A Link Between FPTs and VBDs for Running Backs (1 Viewer)

[Aabye]

The thread Jeff started about RBBC got me thinking about the link between how many fantasy points RBs score and how many VBD points they score. Here's my thought process:

(1) Let's assume, for the sake of the argument, that RB fantasy scoring is going down somewhat. This hypothesis has some data behind it. From 2002-2006, the top 24 RBs in the league produced >5000 total fantasy points every season. It hasn't happened since then and in the other 20 seasons from 1985-2010, it's only happened two other times.

(2) VBD theory tells us that it's not raw fantasy points that matter, but rather points over the baseline. So even though a QB score more fantasy points than a TE, the TE might be more valuable for fantasy owners. In other words, we can't judge the fantasy value of a player by simply looking at his FPT scoring.

So the question is: Even if we grant that RB fantasy scoring is trending downward, should we worry that RB VBD points might also be trending downward?

It was my hypothesis that RB VBD scoring would be independent of RB FPT scoring.

To test the hypothesis, I took every season from 1985-2010 (excluding 1987) and calculated the total FPT scored and total VBD points scored by the top 24 RBs during each of those seasons. I graphed the results, which you can find here. I used a baseline of RB24 and compiled the data by using the Historical Data Dominator.

The results surprised me. There appears to be a link between total FPTs scored by RBs and the fantasy value of the RB position overall. As FPT scoring goes up, the VBD value of RBs overall seems to go up as well. As RB FPT scoring goes down, RB VBD value drops as well. So if RB scoring is indeed trending downwards, we might find that RB fantasy value is dropping along with it.

 

[Jeff Pasquino]

The thread Jeff started about RBBC got me thinking about the link between how many fantasy points RBs score and how many VBD points they score. Here's my thought process:

(1) Let's assume, for the sake of the argument, that RB fantasy scoring is going down somewhat. This hypothesis has some data behind it. From 2002-2006, the top 24 RBs in the league produced >5000 total fantasy points every season. It hasn't happened since then and in the other 20 seasons from 1985-2010, it's only happened two other times.

(2) VBD theory tells us that it's not raw fantasy points that matter, but rather points over the baseline. So even though a QB score more fantasy points than a TE, the TE might be more valuable for fantasy owners. In other words, we can't judge the fantasy value of a player by simply looking at his FPT scoring.

So the question is: Even if we grant that RB fantasy scoring is trending downward, should we worry that RB VBD points might also be trending downward?

It was my hypothesis that RB VBD scoring would be independent of RB FPT scoring.

To test the hypothesis, I took every season from 1985-2010 (excluding 1987) and calculated the total FPT scored and total VBD points scored by the top 24 RBs during each of those seasons. I graphed the results, which you can find here. I used a baseline of RB24 and compiled the data by using the Historical Data Dominator.

The results surprised me. There appears to be a link between total FPTs scored by RBs and the fantasy value of the RB position overall. As FPT scoring goes up, the VBD value of RBs overall seems to go up as well. As RB FPT scoring goes down, RB VBD value drops as well. So if RB scoring is indeed trending downwards, we might find that RB fantasy value is dropping along with it.

First of all, I wanted to say that I appreciate your contributions of late Aaybe. Very solid all around, so thanks for that.I think the concern is valid, but it is to be expected. It's just simple math (at least that is my view on it). As fantasy points get bigger, VBD will get bigger, and vice versa.

The simplest way I can explain it is with two examples - but the two share the same trait.

I will say that every RB is 5% better than the previous one. Now, example one will have RB24 = 100 points, and example two will have RB24 = 125 points.

FPTs VBDRB1 307.2 207.2RB2 292.5 192.5RB3 278.6 178.6RB4 265.3 165.3RB5 252.7 152.7RB6 240.7 140.7RB7 229.2 129.2RB8 218.3 118.3RB9 207.9 107.9RB10 198.0 98.0RB11 188.6 88.6RB12 179.6 79.6RB13 171.0 71.0RB14 162.9 62.9RB15 155.1 55.1RB16 147.7 47.7RB17 140.7 40.7RB18 134.0 34.0RB19 127.6 27.6RB20 121.6 21.6RB21 115.8 15.8RB22 110.3 10.3RB23 105.0 5.0RB24 100.0 0.0 4450.2 2050.2

	FPTs	VBDRB1	383.9	258.9RB2	365.7	240.7RB3	348.2	223.2RB4	331.7	206.7RB5	315.9	190.9RB6	300.8	175.8RB7	286.5	161.5RB8	272.9	147.9RB9	259.9	134.9RB10	247.5	122.5RB11	235.7	110.7RB12	224.5	99.5RB13	213.8	88.8RB14	203.6	78.6RB15	193.9	68.9RB16	184.7	59.7RB17	175.9	50.9RB18	167.5	42.5RB19	159.5	34.5RB20	151.9	26.9RB21	144.7	19.7RB22	137.8	12.8RB23	131.3	6.3RB24	125.0	0.0			5562.7	2562.7

With RB24=100, the Total of RB1-RB24 =4450 and the toal VBD = 2050.

With RB24=125, the Total of RB1-RB24 =5563 and the toal VBD = 2563.

The difference is the quantity of FPTs in each 10% step.

 

[Aabye]

First of all, I wanted to say that I appreciate your contributions of late Aaybe. Very solid all around, so thanks for that.

I think the concern is valid, but it is to be expected. It's just simple math (at least that is my view on it). As fantasy points get bigger, VBD will get bigger, and vice versa.

The simplest way I can explain it is with two examples - but the two share the same trait.

With RB24=100, the Total of RB1-RB24 =4450 and the toal VBD = 2050.

With RB24=125, the Total of RB1-RB24 =5563 and the toal VBD = 2563.

The difference is the quantity of FPTs in each 10% step.

Jeff,Thanks. The RBBC stuff is really interesting, and thanks for posting your study.

Here's a little more to explain why I wasn't necessarily expecting what I got:

It's possible that increasing total FPT will increase VBD, but it's not necessary. If each player sees his production increase by a uniform %, then we will see increased VBD scoring along with FPT scoring. But that's not the only way for RBs to increase their FPT scoring overall. In fact, I think it's rather unlikely that it would work like that. Here are a few other possibilities:

(1) There are more feature RBs, which might bunch up the lower end of the RB spectrum. So if the NFL moved away from RBBC, it's possible that RBs 18-24 might see an uptick in their FPT production while RBs 1-8 remain fairly constant. This would increase FPT but decrease VBD, since it would crunch together RB1 and RB24 somewhat.

(2) There are more great backs in the NFL. If there are a few more uber-productive backs, then RBs 1-8 might account for the largest % of the increase and sort of "run away from" the lower-level RBs. This would increase both FPT and VBD since it would spread out RB1 and RB24.

My hypothesis about the current NFL as opposed to the NFL of 2006 is this:

There are fewer great backs and fewer feature backs overall. Both of those reductions should result in a decrease in FPT. But one of those differences should increase VBD and one should decrease it. I guessed that those factors would more or less cancel out, so we would see a decrease in FPT but no change in VBD scoring.

The data doesn't indicate that. I'm not sure how to explain the correlation yet. Still thinking about it.

*note about the idea of uniform increases or decreases in RB production*

I don't think that uniform increases or decreases are the most likely explanation for why we see the correlation that we do.

You and I both looked up total RB touches for the RBBC topic and neither of us found any significant difference from year to year. Additionally, I know of no evidence for the hypothesis that per-touch productivity has changed much either. It seems unlikely that increases or decreases in FPT scoring would be the result of uniform increases or decreases in FPT production since a uniform increase or decrease would probably trigger a difference in either total RB touches or per-touch production, neither of which we've seen.

In other words, if all RBs are scoring more, they are either touching the ball more or doing more with the touches that they're getting. Neither of those things seems to be happening. Similarly, if all RBs are scoring less, they are either touching the ball less or doing less with the touches that they're getting. Neither of those things seems to be happening either. So the fluctuations in RB scoring are more likely the result of some RBs touching the ball less or doing less with their touches, rather than all RBs touching the ball less or doing less with their touches.

 

[Aabye]

Here's the data for WRs from the same time period (1985-2010). Same pattern.

 

[Aabye]

Here's the data for quarterbacks and tight ends.

These two positions show no correlation whatsoever between total FPTs scored and total VBD points scored.

So the question is: why is there a correlation between FPTs and VBDs for RBs and WRs but not for QBs or TEs?

 

[Hoosier16]

Here's the data for quarterbacks and tight ends.

These two positions show no correlation whatsoever between total FPTs scored and total VBD points scored.

So the question is: why is there a correlation between FPTs and VBDs for RBs and WRs but not for QBs or TEs?

Maybe it's a sample size issue. For TEs and QBs, you are likely using a baseline of 12, meaning the top 2 or 3 players will have an overall greater effect on the total VBDs. For example, Brady had over 30% of the VBDs in his monster year, while Chris Johnson was responsible for just over 15% of the VBDs in his big year.

 

[Aabye]

Here's a little bit more: LINK

If we look at all of the groups over that time, there's no apparent correlation between one group's VBD scoring and any other group's VBD scoring. In other words, it's not as if when RB scoring drops off, WR scoring picks up the slack. They seem to be completely independent.

What the hell is going on here?

 

[Aabye]

Here's the data for quarterbacks and tight ends.

