Fantasy Success by NFL Draft Round (1 Viewer)

[ZWK]

Players who are taken early in the NFL draft tend to have better NFL careers, and better fantasy careers, than players who are taken later. But how big of a difference is there? And how does it vary across positions? Does a typical 3rd round RB provide more fantasy value than a typical 2nd round WR, QB, or TE? What about 4th round RBs? There's been some discussion of this in other threads, and I decided to look at some data.

When you're making picks in your dynasty rookie draft, obviously you should be paying attention to the particular player, his situation, and your league's rules and customs. But it helps to have some sense of the baseline for what you can typically expect from a player at that position taken in that round of the NFL draft. So I looked at every RB, WR, TE, and QB drafted from 1993 to 2006, and measured the success of their fantasy career based on their cumulative fantasy points over baseline for their career (VBD, taken from PFR - it's the number that's 642 for Tiki Barber at the bottom of his player page).

The full results are in this google spreadsheet, but I'll save the detailed explanation for the end of this post. Here's the quick version, a ranking of position/round combos based on the percentage of players who had at least a decent fantasy career (50+ cumulative VBD). 1st round RBs come first, since 58% of RBs drafted in the first round of the NFL draft from 1993-2006 ended up with at least 50 career fantasy points over baseline. Here's the ranking:

rd 1 RB (58%)

rd 1 WR (46%)

rd 1 TE (44%)

rd 1 QB (36%)

rd 2 RB (29%)

rd 2 QB (27%)

rd 2 WR (22%)

rd 3 RB (20%)

rd 6 QB (14%) [this is a fluke]

rd 3 WR (14%)

rd 4 RB (13%)

rd 2 TE (11%)

rd 5 RB (10%)

rd 3 TE (8%)

rd 4 WR (8%)

rd 5 QB (6%)

rd 6 RB (6%)

rd 3 QB (6%)

rd 4 QB (5%)

rd 4 TE (4%)

and that's enough

RBs do have an advantage over other positions, but not as big as I expected - not even a full round. Round 2 RBs are less likely to be valuable than round 1 players at any position, round 3 RBs are less likely to have value than round 2 WRs or QBs, and round 4 RBs pan out less often than round 3 WRs. Value drops off gradually for RBs and WRs, but at QB there's a huge dropoff in value after round 2 and at TE after round 1. Although I wouldn't trust those TE results too much, since there have been big changes at the position since 1993 (also, blocking TEs are mixed in with receiving TEs, and there's a small sample size with only 15 guys meeting the 50+ VBD threshhold).

One complaint you might have about this ranking: fantasy is about hitting homeruns, not just finding decent contributors. It's about getting studs like Tomlinson, Moss, Manning, or Gonzalez. And, relatedly, not all first round picks are the same - players taken early in the first round (like Manning 1st overall, Tomlinson 5th, and even Gonzalez 13th) could have a much better chance to be studs than guys taken at the end of the round.

Here's another ranking that takes that into account. First, I based the ranking of each position/round combos on the average VBD of the players taken in that round. For instance, second round RBs averaged 82 career VBD - that's averaging together the 642 from Tiki Barber, the 43 from Anthony Thomas, a bunch of 0's from guys who never panned out, and so on. And I also split up the first round into smaller segments. At RB & TE I split it in half - top 16 picks I'll call round 1a and picks 17+ I'll call round 1b. At QB the biggest gap seems to be between QBs taken 1st overall and the rest, so I split the first round into pick 1 vs. picks 2+. For WRs I left the 1st round all together, since late first round WRs actually have a slightly better track record than early first rounders. Here's that ranking:

rd 1a RB (303 vbd)

pk 1 QB (233 vbd)

rd 1a TE (193 vbd)

rd 1b RB (143 vbd)

rd 1 WR (119 vbd)

rd 2 RB (82 vbd)

pk 2+ QB (73 vbd)

rd 2 QB (63 vbd)

rd 3 RB (60 vbd)

rd 2 WR (47 vbd)

rd 3 WR (46 vbd)

rd 1b TE (40 vbd)

rd 6 QB (30 vbd)

rd 6 RB (23 vbd)

rd 4 RB (23 vbd)

rd 3 TE (19 vbd)

rd 5 RB (18 vbd)

rd 2 TE (16 vbd)

