Which Positions Sizzled during Best Ball Summer ‘21 and ‘22?

Author: Bill Jones

Introduction

Hey there, fantasy football analytics fam! Welcome back to the blog. Today, we're diving back into Best Ball summer as I continue my exploration of the Best Ball Mania data provided through the Best Ball Data Bowl. Best Ball is the perfect format for those looking to draft during the summer fantasy downtime and offers a twist on traditional fantasy football by eliminating the need for weekly roster management. I’m a little over 100 drafts in on Underdog ($5 and $7 Puppy and Poodle tournament are too hard to resist) and I feel like I have a pretty good sense for the current draft market. The most annoying thing about drafting on Underdog right now is the current Wide Receiver landscape, as Wide Receivers fly off the board at an outrageous rate (Adam Levitan’s tweet demonstrates this perfectly). This wide receiver obsession got me wondering and I dug into the data to see what historical BBMs could tell us about the importance of scoring by position by employing logistic regression models on the data from the past two Best Ball Mania tournaments. Without further ado, here is what I have discovered.

BBM Basics

For those who may be less familiar with Best Ball Mania (BBM) by Underdog, it’s a unique and immensely popular fantasy football tournament. Best ball is a fantasy football format where participants draft a team and then have no in-season management requirements like waiver wire additions or setting lineups. Instead, the highest-scoring players are automatically placed in the starting lineup, and leagues are decided based on cumulative season points rather than head-to-head matchups.

BBM takes best ball to another level with its distinct two-stage structure: the regular season and playoffs. During the 14-week regular season, teams compete against others in their draft, with the top two teams advancing to the playoffs. The playoffs then consist of a series of single-week DFS-style competitions, where teams from different drafts, potentially with overlapping players to your roster, vie for advancement to the next round.

Data Source

Before I jump back into analytics, I would like to acknowledge the data source. The data used for this analysis was obtained from Underdog Fantasy and is the basis for the Best Ball Data Bowl. There have been three BBMs but the Best Ball Data Bowl competition states the competitors should only be using the BBMII and BBMIII data as the BBMI data is a bit of a mess, so I’ll be following the same rules here.

Logistic Regression

To shed light on the importance of scoring by position, I utilized a logistic regression analysis using data from the past two Best Ball Mania tournaments. Logistic regression is a statistical modeling technique used to analyze the relationship between independent variables and a binary outcome variable. In the context of the Best Ball Mania data analysis, I will be employing logistic regression to determine the most influential position group (independent variables) in terms of Best Ball Mania advancement (dependent variable). To interpret the output of a logistic regression model, we need to consider several components: coefficient estimates, standard errors, z-values, and p-values.

Coefficient Estimates: These values represent the estimated impact of each independent variable on the log-odds of the dependent variable. Positive coefficients indicate that an increase in the independent variable leads to an increase in the log-odds (and thus the probability) of the event occurring, while negative coefficients indicate the opposite. The magnitude of the coefficients reflects the strength of the relationship between the independent variable and the dependent variable.

Standard Errors: These values estimate the variability or uncertainty associated with the coefficient estimates. Smaller standard errors indicate greater precision in the coefficient estimates, while larger standard errors suggest more uncertainty.

P-values: P-values (pr(>|t|)) assess the statistical significance of the coefficient estimates. This figure indicates the probability of observing a coefficient as extreme as the one estimated if there was no true relationship between the independent variable and the dependent variable. Typically, a significance level (e.g., α = 0.05) is chosen, and if the p-value is below this threshold, it is considered statistically significant, suggesting a meaningful relationship between the variable and the outcome. A smaller p-value indicates stronger evidence against the null hypothesis (no relationship) and suggests that the coefficient estimate is statistically significant.

Z-value: Z-value refers to the coefficient's z-score associated with a particular predictor variable. In logistic regression, predictor variables are often represented by regression coefficients, denoting the relationship between the predictor variable and the log-odds of the outcome variable. The z-value represents the number of standard deviations that the coefficient estimate deviates from its expected value under the null hypothesis of no relationship between the predictor variable and the outcome variable. It is calculated by dividing the coefficient estimate by its standard error. Typically, if the absolute value of the z value is greater than 1.96 (assuming a significance level of 0.05), the coefficient is considered statistically significant.

By examining these components, you can assess the significance and direction of the relationships between the independent variables and the dependent variable in the logistic regression model.

Analysis

Data Processing

To facilitate the analysis, some light data transformations were needed. The data BBMII and BBMIII data was summarized by tournament entry and position and totaled pick points. This data was then unpivoted and joined with the playoff data to create an Advanced to Playoffs column as a binary result of advanced = 1 and did not advance = 0. The final cleaned data looked as such:

With that now explained, let’s look at the logistic regression results for BBMII and BBMIII. 

BBMII

QB: The coefficient estimate is 0.02955, with a highly significant t-value (97.616) and a p-value < 2e-16. This indicates that the QB position group has a statistically significant impact on the log-odds of success in BBMII.

RB: The coefficient estimate is 0.03226, with a highly significant t-value (143.734) and a p-value < 2e-16. The RB position group also has a statistically significant effect on the log-odds of success in BBMII.

WR: The coefficient estimate is 0.03192, with a highly significant t-value (145.741) and a p-value < 2e-16. Similarly, the WR position group shows a statistically significant impact on the log-odds of success in BBMII.

