I thought I would share my thinking around a stats model that I designed to show player performance. I've named it the PDB Model. (or PDB for short)
As it's something that took me a while to work out, I won't give away everything, but here is the basis of it and how it may show who is currently playing well, or who is losing form. I also want to see how accurate it may be as a prediction model too.
I will use the Challenge Tour as my example, as there have already been five events, so quite a bit of data.
It is important to recognise who won and who got to the later stages, however this doesn't necessarily mean that the best player won. So as an outright prediction model, this may not be useful. Although, maybe in time, my PDB Model will align. For now I would be looking at single match prediction .
Some weight is given to the round a player got to, as that does show performance, but other factors will show a truer reflection, in my opinion.
For example, when looking at the stats for averages for the first 5 events, you will see that out of the top 10 players, only 3 of the players are in the top 10 on the Order of Merit. I have been told by several experienced players (both pro and amateur), that you shouldn't look too much at averages. The win is the important thing. And this is true. Firstly, a losing average, doesn't give true result, as you may not have had darts at a double, so the average remains high. If you miss, say, 6 darts at a double, the average will tumble. Secondly, you could have a poor winning leg, but a great losing leg, in terms of average, and this will give you a lower overall average than you might expect. Therefore an average is more of a guide. It does have merit, and is included in the model, but is certainly not the be all and end all.
My initial thoughts for power scoring, were that it holds a decent amount of weight when it comes to winning matches. The idea is that a player can get to a finish with extra darts in hand to hit the double.
At an amateur level, I think this is true, but as the standard improves, then maybe less so. For the frequent 180 hitters, this didn't have too much impact, with only Beau Greaves and Mervyn King making the top 10 in the 180 frequency category and overall Order of Merit.
Things become slightly more interesting in the 140 category where the top 4 players in this category make the top 10 in the Order Of Merit. So should more weight be granted here? Obviously this is only 5 events, with limited data, but will be interesting to see if this changes.
My model is not set in stone. It's a work in progress. The aim is probably impossible, but it's to try and get as an accurate performance/prediction model as possible.
I have a couple more factors such as previous form and do the trends show a player improving or declining over time. More data is needed to tweak these results, but they are certainly things to consider.
As Bobby George famously said 'trebles for show, doubles for dough', I think it is necessary to look at checkout percentages and how to weigh them in the model
I was surprised to see, that so far, only two players, Stefan Bellmont and Bean Greaves, are in the top 10 in this category. They did win 3 of the 5 events between them, so it is important, however, beyond these players, it doesn't appear to have that much of an impact.
So, I set up my own PDB rankings to show what the model thinks are the top 10 performers so far. The maximum score a player can get is 800. (Someone has already done this, but only in a single event - more below)
Beau Greaves 250
Stefan Bellmont 146.7
Mervyn King 110
Tom Sykes 96.7
Andreas Harrysson 76.7
Graham Hall 73.3
Martin Grearson 73.3
Carl Sneyd 70
Ryan Harrington 66.7
