PROJECTING UPSIDE / DOWNSIDE : Seeking Feedback (1 Viewer)

[[icon]]

THE PROBLEM:

Not all projections are created equal. While standard projections provide you with a concrete value such as number of yards and TDs a player is expected to get... you have to read much deeper into a player to get an idea of what's behind those numbers. Sure.. he's projected to hit these levels.... but what about upside? Downside? Is this a number that he's a lock for.. or a best guess?

These sorts of things are factors that, IMHO, haven't been addressed in the FF universe as of yet. With Footballguys being among the industry leaders I'd love to see them impliment a means to show these factors in a concise, easy to understand manner.

CONCERNS:

ª Data Overload : You want to provide more information without having to search through column after column of data

• Additional Workload : This is likely something that would be impossible to do for all players.

• Lack of precision : This is more of a feel thing.. showing ballpark ranges for upside/downside as opposed to precice data points.

PROPOSED SOLUTION:

Obviously there can be such a thing as data overload and while grids of numbers upon numbers are good... I feel this sort of thing might be best respresnted visually. My idea would be a chart showing a Bell-curve sort of view. The two axis showing probability and the quantity (TD or Yards). The crest would be indicative of the likely FBG projection... with slopes to the left or right indicating a probable termination point for upside and downside. The degree of slope and distance from the FBG Projection would indicate the range up upside and downside as well as the probability of numbers along that range.

IMHO it's a rather eliegant means to get an at-a-glance take on not only a player's projected output, but he range of possibille outcomes. Factors that would need to be incorporated into this chart would be:

ª Injury Risk

• Role Stability

• Holdout Risk (them or others around them)

etc.

A SAMPLE

LINK TO SAMPLE CHARTS >

There are a few generic charts simply to show how the charts might look. I've added a few bullet points below to indicate pointers on what to read into the charts.

Once you're familiar with the player and the charts the bullets wouldn't really be needed. Seeing signifcant downside in LJ would be understood as a measure of the staff's read on his potential holdout or injury risk. A high confidence, low upside/downside chart might be indicative of a Harrison type player. A low output, high upside chart could be indicative of an unproven guy who FBG likes to breakout... or a guy who's favored to set himself apart in a RBBC battle. I see these primarly as a great means to break apart the tiers of guys projected in the same range. See who's riskier... who has more upside, etc.

Obviously these couldn't be done for every player. I'd see this as something that's added to player spotlights or even for the top players at each position (ie Top 20 QB.... 40RB....60WR....20TE). In addition the probability range of 25-75% makes it clear that these are estimates and are not to be considered the absolute upside/downside values of a player.. but more of a likely range with embedded probabilty of landing at various points within that range. The arcs don't need to be that precise... no need to plot 10 points along each arc. I think simply coming up with a projection, an upside and a downside number and then perhaps tweaking the shape of the curve on either side should provide more than enough data.

Curious what other FBGs might think of this sort of system and what it's value might be when preparing for draft day.

THoughts? Suggestions? Feedback?

 

[Rounders]

Could be an interesting tool to add to the tool belt.....

 

[[icon]]

Could be an interesting tool to add to the tool belt.....

That's essentially what I'm viewing it as. A quick at-a-glance way to get a "big picture" viewpoint on a player. I think something like this is perfect for player spotlights.... but curious what others think about it.

 

[Curado Kev]

I think it would a GREAT idea, for player spotlights for a quick visual. Plus, you could have FBG projections and a another colored line for Message board projections.

If you want to get really fancy, have the different experts lines in their for projections. I feel the most important part of the graph is the "slope" away from the projected lines.

I think most of us want to see quickly if there is more upside or downside risk and that would do it.

Good idea, Ckev

 

[BigJ]

Very, very good idea Icon. Seriously this would take what is already a great set of projections and tools by the website to a whole new level.

I find that most, including the Footballguys kind of "tweak" their projections of a player based on his holdout status, injury history, ect. This would take that random "tweaking" away and allow the true projection of a player to be done. And then you could put in your confidence of this ranking based on other these factors.

Nice Icon, very nice man!

 

[gheemony]

Take a look at Baseball Prospectus for how they do it. They put a percentage likelihood that the player will meet their performance of the prior year. You could do the same by specifying the likelihood that they'll hit the projected number. Take a look at Footballoutsiders.com KUBIAK projections or their Pro Football Prospectus 2007 - they give risk ratings to all of their projections. And finally, an option is to include something that is based on standard deviation of the projection, but that only works for projections spit out by a computer - not by a human. If by a human, the human is just plain guessing about the risk.

 

[[icon]]

Take a look at Baseball Prospectus for how they do it. They put a percentage likelihood that the player will meet their performance of the prior year. You could do the same by specifying the likelihood that they'll hit the projected number. Take a look at Footballoutsiders.com KUBIAK projections or their Pro Football Prospectus 2007 - they give risk ratings to all of their projections. And finally, an option is to include something that is based on standard deviation of the projection, but that only works for projections spit out by a computer - not by a human. If by a human, the human is just plain guessing about the risk.

