Show the Work
Methodology v2.0.0 · Inclusion and placement, measured separately.
Who’s In? Where Do They Rank?
These are two different questions. Ranked By shows the percentage and number of available voters who include a team. Average When Ranked shows where those voters place it. Unranked (NR) has no numeric rank.
If six of twelve voters rank a team #20 and six leave it out, we show 50% inclusion and an average of #20 among the six who ranked it. We do not turn the omissions into #26 or average them into #23.
Consensus and Points
Published AP ranks are authoritative for real imports. The demo calculates consensus from synthetic ballots: 25 points for first, down to one for 25th; omissions earn zero. Equal points share a rank and the next position is skipped. All teams tied at #25 remain included. This scoring rule does not assign a rank to omitted teams.
Placement Spread and Vote Ranges
Ranked variance = sum of (rank − ranked average)² ÷ ranked ballots
Ranked SD = √ranked variance
SD measures spread among voters who rank the team. A higher SD means less placement agreement within that group. It does not measure how many voters omit the team. If all six voters above choose #20, ranked SD is zero even though inclusion is split 50/50. With one ranked ballot SD is also zero; check the sample count.
Every vote range uses the same #1–#25 scale. The darker line spans numbered votes, and the dot shows their average. Omissions are shown separately. The histogram plots numbered votes; its exact-count table also includes NR.
Placement Differences and MAD
Placement MAD = sum of |deviation| ÷ teams ranked by both
We calculate a difference only when both the voter and consensus rank the team. Consensus #20 and a vote of #15 gives +5; consensus #12 and a vote of #18 gives −6. Positive means above consensus.
Placement MAD averages the absolute differences across shared teams. We show the number compared because it can differ between voters. Placement outlier lists use these numeric differences only. No shared teams means no placement score.
Added and Omitted Teams
Added means a voter ranks a team outside the consensus Top 25. Omitted means the voter leaves out a consensus Top 25 team. A swap usually counts as one addition and one omission. Ties at the consensus cutoff can make those counts unequal.
The inclusion leaderboard sorts by the total of additions and omissions, while showing each separately. Individual inclusion differences are listed by voter and team, with no invented rank distance. A low placement MAD can coexist with many inclusion differences: read both measures together.
No Single Chaos Score
We retired the former Chaos Percentile rather than blend placement and inclusion with arbitrary weights. Version 2 scores cannot be directly compared with the previous NR=26 calculations. Consensus ranks and points are unchanged.
Missing Data and Archive Averages
A missing ballot is unavailable data, not an unranked vote for every team. No available ballots means no inclusion percentage. If nobody ranks a team, inclusion is 0% and placement statistics are unavailable.
Archive placement MAD averages eligible weekly scores with equal weight and shows the scored ballot count. Average shared-team counts and additions/omissions use all available ballots. Changing comparison sets and incomplete archives limit comparisons across voters and seasons.
Agreement Is Not Accuracy
These statistics describe voting patterns, not journalists’ motives or team quality. Historical tendency labels require at least 10 distinct polls and verified metadata. Selection effects, repeated observations, schedule strength and multiple comparisons need further analysis. Consensus includes the voter being assessed; leave-one-out analysis is future work. Predictive scores remain deferred until a benchmark and evaluation protocol are defined.