Can Uniform Color Color Aggression? Quasi-Experimental Evidence From Professional Ice Hockey

Black, White, and Aggression

Although the experimental social cognition literature shows some convincing color–valence associations, rarely have they been tested in applied settings with behavioral aggression. The most comprehensive investigation to date may be Frank and Gilovich (1988). In Study 1, they first showed that people unfamiliar with the National Hockey League (NHL) rated black jerseys as significantly more malevolent than nonblack uniforms. We performed a team-level secondary analysis of the mean ratings (their Table 1), which showed a significant difference (M _diff = 0.92, t ₁₉ = 4.84, p < .05, d = 2.22); teams that wore black jerseys were perceived as more malevolent than teams that did not (Ms = 4.82 vs. 3.89, respectively).

Frank and Gilovich (1988, Study 2) also examined the effect of black jerseys on penalty minutes per game in 21 NHL teams from the 1970–1971 to the 1985–1986 regular seasons. Our secondary analysis of the standardized (z-scored) team means (their Table 3) showed a relation between black jerseys and aggression (M_{z diff} = 0.94, t ₂₁ = 3.82, p < .05, d = 1.67); teams that wore black jerseys were assessed relatively more penalties than teams that did not (M_{z s} = 0.79 vs. −0.16, respectively). Of the two teams that switched to black jerseys during the 16-season sample, one was penalized more (p < .05, d = 1.26) but the other was not (p > .05, d = 0.37). One limitation of this study is that only 2 (9.5%) of 21 teams switched to black uniforms during the 16 seasons, which provides a novel—but underpowered—test of the black uniform effect.

Frank and Gilovich (1998, Studies 3 and 4) even conducted experiments to test the effect of black (vs. white) uniforms on aggression. In Study 3, 20 high school football referees watched videotaped, staged football plays with ambiguously aggressive acts that could be assessed a penalty. Each play was carefully choreographed so that only the offending team’s uniform color (black vs. white) was manipulated. Referees rated the act to be more aggressive—and were more likely to penalize the team—when the team wore black versus white (both ds = 1.20). In Study 4, the 6 three-person groups that were randomly assigned to wear black jerseys chose more aggressive games to play with a fictional competing group than did the six three-person groups that wore white (d = 1.94).

Although Frank and Gilovich’s (1988) two experiments (Studies 3 and 4) suggest either a black–aggression causal link—or a white–nonaggression one—their Study 2 could only show correlation and temporal precedence for a black uniform effect in the NHL. Recently, Caldwell and Burger (2011) sought to replicate this black uniform effect by comparing a subset of NHL games from the 2008–2009 and 2009–2010 seasons in which the eight teams that used black “third” jerseys played the same opponents in a pair of home games—once while wearing their normal dark-colored home jerseys and once while wearing their black third jerseys. Over 2 seasons, 102 pairs of games met these specific criteria for the 8 teams of interest. Although they tested multiple penalty measures related to aggression (e.g., penalty minutes at home and away games, severe penalties involving roughing and fighting), none produced a significant difference between black and dark-colored jerseys. For example, a paired-samples t-test comparing home penalty minutes assessed in games when black jerseys were worn versus when dark-colored jerseys was nonsignificant (Ms = 13.70 vs. 13.12 min, respectively, d = .09). ¹

Although Caldwell and Burger’s (2011) study featured focused comparisons, its analyses were flawed because they were done at the game level instead of the team level. That is, they ignored the inherent dependency of pairs of games nested within teams. The independence-of-errors assumption is violated whenever nested data structures are ignored. Violation of the independence-of-errors assumption can dramatically alter statistical conclusions, yielding both Type I and Type II errors. Statistically, Caldwell and Burger tested an erroneous model in which all 102 game pairs were assumed to come from 102 different teams, when in fact they came from only 8 teams. Accordingly, their conclusions must be called into question.

