Starting in version 2.3, Team Up can calculate an Event Performance Rating (EPR) for each player in an event, which serves as a numerical measure of each player's performance in the event.

​

The concept is that the EPR attempts to take into account how well each player performed given who they partnered with and who they played against. For example, if you have a good result partnering with a player who isn't playing well, that would boost your EPR. On the other hand, if you have a bad result partnering with a player who is playing well, that would hurt your EPR.

​

The basic model is that you add up the EPRs of each player on a team to get a team EPR. The team with the higher EPR is expected to win and the ratio of the EPRs should equal the ratio of the points scored.

​

For example, if Team W has a team EPR of 30 and Team L has a team EPR of 20, the ratio of the team EPRs is 30/20, or 1.5. In a game to 15, Team W is expected to get 15 points and Team L is expected to get 10 points. Assuming Team W wins but Team L scores more than 10 points, Team W underperformed and Team L overperformed.

​

To calculate the EPRs for all the players in an event, Team Up looks at all the games played and attempts to pick EPRs for all the players such that underperformance and overperformance is minimized over all games. Sometimes it does a good job and sometimes it doesn't! The header of the Records screen shows an estimate of the error in the process. Next to the word "EPR" is the sample standard deviation when two or more games are used in the calculation. To calculate the standard error, all winner scores are scaled to 15 and loser scores are scaled by the same amount. The approximate average of all EPRs in an event is 15.

​

In the example Records screen on the right, we see that Jen Brannen appears to be having a great day, with a record of 3-0. But her EPR is just 14.0, which is about average. Why? Because she won games in which her partners were Elliot Walsh and Alan Jones, who have the highest EPRs of the event. In other words, she won mainly because she happened to partner with the two best players, not because she had an amazing day.

​

Equalize EPR vs. Equalize on Points

​

On the Edit Configuration screen, you can choose "Equalize EPR" as a priority for creating matches when Generate is selected. With this option, Team Up will try to equalize the team EPRs of the two teams in any matches.

​

If you want to create equally matched teams, Equalize EPR is probably better than Equalize on Points because the EPR calculation takes into account all the partners and opponents that a player has played with or against in an event.

​

Negative EPR​

​

Sometimes a player's EPR can go negative if they are having a bad day. Theoretically, that means a team EPR could also be negative. However, that breaks the model because if the ratio of two EPRs is negative, there is no way to use that ratio to estimate the expected score of the losing team. To deal with this sub-optimal situation, the model simply uses an EPR of 1.0 for any team EPR that is less than 1.0.

​

Teams with Unequal Numbers of Players

​

Although the Generate screen only generates teams with equal numbers of players (e.g., 2 vs 2, 3 vs 3, etc.), you can use the Choose Teams screen to create teams with unequal numbers of players, like 3 vs 2. The model handles this the same way as ever: it adds up the EPRs of both teams and creates their ratio. For example, if Team W has three players with EPRs 10, 15, and 20, while Team L has two players with EPRs 10 and 20, the team EPRs would be 45 and 30. So the expected score in a game to 15 would be 15 to 10.

​

In other words, if five equally-skilled players play 3-on-2, the expected score would be 15-10. Is this a good assumption? It probably depends on playing level. For open-level players, maybe the score would be more lopsided. For B-level players, almost anything could happen. At any rate, this is the simplest assumption and it works reasonably well for the occasional situation in which the model must handle that situation.

​​

Rally Scoring vs. Old-School Scoring

​

The model treats all games as if they are rally scoring. However, the model assumptions don't really work well for old-school scoring (in which you can only win a point when you serve). To see why, consider two games that end 15-13. Game #1 used rally scoring and game #2 used old-school scoring. Both games were close, but game #2 was potentially far closer. Game #1 had exactly 28 serves. Game #2 could have had many more. For example, if game #2 had had 70 serves, the equivalent rally scoring score would have been 36-34, not 15-13. That's a ratio of 1.06 (game #2) versus a ratio of 1.15 (game #1), so game #2 was actually a closer match. Unfortunately, the EPR model has no way of knowing that game #2 was much closer than game #1, and so it treats them as equally close.