The effect of contracts on player performance (1 Viewer)

[Birdnals]

Hey all. I'm looking to do a study analyzing the effects of player contracts on performance. Specifically, I want to analyze

1) Players playing in a contract year

2) Players in their final year of their rookie deal

3) Performance in first season of new contract with same team

4) Players in first season on new contract with a new team

My plan is to use some simple t-scores to see if there is a significant difference (statistically speaking) between these types of players and all others and then use some regression methods to find variables which may predict variances from the norm (either poor or good).

I need some help though and I know of no better place to look than the SP. What I need is a centralized place to find accurate contract info. Rotoworld is pretty good but if I didn't know if anyone had any other suggestions. I'm also looking for suggestions for the project such as variables to look at (I'm thinking days missed from training camp, age, experience, where contract ranks financially amongst others in their positions, length of contract) or any other suggestions.

I hope to be doing a few things like this over the summer. I'm a statistics major in school and see this as a way to help me get a better understanding and apply what I'm learning in the classroom to something I love. I'm planning on looking for ways to identify mid/late round value, injury risk, and analyze early season trends (flash in the pan plavers vs. sustainable output free agents), and preseason projections. Again, if anyone has any suggestions or requests for analysis let me know and I'll see what I can do. Thanks in advance!

 

[Lavachebeadsman]

I've always used CapGeek. Let me know if there is anyway I can help this study. I'm pretty interested.

 

[DoubleG]

A couple of quick points (and I am not trying to burst your proverbial bubble):

1) If I recall correctly, there has been shown to be little to no correlation in contract year performances. Many people have suffered confirmation bias when looking for it - but statistical analysis has shown very little change in most players stats in a contract year.

2) In regards to points 2, 3 & 4, beware of the fact that there are numerous other factors other than contract status that affect player performance. For example, a players better performance in the final year of his rookie deal may simply be a function of his maturity, his coaching staff trusting him with a larger workload, his becoming more acclimated to a full NFL season and offseason work needed, etc. Relation does not equal causation. In fact, contract status may simply be coincidental. The same holds true with playing for new teams. A RB who goes from being a COP back with his previous team to a starter on a new team with a better offensive line might drastically increase productivity - having nothing to do with contract. Even if he stays with the same team, he might land a new contract because he is now the starter.

I would suggest that there are a great many statistical correlations in the NFL that can be studied - but they are not related to contract status.

 

[Birdnals]

A couple of quick points (and I am not trying to burst your proverbial bubble):

1) If I recall correctly, there has been shown to be little to no correlation in contract year performances. Many people have suffered confirmation bias when looking for it - but statistical analysis has shown very little change in most players stats in a contract year.

2) In regards to points 2, 3 & 4, beware of the fact that there are numerous other factors other than contract status that affect player performance. For example, a players better performance in the final year of his rookie deal may simply be a function of his maturity, his coaching staff trusting him with a larger workload, his becoming more acclimated to a full NFL season and offseason work needed, etc. Relation does not equal causation. In fact, contract status may simply be coincidental. The same holds true with playing for new teams. A RB who goes from being a COP back with his previous team to a starter on a new team with a better offensive line might drastically increase productivity - having nothing to do with contract. Even if he stays with the same team, he might land a new contract because he is now the starter.

I would suggest that there are a great many statistical correlations in the NFL that can be studied - but they are not related to contract status.

Totally agree. The first thing you learn as a statistics major is that correlation does not equal causation (which is also why I'm excited for the upper level classes where you being to examine causal inference!). I'm not sure what to expect as there are numerous examples, for example, of players who have performed incredibly after signing a new contract. What I'm looking to study is the ones who break the mold, the outliers. Are there any common characteristics of those who perform poorly/incredibly? For example, you mention age. Do younger players who bust after signing a new contract bust a higher rate than older ones? What about ones with more NFL experience (to account for players being drafted at different ages and number of offseasons)? You've also posed some other questions worth looking at for both this study and future ones like "Does the number of Pro Bowlers on the offensive line effect Stat x, y, or z and can the effect be quantified?"

By no means am I trying to suggest that fantasy football, or anything, can be deduced to numbers or a formula. I am interested though in looking for trends and maybe debunking a few myths along the way. To paraphrase Mark Twain, most people use statistics for a crutch to support their point rather than for the light they shed. I really hope to be in the camp of the latter and hopefully y'all in the SP will be able to help me as I go and provide an honest check on my work and findings.

 

[Birdnals]

I've always used CapGeek. Let me know if there is anyway I can help this study. I'm pretty interested.

Looks like exactly what I need but unfortunately that site is only for NHL.

