This is really brilliant, @Callumcafc - thoroughly absorbing and instructive set of data, which will be really interesting to track of the course of a season.
One thing: if I understood your initial post correctly, I think the following bit is wrong:
To calculate Charlton's SOS, we take only the teams we have beaten (Derby & Plymouth) who have seven wins, two draws and three losses across 12 games.
That should be SOV (not SOS). If I am wrong, then I haven't properly understood it at all and you can ignore this post..!
This is really brilliant, @Callumcafc - thoroughly absorbing and instructive set of data, which will be really interesting to track of the course of a season.
One thing: if I understood your initial post correctly, I think the following bit is wrong:
To calculate Charlton's SOS, we take only the teams we have beaten (Derby & Plymouth) who have seven wins, two draws and three losses across 12 games.
That should be SOV (not SOS). If I am wrong, then I haven't properly understood it at all and you can ignore this post..!
Nope you're right - thanks for catching it. First post has been edited now.
Once we get halfway through the season and everyone has played each other does the SOS stat not become obsolete?
Then presumably the SOV stat will start looking pretty similar to the actual league table?
Love a good stat though so looking forward to the update. Reckon 10-15 games in would be a good reference point
Yep, as a quirk of everyone playing each other twice, it should start to tend that way at the halfway point, but once we get to the 30-35 game point, it should start to diverge again and eventually come back together by the end of the season.
Once we get halfway through the season and everyone has played each other does the SOS stat not become obsolete?
Then presumably the SOV stat will start looking pretty similar to the actual league table?
Love a good stat though so looking forward to the update. Reckon 10-15 games in would be a good reference point
Yep, as a quirk of everyone playing each other twice, it should start to tend that way at the halfway point, but once we get to the 30-35 game point, it should start to diverge again and eventually come back together by the end of the season.
Doesn't this work in American sport because the league isn't play everyone once at home and once away?
whilst it may not tell us as much the further into the season we go, this is definitely worth showing the lads to point out the relative difficulty of the starting fixtures and that we are theoretically punching above our weight as things stand
Once we get halfway through the season and everyone has played each other does the SOS stat not become obsolete?
Then presumably the SOV stat will start looking pretty similar to the actual league table?
Love a good stat though so looking forward to the update. Reckon 10-15 games in would be a good reference point
Yep, as a quirk of everyone playing each other twice, it should start to tend that way at the halfway point, but once we get to the 30-35 game point, it should start to diverge again and eventually come back together by the end of the season.
Doesn't this work in American sport because the league isn't play everyone once at home and once away?
They came up with it as a tiebreaker when trying to decide which teams are more deserving of advancing to the playoffs, for example. Because you get a lot of teams with the same W/L records when a season is only 17 games long.
It doesn’t mean it can’t be adapted and be valuable in other sports though.
Once we get halfway through the season and everyone has played each other does the SOS stat not become obsolete?
Then presumably the SOV stat will start looking pretty similar to the actual league table?
Love a good stat though so looking forward to the update. Reckon 10-15 games in would be a good reference point
Yep, as a quirk of everyone playing each other twice, it should start to tend that way at the halfway point, but once we get to the 30-35 game point, it should start to diverge again and eventually come back together by the end of the season.
Doesn't this work in American sport because the league isn't play everyone once at home and once away?
It's not just a case of you don't play everyone in the league, or even your conference; the fixture is compiled based on a rotating formula which is partly driven by the division you're based in and partly by where you finished last year. This means that teams in the same division don't have comparable fixture lists, let alone any other team - this is where the "Strength of Schedule" comes in so that there's a measure of comparability between clubs' schedules. The idea is you play every club in the league at least once over a four-year period.
The poorer-performing teams have easier match-ups (in theory) compared to the top teams which is all part of the NFL's equalisation policy as it's not a healthy league if it's the same two or three clubs winning year after year after year.
A similar thing happens in the Australian Football League: there are 18 teams and the regular season is 22-games long. Each team plays every other team and then there are five other matches to play made up of a combination of regular "double-ups" (for example, the two West Australian sides will play each other twice in a season, as will the two South Australian teams, the two Sydney sides, and the two Queensland clubs) and matches determined by where you finished on the previous season's ladder so that if you finished near the bottom you wouldn't play as many teams from the top echelon twice compared to the middle or bottom sections. Again, it's all about equalisation and trying to keep the league as competitive throughout.
