some comments.......

I think the methodology is fine, the points totals/distributions look very reasonable based on each finishing position.

Your results show the problem when dealing with small numbers.

Toss a coin 5 times and you could get 4 heads and 1 tails
Toss a coin 500 times and its virtually impossible to get 400 heads and 100 tails with a fair coin toss.

Looking at the B'mouth Data , their finishing position is based on very recent form, yet the bookmakers still rate their chances highly as they are basing their data on quality of opposition in the run in.
They are also basing their chances of WLD on their seasonal form, not recent form.
Seasonal form is more balanced over 36 games. It has the correct balance of playing a wide variation in teams home and away.
If the run in is perceived as easier you would expect anti-correlation as you have had a harder time than average in the previous 36 games.
Recent form could be based on playing a smaller subset which could be tougher or easier than the average match.

Posted By: usacanary, Mar 9, 12:40:03