 

[Lavachebeadsman]

I've always used CapGeek. Let me know if there is anyway I can help this study. I'm pretty interested.

Looks like exactly what I need but unfortunately that site is only for NHL.

Is it really? ####, that's the only sport I can remember using it for, I just assumed they would have NFL contracts. Lemme do some searching, brotha.

 

[Lavachebeadsman]

Here ya go dude, this gives individual cap numbers for all NFL players. http://www.spotrac.com/cap-tracker/nfl/

 

[Birdnals]

Here ya go dude, this gives individual cap numbers for all NFL players. http://www.spotrac.com/cap-tracker/nfl/

Heck ya! Thanks!

 

[mjr]

Certainly would be an interesting analysis, but work ethic/attitude is going to have to be considered if you want an accurate read on it (particularly in the "Performance in first season of new contract with same team" stat). Just look at the difference between Chris Johnson and Lynch's performances the year after they got their new contracts. Maybe there's a pattern in the size of the contracts or the history of the team itself there.

 

[Birdnals]

Certainly would be an interesting analysis, but work ethic/attitude is going to have to be considered if you want an accurate read on it (particularly in the "Performance in first season of new contract with same team" stat). Just look at the difference between Chris Johnson and Lynch's performances the year after they got their new contracts. Maybe there's a pattern in the size of the contracts or the history of the team itself there.

CJ and Lynch are definitely the first two that come to mind. I think work ethic and attitude might be the biggest predictor of success/failure here and is essentially what I am trying to quantify. Age/experience are going to be the first things I look at, (Johnson signed his new contract 2 years after his rookie deal, Lynch 4 years after his). Length and size of contract also differ greatly. Lynch's 4 year deal leave him little wiggle room for a bad season if he wants to get another decent contract. A 6 year deal like the one CJ could make him feel like he has more breathing room for mistakes. This is a sample size of 2 though so we'll see how it plays out when we start factoring in larger data sets. I think other things to look at might be days missed from training camp and OTA attendance. Any other suggestions?

 

[Birdnals]

I believe I am ready to start compiling data for the study. I’m going to examine the following variables to see what kind of correlation there is between a player’s contract and their on field performance. Specifically, I’ll be looking for any outliers and what they have in common. I’ll be examining the following:

Position

Age

NFL Experience

# of NFL Contracts

Free agent or extension?

Years left on remaining contract

New team?

OTA participation?

Did the player hold out?

If the player held out, for how many days?

Guaranteed money in new deal

Average salary of new deal

Percentile quadrant new deal represents amongst players in same position

Length of new contract

Change in number of Pro Bowlers on offensive line (+/-)

Change in number of Pro Bowlers at skill positions (+/-)

Previous seasons fantasy points

Standard deviation in career from season to season

I’ll then look to see how many points they scored in the first season with their new contract and how this compares to their previous seasons (within a given amount of standard deviations).

I will use t-tests to determine which of the variables may be statistically significant. After identifying players who performed above or below one standard deviation I will look for common trends in these variables that separate them from the median.

Using this information, I’ll create a linear regression model to try and predict how players under new contracts this year will perform in regards to their career standard deviation. If possible, I will use a logistic regression model to predict the odds of a player performing within their standard deviation as well as the odds of them performing either above or below it.

I’m going to try and start collecting data tonight and will hopefully have some initial findings to report on next week. Yes, I know that correlation does not indicate causation. There are many things beyond quantifiable variables that will effect a player’s performance. I am mostly doing this for fun and as a way to connect my classroom material to the real world in an applicable way. Any significant findings will just be a cherry on top.

If anyone has any suggestions or critiques, I’d love to hear them. I’ll let you guys know once I’ve built my dataset and if we want to try and find anything else with it shouldn’t be too difficult to do.

 

[Chase Stuart]

Have you looked into any of the studies on this topic? I would probably google the topic and read a few of those before proceeding.

 

[Birdnals]

Have you looked into any of the studies on this topic? I would probably google the topic and read a few of those before proceeding.

Chase,

I did look around for some other studies. Most of them look to see if there is a statistically significant difference in performance. I'm hoping to go beyond this and look for indicators that might identify potential outliers for us. Results have varied and it looks like the most conclusive study was done as part of the NFL Network's Freakonomics segment. His study looked at running backs and did find some evidence of decline. I assume I'll find this as well, and hopefully I can find some clues as to why this happens allowing us to identify players who do better or worse than the median. Another study from an economics student at Brown found no "contract year phenomenon" as he called it. Again, I expect to find something similar but can hopefully identify potential outliers. Unfortunately, neither of the presentations are very thorough from a technical standpoint (I'm sure they did some thorough technical work, it's just not disclosed in their presentation) and neither provide and resources. Another problem I've seen from their studies is that sample sizes are relatively small (n<30). If I can't gather large sample sizes either, I'll use bootstrapping methods to create them. I think there will be enough data to use bootstrapping in an accurate fashion. I'm also thinking of contacting the man behind the NFL Network study and seeing if he can shed any light as to what he did and if he would have any further advice for my own study. If you have any other suggestions, or know of any other studies to look at, please pass them!

