The model was based on identifying & compiling a number of factors that predicted finishing position (eg. the relationship between goals scored & league rank and others), deriving their relative effectiveness, then blending them together proportionally to create one master figure based on last year's numbers.
To this heady mix, I then added or subtracted a percentage based on 3 factors: managerial change, personnel change and participation in Europe. This bit was, well, subjective.
At this point Man City were projected for 160 unit points so a degree of levelling was incorporated (based on historical points totals) to allow the prediction to actually look like a functioning and viable league table.
And, hey presto:
At this point, I should probably state that I'm not wildly enamoured with the model's results, they're too dull!
On the positive side of things, a large part of this resembles the betting market for the league so, I must be in good company if I can model something that looks similar to the pro-odds compilers. But, for me there's not much fun in that.
So by way of a rider, I thought i'd add that personal bias (and knowledge!) predicts a far less average season for Tottenham, i'm not as convinced as the model that Man Utd will emerge so freely from their slump and I think Sunderland are better than I've predicted. I personally think it'll be tighter at the top too, the model loves Man City and isn't too enthusiastic about Arsenal. My view is that they will be a lot closer with Chelsea right beside them. Also, pity Palace: Pulis leaving has taken them from looking up the table to very firmly looking down.
Anyway: it was a fun project & i'll be interested to look back on it in 9 months time.
Thanks to Simon for the impetus.
James @ The Big Ripple