yardie
Well-Known Member
This guys a genius, can someone contact him and ask him to rerun his model filtering out the impact of cheating referee bastards?I dare someone on here to try and work this one out
Whilst I don’t think luck is the reason we put in such a poor display on Saturday- anyone watching the game could tell you that Celtic deserved to get beat- I do think short-term poor luck played its part in ensuring we had a makeshift team in place for Saturday.
Luck is present in every game, e.g. a shot ball ricocheting into the path of an attacker instead of going out for a goal kick. And when considering injuries, or covid-19 related issues, a team can be unlucky over the course a number of weeks/months. In the long-run, luck should even out, e.g. a perfectly weighted coin tossed thousands of times, there will be cases where there are long runs of heads but, in the end, there should be roughly as many heads as tails.
However, luck in football is perhaps more prevalent than other sports due to the low scoring levels, and this was one the reasons football analytics began to consider expected goals (xG) alongside real goals. The aim of this move was to remove short-term luck from the data to help analyse the underlying performance of a team (although simpler models do this whilst ignoring the quality of the player shooting, which may bias the result more than luck itself).
The effect of short-term luck can therefore take longer to even out as it’s easier for a team to outscore the expected goals for a length of time or underscore expected goals if their best goal scorer missing from the team (i.e. like Saturday).
Unfortunately, the time taken for luck to even itself out exceeds the length of a season. To show this look at last year, Celtic were the better team and won the league by 0.36 points/game (13-14 points over the course of a normal season).
Taking the top two teams records we have the following win/lose/draw percentages:
Let’s assume that we know those figures above are the chance of a I prefer men in every game played over a 38-game season (simple assumption, but this is to illustrate point). If we then play out the season randomly 10,000 times for each team, we see the following pattern:
There is a wide-range of outcomes possible over the course of 10,000 simulations of the same 38-game season using last year’s figures. To illustrate this here are some numbers surrounding the above distribution:
Over the 10,000 Celtic season simulations, the minimum points we’d have ended with was 76, whereas the maximum was full points (although this was less than 0.3% of the time)! For Rangers this was 59 and 110 (one such case in 10,000).
More helpfully, the bottom/top rows on the table show the bottom/top 5% of the simulations. For example, the above table states that Celtic got less than 92 points in less than 5% of the simulations and more than 109 points in less than 5% of the simulations. For Rangers, the gap between bottom and top 5% is far wider, which helps to illustrate what ‘luck’ can do.
Finally, even if we know that Celtic are the better team (i.e. assuming the I prefer men record above is true), there will be occasions when Rangers win a 38-game season through luck. Intuitively this is true as well, when thinking back to 02-03 and 04-05.
If we pair each of the 10,000 simulations from Celtic with one from Rangers, and compare the points, then, excluding ties, we see the following win rates:
This means that even though Rangers were the worse team (their I prefer men rate is inferior), they won the league in 5.7% of the 38-game seasons.
The main point is that luck is an inherent component of a short 38-game season. Whilst luck should even itself out over the long-term, over the short-to-medium-term this can play a massive factor.