During last year’s dance, I fiddled around with an Excel model that order the teams by 10 key performance factors…then apply some level of “chaos” to the numbers as a way of simulating the crazy stuff that can happen in the tourney.

This was inspired by a couple circumstances. First, I’ve got a bartender friend named Josh (he’ll never read this!) who asks me every tourney whether I’ve created an “automatic pick” bracket. Secondly, many of the stats models spit out the same picks…and they tend to bias my personal choices. By adding the element of chaos, I get out of some bracket picking ruts that I would otherwise remain in.

So the Brackomatic Excel sheet solves both those problems. It gives Josh the automatic picks he needs (sorry…it doesn’t plop them into a bracket) while enforcing a certain amount of randomness to the teams’ fates.

How much randomness? There’s the rub. Based on some of the work I did to normalize the impact of coaching against team performance, I’ve weighted overall Pythag efficiency to be 67% of the equation, coaching to be 11%–and chaos to have a 22% impact on outcomes. The first two figures are based to an extent on real numbers. The third is a guess. You’re welcome to adjust the Excel model up or down to accommodate your impression on the role of chaos in the tourney.

Using the Brack-o-Matic is easy. Insiders can find it in the TIPS+ section. Just open the Excel file and filter by the “Prospect” column. The teams will be rank ordered from the champion on down. If you filter again, a new order of teams is created. Pay no attention to the final number in the Prospect column, since once you okay the filter, it generates a new value in the Chaos column.

When I publish the new wave of bracket models, you’ll see a new model, #13, called Brack-o-Matic. It’s based on the first time I filtered on this Excel sheet.

Would it be worth simulating this a few thousand (or more) times to see what average ranking each team gets over a large sample size? Not sure if this is even possible in excel.

Yes, a true Monte Carlo simulation is really the way to go here…and the results would validate the formula. That’s a project for the off season. I know that ESPN uses simulation models for their game analyses–not just the tourney, but other games as well.

I consider myself pretttty handy with Excel (maybe I shouldn’t…), but can you explain the formula used to calculate the “Coach” column, and how to raise/lower the 22% “Chaos” weighting?

I haven’t commented on any of your posts yet, but I wanted to take this time to thank you for all of the work you’ve done. As this is my first year, I wasn’t sure what all to expect, so I ended up manually building the brackets based on the excel data you provided in the wee hours of Sunday night. Little did I know how much of this you would end up doing “for us”!

As an excel jockey, I’ve been playing around building models of all kinds throughout the whole season using your insights. I know a lot of people here plan on splitting the $1B bracket money with you when they win, but I’m not going to go so far as to guarantee that I will be able to do the same. However, with all of the money I’m going to make from my annual Calcutta auction, I can guarantee you’ll have a customer for life.

(By the way, you reeeally need to get into a Calcutta auction group if you’re not already…the thrills of auction modeling with real money are way better than just a “simple” bracket!)

Raising the chaos factor is easy. Just randomize numbers between 0 whatever. I did .200. As for coaching, it’s a long explanation. But the possible range of PASE values is -3.35 in the case of a 1 seed losing in round 1…and 6.00 for a 16 seed winning it all. That’s a 9.35 gap, and I take the percentage of where a coach falls on that range. That formula is: (x+3.35)/9, then to normalize with Pythag I divide by 2. That divisible is the SWAG, so play with that if you want. Without the divisor, coaching is heavily weighted…and Shaka Smart comes up often.

I’m all about the SWAG.

Clearly my math skills are dulling after 2 days of nonstop crunching…I kept looking for some weighting of .22 to show up in a formula, not factoring in how averaging the three columns and the estimating would provide the range of weighting itself.

Peter- The office pool that I participate in gives a seed differential bonus with no multiplier. I have never played with a seed differential bonus before and was wondering if you have ever created any models that factor this in. Obviously this increases the payoff for picking upsets so more should be selected but I am struggling to weigh the payoff versus the risk being taken by selecting a lower seed. Do you have any advice?

Hi Mark – Not sure exactly what the rules are. How many points do you get for picking a 2v15 right if it’s a two? How many if you’re right on a 15? The short answer is that I haven’t done any rigorous analysis into this. But perhaps you could multiply the differential points by the average wins per seed to find out which matchups are the sweet spot.