rd 4 WR (14 vbd)

and we'll stop there

One stud can have a big impact on these rankings, like Manning for the 1st pick QBs and Gonzalez for the early 1st round TEs, which could be seen as a disadvantage of this ranking method (if it's just random variation with a small number of players skewing the numbers) or an advantage if it actually reflects where you're most likely to find a stud. An elite QB or TE is very valuable, and it makes sense that the very best prospects go 1st overall at QB and in the early first at TE. If you buy these numbers, it suggests that the best QB and TE prospects should be considered up there with everyone besides the very best RB prospects (and if you can distinguish the Peyton Mannings from the Alex Smiths, that's even better). The numbers show a big dropoff at QB after the #1 overall pick, then other first rounders are similar to second rounders, and then there's another big dropoff. At TE there's a big dropoff during the first round, and another drop at the end of the round.

RBs have a bigger advantage here than in the other ranking, and you can see why the top RBs are generally the top fantasy rookie picks. Round 3 RBs are close to round 2 QBs and ahead of rd 2 WRs, although there's a big dropoff after round 3 for RBs.

Methodology (and explanation of the data spreadsheet):

I used PFR's database to look at all RBs, WRs, QBs, and TEs taken in the 14 drafts from 1993 to 2006. I started in 1993 since the draft was much longer before then (I left out the 1993 8th rounders), and went up to 2006 since 5 years is enough to have a pretty good measure of a player. Using PFR's season fantasy stats on the player pages, I kept track of 3 things for each player. First, whether the player was ever a fantasy starter in his career (meaning a top 24 RB, top 36 WR, top 12 QB, or top 12 TE). Second, whether the player was a fantasy starter in the first 3 years of his career. And third, the player's total VBD for his career (through 2010), as calculated by PFR, relative to RB 24, WR 30, QB 12, or TE 12.

The spreadsheet compares players by position and round - the first row looks at all RBs taken in round 1, and so on. One column of the spreadsheet contains the average VBD for players at that position in that round. The next column contains the proportion of the players who were ever a fantasy starter. The next contains the proportion of players who were a fantasy starter in the first 3 years of their career. The next two columns contain the proportion of players who had 50 VBD or more for their career and who had 100 VBD or more. The final column gives the number of players who were included in that row's data. The last several rows of the spreadsheet divide up the first round into smaller chunks.

 

[_lurker_]

 

[Multiple Scores]

Awesome post.

 

[UnknownCoach]

Another reason why top rookie RBs are the best choice is that virtually every 1a RB who busts does so because of either injury or off field problems. RBs are easy to scout and you can practically just rank them the same way the NFL drafts them. The only big difference between NFL scouting and fantasy is size. An NFL team will rank larger backs higher and smaller backs lower when you translate their fantasy potential. That's because fantasy and real football are two very different games, not because the scouts are wrong. You also have to add some fantasy value to starter prospect backs who come out of college with receiving skills.

On the other hand, its much more diffacult to scout the college passing game for NFL success. That's why receivers and especially QBs are frequently busts.

Its not rocket science. Dont worry about what happened one year or the other. Dont believe that you are smarter than the NFL scouts or that some randomly pulled past stat determines what will happen in the future. Just trust the NFL scouts and realize that college football, NFL football, and fantasy football are three different things.

 

[drunken slob]

Great info

Thx

 

[sihaokills]

this is great

 

[guderian]

 

[Aabye]

ZWK,

This is a really cool study. I did something eerily similar (split up players by position and draft round, sorted by career VBD, looked at which players scored >50 career VBD points, looked at how quickly players started producing, etc) as a part of a pretty extensive attempt to create a quantitative model for dynasty player values. My study took into account drafts between 1985 and 2000 and I got similar bust rates and career averages (well, my averages are a little higher but that's to be expected since you're counting plenty of active players who are still scoring points).

I split the groups up a little bit differently, and got this:

Group (sample size) - average career VBD

Top 10 RB (22) - 251

1st round RB* (41) - 134

2nd round RB (49) - 85

3rd round RB (50) - 43

Top 5 WR (5) - 90

1st round WR* (49) - 183 <- Jerry Rice is in here, which basically messes up the set

2nd round WR (58) - 50

Top 5 QB (15) - 192

1st round QB* (17) - 52

Top 20 TE (9) - 166

*"1st round RB" means all 1st round RBs not selected in the top 10, likewise for other groups.

