(12-02-2019 09:27 PM)FadedCrown Wrote: Don't get the hard on for RPI here.
If the basis of your remark is that
the RPI isn't rating the American as highly as the Massey Composite is (6 vs 8 top 100 American teams), then
that's an observable fact.
That's why so many different rating systems are included in the initial post above.
The RPI ratings continue to be of some interest to many readers because they are familiar and understandable to most fans (e.g., are highlighted in the CBS Sports conference standings page), are based on a fully transparent mathematical formula that
permits comparisons across conferences, sports (FB & BB), seasons and decades, and are
designed to reflect each team's strength of schedule.
--If the RPI were no longer of interest, sites such as realtimerpi would no longer get the numbers of hits that they continue to receive.
In addition,
like the NET ratings (after January), the RPI ratings provide a useful contrast with the Massey Composite ratings, because they highlight the impact of strength of schedule on W-L records, margins of victory, etc.
The Massey Composite is perhaps the most reliable indicator because it integrates data from 26 different rating systems. However, the NET and RPI are both of special interest because they aren't factored into the Massey Composite, and
place highly contrasting emphases on scoring (NET) or SOS (RPI).
Later in the season, the discrepancies between Massey Composite, the NET, and the RPI will be of increasing interest, [u]especially among critics of the NET (including selection committee members, who used SOS over NET in picking the "last 8 teams in" last March).
[/u]--The Massey Composite rankings are likely to fall between the NET rankings (emphasizing scoring regardless of SOS) and the RPI rankings (emphasizing SOS over scoring).
Of further interest,
the RPI rankings are very informative when compared to the later-season NET rankings, which have been strongly criticized for over-emphasizing scoring "efficiency" (FG%, FT%, and points per minute) - - a mathematical proxy that, by design, offsets (by overwhelming) the misleading "10 pt. scoring margin limitation" in the NET formula.
--The basis of this criticism has been made crystal clear as a result of the fact that
some teams' 2018-19 NET ratings were boosted by as many as 80 rankings (e.g., from #110 to #30) because they hit a very high percentage of their shots and scored a huge number of points in numerous games against bottom 300 teams.
The RPI, which factors in strong SOS, as well as W-L records, and which
is not strongly influenced by scoring margin or its highly-correlated proxy variable known as "scoring efficiency."
From this standpoint, when February approaches,
the RPI is a useful corrective to the NET and similar rankings which emphasize scoring and winning percentages over strength of schedule.
The selection committee members are and will be continuing to monitor the discrepancies between the NET and the other well recognized rating systems. It is most likely that
they will, once again, use alternative ratings to balance out the imperfections of the NET formula, and that the formula will eventually have to be revised to include a much stronger adjustment for strength of schedule.
Fans of the game can apply the same logic in Jan/Feb/March
by computing the mean (average) of the NET, Massey Composite, and RPI ratings.
--The NET, due to being stipulated by the NCAA as their key indicator, the Massey Composite due to being among the most reliable indicators by combining many (i.e., 26) different predictive ratings, and
the RPI being included to correct the NET data with a stronger indicator of strength of schedule.
It is a working hypothesis, at this point, that the mean of NET, RPI, and Massey Composite rankings - - by virtue of incorporating the strengths of the three different rating systems, while reducing their discrepancies to statistical "noise" (i.e., "error variance") - -
may be a substantially better predictor than the NET, the RPI, and most other
individual rating systems.
--
Whether or not the equally-weighted NET+RPI+Massey Composite Rankings will turn out to out-perform or under-perform the Massey Composite itself, as a predictor of late season team performances will remain to be seen.
--Until then,
the "law of large numbers" suggests that the Massey Composite will tend to be a better and more reliable predictor of team performances than any of the non-composite ranking systems, especially if the 26 systems in the Massey Composite were selected on the basis of a well-informed methodological and statistical rationale.