These two positions show no correlation whatsoever between total FPTs scored and total VBD points scored.

So the question is: why is there a correlation between FPTs and VBDs for RBs and WRs but not for QBs or TEs?

Maybe it's a sample size issue. For TEs and QBs, you are likely using a baseline of 12, meaning the top 2 or 3 players will have an overall greater effect on the total VBDs. For example, Brady had over 30% of the VBDs in his monster year, while Chris Johnson was responsible for just over 15% of the VBDs in his big year.

Yeah, that's a good suggestion. I'm monkeying around with the data now, trying to make some sense of where this discrepancy might be coming from.So far, I've just looked at RBs. I'm seeing:

- a strong connection between overall VBD scoring and (1) how well the 5th best RB scored and (2) how well the 24th best RB scored

- a weaker connection between overall VBD scoring and how well the 12th best RB scored

- no connection between overall VBD scoring and how well the #1 RB scored

I have no idea why I'm seeing this though. Very weird. I'll see how the other positions work to see if I see the same pattern.

 

[CalBear]

Here's a little bit more: LINK

If we look at all of the groups over that time, there's no apparent correlation between one group's VBD scoring and any other group's VBD scoring. In other words, it's not as if when RB scoring drops off, WR scoring picks up the slack. They seem to be completely independent.

What the hell is going on here?

The positions should be independent. VBD is a measure of how many more points the above-baseline players at a position score, relative to the baseline player at that position. So in years like 1995 where there are a number of WRs going nuts, WR VBD numbers are going to be off the charts (as you can see in your graphs). But there's no particular reason why RB VBD should be significantly affected by that; RB VBD is measured relative to other RBs, not to WRs. Now, WR value was higher relative to RBs in 1995; you can measure a position's value by its proportional contribution to total VBD points, and that measurement is dependent on relative performance.

 

[Aabye]

Here's a little bit more: LINK

If we look at all of the groups over that time, there's no apparent correlation between one group's VBD scoring and any other group's VBD scoring. In other words, it's not as if when RB scoring drops off, WR scoring picks up the slack. They seem to be completely independent.

What the hell is going on here?

The positions should be independent. VBD is a measure of how many more points the above-baseline players at a position score, relative to the baseline player at that position. So in years like 1995 where there are a number of WRs going nuts, WR VBD numbers are going to be off the charts (as you can see in your graphs). But there's no particular reason why RB VBD should be significantly affected by that; RB VBD is measured relative to other RBs, not to WRs. Now, WR value was higher relative to RBs in 1995; you can measure a position's value by its proportional contribution to total VBD points, and that measurement is dependent on relative performance.

Right. That's what I was thinking. But the link between FPT and VBD scoring for RBs and WRs threw a monkey wrench into that independence hypothesis. My original hypothesis was that FPT and VBD scoring should be independent. That is born out in the case of QBs and TEs but not for RBs and WRs.So I thought that if FPT scoring and VBD scoring are linked for RBs and WRs, maybe they also affect one another. The intuition is simply "if WRs are gaining more yards, scoring more TDs and scoring more FPT and VBD, maybe RBs are declining somewhat to offset this. It turns out that's not the case.

My problem isn't that any particular one of these findings is weird. Rather, it's that I'm having trouble figuring out a coherent story for why FPT and VBD are sometimes correlated and sometimes aren't.

 

[CalBear]

Here's a little bit more: LINK

If we look at all of the groups over that time, there's no apparent correlation between one group's VBD scoring and any other group's VBD scoring. In other words, it's not as if when RB scoring drops off, WR scoring picks up the slack. They seem to be completely independent.

What the hell is going on here?

The positions should be independent. VBD is a measure of how many more points the above-baseline players at a position score, relative to the baseline player at that position. So in years like 1995 where there are a number of WRs going nuts, WR VBD numbers are going to be off the charts (as you can see in your graphs). But there's no particular reason why RB VBD should be significantly affected by that; RB VBD is measured relative to other RBs, not to WRs. Now, WR value was higher relative to RBs in 1995; you can measure a position's value by its proportional contribution to total VBD points, and that measurement is dependent on relative performance.

Right. That's what I was thinking. But the link between FPT and VBD scoring for RBs and WRs threw a monkey wrench into that independence hypothesis. My original hypothesis was that FPT and VBD scoring should be independent. That is born out in the case of QBs and TEs but not for RBs and WRs.So I thought that if FPT scoring and VBD scoring are linked for RBs and WRs, maybe they also affect one another. The intuition is simply "if WRs are gaining more yards, scoring more TDs and scoring more FPT and VBD, maybe RBs are declining somewhat to offset this. It turns out that's not the case.

My problem isn't that any particular one of these findings is weird. Rather, it's that I'm having trouble figuring out a coherent story for why FPT and VBD are sometimes correlated and sometimes aren't.

FPT and VBD are definitely correlated in the large sense. If you double the value of yards and TDs, both FPT and VBD will rise. If you increase the value of passing TDs, both FPT and VBD for QBs will rise. There are edge cases where this might not be true--basically, when the baseline player at the position gains as much as the rest of the field. In the typical case, the baseline player won't benefit as much, because the VBD curve isn't linear. Although it tends to be close to linear for QB1-12, which might be one reason why your QBs aren't behaving the same way.You are also getting a conflating factor by using each year's baseline player instead of a Mendoza Line; the number of points represented by VBD changes from year to year. If you take the average points scored by the baseline player at a position over a 10-year period, and use that as the VBD baseline instead of the actual points scored that year, you may get more comprehensible results.

 

[Aabye]

FPT and VBD are definitely correlated in the large sense. If you double the value of yards and TDs, both FPT and VBD will rise. If you increase the value of passing TDs, both FPT and VBD for QBs will rise. There are edge cases where this might not be true--basically, when the baseline player at the position gains as much as the rest of the field. In the typical case, the baseline player won't benefit as much, because the VBD curve isn't linear. Although it tends to be close to linear for QB1-12, which might be one reason why your QBs aren't behaving the same way.

Yeah, I think this is the point that Jeff was making as well. While true, what I'm saying is that even if I don't change the value of yards or TDs, I'm still getting a correlation between yearly FTP scoring and VBD scoring for both RBs and WRs. It's not due to uniform expansion or contraction. It's due to something else.I think that people are viewing this correlation as obvious because of the uniform expansion example, which is misleading. Uniform expansion is almost certainly irrelevant here. While everyone seems to see this link as intuitively obvious, I was pretty surprised to find it. I don't see it at all for either QBs or TEs and I can't explain where it comes from for RBs or WRs.Hoosier16 suggested that maybe the discrepancy was due to a smaller data set for both QBs and TEs. So the idea is that the highest scoring QB has a much larger effect on overall yearly VBD than the highest scoring RB. So I just checked to see if that might explain it. It turns out that it doesn't. There's no good correlation between how QB1 scores and how high overall QB VBD is in a given year. I ran the same test for TEs and found a weak correlation between TE1 and overall yearly TE VBD, but it's not a great indicator. The question is why I'm seeing this correlation for RBs and WRs but not for QBs or TEs. I can't figure out a sensible answer and I've now run through a number of possible explanations, none of which has been satisfactory.

You are also getting a conflating factor by using each year's baseline player instead of a Mendoza Line; the number of points represented by VBD changes from year to year. If you take the average points scored by the baseline player at a position over a 10-year period, and use that as the VBD baseline instead of the actual points scored that year, you may get more comprehensible results.

I'm not sure I understand this.

 

[CalBear]

You are also getting a conflating factor by using each year's baseline player instead of a Mendoza Line; the number of points represented by VBD changes from year to year. If you take the average points scored by the baseline player at a position over a 10-year period, and use that as the VBD baseline instead of the actual points scored that year, you may get more comprehensible results.

I'm not sure I understand this.

You're using a moving baseline to calculate your VBD numbers. Here are the last few years for RBs:2010: Blount (138 points)2009: Turner (151 points)2008: Stewart (148 points)2007: Bush (136 points)2006: Dunn (161 points)2005: Foster (143 points)So your baseline can move by 10-15% per year. If you use the average of these (146) as the baseline, instead of moving the baseline every year, you'll have better trend comparisons.As per discussion in the other thread, I'm not sure that worst-starter is the correct baseline for your exercise. Certainly the worst starter has a lot of value compared to a replacement-level player (the Mendoza line).

 

[Aabye]

You are also getting a conflating factor by using each year's baseline player instead of a Mendoza Line; the number of points represented by VBD changes from year to year. If you take the average points scored by the baseline player at a position over a 10-year period, and use that as the VBD baseline instead of the actual points scored that year, you may get more comprehensible results.

I'm not sure I understand this.

You're using a moving baseline to calculate your VBD numbers. Here are the last few years for RBs:2010: Blount (138 points)2009: Turner (151 points)2008: Stewart (148 points)2007: Bush (136 points)2006: Dunn (161 points)2005: Foster (143 points)So your baseline can move by 10-15% per year. If you use the average of these (146) as the baseline, instead of moving the baseline every year, you'll have better trend comparisons.As per discussion in the other thread, I'm not sure that worst-starter is the correct baseline for your exercise. Certainly the worst starter has a lot of value compared to a replacement-level player (the Mendoza line).

