I'll be in Tucson on Tuesday. You can click on the "Email Me" link to send me their schedule.

I meant to get back to this sooner... Total CONCACAF coefficient is Country Coefficient + Club Coefficient. Country coefficient is the average performance of all teams in CONCACAF club competition over the past five seasons, and Club coefficient is the performance of the specific club in the Champions League proper (group stage and beyond). It's the same concept as UEFA's coefficients. Thing to remember is that UEFA view their coefficient system less as a ranking of teams and more as a way of seeding top teams and determining who goes into which bowl for group stage drawings. I attempt to do the same thing here.

I was most surprised by the result in Grenada. But beside Rocastle and Joseph, they don't have much else (domestic league, lower reaches of English footy). Still, I'm shocked. And Nicaragua just scored in Dominica, so the hits keep on coming!

Yes. They're in the "Soccer Pythagorean: Theory" category. You should find a link to the journal paper I wrote and an explanation of the Pythagorean in more accessible language.

The SQL code is currently written for PostgreSQL (MySQL does auto-increment of primary key IDs differently), and the code in the application assumes a PostgreSQL driver. I decided on PostgreSQL instead of MySQL because of its better handling of relational databases (especially views), but now that I have become more aware of MySQL 5 I'm no longer as wedded to Postgres as I once was. I need to sit down with my SQL code and figure out how much work it will require to write a MySQL or SQLite equivalent. Hopefully it won't be much, but I imagine that I will have to place more stringent checking in the application layer.

Hi Matt, I don't have an entity relationship map per se, but I have drawn up such a description in the database documentation. If you download the code repository, the database-specific documentation will be in the docs/devel/ directory. I did draw some entity relationship diagrams when I was designing the database, but at the time I didn't know of a drawing package that would allow me to draw them with minimal pain. I gave a lot of thought to the issue of modeling player involvement with teams, and I concluded that the best strategy for a model would be to remove an explicit affiliation between the player and the team, and establish 1:N relationship between the player and other characteristics that you'd like to track (e.g. injuries, or height/weight). The inclusion of dates is necessary for making future queries. The database documentation could use some images of the entity-relationship diagrams, but that task is low on my list of priorities.

The latter. I have had publicly available match data for a while, but it wasn't well organized and I wanted to enter information once and access it many times. With time I would like to have a repository (or repositories) of match result data, but I would need (a) a web-based data entry application and (b) a heavy-duty server. I'm working on (a), I haven't figured out the right strategy for realizing (b). Cricinfo's experience would be very valuable in that respect.

Thanks, Simon! If you have a chance join the developers' mailing list. I have started working on version 1.1 and I am attempting to build a PyQt executable, so I could use the additional feedback.

That's actually a much better option than Cancún. Less than a 60 minute flight away, and it would guarantee a pro-Haitian crowd (and money for Tempête's and CONCACAF's coffers).

You can't find any information about goals probability added because no one has developed it yet. Farhan Zaidi of the Oakland A's felt that it would be the ultimate kind of metric; I mentioned something similar to that (expected goal value) in my Moneyball and Soccer post a couple of years ago.

Thank you very much! It is gratifying to read from supporters about how the projections mesh well with how they viewed their team's performance. One possible extension is to create home and away Pythagoreans to drill down into the source of a team's performance, but the resulting table won't fit on the current page layout. I could write CSS to fix the problem or create a new page, but it's low on my priority list.

You know, every time I think of writing an update to this saga, a new twist appears. But I will say this: not only is the Warner/Blazer rift real, the Caribbean/rest of CONCACAF rift is real. But after Grondona's comments about not voting for the USA 2022 bid, I don't believe that CONMEBOL is a likely destination.

The idea is that the value of a goal is approximately a product of the quality of the opponent (for which league position acts as a proxy) and the interim value of the goal toward the maximum result (a win). It's true that scoring against the top teams in the league will always be difficult, but a late goal scored at 3-0 or 0-4 won't matter that much in the result. Thanks for the comments.

Thanks for the advice, but I've read both of those books years ago, and several others. As a matter of fact, I wrote about those books on this blog. These were ideas I was thinking about that were confirmed by reading those books.

Chris, You make a good point about the "marginal utility" of a goal - one scored to win the game against any team is much better than one scored as a mere concession no matter who the opponent. I'd like to tackle that in a future iteration, but I'm not sure the humble spreadsheet can handle such a task. And thanks for the advice on Transfermarkt.

No offense taken. I've seen much worse at scientific conferences! I think it is relatively easier to derive a priori metrics from in-match data than historical data (I made a distinction between the two in a previous post). With in-match metrics you might be able to use some sort of underlying physical model, but when you're developing a metric from goals scored, or fouls, or free kicks, or corners, or whatever, there's no underlying model for that. You have to start somewhere, so you assume some mathematical or statistical model that best fits your intuition and proceed from there. And if the results don't make sense, then change the model. Later on I'll look at how the weighted goals metric changes for linear and rational weighting functions.

I never claimed that my weighting was more valid than any other approach. I just made an assumption about comparative team strength and looked for a math function that best captures that. In my opinion, a logistic-like function does the job better than a simple linear function like y = x, or a simple nonlinear one like y = 1/x). That's just intuition on my part, I don't have anything to back it up at the present time, but it would be worth taking the time to test that expectation. You are correct when you say that this approach may not necessarily be sound just because a player like Messi is at the top of the weighted goalscoring list. The challenge of these approaches is to develop results that are a little bit surprising, but not so surprising that the credibility of the metric is questioned. All of these derived metrics have the same issue; they attempt to estimate something unknowable (not even to 90% accuracy in most cases) and an a priori derivation of these metrics isn't possible. I'm just taking a tiny step here. There's a lot more work to be done.

That's a good idea. I'll incorporate it in my implementation for league competitions. For Champions League group stages I felt it was a better idea to use mid-table at the beginning of the stage, but perhaps I could use club coefficient rankings.

By that I mean that FIFA and its confederations tend to protect their own until they die or are no longer useful, whichever comes first. Any reform of these international organizations will occur only from within and from the very top. The threat of cutting off a federation from all international competition up to and including the World Cup is enough to keep troublemakers and whistleblowers in line. Welcome to international sport politics.

Two words: Guild Loyalty.

I'm not sure the Caribbean should get more than three places in the Champions League, but you could make the case that they should get a direct entry to the group stage. Right now only Puerto Rico and Trinidad & Tobago have teams with a legitimate shot to qualify for the Champions League, much less advance to the group stage. In light of that I would be very interested to see if Tempete can qualify for the CCL this time around (they were 4th in the Caribbean championship last season). As for slot allocation, I do know that CONCACAF used their own ranking system to reallocate berths previously given to Belize and Nicaragua when both countries had stadium issues. Their rankings just happened to be very similar to mine. Make of that what you will.

Or even host...

If that's true, it also means that Monterrey fans won't be entirely in their own section. Not that I'm expecting any crowd trouble, but that I don't like that particular development nevertheless. But just 100 tickets for the away team? Was that all Monterrey requested or the minimum that RSL was entitled to make available? Either way, that's an extremely small number considering the number of Rayados fans in this country (or any other Mexican Primera club for that matter).

Yeah, Aguilar is becoming a key figure, just as I feared.

Yes.