LPGas
Stuart Taylor
Joined: May 2014
Posts: 1,240
|
Post by LPGas on Mar 5, 2016 14:20:45 GMT
If I could predict football matches I would be a millionaire
|
|
old_fogey
David Williams
Joined: May 2014
Posts: 18
|
Post by old_fogey on Mar 5, 2016 14:27:57 GMT
You only need to predict them better than a bookmaker. Of course you also need them to keep taking your bets, once you've shown them you can?
|
|
irishrover
Global Moderator
Joined: June 2014
Posts: 3,372
|
Post by irishrover on Mar 5, 2016 15:42:35 GMT
Some quality statistical analysis on the forum - well done sir! A cursory look at our fixture list suggested to me that it's hard to see anyone else having an easier run in. That's kind of what this analysis suggests. It's interesting that in your scenario we would struggle to make the playoffs if T10 teams were to do poorly against the rest on block. I assume that is because we are a T10 team and we have a disproportionate number of games remaining against those sides, therefore in that scenario we would do worst proportionally than other T10 teams who had fewer games v the rest. Good news for us is that is probably the least realistic of your assumptions. If you wanted to be a bit more fancy/are completely bored you could replace your ideal assumptions with known priors. Re; we already know how T10 teams are doing against bottom 14 sides - we can figure out what the average points per game is in those matchups and extrapolate to the rest of the season. You could even be really fancy and control for how sides do home vs away. Then you are calculating based on known properties rather than ideal theoretical scenarios. Of coruse if you were being really, really fancy, you could move from fixed effects (assuming all your observations will behave equally) to random effects (allowing the effect of each observation to vary) and work that out for each club. You'd say goodbye to your weekend but you could feel justifiably smug doing it and I'm pretty sure it would be good news for Rovers. My god - I've managed to get the language of Bayesian statistics and multi-level modelling onto a Bristol Rovers fans forum! I don't know about multi-level modelling, but we essentially have 24 non-independent, non-Markovian stochastic processes describing the game by game change in the number of points gained by the clubs in the division (that's racked up the amount of statistical terminology ). Anyone able to come up with a sensible model of this and to run the tens of thousands of simulations necessary to provide insight into our likely finishing position, within a weekend, is more than justified in being smug. Things however, are not so simple . One thing that two seasons ago will have taught us, amongst others, is that GD (and I don't mean our ex-director!) is important. We really need to treat the random variables being modelled as multivariate (well bivariate at least) rather than univariate, and model both points obtained and GD. (amount of stats termnology still increasing ). Anyway, enough of this frivolity! If I were a betting man I would be fairly sure that the outcome of any sensible stochastic modelling based on estimates gained from the current season, would show:- 1. We will almost certainly finish somewhere between 2nd and 12th (incl). 2. We are most likely to finish somewhere between 5th and 9th (incl), probably being in rather than out of playoffs as the more likely - which I think is what GasMacc1 is suggesting.
Of course being more likely doesn't mean it is bound to happen!!
Ah well, time to go and lie down in a darkened room for an hour to prepare myself to be put through the mincer listening to the game on Gas Player. True - but even in a multivariate environment you would still have unobserved variables biasing your estimations. For example, I always think the key in lower League football is keeping your players fit down the stretch. When there isn't much between the sides it often comes down to whose goal-scorers and key performers stay on the pitch. All with a whole range of other potentials (dodgy referees, call offs, food poisoning, players falling out with each other) Can't really model that - hence why football is exciting and unpredictable whereas statistical analysis is more often than not boringly predictable no matter what fancy method is employed. Football>Statistics on the whole unfortunately for me because I should really be doing the latter this Saturday afternoon.....
You're right though I think - it is the non-independence that makes team sport much harder to model in that way. I once saw a very lazy student present his plan to statistically explain the phenomena for why teams who are bottom at Christmas always get relegated. He droned on for ages presenting various spurious methods for explaining what he claimed was a 'statistical mystery'. Someone put their hand up and, not unreasonably, pointed out that the reason teams that are bottom at Christmas get relegated is because they have already proven themselves to be not very good at football!
