Grading Andrew Ference

The Buyout Proposition

This analysis is a direct response to recent articles written by Alan Hull that claim that Andrew Ference will be a buyout candidate by the end of this season. In his latest article, Alan writes:

I feel like I don't need to re-state my issues with the Ference deal, but to summarize, the Oilers needed a top pair (or at least top 4 Dman) and instead signed Ference, a bottom-pairing player at the time he was signed, to a 4 year deal at $3.25M per year and a full NMC. Ference was 34 years old at the time and his decline from effectiveness was totally foreseeable, but the Oilers signed the deal anyway. Making this a 2 year deal still wouldn't have made the signing totally excusable, but it would have greatly minimized the problems that this deal creates. (If only someone could have foreseen this problem!)

Serious statements such as these merit a serious quantitative analysis. So, here goes.

The Gaussian Standardization

The principal challenge of a quantitative analysis of Ference's performance of the past five years is that he has gone from one of the most successful to one of the least successful teams in hockey. Thankfully, statistics has ways of exorcizing such demons of inequality.

Usually in the Corsiverse we discuss raw scores, percents, and averages. We might, for example, say that Andrew Ference had a Corsi-for percentage (CF%) of 43 last season. On it's own this number only means so much. It tells us that he was on the ice for more Corsi events against than for. It doesn't tell us where Ference placed relative to the other Oilers or relative to other players in the league. One way to make Corsi a bit more meaningful is to rank a player's score in relation to other players. We can state that last year Ference ranked 17th on a team that was 28th in the league with a collective CF% of 44.3. This gives us a better idea about what Ference's score actually means (it wasn't that great). A numerical short-cut to this "more meaningful" number is to normalize and standardize so that the number itself represents its context. Such procedures are common to economics and the sciences.

Normalization fits data into a scale, often from 0-1. I decided to normalize the raw scores in this analysis to have values between -1 and 1. I suppose this is just a preference of mine, however, my goal was for "0" to represent the team averages. After normalizing the data I standardized them, meaning I fit them to a curve with a Gaussian distribution. The average was chosen as 0 and first standard deviations were selected to be 1 and -1 respectively. Note that I didn't precisely filter outliers. I chose my criteria for selection of players, which was 100 minutes 5v5 start-zone adjusted.

What this procedure means is that each player's possession, quality of competition, and time-on-ice were calculated in relation to his teamates' for each season. This allows me to plot Ference's performances across seasons with both teams on the same graph: it provides a method of teasing-out Ference's performance in relation to his colleagues on Boston and Edmonton. It grades Ference "on the curve".

The Standardization Presentation

The following are interactive charts displaying standardized CF% (abcissa, Y axis) plotted against standardized OPPCF% (ordinate, X axis). Standardized TOI values correspond to bubble size. Bubble color indicates raw PDO with the exception of Ference's bubble in each chart which is orange. Lines at zero represent team averages, all other lines indicate standard deviations. All data is 5v5 zone-start adjusted from Puckalytics.The higher the bubble, the better the player's possession. The farther the bubble is to the right, the higher the quality of competition as measured by Corsi. The bigger the bubble the greater the relative TOI. The darker the bubble, the lower the PDO score.

Tables may be viewed here.

Tables may be viewed here.

Tables may be viewed here.

Tables may be viewed here.

Tables may be viewed here.

Tables may be viewed here.

The Standardization Argumentation

When standardized, Ference's possession and quality of competition metrics are relatively consistent. He's a kind of "C" student. His possession scores are below-average in "C-" territory between -1 and 0, but close to 0. The Bruins mostly deployed Ference against competition that was approximately equal-to Ference in possession skill. However, the Oilers played Ference against players in "C+" territory in 2013-14 (between 0 and 1).

There is an argument to be made that Ference's 2013-14 performance was one of his best of the past five seasons. He's played tougher competition while maintaining his relative possession score and he's done it while paired with Justin Schultz. Here are Ference's TOI statistics with a TOI% calculated against the total ice-time played by all defensemen.

Ference definitely isn't a first-pairing or top-four defenseman. As a young crop of new talent comes forward and management continues to paint itself into corners he may very well face that buyout. But Ference's age doesn't appear to have caught-up with him yet. In my opinion he's done what's been asked of him to the best of his ability and the results he's achieved are consistent with his play in Boston.