Clutching Claude

Over at Lowetide's, an argument recently developed over the nature of "clutch" performances: some players are reputed to perform at their best in key games. I was one of several commenters taking the position that this was probably illusionary. One commenter challenged this view by asking "How else do you explain Claude Lemieux's playoff stats?" My response was as follows:

Regular season PPG: 0.647
Playoffs PPG: 0.675

What's to explain?

That was a little glib, though, primarily because scoring tends to go down significantly during the playoffs (each team is playing against better teams on average than they do in the regular season, after all); a quick analysis of the 2008-2009 season showed an 11% decrease (limiting the analysis to teams that qualified for the playoffs), so for Lemieux to have increased his scoring by 4% in those circumstances is actually pretty impressive, on its face.

I'm of the view that, when it comes to sports performance, statistics can tell the whole story, but sometimes you need a lot of them, so I decided to look into Lemieux's career in more detail. The first things that struck me as surprising had nothing to with my hypothesis, but I thought I'd mention them anyway: first, Lemieux only spent four seasons with the Avalanche (as an Oilers' fan, I suppose I have observer bias from seeing him so often during those four years), and second, the Devils of the early nineties were a higher-scoring team than the Avalanche of the late nineties. Anyway, onward.

If we're going to test the hypothesis that Lemieux was a clutch playoff performer, we need to compare his playoff production to his regular season production, and then adjust it for the decreased scoring during the playoffs. That last bit is the tricky part: you can't just apply an across-the-board decrease, because the decrease hasn't been the same from era to era and, more importantly, isn't the same from team to team: teams eliminated in early rounds, unsurprisingly, see a much bigger dropoff in scoring than those eliminated in later rounds. So what I decided to do was compare Lemieux's shift in production each year to his team's shift in production during the same year. This will produce some small sample sizes, especially in years in which Lemieux's team was eliminated early, but we'll worry about that later.

Claude Lemieux, it's worth noting at this point, was part of some pretty good hockey teams. Of course, he was a pretty good hockey player, so all else being equal a team with Claude Lemieux was better than a team without him, but he won the Stanley Cup with three different teams, and played with goalies like Patrick Roy (in both Montreal and Colorado) and Martin Brodeur, and scorers like Peter Forsberg and Joe Sakic. Good teams. This is supported by the fact that he appeared in 18 seasons' playoffs on six different teams; only the 2000-2001 and 2002-2003 Phoenix Coyotes failed to qualify. Moreover, despite the fact that in any given playoffs half of the teams are eliminated in the first round, Lemieux's teams advanced to the second round twelve times, to the third round nine, and to the Stanley Cup Finals five times.

Anyway, here are the annual production shifts for Lemieux and the teams for which he played (team stats are goals per game, while player stats are points per game). I've bolded the years that Lemieux performed better than his team.

1986 (Montreal):
Team: 4.13 to 3.08 (25% decrease)
Player: 0.30 to 0.80 (167% increase)

1987 (Montreal):
Team: 3.46 to 3.94 (14% increase)
Player: 0.70 to 0.76 (9% increase)

1988 (Montreal):
Team: 3.73 to 3.36 (10% decrease)
Player: 0.78 to 0.45 (42% decrease)

1989 (Montreal):
Team: 3.94 to 3.19 (19% decrease)
Player: 0.74 to 0.39 (47% decrease)

1990 (Montreal):
Team: 2.93 to 2.64 (10% decrease)
Player: 0.46 to 0.36 (22% decrease)

1991 (New Jersey):
Team: 3.40 to 3.00 (12% decrease)
Player: 0.60 to 0.57 (5% decrease)

1992 (New Jersey):
Team: 3.61 to 3.57 (1% decrease)
Player: 0.92 to 1.00 (9% increase)

1993 (New Jersey):
Team: 3.67 to 2.40 (35% decrease)
Player: 1.05 to 0.40 (62% decrease)

1994 (New Jersey):
Team: 3.64 to 2.60 (29% decrease)
Player: 0.56 to 0.90 (61% increase)

1995 (New Jersey):
Team: 2.83 to 3.35 (18% increase)
Player: 0.42 to 0.80 (90% increase)

1996 (Colorado):
Team: 3.98 to 3.64 (9% decrease)
Player: 0.90 to 0.63 (30% decrease)

1997 (Colorado):
Team: 3.38 to 3.12 (8% decrease)
Player: 0.62 to 1.35 (118% increase)

1998 (Colorado):
Team: 2.82 to 2.29 (19% decrease)
Player: 0.55 to 0.86 (56% increase)

1999 (Colorado):
Team: 2.91 to 2.79 (4% decrease)
Player: 0.62 to 0.74 (19% increase)

2000 (New Jersey):
Team: 3.06 to 2.65 (13% decrease)
Player: 0.54 to 0.43 (20% decrease)

