2016 Supercross Predictions - 450SX Update going into Week 6: San Diego 2
2/12/2016
This week, for the first time we'll also have a "this week" ranking. It's still very basic -- a much more involved ranking is in the works -- but we'll get to that. First, the updated full-season prediction:
At the top, there are very few changes, once again. Ryan Dungey with a big lead and Ken Roczen comfortably in second. Dungey's projected lead is slightly slimmer (55 points vs 64 last week), and Roczen's is slightly larger (33 vs 29). The only real news is that Jason Anderson passed Cole Seely to take over 5th (sounds like we're watching a race). Some might think that Anderson should be higher up and/or closer to Eli Tomac, but Anderson's 2015 still weighs pretty heavily in his projections, and 2015 for Anderson wasn't great.
Trey Canard falls to 9th, due to missing 2 races' worth of points more than his expected performance the rest of the way. Canard trails Marvin Musquin and Davi Millsaps, who both trail Seely and the Top 5 by a considerable margin. And Justin Brayton is a fraction of a point behind Canard. Brayton has outperformed our projections for him so far this season, even if we hadn't included a downward adjustment for his older age.
Christophe Pourcel made a big jump, up to #11, based on consistent results between 12th and 14th over the last 4 weeks, but the jump is larger than normal because he doesn't have a 450SX track record, so every 2016 race has more impact for him. The same is generally true for Justin Bogle and Lawson Bopping, who moved into the rankings at #19 and #20, respectively (Bopping was also helped by returning from injury in San Diego 2, which was a week earlier than we expected).
Another note, sort of sad, is that James Stewart falls out of the Top 20 since he's out indefinitely. Even if we predicted him returning at Week 10, he still wouldn't make the cut because his 2 DNF's in 2016 are really weighing down his projection. If he does come back at some point and puts in a reasonably good finish, I think he at least gets back on the list. But not for now.
The season predictions above are based on the rider's history (at this point, 2016 through 2013), the rider's age (see Aging Curves), and the rider's experience (or lack thereof) where we regress the prediction toward the mean (in the case of 450SX riders, that's almost always down) because "average" is typically a better prediction than "whatever the extremely small amount of data says" (see Footnote). We want to add a separate MotoXGraphs prediction that takes into account if the rider was recently injured and is likely to be affected. While eventually a much more complicated analysis on injuries is hopefully happening, right now the "recently injured" effect is pretty basic, and more of an indicator that the rider was recently injured than an actual, precise prediction of how that recovery from injury will affect the rider. In the second column, the adjustment has been applied; as I said, this is a simplistic version and hopefully more involved and accurate iterations will follow:
So, for instance Eli Tomac, having been "recently injured", we're showing here that we should take into account that his current riding performance is not expected to be as high as if he hadn't been injured (i.e. his history/age/experience projection). This adjustment drops Tomac down below Chad Reed by a fraction of a point. Again, this is only meant to be an indicator that Tomac is dealing with an injury, not a precise prediction that the injury will drop him below Reed.
Same with Trey Canard, who has a more direct case, I think. Returning this week from injury, our injury adjustment points out that he may not be riding at full strength, which hurts his rankings -- causes his ranking to drop by 2 spots after applying the injury affect.
Which brings us to the "this week" ranking. The idea is to show only the estimated "true talent"of the rider. By "true talent" we just mean our best guess at the actual ability of the rider, based on what the data is telling us. So, where the "true talent" measure in the chart below differs from the season projection above is that the points in the standings variable has been stripped out, so all we're left with is rider history + age effect + experience. While the season projection above thinks Trey Canard is going to finish below Marvin Musquin and Davi Millsaps because they already have a big lead in points, the True Talent projection below thinks that in any one race Canard is expected to finish above those two.
In addition, there's an accompanying column that takes into account the "recovering from injury" factor that was discussed above. Again --even more important than above, I think-- this is just an indicator, not an exact prediction of the effect of the injury. While the true talent estimate for a specific week is fairly simple, we do hope to expand it to include location/city, the course format, and perhaps beginning/middle/end of the season, to the extent that those are shown to be applicable. It could be really useful for motocross/supercross Fantasy Leagues, but the actual predictive power of this True Talent measure hasn't been tested yet for that purpose, so use at your own discretion.
(Broc Tickle is scratched out above because he's not riding in week 6 at San Diego 2. James Stewart doesn't show on this list because he's not in the Top 20 right now, but if he did, he'd be ranked around #11, behind Millsaps and above Brayton -- not because Stewart would be expected to finish #11 exactly, but because he has a pretty equal chance of finishing in the Top 5 as not finishing at all.)
Footnote: (adapted from the link above regarding the definition of true talent) Let's say I asked you to predict a rider's finish in the next few races, but all you knew about this rider was that he had won the previous race. Well, you probably wouldn't guess that he'd will all of the next races. You'd figure that the rider was probably around average, but he had won at least one race, so the answer is probably somewhere in between "average" and "wins all the races". And statistically speaking you'd be best off to guess closer to "average".
Now what if I told you that he had won 5 out of the last 6 races? Here, we have more data to go off of, so now we start to estimate that the rider is closer to the top of the class than just "average". Maybe we'd guess that he'd finish in the Top 3. What if he'd won 9 of the last 10? Then the best prediction might actually be that he'd win the next race. So, the moral of the story is that the more data we have, the more accurate we can be, of course; but the less data we have, the better off we are just guessing "average" rather than relying too much on only the limited data.
Posted by: SagehenMacGyver47 ::: As always – Feedback welcomed
Previous: How much slowing? -- Follow Up | Back to Blog | Next: 2016 Week 6 - By-The-Numbers Recap: San Diego 2