With our football everyone plays everyone else so everyone has the same strength of schedule at the end of the season. Where it would be more informative would be if it your performance-to-date was based on the position of the opponents at the time of playing them.
For example, it's normally the case to have played a team or two twice before you've even played some clubs. When you play those particular sides they could be high up in the league so you could have matches against a top-four side ... but after you've played them twice the arse falls out of their season and they drop away to flirt with relegation. If you look at you results at season's end you think "crikey, we got trounced a couple of times by those muppets and they finished 19th, no wonder we didn't go up as champions" when in reality they could have been the top side when we played them.
Having said that, bugger knows how you'd go about compiling a table based on that!
SOS table (ie who's had the hardest fixtures so far):
SOV table (ie how good are the teams you've got points against):
Plus a bonus SOV minus SOS table (ie who's making it most difficult for themselves by beating good teams but dropping bad performances against bad teams): My personal interpretation of this third table is that the teams placed higher are harder to predict how the rest of their season will unfold, teams lower are more predictable and are likely to have found their natural level after nine games because they're putting together points against teams worse than them and dropping points to teams better than them.
Good stuff. Hopefully we can stay within a respectable points difference to the playoffs going into November as we have a tasty run of 5 fixtures that in all honesty we have a chance of winning all 5. More likely we will manage it then proceed to draw 5 games but until mid november I'm not ruling us out of a top 6 finish despite where we currently find ourselves.
Accy and wycombe with easy starts to their season… obviously
Teams that Accrington have faced have 7 wins, 10 draws and 13 defeats.
Teams that Wycombe have faced have 9 wins, 12 draws and 15 defeats.
By contrast, Charlton have faced teams with 17 wins, 9 draws and 9 defeats.
It's easy to tell from those figures that Charlton have had harder matches so far, the SOS/SOV numbers just quantifies everything with a score between 0 and 1.
So, according to your calculations, we're nailed on for promotion?
What a surprise that we're joint top of the "making life difficult for ourselves" table.
I see it slightly differently. It's the league table of unpredictability. And we're exactly where one might expect them to be. That is, one place lower than where everyone assumes we would be.
I always like a table are top of, or nearly top. But I can't get around the idea that we've had a tough start because we didn't beat teams whereas Ipswich had an easy start because they beat their opponents. I think Ipswich are a better team a.t.m. and that explains that. But thank you for cheering me up, cos I would like to believe....
The strength of schedule is sort of helpful over a short period - as a way to compare form, it adds weight based on opponents' league position (but not form). The strength of victory less so, as it only considers victories - a single total outlier victory against a top side will provide a wholly skewed result, by taking no allowance of the other games against even the weakest opposition. As if we'd recognise that sort of scenario
What a surprise that we're joint top of the "making life difficult for ourselves" table.
I see it slightly differently. It's the league table of unpredictability. And we're exactly where one might expect them to be. That is, one place lower than where everyone assumes we would be.
That's about the most unpredictable thing ever.
Or maybe the most predictable thing ever, for Charlton fans.
Lots of back and forth about how hard/easy we've had it so far this season, so I wanted to build a SOS/SOV picture (based on an NFL/american sports stat) which looks at your opponents win-draw-loss records to build an overall picture of how strong your 'schedule' has been, while taking only the records of the teams you have beaten to create a 'strength of victory' number. The quirk with this stat is that after an entire season, everyone's SOS will be the same. But during a season, it can help to determine who's played the most difficult teams.
For example, with Charlton, our opponents so far have been: Accrington, Derby, Sheff Wed, Plymouth, Cambridge and Wycombe.
These six teams have 17 wins, 9 draws and 9 losses across a total of 35 games played.
To calculate Charlton's SOS: a win is worth 1, a draw is worth half and a loss is worth zero, divided by the overall total of games.
17+(9/2)+(9 times 0) / 35 = 0.614 SOS
To calculate Charlton's SOV, we take only the teams we have beaten (Derby & Plymouth) who have seven wins, two draws and three losses across 12 games.
7+(2/2)+(3 times 0) / 12 = 0.666 SOV
I'm working on building SOS and SOV tables and will post them later. (Now posted below)
I always like a table are top of, or nearly top. But I can't get around the idea that we've had a tough start because we didn't beat teams whereas Ipswich had an easy start because they beat their opponents. I think Ipswich are a better team a.t.m. and that explains that. But thank you for cheering me up, cos I would like to believe....
