The method we used to correct for survivorship bias seems to have done a reasonable job given the limitations of the data available. However, the delta method will always have flaws – even after addressing this bias. Lichtman has pointed out that an aging curve constructed using the delta method represents a “hybrid” player, and this by itself is an issue. Additionally, our data starts in 2008, so we are unable to include the full career of any player who has played more than 8 years in the NHL (we cannot include Jagr’s entire career, for instance). I think it’s important to note that this is an average change and not the expected change for every single player. Additionally, with any analysis of player aging there will always be players who “defy” the curve, so to speak, and that should be taken into account when looking at any given player. Regardless, I do think this approach could be applied to projection models for WAR as a potential age adjustment (like we used in the Marcel projection above). While it’s not ideal (it seems no age adjustment ever is), I think it’s more accurate than the age adjustment Marcel utilizes.
Given what we know about NHL skater aging based on various methods and approaches (which I covered in more detail in part 1), I find it very interesting that this WAR model, for the most part, lines up with the generally held belief that skaters peak around 24-25 years old and decline gradually after that (there does seem to be some indication here that the decline of players over 30 might not be as steep as is often assumed). While some of the individual components do not show this, the EV Overall WAR and Overall WAR numbers do – I might argue they strengthen this notion. As we get more data, these curves may change, but for right now EV Overall WAR and Overall WAR seem to support what we already know. From an aging perspective, this metric makes sense. In my opinion, that alone lends credence to its validity.