TE: The coefficient estimate is 0.03101, with a highly significant t-value (103.483) and a p-value < 2e-16. The TE position group is also found to have a statistically significant effect on the log-odds of success in BBMII.

FB: The coefficient estimate is -0.2058, with a non-significant t-value (-0.862) and a p-value of 0.389. The FB position group does not reach statistical significance, indicating that it does not have a significant impact on the log-odds of success in BBMII.

BBMIII

Interpretation of the results:

QB: The coefficient estimate is 0.04695, with a highly significant t-value (198.414) and a p-value < 2e-16. This indicates that the QB position group has a statistically significant impact on the log-odds of success in BBMIII.

RB: The coefficient estimate is 0.04731, with a highly significant t-value (225.556) and a p-value < 2e-16. The RB position group also has a statistically significant effect on the log-odds of success in BBMIII.

WR: The coefficient estimate is 0.04710, with a highly significant t-value (225.706) and a p-value < 2e-16. Similarly, the WR position group shows a statistically significant impact on the log-odds of success in BBMIII.

TE: The coefficient estimate is 0.04728, with a highly significant t-value (195.735) and a p-value < 2e-16. The TE position group is also found to have a statistically significant effect on the log-odds of success in BBMIII.

FB: The coefficient estimate is -0.02306, with a non-significant t-value (-0.199) and a p-value of 0.842. The FB position group does not reach statistical significance, indicating that it does not have a significant impact on the log-odds of success in BBMIII.

Takeaways

1. Why is anyone drafting a fullback? I guess maybe there is an argument to be made for it being a sharp uniqueness/playoff week leverage play, but it would be a pretty gross argument to make so I’ll let someone else do that if they want.

2. I was hoping to write “strong wide receiver rooms emerged as the key to success in BBMIII, making the current obsession with wide receivers on Underdog seem reasonable” but this is not the case. Variable significance (z value) for both wide receiver and running back scoring in both BBMII and BBMIII was almost identical as well as quarterback and tight end scoring was also statistically significant for both tournaments. 

3. When looking at the interactions (excluded explanation and detailed results because this is already 2x my preferred post length) the 2x interactions were all negative while the 3x interactions were all positive. This would give support to the idea of needing to hit on multiple position groups of your roster to advance and constructions around two mediocre position groups being dragged along by two elite position groups isn't ideal.

4. I look forward to getting into the weeds about what drove these teams to success by position group. I am already mostly done with the wide receiver position and am incredibly excited to share my results. 

5. I generally agree with the current macro level market sentiment of prioritizing wide receivers in the early rounds. As I will show in the upcoming wide receiver post, early round wide receivers produce at a different level than what can be found in later rounds (even if nailing a breakout player) and therefore, I plan to prioritize zero and hero running back roster constructions in my portfolio. Nevertheless, it has been well noted that what was optimal for BBMI, BBMII, and BBMIII differed, so I will also maintain some exposure to the more traditional running back centric roster constructions. 

6. With a need to have a well-rounded team but the wide receiver market being out of control, this is how am I attacking the wide receiver market this best ball season.

a. Currently I am building my teams around early wide receiver-heavy builds as I believe there is big value to be had if and when uncertain running back situations become certain, or RB2s become RB1s due to unfortunate injuries. Additionally, based upon the current market, if one doesn’t prioritize wide receivers early then they will quickly see how gross the wide receiver options get quickly. At pick 100ish the wide receiver position really drops off a cliff, which aligns nicely with the zero/hero running back window for those that went heavy wide receiver early. In this range there is a list of strong RB2s that I love that have strong value in their current role but could be diamonds if only a few small things break their way (Javonte = good injury news, Kamara = no/limited suspension, Charbonnet & Dillon = Walker or Jones injury, etc.). 

b. I am going to wait to fill out my elite running back/hyper fragile profile until later in the summer. As the best ball season progresses, casual drafters enter the picture as well as running back situations will begin to be clearer and I believe the market will shift to a better environment for filling out my portfolio with the more running back-centric constructions. This concept hopes to take advantage of a shifting draft board where more running backs are picked in the historical running back deadzone, pushing wide receiver talent down the draft board to allow for the elite running back teams to not be as boxed out at the wide receiver position as if the same build were attempted today. 

c. Mike Leone's Best Ball Manifesto would indicate that drafting earlier in the summer is still better for the "hyper fragile" roster construction. I would counter that we are working in a very different draft market this year but if my theory is incorrect, I can’t see an environment the wide receiver avalanche becomes more outrageous than it currently is. This gives me comfort that when I do go to fill out my running back centric portfolio the market will either be as it is now (with several months of additional data to weigh into my selections) or better for drafting that build.

Conclusion

In conclusion, I am a nerd and I had way too much fun working through this project. But the adventure doesn't end here! I look forward to expanding this project and delving into other fascinating data modeling considerations (working with imbalanced data so I can use 1 model for BBMII and BBMIII, removing anomalous data points, etc.). Thank you for joining me on this data-driven journey, and I hope you found this analysis insightful and informative. Stay tuned for more football and fantasy football analytics. Until then, may your fantasy teams thrive, and your player takes never be wrong!

*This blog post was enabled by ChatGPT. The text was generated by me, and the content is my own, but some sentences and wording were provided by the model. I take full responsibility for all information produced in this blog. More information about OpenAI and their technology can be found at https://openai.com.*