Are There Real Stat Padders?

Last time, we found that teams don't carry a killer instinct: once you account for how good a team is, there's no lasting tendency to run up the score. But a team is a lot of people. What about the individuals? Is there a kind of player who comes alive after a game is decided, a stat padder, whatever his team is doing?

It's a harder question than it sounds, because a player on a dominant team will have more of his points come on padded goals simply because his team pads a lot. So we didn't compare players to the league. We compared each one to his own teammates.

How we tested it

For every player on every team in every season, we looked at the goals he was involved in, as the scorer or an assister. Each is one point. Then we asked what share of those points came on padded goals, the ones the winning team scored after the game was already decided.

The comparison is the padded share of his teammates' goals, only in the games he actually played. His own goals are left out of that baseline. The gap between what he had and what his teammates' goals would predict is his excess.

A single season can't settle this, since a few points more or fewer in the padded column is mostly luck. So we asked four things: do teammates differ by more than chance would produce, does a player's excess carry over from year to year, does it show up across a whole career, and does it follow a player to a new team?

Do teammates really differ?

A little, yes. If padded goals were handed out at random among a team's goals, the spread between teammates would have a variance of 1.36. The real spread is 1.53, about nine standard deviations higher. And 10.4% of player seasons landed beyond two standard deviations from their teammates, against 8.3% under pure chance.

So there's something there. The question is what, and how big.

Does it follow a player?

Only faintly from one year to the next. A player's excess carries over to the following season at 0.08, on a scale where 1 is perfect carryover and 0 is none. Over a whole career it's clearer: when we split each player's seasons into two interleaved halves, the two halves agree at 0.29, which works out to about 0.45 for a full career.

That's a real, repeatable component. It's also small. And it doesn't survive a change of team in any way we can detect: across the 749 cases where a player moved teams between seasons, the carryover is 0.03, with a range of −0.05 to +0.10. That's too few cases to tell the difference between "none" and "a little", so we can't say it follows a player in the way a skill would.

A lot of it is empty-net goals

Empty-net goals count as padded under the rule, and they come from a particular situation: a team protecting a lead in the last minutes, with a certain kind of player on the ice. The play-by-play marks these goals, so we ran everything again with every empty-net goal removed.

The carryover from year to year drops from 0.08 to 0.03. The career-half agreement drops from 0.29 to 0.17. The number of players far from their teammates falls from 104 to 42, against about 11 expected. Something is left, which shows it isn't only empty nets. But a large part of the apparent "padder" signal is lead-protection scoring.

Who stands out

Here are the six players with the most padding relative to their teammates, among players with at least 300 career points.

And here are the six with the least padded points relative to their teammates, among players with at least 300 career points.

The differences are in percentage points of padded share, calculated from unrounded figures, so they can differ by 0.1 from subtracting the two columns beside them. At the extremes they run from about 5 to 10 points either way. Two things stand out.

The high end looks like role players. Two of them, Guy Carbonneau and John Madden, are best known as defensive forwards and each won the Selke Trophy, though that comes from outside our data. That fits the empty-net result: these are the players you'd expect to be on the ice protecting a lead.

The low end is stars. Sakic, Bourque, Boyle, Hedman and Murphy produced a lot of their points while games were still live. Defensemen, as a group, are about 3 points less padded than their teammates (2.4 with empty-net goals removed), against about half a point for forwards. We don't know why, and we'd rather not guess.

So, are there real stat padders?

Barely, and not the way the phrase suggests. There is a small, repeatable difference between players, and some of it is clearly empty-net scoring. But it looks like role, not a talent for padding. The players with the most padding are mostly the ones deployed to protect a lead, and the players with the least are mostly the ones who produce while the game is still live.

With more than a thousand players, a few are bound to look extreme by luck alone. In our shuffle tests, the most extreme player at the high end landed anywhere from about 3.3 to 5.9 standard deviations above his teammates, depending on the shuffle, and the real top player, at 5.0, falls inside that range. We can't say for certain that any one of them is a true padder. The clearest individual result is at the other end: Dan Boyle's low padding, 5.5 standard deviations below his teammates, is further out than any of 80 shuffled versions.

What this doesn't show

The baseline can't see ice time, which line a player was on, or power-play usage, and those are real roles that "excess" partly measures. We only see goals a player was involved in, so a defender's shift without a point is invisible. Moving teams is a small sample. And "padding" isn't a judgment: a player who scores when his team already leads is doing exactly what he's asked to do.

For the team-level version of this question, see Is Killer Instinct Real?. To look up any player's own numbers, try the Season Leaders and All-Time Leaders, and the FAQ for how "decided" is defined.