While I agree that a human can't predict with complete accuracy I'd argue that give nthe shorter career lifespan as well as greater impact of surrounding players (ie impact of OL on a RB's numbers) there isn't really an algorithm that can claim to make an accurate assesment either.... hence why we've not seen anything like that from the guys at FBG.I would wager, however that some of the better minds at FBG (ie dodds and co) can evaluate risk quite well and are likelymore than capable of providing a reasonable spectrum of upside/downside that could come out of the likely scenarios the given player could encounter during the upcoming season.Obviously this chart system doesn't claim to predict injury, nor does it claim to predict breakouts... but what it can do is help the user get a quick idea, at a glance, of what range of production they can expect out of a player as well as a relative likelyhood of any plot on that spectrum coming in. This is less about the confidence rating in a prediction (ala BP) and more about that range of upside/downside. Sure it's not 100% but anyone pretending to be able to predict anything with complete accuracy in this sport are kidding themselves.

 

[David Dodds]

Take a look at Baseball Prospectus for how they do it. They put a percentage likelihood that the player will meet their performance of the prior year. You could do the same by specifying the likelihood that they'll hit the projected number. Take a look at Footballoutsiders.com KUBIAK projections or their Pro Football Prospectus 2007 - they give risk ratings to all of their projections. And finally, an option is to include something that is based on standard deviation of the projection, but that only works for projections spit out by a computer - not by a human. If by a human, the human is just plain guessing about the risk.

While I agree that a human can't predict with complete accuracy I'd argue that give nthe shorter career lifespan as well as greater impact of surrounding players (ie impact of OL on a RB's numbers) there isn't really an algorithm that can claim to make an accurate assesment either.... hence why we've not seen anything like that from the guys at FBG.I would wager, however that some of the better minds at FBG (ie dodds and co) can evaluate risk quite well and are likelymore than capable of providing a reasonable spectrum of upside/downside that could come out of the likely scenarios the given player could encounter during the upcoming season.Obviously this chart system doesn't claim to predict injury, nor does it claim to predict breakouts... but what it can do is help the user get a quick idea, at a glance, of what range of production they can expect out of a player as well as a relative likelyhood of any plot on that spectrum coming in. This is less about the confidence rating in a prediction (ala BP) and more about that range of upside/downside. Sure it's not 100% but anyone pretending to be able to predict anything with complete accuracy in this sport are kidding themselves.

I don't see us getting this done this year, but this is really good stuff. I am going to add it atop my list of things we discuss in the offseason. I think between the different stats minds we have here that we can figure out the best way to display this type of data betwen the different projectors as well.Thanks again for doing this. We are always looking to present the data in the most logical format.

 

[Bob Henry]

Good stuff, icon..

 

[AB in DC]

This is a bit much. I think it would be much easier, and just as helpful, to display something like an error bar for each estimate. You can show, on a single axis, what the expected production would be at 1 SD above or below the mean. And you can show it all on a standard page of projections for a number of players, without having to create separate graphs for each player.

I don't know if I'm explaining this very well, so let me try this graphically:

1	Peyton Manning IND / 6	344.9				|  X  |2	Carson Palmer CIN / 5	 305.2			|   X  |3	Tom Brady NE / 10		 299.1			|   X   |4	Marc Bulger STL / 9	   290.1		  |	X   |5	Donovan McNabb PHI / 5	287.9		 |	 X	 |

 

[Anthony Borbely]

 

[[icon]]

This is a bit much. I think it would be much easier, and just as helpful, to display something like an error bar for each estimate. You can show, on a single axis, what the expected production would be at 1 SD above or below the mean. And you can show it all on a standard page of projections for a number of players, without having to create separate graphs for each player.I don't know if I'm explaining this very well, so let me try this graphically:

1	Peyton Manning IND / 6	344.9				|  X  |2	Carson Palmer CIN / 5	 305.2			|   X  |3	Tom Brady NE / 10		 299.1			|   X   |4	Marc Bulger STL / 9	   290.1		  |	X   |5	Donovan McNabb PHI / 5	287.9		 |	 X	 |

I see what you're saying. That is a great simplified version of the chart above however it's not really giving the same amount of information. The slope of the arc could be a factor. In addition other elements such as various lines for various mod projections (hell they could heven have a customizeable overlay by clicking various buttons to view/hide lines for each mod or his actual stats from various years. In addition the background could be colored to represent league averages, percentiles, etc. I agree the simple bar would convey the range but would also be missing a bit of the detail that could be provided in a chart like above. Hell, they could use them both in various applications once the data points are marked.