I'm still not sure I get it. I'm looking at yearly correlations between FPTs and VBDs. How is standardizing VBD baselines going to help?Here's something else strange:When the baseline is higher, VBD scoring goes higher too. That's extremely weird. Shouldn't a higher baseline value drive VBD values down? It seems like it should but just the opposite happens. So if I switch to a standardized baseline, won't that just make the yearly VBD fluctuations even bigger?As for using a suboptimal baseline, you may be right. I don't really know anything about the strengths and weaknesses of using different baselines. I'll head over to the RBBC thread and take a look if there's discussion of different baselines there.

 

[CalBear]

I'm still not sure I get it. I'm looking at yearly correlations between FPTs and VBDs. How is standardizing VBD baselines going to help?

It will remove the variability in VBD based on the performance of the tail end of the population. Basically, the fact that RB24 scored 161 points in 2006 and 136 points in 2007 is noise in your results; Warrick Dunn in 2006 was more valuable than Reggie Bush in 2007, despite the fact that they both ended up as RB24. Standardizing the baseline would give Dunn credit for being more valuable, and would raise the VBD totals for 2006 relative to 2007, which is probably the right thing to do.

Here's something else strange:When the baseline is higher, VBD scoring goes higher too. That's extremely weird. Shouldn't a higher baseline value drive VBD values down? It seems like it should but just the opposite happens. So if I switch to a standardized baseline, won't that just make the yearly VBD fluctuations even bigger?

You may be running into sample size issues here. 2006 had a high baseline (highest of the ones I looked at), and it also had Tomlinson's 31-TD season. The two aren't related, but when you stick a 431-point RB season into your spreadsheet, it's going to make it look like seasons with high baselines also have high VBD totals.

 

[Aabye]

You may be running into sample size issues here. 2006 had a high baseline (highest of the ones I looked at), and it also had Tomlinson's 31-TD season. The two aren't related, but when you stick a 431-point RB season into your spreadsheet, it's going to make it look like seasons with high baselines also have high VBD totals.

I did check yearly VBD numbers against other possible indicators, including top scoring RBs. There isn't any good correlation there at all. Years with a great RB season aren't more likely to be high-VBD years than years without a great RB season. For instance, 2006 wasn't a particularly high-VBD season, despite Tomlinson's record-setting campaign. RB1 FPT scores turn out to have pretty much nothing at all to do with how many total VBD points are scored by RBs in a given season.The baseline is a pretty weak positive correlation but the mere fact that it's positively correlated is what's weird. It's not one outlier season that's producing this effect. It's 25 seasons. It's actually a better correlation than the top RB scorer. So far, the best correlation I have found, by far, is how well the 5th best RB did. If the 5th best RB had a historically high FPT number, then that year was extremely likely to have a high total VBD value. I have no idea why. Could be significant, could be noise.

 

[Aabye]

I'm still not sure I get it. I'm looking at yearly correlations between FPTs and VBDs. How is standardizing VBD baselines going to help?

It will remove the variability in VBD based on the performance of the tail end of the population. Basically, the fact that RB24 scored 161 points in 2006 and 136 points in 2007 is noise in your results; Warrick Dunn in 2006 was more valuable than Reggie Bush in 2007, despite the fact that they both ended up as RB24. Standardizing the baseline would give Dunn credit for being more valuable, and would raise the VBD totals for 2006 relative to 2007, which is probably the right thing to do.

Apologies for being dense here. Perhaps just this will help me figure out why I should use standardized VBD baselines:Why is 2006 Warrick Dunn more valuable than 2007 Reggie Bush?

 

[CalBear]

I'm still not sure I get it. I'm looking at yearly correlations between FPTs and VBDs. How is standardizing VBD baselines going to help?

It will remove the variability in VBD based on the performance of the tail end of the population. Basically, the fact that RB24 scored 161 points in 2006 and 136 points in 2007 is noise in your results; Warrick Dunn in 2006 was more valuable than Reggie Bush in 2007, despite the fact that they both ended up as RB24. Standardizing the baseline would give Dunn credit for being more valuable, and would raise the VBD totals for 2006 relative to 2007, which is probably the right thing to do.

Apologies for being dense here. Perhaps just this will help me figure out why I should use standardized VBD baselines:Why is 2006 Warrick Dunn more valuable than 2007 Reggie Bush?

The short answer is, because he scored 25 more points. Assuming the cost of each was similar, Warrick Dunn provided more value.Another way to look at it is point differential with the rest of the population. The 23 RBs ahead of Dunn combined for 1377 VBD points relative to him--that is, if you subtract Dunn's total from theirs, the aggregated total is 1377. If you continue that analysis for the RBs behind Dunn--subtracting their totals from his, call it NVBD--you have to go down to RB51 for Dunn to break even (1404 NVBD points). Whereas Bush, even though the players ahead of him only scored 1255 VBD points relative to him, has to go down to RB58 to exceed 1255 NVBD points.Or to put it in lay terms: Dunn provided more value against whomever you might have replaced him with than Bush did. That's why you probably want to look at using a lower baseline.

 

[Aabye]

The short answer is, because he scored 25 more points. Assuming the cost of each was similar, Warrick Dunn provided more value.

Another way to look at it is point differential with the rest of the population. The 23 RBs ahead of Dunn combined for 1377 VBD points relative to him--that is, if you subtract Dunn's total from theirs, the aggregated total is 1377. If you continue that analysis for the RBs behind Dunn--subtracting their totals from his, call it NVBD--you have to go down to RB51 for Dunn to break even (1404 NVBD points). Whereas Bush, even though the players ahead of him only scored 1255 VBD points relative to him, has to go down to RB58 to exceed 1255 NVBD points.

Or to put it in lay terms: Dunn provided more value against whomever you might have replaced him with than Bush did. That's why you probably want to look at using a lower baseline.

I still don't know why you're recommending this but it's probably best to just do it and see what the results are.So here's what I did:

I edited the RB Document to add a new column called "AVBD." AVBD calculates VBD by comparing the FPTs scored during a given year to a new baseline (ABASE) that is calculated by averaging the yearly baselines from the year in question and the two closest seasons in either direction. So, for instance, the baseline for 2006 RBs is the average of the FPTs scored by the 24th RB from 2004, 2005, 2006, 2007, and 2008. As you said, this does reduce yearly baseline variance:

YEAR BASE ABASE2010 141 146.42009 150.2 143.752008 148 147.22007 135.8 147.622006 161 149.062005 143.1 147.862004 157.4 151.162003 142 145.962002 152.3 149.162001 135 143.682000 159.1 140.761999 130 137.181998 127.4 135.281997 134.4 131.461996 125.5 130.961995 140 127.31994 127.5 122.381993 109.1 121.021992 109.8 117.981991 118.7 116.521990 124.8 122.661989 120.2 128.281988 139.8 134.661986 137.9 134.761985 150.6 140.74The AVBD doesn't change much, at least in terms of the correlation that I'm seeing between FPT and VBD. It just strengthens the correlation further. This should be fairly clear, since the more I reduce the variance of ("flatten") the baseline, the more VBD is going to just match up with FPT.If it's not obvious, consider the limiting case in which the baseline is constant for every year. Here, the correlation between VBD and FPT will be perfect.

My goal here isn't to figure out the "correct" value of RBs. I just want to explain why this correlation between FPTs and VBDs exists for RBs and WRs but not for QBs or TEs.

One suggestion was that because QBs and TEs have only 12 scorers as compared to 24 RBs and 30 WRs, outliers might become more of a problem. So I checked to see if variance among QB1s might be correlated with QB VBD overall. It wasn't. Similarly, the highest scoring TE had no effect on the total number of TE VBD points scored during a given year. So that suggestion, while reasonable, doesn't seem to answer the question.

I'm looking at other indicators but so far, I can't figure it out.

 

[Aabye]

I tried to figure out what sorts of things might correlate to overall fantasy value for a position (when fantasy value goes up, what else goes up?).

For RBs and WRs, overall fantasy scoring is positively correlated with VBD scoring, but I wanted to go a little deeper. Which RBs were responsible for the increases or decreases? Here were a few possibilities that I looked at:

(1) Maybe years with a great RB season (e.g. Tomlinson's 2006 season) have higher VBD scoring.

(2) Maybe years with a good middle RB (RB12) have higher VBD scoring.

(3) Maybe years in which the top 25% of RBs are a lot better than the other RBs have higher VBD scoring.

(4) Maybe years in which the middle RBs (25th-75th%) are very good have higher VBD scoring.

(5) Maybe years in which the baseline is lower (RB24 is worse) have higher VBD scoring.

So I checked each one of these over the last 25 seasons. Here are the results.

(1) RB1 scoring is not correlated with overall VBD scoring. I checked RB1's FPTs against yearly VBD values and found no correlation.

(2) RB12 scoring is not correlated with overall VBD scoring.

(3) I checked the total VBD% that RBs 1-6 contributed to overall RB scoring. If that % is higher, that means that RBs 1-6 are pulling away from the rest of the pack. It turns out that total VBD% for RBs 1-6 is not correlated with overall VBD scoring.

(4) I checked the total VBD% for RBs 7-18 (the middle 12 RBs). Again, there is no correlation.

(5) I checked baseline values (RB24 FPTs in a given year) against overall VBD scoring and found that (strangely) as the baseline increased, VBD scoring increased as well. The correlation is weak-ish but it's noteworthy that it's a positive correlation rather than a negative correlation.

Finally, I found something that was really astonishing: The FPTs that RB5 scored in a given year were strongly correlated to the fantasy value of the position as a whole.

