The virtual ink wasn't dry on Ratings Trends before the obvious extension to performing the same analysis for every rating that I'd done for the Majority Consensus came to mind. It takes a few minutes on a six year-old computer but it's worth it because as a byproduct there's a convenient report of team rank trends for each of the ratings Dr. Massey includes in his summary. In that report the team name links to the schedule page that includes team and opponents' Majority Consensus ranks.
Presenting the temporal correlations for 100+ ratings is something of a challenge. My first cut included all ratings in one very large (about 6MB) file with an index at the top that just shows the weeks for which I have data.I changed it to reference one page per ranking, which is slower to upload, but much faster to download (you're welcome.)
The alternate index pesents the distance between successive rankings for each rating.
The measure of week-to-week variance listed in the reports is the distance between week N and week N+1. This is defined as the total number of team-pair swaps it takes to turn one ranking into the other.
- The team reports show the number of swaps involving the team. This is not the same as the team's rank difference: if five teams "jumped" the team and two teams dropped below the team from above, the swap-count for the team is 7, even though the rank difference for that team is at most 3 spots.
- The distance between a ranking's week N and week N+1 values is half the sum of the distance contributions by teams, since a swap is counted once for each team involved.
- For rankings that allow ties, if a team-pair is tied in either week N or N+1 it is not counted as a swap even if they are different in the other week.
- Since there are ½ × 128 × ( 128 − 1 ) = 8,128 team-pairs, the scale for distance is [ 0 , 8,128 ], ranging from an unchanged list to a reversed list. Lower numbers mean the rankings are more similar.