Mis-seedings may foretell tourney madness better than KenPom curve

Armed with the last 11 years of Selection Sunday data based on KenPom’s new mismatch-free formula, I’ve been doing some analysis into what I call the “quality curve” of the 52 teams. The more I look at the numbers, the more convinced I am that, while the relative quality of the tourney field has some impact on whether the dance will play out chalky or chaotic, it’s the mis-seedings that have a bigger influence on tourney madness.

If teams were seeded strictly according to Pythag, the first four teams on the KenPom list would be one seeds, the next four would be two seeds, and so on. The top 52 teams would thus occupy the top 13 seed positions. (I’m ignoring the longshot 14-16 seeds for this analysis.) I ranked the Pythag values of the top 52 teams. Then I calculated the averages in 13 four-team groupings for the 2007 tourney, when the Madometer read an all-time low 4.1% deviation of higher-seed supremacy, and the 2013 tourney, when the dance reached record 20.8% madness. Finally, I did the same calculations for today’s top 52 KenPom teams. Then I plotted the results on a line chart. Here’s what it looked like:

Top_52_Curve

If the Selection Committee had used KenPom data to seed the 2007 and 2013 tourneys, and did the same for this year, the three quality curves wouldn’t be dramatically different. There are, however, some differences that might’ve foretold the chalkiness of 2007 and the chaos of 2013. You may have to squint at the lines above, but this is what I see:

  • 2007 (in blue) would have slightly weaker one seeds than 2013 (in orange), stronger two seeds, markedly weaker three and four seeds, and similar five and six seed.. You might conclude from this that the one and two seeds would have less trouble in Sweet 16 matchups against their most likely seed rivals. In fact, the top two seeds were 7-0 in 2007 and 2-3 in 2013.
  • While five and six seeds would almost be the same, 2007 would have stronger 11 seeds and slightly better 12 seeds. This might be expected to lead to more 5v12 and 6v11 upsets in 2007. But it didn’t work out that way. In those matchups, the higher seed was 6-2 in 2007 and 4-4 in 2013. We’ll get an idea why in a minute.
  • Seven through 10 seeds would be generally weaker in 2007, making second round matchups versus one and two seeds easier. Indeed, one and two seeds were 7-1 in round two of the 2007 dance, and 5-1 against those same seeds last year.

Of course, the teams weren’t seeded according to Pythag in 2007 and 2013—and they won’t be this year either. Instead “mis-seedings” contributed to easier paths for the highers seeds in 2007 and tougher upset matchups in 2013. Take a look at this chart comparing the average Pythag values of the top 13 seeds in 2007 with their 2013 counterparts:

Top_13seed_Curve

The way the seeding shook out, the top seeds in 2007 (in blue) were actually a touch stronger than the 2013 top seeds (in orange), as were the two seeds. And the three, four and five seeds were all historically weak—much, much weaker than average (in white). That paved the way for seven one and two seeds and a single three seed in the 2007 Elite Eight.

As for the 4v13, 5v12 and 6v11 matchups in the two years, while it’s true that fours and fives were weak in 2007, so were 12 and 13 seeds. And six seeds were stronger than average that year, while 11 seeds were considerably weaker. The result: the higher seeds were 10-2 in the first round of the 2007 tourney.

Now let’s look at 2013. Notice how much stronger three, four and five seeds are compared to 2007? This might explain why we saw four teams with these seeds in the Elite Eight last year. But there were also a good number of 4v13, 5v12 and 6v11 upsets in 2013. The higher seeds were just 7-5 in these matchups. That might be because, the way the seeding fell out, 11 and 12 seeds were remarkably strong. You got a hint that this might be the case in the “top-52” curve above, the seed selections amplified the strength of these seeds.

So what does all this have to do with 2014? The key takeaway is that, while we might be able to make some generalizations about the relative madness of the tourney from the top-52 curve (the first chart), we really have to wait and see where the Selection Committee slots the teams into the bracket.

If I look at the red line in the first chart (that shows today’s top 52 KenPom teams in 13 four-team averages), I see these things:

  • Top seeds are much weaker than either 2007 or 2013, two and three seeds are about the same as last year, and four seeds are nearly as bad as the 2007 crew. This says to me that, while the top seeds may skate by soft four and five seeds, they could have trouble against two and three seeds.
  • 11, 12 and 13 seeds are all weaker than 2007 and 2013. And their higher seed opponents look at least average. Does this mean that we won’t have as many 4v13, 5v12 and 6v11 shockers as we had last year?
  • Six, seven and eight seeds are all slightly stronger than the other two tourneys. Does that spell trouble for the higher seeds in round two?

None of these questions can really be answered until the teams get seeded and we have the kind of view you see in the second chart. I’ll say one thing though: bracketologists Joe Lunardi and Jerry Palm have Arizona, Florida and Wichita State as their top seeds, with a disagreement between Kansas (Joe) and Villanova (Jerry). Those are the first, third, sixth and either seventh or eight best Pythag teams. So…it’s likely that the actual top seeds will be much weaker than the already weak “optimal” top seeds.

And this is frightening.

This entry was posted in Measuring Madness. Bookmark the permalink.

14 Responses to Mis-seedings may foretell tourney madness better than KenPom curve

  1. Gary Diny says:

    Pete,

    Great insight as always. Do you have any thoughts/analysis on 11-14 seeds that pull off upsets that are under seeded based on the efficiency data (for the years that the data exists)?? Similarly for higher seeds that get upset? Is there an efficiency gap between the competing teams that is too narrow? Efficiency numbers that reflect a grossly (more than 1 seed line) over or under seeded team that was involved in an upset?