What this means is that if you chart the total FPTs scored by the 5th RB, you get a curve that looks almost identical to the curve that you get when you chart the total VBD value of all RBs.

Since it worked for RBs, I checked it against WRs too and got a similar correlation. I then tried it for QB3 and TE3. TE3 didn't correlate particularly well but the correlation for QB3 was absolutely amazing, especially since overall QB FPT and VBD scoring aren't correlated at all.

Here's the CHART.

If anyone has a suggestion as to why I might be finding this correlation, let me know. It's across 3 positions for 25 seasons, so it's probably not an accident. Keep in mind that when I checked other particular RBs, I got no correlation at all. And when I checked RBs 1-6 inclusive, I also got no correlation. So why is RB5 such a good indicator?

A shiny gold star to anyone who can figure this out. It's got me totally stumped.

 

[Hoosier16]

Hoosier16 suggested that maybe the discrepancy was due to a smaller data set for both QBs and TEs. So the idea is that the highest scoring QB has a much larger effect on overall yearly VBD than the highest scoring RB. So I just checked to see if that might explain it. It turns out that it doesn't. There's no good correlation between how QB1 scores and how high overall QB VBD is in a given year. I ran the same test for TEs and found a weak correlation between TE1 and overall yearly TE VBD, but it's not a great indicator.

I was thinking about testing it in a little different way but your latest (RB5, QB3) may be doing something similar. Instead of testing the correlation between the top player at each position and the yearly VDB, I'd remove the top player at each position from the data. Then, apply your original testing. This should remove the "Brady effect" (usually only one strong outlier per year).

 

[GregR_2]

One thing that might help you get a better feel for the data you're working with would be to graph each year's positional curve and compare them, rather than rolling everything up to one number.

I'd graph player fantasy point vs their rank in position for each year, and then overlay a few of them to see how they differ from one season to the next. Then do the same for VBD values. You may notice a trend (comparatively flatter scoring curves lead to X, more exponential curves lead to Y), or you might notice things like where that 5th RB who is the best predictor lies on the curve and why that might be more reflective of the result (is he at the end of the linear portion of the curve and so is a good predictor of the bottom portion that might drive the overall result?). You also might notice general differences between QB and TE versus RB and WR that might shed some light.

 

[Aabye]

One thing that might help you get a better feel for the data you're working with would be to graph each year's positional curve and compare them, rather than rolling everything up to one number.I'd graph player fantasy point vs their rank in position for each year, and then overlay a few of them to see how they differ from one season to the next. Then do the same for VBD values. You may notice a trend (comparatively flatter scoring curves lead to X, more exponential curves lead to Y), or you might notice things like where that 5th RB who is the best predictor lies on the curve and why that might be more reflective of the result (is he at the end of the linear portion of the curve and so is a good predictor of the bottom portion that might drive the overall result?). You also might notice general differences between QB and TE versus RB and WR that might shed some light.

Good idea. Will do.

 

[ZWK]

Here's one idea. Looking at how fantasy points scored changes over time, you can break that down into two things: a long term trend and year-to-year variation. At RB, the trend is that fpts fell from 1985 to the early 90s, then rose till the early 2000s, and then have been falling again. That reflects meaningful change in how the position is used. But there are also spikes and dips from year to year - there's a big spike in RB fpts in 2000, then a dip down in 2001, then a spike back up in 2002. Those spikes and dips are basically just random variation - in this case a lot of RBs happened to have big years in 2000 & 2002, but not so many did in 2001.

The two different causes of fantasy points could have different effects on VBD. With random year-to-year variation, I'd expect fpts and VBD to be correlated. A spike in fpts generally comes from a few players having big years, not from a whole lot of players having slightly better than usual years (the law of large numbers applies), so it should involve a spike in VBD. But with the trend over time, the relationship is not so clear. There could be an upward trend over time in fpts because the elite players are doing even better (which would produce a positive correlation with VBD), or because the rest of the pack is catching up to the elite players (which would produce a negative correlation with VBD).

How could we look at this in the data? My first thought was to look at the correlations with the year, but that would only work if there was a steady trend in fpts (either up or down) over the whole time period, and that's not what happened (at least not at RB). My second idea was to try to smooth out the trend lines. I did that in a simple way: just take the five-year average. And that worked. You can see on this graph, which has RB fpts in blue, that the 5-year average (in black) captures the trend over time without all the jittery ups & downs from year-to-year variation.

RB VBD is on the same graph in red, also with a smoothed out version showing 5-year averages. We can compare RB fpts & VBD in two ways using this graph. First, we can compare trend lines: do the two trend lines show the same pattern? And you can see that they do - both black trend lines go up until the early 2000s and then start declining. The VBD changes more gradually, but it's a similar pattern. Second, we can compare the deviations from the trend lines: in years when the fpts spike up above the trend line, or dip down below it, does VBD also spike up or dip below its trend line? And the answer there, too, is yes. Most random good years for RB fpts (up above the trend line) are also good years for RB vbd (up above that trend line), although this doesn't hold every single time.

And, instead of just eyeballing the graph, we can show these relationships with numbers. The correlation between the fpts trend line and the VBD trend line is .79, which confirms that the two trend lines have the same shape. For each year, we can also calculate a deviation in fpts (that year's fpts minus the trend line's value for that year) and a deviation for VBD, and the correlation between those two numbers is .63. That confirms that the year-to-year variation in VBD matches the year-to-year variation in fpts.

What about the other positions? This is where it gets interesting. Here are the correlations in the deviation for the 4 positions (reflecting year-to-year variation):

RB .63

WR .60

TE .61

QB .69

All basically the same. An unusually good year for fpts (better than the surrounding years) is also an unusually good year for VBD, at every position. But what about the correlations between the trend lines:

RB .79

WR .88

TE .44

QB -.44

That's where the difference shows up. At RB & WR the the trends match up, but at QB they go in opposite directions. TE is in between, but closer to RB & WR. Here's the graph for QB - you can see that the deviations still match up (years above the trend line on fpts are also generally above on VBD), but the trend lines diverge (especially from 1996-2003, when fpts shoot up & VBD drops). I think what's happening is that at RB & WR, trends towards heavier usage influence the players at the top more than anyone else (it's the elite RBs who become workhorses), but at QB there has been a more dramatic leaguewide change towards passing. It used to be that a few elite passing teams were throwing a lot, but now everyone's doing it and there are lots of good passing teams, so the catch-up by the rest of the league is happening faster than the elite few can pull away. TE is a mix of elite improvement and a trend towards more widespread usage of TEs as receivers.

It would be interesting to see this type of analysis for other statistics besides players' fantasy points, like team rushing fantasy points or team passing attempts. That might help clarify what the relevant difference is between the positions.

 

[Jeff Pasquino]

One thing that might help you get a better feel for the data you're working with would be to graph each year's positional curve and compare them, rather than rolling everything up to one number.I'd graph player fantasy point vs their rank in position for each year, and then overlay a few of them to see how they differ from one season to the next. Then do the same for VBD values. You may notice a trend (comparatively flatter scoring curves lead to X, more exponential curves lead to Y), or you might notice things like where that 5th RB who is the best predictor lies on the curve and why that might be more reflective of the result (is he at the end of the linear portion of the curve and so is a good predictor of the bottom portion that might drive the overall result?). You also might notice general differences between QB and TE versus RB and WR that might shed some light.

Good idea. Will do.

GregR knows things. Great to see him in this thread.I'd also send a message to Doug Drinen and even Chase Stuart - those guys are deep into these sorts of analyses.ETA - Never mind, they may be more likely to visit here if I let them know.

 

[Aabye]

One thing that might help you get a better feel for the data you're working with would be to graph each year's positional curve and compare them, rather than rolling everything up to one number.I'd graph player fantasy point vs their rank in position for each year, and then overlay a few of them to see how they differ from one season to the next. Then do the same for VBD values. You may notice a trend (comparatively flatter scoring curves lead to X, more exponential curves lead to Y), or you might notice things like where that 5th RB who is the best predictor lies on the curve and why that might be more reflective of the result (is he at the end of the linear portion of the curve and so is a good predictor of the bottom portion that might drive the overall result?). You also might notice general differences between QB and TE versus RB and WR that might shed some light.

Good idea. Will do.

GregR knows things. Great to see him in this thread.I'd also send a message to Doug Drinen and even Chase Stuart - those guys are deep into these sorts of analyses.ETA - Never mind, they may be more likely to visit here if I let them know.

Yeah, this place is loaded with really smart fellas. I'm don't know much when it comes to statistics, so I'm sorta fumbling around. I'm confident that others can do a much better job of explaining what is going on. I'm consistently impressed with the level of discourse that you can find in the Shark Pool.

 

[Aabye]

Here's one idea. Looking at how fantasy points scored changes over time, you can break that down into two things: a long term trend and year-to-year variation. At RB, the trend is that fpts fell from 1985 to the early 90s, then rose till the early 2000s, and then have been falling again. That reflects meaningful change in how the position is used. But there are also spikes and dips from year to year - there's a big spike in RB fpts in 2000, then a dip down in 2001, then a spike back up in 2002. Those spikes and dips are basically just random variation - in this case a lot of RBs happened to have big years in 2000 & 2002, but not so many did in 2001.

