How Do I Use This Graph?
There are two ways to use this graph: identify uncertainty between players or identify tiers of players.
To identify uncertainty between players, examine an individual plot point and note how tall the error bar (vertical bar extending from dot) is. The taller the bar, the more uncertain their projection is. The top of the bar is the most likely “high” score, while the lower bar is their most likely “low” score.
To identify tiers of players, examine groups of players by color. If you’re trying to decide between two players, tend to choose the player with the “hotter” color (red>yellow>green>teal>blue>pink). The players in the same color group are within the same tier, when players are in a same tier – they essentially are a wash between each other due to a randomness. This where I’d recommend to use some intuition and anything else (flip a coin?) to make a call.
How Was This Calculated?
This data is compiled from
7 8 different sources projecting actual fantasy football points (PPR) each week. Sources: CBS Sports (3 Analysts), ESPN, FFToday, Bloomberg Sports, FantasySharks, Fox Sports.
The vertical axis is the player’s Fantasy PTS (FPTS) projection. The dot represents their most likely outcome, while the bars represent the standard deviation in their projections. The taller the bar, the more uncertainty in the projection.
The horizontal axis is player’s average rank. This is looking at FantasyPros aggregate rankings (wisdom of the crowd). This helps give a sense of how the “experts” would sort the players.
The colors distinguish tiers of players. This is determined by a clustering algorithm which is clustering players by their projected output. Players in the same color, are expected to perform roughly the same as other players in their color – but they are expected to perform differently than tiers different from theirs.
The scatterplot for each player represent their “most likely” outcome. The 8 projections are analyzed using the Hodges-Lehmann estimator, to create a more robust average. The Hodges-Lehmann estimator is the median of all pairwise means and is robust to outliers. It is better than a mean because the mean is greatly influenced by outliers.
The tiers are determined by using a machine learning clustering algorithm called a gaussian mixture model. Thank you to Boris for introducing me to the idea.
The upper and lower limits of the error bar represent 1 standard deviation away from the Hodges-Lehmann median.