Recency Bias in Dynasty Fantasy Football (2024)

With the NFL Combine in the books and the NFL Draft looming, now is a great time to look at dynasty fantasy football rankings. It’s also a time when we may fall victim to recency bias.

Recency bias is the over-favoring of recent information over older information. Common examples include down-ranking players who spent much of the previous season injured, assuming players will repeat last season’s career highs and fading rookies who didn’t break out.

Let’s look at a few years of preseason dynasty rankings to determine how heavily recency bias plays a factor in expert rankings.

Recency Bias in Dynasty Fantasy Football

At the risk of baking recency bias in our analysis, we’ll focus on the 2020 consensus preseason dynasty ranks. 2020 is the most recent year with four full seasons of data to follow. The basic framework for evaluating the 2020 dynasty ranks will be to see if they would have been more accurate with more consideration for recent information (coming off 2019), or more accurate with less consideration for 2019.

To approximate player success from 2020 to now, rather than take points scored for each season, we’ll use preseason half-PPR ranks (roughly converted to fantasy wins) for each player each year. This might sound odd but it works fairly well at measuring success: If you have a bunch of early-round full-season picks on your dynasty team each year, you know you have a good dynasty team. Plus, this method equalizes across positions and takes out some undue injury luck.

Recency Bias Check #1 – The Control Group

The first way we check for bias is by seeing if we can build a linear regression model from full-season fantasy rankings in the few years leading up to 2020 that beats the 2020 dynasty rankings.

To do this, we need a set of players ranked every preseason from 2017-2020 (52 players made the cut). Then we use Excel, or any statistical tool, to make two linear regression models — one using just 2020 dynasty ranks and one using full season ranks from 2017-2020. The model with the higher correlation with fantasy wins is theoretically the more accurate model.

The results? Dynasty ranks aren’t looking too bad…

Model	Target	Calculated Regression	Correlation
2020 Dynasty Ranks	Expected Fantasy Wins, 2020-2024	1.49 – dynastyrank * 0.011	0.640
2017-2020 Full Season Ranks	Expected Fantasy Wins, 2020-2024	1.33 – 2020rank * 0.016 – 2019rank * 0.000 – 2018rank * 0.0004 + 2017rank * 0.007	0.641

The way to read the calculation here is that for dynasty ranks the model says you can get fantasy wins over 2020 to now by taking the dynasty rank in 2020 (e.g. No. 1 overall), multiplying it by 0.011, and adding it to 1.49 wins. Said model would be about 64% accurate.

Meanwhile, if you took the four years of full-season ranks, you can barely get the correlation any higher. This is a sign the information we had from 2017-2020 was being properly incorporated into the 2020 dynasty ranks.

Recency Bias Check #2 – Combining Forces

Another way to look at recency bias is to run a regression where you try to predict long-term success using only two variables — the dynasty ranks and the regular full-season ranks for the same year.

Here’s how this one works. Full-season ranks need to weigh recency more highly than dynasty ranks. This is because recent injuries may linger into the upcoming season, most coaching staffs stay intact, most players don’t change teams, etc. Thus, full-season ranks are a decent proxy for “recency” in information.

Our regression model will start with dynasty ranks and use full-season ranks to try and become more accurate. This will tell us if the dynasty ranks need more or less recency bias.

If boosting highly ranked full-season players improves ranking accuracy, then it means recency was not weighed heavily enough. If down-ranking the highly-ranked full-season players helps accuracy, it means recency was weighed too much.

Instead of showing the full regression equations here, you will see the full season ranks weighting as a percentage of the dynasty ranks weighting. Bigger numbers mean full season ranks were used more, negative means the model had to reduce recency bias, and positive means the model had to increase recency bias.

Ranking Year	Prediction Window	Full Season Adjustment
2020	2020-2024	-10%
2020	2021-2024	-16%
2019	2019-2024	-2%
2019	2020-2024	-9%

Was there recency bias in the 2020 dynasty ranks? A little bit. But the bias was miniscule. A full season adjustment of -10% means a player ranked 10 spots higher than another in full-season ranks should maybe get docked one spot if you’re comparing their dynasty ranks. This is to account for the bias the full-season ranks have on dynasty.

Never Tell Me The Odds

If this is too much math for your blood I applaud you for making it this far. I have the antidote, though. If you want to see who I think would win in a game of football between actual bears and lions, you can watch this video that involves way less math.