check for Best Tournaments
As a second robustness check, we apply models (1) and (2) exclusively to the “best” tournaments, defined as those recognized by professionals as the most prestigious. These are high-level tournaments featuring strong players and teams, ensuring a consistent quality of participants and competition. Historically, these tournaments were primarily held in LAN format before the COVID-19 pandemic, which led to their cancellation or transition to online formats. The shift in format due to the pandemic is exogenous, which suggest a more accurate comparison of player performance across formats Thus, the subsample includes the top tournaments 2 held between June 2017 and October 2021, excluding the post-COVID period. During this later phase, while several top tournaments continued online, others returned to LAN formats, resulting in a mix of formats that could confound analysis. Consequently, within the model we compare behavior of relatively strong players in both online and offline settings, using a method analogous to the “difference-in-difference” approach. Here, the official declaration of COVID-19 as a global health emergency by the World Health Organization on January 30, 2020, serves as the treatment variable.
The restricted sample consists of 80 tournaments, among which only 8% were held online, a proportion that does not match the online-to-offline ratio in the original dataset. The average players’ ratings in this subsample are 1.10 for LAN and 1.09 for online tournaments, respectively. This difference, along with the average team ratings, is not statistically significant, suggesting that the players in this subsample are of equal talent.
The model results are presented in Table 5, with Models 1 and 2 examining the normalized spread and Models 3 and for as metrics of price structure inequality, respectively. The effect of the normalized spread aligns with previous findings: the greater the normalized spread is, the better the performance exhibited by players. The interaction term between the normalized spread and the binary variable for online tournaments is not significant solely but is jointly significant at the 5% level. Thus, in online conditions, the normalized spread continues to positively influence performance, consistent with our previous finding.
For Models 3 and 4, the HHI does not significantly impact player ratings, nor does the interaction term of HHI significantly affect player ratings. One possible explanation is that the tournaments within this subsample are homogeneous and thus have a similar prize structure. This is partly confirmed by the data; except for one winner-take-all tournament, the HHI ranges from 0.14 to 0.6, with smaller variations in online conditions—from 0.22 to 0.34.
Conclusion and Discussion
This research delves into the potential implementation of tournament-like incentive design in the realm of remote work, using eSports as a case study. While eSports is computer-centric by nature, distinctions between offline and online competition formats, such as location and technical requirements, significantly influence player perceptions of reward systems. Leveraging a distinct dataset of individual eSports players who have competed in both LAN and online tournaments, we investigated how online conditions moderate the impact of prize spread on player performance.
As a preliminary step, this study supported the predictions of Lazear and Rosen (1981) tournament model in the context of eSports by providing empirical evidence on the positive and significant impact of prize spread on individual performance. The findings are consistent with the previous studies such as Mao (2023). The incentive effect of tournament-style rewards in eSports remains stable across various scenarios, including versatile players who participated in both online and offline competitions within a single year, as well as top tournaments that were previously held in LAN settings before the pandemic and have transitioned to online formats since January 2020.
The main contribution consists of comparing the magnitude of incentive effect in LAN and online conditions. Being more prestigious, having, on average, higher spreads and prize pool, LAN tournaments provide lower relative spreads and prize distribution. In other words, online conditions represent on average more convex prize structures in terms of tournament theory. Such a distinction causes differences in the magnitude of the reward effect. Empirical evidence shows that a tournament-style incentive design leads to lower individual productivity in online environments. This effect is robust across varying prize spreads and overall prize inequality between stages, and is consistent across players of different skill levels. It seems that we have identified one potential mechanism for decreased individual productivity in online conditions.
Such findings have practical implications for the development of human resource management practices. Taking into account that eSports players are mostly representatives of digital natives, it appears relevant to notice that they are sensitive to reward structures, such as prize differentials and prize inequality. This sensitivity depends on the work format. Remote working conditions challenge individual performance and require a different incentive design in comparison with offline conditions.
Footnotes
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This paper is an output of a research project implemented as part of the Basic Research Program at the HSE University.
