The Puck Stops Here

In my earlier piece assessing shot quality I wrote about the NHL's real time scoring system (RTSS), which has useful data for sabermetrics and hockey studies, but much of this data is poorly tabulated and must be extracted from the various NHL games play-by-play summaries and once that data is extracted it is often of questionable quality including some events that are clearly incorrect (such as 60 foot wrap around shots - which are clearly coding errors) but nevertheless is a useful data set to extract averages over entire seasons. In order to help NHL hockey be better analyzed, systematic improvement of this data set would be useful.

One simple method of improvement would be clearly marking which shots in an NHL game are rebounds. This could be done by the scorer of the game when shot data is inputted with little additional effort. In order to attempt to extract which shots are rebounds, it is common practise to record all shots that come shortly after a previous shot with no other intervening event occurring between the shots (such as a faceoff). However, there are different standards used by different analyses to determine which shots are rebounds.

In his shot quality study, Alan Ryder of hockeyanalytics.com uses the definition that a rebound is any shot two seconds or less after a previous shot with a distance of less than 25 feet. He shows these shots are for more likely to score then a "standard shot". In 2002/03 41.2% of these rebounds score on power plays and 34.8% at other times. This is a rather restrictive rebound definition and may miss some rebounds (rebound shots may be taken three or more seconds after the initial shot), but clearly captures those rebounds where the goalie is likely still out of position from the initial shot.

Recently, On the Forecheck has written about rebounds in the 2006/07 season on both the team and individual level. He uses a much more liberal definition of rebound, where a rebound is any shot with 5 seconds of a previous shot with no intervening event reported. This definition will capture some rebounds the Ryder definition misses, but may also gather false positives when a team attempts to clear the puck after a shot and loses it to the attacking team that quickly shoots or on an individual level when a player picks up a rebound and passes the puck to a second player who then shoots.

This confusion over what is and is not a rebound could easily be removed if the NHL noted rebounds in their real time scoring. Nevertheless, some conclusions can be made from the data. Certain teams (Buffalo, Colorado, NY Rangers took more rebounds then others (Edmonton, Chicago, Minnesota). Certain teams (Pittsburgh, Ottawa, NY Islanders) gave up more rebounds then others (Detroit, Tampa Bay, San Jose). Certain players were better at taking rebounds (Martin St Louis, Alexander Frolov, Jaromir Jagr) then others. The uncertainties in the exactly which shots are rebounds (would we see qualitatively different results with a different definition of rebound) makes it hard to push this analysis to any meaningful next level. However, there are some counterintuitive results. For example, I would have guessed that Tampa Bay being a weak team in terms of goaltending would give up a lot of rebounds. They do not. I would have imagined that power forward players who are able to play in the slot without being removed by defences pick up the most rebounds. None of the leaders in rebounds are clearly known as power forwards. What this all means is hard to tell without better data to study.

The NHL's real time scoring system has the potential to be very valuable to understanding some of the details of what happens in a hockey game statistically. In order to get to this point, we need more accurate data kept by the league and we need a better record of the important events in a game. One simple way to improve this record is to record when a rebound occurs, so we do not need to guess.