Bust Rates (Bust = >50 career VBD points):

Top 10 RBs: 11/22 50%

1st round RBs: 22/41 54%

2nd round RBs: 35/49 71%

3rd round RBs: 42/50 84%

Top 5 WRs: 2/5 40%

1st round WRs: 26/49 53%

2nd round WRs: 43/58 74%

Top 5 QBs: 7/15 47%

1st round QBs: 13/17 76%

Top 20 TEs: 3/9 33%

A couple of things:

(1) the only place our numbers are significantly off in with 1st round WRs. Jerry Rice is an outlier but even if I take him out completely, I get a career VBD average of 154, which is still higher. I wonder why that is. Is it just statistical noise or is there something else going on?

(2) one or two players can wildly skew the numbers we're getting. I'm not well trained in statistics so I don't know how much confidence I can put into these numbers. The standard deviations for most groups are gigantic and when I tried to do a confidence interval, that was huge too. So I do worry that for all of the data mining you've obviously done, there might not be any strong conclusions to draw. For instance:

My top 10 RBs category has a standard deviation of approx. 375 and for a confidence level of 90%, my range for the true average of top 10 RBs is 119-382 (meaning that I can say, with 90% confidence, that the actual average career VBD for top 10 RBs falls between 119 and 382 VBD points). You're using the same numbers, so you'll have the same huge SDs and ranges.

So an open question to you or anyone else who might know about this:

How much confidence can we place in these numbers? Any idea?

Again, really cool post.

 

[gianmarco]

 

[WVU Alum]

ZWK and Aabye hurt my brain.

 

[Hoosier16]

Here are a couple of questions and suggestions:

[*]What was the scoring system?

[*]Where were your VBD cutoffs for each position (I see ZWK lists his but he lists 2 different cutoffs for WR - 30 and 36).

[*]Have you looked at median (although as I write this, I can see that may not be very helpful with the bust rates you list)?

[*]Have you looked at expanding the VBD cutoffs? Bye weeks and distribution of players affect how many players get used at each position. I'd look at expanding those by about 50% to account for that - maybe RB36, WR54, . . .

[*]For Aabye, why top 10 RB, but only top 5 WR?

Otherwise, great stuff!

 

[coolnerd]

with all the numbers

 

[loose circuits]

My top 10 RBs category has a standard deviation of approx. 375 and for a confidence level of 90%, my range for the true average of top 10 RBs is 119-382 (meaning that I can say, with 90% confidence, that the actual average career VBD for top 10 RBs falls between 119 and 382 VBD points). You're using the same numbers, so you'll have the same huge SDs and ranges.So an open question to you or anyone else who might know about this:How much confidence can we place in these numbers? Any idea?Again, really cool post.

this is the best post in this whole thread. I believe it is where the truth lies. There are far too many factors to accurately predict rookies performance based off completely different players. Every situation is different and that is the beauty of fantasy football. What are your critical values for a 98 or 99% confidence interval?

 

[Multiple Scores]

If any of this actually made sense to you, why did you blow the 1.9 rookie pick on Helu, a 4th round scrub?Me thinks you don't get any of this.

My top 10 RBs category has a standard deviation of approx. 375 and for a confidence level of 90%, my range for the true average of top 10 RBs is 119-382 (meaning that I can say, with 90% confidence, that the actual average career VBD for top 10 RBs falls between 119 and 382 VBD points). You're using the same numbers, so you'll have the same huge SDs and ranges.So an open question to you or anyone else who might know about this:How much confidence can we place in these numbers? Any idea?Again, really cool post.

this is the best post in this whole thread. I believe it is where the truth lies. There are far too many factors to accurately predict rookies performance based off completely different players. Every situation is different and that is the beauty of fantasy football. What are your critical values for a 98 or 99% confidence interval?

 

[ZWK]

with all the numbers

If the numbers are hard to follow, just look at the rankings. The first set of rankings say that first round RBs turn out to have the most fantasy value, on average, followed by first round WRs, then first round TEs, and so on. The second set of rankings looks at the same question in a slightly different way, and finds that early first round RBs tend to have the most fantasy value, followed by quarterbacks drafted #1 overall, followed by early first round TEs, then late first round RBs, and so on.If you want to get a little fancier, look at the number in parenthesis next to each one - if there's a big gap in the numbers that means that there's a big gap between the positions, and if the numbers are close then they're close. For instance, in the first set of rankings, round 2 WRs finished ahead of round 3 RBs but they were very close.

