Daniel Wilco | NCAA.com | March 11, 2018

This ranking is the best single-predictor of NCAA success

High Five: 15 over 2 Tournament Upsets

Seed. AP Rank. RPI. The big three. When it comes to March, few statistics or rankings are more well-known and all-encompassing than those. But which should you fall back on when filling out your bracket? Which is the best predictor of March Madness success?

MARCH MADNESS SHOP

To find out, we looked at the past eight years of data for the three rankings (Seeds 1-16, AP rank 1-25, and RPI rank 1-15), and compared it to NCAA tournament wins.

Here’s how the three rankings stack up, comparing rankings to average wins per tournament.

	Seed	AP rank	RPI rank
1	3.13	2.75	2.75
2	2.22	2.75	2.50
3	1.66	3.13	3.00
4	1.84	2.75	3.25
5	1.00	2.25	2.63
6	0.72	3.50	3.38
7	1.13	3.00	2.00
8	0.97	0.88	2.38
9	0.47	2.38	1.75
10	0.59	1.60	1.13
11	0.97	2.50	1.38
12	0.50	1.13	1.00
13	0.22	1.63	0.71
14	0.19	1.38	2.88
15	0.16	2.00	1.00
16	0.00	1.57	N/A*
17	N/A	1.50	N/A*
18	N/A	1.25	N/A*
19	N/A	1.38	N/A*
20	N/A	1.13	N/A*
21	N/A	0.88	N/A*
22	N/A	0.88	N/A*
23	N/A	1.25	N/A*
24	N/A	0.29	N/A*
25	N/A	1.38	N/A*

*While RPI ranks all 351 teams, we only looked at the Top 15 for this story.

The next step is to determine which of these is the best predictor. We can do that by using a linear regression to model the association between each ranking and the number of NCAA tournament wins. Here’s how each ranking’s linear regression looks:

Next, we look at the coefficient of determination — also known as R-squared — for each. Put simply, the R-squared value is a percentage between 0 and 100 that tells how much of the variance of the data is predicted by the linear regression model. A perfect R-squared value is 1.0 (or 100 percent), while the worst possible is 0.0 (0 percent).

Rank	R-sqaured value
Seed	0.80
AP	0.64
RPI	0.48

So, we have our answer. The best predictor of March Madness success, unsurprisingly, is a team’s seed. If you had to choose just one number to use to fill out your bracket, you’d be hard pressed to find a better one than seed. Which begs the question, how would your bracket do if you filled it out just by picking the better seed? We answer that here.