Abstract
We examine whether the three-point rule—the increase in rewards for a win from two to three points that the Fédération Internationale de Football Association (FIFA) adopted in 1995—makes Bundesliga games become more exciting. Using regression discontinuity design as the empirical strategy, we do not find evidence that the three-point rule makes games more decisive, increases the number of goals, or decreases goal differences. We only find some evidence that the three-point rule increases the second-half goals of losing first-half teams. Overall, our results suggest that, in the case of Bundesliga games, the three-point rule does not work as FIFA intended.
Introduction
The Fédération Internationale de Football Association (FIFA) adopted the three-point rule, raised the rewards for a win from two to three points, in 1995 to make soccer games more exciting. The extra point, FIFA hoped, would induce teams to play more aggressively. Teams would abandon overly defensive plays, they would attempt to score more goals, games would become more decisive, and fans would crowd stadiums or be glued to their TV screens watching exciting soccer games. 1
The literature is, however, mixed on whether the three-point rule works. Brocas and Carrillo (2004), for example, theoretically show teams may play more defensively under the three-point rule; Haugen (2008), on the other hand, shows the opposite. Empirical findings also vary by sample of games, measure of outcomes, type of competitions, and empirical strategy. Dilger and Geyer (2009), for example, using a difference-in-differences approach, find the three-point rule decreases the probability of tied games in Bundesliga games. But, Guedes and Machado (2002), using regression control strategy, find the three-point rule does not affect teams’ offensive moves in the Portuguese first division league except those of underdog teams. Moreover, the direct effects are not as intended: Underdog teams become more defensive.
Two concerns complicate the estimation of the effects of the three-point rule. One, the theoretical effects of the three-point rule may be subtle—they may vary by the phase of games or the interim results, which means the results may also vary by measure of outcomes (Brocas & Carrillo, 2004; Guedes & Machado, 2002; Haugen, 2008). Two, more importantly, it is difficult to ensure that games under the three-point rule are compared with the right control group; if not, the estimates of the counterfactuals and those of the effects will be biased.
This article contributes to the literature by addressing the previously mentioned two issues. One, we analyze the Bundesliga whose game archives are rich in detail and publicly available, which we use to create various measures of outcomes such as goal differences in decisive first-half games or the number of goals by winning or losing first-half teams, in addition to the widely used outcomes such as the number of goals and whether games are decisive. Two, we use regression discontinuity (RD) design as the empirical strategy: We compare games around the time when FIFA adopted the three-point rule in 1995—games that, we argue, are similar except for the three-point rule. The RD design, hence, provides good counterfactuals to which we compare games under the three-point rule, which means our estimates eschew omitted variable bias problem unlike those from regression control strategy, difference-in-differences approach, or fixed effects models. Unlike these empirical strategies, the RD design excludes the effects of other changes in the rules of the game (e.g., back-pass rule or offside rule) or those of characteristics of the games (e.g., tactical evolutions, ball technology, or international player mobility) during the period of analysis so that we can attribute observed outcome differences at the time of the rule change in 1995 to the three-point rule.
We do not find evidence that the three-point rule makes Bundesliga games more decisive, increases the number of goals, or decreases goal differences. The only statistically significant result we find is that the three-point rule increases the second-half goals of losing first-half teams. Overall, our results suggest that, in the case of Bundesliga games, the three-point rule does not work as FIFA intended.
We proceed as follows. We review the literature and describe the empirical strategy and data in the next two sections. Then we discuss the results. The last section concludes.
