I’ve been too busy, and the game weeks have been coming too quickly for me to update these lately, but I have been thinking about it and have modified the formula slightly.
For full disclosure, here are the last two GWs results. Nothing spectacular I’m afraid!
In GW15 the model had 1 exact correct score, and another 5 games where it was just a single goal away. In GW16 it only had two games where it was a single goal away – a little disappointing.
Up until now I’ve been using data for the entire season, which has been fine as I needed the data to build. However now we’re 16 weeks in, the early season results can influence too heavily the last few games form, so I’ve tweaked it to include just the previous 8 games.
Using the previous model, for GW17 we had the following predicted scores:
And using just the last 8 games data, we have this:
There aren’t huge differences, but that’s to be expected – generally the good teams are still good, and the bad teams are still bad, but there are subtle differences.
The Utd-Bournemouth game has changed from 3-0 to 2-1. I might not agree with that, but that highlights the drop off from Utd’s early games where they were putting 3/4 past teams each week. Brighton are also thought to score at Wembley, and this is because in their 8 away this season they’ve scored 5 goals, but 4 of those have come in the last 4 games.
I’ll monitor both outcomes for a few weeks and see which works better, but I feel recent form should become the better model.