The two different causes of fantasy points could have different effects on VBD. With random year-to-year variation, I'd expect fpts and VBD to be correlated. A spike in fpts generally comes from a few players having big years, not from a whole lot of players having slightly better than usual years (the law of large numbers applies), so it should involve a spike in VBD. But with the trend over time, the relationship is not so clear. There could be an upward trend over time in fpts because the elite players are doing even better (which would produce a positive correlation with VBD), or because the rest of the pack is catching up to the elite players (which would produce a negative correlation with VBD).

How could we look at this in the data? My first thought was to look at the correlations with the year, but that would only work if there was a steady trend in fpts (either up or down) over the whole time period, and that's not what happened (at least not at RB). My second idea was to try to smooth out the trend lines. I did that in a simple way: just take the five-year average. And that worked. You can see on this graph, which has RB fpts in blue, that the 5-year average (in black) captures the trend over time without all the jittery ups & downs from year-to-year variation.

RB VBD is on the same graph in red, also with a smoothed out version showing 5-year averages. We can compare RB fpts & VBD in two ways using this graph. First, we can compare trend lines: do the two trend lines show the same pattern? And you can see that they do - both black trend lines go up until the early 2000s and then start declining. The VBD changes more gradually, but it's a similar pattern. Second, we can compare the deviations from the trend lines: in years when the fpts spike up above the trend line, or dip down below it, does VBD also spike up or dip below its trend line? And the answer there, too, is yes. Most random good years for RB fpts (up above the trend line) are also good years for RB vbd (up above that trend line), although this doesn't hold every single time.

And, instead of just eyeballing the graph, we can show these relationships with numbers. The correlation between the fpts trend line and the VBD trend line is .79, which confirms that the two trend lines have the same shape. For each year, we can also calculate a deviation in fpts (that year's fpts minus the trend line's value for that year) and a deviation for VBD, and the correlation between those two numbers is .63. That confirms that the year-to-year variation in VBD matches the year-to-year variation in fpts.

What about the other positions? This is where it gets interesting. Here are the correlations in the deviation for the 4 positions (reflecting year-to-year variation):

RB .63

WR .60

TE .61

QB .69

All basically the same. An unusually good year for fpts (better than the surrounding years) is also an unusually good year for VBD, at every position. But what about the correlations between the trend lines:

RB .79

WR .88

TE .44

QB -.44

That's where the difference shows up. At RB & WR the the trends match up, but at QB they go in opposite directions. TE is in between, but closer to RB & WR. Here's the graph for QB - you can see that the deviations still match up (years above the trend line on fpts are also generally above on VBD), but the trend lines diverge (especially from 1996-2003, when fpts shoot up & VBD drops). I think what's happening is that at RB & WR, trends towards heavier usage influence the players at the top more than anyone else (it's the elite RBs who become workhorses), but at QB there has been a more dramatic leaguewide change towards passing. It used to be that a few elite passing teams were throwing a lot, but now everyone's doing it and there are lots of good passing teams, so the catch-up by the rest of the league is happening faster than the elite few can pull away. TE is a mix of elite improvement and a trend towards more widespread usage of TEs as receivers.

It would be interesting to see this type of analysis for other statistics besides players' fantasy points, like team rushing fantasy points or team passing attempts. That might help clarify what the relevant difference is between the positions.

See, this is just fantastic work. Thanks ZWK.

 

[ZWK]

I've put everything I used for that post online:

Data spreadsheet

RB graph

WR graph

TE graph

QB graph

The graphs have fantasy points in blue, VBD in red, trend lines (5-year average) in black.

Also, one last thing we can do is to see how sensitive VBD is to changes in fpts. You can see on the graphs that fpts vary a lot more than VBD - when the fpts trend line goes up by a lot, the VBD trend line only goes up by a little. How strong is that relationship? A linear regression (using the smoothed data - that is, the 5-year averages) shows that each 1 fpt increase in the scoring trend at that position is associated with a:

.30 VBD increase for RBs

.27 VBD increase for WRs

.09 VBD increase for TEs

-.21 VBD drop for QBs

 

[Aabye]

I've put everything I used for that post online:

Data spreadsheet

RB graph

WR graph

TE graph

QB graph

The graphs have fantasy points in blue, VBD in red, trend lines (5-year average) in black.

Also, one last thing we can do is to see how sensitive VBD is to changes in fpts. You can see on the graphs that fpts vary a lot more than VBD - when the fpts trend line goes up by a lot, the VBD trend line only goes up by a little. How strong is that relationship? A linear regression (using the smoothed data - that is, the 5-year averages) shows that each 1 fpt increase in the scoring trend at that position is associated with a:

.30 VBD increase for RBs

.27 VBD increase for WRs

.09 VBD increase for TEs

-.21 VBD drop for QBs

More great stuff. Thanks for all of your work on this, ZWK.It looks like you copy/pasted your data from my documents. I am aware of a couple of errors when I was putting it all together:

1997 RBs: FPT should be 4546.1, not 4413.9. VBD should be 1373.3, not 1322.7

1993 RBs: FPT should be 3862.6, not 3971.7. VBD should be 1174.6, not 1244.4.

Minor glitches resulting from sloppy copy/pasting. Sorry about that. Shouldn't mess with the numbers too much. I've double checked all of the other numbers for RBs, WRs, QBs, and TEs.

***

So you suggested a possible explanation for the strong correlation at RB and WR: Overall FPT and VBD are driven up by the top-tier players increasing their production. At QB, FPTs are driven up by increased scoring by lower-level QBs. TEs are a mix, so we get noisy results.

I did the following for RBs, and I'll do it for the other positions as well:

I chunked up yearly VBD scoring in the following way:

(1) percentage of VBD points scored by the top 6 RBs (RB1-RB6)

(2) percentage of VBD points scored by the next 6 RBs (RB7-RB12)

(3) percentage of VBD points scored by the next 6 RBs (RB13-RB18)

(4) percentage of VBD points scored by the last 6 RBs (RB19-RB24)

I then graphed those values and checked for correlation with overall VBD and found nothing (CHART). It looks like higher FPTs and VBDs are not generated by elite backs running away from the field. That correlation is slightly negative. There's a weak positive correlation with VBD and % scored by RBs 7-12. Just for fun I checked RB1-12 against total VBD and there is no correlation there. I checked correlations against FPTs too, but those are (predictably) nonexistent.

So if I'm interpreting this correctly, it looks like RB FPT and VBD increases are not driven by increased production by top-tier RBs, defined as RBs 1-6.

**EDIT**

Updated the spreadsheet with tabs for WR, QB, and TE. WRs are divided into 1-6, 7-12, 13-18, 19-24, and 25-30. QBs and TEs are divided into 1-3, 4-6, 7-9, 10-12. I did not find a strong correlation between any group and position VBD scoring or FPT scoring.

So it looks like the suggestion that RB and WR VBD increases are driven by increased relative production by the top-tier producers is incorrect.

 

[Chase Stuart]

I'm not sure I totally understand the question. There should be a link between FP and VBD values for RBs, as there should be for any non-linear data. Running back scoring is highly correlated with touches, even moreso than QB scoring. http://www.pro-football-reference.com/blog/?p=1133

That said, there's a key distinction between rushing and passing. Allow me to make some gross generalizations. When you're a good passing team (measured by ANY/A), you probably won't need to pass that often. You'll score points and you'll be winning, so you'll stop throwing. If you're a bad passing team, you'll throw some INTs, have a bunch of three and outs, and are likely to have to pass a bunch in the fourth quarter. That's exactly why raw passing yards is a bad way to measure passing production -- bad passing teams are likely to have a bunch of passing yards late in the game and good passing teams aren't going to accumulate many passing yards at the ends of games. That means passing yards doesn't accurately measure passing ability, since it penalizes good passing teams and helps bad passing teams.The opposite happens with rushing. If you're a great rushing team, you'll keep rushing. If you're a bad rushing team, you'll stop rushing. If you're effective running the ball through three quarters, you may not average a high YPC in the fourth quarter, but you'll keep running. If you've got 14 carries for 42 yards after three quarters and are down by 10 points, you're unlikely to have a 100 yard day. So in that sense, raw rushing yards tells a pretty good story. And that's why raw rushing yards and adjusted yards per carry over 3.0 YPC won't bring out too many differences. While I might have a slight preference for AYPC over 3.5 or 3.0 yards per carry, rushing yards hits the nail pretty well (especially if you just use rushing yards by running backs).

Weaker quarterbacks can shoot up the rankings a bit if they're on bad teams that need to throw a lot -- that's one of the reasons that QB scoring data can look more linear than running back data. But that's just because FP is a better measure of true value for RBs than QBs.For example, using QB12 as the baseline and ignoring all strike seasons:-- the correlation coefficient between VBD and FP from '85 to '09 is 0.24-- the CC between VBD and FP from '70 to '09 is 0.20While at RB, using RB24 as the baseline and ignoring all strike seasons:-- the CC between VBD and FP from '85 to '09 is 0.65-- the CC between VBD and FP from '70 to '09 is 0.70But now... let's grade all QBs by AY/A -- that is, (Passing Yards + 20*PassingTDs - 45*INTs)/Attempts) for all quarterbacks who met the league minimum in terms of attempts. If we then sort all qualifying quarterbacks by AY/A, and use QB12 as the baseline for AY/A, we can run the same sort of analysis. Of course, because AY/A has risen so dramatically over the years due to various rule changes, you need to compare like years to like years:From '70 to '77, the CC between the sum of all QB AY/A values in the top 12 and the sum of their "VBD" AY/A values was 0.40Passing was in flux for the first few years after the big rules changes of '78, but I think you can point to the late '80s as the time when passing games really took advantage of the rule changes. From '88 to '00, the CC was 0.64. Then from '01 to '09, when passing has gone through the roof (I don't have '10 data in my database yet), the CC was 0.63.The point is we should expect any measure that accurately measures ability to appear as exponential or logarithmic functions, not linear ones. And when that's the case, FP and VBD numbers should be correlated. With QBs, because of the loose correlation between QB ability and QB FP, it loses that true connection.Does that help? Again, I'm not totally sure I understand the question.