Literature Review
Theoretically, the effects of the three-point rule are subtle. Guedes and Machado (2002), for example, show that the effects of the three-point rule may vary by the relative strength of teams in matches. If the two teams in a match are equally strong (or equally weak), the three-point rule induces more offensive plays. Underdog teams, however, may play more defensively under the three-point rule if the opponents are strong. Brocas and Carrillo (2004) argue that the effects of the three-point rule depend on the dynamics of games: They show that, unlike the two-point rule, the three-point rule induces teams to play more offensively in the second half compared to the first half; in particular, if games are tied, the three-point rule induces teams to play more defensively in the first half and more offensively in the second half. In contrast, Moschini (2010), using numerical analyses of a two-stage game, shows the three-point rule increases the number of goals per game and the probability of decisive games under quite general parametric assumptions. Haugen (2008) also shows that teams do not play more defensively under the three-point rule. If anything, they play more offensively, which means that, overall, the three-point rule leads to more offensive plays. 2
Some empirical articles show the three-point rule works. Moschini (2010), for example, finds that, using fixed-effects models of games in 35 countries over 30 years, the three-point rule increases the number of goals and decreases the probability of tied games. His analyses by country show that the three-point rule in many non-European countries makes games more exciting, but in European countries, the magnitude of the effects is smaller and in some cases the signs are the opposite. 3 Aylott and Aylott (2007) find that, in league games in seven countries, the number of goals increases in 2–3 years after the rule change except in Germany. 4 An “exciting index,” which they define as a function of the number of goals and whether games are decisive, also increases in all countries although it levels off after 4–5 years. Dilger and Geyer (2009) also find that, using a difference-in-differences approach, the Bundesliga’s fraction of tied games and goal differences in decisive games decreases after FIFA adopted the three-point rule.
However, some other articles show mixed results. Guedes and Machado (2002), for example, find that, in the Portuguese first division league, the three-point rule affects the offensive moves of underdog teams only: Underdog teams become more defensive. Palacious-Huerta (2004), who analyzes the structural breaks in the English league games during the period of 1982-1996, finds that the three-point rule and the back-pass rule affect the variability of goals, but not their averages.
We want to provide a more convincing empirical evidence of the effects of three-point rule: Following Dilger and Geyer (2009), we examine Bundesliga games to see whether their favorable results are robust to the use of an RD design as the empirical strategy. Perhaps, the RD design would provide a better control group for the games under the three-point rule than Dilger and Geyer’s control group, that is, the German Cup games, which may have different characteristics than those of Bundesliga games. We also see the fractions of tied games in Bundesliga and German Cup games diverge before the rule change, which may compromise the use of difference-in-difference or fixed-effects models as the empirical strategy. 5 Moreover, after FIFA adopted the three-point rule, the fraction of tied games in the Bundesliga does not decline immediately, while the fraction of tied games in German Cup games increases for some reasons over time, which means the favorable results that Dilger and Geyer identify may be driven by games played long after the rule change, not necessarily by the three-point rule only.
Empirical Strategy and Data
Empirical Strategy
We use RD design to identify the effects of the three-point rule on whether soccer games become more exciting. Winners in the 1994-1995 or earlier seasons get two points; those in the 1995-1996 or later seasons three points. There is, therefore, a deterministic and discontinuous treatment of wins—the points awarded to winners—between the 1994-1995 and 1995-1996 seasons, which fits an RD design. 6
Identification relies on the plausible assumption that characteristics of games, players, and teams—the skill sets of players, coaches’ philosophy of plays, teams’ popularity and finances, stadiums’ size and shape, and so on—around the discontinuity in the 1994-1995 and 1995-1996 seasons are similar on average. There are no major (exogenous) shifts in coaches’ philosophies or tactics during the 1994-1996 seasons except changes in coaches’ approaches induced by the three-point rule. Most players play in both seasons; most coaches coach the same teams in both seasons; most teams play in both seasons. The rules of play are also identical except for the three-point rule. Therefore, if we see discontinuities in some measures of outcomes between the 1994-1995 and 1995-1996 seasons, we can attribute the discontinuities to the three-point rule. 7
Formally, we estimate the effects of the three-point rule using the following regression
where yi is a measure of outcome of game i such as whether game i is decisive or the number of goals scored in the game; D is an indicator equals 1 for games in the 1995-1996 or later seasons, it equals 0 otherwise; f(season) is a polynomial function of season, the assignment variable; X is a vector of control variables such as the round the game is played and the identities of home and away teams; and ∊ is the error term. In the basic specifications, we use the quartic polynomial function of season, though in some specifications we also use its cubic or quintic function. As part of robustness checks, we use round instead of season as the assignment variable. Because the data fit an RD design, we do not have to include X in Equation 1—its inclusion would not change the estimate of β; but we do include the vector of control variable X in some specifications to increase the precision of the estimate.