Here are a couple of questions and suggestions:

[*]What was the scoring system?

[*]Where were your VBD cutoffs for each position (I see ZWK lists his but he lists 2 different cutoffs for WR - 30 and 36).

[*]Have you looked at median (although as I write this, I can see that may not be very helpful with the bust rates you list)?

[*]Have you looked at expanding the VBD cutoffs? Bye weeks and distribution of players affect how many players get used at each position. I'd look at expanding those by about 50% to account for that - maybe RB36, WR54, . . .

[*]For Aabye, why top 10 RB, but only top 5 WR?

Otherwise, great stuff!

My VBD numbers were taken directly from the PFR player pages, so it's their scoring system. 1/10 yd rushing & receiving, 1/25 yd passing, 6/td rushing receiving or passing, -2/int. Baseline is RB 24, WR 30, QB 12, and TE 12.If I'd done it from scratch I would've used WR 36 as the baseline, but having to calculate career VBD rather than just taking it from PFR would've been a lot more work (probably double the work). When I looked at the percentage of players who were ever a fantasy starter (or who were a fantasy starter in their first 3 years), I did use top 36 WR instead of 30 since it was easy to choose my own cutoff for that.

Aabye, that's neat stuff. It looks like we were thinking along very similar lines. You're right that there are going to be pretty wide confidence intervals, especially for the VBD average. The VBD average is less statistically reliable than the hit rate stats (percent of players with 50+ VBD, or 100+ VBD, or who were ever a fantasy starter, or who were a fantasy starter in their first 3 years), since it's more susceptible to a small number of extremely high scoring players. The fact that I have a bunch of active players in my sample is also more of a problem for the VBD average than for the other statistics. Guys like Larry Fitzgerald and Philip Rivers (both class of 2004) could accumulate a lot more VBD over the next several years, but they already count as hits on all of the other measures. If I was only looking at VBD average I wouldn't have gone all the way up to the 2006 draft, but for the other stats the 2006 cutoff should work okay (a few players might come out of the woodwork to cross the 50 VBD threshold, but not many). It is worth keeping in mind that RBs are probably overrated relative to other positions on my rankings by VBD average (since they tend to have the shortest careers, so the active ones don't have much more VBD to accumulate) and QBs are probably the most underrated (since they have the longest careers and thus the most VBD left to accumulate).

I have more to say about how confident we can be in these results, given the smallish sample sizes and the noise in the data, but I'm going to look at some more numbers first. I'll try to post about it tonight.

 

[Aabye]

Here are a couple of questions and suggestions:

[*]What was the scoring system?

[*]Where were your VBD cutoffs for each position (I see ZWK lists his but he lists 2 different cutoffs for WR - 30 and 36).

[*]Have you looked at median (although as I write this, I can see that may not be very helpful with the bust rates you list)?

[*]Have you looked at expanding the VBD cutoffs? Bye weeks and distribution of players affect how many players get used at each position. I'd look at expanding those by about 50% to account for that - maybe RB36, WR54, . . .

[*]For Aabye, why top 10 RB, but only top 5 WR?

Otherwise, great stuff!

It looks like ZWK's answers to 1, 2, and 5 are identical to my own. The others:3. Because of the high bust rates, most of the medians are 0.

5. Short answer: I made the cutoffs based mostly on intuition about where elite talent at each position tends to go in the NFL draft. NFL GMs like LTs, DEs, and QBs. They don't like RBs, TEs, OGs, or Cs. For whatever reason, elite WR prospects seem to be a part of the "liked" group. There wasn't anything scientific about it, really.

 

[Aabye]

My top 10 RBs category has a standard deviation of approx. 375 and for a confidence level of 90%, my range for the true average of top 10 RBs is 119-382 (meaning that I can say, with 90% confidence, that the actual average career VBD for top 10 RBs falls between 119 and 382 VBD points). You're using the same numbers, so you'll have the same huge SDs and ranges.

So an open question to you or anyone else who might know about this:

How much confidence can we place in these numbers? Any idea?

