Tuesday, May 25, 2010

Rotating League Name week 6 recap

Chapter 6: In which I crap my pants

Whenever I listen to audio books, I always want the old Brit speaking to say that title. Anyway, here's stuff you might care about.

Decided to see how well overall performance can predict weekly performance in a given matchup, so here's a picture of what I found:

The numbers shown for each week 6 matchup shows how many categories were won by the team with the better overall numbers through week 5. That is, if team A had more RBI year-to-date than team B, and team A beat team B in RBI, that got scored a 1 or W for prediction. If either the YTD or matchup ended in a tie, it was scored a .5 (used the same rules as our standings as ties counting half a win and half a loss).

Of note is the Kid-Pawn matchup where every single hitting category went against the past performance (4-2 Bartha, instead of 2-4). At the same time, every pitching category in that matchup went exactly according to the past (though Bartha almost blew it trying to chase Saves history - which incidentally was the only best/worst change from last week, so I'm not bothering to update it this week).

At the bottom of the picture are measures of how well YTD predicted this week's outcome. I actually ran the figures using the current data (through week 6) first because: 1) I thought it would be easier and 2) I forgot that I had the cumulative stats through week 5 handy in the same spreadsheet.

In general, the predictive power based on this one small subset was somewhat better than just flat out guessing (.500). As you might expect (or maybe not, who am I to know what you expect?), adding this week's cumulative stats produces better overall guess, but not that much better than before. The pitching categories seem to follow form a little closer, and so are more predictive (the hitting categories were dead even). I believe this is partly because the league is set up with extreme hitting strategies (this year at least) and because the presence of several rate-based categories means that one good or bad week is more mitigated than in the counting stats. Indeed, when I separated out the rate categories, they were correct 2/3 of the time, versus around 52% each for the positive and negative (and net) counters.

So, after showing that the predictive power is potentially a problem (I mean limited, but I was on a roll), I decided the best way to use this information would be to make predictions for the next week. Here that is:

Yes, I'm going out on a limb and saying that these scores will be closer to correct overall than just picking 6-6-0 for each matchup. Depending on whether this is true, there may or may not be an update on this next week.

My bold predictions (or those of this system I'm using if they are far off) include 4 out of 6 matchups ending 5-7. Shown in gray, there's only one predicted upset (Rod Beck's Counting). If nothing else, this little exercise allows me time to sit around thinking of (good?) ways to combine team names.

No comments: