RPI for 2008

Hahaha its almost as bad as "The Ohio State University"

Or, wrt tonight's action, The University of Texas.

I always carefully and conspicuously leave off the article when writing to colleagues there.

I probaly should not admit this. I have read all of this and the bigsoccer thread. I still do not understand the RPI.

Don't worry I dunked my toe in the dark pool that is known as the NCAA RPI and was soundly thrashed by the Pilot Nation hooligans wielding slide rules and pocket protectors.

Well, there is not only the Big 11, but the Atlantic 10 actually has 12 teams...

Haha at least the Pac-10 was honest when adding the Arizona schools back in the day...

Hey Auto Pilot - I think pocket protectors are neato, but what's a slide rule?

Anyone see where I left my abacus?

Now I have really dated myself. Before computers the smart kids used to carry slide rules around. Read about them

HERE

Try calculating RPI with that....

The results might be just as meaningful

fan from afar wrote:Hey Auto Pilot - I think pocket protectors are neato, but what's a slide rule?

you're kidding, right?




I just counted up--I have 7 different kinds. I still can use them all. oooooh- I just outed myself as a TRUE geek.

And I thought my TI-83 was out of date.... I mean it doesn't even have the built in USB connection...

Just kidding.

7 different kinds of slide rules, or pocket protectors? Slide rules are ok, but pocket protectors - Nerd Alert!!

And the UBERgeeks had these



Only two of us had 'em when I went to college. They cost about $200 in 1965. A Ford Mustang cost $2000 back then.

fan from afar wrote:Nerd Alert!!

I Prefer

Wow!! That thing is pretty phenomenal... And I'm not really a numbers guy either...

My aunt used to have one of these that I used to play with for hours when i was little:

DTR- your sig is my 2nd fave Mitch Hedberg quote behind this one.


An escalator can never break. It can only become stairs. You would never see an "Escalator Temporarily Out Of Order" sign, just "Escalator Temporarily Stairs. Sorry for the convenience."

-Mitch Hedberg

I know we're all engaging in what I once heard called "displacement activity," in anticipation of Friday's game. (The illustration I saw of displacement activity was the frantic activity of lab rats when put under stress -- don't take it personally, that's just the image that comes to mind.) So, my displacement activity has been analyzing the RPI in terms of how well it correlates with the actual results of this year's regular season games. I've come up with a pretty nifty system -- if you ask me -- for doing the analysis. I'll publish a lot about it in detail later. But here's an initial tidbit:

The NCAA's RPI matches the 2008 regular season game results, for all regions, 76.1% of the time. In other words, on average teams are losing 76.1% of the games the RPI says they should win; and are winning 76.1% of the games the RPI says they should lose.

However, for the West Region:

(1) The RPI matches the West Region teams' winning of games 81.6% of the time. In other words, on average West Region teams are losing 18.4% of the games the RPI says they should win (as compared to teams losing nationally 23.9% of the time); and

(2) The RPI matches West Region teams' losing of games 59.4% of the time In other words, on average West Region teams are winning 40.6% of games the RPI says they should be losing (as compared to teams losing nationally 23.9% percent of the time).

Here are two posts I entered recently on BigSoccer's RPI "analysis" thread, based on some analysis I did of how well the RPI's rankings for the top 120 teams correlate with their actual inter-conference results during the 2008 regular game season. Warning: It's long ....

Recently I took a look at how the RPI does when looking only at the top 120 teams, on a regional basis. What I wanted to find out was whether the RPI's rankings for the top 120 teams correlate better with the actual regular season results of those teams in inter-regional games than the rankings correlate with the actual results for all 320 teams. What I learned was that they do correlate better, although not a whole lot better. I also wanted to find out how the RPI did in ranking the top 120 teams, on a regional basis, to see if the RPI discriminates against some regions in favor of others, as it does when one looks at all 320 teams.

I first looked at the overall accuracy of the RPI as it correlated with actual results of the top 120 teams in inter-regional games. In the following table, the percentages indicate (1) the percentage of time the regular season game results matched the teams' relative RPIs, without regard to how great or small the differences in RPIs were; (2) the percentage of time games were tied even though one team had a higher RPI than the other; and (3) the percentage of time the game results were contrary to the teams' RPI rankings because the lower-ranked team won:

Unadjusted RPI, top 120 only:

RPI correct: 76.7%

RPI incorrect, tie: 10.9%

RPI incorrect, lower-ranked team won: 12.4%

Unadjusted RPI, all 320 teams:

74.0%

10.7%

15.3%

Massey, all 320 teams:

73.9%

10.7%

15.4%

Adjusted RPI, all 320 teams:

73.9%

10.7%

15.4%

Jones, all 320 teams:

73.3%

10.7%

15.9%

Non-Conference RPI, all 320 teams:

69.4%

10.7%

19.9%

What this shows is that the unadjusted RPI correlated better with regular season results for the top 120 teams in inter-regional games than it did when looking at all 320 teams: a 76.7% correlation, as compared to a 74.0% correlation. Presumably there would be similar differences for the other rating systems.

I next took a look at how the numbers for the top 120 broke down on a regional basis. First, here's some more general regional information that's interesting at least to me. When I looked at the top 120 teams, based on the adjusted RPI rankings for 2008, here's how the teams split by region:

Central Region: 19 teams (of 57 total), or 33% of the region's teams.

Great Lakes: 23 teams (of 60), or 38%.

MidAtlantic: 14 teams (of 48), or 29%.

Northeast: 16 teams (of 41), or 39%.

Southeast: 25 teams (of 61), or 41%.

West: 23 teams (of 51), or 45%.

The dispersal of the top 120 teams among the regions, in relation to the number of teams from each region, suggests that the strengths of the regions are in the order of (1) West, (2) Southeast, (3) Northeast, (4) Great Lakes, (5) Central, (6) MidAtlantic. But, that does not take into account how well the actual regular season results of the top 120 teams in inter-regional games correlate with their RPI rankings. Here's how the six regions' actual regular season results in inter-regional games correlated with their RPI rankings. As with information provided earlier on this thread, a region's 100% result would indicate that it's teams, in games in which they had the higher RPI, won the % of such games that the average top 120 team won and, in games in which they had the lower RPI, lost the % of such games that the average top 120 team lost. If a region's percentage is above 100%, it means that the region's teams did better overall than the "average" team would have done. If a region's percentage is below 100%, it means that the region's teams did more poorly overall than the "average" team would have done. Put differently if a region's percentage is above 100%, then the RPI on average under-rates the region's teams. If a region's percentage is below 100%, then the RPI on average over-rates the region's teams.

Here is how inter-regional regular season results correlate with teams' RPIs, on a region-by-region basis:

MidAtlantic: 85.8%
Central: 95.6%
Northeast: 97.4%
Great Lakes: 105.6%
West: 106.1%
Southeast: 111.3%

What this appears to demonstrate is that when looking only at the top 120 ranked teams, the RPI underrates the Southeast, West, and Great Lakes regions' teams; slightly overrates the Northeast region's teams; overrates the Central region's teams; and significantly overrates the MidAtlantic region's teams.

Part of my study work has been to see if there is some explanation for these apparent overratings and underratings in how the different regions' matchups occur in inter-regional games. For example, are some regions playing inter-regional games where the opposing teams have closer RPIs than other regions? My conclusion so far has been that how the matchups occur does not provide an explanation: the matchup patterns are not particularly different from region to region.

Here is an additional note:

In theory, if Region A is stronger on average than Region B, one would expect the RPI to underrate Region A and overrate Region B. This is because, absent inter-regional games, the RPI would rate the teams in Region A exactly the same as it rates the teams in Region B. (This would be true for Massey and Jones, too.) This tendency of the RPI to rate two regions the same is moderated only if there are inter-regional games; and the moderation only is to the extent of the number of inter-regional games. I've demonstrated elsewhere that there are not enough inter-regional games in Division I Women's Soccer to fully overcome the tendency of the RPI to rate the different regions the same.

If we assume that the percentage of a region's teams in the RPI's top 120 is a beginning point for evaluating that region's overall strength, then as suggested in the previous post, the regions' average strength rankings are as follows, from strongest to weakest:

West (45%)
Southeast (41%)
Northeast (39%)
Great Lakes (38%)
Central (33%)
MidAtlantic (29%)

In theory, what one then would expect is that the RPI underrates the top teams on this list and overrates the bottom teams.

In fact, the extent of underrating and overrating comes close to following the above strength rankings, although not exactly matching them. The numbers are as follows, with 100% being the average region's performance in relation to its teams' RPIs, anything over 100% meaning the RPI underrates the region's teams, and anything under 100% meaning the RPI overrates the region's teams:

Southeast (111.3%)
West (106.1%)
Great Lakes (105.6%)
Northeast (97.4%)
Central (95.6%)
MidAtlantic (85.8%)

These two lists' orderings of teams are close enough to be consistent with what one theoretically would expect. In other words, they appear to be further confirmation that the RPI underrates strong regions and overrates weak regions. By being restricted to the top 120 RPI-rated teams, however, they deal more directly with the question of the ability of the RPI to properly rank those teams that really are the ones under consideration in the NCAA Tournament at large selection and seeding process.

In relation to the actual strengths of the regions, so far as the regions' teams in the top 120, these numbers lead to the following conclusions:

1. Not only are the Southeast and West Regions the strongest regions as rated by the RPI, but they are even stronger than the RPI indicates. I think it's hard, using the above numbers, to say which region is stronger.

2. The next strongest region probably is the Great Lakes Region, followed by the Northeast Region.

3. The Central Region not only is weaker than those four regions as rated by the RPI, but is even weaker than the RPI indicates.

4. The MidAtlantic Region not only is significantly weaker than any of the other five regions as rated by the RPI, but is much weaker than even the RPI indicates.