Aggression and Penalties in the NHL

If caught by referees, perpetrators of on-ice aggression can receive penalties that range from spending 2 min in a penalty box to a game misconduct (ejection and suspension from the next game). Five-minute major penalties are assessed for violent infractions such as fighting, charging, and spearing (stabbing an opponent with the stick blade). Especially violent major infractions may draw a 10-minute penalty in conjunction with a game disqualification or game misconduct. Two-minute minor penalties are assessed for aggressive—but not necessarily violent—infractions (e.g., elbowing, kicking, kneeing, tripping, high-sticking). Minor penalties are far more frequent than major ones.

Uniform Color in the NHL: A Unique Natural Experiment

Currently, all NHL teams have at least two jerseys—a white jersey and a colored jersey featuring the primary team color—and about 80% of teams have a “third” or “alternate” jersey that features a secondary team color or logo.

Over the course of the past 25 years, numerous NHL teams have changed their jersey designs, logos, and—crucially for our purposes—uniform colors. With a much longer time horizon to draw from than Frank and Gilovich (1988), we were able to examine the effect of wearing—and switching to—black jerseys on aggression (measured using penalty minutes; Study 1).

The NHL additionally mandates whether the home or away teams wear white or colored jerseys, which would normally confound uniform color with home versus away status (which can affect penalties; Glamser, 1990; Lehman & Reifman, 1987). The NHL, however, has changed its guidelines over time. Home teams wore white jerseys between the 1984–1985 and 2002–2003 seasons and colored jerseys between the 2003–2004 and 2009–2010 seasons (Study 2). Starting with the 1995–1996 season, the NHL allowed teams to create and wear third jerseys as an alternative to their normal colored jersey in roughly 10–15 games per season (12–18% of games).

The Present Research: A “Hat Trick” of Goals

The goals of the present research were threefold. First, we sought to provide a quasi-experimental test of the black uniform effect—comparing penalties assessed to teams wearing black versus colored jerseys—across time (25 years) for all 30 NHL teams (Study 1). Second, we aimed to analyze penalty minutes at home games to test the effect of an NHL-wide quasi-experiment—teams switched from wearing white to colored jerseys at home games starting in the 2003–2004 season (Study 2). Thus, Study 1 tested the effect of black (vs. colored) jerseys, whereas Study 2 tested the novel effect of white (vs. colored) jerseys. Third, we sought to use multilevel modeling (MLM) to account for the nonindependent structure of data where seasons were nested within teams (Studies 1 and 2).

Study 1: Black Versus Colored Uniforms

A limitation of Frank and Gilovich (1988, Study 2) was that only 2 of 21 teams switched to black jerseys in their 16-year sample. A stronger quasi-experimental test of this effect would involve multiple switching replications (Shadish, Cook, & Campbell, 2002) of the black-versus-colored uniform effect in several teams. To this end, we analyzed the largest data set to date and hypothesized that black uniforms would be related to more penalty minutes than colored (nonblack) uniforms for the average team.

Results and Discussion

On average, the 30 teams were significantly more aggressive (i.e., assessed more penalty minutes) during seasons in which they wore black jerseys (β₂₀ = 0.098, t ₂₉ = 3.21, p < .05, d = 1.19; Figure 1 , top), and penalty minutes decreased between 1984–1985 and 2009–2010 (β₁₀ = −0.026, t ₂₉ = −18.20, p < .05, partial correlation [pr] = −.96). The average team was assessed more PIM in seasons when they wore black jerseys (M = 1,528 min) versus when they wore other colored jerseys (M = 1,386 min), a difference of 142 min (10.2%)—1.73 min/game (104 s). This effect remained significant even after controlling for the lagged, autoregressive effects of PIM (i.e., the previous season’s PIM; see West & Hepworth, 1991). Moreover, for the 10 teams that switched to—or from—black jerseys, the effect was marginally significant despite less power, and the effect size was larger (β₂₀ = 0.076, t ₉ = 1.89, p < .10, d = 1.26). For illustrative purposes, Figure 1 (bottom) shows the black uniform effect on residual PIM (controlling for season) for the first team (alphabetically) in the data set, the Anaheim Ducks (see Figure 1 by Frank & Gilovich,1988).

OPEN IN VIEWER

Figure 1.