The coefficient of interest is that of D. If the three-point rule makes games more exciting, we expect β to be positive in regressions whose dependent variable is the number of goals or whether games are decisive. We expect it to be negative in regressions of goal differences (because lopsided games are boring). Along the lines of Brocas and Carrillo’s (2004) results, we expect β to be positive (negative) in regressions of second-half goals of losing (winning) first-half teams.
Data
We get the data from Deutcher Fussball-Bund, which archives Bundesliga games since the 1960s. 8 We use games played during the period of 15 years before and after the 1995-1996 season, that is, the 1980-2010 seasons. The sample includes 9,560 games played in 31 seasons with 34 rounds each. 9
We create five sets of outcome measures, whether games are decisive, the number of goals, goal differences, second- and first-half comparisons, and second-half goals. We look at whether games are decisive, the number of goals, and goal differences because the objective of the three-point rule is to promote attacking plays, which lead to more decisive games, larger number of goals, and smaller goal differences. Further, to test some aspects of the theoretical predictions of the three-point rule by Brocas and Carrillo (2004), to see whether game dynamics change from the first to the second half, we compare first- and second-half outcomes and examine second-half goals.
We define decisive games (at halftime and full time, for all games and for a subsample of tied first-half games) equal to 1 if a game is decisive and 0 otherwise; number of goals equal to the sum of goals scored in a game, both by home and away teams (we consider the number of goals in the first half, in the second half, and at full time); goal differences equal to the absolute value of the difference between goals scored by the home and away teams (we consider goal differences in the first half, in the second half, at full time; we also look at goal differences in a subsample of decisive first-half games). We define second- and first-half comparisons equal to the difference between the number of goals scored in the second half less than that in the first half. Finally, we define second-half goals by a subsample of teams or games: second-half goals by winning first-half teams, by losing first-half teams, and in tied first-half games.
The summary statistics in Table 1 presents mixed evidence on whether the three-point rule makes soccer more exciting. The fraction of decisive games in the 1995-2010 seasons at full time, for example, is higher than that in the 1980-1994 seasons, though the difference is insignificant statistically (Panel A). The number of goals declines on average after the adoption of the three-point rule (Panel B), so do goal differences (Panel C). Winning first-half teams score more second-half goals, losing first-half teams score fewer goals in the second half, and the number of goals in tied first-half games declines (Panel D).
Summary Statistics.
Note. The number in each cell is the mean. The figures in parentheses are standard deviations.
Results
The Number of Goals and Whether Games Are Decisive
Graphs in Figure 1 illustrate the effects of the three-point rule on the number of goals and whether games are decisive. They plot the average number of goals or the fraction of decisive games by season. The vertical dash line indicates the discontinuity, the season after which the Bundesliga uses the three-point rule. The graphs also fit a quartic polynomial function of season, the assignment variable, that may jump between the 1994-1995 and 1995-1996 seasons.
Decisive games and number of goals at halftime and full time.
Teams score more goals and games are more decisive in the 1980s: The trend lines decline slightly. The trend lines seem to rise between the 1994-1995 and 1995-1996 seasons, which suggests the three-point rule increases the number of goals and makes games more decisive. The magnitude of the jumps is small, however, and they are also insignificant statistically.
Table 2 confirms the trend lines in Figure 1. Teams score fewer goals under the three-point rule as column 1 of Panel B shows. There is no evidence that the three-point rule increases the number of goals, however, once we control for the quartic polynomial of season (column 2): The estimates are positive, but they are insignificant statistically with standard errors bigger than the estimates. We get similar results after we add round dummies and home and away teams’ dummies (columns 3–4). As we expect, because the data fit an RD design, the estimates are stable across the different specifications in columns (2–4).
The Number Goals and Whether the Games Are Decisive.
Note. The number of oberservations is about 9,560. The number in each cell is the estimate of three-point rule from a regression of a measure of outcome, which is listed on the left column, on three-point rule and a set of control variables listed at the bottom rows. Three-point rule of a game equals 1 if it is in the 1995 or later seasons; it equals 0 otherwise. The figures in parentheses are robust standard errors clustered by season.
* and ** denote statistical significance at a level of 5% and 1%, respectively.