It's a common criticism of comparing fixture lists in this way - of course winning teams make their games against opponents look easy.
SOS aims to get around that problem by collecting the wins, draws and losses of all nine of, for example, Charlton's opponents. By looking at the W/D/L column of the league table, you can find that the nine teams Charlton have faced have collectively won 33 games, drawn 22 and lost 26 (81 total games from 9 opponents who played 9 games each). This is clearly a better record of results than for example Exeter's opponents who have collectively won 24 games, drawn 21 and lost 36 across 81 games so far.
But giving each team a string of numbers like 33/22/26 or 24/21/36 is weird and hard to compare, so you need to standardize that somehow to put into a neat table or graph.
We do that by taking (in Charlton's example) the 33 wins plus half of the 22 draws and dividing by the total number of games, 81, to pull together a number that represents how good the 9 teams faced are. 44/81 = 0.543. For Exeter, the number is 34.5/81 = 0.426. This is why, once everyone has played each other, everyone's number will converge the middle around Christmas, and then teams will begin to spread through January and February, finally coming together for a second time when everyone has played each other twice by the end of May.
The fun thing with stats and numbers is that you can always find new ways to look at the data and so after thinking a bit more about your post, I've created a SOS-1 table (can't think of a better name) where we discount the results obtained against your team. For example, where Charlton's opponents record was previously 33/22/26 for SOS, removing results achieved against Charlton gives us new inputs of 31/17/24 for SOS-1.
Here is what that SOS-1 table looks like:
Hopefully this satisfies those who dislike the inclusion of direct results and the effect they might be having on the scoring method.
An interesting quirk of SOS-1 is that the ratings will not all converge to exactly 0.500 at the end of the season like they do for SOS. In fact teams with higher SOS-1 values will be higher in the table on average. It's something you expect to happen if you're thinking logically (teams higher in the table tend to win more and lose less) but I had to double check it with real numbers.
By using W/D/L data from last season, I found that SOS-1 ratings do get close to 0.500 andcrucially you can prove that higher SOS-1 values align almost exactly with end of season league position. For example, Wigan's end of season SOS-1 value was 0.509 while Crewe's was 0.489. Charlton in 13th had an SOS-1 value of 0.498.
An interesting quirk of SOS-1 is that the ratings will not all converge to exactly 0.500 at the end of the season like they do for SOS. In fact teams with higher SOS-1 values will be higher in the table on average. It's something you expect to happen if you're thinking logically (teams higher in the table tend to win more and lose less) but I had to double check it with real numbers.
By using W/D/L data from last season, I found that SOS-1 ratings do get close to 0.500 andcrucially you can prove that higher SOS-1 values align almost exactly with end of season league position. For example, Wigan's end of season SOS-1 value was 0.509 while Crewe's was 0.489. Charlton in 13th had an SOS-1 value of 0.498.
Wouldn't the SOV have a meaningful use against the form table?
Your last six games could be the best, or worst, 6 teams but if you had a rolling 6 games SOV you would actually have a good idea how "in form", or not, a team really are.
The SOS-1 table gives a much better indication of our relative fixture difficulty I feel.
To take that further and to a simpler place you could convert the SOS-1 chart to a simple average points per game of teams faced / average points per game of teams yet to face. This would cause some movements to the SOS-1 table as it weights draws as 1 point Vs a share of points.
Good stuff. Hopefully we can stay within a respectable points difference to the playoffs going into November as we have a tasty run of 5 fixtures that in all honesty we have a chance of winning all 5. More likely we will manage it then proceed to draw 5 games but until mid november I'm not ruling us out of a top 6 finish despite where we currently find ourselves.
Don't forget that you are talking about Charlton. Experience has taught me that a run of games where we should win every one and do just doesn't happen.
Comments
One thing: if I understood your initial post correctly, I think the following bit is wrong:
That should be SOV (not SOS). If I am wrong, then I haven't properly understood it at all and you can ignore this post..!
Then presumably the SOV stat will start looking pretty similar to the actual league table?
Love a good stat though so looking forward to the update. Reckon 10-15 games in would be a good reference point
whilst it may not tell us as much the further into the season we go, this is definitely worth showing the lads to point out the relative difficulty of the starting fixtures and that we are theoretically punching above our weight as things stand
It doesn’t mean it can’t be adapted and be valuable in other sports though.
