www.hockeythink.com

Single-Season Gains and Losses

Copyright Iain Fyffe, 2003

In their opus The Hockey Compendium, Klein and Reif discuss the topic of single-season increases and decreases in team winning percentage. Besides presenting lists of large single-season gains and losses, they offer one observation: large declines in winning percentage are generally tied to deterioration of defence rather than offence.

I thought that further analysis was called for. I compiled a list of teams that gained or lost .100 or more winning percentage compared to the previous season, from 1992/93 to 2000/01. In these years, there were 29 "gainers" (which is 58 half-gainers, I suppose), and 27 "losers".

1992/93 gainers: Que, Tor, Pgh, Bos, Cgy

1992/93 losers: NYR, Edm, SJ

1993/94 gainers: SJ, NYR, NJ, TB

1993/94 losers: Wpg, Que, LA, Chi, Pgh

1994/95 gainers: Que, Phi, Det

1994/95 losers: NYR, Sal, NYI, Mtl

1995/96 gainers: TB, Mtl

1995/96 losers: SJ, StL

1996/97 gainers: Dal, Ott, Buf, NJ

1996/97 losers: Det, Bos, Pgh

1997/98 gainers: Bos, LA, Wsh

1997/98 losers: TB, Fla, Ana, NYR

1998/99 gainers: Tor, Ott, Ana

1998/99 losers: Wsh, LA

1999/00 gainers: Wsh, StL, LA, Van

1999/00 losers: Bos, Dal

2000/01 gainers: Col

2000/01 losers: Fla, Ana

Looking over this list, the more perceptive among you may already see where this is headed. The average gain of the gainers is .158 winning percentage (WP), while the average loss of the losers is .146 WP. Nothing interesting there. The interesting thing happens when we look at the season following the big gain or loss.

Gainers, on average, become losers the subsequent season (not necessarily .100 losers, but losers nonetheless), losing an average of .053 WP. More strikingly, losers become gainers the subsequent season (again, not necessarily .100 gainers), gaining an average of .092 WP. This is not an illusion created by a few teams turning it around and masking the results of other teams. Particularly among the losers, the effect is persistent. Put another way, the gainers tend to lose 34% of their gains the nest season, while the losers regain a full 63% of their losses, on average. A team that lost 24 points (in an 82-game schedule) regains 15 of them the next season.

What all this means is that one of the more important factors in big gainers and (especially) big losers is sample size. That is, it seems 82 games are not enough to get a truly reliable indication of team quality. The season is short enough for flukes to occur. (And please don't think I'm advocating a longer season; the season is too long already, I'm just saying flukes will happen.) The poster child for this is the 1996/97 Bruins. They lost .183 WP from 1995/96, then gained back exactly that amount in 1997/98! That's a strong indication that 1996/97 was a fluke for Boston.

In summary, sample size and its implications is likely the most-neglected concept in sports analysis. Do not let yourself be fooled that something real has happened when it is only a trick of small sample size.

Reference

Klein, Jeff and Karl-Eric Reif. 2001. The Hockey Compendium. Toronto: McClelland & Stewart Ltd.