Bad at Math: Predicting the rest of the season

By: James | July 15th, 2008

More fun with statistics: I’ve been playing with the numbers available at mlsnet.com and wanted to see if there was a simple-yet-interesting method for predicting the rest of the season. I’m sure there are agreed upon methods for making predictions, but I decided to not do any research. I’m forging new territory in bad decision-making.

I was going to explain why I chose what I did, but I couldn’t do it without sounding like Yoda. The short form: It’s about the goals. (That reminds me of that overstock.com chick butchering the word gold: “It’s about the goallde.” Just me? Turns out she’s German.)

So here goes:
Home games—-
((Crew Home Goals For * Opponent Away Goals Allowed) - (Opponent Away Goals For * Crew Home Goals Allowed)) + (Crew Home Percentage - Oppenent Away Percentage)

Away games—-
((Crew Away Goals For * Opponent Home Goals Allowed) - (Opponent Home Goals For * Crew Away Goals Allowed)) + (Crew Away Percentage - Oppenent Home Percentage)

Here’s the calc for this Thursday’s Kansas City Game. All of the numbers come from here.

Unweighted: ((1.63 * 1.13) - (.63*1.25)) = 1.0544
Weighted: ((1.63 * 1.13) - (.63*1.25)) + (.668 - .375) = 1.3674

The resulting numbers are basically a predicted goal differential. If the predicted differential is less than -1 the Crew should lose. Greater than 1, they win. Between -1 and 1, they tie. In the above example, Columbus beats Kansas City.

The percentage bit is optional. I was trying to weight the value with the difference in result percentages: teams that win more get a slight bump over teams that win less. The max bump in the league would be for RSL @ New England for .579) Without the weighting the Crew end the season with 49 points. With the weighting it’s 51.

Here’s how the numbers work for the 7/27 game in Colorado (it’s not pretty):
Unweighted: ((1.38 * .63) - (1.88*1.5)) = -1.9506
Weighted: ((1.38 * .63) - (1.88*1.5)) + (.5 - .688) = -2.1386

Here are the games, calculated goal differential (weighted), and predicted result:

Kansas City Wizards 1.3674 W
@ Colorado Rapids -2.1386 L
@ Houston Dynamo -0.908 T
FC Dallas 1.2119 W
Real Salt Lake 2.8643 W
@ FC Dallas -0.0336 T
New England Revolution -0.0247 T
@ Toronto FC -1.843 L
New York Red Bulls 2.5157 W
@ New England Revolution -0.8156 T
Los Angeles Galaxy 1.5206 W
@ Chicago Fire -0.765 T
@ New York Red Bulls -0.769 T
D.C. United 3.0747 W

Again, that’s 24 more points this season, ending with 51. Without the weighting the Crew ties FC Dallas at home and ends the season with 49 points.

I ran the same numbers for New England. Weighted, they end the season with 55 points. Unweighted, 53 points. They lose @Toronto, @Colorado, & @DC. They tie Columbus twice, and tie Chicago and KC. Their wins come against RSL, LA, ChivasUSA, San Jose, and @home vs. DC.

That puts Columbus 4 points shy of the Supporters’ Shield.

So how have I screwed this up? There must be a sabremetrician out there ready to try his luck on soccer. If I were a betting man I’d know how to do this. What’s the line?