Not only that the estimates are insignificant statistically, the magnitude of the effects is also small. The number of goals in the first half, for example, increases by 0.02, which is about 1.6% increase. The increase in the number of goals in the second half is larger, 0.12 goal (7%), which means the three-point rule increases the number of goals in the second half by one goal in every eight games on average. Games do not become more decisive either. The fraction of decisive games at full time, for example, increases by only 1 percentage point (1%). The increase in the fraction of decisive games at halftime is larger, however, about 3 percentage points (5%). Again, none of the estimates is significant statistically.
Goal Differences
Graphs in Figure 2 show the three-point rule does not seem to lower goal differences: The trend line does not fall between the 1994-1995 and 1995-1996 seasons. In fact, it rises for goal difference in the second half. All estimates seem to be insignificant statistically because of the large variations in the average differences.
Goal differences at halftime and full time.
Panel A of Table 3 shows that goal differences are smaller under the three-point rule (column 1), but none of the estimates is significant statistically once we include the quartic polynomial of seasons in the regression (column 2). The magnitude of the effects also remains similar when we include round dummies and home and away teams’ dummies (columns 3–4). Again, the estimates are insignificant statistically, which means that there is no evidence that the three-point rule decreases goal differences in the first half, in the second half, at full time, or in decisive first-half games.
Goal Differences.
Note. The number of oberservations is about 9,560. The number in each cell is the estimate of three-point rule from a regression of a measure of outcome, which is listed on the left column, on three-point rule and a set of control variables listed at the bottom rows. Three-point rule of a game equals 1 if it is in the 1995 or later seasons; it equals 0 otherwise. The figures in parentheses are robust standard errors clustered by season.
* and ** denote statistical significance at a level of 5% and 1%, respectively.
The small estimates means the effects are probably indifferent from zero. The effects on goal differences in decisive first-half games in particular are very small, 0.1–0.2 percentage point. Only the estimates of the effects on goal differences in the second half are large, 0.13 goals (13%), though they are insignificant statistically.
There is no evidence that the three-point rule reduces the difference between the number of goals in the second and first half either (Panel B). The estimates are positive and large, 0.1 goal (24%), but insignificant statistically.
Final Outcomes Given First-Half Outcomes
It is possible that the three-point rule induces teams to change their strategies given early outcomes of games. For example, teams may play more offensively in the second half if games are tied at the half (Brocas & Carrillo, 2004). A losing team in the later stages of a game may attack aggressively because it does not have much to lose; on the contrary, a winning team may play defensively to prevent equalizers to protect their lead.
Graphs in Figure 3 illustrate the effects of the three-point rule on second-half goals by winning and losing teams as well as outcomes of tied first-half games. The trend lines seem to rise between the 1994 and 1995 seasons, which indicate the three-point rule induces teams to change their strategies later in games.
Second-half goals and decisive games given first-half results.
Table 4 presents the estimates of the jumps. All RD estimates in columns (2–4) are positive, but only the effects on second-half goals by losing first-half teams are significant statistically. When we include season quartic polynomial in column (2) or the polynomial of round dummies in column (3), the estimate is significant at 5% level. When we include home and away teams’ dummies further in column (4), it becomes significant at 1% level. The estimates are not large, but not trivial either, 0.12 goal (15%), which means losing first-half teams score one more goal in the second half in 8-9 games on average.
Final Outcomes Given First-Half Outcomes.
Note. The number of oberservations varies from 3,800 to 5,800. The number in each cell is the estimate of three-point rule from a regression of a measure of outcome, which is listed on the left column, on three-point rule and a set of control variables listed at the bottom rows. Three-point rule of a game equals 1 if it is in the 1995 or later seasons; it equals 0 otherwise. The figures in parentheses are robust standard errors clustered by season.
* and ** denote statistical significance at a level of 5% and 1%, respectively.
While there is no evidence of winning first-half teams scoring more goals in the second half, there is also no evidence that they play more defensively either: The estimates are positive with standard errors about twice as large. Overall, there is no evidence that the three-point rule makes tied first-half games more decisive or that it increases second-half goals in tied first-half games.
Robustness Checks
We do some robustness checks. One, we examine a subsample of games in the second half of seasons only to focus on games that matter the most for teams that fight for the championship or those battling against relegation. Two, we use alternative polynomial of season. Three, we use round instead of season as the assignment variable to increase the similarity of games near the discontinuity in 1995.