2002 (Phoenix):
Team: 2.78 to 1.40 (50% decrease)
Player: 0.50 to 0.00 (100% decrease)

2003 (Dallas):
Team: 2.99 to 2.83 (5% decrease)
Player: 0.19 to 0.14 (26% decrease)

2009 (San Jose):
Team: 3.13 to 1.67 (47% decrease)
Player: 0.06 to 0.00 (100% decrease)

So of those eighteen playoff performances, he exceeded his team's performance eight times, and fell short of it ten times. That's about what you'd expect from a random sampling - slightly worse, actually. Of course, some of the times he exceeded his team's performance were pretty epic: 1986, 1995, and 1997, for example. But viewed as a whole, that's not the record of a "clutch" player, that's a "streaky" player: sometimes he's really on, and sometimes he's really off. If you divided his regular season play into twenty game segments, I have no doubt that you'd find some segments that looked a lot like 1986, and others that looked a lot like 1989. When you take a large sample size like the career of Claude Lemieux, sometimes things will go really well for periods of time, and other times they'll go really badly.

But let's be fair: the above treats all playoff years the same, which is misleading. After all, 2009 looks a little less like a total collapse on Claude Lemieux's part when you realize that he played all of one post season game. So let's sort the above years, not chronologically, but by the number of post-season games in which Lemieux played. Then, if we see that the bolded seasons all appear at the top, we can decide that he really is clutch.

2000 (New Jersey):
Team: 3.06 to 2.65 (13% decrease)
Player: 0.54 to 0.43 (20% decrease)

1986 (Montreal):
Team: 4.13 to 3.08 (25% decrease)
Player: 0.30 to 0.80 (167% increase)

1994 (New Jersey):
Team: 3.64 to 2.60 (29% decrease)
Player: 0.56 to 0.90 (61% increase)

1995 (New Jersey):
Team: 2.83 to 3.35 (18% increase)
Player: 0.42 to 0.80 (90% increase)

1996 (Colorado):
Team: 3.98 to 3.64 (9% decrease)
Player: 0.90 to 0.63 (30% decrease)

1999 (Colorado):
Team: 2.91 to 2.79 (4% decrease)
Player: 0.62 to 0.74 (19% increase)

1989 (Montreal):
Team: 3.94 to 3.19 (19% decrease)
Player: 0.74 to 0.39 (47% decrease)

1987 (Montreal):
Team: 3.46 to 3.94 (14% increase)
Player: 0.70 to 0.76 (9% increase)

1997 (Colorado):
Team: 3.38 to 3.12 (8% decrease)
Player: 0.62 to 1.35 (118% increase)

1988 (Montreal):
Team: 3.73 to 3.36 (10% decrease)
Player: 0.78 to 0.45 (42% decrease)

1990 (Montreal):
Team: 2.93 to 2.64 (10% decrease)
Player: 0.46 to 0.36 (22% decrease)

1991 (New Jersey):
Team: 3.40 to 3.00 (12% decrease)
Player: 0.60 to 0.57 (5% decrease)

1992 (New Jersey):
Team: 3.61 to 3.57 (1% decrease)
Player: 0.92 to 1.00 (9% increase)

1998 (Colorado):
Team: 2.82 to 2.29 (19% decrease)
Player: 0.55 to 0.86 (56% increase)

2003 (Dallas):
Team: 2.99 to 2.83 (5% decrease)
Player: 0.19 to 0.14 (26% decrease)

1993 (New Jersey):
Team: 3.67 to 2.40 (35% decrease)
Player: 1.05 to 0.40 (62% decrease)

2002 (Phoenix):
Team: 2.78 to 1.40 (50% decrease)
Player: 0.50 to 0.00 (100% decrease)

2009 (San Jose):
Team: 3.13 to 1.67 (47% decrease)
Player: 0.06 to 0.00 (100% decrease)

Of his "clutch" performances three took place in low sample size situations, in which his team was eliminated in the first round. If we confine our analysis to years where Lemieux's team qualified for the Stanley Cup finals, we see two "clutch" performances and three sub-expectation performances.

Claude Lemieux built his reputation as a playoff performer on his performances in 1986, 1999, and to a lesser extent 1994 and 1997. Some of those performances were remarkable, so it's only natural that they'll remain in our minds while, for example, his four series 1989 vanishing act is forgotten. But viewed in the context of his entire career, these were less "clutch" performances and more instances of a player's hot streaks happening to coincide with big games.

But what about that 4% overall career increase? Numbers are deceiving: throughout his career, Claude Lemieux played approximately one playoff game for every twenty regular season games he played. During his last four seasons, when his skills had faded, he played 214 regular season games and 13 playoff games. His consistently low points per game total during this period therefore dragged his regular season points per game total down significantly, but had very little effect on his playoff total. If we ignore those seasons, his regular season points per game climbs to 0.697, while his playoff points per game goes to .710 - leaving the increase at under 2%.