|
|
|
Post by Ilfracombe Gas on Mar 6, 2016 10:10:10 GMT
Dont forget the Trevor Kettle factor as Accrington found out
|
|
Igitur
Joined: June 2014
Posts: 2,294
|
Post by Igitur on Apr 4, 2016 10:57:21 GMT
FixturesRovers have the fewest games (3) against T10 teams: Wimbledon (H); Mansfield (H) and Carlisle (A). Wycombe have the most (7): three at home, four away. All of the other T10 teams have five games, except for Portsmouth and Wimbledon, each with six. MethodFirst, I should point out that this is not an attempt to predict results! I’m merely trying to gauge the influence on the range of possible outcomes of the fact that matches between T10 teams must result in dropped points…and, knowing Rovers have fewer such matches to play, to what extent Rovers could benefit from that fact. Clearly, a superior run of form between now and the end of the season will stand any club in good stead…but this analysis is going to assume that each club shows the same form as each other. By taking form out of the reckoning, we can see what effect the inter-T10 matches in isolation might have. So… I’m going to explore extreme outcomes of: - games between T10 teams, and
- games between a T10 team and any of the other 14 teams in League Two.
Consider a match between any T10 teams. The lowest total of points gained by those two teams is two - from a draw. So Extreme Scenario A is when ALL inter-T10 games end in a draw and each team gets just one point per inter-T10 game. The highest total is three - if one teams wins. So the highest total of points gained by those two teams is three (one gets three, the other gets none). But, since I’m taking form out of the equation, let’s assume that the T10 teams all achieve one and a half points per game on average, achievable by winning one and losing one. So Extreme Scenario B is when the T10 teams get 1.5 points per inter-T10 game. Now consider games between a T10 team and any of the other 14 teams in League Two. The extreme here - Extreme Scenario C - is that all such games end in a win - and three points - for the T10 team. The final Extreme Scenario D is that all such games end in a draw. Why not a defeat? Because that would cause non-T10 teams to overtake the T10 teams…and would mess up this analysis! Let’s also consider some interim scenarios: that the inter-T10 teams gain an average of 1.33 points per game (as they would get from W-D-L sequences); and that T10 teams tend to get a healthy, but not unrealistic 1.75 points per game against the non-T10 teams they have still to play. In summary (and giving the scenarios names): Scenario | Name | Definition | A | STIFLE | All inter-T10 matches end in a draw | Interim 1 | FAIR | Each T10 team gets an average 1.33 points per inter-T10 game | B | COLLUDE | Each T10 team gets an average 1.5 points per inter-T10 game | C | DOMINATE | T10 teams win every game against non-T10 teams. | Interim 2 | NORMAL | T10 teams average 1.75 points in games against non-T10 teams. | D | CHOKE | T10 teams draw every game against non-T10 teams. |
Outcomes
The following table shows, for each scenario, whether Rovers would finish in the play-offs and by how many points clear of the 8th place team COLLUDE 1.5 ppg v T10 | 8th (0 clear) | 5th (2 clear) | 5th (3 clear) | FAIR 1.33 ppg v T10 | 7th (0 clear) | 5th (2 clear) | 5th (6 clear) | STIFLE 1 ppg v T10 | 6th (1 clear) | 5th (3 clear) | 5th (7 clear) | | CHOKE 1 ppg v non-T10 | NORM 1.75 ppg v non-T10 | DOMINATE 3 ppg v non-T10 |
GasMacc1, your stats are fantastic, even the attendance ones, do these predictions still hold good?
|
|
|
Post by a more piratey game on Apr 4, 2016 12:10:22 GMT
I don't know about multi-level modelling, but we essentially have 24 non-independent, non-Markovian stochastic processes describing the game by game change in the number of points gained by the clubs in the division (that's racked up the amount of statistical terminology ). Anyone able to come up with a sensible model of this and to run the tens of thousands of simulations necessary to provide insight into our likely finishing position, within a weekend, is more than justified in being smug. Things however, are not so simple . One thing that two seasons ago will have taught us, amongst others, is that GD (and I don't mean our ex-director!) is important. We really need to treat the random variables being modelled as multivariate (well bivariate at least) rather than univariate, and model both points obtained and GD. (amount of stats termnology still increasing ). Anyway, enough of this frivolity! If I were a betting man I would be fairly sure that the outcome of any sensible stochastic modelling based on estimates gained from the current season, would show:- 1. We will almost certainly finish somewhere between 2nd and 12th (incl). 2. We are most likely to finish somewhere between 5th and 9th (incl), probably being in rather than out of playoffs as the more likely - which I think is what GasMacc1 is suggesting.
Of course being more likely doesn't mean it is bound to happen!!
