I read the paper† and I have to admit the statistics ("multilevel generalized estimating equations with a negative binomial distribution and a robust estimator covariance") is over my head, but to me it kind of sounds like they couldn't find any results with a simple model, so opted for something more complex. Also some of these results are kind of fishy.
Across all countries, FWHR only had a marginally significant association with fouls committed (B=.291, 95 % CI: [−.040, .622], p=.085). When these effects were examined separately for defenders, midfielders, and forwards, the relationship between FWHR and fouls were in a negative, nonsignificant direction in defenders (B=−.361, 95 % CI: [−.841, .119], p=.140), a positive direction for midfielders (B=.449, 95 % CI: [.015, .884], p=.043), and a more robust positive direction for forwards (B=.935, 95 % CI: [.401, 1.469], p=.001).
So if you are a defender FWHR and fouls are negatively correlated, but if you are a forward FWHR and fouls are positively correlated? That just seems confusing, and doesn't at all support the conclusion that "high FWHR is associated with more aggressive behavior."
Across all countries, FWHR only had a marginally significant association with fouls committed (B=.291, 95 % CI: [−.040, .622], p=.085). When these effects were examined separately for defenders, midfielders, and forwards, the relationship between FWHR and fouls were in a negative, nonsignificant direction in defenders (B=−.361, 95 % CI: [−.841, .119], p=.140), a positive direction for midfielders (B=.449, 95 % CI: [.015, .884], p=.043), and a more robust positive direction for forwards (B=.935, 95 % CI: [.401, 1.469], p=.001).
So if you are a defender FWHR and fouls are negatively correlated, but if you are a forward FWHR and fouls are positively correlated? That just seems confusing, and doesn't at all support the conclusion that "high FWHR is associated with more aggressive behavior."
† http://www.researchgate.net/publication/264545489_An_Examina...