    Thanks

    Gary

  2. Justin says:

    Pete,

    Can you add the data for 2014 on Monday after the NCAA bracket is announced?

    – Justin

  3. larry k says:

    Identifying ‘inflated’ and ‘undervalued’ seeds has always been the key to bracket success… unfortunately, until peter and ken pom have come along, it was not easy to quantify the values that were the foundation of seed ratings… what i had done in an earlier life was, just before selection sunday, sit down and value the teams myself from 1 to n … usually not caring much past 40 …
    i think the analysis here is fantastic…. as someone else has already mentioned though, i think we could find even greater value if somehow we could ‘weight’ the data based upon time of occurrence, as i do believe that recency of the numbers might add value to the final concensus…

    perhaps one way of doing this would be to calculate the values on a running basis, say weekly…. then we could see if the last 3 or 4 weeks data is showing an upward or downward trend… maybe like stock market technical analysis… :-)

    good work pete,
    larry k

    • ptiernan says:

      The only real “recency” stat I track is wins in last 10…and that does have some impact on tourney performance.

  4. Blazers says:

    Remember, the selection commitee are not to be trusted. Back in 2012, Missouri should’ve been a number 1 seed instead of Michigan State. And we know what happened then (Norfolk St. would’ve won by more points had they hit their free throws). Then there was last year Wisconsin seeded 5 for a 23-11 record, Oregon underseeded, and the list goes on. There is a good trend of powerhouse team winning their conference tourney and advancing to at least the round of 32 (except Missouri of 2012, though I have a suspicion that the Big 12 gave it to them because they are leaving to the SEC). This March, expect multiple eyebrow raise in terms of team seeding, especially the number 1 seed. It could be another 15 seed over 2 seed or 14 over 3 and maybe this is the year when a 16 defeat a 1 seed.

    • ptiernan says:

      No argument on the potential craziness of the seedings. A lot of really good teams bunched together…but few great teams.

    • Tom says:

      Really, a lot of the strength of the bottom quadrant of the tournament depends on the vagaries of one-bid conference tournaments. A bunch of upsets could lead to an unusually weak bottom quadrant.

  5. Blazers says:

    Let’s update potential low seed capable of pulling an upset -> Mercer, perhaps FGCU again, Manhattan, Green Bay, UC Santa Barbara and maybe even UC Irvine. Georgia State may be one of those seeded 15-16 that may scare us (they have Ryan Harrow transfered from Kentucky of last season on that squad).

    Alert: **** Southern U of the SWAC is out ***** The NCAA suspend them and they won’t come into tourney time.

    • Andy says:

      Watch out for Harvard too after they clinched an auto bid. They returned everyone from the team that knocked out New Mexico last year, their numbers look a lot better than last year, and they have some major momentum going for them. Let’s pretend they are a 12 seed. Obviously, they were in last year’s tournament. But from my math, I think their frontcourt contributes to over 50% of their points.

      Then again, there is the point of being seeded off from all the predictions. This is such a case where I think it may very well happen. Call it a hunch, but I think the committee is going look at Harvard as only losing four games (barring a loss tomorrow, which I think is doubtful) and what they did last year, and reward them with a 10 or 11 seed instead. My only slight concern is they average 72.9 points per game, just missing the 73-point threshold. However, there is a lot of mediocrity out there, and Tommy Amaker in three tournament appearances has overachieved twice, both times rather impressively. I mentioned last year, but he also took a 10-seed Seton Hall team in 2000 that missed the tournament the year before to the Sweet 16. Call me crazy, but if they get just the right draw, they could go quite far, maybe being this year’s Cinderella team. If not, they can still likely pull off at least one upset, if not two.

      • Blazers! says:

        Thank you Andy. You shed some light to why 3 NM lost to them (aside from them being in the Mountain West). I never took the time to do some coach background and never knew about Amaker. I forgot to put Harvard on the list and they will do some damage (they reinstated the two players that they suspended a year ago back in addition to returning key players).

        I have a hard time believing Louisville is going back to the Final Four because of the “champion lingering effect” as I call. Duke of 2011 could’ve repeat (the addition of Kyre Irving) but they just disappear at the second half of that Arizona game. From my previous post, I mentioned about back-to-back Final Four appearance and I am going with Witchita State (they won’t be the number 1 overall seed, they may be a third or fourth number 1 seed; Florida and Arizona are frontrunners for that spot). Watch out for the selection committtee intentionally putting bad teams in the same region so WSU can literally walk to the Final Four.

        • Tommy says:

          Starting to realize my Jayhawks might have issues. According to the Contender/Pretender section, low scoring margin makes for a No. 1 seed flag, more than one freshman makes for a No. 2 seed flag. And defense would not qualify KU as a contender according to the kenpom test at the moment.

  6. Justin says:

    2014 #s..10’s look low, 7’s look strong.

    1 0.9458
    2 0.9060
    3 0.8983
    4 0.9056
    5 0.8625
    6 0.8456
    7 0.8480
    8 0.8293
    9 0.8420
    10 0.8011
    11 0.8301
    12 0.7859
    13 0.7139
    14 0.6514
    15 0.5439
    16 0.4246

Leave a Reply

Your email address will not be published. Required fields are marked *

* Copy This Password *

* Type Or Paste Password Here *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>