Ah well, time to go and lie down in a darkened room for an hour to prepare myself to be put through the mincer listening to the game on Gas Player. True - but even in a multivariate environment you would still have unobserved variables biasing your estimations. For example, I always think the key in lower League football is keeping your players fit down the stretch. When there isn't much between the sides it often comes down to whose goal-scorers and key performers stay on the pitch. All with a whole range of other potentials (dodgy referees, call offs, food poisoning, players falling out with each other) Can't really model that - hence why football is exciting and unpredictable whereas statistical analysis is more often than not boringly predictable no matter what fancy method is employed. Football>Statistics on the whole unfortunately for me because I should really be doing the latter this Saturday afternoon.....
You're right though I think - it is the non-independence that makes team sport much harder to model in that way. I once saw a very lazy student present his plan to statistically explain the phenomena for why teams who are bottom at Christmas always get relegated. He droned on for ages presenting various spurious methods for explaining what he claimed was a 'statistical mystery'. Someone put their hand up and, not unreasonably, pointed out that the reason teams that are bottom at Christmas get relegated is because they have already proven themselves to be not very good at football!
my friends often think that this forum is full of idiot football fans talking nonsense
I think the above proves, at least, that there's some sophisticated nonsense on here too
|
|
GasMacc1
Les Bradd
Joined: May 2014
Posts: 1,423
|
Post by GasMacc1 on Apr 4, 2016 17:55:02 GMT
FixturesRovers have the fewest games (3) against T10 teams: Wimbledon (H); Mansfield (H) and Carlisle (A). Wycombe have the most (7): three at home, four away. All of the other T10 teams have five games, except for Portsmouth and Wimbledon, each with six. MethodFirst, I should point out that this is not an attempt to predict results! I’m merely trying to gauge the influence on the range of possible outcomes of the fact that matches between T10 teams must result in dropped points…and, knowing Rovers have fewer such matches to play, to what extent Rovers could benefit from that fact. Clearly, a superior run of form between now and the end of the season will stand any club in good stead…but this analysis is going to assume that each club shows the same form as each other. By taking form out of the reckoning, we can see what effect the inter-T10 matches in isolation might have. So… I’m going to explore extreme outcomes of: - games between T10 teams, and
- games between a T10 team and any of the other 14 teams in League Two.
Consider a match between any T10 teams. The lowest total of points gained by those two teams is two - from a draw. So Extreme Scenario A is when ALL inter-T10 games end in a draw and each team gets just one point per inter-T10 game. The highest total is three - if one teams wins. So the highest total of points gained by those two teams is three (one gets three, the other gets none). But, since I’m taking form out of the equation, let’s assume that the T10 teams all achieve one and a half points per game on average, achievable by winning one and losing one. So Extreme Scenario B is when the T10 teams get 1.5 points per inter-T10 game. Now consider games between a T10 team and any of the other 14 teams in League Two. The extreme here - Extreme Scenario C - is that all such games end in a win - and three points - for the T10 team. The final Extreme Scenario D is that all such games end in a draw. Why not a defeat? Because that would cause non-T10 teams to overtake the T10 teams…and would mess up this analysis! Let’s also consider some interim scenarios: that the inter-T10 teams gain an average of 1.33 points per game (as they would get from W-D-L sequences); and that T10 teams tend to get a healthy, but not unrealistic 1.75 points per game against the non-T10 teams they have still to play. In summary (and giving the scenarios names): Scenario | Name | Definition | A | STIFLE | All inter-T10 matches end in a draw | Interim 1 | FAIR | Each T10 team gets an average 1.33 points per inter-T10 game | B | COLLUDE | Each T10 team gets an average 1.5 points per inter-T10 game | C | DOMINATE | T10 teams win every game against non-T10 teams. | Interim 2 | NORMAL | T10 teams average 1.75 points in games against non-T10 teams. | D | CHOKE | T10 teams draw every game against non-T10 teams. |
Outcomes
The following table shows, for each scenario, whether Rovers would finish in the play-offs and by how many points clear of the 8th place team COLLUDE 1.5 ppg v T10 | 8th (0 clear) | 5th (2 clear) | 5th (3 clear) | FAIR 1.33 ppg v T10 | 7th (0 clear) | 5th (2 clear) | 5th (6 clear) | STIFLE 1 ppg v T10 | 6th (1 clear) | 5th (3 clear) | 5th (7 clear) | | CHOKE 1 ppg v non-T10 | NORM 1.75 ppg v non-T10 | DOMINATE 3 ppg v non-T10 |
GasMacc1, ... do these predictions still hold good? To quote Wael al-Qadi “Thank you for your kind words”! To be clear, the analysis was not a prediction. I had noticed that Rovers had already played most of the teams in the top 10 home and away, whereas the others in the top 10 had still to play each other more often. Instinctively, we assume that puts Rovers at some advantage, as the other top 10 teams must drop points when they play each other. But I wanted to try to assess the size of that advantage. In the next two posts, you can see the tables for (a) inter-T10 matches and (b) matches between T10 and non-T10 teams, since 4th March. The average points per team per inter-T10 match is 1.37. The average points per team per match against non-T10 teams is 1.54. Referring to the “Outcomes” matrix (above), that combination corresponds roughly to the “NORM”/“FAIR” square. That indicated, solely through the fixture-schedule, Rovers would be two points clear of the eighth place team. (We were already one point clear on 4th March). Today, Rovers are actually ELEVEN points clear of eighth place. So the extra nine points have come from Rovers superior form in the past month. Since 4th March, Rovers have achieved 2.00 points per game against T10 teams, 0.63 points per game better than the average among T10 teams. Rovers have also achieved 3.00 points per game against non-T10 teams, 1.46 points per game better than the average among T10 teams.