Table 5 presents the estimates on key outcome measures using games in the second half of seasons only. Overall, the results are robust. There is no evidence that the three-point rule makes games more decisive or decreases goal differences. There is also no evidence that it induces winning first-half teams to play more defensively or increases the second-half goals in tied first-half games. Nonetheless, we do find some evidence that losing first-half teams score more goals in the second half; the estimates are similar to those in Table 4 and significant statistically at 5% level when we include round as well as home and away teams’ dummies in the regression (column 4 of row 5).
Using a Sample of Second Half of Seasons Only.
Note. The number of oberservations varies from 2,000 to 4,800. The number in each cell is the estimate of three-point rule from a regression of a measure of outcome, which is listed on the left column, on three-point rule and a set of control variables listed at the bottom rows. Three-point rule of a game equals 1 if it is in the 1995 or later seasons; it equals 0 otherwise. The figures in parentheses are robust standard errors clustered by season.
* and ** denote statistical significance at a level of 5% and 1%, respectively.
Table 6 presents the estimates of the effects of the three-point rule using alternative function of season (cubic and quintic polynomial of season) in columns (1–2) and alternative assignment variable, round, instead of season in columns (3–5). Overall, the basic results are robust. There is no evidence that the three-point rule makes games more decisive, increases the number goals, or decreases goal differences. However, we find some evidence that losing first-half teams score more second-half goals; the estimates are stable across the various specifications and are significant statistically at 5% level except when we use quintic polynomial of round (column 5 of row 5).
Using Alternative Polynomial of Season and Using Round as the Assignment Variable.
Note. The number of observations varies from 3,800 to 9,650. The number in each cell is the estimate of three-point rule from a regression of a measure of outcome, which is listed on the left column, on three-point rule and a set of control variables listed at the bottom rows. Three-point rule of a game equals 1 if it is in the 1995 or later seasons; it equals 0 otherwise. The figures in parentheses are robust standard errors clustered by season.
* and ** denote statistical significance at a level of 5% and 1%, respectively.
We check whether our results are robust to the use of a linear function of the assignment variable instead of higher polynomial functions of it. Overall, column (6) shows the results are robust. Almost all estimates are insignificant statistically; only the effect on whether games in tied first-half games are decisive is significant (at 5% level), though it is probably because a linear function of seasons does not fit this measure of outcome well, as the bottom-left graph in Figure 3 shows. (The outcome had declined in the 1980s and increased in the early 1990s before the Bundesliga used the three-point rule in 1995, which a linear function cannot capture well.)
We also examine whether our results apply in other countries: We analyze the effects of the three-point rule in major leagues in England, Spain, Portugal, Italy, Argentina, and Mexico. 10 Figures 4–5 and Table 7, which are analyses of games in 20 seasons around the discontinuity in each country, show our basic results are quite robust: We do not find evidence that the three-point rule increases the number of goals per game except in Argentina; we do not find it increases the fractions of decisive games either except in Italy. The statistically significant finding in Italy seems real; as the bottom-left graph in Figure 4 shows, there is a clear jump in the fractions of decisive games at the discontinuity. The three-point rule does not seem to affect the number of goals in Italy, however, as the bottom-left graph of Figure 5 shows. The statistically significant effect on the number of goals in Argentina is perhaps because our use of the cubic- and quantic polynomial function of seasons does not fit the outcome well; the average goals per game had increased since 1991, long before Argentina used the three-point rule in the 1995-1996 season. Therefore, our basic results (that there is no evidence the three-point rule in Germany makes games more exciting) seem to apply quite generally in other countries. 11
Decisive games in six other leagues.
The number of goals in six other leagues.
Whether Games Are Decisive and the Number of Goals in Six Other Leagues.
Note. The data include all games in the 20 seasons around each discontinuity. The number in each cell is the estimate of the three-point rule (in a league in the country listed on the left column from a regression of a measure of outcome listed on the top row, whether games are decisive or the number of goals per game) on the three-point rule and a set of control variables listed at the bottom rows. Three-point rule of a game equals 1 if it is in the 1981 (England), 1994 (Italy), or 1995 (other countries) season or later; it equals 0 otherwise. The figures in parentheses are robust standard errors.
