flickr |
Welcome back, to another round of ShotCaller. On Friday morning, I made the following prediction:
The Prediction:
During the 1st quarter of the New York Knicks sixth game of the year, against the Brooklyn Nets on November 7th, Carmelo Anthony is expected to take 6 shots; those shots are expected to occur within each of the red ovals.
And here’s what happened…
The Result:
Well, that’s not so bad at all, and it’s certainly an improvement from the first one. But how good is it really?
For starters, the 6/6 is a strong win for the Prediction Crowd. I laid out in decent detail why Melo would take exactly 6 shots during that 1st quarter. It would appear that everything held true to form. So, if you’re scoring at home, the combination of a) career-long trends, b) game-to-game differences, and c) the opponent, all appear to have strong predictive value — especially when examined with appropriate weighting in conjunction with one another. Does this mean I’ll get every 1st quarter correct going forward? No, not a chance. But it does mean I’ve unlocked at least a piece — an important piece, no less — to understanding certain behavioral rules to decisions made in a pretty dynamic environment. Applying this process to other players as the season continues is a challenge I will (eventually) accept.
Also, this is the semblance of an early trend worth monitoring:
This is just a micro-view of a marathon; not even 10% of the season has occurred. Arguably any six-game snapshot of Melo’s career would appear to be trending in a particular direction; it’s just so interesting when it’s the start of a new season. So there are likely to be periods of simple up-down smoothing preceded by volatility. The length of these periods, and their predictability will be a fascinating study going forward.
Let’s also discuss the shot locations, of course. I was a bit giddy during the game because at a quick glance it appeared that four of the six shots were bulls-eyes. Upon further review… I came close, but there’s still work to be done. Technically, there are zero bulls-eyes; none of the shots fell directly in any of the predicted areas. However, 0/6 doesn’t really do this performance justice either, does it? Let’s measure:
- The shot at the rim was about 2ft off
- The two midrange jumpers were less than 1ft combined from that predicted area
- The above-the-break 3pt shot from the left side (same side as the shots above) was about 2.5ft away
Running total: 4 shots combined for less than 5.5ft away from predicted areas.
As for the other two shots:
- The right-wing 3pt shot was about 10ft away from the predicted right-wing 3pt shot (right track, wrong train)
- The top-of-the-key jumper has no noteworthy/deserving matched prediction; all indications had me along the baseline, nowhere near the middle of the court
My original metric (remember 20% from the first evaluation??) is whack, and already needs an overhaul. I’m hitting the drawing board this week, in an effort to capture count and distance. How do we describe the performance of this prediction? It’s 6/6 and it’s 0/6 at the same time. It’s so very close on 4 of 6 shots. It’s way off on 1 shot. Most important (to me, at least): it’s substantially better than the first prediction. We’re learning, and the process should be better every time. The question becomes how long until it’s iron-clad and usable.
And on that note, I should probably just end this with a song.
Data and photo support provided courtesy of NBA.com, Basketball-Reference.com, and Darryl Blackport.