The Points Above Replacement (PAR) statistic is modeled after baseball’s Wins Above Replacement (WAR) stats, and is calculated to have similar properties. Thus, each player is assigned a PAR value that estimates that player’s contributions to his team’s league point total. The PAR stats are not meant to be predictive, or forecast future performance, but are instead of an evaluation of the player’s contribution to league performance so far.
The model is based on widely available player data that considers players’ game actions and performance, playing time, positions and their influence on the team’s points. The performance measure is a calculation created based on widely-available consolidated performance measures, and adjusted to fit the purposes of the PAR. Depending on the league and season, the model accounts for 73% to 88% of the variability in league points. Thus, while it does not match results perfectly, it does rather well, and can serve as a useful starting point for discussions.
Each player’s score is then compared to a hypothetical “replacement” player. As with the WAR, the replacement player is set at a level that could be procured with relatively little cost to the team. An “average” player is likely getting playing time for a competitor, and would have significant transaction costs. In baseball, the goal is to set the replacement player at a level of a fringe player, one of the last couple to make the team, or a minor league player likely to be called up.
For the purposes of the PAR, each league’s “replacement” player is set at the representative level of the bottom three teams. For example, in the English Premier League, the bottom three teams have averaged 31.8 points since 1995. As a result, the EPL replacement player is set at the performance level required to expect 31.8 points (or .83 pts/game for partial seasons). Another way of stating it, if a team was filled with only with the hypothetical “replacement players”, the model would predict the team would earn 31.8 points.
Some key attributes of the PAR:
- A player’s position has a large impact on the PAR, and the calculation can only handle one position description. Thus, a midfielder with a PAR of +1.2, may get a +2.0 if they were a forward. Because “actual” playing position may change from game to game, or even within a game, no efforts were made to adjust a player’s position description. Thus, some hard-to-describe players may justifiably take issue with their result.
- Further, the default score is 0. A player with no time on the field cannot help (or hurt) a team, and is not considered to have any influence on the league table.
- Similarly, playing time matters on the PAR. A great player with two games on the field may not affect the table standings as much as a player who makes a smaller contribution every game. And the PAR also reflects the ability to get on the field. Thus, only those players with lots of playing time can earn the highest, or the lowest, PAR scores. For related reasons, only those players with 500 league minutes are included in the PAR table. Players with less time typically have PAR values below 1.
- While the process for creating the statistic is the same for each league, the actual coefficient values are specific to the particular league, year and team. Thus, one cannot assume that a player (at their reported performance level) would achieve the same PAR in a different league, team or year. But instead, the PAR reflects their value in the given league campaign.
- Finally, as with the WAR, adding the PAR scores for a team should equal the team’s points above and beyond the “replacement” level, although this won’t be an exact match. And each player’s PAR shows how much they are helping the team get above what a typical, relegated team might earn.