The top panel shows a predicted-scores spaghetti plot for the black jersey effect on penalty infraction minutes (PIM) for all 30 teams. The thin gray lines show the differences for the 10 teams that changed jersey colors to—or from—black at one or more times between the 1994–1985 and 2009–2010 seasons. The thick gray line shows the difference between the 3 teams that wore black jerseys for all seasons and the 17 teams that wore no black jerseys for all seasons. The thick black line shows the difference averaged across all 30 teams, which is the effect of interest. The bottom panel shows an example of the black jersey effect on penalty minutes for the first team in the data set (Anaheim Ducks). The y-axis shows standardized (z-scored) residual PIM, controlling for season. Gray bars show seasons when the team wore jersey colors other than black; black bars show seasons when the team wore black jerseys.

The data we obtained also had bench minor infraction (BMI) penalty minutes. BMI penalties typically result from personnel errors such as illegal substitutions or too many men on the ice; they are 2-minute penalties assessed against the whole team rather than an individual player, and accounted for only 1.14% of all PIM. Distinguishing between BMI and non-BMI penalties is important for two reasons. First, non-BMI penalties are typically aggressive acts (fighting, roughing, elbowing, kicking, kneeing, tripping), whereas BMI penalties are typically nonaggressive mental lapses (being on the ice when one should not). Second, a quasi-experiment can benefit from the addition of a nonequivalent dependent variable because it helps rule out threats to internal validity and helps bolster the construct validity of the effect (Shadish et al., 2002). We thus expected a black uniform effect for aggressive (non-BMI) penalties but not for nonaggressive (BMI) ones. BMI and non-BMI penalties were not significantly correlated. Analysis of non-BMI penalty data showed a similar effect to that of all PIM data (β₂₀ = 0.102, t ₂₉ = 3.23, p < .05, d = 1.20). This effect remained significant even after controlling for the lagged effect of non-BMI PIM. In contrast, analysis of BMI penalty data showed no such black versus nonblack jersey difference (β₂₀ = 0.0076, t ₂₉ = 0.18, p > .05, d = .07). Collectively, these ancillary analyses supported a key dissociation: the black-versus-colored uniform effect generalized to aggressive penalties but not nonaggressive ones.

Study 2: White Versus Colored Uniforms

In Study 2, we used MLM to test the average effect of multiple, team-level quasi-experiments to examine a link between wearing white-versus-colored jerseys and aggression at home games. As noted above, the NHL conducted a league-wide quasi-experiment starting with the 2003–2004 season: all 30 teams switched from wearing white jerseys to colored jerseys at home games. Thus, we tested the NHL’s jersey-color quasi-experiment and hypothesized that the switch from white to colored jerseys would be related to increased penalty minutes at home for the average team.

Results and Discussion

In a preliminary analysis, we replicated our Study 1 results, showing that teams were significantly more aggressive (i.e., assessed more penalty minutes) at home games during seasons in which they wore black jerseys (β₄₀ = 0.059, t ₂₉ = 2.06, p < .05, d = .77) for the average team, controlling for season and Away PIM.

Regarding the main analysis, teams were significantly more aggressive (i.e., assessed more penalty minutes) at home games after they switched to wearing colored jerseys at home during the 2003–2004 season (vs. wearing white jerseys at home before 2003–2004; β₃₀ = 0.091, t ₂₉ = 2.23, p < .05, d = 0.83; Figure 2 ; this effect was not significantly moderated by season). In addition, penalty minutes decreased between 2000–2001 and 2009–2010 for the average team (β₂₀ = −0.018, t ₂₉ = −2.43, p < .05, pr = −.41), and Away PIM were positively related to Home PIM (β₁₀ = 0.00074, t ₂₉ = 5.00, p < .05, pr = .68). The average team was assessed more Home PIM in seasons when they wore colored jerseys at home in 2003–2004 (M = 615 min) than when they wore white in 2002–2003 (M = 561 min), a difference of 54 min (9.6%)—1.32 min/game (79 s). This effect remained significant even after controlling for (a) the lagged effects of Home PIM and Away PIM (see West & Hepworth, 1991) and (b) seasons in which teams wore black (vs. colored) jerseys (Study 1).

OPEN IN VIEWER

Figure 2.