 

[AB in DC]

This is a bit much. I think it would be much easier, and just as helpful, to display something like an error bar for each estimate. You can show, on a single axis, what the expected production would be at 1 SD above or below the mean. And you can show it all on a standard page of projections for a number of players, without having to create separate graphs for each player.I don't know if I'm explaining this very well, so let me try this graphically:

1	Peyton Manning IND / 6	344.9				|  X  |2	Carson Palmer CIN / 5	 305.2			|   X  |3	Tom Brady NE / 10		 299.1			|   X   |4	Marc Bulger STL / 9	   290.1		  |	X   |5	Donovan McNabb PHI / 5	287.9		 |	 X	 |

I see what you're saying. That is a great simplified version of the chart above however it's not really giving the same amount of information. The slope of the arc could be a factor.

If you assume that the projection is a median point estimate (which I think it should be), the slope can be inferred from the length of the error bars. Looking at your sample charts, it appears that you're not assuming a median, in which case I'd agree with you, but I think it makes more sense to do so.

 

[[icon]]

This is a bit much. I think it would be much easier, and just as helpful, to display something like an error bar for each estimate. You can show, on a single axis, what the expected production would be at 1 SD above or below the mean. And you can show it all on a standard page of projections for a number of players, without having to create separate graphs for each player.I don't know if I'm explaining this very well, so let me try this graphically:

1	Peyton Manning IND / 6	344.9				|  X  |2	Carson Palmer CIN / 5	 305.2			|   X  |3	Tom Brady NE / 10		 299.1			|   X   |4	Marc Bulger STL / 9	   290.1		  |	X   |5	Donovan McNabb PHI / 5	287.9		 |	 X	 |

I see what you're saying. That is a great simplified version of the chart above however it's not really giving the same amount of information. The slope of the arc could be a factor.

If you assume that the projection is a median point estimate (which I think it should be), the slope can be inferred from the length of the error bars. Looking at your sample charts, it appears that you're not assuming a median, in which case I'd agree with you, but I think it makes more sense to do so.

A median makes sense for players who are established and show little chance for a significant change that could impact thieir output (ie injury, role change, etc). IMHO the major value here is to not show a simple high-low-med projection.. but to try to pinpoint upside and downside... something a simple centerpoint +/- 1 SD will not reflect. Granted... this is an inexact science but it could prove hghly valuable when drafting and debating between a few guys in a specific tier.Not saying a simple median won't work for most players.. .I'd wager that 60%+ of the players would likely be best represented by a median setup like you described. However the real value here is in reflecting those with risk/reward factors that can't be represented by a simple standard deviation.

 

[AB in DC]

Not saying a simple median won't work for most players.. .I'd wager that 60%+ of the players would likely be best represented by a median setup like you described. However the real value here is in reflecting those with risk/reward factors that can't be represented by a simple standard deviation.

Sorry, I shouldn't have used SD -- SD implies the same difference between upside and median as between median and downside.I'm thinking of a way to show both the upside and the downsize to varying magnitudes. Something like 75th, 50th, and 25th percentiles, so that a player with high upside should show a larger difference between upside and median than between median and downside.

 

[[icon]]

Not saying a simple median won't work for most players.. .I'd wager that 60%+ of the players would likely be best represented by a median setup like you described. However the real value here is in reflecting those with risk/reward factors that can't be represented by a simple standard deviation.

Sorry, I shouldn't have used SD -- SD implies the same difference between upside and median as between median and downside.I'm thinking of a way to show both the upside and the downsize to varying magnitudes. Something like 75th, 50th, and 25th percentiles, so that a player with high upside should show a larger difference between upside and median than between median and downside.

Gotcha. It would be interesting to find a way to calculate these ranges. I don't think any computerized simulations can be used... or perhaps you start with computerized data then edit it based on information not included (ie news coming out of camp... injuries being reported....competition for job, etc).

 

[Chaos Commish]

These sorts of things are factors that, IMHO, haven't been addressed in the FF universe as of yet.

I wouldn't go that far. We are limited by static numbers when we rank and project in simple terms, but I know several who consider "ranges" based on "intangibles" into their projections. At another site we had a long geeky thread (Mike?) discussing the value of ranges compared to set figures. Bramel talks about grouping players into ranges. Baseline formulas often consider projected point ranges for tiering. I have always used a homemade code (shorthand) in tiers to express intangible ideas that create differing values for players projected very similarly.I agree your graphs are elegant and would love a tool to create those. :(

 

[[icon]]

These sorts of things are factors that, IMHO, haven't been addressed in the FF universe as of yet.

I wouldn't go that far. We are limited by static numbers when we rank and project in simple terms, but I know several who consider "ranges" based on "intangibles" into their projections. At another site we had a long geeky thread (Mike?) discussing the value of ranges compared to set figures. Bramel talks about grouping players into ranges. Baseline formulas often consider projected point ranges for tiering. I have always used a homemade code (shorthand) in tiers to express intangible ideas that create differing values for players projected very similarly.I agree your graphs are elegant and would love a tool to create those.

Sorry.. yeah I wasn't trying to imply that this is the first time anyone had considered upside or even tried to quantify it in projection form. I just hadn't seen any websites of this nature that really addressed it in this manner. I could be wrong though, as this is the only "pay" site I subscribe to.