 

[Aabye]

Chase,

I think we're going in two different directions.

Here's the project:

I wanted to see if there was a link between (1) how many fantasy points the entire group of top 24 RBs scored in a given year a (2) how many VBD points that same group of RBs scored during that year. I gathered data for each season between 1985-2010. For example, here's the data for 2007:

2007 NAME FPT VBD1 LaDainian Tomlinson 307.8 1722 Brian Westbrook 282.4 146.63 Adrian Peterson 238.9 103.14 Clinton Portis 235.9 100.15 Joseph Addai 233.6 97.86 Jamal Lewis 221.2 85.47 Marion Barber 197.7 61.98 Willis McGahee 191.8 569 Frank Gore 189.8 5410 Edgerrin James 184.8 4911 Earnest Graham 182.2 46.412 Marshawn Lynch 176.3 40.513 Maurice Jones-Drew 171.5 35.714 Steven Jackson 167.4 31.615 LenDale White 164.2 28.416 Willie Parker 160 24.217 Ryan Grant 158.3 22.518 Fred Taylor 156 20.219 Kenny Watson 155.7 19.920 Brandon Jacobs 154.5 18.721 Chester Taylor 154.5 18.722 Thomas Jones 145.6 9.823 Justin Fargas 143.7 7.924 Reggie Bush 135.8 0 TOTAL 4509.6 1250.4I went through every season from 1985-2010 and recorded each year's total FPT and VBD for the top 24 RBs. In 2007, the FPT total was 4509.6 and the VBD total was 1250.4.I then plotted those yearly totals on a graph to see if there was any correlation between FPTs scored during a given year and VBDs scored during that same year. I also did this for the other offensive positions as well, using standard baselines of WR30, QB12, and TE12.

Here are the results:

RBs

WRs

QBs

TEs

***

I found that there was a strong positive correlation between FPTs and VBDs scored by RBs and WRs, but I didn't see the same thing for QBs or TEs.

So now I had a new question: why do we see this strong correlation for RBs and WRs but not for TEs or QBs?

Hoosier16 suggested that perhaps the difference comes from the fact that there are fewer QBs and TEs, so specific players will have a much greater effect on the total FPT and VBD scoring. I checked to see if the VBD scoring of the top RB, WR, QB, or TE was correlated with total VBD scoring for that year and didn't find much. The top player doesn't have much to do with overall scoring.

***

ZWK had the idea of separating out broad trends in fantasy scoring from the year-to-year fluctuations. Post #24 is his attempt to do this. What he found was that year-to-year fluctuations are more or less the same for every position (CC of between .60 and .69). The difference comes out when we look at the longer term trends. There, the correlation between FPT and VBD is very strong for RBs and WRs (.79 and .88, respectively), considerably weaker for TEs (.44) and negative for QBs (-.44). This is an incredibly useful piece of information, but it doesn't yet answer the "why?" question.

***

ZWK suggested that perhaps the long term trends were due to different factors. He writes:

I think what's happening is that at RB & WR, trends towards heavier usage influence the players at the top more than anyone else (it's the elite RBs who become workhorses), but at QB there has been a more dramatic leaguewide change towards passing. It used to be that a few elite passing teams were throwing a lot, but now everyone's doing it and there are lots of good passing teams, so the catch-up by the rest of the league is happening faster than the elite few can pull away. TE is a mix of elite improvement and a trend towards more widespread usage of TEs as receivers.

I checked to see if when RB1-6 comprised a larger % of the yearly VBD scoring we would see increases in FPT and/or VBD scoring. It turns out that there is no correlation.I did the same for many other groups (see post 29) and found no good correlation. In other words, it looks like we can't explain increases in fantasy scoring at any position by either "elite improvement" or "catch up."

So I think the question still stands.

Does this make sense? I get somewhat confused myself.

 

[Brothers Mayhem]

Forgive me for jumping in the middle of something that is way over my head - but I am curious as to what you are hoping to gain from the result of your entire theory to start with? I mean, if I am understanding correctly, you are hoping to learn something in regards to strategical drafting utilizing either the FTP system or a VBD system. I guess where I am at a complete loss here is that when you highlight one correlation to another - how do we apply these findings to garner an edge on draft day?

Again, it's not my intention to interrupt a well structured study you guys have going here - on the contrary, I find it extremely interesting. Just trying to gain some perspective on where this info is suppose to guide me. Better put - How do I apply your findings?

 

[Aabye]

Forgive me for jumping in the middle of something that is way over my head - but I am curious as to what you are hoping to gain from the result of your entire theory to start with? I mean, if I am understanding correctly, you are hoping to learn something in regards to strategical drafting utilizing either the FTP system or a VBD system. I guess where I am at a complete loss here is that when you highlight one correlation to another - how do we apply these findings to garner an edge on draft day?

Again, it's not my intention to interrupt a well structured study you guys have going here - on the contrary, I find it extremely interesting. Just trying to gain some perspective on where this info is suppose to guide me. Better put - How do I apply your findings?

I can't speak for anyone but myself here.Originally, I started with a pretty simple hypothesis: that FPT and VBD scoring should be more or less independent. If that had been the case, it would have suggested that fantasy worries like "if we see more RBBCs, RBs as a group will become less valuable" were baseless and we could do some interesting stuff from that point (e.g. maybe the value stays the same but the weighting gets "shifted" so that the top backs are worth way more and the lower backs are worth way less.)

That turned out to not be the case, at least for RBs and WRs. FPTs and VBDs are linked, so now we have a new problem: VBDs do fluctuate, so why do they fluctuate?

Now things get murky, at least in terms of what we hope to gain. That's going to depend on where the correlation is coming from. It's possible that, in the end, we can't figure out what's driving the trend lines so we can't glean anything directly useful. It's also possible that we find something interesting and we can use that connection to learn something interesting to apply to FF strategy.

I'd say that so far, you really can't apply the findings in a useful way. But we're sorta in the middle of things, so if we can sort it out more clearly, there may be something useful at the end of it.

 

[kerntuber]

Aabye, I have been trying to follow along to see what comes out of this discussion. Regarding your question related to why QB's and TE's do not follow the same trend as the RB's and WR's, I do believe that it is related to your sample sizes for each position. I recreated your initial chart in part (1985-2002) but only took the top 12 RB's and saw similar variances to the QB's and TE's in that there were years where increases to FPT didn't match the trend increase to VBD. If you still have your original docs, can you try to confirm this for the entire study and all positions? I believe that there is a correlation of FPT and VBD and that the larger the sample you take (top 50 at each position) the more definative the link between the two. I'm no math wizard, but it seems that the two should follow each other given a broad enough sampling to weed out the anomolies. Not sure if that helps or hinders the process. These are the results I had from my sampling at the top 12 RB's:

FPT VBD

1985 2935.4 1128.2

1986 2705 1050.2

1988 2636.9 959.3

1989 2696.1 1253.7

1990 2411.1 913.5

1991 2329.5 905.1

1992 2768.1 1450.5

1993 2216.4 872.4

1994 2601.7 1071.7

1995 2875.8 1195.8

1996 2697.1 1191.1

1997 2686 1099.6

1998 2972.5 1443.7

1999 2764.9 1204.9

2000 3181.9 1272.7

2001 2839.9 1219.9

2002 3278.6 1451

 

[Aabye]

Aabye, I have been trying to follow along to see what comes out of this discussion. Regarding your question related to why QB's and TE's do not follow the same trend as the RB's and WR's, I do believe that it is related to your sample sizes for each position. I recreated your initial chart in part (1985-2002) but only took the top 12 RB's and saw similar variances to the QB's and TE's in that there were years where increases to FPT didn't match the trend increase to VBD. If you still have your original docs, can you try to confirm this for the entire study and all positions? I believe that there is a correlation of FPT and VBD and that the larger the sample you take (top 50 at each position) the more definative the link between the two. I'm no math wizard, but it seems that the two should follow each other given a broad enough sampling to weed out the anomolies. Not sure if that helps or hinders the process. These are the results I had from my sampling at the top 12 RB's: FPT VBD1985 2935.4 1128.21986 2705 1050.21988 2636.9 959.31989 2696.1 1253.71990 2411.1 913.51991 2329.5 905.11992 2768.1 1450.51993 2216.4 872.41994 2601.7 1071.71995 2875.8 1195.81996 2697.1 1191.11997 2686 1099.61998 2972.5 1443.71999 2764.9 1204.92000 3181.9 1272.72001 2839.9 1219.92002 3278.6 1451

Yeah, good idea. Let me check.

