Odds and Expectations Ahead of Man City-Liverpool

There’s a reason why casinos make money (well, at least most of them): many people don’t understand odds. The same is true for lotteries the world over, as well as sports betting agencies.

While we might be able to consciously tell ourselves that in a given football match the three possible results—win, lose, or draw—are not equally likely, there is still a deep feeling that all three outcomes are (though for Everton Football Club that certainly appears to be the case). It is precisely these feelings that get gamblers in trouble.

I bring all of this up because of some misconceptions I’ve seen on here and Twitter over the FiveThirtyEight Club Soccer Predictions. The misconceptions seem to fall into two opposing categories, either those who see them as “Gospel Truth” or “Complete Bullshit.” As of right now, the site is listing the Reds as 78% favorites to take the league title.

To some, these numbers are meaningless and seemingly pulled out of thin air. The site’s apparent “failure” in their political predictions (more on this in a bit) gives these skeptics a rather large pole with which to beat them.

I’m not here to explain every last detail of their models, besides, they do a better job of it themselves. Rather, I’d just like to point out the fact that they’re open about their methodology. If you disagree with one or more of the metrics they use for their predictions (e.g. non-shot expected goals), there is a discussion to be had. Perhaps better models can be built with different data, or different weights given to the various metrics.

(Hang with me for a second while I relive the horror of 2016, it will be relevant to Liverpool Football Club, I promise):

On to the political point: the models they use for sports predictions are completely different from their political ones. And as for 2016 (the one which draws the most ire from critics), the writers for FiveThirtyEight repeatedly warned that Trump winning was a real possibility, even writing the weekend before the election that “Trump Is Just A Normal Polling Error Behind Clinton.” Their model reflected this, showing roughly a 30% chance for the Orange One in the days leading up to the election. (The model is often misremembered for giving Clinton huge, 95%+ odds of winning, but that’s simply revisionist history).

This was up from weeks prior, namely because of a large polling hit following the infamous Comey letter. Prior to the letter, Clinton appeared to be headed for a victory. In fact, three weeks before the election, Nate Silver wrote the following:

We’re getting to the point where a Clinton loss would require either an “October surprise” — maybe Wikileaks has something more damaging up its sleeve than what it’s shown so far, although even then it could be drowned out by all the news Trump is generating — or a significant polling error.

Clinton’s lead was 6 to 7 percentage points in the polls, translating to a 83-86% chance of winning in their model. Then the Comey letter hit (just about the definition of an “October Surprise”), cutting her lead in half, reducing her odds, and opening herself up to a polling error (and the goddamned Electoral College).

So how does this translate to Liverpool’s odds? (See, I told you I’d bring it back around). At 78% favorites, the model is telling us that we need a few things to go horribly wrong, just like in the 2016 example, in order to lose the title. If Liverpool lose against Manchester City, expect to see those odds to drop significantly (probably closer to the mid-to-high 60’s, I’d guess). But Liverpool would still be favorites. The model even accounts for this possibility, making City sizeable 44% favorites for winning at home (compared to 32% to lose or 24% to draw). However, Liverpool’s 4-point buffer would give the Reds some margin for error in the run-in. This is to say that like Liverpool in 2013/14, City very well could win the battle, but lose the war.

A Liverpool draw (or especially a win)? I’d expect the Reds to become even bigger favorites. With one result, City would lose their best chance to pull the Reds back closer to touching distance, and would need to better Liverpool’s results in 3-4 matches (if it’s a draw) or 4-5 matches (if it’s a Liverpool victory).

Some see this model (and the aforementioned political one) changing to real world events as a knock against the model. It shouldn’t be. When playing poker your odds can change—often drastically—with a single card. In a tight title race, a single result can have a major impact. Indeed, it would be a much worse failure if the model still showed City as big favorites. If bookies didn’t change their odds when circumstances dictated, they’d be out serious coin.

My point, as rambling and meandering as this post ended up being, is that you can’t just take these numbers on face value without contextualizing them. Liverpool are 7 points clear at the top, having played more than half their games, and are in an excellent position to take the league title; this is something that should be clear even without a computer model.

But, as should be obvious to everyone, it’s not a sure thing. On paper, 78% sounds great. In reality, it’s worse odds than you get “playing” Russian roulette. Until that number hits 100%, mathematically guaranteeing Liverpool their first league title in 29 years, no one should be happy with simply “good odds.” Klopp & Co. simply need to keep winning games until that happens.