|
|
GasMacc1
Les Bradd
Joined: May 2014
Posts: 1,423
|
Post by GasMacc1 on Apr 4, 2016 18:06:04 GMT
Performance of T10 in inter-T10 matches (from 4th March to 4th April 2016)
At 4th April | Total T10 games | T10 games so far | W | D | L | Points | Av Pts | + / - | Portsmouth | 6 | 3 | 2 | 1 | 0 | 7 | 2.33 | 0.97 | Rovers | 3 | 3 | 2 | 0 | 1 | 6 | 2.00 | 0.63 | Accrington | 5 | 4 | 2 | 1 | 1 | 7 | 1.75 | 0.38 | Leyton Orient | 5 | 2 | 1 | 0 | 1 | 3 | 1.50 | 0.13 | Oxford | 5 | 3 | 1 | 1 | 1 | 4 | 1.33 | -0.03 | Carlisle | 5 | 3 | 1 | 1 | 1 | 4 | 1.33 | -0.03 | Wycombe | 7 | 4 | 1 | 2 | 1 | 5 | 1.25 | -0.12 | Wimbledon | 6 | 3 | 1 | 0 | 2 | 3 | 1.00 | -0.37 | Plymouth | 5 | 2 | 0 | 1 | 1 | 1 | 0.50 | -0.87 | Mansfield | 5 | 3 | 0 | 1 | 2 | 1 | 0.33 | -1.03 | TOTAL Games | 26 | 15 | 11 | 8 | 11 | 41 | 1.37 | - |
|
|
GasMacc1
Les Bradd
Joined: May 2014
Posts: 1,423
|
Post by GasMacc1 on Apr 4, 2016 18:17:37 GMT
Performance of T10 in matches against non-T10 teams (from 4th March to 4th April 2016)
At 4th April | Played | W | D | L | Points | Av Pts | +/- | Rovers | 4 | 4 | 0 | 0 | 12 | 3.00 | 1.46 | Accrington | 3 | 2 | 1 | 0 | 7 | 2.33 | 0.79 | Oxford | 4 | 2 | 2 | 0 | 8 | 2.00 | 0.46 | Portsmouth | 3 | 2 | 0 | 1 | 6 | 2.00 | 0.46 | Plymouth | 4 | 2 | 0 | 2 | 6 | 1.50 | -0.04 | Wycombe | 3 | 1 | 1 | 1 | 4 | 1.33 | -0.21 | Leyton Orient | 4 | 1 | 1 | 2 | 4 | 1.00 | -0.54 | Wimbledon | 3 | 1 | 0 | 2 | 3 | 1.00 | -0.54 | Carlisle | 4 | 0 | 3 | 1 | 3 | 0.75 | -0.79 | Mansfield | 3 | 0 | 1 | 2 | 1 | 0.33 | -1.21 | TOTAL Games | 35 | 15 | 9 | 11 | 54 | 1.54 | - |
|
|
GasMacc1
Les Bradd
Joined: May 2014
Posts: 1,423
|
Post by GasMacc1 on Apr 4, 2016 18:23:48 GMT
When I posted the original analysis (4th March), most Gasheads would have been hoping to protect a play-off place. Now, one month on, if only we could find some statistical basis for proving that Rovers would finish in the top three! Let’s consider only Oxford, Rovers, Plymouth, Accrington and Portsmouth - the five teams most likely to be contesting second and third automatic places and call them the T5. There have been three inter-T5 games since 4th March. Inevitably, that means that the best they could do would be to pick up an average of 1.5 points per game (one team beating the other). If they had been playing non-T5 teams, they could have picked up nine points from those games, an average of 3 points per T5 team per game. In the event, Plymouth drew at home to Oxford, Accrington beat Plymouth, and Portsmouth won at Accrington! So, the average haul of points per inter-T5 game so far has been only 1.33 per team. The only remaining inter-T5 match is Portsmouth versus Plymouth, scheduled for 16th April. Strategically, a draw would be the best result. But now we’re coming down to individual games, who knows?!
|
|
|
Post by DudeLebowski on Apr 4, 2016 18:24:08 GMT
Looks like the Matrix!