Again, really cool post.

this is the best post in this whole thread. I believe it is where the truth lies. There are far too many factors to accurately predict rookies performance based off completely different players. Every situation is different and that is the beauty of fantasy football. What are your critical values for a 98 or 99% confidence interval?

My range at 99% confidence is 45-457, so the confidence interval is +/- 206.Two Thoughts:

(1) If we think of these numbers as a tool for accurately predicting how many VBD points a specific rookie player will score, we're frequently going to be way off. But if we think of the numbers as giving us a fair price for a specific rookie player, then it's not so bad.

So think about buying a lottery ticket in a fair lottery. There are 3 prizes (100 dollars, 60 dollars, and 40 dollars) and you sell 100 tickets. Since it's a fair lottery, each ticket will cost $2. That $2 isn't a good predictor of what your ticket is actually worth, but it's perfect for describing what you should be willing to pay to play the game.

(2) I'll start by saying that I don't know if the following actually characterizes your position, so apologies if I've got you wrong here:

Sometimes, people are resistant to attempts to quantitatively analyze players in a dynasty format. I've heard a fair number of folks say what you're saying here ("far too many factors" and "every situation is different"). This puzzles me. After all, we think that it's perfectly legitimate to price life insurance, even though the same criticisms apply there.

It is admittedly difficult to quantitatively analyze players, since there are a lot of factors and the data sets aren't very big. But it's certainly not impossible. The relevant question isn't whether we can use these numbers to make better predictions (of course the numbers help), it's just a question of how well we think we can price player values based on the available data. Can we get pretty close or are we just stuck in a world of insufficient data sets and low confidence?

 

[Aabye]

I have more to say about how confident we can be in these results, given the smallish sample sizes and the noise in the data, but I'm going to look at some more numbers first. I'll try to post about it tonight.

One thought I had is that while individual data sets might be fairly uncertain when taken alone, perhaps we can kinda look at all of the data sets together to increase our confidence.For instance, it seems intuitively obvious that players drafted higher should be more valuable. We both get the result that 1st round RBs are more valuable than 2nd round RBs, who are more valuable than 3rd round RBs, which matches up with the intuition. So can we use the fact that our data sets are well-ordered with respect to that intuition to lend additional confidence to the numbers we're generating?It's times like this that I wish I knew something about statistics.

 

[loose circuits]

My top 10 RBs category has a standard deviation of approx. 375 and for a confidence level of 90%, my range for the true average of top 10 RBs is 119-382 (meaning that I can say, with 90% confidence, that the actual average career VBD for top 10 RBs falls between 119 and 382 VBD points). You're using the same numbers, so you'll have the same huge SDs and ranges.

So an open question to you or anyone else who might know about this:

How much confidence can we place in these numbers? Any idea?

Again, really cool post.

this is the best post in this whole thread. I believe it is where the truth lies. There are far too many factors to accurately predict rookies performance based off completely different players. Every situation is different and that is the beauty of fantasy football. What are your critical values for a 98 or 99% confidence interval?

My range at 99% confidence is 45-457, so the confidence interval is +/- 206.Two Thoughts:

(1) If we think of these numbers as a tool for accurately predicting how many VBD points a specific rookie player will score, we're frequently going to be way off. But if we think of the numbers as giving us a fair price for a specific rookie player, then it's not so bad.

So think about buying a lottery ticket in a fair lottery. There are 3 prizes (100 dollars, 60 dollars, and 40 dollars) and you sell 100 tickets. Since it's a fair lottery, each ticket will cost $2. That $2 isn't a good predictor of what your ticket is actually worth, but it's perfect for describing what you should be willing to pay to play the game.

(2) I'll start by saying that I don't know if the following actually characterizes your position, so apologies if I've got you wrong here:

Sometimes, people are resistant to attempts to quantitatively analyze players in a dynasty format. I've heard a fair number of folks say what you're saying here ("far too many factors" and "every situation is different"). This puzzles me. After all, we think that it's perfectly legitimate to price life insurance, even though the same criticisms apply there.

It is admittedly difficult to quantitatively analyze players, since there are a lot of factors and the data sets aren't very big. But it's certainly not impossible. The relevant question isn't whether we can use these numbers to make better predictions (of course the numbers help), it's just a question of how well we think we can price player values based on the available data. Can we get pretty close or are we just stuck in a world of insufficient data sets and low confidence?