The top panel shows a predicted-scores spaghetti plot of interrupted time series analyses for all 30 teams (thin gray lines) and the average team (thick black line) in which penalty infraction minutes (PIM) at home games (Home PIM) were regressed onto PIM at away games (Away PIM), season, and the policy switch from wearing white to dark-colored jerseys at home games, which occurred between the 2002–2003 and 2003–2004 seasons; this is the effect of interest. The bottom panel shows a magnification of the top panel at the point of greatest interest: the jersey policy switch between the 2002–2003 and 2003–2004 seasons. Here, the predicted-scores spaghetti plot shows the difference in Home PIM between the 2002–2003 (white jerseys at home) and 2003–2004 (colored jerseys at home) seasons for all 30 teams (thin gray lines) and the average team (thick black line). These difference slopes are enlargements of the intercept shifts (regression discontinuities) shown in the top panel.

In the latter analysis, the effect of black jerseys was reduced to nonsignificance (β₄₀ = 0.052, t ₂₉ = 1.24, p > .05, d = 0.46); however, this was a weak test of the effect because all teams wore white at home games before the policy switch; there was no black-versus-color effect to test then. A more robust test—the simple black-versus-color effect after the switch—showed a marginal effect (β₄₀ = 0.087, t ₂₉ = 1.70, p < .10, d = 0.63), suggesting that the association between white jerseys and less aggression may be stronger than that between black jerseys and more aggression; however, one would expect a weaker effect across 6 seasons versus 25 (see Study 1).

Quasi-experiments can benefit from the addition of nonequivalent dependent variables because they address possible threats to validity such as history (Shadish et al., 2002). The data we obtained (2001–2010 seasons) had not only Home PIM and Away PIM, but also the number of goals each team scored per season at home and away games. Because the uniform change should affect aggression (penalties) but not performance (goals) in home games, we expected a null effect for the latter, despite the fact that these variables were positively correlated for the average team (r = .40, p < .05). To this end, we re-ran the above model substituting home and away goals for home and away PIM, respectively. In contrast to Home PIM, analysis of Home Goals showed no such difference relating to the 2003–2004 switch in home jersey color (β₂₀ = −0.0019, t ₂₉ = −0.07, p > .05, d = −0.02). This ancillary analysis supported a key dissociation: the white-versus-color uniform effect generalized to aggression (penalties) but not performance (goals).

General Discussion

Collectively, the present results show links between black (vs. colored) uniforms and more aggression, and between white (vs. colored) uniforms and less aggression. To be sure, one issue these studies cannot empirically resolve is whether (a) black jerseys are causing more aggression, (b) white jerseys are causing less aggression, (c) both are affecting aggression in opposite directions, or (d) the supposed causal link is spurious (a “third-variable” problem). Nevertheless, the findings of Study 2 suggest that the white-jersey–less-aggression association may be stronger than that of the black-jersey–more-aggression one. Although true experiments are difficult to conduct in settings that involve real physical violence such as professional hockey, we feel that these quasi-experimental findings—for which temporal precedence was established—come closer to establishing a causal link than do one-time correlational studies (see Kenny, 2004). Together with Frank and Gilovich’s (1988, Studies 3 and 4) experimental research on uniform color and aggression, our quasi-experimental findings point to a possible causal color–aggression link in professional hockey but not a definitive one.

The effect sizes were large by convention (Cohen, 1988), ranging from d = 1.19 (Study 1) to d = 0.83 (Study 2). On the whole, our effect sizes were comparable to those for the black (vs. white) uniform effect on referees in Frank and Gilovich’s (1998) research (ds = 1.67, 1.20) but not in Caldwell and Burger’s (2011) study (d = 0.09). One reason why the effect size of Caldwell and Burger’s study may be small—in addition to its analytic problems of ignoring nonindependent data—may be that its sample included too few games to iron out many game-level idiosyncrasies. Specifically, their study included data from only 204 games, which is less than 0.4% of the present research’s sample based on 52,098 games. Given the central limit theorem and the law of large numbers, effect sizes based on a larger sample are generally more stable and replicable than those based on smaller ones.