 

[Birdnals]

Done with my initial data mining. I looked for players since 2007 who signed a new contract and were above a certain fantasy rank the year before they signed their new deal (top 24 QB, top 36 RB, top 48 WR, top 24 TE). I ended up with a total number of 89 players, 16 QB, 22 RB, 17 TE, and 34 WR. I haven't finished collecting all the variables or run any statistical tests but there are some initial findings. Please note when I say "contract year" I am just referring to the year before they signed a new deal. It doesn't necessarily mean that their contract was expiring at the season's end.

• Players seem to peak in the year before they signed a new deal. Only 27/89 players improved after signing a new deal.
• 30 of the 89 players signed extensions. Only 8 of them improved from the after signing. 18 of the 30 had below average numbers the next season. Players who signed with new teams had similar numbers.
• 25 players had a below average year in their contract year. 17 of them continued a downward trend in the first season of their new deal.
• Age doesn't seem to matter in regards to improving the year following a new contract.

• Roy Williams was the greatest example of signing a new contract and tanking it. After signing a 54 million dollar contract with Dallas in 2008, his production dropped by more than 8 standard deviations.
• On a PPG basis, Andre Johnson improved by almost 4.5 standard deviations after inking his 70 million dollar deal with Houston in 2007. Unfortunately, he only played in season games that season.

Over the next couple of days I'm going to look at some other variables and start running some tests to see if there are any distinguishing features of those who under/over perform and depending on how that goes see if I can build a logarithm to predict under or over performance.

 

[Chase Stuart]

A couple of thoughts.

You will want to measure your results against a control group -- i.e., players in the top 24, top 36, etc., who were not in a contract year. Saying only 27/89 improved doesn't have any meaning without a baseline to compare it against.

You should understand that when data mining, you won't be able to infer any predictive power from your results. If you want to data mine to find out what's happened historically, that's fine, but then you won't have any data to test your results on. It would be like saying "after going through the data, players named Gonzalez are awesome." One way to test whether the results are predictive would be to split the data in half, data mine one half, and then test your results on the other half.

 

[Birdnals]

A couple of thoughts.

You will want to measure your results against a control group -- i.e., players in the top 24, top 36, etc., who were not in a contract year. Saying only 27/89 improved doesn't have any meaning without a baseline to compare it against.

You should understand that when data mining, you won't be able to infer any predictive power from your results. If you want to data mine to find out what's happened historically, that's fine, but then you won't have any data to test your results on. It would be like saying "after going through the data, players named Gonzalez are awesome." One way to test whether the results are predictive would be to split the data in half, data mine one half, and then test your results on the other half.

Thanks for the input, Chase. The control group aspect is a valid point. Since the data appears to be normally distrinbuted, I was planning on using those who finished within one standard deviation, which is the norm, as my control group against players who finished +/- one standard deviation. What are your thoughts on that?

As for predicitive power, the way I would test my results would be by examining the residuals and applying a goodness of fit test on my final model.

 

[Chase Stuart]

A couple of thoughts.

You will want to measure your results against a control group -- i.e., players in the top 24, top 36, etc., who were not in a contract year. Saying only 27/89 improved doesn't have any meaning without a baseline to compare it against.

You should understand that when data mining, you won't be able to infer any predictive power from your results. If you want to data mine to find out what's happened historically, that's fine, but then you won't have any data to test your results on. It would be like saying "after going through the data, players named Gonzalez are awesome." One way to test whether the results are predictive would be to split the data in half, data mine one half, and then test your results on the other half.

Thanks for the input, Chase. The control group aspect is a valid point. Since the data appears to be normally distrinbuted, I was planning on using those who finished within one standard deviation, which is the norm, as my control group against players who finished +/- one standard deviation. What are your thoughts on that?

As for predicitive power, the way I would test my results would be by examining the residuals and applying a goodness of fit test on my final model.

I don't think how you measure against a control group is very important; it's just that some measure needs to be there. Within one standard deviation seems fine, although it couldn't hurt to test out a few different methods.

Examining the residuals and applying a goodness of fit will only tell you about correlation, which won't necessarily tell you the predictive power of the system.

 

[Lord Fantasy]

 

[Donsmith753]

Hernandez is an interesting case...