Anyway, here's how it's going to go....
Cobblers - Win Yeovil - Win Stevenage - Win Exeter - Draw York - Win Daggers - Win
Promoted automatically on 87 points, thank you and goodnight!
|
|
|
Post by Richard on Apr 4, 2016 19:10:49 GMT
Looks like the Matrix! Anyway, here's how it's going to go.... Cobblers - Win Yeovil - Win Stevenage - Win Exeter - Draw York - Win Daggers - Win Promoted automatically on 87 points, thank you and goodnight! I'm going... Northampton - Loss Yeovil - Draw Stevenage - Win Exeter - Win York - Win Dag and Red - Win Promoted on 84.
|
|
|
Post by pirate49 on Apr 4, 2016 20:46:28 GMT
Have you factored in:
- heavy rain - injuries to key players - a plethora of red cards - dog/beachball/streaker/fox etc on the pitch - spot betting - an outbreak of diarrhoea - Acts of God
??
(P.S......I think we'll be 4th!)
|
|
Bridgeman
Alfie Biggs
Joined: May 2014
Posts: 3,549
|
Post by Bridgeman on Apr 4, 2016 21:07:28 GMT
If only we hadn't lost to Carlisle we'd all be on the promotion bus and not having this discussion on this superb piece of analysis
|
|
|
Post by Blue Mist on Apr 4, 2016 21:57:42 GMT
Who remembers Dover away last season? Statistically that was a punch in the c o c k.
|
|
Bridgeman
Alfie Biggs
Joined: May 2014
Posts: 3,549
|
Post by Bridgeman on Apr 4, 2016 23:21:56 GMT
Who remembers Dover away last season? Statistically that was a punch in the c o c k. Yeah, I wish somebody had warned me to wear a cricket box......
|
|
|
Post by DudeLebowski on Apr 5, 2016 0:45:14 GMT
Who remembers Dover away last season? Statistically that was a punch in the c o c k. I prefer to just remember it as the best pre match pub atmosphere I've ever known when on the road with the gas! Sunshine, packed boozer, BBQ & thatchers gold especially for us!
|
|
|
Post by mehewmagic on Apr 5, 2016 1:29:17 GMT
Performance of T10 in inter-T10 matches (from 4th March to 4th April 2016)
At 4th April | Total T10 games | T10 games so far | W | D | L | Points | Av Pts | + / - | Portsmouth | 6 | 3 | 2 | 1 | 0 | 7 | 2.33 | 0.97 | Rovers | 3 | 3 | 2 | 0 | 1 | 6 | 2.00 | 0.63 | Accrington | 5 | 4 | 2 | 1 | 1 | 7 | 1.75 | 0.38 | Leyton Orient | 5 | 2 | 1 | 0 | 1 | 3 | 1.50 | 0.13 | Oxford | 5 | 3 | 1 | 1 | 1 | 4 | 1.33 | -0.03 | Carlisle | 5 | 3 | 1 | 1 | 1 | 4 | 1.33 | -0.03 | Wycombe | 7 | 4 | 1 | 2 | 1 | 5 | 1.25 | -0.12 | Wimbledon | 6 | 3 | 1 | 0 | 2 | 3 | 1.00 | -0.37 | Plymouth | 5 | 2 | 0 | 1 | 1 | 1 | 0.50 | -0.87 | Mansfield | 5 | 3 | 0 | 1 | 2 | 1 | 0.33 | -1.03 | TOTAL Games | 26 | 15 | 11 | 8 | 11 | 41 | 1.37 | - |
It's early April and mansfield STILL hold the unwanted accolade of not beating a team in the top 14. until recently they had won every game, bar one draw, against the bottom 10, but even that has now slipped, with 3 losses against them. 1 win in 10, and dropping from 4th to 13th is potential sacking form.
|
|
|
Post by gashead1979 on Apr 5, 2016 7:47:20 GMT
Personally I'm worried about Portsmouth crashing the party.
Very likely to win 3 in a row for the first time this season, they have the players and strength in depth.
With what they have, they should really be doing what Northampton are.
|
|