I agree with some of what you are saying, but I disagree that you should write players off because they don't fit into the mold when their situation seems to be much better than people drafted ahead of them like Helu who has less competition than most other rookie RB's. In other words, it's useful as reference but I don't believe we should always draft along the lines of how NFL GM's see it which is basically what the stats say to do although they do show lots of variance which means any route you go has risk. If you wanted to use this as gospel, guess there is no reason to do any research since you would just be using the draft order as reference

 

[Aabye]

I agree with some of what you are saying, but I disagree that you should write players off because they don't fit into the mold when their situation seems to be much better than people drafted ahead of them like Helu who has less competition than most other rookie RB's. In other words, it's useful as reference but I don't believe we should always draft along the lines of how NFL GM's see it which is basically what the stats say to do although they do show lots of variance which means any route you go has risk. If you wanted to use this as gospel, guess there is no reason to do any research since you would just be using the draft order as reference

Oh, I don't think that the numbers themselves tell the whole story or that fantasy draft order should just mirror NFL draft order. I think that the numbers provide a good starting point, after which you have to "correct" for factors that are less easily quantified (character concerns, concerns with development, situation, etc.). It's not about writing players off, but rather about generating realistic fantasy expectations for incoming rookies.My honest opinion is that these sorts of stats probably matter more than people think they do. People like to do their own scouting. And people also famously overestimate their ability to predict success (see: sports books in Las Vegas, Nevada). I am less confident in my ability to do non-quantitative scouting, so I tend to stick a little closer to draft order. For instance, the fantasy community this year is way lower on the 1.01 QB (Cam Newton) than they were on the 1.01 QB last year (Sam Bradford). I realize that these two players are very different sorts of QBs, so a gifted scout could probably tell me more about who has better fantasy prospects. But I honestly don't know enough to make reasonable predictions based on these scouting distinctions, so I am tempted to rate the two guys as very similar as prospects.

 

[Biabreakable]

I have more to say about how confident we can be in these results, given the smallish sample sizes and the noise in the data, but I'm going to look at some more numbers first. I'll try to post about it tonight.

Curious if you ever got back to this?

Great work. I was just looking up confidence coefficients and essentially any result of 95% is considered reliable or very high confidence in the accuracy of the data.

One way to normalize the data would be to take the total number of players drafted each round in your sample and balance them so that you are not creating a frequency bias in the final numbers. The sample sizes are too small to really be concerned with this but I do think it would be technically more correct, although perhaps not different or significantly different than the results you already found.

 

[ZWK]

I have more to say about how confident we can be in these results, given the smallish sample sizes and the noise in the data, but I'm going to look at some more numbers first. I'll try to post about it tonight.

Curious if you ever got back to this?

Great work. I was just looking up confidence coefficients and essentially any result of 95% is considered reliable or very high confidence in the accuracy of the data.

One way to normalize the data would be to take the total number of players drafted each round in your sample and balance them so that you are not creating a frequency bias in the final numbers. The sample sizes are too small to really be concerned with this but I do think it would be technically more correct, although perhaps not different or significantly different than the results you already found.

When I looked at the numbers re confidence, they ended up not being that informative. The 95% confidence interval for a single group (e.g. 2nd round RBs) is very wide, but that involves treating each group (of position & round) as a separate category, which is throwing away a lot of information. The numbers that I gave here should be unbiased estimates (except for issues like still-active players), because I made them just by averaging together all the relevant historical data, but the error in any one number could be pretty large. Still, the overall pattern is fairly clear, and I suspect that it would take an unlikely amount of random errors to create the appearance of this pattern when it wasn't really there.

I think it's safe to take it as given that, within a position, there will be some kind of downward trend with later draft picks having less VBD. Questions to ask include how quickly that decline happens, what shape it has, and how it differs between different positions. The way that I presented the data in this post gives a rough sense of the patterns, but it's not set up in a good way for doing statistical testing. (My more recent posts on generic rookie rankings are based on analyzing the same data set in a way that was a bit more statistically sophisticated, looking at the gradual decline with draft pick rather than categorizing players by rounds, but I don't think I looked that closely at the uncertainties / error bars when I did those analyses either.)