Do we have a good method of predicting relegation? Assessing last year's results

Earlier today I posted my first numbers-based look at the relegation race for the end of the 2014-15 Premier League season. Give it a read if that sort of thing interests you! But this isn't about that piece. Or rather, it's only tangentially about that.

I went ahead and posted that on Facebook and one of our readers there remembered that I had done a similar exercise last year. She then wondered how accurate it had been. And that got me curious. I posted these things like crazy last year, but I never went back and did a recap of how well they had turned out. So I decided to take a look since one of the things I am striving for in this project is transparency.

At the time (April 25), Aston Villa and the rest of the league had either 3 or 4 matches remaining. I focused on six teams near the bottom of the table and predicted that 36 points would mean safety while 35 points would lead to relegation. I also said that the bottom three would be Fulham, Norwich, and Cardiff (18th, 19th, and 20th, respectively).

By the time the season was over, both of those predictions more or less came true. West Bromwich Albion were the 17th-placed team and survived in the Premier League with 36 points. Anything 35 or below would have meant relegation. But, to be fair, West Brom could have gotten 35 or even 34 points and been safe, because the 18th-placed team finished with 33. And as for those bottom three, while I got all of the teams, I messed up the order. Thanks to an incredible late-season collapse, Fulham managed to be worse than Norwich City.

I did get a couple of things wrong, though. For instance, I said that Sunderland only had a 32.16% chance of hitting 36 points. Instead, they put together a baffling run of results and secured three wins in their last four matches (over Cardiff, Manchester United, and West Brom) to end on 38 and in 14th. I also didn't even think of Hull City as a relegation contender and they did their best to prove me wrong by going 0-1-3 over their run-in. Their one draw actually came against Fulham, but even had the Cottagers won that match it would not have changed much except finishing order.

Nevertheless, at least within the realm of just a few remaining matches, I think that last year's model showed that using odds as a proxy for probabilities and calculating the end-of-season run-in from there is a decent model. Of course, that was only one year. Once this season is officially over, we can look back at our first run now and see if that success was sustainable!