 

[Aabye]

Aabye, I have been trying to follow along to see what comes out of this discussion. Regarding your question related to why QB's and TE's do not follow the same trend as the RB's and WR's, I do believe that it is related to your sample sizes for each position. I recreated your initial chart in part (1985-2002) but only took the top 12 RB's and saw similar variances to the QB's and TE's in that there were years where increases to FPT didn't match the trend increase to VBD. If you still have your original docs, can you try to confirm this for the entire study and all positions?

Yeah, good idea. Let me check.

Instead of adding more QBs or TEs, I reduced RBs and WRs. I don't have the information for the other QBs and TEs, so it's way more work to add those than to just cut down RBs and WRs. So for every group, I used a baseline of the 12th player. Got these results:RB CC: .57314WR CC: .29874QB CC: .17868TE CC: .3375So yeah, I think you're right. I'm a little surprised how high the CC is for RBs, actually. Thought it would probably be less.Now, if we move the baseline down to RB50 or whatever, we're going to see a huge correlation between FPT and VBD because when you go that far down, you get such a flat baseline that FPT fluctuation is basically the only thing that matters for VBD fluctuation.

 

[ZWK]

Aabye, I have been trying to follow along to see what comes out of this discussion. Regarding your question related to why QB's and TE's do not follow the same trend as the RB's and WR's, I do believe that it is related to your sample sizes for each position. I recreated your initial chart in part (1985-2002) but only took the top 12 RB's and saw similar variances to the QB's and TE's in that there were years where increases to FPT didn't match the trend increase to VBD. If you still have your original docs, can you try to confirm this for the entire study and all positions?

Yeah, good idea. Let me check.

Instead of adding more QBs or TEs, I reduced RBs and WRs. I don't have the information for the other QBs and TEs, so it's way more work to add those than to just cut down RBs and WRs. So for every group, I used a baseline of the 12th player. Got these results:RB CC: .57314WR CC: .29874QB CC: .17868TE CC: .3375So yeah, I think you're right. I'm a little surprised how high the CC is for RBs, actually. Thought it would probably be less.Now, if we move the baseline down to RB50 or whatever, we're going to see a huge correlation between FPT and VBD because when you go that far down, you get such a flat baseline that FPT fluctuation is basically the only thing that matters for VBD fluctuation.

Do you have the data for this, so that I can try my thing of separating it into the trend and the year-to-year variation?

 

[Aabye]

Aabye, I have been trying to follow along to see what comes out of this discussion. Regarding your question related to why QB's and TE's do not follow the same trend as the RB's and WR's, I do believe that it is related to your sample sizes for each position. I recreated your initial chart in part (1985-2002) but only took the top 12 RB's and saw similar variances to the QB's and TE's in that there were years where increases to FPT didn't match the trend increase to VBD. If you still have your original docs, can you try to confirm this for the entire study and all positions?

Yeah, good idea. Let me check.

Instead of adding more QBs or TEs, I reduced RBs and WRs. I don't have the information for the other QBs and TEs, so it's way more work to add those than to just cut down RBs and WRs. So for every group, I used a baseline of the 12th player. Got these results:RB CC: .57314

WR CC: .29874

QB CC: .17868

TE CC: .3375

So yeah, I think you're right. I'm a little surprised how high the CC is for RBs, actually. Thought it would probably be less.

Now, if we move the baseline down to RB50 or whatever, we're going to see a huge correlation between FPT and VBD because when you go that far down, you get such a flat baseline that FPT fluctuation is basically the only thing that matters for VBD fluctuation.

Do you have the data for this, so that I can try my thing of separating it into the trend and the year-to-year variation?

Sure thing: LINK

 

[ZWK]

Correlations in the deviation (reflecting year-to-year variation) are basically the same as all the others, .61 for RBs and .64 for WRs (the previous ones were .63 for 24 RBs, .60 for 30 WRs, .61 for 12 TEs, and .69 for 12 QBs).

The correlations between the trend lines are .84 for RBs (basically the same as with 24 RBs, when it was .79) and -.42 for WRs (with 30 WRs it was .88, the highest correlation of all). Each additional fpt is associated with a .33 VBD increase for RBs (vs. a .30 VBD increase with the RB24 baseline) and a -.11 drop in VBD for WRs (vs. a .27 VBD increase with the WR30 baseline). So, move the baseline from RB24 to RB12 and nothing changes, but move it from WR30 to WR12 and the direction of the relationship between fpts & VBD flips.

 

[Jeff Pasquino]

I've been following this discussion off and on - can someone boil some practical application of the results, if any?

Not trying to belittle the discussion at all - just trying to wrap my mind around the usage of the outcome.

 

[kerntuber]

Jeff, I've been following as well and I won't speak for the others, but I think there is some anecdotal interest in knowing where there are increases and decreases in total fantasy points between different positions and total VBD points. It may help in a draft to know that there are growing or shrinking points projected for the top 20 RB's or top 30 WR's. I'm not sure that it is something different or more useful than the draft dominator app. I think the mechanics of the draft dominator program take the subtle differences between the positions, what's left in the draft pool, and give you the values to your team. The outcome of this exercise may give you some overall shifts in your draft strategy of how to target or start your draft based on the combined totals for each position. Maybe some of the others have some different thoughts they could share.

 

[Aabye]

I've been following this discussion off and on - can someone boil some practical application of the results, if any?Not trying to belittle the discussion at all - just trying to wrap my mind around the usage of the outcome.

Jeff,At this point the best practical application is probably ZWKs insight that if we remove yearly "noise", RB and WR FPT trends match up with VBD trends, TE trends are only loosely correlated, and QB trends are negatively correlated. So:(1) as RBs as a whole score more FPTs, they score more VBD points as well(2) as WRs as a whole score more FPTs, they score more VBD points as well(3) as TEs as a whole score more FPTs, they seem to score slightly more VBD points(4) as QBs as a whole score more FPTs, they seem to score fewer VBD pointsWe could probably get a lot more practical application if we could figure out why that's the case. So far, no explanation has been found. Still working on it and thinking about it.At this point, there's really no way to tell if the explanation will turn out to have interesting fantasty implications or not.

 

[Aabye]

Jeff, I've been following as well and I won't speak for the others, but I think there is some anecdotal interest in knowing where there are increases and decreases in total fantasy points between different positions and total VBD points. It may help in a draft to know that there are growing or shrinking points projected for the top 20 RB's or top 30 WR's. I'm not sure that it is something different or more useful than the draft dominator app. I think the mechanics of the draft dominator program take the subtle differences between the positions, what's left in the draft pool, and give you the values to your team. The outcome of this exercise may give you some overall shifts in your draft strategy of how to target or start your draft based on the combined totals for each position. Maybe some of the others have some different thoughts they could share.

Yeah, this is well put.

 

[Jeff Pasquino]

Jeff, I've been following as well and I won't speak for the others, but I think there is some anecdotal interest in knowing where there are increases and decreases in total fantasy points between different positions and total VBD points. It may help in a draft to know that there are growing or shrinking points projected for the top 20 RB's or top 30 WR's. I'm not sure that it is something different or more useful than the draft dominator app. I think the mechanics of the draft dominator program take the subtle differences between the positions, what's left in the draft pool, and give you the values to your team. The outcome of this exercise may give you some overall shifts in your draft strategy of how to target or start your draft based on the combined totals for each position. Maybe some of the others have some different thoughts they could share.

Yeah, this is well put.

So - if I may - if the data supports it (and I'm waiting to see if that is or not), we're talking about separations in tiers of players expanding or contracting. Is that it? Or is it a comparison across different positions?

 

[ZWK]

I've been following this discussion off and on - can someone boil some practical application of the results, if any?

Not trying to belittle the discussion at all - just trying to wrap my mind around the usage of the outcome.

This is a good question to be asking. I think the main practical point is to better understand the relative fantasy values of the different positions, and how they have changed over time. One way to look at the fantasy values of the positions is with the total VBD points scored by each position. Here is another graph which looks at that more directly - it shows what percent of the total VBD points (for all offensive positions) were scored by each position, using the 5-year average trend lines to smooth out the year-to-year fluctuations.You don't have to look at these long term trends in order to get the relative values of the positions correct. If you make good projections and do the VBD calculations, then the relative values for next year should naturally fall out of the calculations. But it can help to be able to look at the trends to see how the relative values have changed over time, in order to better update your understanding and your strategies which you may have developed several years ago when things were different. Seeing the trends can also help you predict what is going to happen over the next several years (which is particularly relevant for dynasty).

You could just look at the VBD trend, but if you can also explain why the trend has happened then you'll have a better understanding of it, and maybe be better able to predict what will happen over the next several years. An obvious place to look for an explanation of the VBD trend is at total fpts scored by each position. It seems plausible that RBs will account for more VBD when they're scoring more fpts, and for less VBD when they're scoring fewer fpts, so changes to RB fpts will explain the trend in RB VBD. And with RBs, that does seem to be the case. That's also how it works with WRs (at least as long as you use a baseline of WR30 and not WR12). But with QBs (and, to a lesser extent, TEs) that's not what has happened. This thread has mostly been focused on figuring out why that is.

 

[Aabye]

Jeff, I've been following as well and I won't speak for the others, but I think there is some anecdotal interest in knowing where there are increases and decreases in total fantasy points between different positions and total VBD points. It may help in a draft to know that there are growing or shrinking points projected for the top 20 RB's or top 30 WR's. I'm not sure that it is something different or more useful than the draft dominator app. I think the mechanics of the draft dominator program take the subtle differences between the positions, what's left in the draft pool, and give you the values to your team. The outcome of this exercise may give you some overall shifts in your draft strategy of how to target or start your draft based on the combined totals for each position. Maybe some of the others have some different thoughts they could share.

