APBRmetrics

[RobbieStats]

I'd like to measure the volatility between two college basketball polls (e.g., last week's AP poll vs this week's poll, or the preseason AP poll vs this week's poll). Has anyone heard of such a thing?

The formula I came up with is straightforward:

RV = total position changes / total possible position changes

The numerator is calculated by comparing the number of position changes between two polls. Duke going from 8 to 10 counts as 2 position changes. Dayton going from unranked to 20 would count as 6 position changes. etc.

For the denominator, I need to calculate the total number of changes that would result from all of the teams in poll1 being replaced by new teams in poll2.

For a top X poll, I came up with this: (summation from n=1..X) * 2
So the AP top 25 poll would be this: (summation from n=1..25) * 2 = 650

For something like the RPI where all teams are always inclusive, the denominator would be: (summation from n=1..X)
where X is the total number of teams (341 in Division I men's college basketball this year). No need to multiply it times two since there couldn't be a brand new set of teams to replace the current ones.

Does this sound reasonable?

I have a graph of the AP Poll Volatility for this season in the bottom right corner of this page: http://statsheet.com/mcb/rankings/motion

Thanks,
Robbie Allen

[Mountain]

Sounds reasonable to me and I do think it could have interest and reveal things about team play quality and polling volatility over time. In addition to week to week, total change from pre-season might be another presentation of total accumulated volatility.