Rise of the underdog or fall of the top dog? A study of Canadian Ice Hockey stats from Gibbs et al, 2012

Gibbs’s hypothesis, from 2012, of an underdog effect is often referenced in many papers relating to investigations into Relative Age Effects (RAEs). The hypothesis is used to explain how levels of RAE fall over time from a peak during adolescence to either a lower RAE, parity, or a RAE reversal at some point in a senior career. Gibbs et al, 2012 offered a peer effect as a reason for the reduced level of Q1 (born Jan-Mar) players found in Olympic and All Star Canadian hockey teams, whereby later born players benefit from higher challenge during development through playing/training with very high numbers of early born.

This study looks at the data for the four Olympic squads studied by Gibbs and the source population of players from whom they were selected. Namely the Olympic years 1998, 2002, 2006 & 2010 (n=64). Source population data was consequently collected for seasons 1997-1998, 2001-2002, 2005-2006 & 2009-2010 (n=1179).

What did Gibbs find?

Q1 %s were 40% in junior hockey. Q1 %s were only 28% in the NHL. Q1 %s were lower again at 17% for Canadian Olympic and All Star teams. A peer effects hypothesis for this decline was offered.

What are some of the problems with these findings and the underdog hypothesis?

From the paper we only know about Q1. We have no idea how the other birth quarters (BQ) fare. It is assumed that ALL later born BQs do better and presumably it is a linear effect in that the later you are born (i.e. Q4) you fare the best as they have had the greatest level of challenge.

What do we actually find?

NHL%s Q1 29, Q2 30, Q3 22, Q4 19.

Olympic %s Q1 19, Q2 36, Q3 28, Q4 17

Difference -10, +6, +6, -2

Q1 produces far fewer (-34%) Olympic players than the source population would suggest. But surprisingly Q4 produces less than expected too, in fact more than 10% less than its original population. Both Q2 and Q3 perform better with +20% and +27% respectively. Perhaps this is more of a Goldilocks Effect?

What else is happening in this data?

RAEs don’t disappear overnight from junior to senior transition. Instead they continue to decline. We can see this in these Canadian ice hockey data by splitting the source NHL population into tertiles by age.

Rookie (age 18-22) Q1 28, Q2 34, Q3 21, Q4 17.

Prime (age 23-25) Q1 31, Q2 29, Q3 21, Q4 19.

Veteran (age 26-38) Q1 29, Q2 27, Q3 24, Q4 21.

Difference (Rookie->Veteran) Q1 +1, Q2 -7, Q3 +3, Q4 +4.

* only for skaters (goalies excluded).

Over a senior career RAE often changes/declines. In the NHL the trend is for Q2 to decline, Q3 & Q4 to rise and Q1 to remain fairly static. Presumably (de)selections are made using a combination of factors, such as (perceived) performance, fitness/injury and toughness/character/leadership for example?

In terms of performance, a quartile analysis of games played and points scored, for skaters only, can shed some light on BQ differences. 

Games Played

Highest Q1 26, Q2 30, Q3 26, Q4 19

2nd Q1 30, Q2 29, Q3 22, Q4 19

3rd  Q1 29, Q2 29, Q3 22, Q4 21

Lowest  Q1 31, Q2 31, Q3 18, Q4 19

Diff (Highest-Lowest) Q1 +5, Q2 +1, Q3 -8, Q4 0

Points Scored

Highest Q1 25, Q2 30, Q3 26, Q4 19

2nd Q1 29, Q2 29, Q3 21, Q4 21

3rd  Q1 32, Q2 28, Q3 20, Q4 19

Lowest  Q1 30, Q2 31, Q3 21, Q4 18

Diff (Highest-Lowest) Q1 +5, Q2 +1, Q3 -5, Q4 -1

There appears to be similar ‘performance’ by BQ using both Games Played and Points Scored for skaters. Q1s are more likely to appear in lower quartiles. Q2 and Q4 are fairly equally spread, whereas Q3 appear more likely to be in the highest quartile. So these data could explain why Q3 are retained during a career but not explain why Q4 are retained. Equally these data could explain a decrease in Q1 career length, but they don’t decline significantly. Equally again it doesn’t explain the -7% decline in Q2. (Please remember the base RAE of the source population is Q1 29, Q2 30, Q3 22, Q4 19).

The under performance of Q1 (and perhaps Q4) could explain their relative under selection at Olympic level and the performance of Q3 for the over selection of Q3 BUT perhaps doesn’t explain the over selection of Q2.

Influence of Age on Olympic Selection

The difference between RAE profiles of Olympic teams and the NHL source population may in reality be more extreme. Olympic players debut on average at age 28. Including multi-Olympians this rises to 29. Only 4 Rookies were Olympians, while 19 were in the Prime phase. The vast majority (41 players) were Veterans who have a more equitable RAE profile.

Veteran (age 26-38) Q1 29, Q2 27, Q3 24, Q4 21.

Difference (NHL->Olympian) -10, +9, +4, -4

Conclusions

The mechanisms of RAE decline are unproven. Natural regression from over selection is probably highly likely. Other factors such as peer effects are probably also at play. The power of each (and other effects) are currently only speculative. The Underdog Effect and other related ‘mechanisms’ such as the Reversal of RAE Advantage are, IMO, not yet accepted in the literature. For every paper that finds evidence for them there are equal numbers of papers that are against. 

This data suggests that while Q1 are less likely to reach elite levels of performance relative to the other BQs, Q4 too, to a lesser extent, are not likely to reach those levels either. Both Q2 & Q3 are shown to be more likely to be selected at Olympic level but only Q3 show a reason for this in terms of both Games Played/Points Scored and a higher retention through to a Veteran phase of the NHL.