Thursday, 26 April 2012

Less Significant (On Average)

Many thanks to George for confirming that my numbers regarding significance are at least close to being accurate. In fact, his numbers came in slightly better than mine.

BigAl stopped by suggesting that my significance test was too simple, and that the average (across all matches) probability of a draw is irrelevant, but yes, I had realized it was of limited value, included it in the post as a starting point before rapidly moving on.

BigAl also notes that “Might be wrong. I seem to remember your 3.54 number was just a guess as you didn’t have all the prices”. Partly right – I don’t have all the draw prices of selections prior to 2011, but since January 1st of that year, over 207 matches, the average price has been 3.54 (actually 3.55 now) and I have made the assumption that this price is at least close for the matches prior to this date. I see no reason why this assumption should be invalid.

Having said that, the (Mean) average draw price is just that, average, and as BigAl says, each match is different, which makes using the (Mean) average a less than perfect strategy. My selection draw prices have ranged from 3.05 to 4.3, so comparing matches where the implied probability of the draw is .3279 with matches where it is .2325 is less than perfect. It only takes a couple of 4.2+ prices to pull that Mean average falsely high. (The average person has less than two legs).

I decided it would be better to use the Median (3.5) and the Mode (3.4) to give the implied probability and when I use 3.45 (iP = .2899) as the average, the probability that these results were achieved by chance goes up to .0834 - no longer at the .05 level of significance, but not bad.

Calculating the Implied Probability from the price available is also problematic - if I can back at 3.5, it's because someone else rates the probability of a draw even lower.

I don't have enough data to look at individual prices yet, but it'll come. The Modal price of 3.4 has 24 selections with 8 winners, with the probability that this occurred by chance a whopping .4101.

Anyway, it was an interesting exercise, and I'll probably look at this some more in the close season.

No joy for two XX Draw Selections yesterday in Serie A, with just one Under winning (Lecce v Napoli). The Extended selections also came up empty this week with two losers at Cagliari on Tuesday and Parma last night. The draw on Parma's game v Palermo was trading at below 3.0 after opening at 3.35, so what that was all about, one can only imagine.Oh I know - it's end of the season Serie A! Unfortunately, the game ended 1-2. 

1 comment:

BigAl said...

I've not done significance tests for years but I'm pretty sure you should be doing a "two tailed" test which, if i remember correctly, doubles the chance results are by chance in this type of analysis.

(not that I agree with the methodology as I believe the implied probability assumptions you have made to be incorrect)