## SPAM: A Retrospective - How Accurate is Spam Anyway?

A poster recently posed the question: just how accurate is SPAM? I actually don't know. I can tell you that in seasons past I have used SPAM for pick 'em pools and it does fine but not amazing (like, not good enough to place bets and make a profit).

But, I was curious, so I looked.

Here's the method I used. SPAM doesn't just predict a winner, it predicts how much they will win by. I loaded up SPAM's predictions for 2020, and the results were interesting. On Win/Loss record, SPAM was just 4-4, incorrectly predicting Iowa wins over Purdue and Northwestern and incorrectly predicting losses to Penn State and Wisconsin. In terms of margin of victory, SPAM very much underestimated Iowa's offense (or defense, it's hard to tell which with this method).

To calculate by how much, I looked at what Iowa's average game outcome was predicted to be compared to what it actually was, but just averaging the score differential. For example, if Iowa was predicted to win Game A by 10 points, and to lose Game B by 4 points, the average outcome between those two games is Iowa by 3 (add the two margins of 10 and -4 to get 6, divide by 2 data points). You can think of this as SPAM's prediction for what the average outcome of an Iowa game would be in 2020.

SPAM predicted that Iowa's average game outcome was an Iowa win by 4.35 points. However, the actual average outcome was an Iowa victory by 15.75 points. This means SPAM underestimated Iowa's performance by a whopping 11.4 points. This might be an artifact of when I developed the algorithm, which was in the middle of the 2013 season shortly before the Ohio State game.

Finally, SPAM also predicted a 4-4 season as the most likely outcome (24.5%). Iowa actually went 6-2, which SPAM had given 20.5% chance.

But that's not all! Let's look at some more seasons.

In 2019, SPAM predicted that Iowa's average game outcome would be a win by 7.1 points. However, Iowa's actual average outcome was a win by 11.5 points, meaning SPAM underestimated Iowa by 4.3 points of win margin per game. SPAM also predicted 8-4 as the most likely outcome (24.33%), but Iowa went 9-3 in the regular season, which SPAM gave only a 16.88% chance. On W/L, SPAM picked 10 games correctly at the start of the season, missing only on PSU and Wisconsin. Iowa lost to both, but SPAM predicted Iowa wins against both, though by slim margins (in Wisconsin's case, only 0.59 points).

In 2018, SPAM predicted an average outcome of a 5.9 point victory, but Iowa's average outcome was a 14 point win, though this was inflated by the 63-0 game against Illinois. On win/loss, SPAM was again 10-2, missing on the Northwestern game (which SPAM had a win but Iowa lost) and the Minnesota game (which SPAM had a narrow defeat but Iowa won by 17). Also of note, SPAM predicted that god-awful Purdue loss correctly.

I didn't post any SPAM results for 2016 and 2017 for some reason. Laziness, probably.

In 2015, SPAM predicted 10-2, and of course, Iowa went 12-0.

So, since 2015, SPAM is I believe 34-10 on win/loss record and has consistently underestimated Iowa by a significant margin. 2014, the first full year when I did SPAM, is the only season I have on record where SPAM over-estimated Iowa, and 2014 was of course truly a terrible year. "That's football." The Hawkslayer Bowl. Awful losses to Maryland, Nebraska, and Minnesota.

So, let's talk 2021, what does this portend for the Hawkeyes this year? Seriously, I have no idea. I think this is one of the most difficult years to predict in this fashion because the projections are based on last year's data, which is all highly suspect because so many teams had so many external factors impacting performance that are highly unlikely to have that kind of impact this year. So, yeah. Don't place any bets based on SPAM this year.

Unless otherwise expressly indicated by BHGP editors, this FanPost is strictly the viewpoint of the author and is not endorsed by BHGP in any way.