The poorer-performing teams have easier match-ups (in theory) compared to the top teams which is all part of the NFL's equalisation policy as it's not a healthy league if it's the same two or three clubs winning year after year after year.
A similar thing happens in the Australian Football League: there are 18 teams and the regular season is 22-games long. Each team plays every other team and then there are five other matches to play made up of a combination of regular "double-ups" (for example, the two West Australian sides will play each other twice in a season, as will the two South Australian teams, the two Sydney sides, and the two Queensland clubs) and matches determined by where you finished on the previous season's ladder so that if you finished near the bottom you wouldn't play as many teams from the top echelon twice compared to the middle or bottom sections. Again, it's all about equalisation and trying to keep the league as competitive throughout.
With our football everyone plays everyone else so everyone has the same strength of schedule at the end of the season. Where it would be more informative would be if it your performance-to-date was based on the position of the opponents at the time of playing them.
For example, it's normally the case to have played a team or two twice before you've even played some clubs. When you play those particular sides they could be high up in the league so you could have matches against a top-four side ... but after you've played them twice the arse falls out of their season and they drop away to flirt with relegation. If you look at you results at season's end you think "crikey, we got trounced a couple of times by those muppets and they finished 19th, no wonder we didn't go up as champions" when in reality they could have been the top side when we played them.
Having said that, bugger knows how you'd go about compiling a table based on that!
SOS table (ie who's had the hardest fixtures so far):
SOV table (ie how good are the teams you've got points against):
Plus a bonus SOV minus SOS table (ie who's making it most difficult for themselves by beating good teams but dropping bad performances against bad teams):
My personal interpretation of this third table is that the teams placed higher are harder to predict how the rest of their season will unfold, teams lower are more predictable and are likely to have found their natural level after nine games because they're putting together points against teams worse than them and dropping points to teams better than them.
That's about the most unpredictable thing ever.
The strength of victory less so, as it only considers victories - a single total outlier victory against a top side will provide a wholly skewed result, by taking no allowance of the other games against even the weakest opposition. As if we'd recognise that sort of scenario
SOS aims to get around that problem by collecting the wins, draws and losses of all nine of, for example, Charlton's opponents. By looking at the W/D/L column of the league table, you can find that the nine teams Charlton have faced have collectively won 33 games, drawn 22 and lost 26 (81 total games from 9 opponents who played 9 games each). This is clearly a better record of results than for example Exeter's opponents who have collectively won 24 games, drawn 21 and lost 36 across 81 games so far.
But giving each team a string of numbers like 33/22/26 or 24/21/36 is weird and hard to compare, so you need to standardize that somehow to put into a neat table or graph.
We do that by taking (in Charlton's example) the 33 wins plus half of the 22 draws and dividing by the total number of games, 81, to pull together a number that represents how good the 9 teams faced are. 44/81 = 0.543. For Exeter, the number is 34.5/81 = 0.426. This is why, once everyone has played each other, everyone's number will converge the middle around Christmas, and then teams will begin to spread through January and February, finally coming together for a second time when everyone has played each other twice by the end of May.
The fun thing with stats and numbers is that you can always find new ways to look at the data and so after thinking a bit more about your post, I've created a SOS-1 table (can't think of a better name) where we discount the results obtained against your team. For example, where Charlton's opponents record was previously 33/22/26 for SOS, removing results achieved against Charlton gives us new inputs of 31/17/24 for SOS-1.
Here is what that SOS-1 table looks like:
Hopefully this satisfies those who dislike the inclusion of direct results and the effect they might be having on the scoring method.
By using W/D/L data from last season, I found that SOS-1 ratings do get close to 0.500 and crucially you can prove that higher SOS-1 values align almost exactly with end of season league position. For example, Wigan's end of season SOS-1 value was 0.509 while Crewe's was 0.489. Charlton in 13th had an SOS-1 value of 0.498.
Your last six games could be the best, or worst, 6 teams but if you had a rolling 6 games SOV you would actually have a good idea how "in form", or not, a team really are.
To take that further and to a simpler place you could convert the SOS-1 chart to a simple average points per game of teams faced / average points per game of teams yet to face. This would cause some movements to the SOS-1 table as it weights draws as 1 point Vs a share of points.