As someone had mentioned, I threw this model into my @Risk module to Monte Carlo the lights out of it.

E8: Florida, Syracuse, Michigan St., Villanova, Arizona, Wisconsin, Louisville, Michigan. (2 #1′s, 3 #2′s, 1 #3, 2 #4′s)

F4: Florida, Michigan St., Arizona, Louisville (2 #1′s, 2 #4′s)

Finals: Florida, Arizona (2 #1′s)

Champ: Arizona (#1)

Incidentally, Louisville was a hair below Arizona in average score, but would technically lose to them based on the regional matchup, which thrust the model’s #3, Florida, into the title game. Also, the top 4 seeds in the MW region finished in the top 8 of the model’s outputs…as if anyone didn’t already know that region was a beast.

I think people could use this data (as with plain ol’ Pythag) to determine regional strength and disparity. At the very least it could help someone determine where to “fit” an upset if they needed one, or whether or not eliminating a favorite in an early round is likely to really hurt them down the line.

Sorry Al.

Sorry, I was not very clear. The scoring the group creator chose is fibonacci (2,3,5,8,13,21), which already favors the lower rounds. The seed differential bonus is added to the points you get for a W. For example, if a 15 beats a 2, you get 2 points for the win and then 15-2=13 additional points for the seed differential bonus. If the 2 seed wins, you just get 2 points. You can really rack up points with upsets. So let’s say according to a prediction model, a 13 seed has a 35% chance of winning versus the 6 seed. If the 6 seed wins I get 2 points, if the 13 seed wins I get 9 points. If I multiply the odds, on average I will get 1.3 points for choosing the 6 seed and 3.15 points for choosing the 13 seed. Is that how you would approach it? The hard part is considering the potentially lost points for later rounds. And I’m not sure how to calculate the odds of let’s say a Wichita State vs. Kentucky matchup in round 2. Anyway, thank you for the feedback, it is much appreciated.

Just realized that it should be 13 vs. 4 (or 11 vs. 6). Sorry about that.

Mark,

If you’ll shoot me your email, I will be happy to provide you with some shell excel analysis that might help you build a model for this. It’s basically KP’s pythag figures (I also have Nate Silver and Sagarin’s numbers) shoveled into a log5 matrix (what KP has on his website to show % chance of making each round). From there, you ought to be able to apply some formulas to score what each pick would be worth.

You’re on the right track quantitatively, but now you need to apply those percent chances of winning out to each round, to come up with an average score.

Also, you’d have to wait on me until tomorrow morning, as I’m hosting my Calcutta auction tonight.

Hey Ryan- thanks for the help man. Definitely don’t want you putting any time into helping me but if you already have a shell of something to work with then that would be great, as I don’t really know how to set it all up. You can use this email that I created just for fantasy: markw20055 @ yahoo.com

I have no idea how to use this to generate a bracket.

You’re talking about the Brack-o-Matic? It just force ranks the teams. Then you fill in your bracket with those teams in order.

I wasn’t sure where to put this information this thread seems like the best one for this.

I am old school when it comes to following Pete’s champ criteria. This has worked every single year going back to 2002 I believe. These are the 13 rules I follow.

Earned a one, two or three seed

Come from a Power conference (ACC, Big East, Big Ten, Big 12, Pac-10 or SEC)

Been led by a coach with more than five tourney trips and at least one Elite Eight run

Either went to the previous yearâ€™s dance or had an All-American

Averaged more than 73 points per game

Allowed fewer than 73 points per game

Owned an average scoring margin of at least seven points per game.

Played a schedule among the 75 strongest in the country

Had at least one junior or senior starter

Counted on starters for at least 73 percent of their points

Relied on their frontcourt for more than 35 percent of their points

Posted at least a .667 road winning percentage

Won at least six of their last 10 pre-tourney games without a losing streak of two or more.

I found 2 teams that pass these rules – Arizona and Wisconsin. I’ve double and triple checked to make sure. If you see I made an error definitely yell out. I just wanted to share this and it seems a thread about building the bracket is a good place for it.

Mark,

I sent you a couple of emails. Enjoy.