*denotes statistical significance at a level of 5%.
Concluding Remarks
There is no evidence that the three-point rule makes Bundesliga games become more decisive, increases the number of goals, or decreases goal differences. Most of the estimates are positive, which means the three-point rule increases the fraction of decisive games (as we expect if the three-point rule works), the number of goals (as we expect), and goal differences (the opposite of what we expect); but their magnitude is small, and none of the estimates is significant statistically.
However, there is some evidence that the three-point rule increases the second-half goals of the losing first-half teams. The estimates are significant statistically and robust across various specifications. They are also large, about 9–12 percentage points, which means the three-point rule increases the second-half goals of the losing first-half teams by 12–15%.
These results show no convincing evidence that the three-point rule makes soccer games more exciting. If anything, the three-point rule seems to induce losing teams to play more aggressively later in the games, which probably also means that winning teams play more defensively to preserve the lead. Not only that the three-point rule fails to increase the number of goals, but it also induces winning teams to play defensively to deny goals.
It is likely, therefore, at the margin, the three-point rule makes stronger teams play conservatively earlier in games and play defensively once they are leading. 12 Weaker teams may play more aggressively if they are losing, though they may find it difficult to score goals because the leading, and usually stronger, teams play more defensively.
Our results differ from some findings in the literature perhaps because we use different methods. Dilger and Geyer (2009) use a difference-in-differences approach as their empirical strategy and the German Cup games as the control group; Moschini (2010) fixed-effects models and games in other countries. But, perhaps, these games are not the appropriate control group of the Bundesliga games; besides, the implicit assumption of difference-in-differences and fixed-effect models that games in the treated and control groups have the same trends, which is untestable, may be unsatisfied. Aylott and Aylott (2007) use time series analyses of the average of an exciting index in seven countries, but their results may be driven by the increasing trends of the index in some countries even before the three-point rule was introduced. Haugen (2008) uses the time series of competitive imbalance among teams, but he does not test the effects of the three-point rule formally. In contrast, our empirical strategy, the RD design, relies mostly on the assumption that the games near the discontinuity are similar, which we think is a plausible assumption. 13 We do not see jumps of the trend lines of outcomes at the discontinuity in 1995. It does not mean that the three-point rule is irrelevant, but it suggests that the three-point rule’s effects may be too small to be identified by the RD design.
Our results are, to some extent, in line with the theoretical predictions of Brocas and Carrillo (2004). We do not find statistically significant evidence that winning teams play more defensively later in games, but we do find that losing teams play more aggressively later in the games.
There are concerns that (1) the three-point rule does not affect teams’ behavior immediately; (2) games further away from the discontinuity may not be comparable and, therefore, should be excluded from the analysis; and (3) the estimates are mostly insignificant because of the lack of power to reject the null hypotheses. If the first is true, it means the RD design would fail to capture the effects of the three-point rule. But, we see that the trend lines are flat throughout the period of analysis, which indicates that this worry is unwarranted. If the trend lines decline, mostly they do in the 1980s; the trend lines in the few years around the discontinuity are mostly flat. To address the second concern, we redo the analyses using fewer number of seasons around the discontinuity; we still find no evidence that the three-point rule makes games more exciting, though the fewer the number of seasons we use, the lower the power will be. The third concern is legitimate, and there is no way for us to address it. 14 The RD design, given our data, may have low power because the number of games per season is limited, only season or round can be used as the assignment variable, polynomial functions do not approximate the trends in outcomes well, or other factors (that we are not aware of) negate the effects of the three-point rule. We hope, in future research, to use more powerful methods and analyze more specific measures of outcomes to conclude whether the three-point rule works.
Overall, our results do not provide evidence that the three-point rule makes Bundesliga games more exciting, which seem to apply quite generally in other countries: We do not find sufficient evidence that it increases the number of goals or that it makes games more decisive. These probably mean leagues and FIFA should consider introducing more winning incentives in soccer games, such as penalty shoot-outs in the event of draws or the controversial golden goal rule, to make soccer games even more exciting. 15
Footnotes
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.