Yeah, this is well put.

So - if I may - if the data supports it (and I'm waiting to see if that is or not), we're talking about separations in tiers of players expanding or contracting. Is that it? Or is it a comparison across different positions?

As far as comparison across different positions, that's where the ZWK stuff is great. It gives us the trend lines. We can't yet explain the reason(s) for the way the trend lines are moving, but at least we have the data.As for separating players at a given position into tiers, I did work on that HERE.. It's confusing to just look at the data, so here's the explanation.

I wanted to see if the larger trends (e.g. RBs are scoring more FPTs and VBDs) could be explained by looking at how the total positional fantasy value was distributed amongst the top 24 RBs. So, for instance, was a good year for fantasy RBs more likely to be one in which the best RBs (the top 6 scoring RBs for a given year) scored a bunch of VBD points? Was it one in which the worst RBs scored fewer VBD points? Here's how I tested it:

(1) I calculated the percentage of total VBD that each top 24 RB was responsible for. For example, in 2010, there were 1423.2 VBD points scored by RBs. Arian Foster had a VBD value of 188.8, so his percentage of the total VBD is about 13.3% (188.8/1423.2).

(2) I then added up the totals for the top 6 RBs, the next 6 RBs (7-12), the next 6 RBs (13-18), and finally the last 6 RBs (19-24). So in 2010, the top 6 RBs had percentages of 13.2, 7.2, 7.1, 7.1, 6.5, and 6.0. That meant that the total% for the top group was about 47.1. In 2009, the total% for top 6 RBs was about 60.4%.

(3) I took these yearly RB totals and graphed them to see if there was any correlation in how fantasy points were distributed in a given year and how good that year was overall. From 2005-2010 the RB distributions were:

RB1-RB6 RB7-RB12 RB13-RB18 RB19-RB24 VBD2010 0.471191681 0.315837549 0.173412029 0.039558741 1423.22009 0.604288826 0.255905512 0.118529067 0.021276596 1193.82008 0.471206922 0.308891541 0.163881844 0.056019693 1340.62007 0.563819578 0.246161228 0.130038388 0.059980806 1250.42006 0.671963843 0.207464645 0.080842688 0.039728823 1371.82005 0.622606587 0.242277253 0.102118969 0.032997192 1566.8You can see that RB distributions jump around quite a bit.I found no good correlation at all between any of the groups I tested and how well RBs did as a whole. That indicates that how fantasy points are distributed has very little to do with how valuable the position is. So tiers expanding or contracting seems to have nothing to do with total fantasy value at a given position. I also checked to see if distribution patterns had anything to do with FPTs. Again, I found no good correlation.

I did the same study for all positions (the spreadsheet has a tab for each position) and found no good correlation anywhere, so the mystery isn't just about RBs. Distribution % just doesn't seem to do anything at all.

This is weird, since expanding or contracting tiers seems like the most reasonable explanation for why positional values vary over time. It just makes intuitive sense to think that the when you have an elite tier of RBs that's scoring way more fantasy points than usual, RBs as a whole are going to have a good year. But it just doesn't look like that's what matters. What does actually matter? Well, that's the mystery.

 

[Jeff Pasquino]

I've been following this discussion off and on - can someone boil some practical application of the results, if any?

Not trying to belittle the discussion at all - just trying to wrap my mind around the usage of the outcome.

This is a good question to be asking.

Thanks - I am trying here.

I think the main practical point is to better understand the relative fantasy values of the different positions, and how they have changed over time.

....

You could just look at the VBD trend, but if you can also explain why the trend has happened then you'll have a better understanding of it, and maybe be better able to predict what will happen over the next several years. An obvious place to look for an explanation of the VBD trend is at total fpts scored by each position. It seems plausible that RBs will account for more VBD when they're scoring more fpts, and for less VBD when they're scoring fewer fpts, so changes to RB fpts will explain the trend in RB VBD. And with RBs, that does seem to be the case. That's also how it works with WRs (at least as long as you use a baseline of WR30 and not WR12). But with QBs (and, to a lesser extent, TEs) that's not what has happened. This thread has mostly been focused on figuring out why that is.

In a layman's attempt (i.e., not using math or data to back subjective views), I would take a stab at the reasoning behind QB and TE values. 1. While the overall league suffers from lack of quality QBs, overall the depth of the QB talent pool for fantasy purposes is much better these days. That's part of the philosophy of "wait on your QB" because QB12 isn't as much of a drop from QB1, and waiting 6-7 rounds is well worth 3-4 points a game or week (if that) while you get better RB and WRs.

2. For TE's, it pretty much goes right along with #1. The league used to be dominated by a few TEs at a given time, but now there are several who can be productive and viable fantasy options (especially in PPR leagues).

3. As for a trend, the passing values have gone up since around 2000-2002 or so, and you can see that trend in the one graph. WRs and TEs are higher in value since that point in time, but QBs are not. I believe that's because the Top 12 QB pool is deeper than before - a direct result of both the NFL becoming more of a passing league along with 12-15 viable passers starting in that decade (and today).

4. I believe that you might see a different trend if the cutoff for QB and TE was lower. Most wouldn't go beyond QB12 or TE12 since most leagues wouldn't care beyond the nominal 12 starters, but you may learn more by re-baselining at 15, 18 or 20.

I'll let you guys who are banging on the stats take a look at the suggestion of #4.

 

[GregR_2]

Here's the data for quarterbacks and tight ends.

These two positions show no correlation whatsoever between total FPTs scored and total VBD points scored.

So the question is: why is there a correlation between FPTs and VBDs for RBs and WRs but not for QBs or TEs?

I did a correlation coefficient (CC) for all four complete data sets. For those not familiar with it, it measures how much two series of numbers trend together. The result ranges from 1 (perfect correlation) to -1 (perfect inverse correlation). The closer the CC is to 0, the less correlation between the data sets.For the complete data set, QB is .18, TE is .34, WR is .66 and RB is .64, matching that WR and RB are correlating much more than QB and TE.

I was wondering if doing a full 25 years of stats in those might be influencing the results. There are a number of rule changes that have affected the passing game over that length of time. Including twice that I recall off the top of my head where cracking down on DB illegal contact resulted in some jumps in the passing game.

Rather than try to track down all the major changes like that, I just did a quick CC for the positions in the 90s, and again for the 00s. Results are quite a bit different. For 2001-2010: QB .70, TE .57, WR -0.14, RB .74. For 1991-2000: QB: .65, TE .52, WR .75, RB .72.

So looking at a 10 year window rather than a 25 year window, QB and TE are now showing similar correlation, but WRs which suddenly shows no correlation at all in the 00s but shows the most correlation 90s?

So then I delved into WRs a little deeper to see why this might bne. I looked at 10 year windows for every year rather than just by decade. So 2010 is a correlation coefficient from 2001-2010. 2005 is the CC for 1996-2005, etc.

2010 -0.14

2009 0.00

2008 0.03

2007 0.05

2006 0.07

2005 0.08

2004 0.48

2003 0.50

2002 0.72

2001 0.65

2000 0.75

1999 0.78

1998 0.77

1997 0.86

1996 0.85

1995 0.88

So up until 2002 the WR numbers over the preceding decade correlate nicely. That correlation drops off a bit in 2003 and disappears once you include 2004. What happened in 2004? That was one of the two times when the NFL started emphasizing calling DBs for pass interference, and passing numbers exploded as a result. But for WR, while total fantasy points exploded to the 2nd highest total in the 25 year span, the fantasy point scoring did not increase comparatively. Without seeing the graph of the positional scoring curve for that year, I'd hypothesize that the rule changed helped bring up the bottom part of the curve more than it did the top. Resulting in a lot more fantasy points added, but a higher baseline which had the net effect of keeping the VBD values low.

So if QBs are affected like this, what about QBs? Again, these are CC based on the 10 years previous to the year given. So the 2009 CC is from 2000-2009.

2010 0.70

2009 0.75

2008 0.68

2007 0.60

2006 0.56

2005 0.24

2004 0.05

2003 -0.04

2002 0.41

2001 0.60

2000 0.65

1999 0.66

1998 0.55

1997 0.57

1996 0.62

1995 0.66

We see a similar trend. The correlation goes away gets hammered right around 2003 and 2004, although QB correlation increases again towards the present while WR stayed inconsistent.

TE also has a drop though it's in 2002 and 2003. RB, by contrast, has the highest correlation for 10 year windows around that same time. RB shows a big jump in 1998.

In summary, what does all of that mean? I'm not entirely sure. I believe the less data used in a CC the more likely it is to correlate, but I think a 10 year window is probably still a large enough data set to be representative. (Doug or Chase agree with that?) But I'd say this look at the data might highlight that you have a number of different factors over the 25 year time span that are influencing the FP and VBD in sometimes conflicting fashions, and 25 years worth might be too big a window to get much meaningful conclusion from.

I'm also not entirely sure of how to apply any of this to fantasy football. When you come down to it, I'm going to use my predictions for the coming season to determine the value between positions. Being aware of some general trends like this could maybe help me test my predictions against historical reality, but I don't know that total VBD points is something I would include in that. I'd rather do it by total leaguewide yards and touchdowns and pass attempts and such.