Each week, quiz obsessives and Only Connect champions
Jamie Karran (@NoDrNo) and Michael Wallace (@statacake) take on the pub quizzes of the world.
Find out every Friday if you could have helped with the questions they got wrong.
Sunday 3 November 2013
No player has won the FIFA World Cup's 'Golden Ball' award more than once
The attendees
1) The doctor
2) The statistician
The ones that got away
1) What is the highest number of goals scored by one player in a FIFA World Cup finals tournament?
2) What name is given to a baby guinea pig?
3) In the nursery rhyme, where does the Muffin Man live?
4) Which general led the victorious army in the final battle of the Second Punic War?
5) Which Vice President of fuel company Exxon was kidnapped and murdered in 1992?
6) Who wrote the play Major Barbara, about an Officer of the Salvation Army who becomes disillusioned with her Christianity?
7) Who is the youngest brother of Backstreet Boy Nick Carter?
8) In which west coast US city would you find the Jack Murphy stadium?
The answers
1) 13 (by Frenchman Just Fontaine in 1958)
2) Pup
3) Drury Lane
4) Scipio Africanus
5) Sidney Reso
6) George Bernard Shaw
7) Aaron Carter
8) San Diego
Poll results: 24 votes (including one impressive 8/8) with the average voter scoring 2.5/8!
The excuses
1) A moderately staggering number (we guessed seven) Fontaine scored four goals in a 6-3 victory over West Germany in the third place playoff, along with a hat trick in a 7-3 drubbing of Paraguay in the group stage.
2) Young animal names are one of those things that appear in pub quizzes and pretty much nowhere else. The male and female names, at least, make use of porcine nomenclature, being boars and sows, respectively.
3) The doctor seemed convinced that (via Shrek) it was Pudding Lane, which is presumably part of his elaborate conspiracy theory that the Muffin Man started the Great Fire of London.
4) A spectacular failure of the 'play the quiz, not the question' strategy. The doctor had the correct answer immediately, but was alas swayed by my "that seems really hard for this quiz...".
5) Apparently no-one in the pub got this one right, to the great surprise of the quizmaster. The story is quite an interesting read, although best saved until after you've checked out the alternate questions below.
6) With little else to go on we played our 'American playwright' card of Tennessee Williams. On the plus side, it allowed me to make hilarious jokes for the rest of the evening about how we hadn't got the playwright right.
7) Presuming that the answer would be a famous Carter, and deploying around 90% of our hip hop knowledge in the process, we stuck down Sean Carter (AKA Jay-Z). On retrospect, quite a funny mis-step, but worth some style points, at least.
8) There only seemed a few options here (mainly in California) but we plumped for one of the wrong ones. Apparently it's called the Qualcomm Stadium at the moment, and has been since 1997. The Jack Murphy in question, meanwhile, was a sportswriter who initially drummed up support for a stadium in the city.
The alternative questions
1) Frank Fontaine is a major antagonist in which video game, whose concept was based on the ideas of objectivism (as popularized by Ayn Rand)?
2) Papua New Guinea occupies half of the island of New Guinea, the other half belonging to which nation?
3) The Drury Lane of Muffin Man fame was, in 1869, home to the first store of which eponymous (and now) nationwide retailer?
4) Scipio defeated the Carthaginian Hannibal Barca, the centre of whose Empire was located at (you guessed it) Carthage, now a suburb of which African capital city?
5) In their ransom note, Reso's kidnappers posed as members of Greenpeace angry (amongst other things) about the sinking of which ship in 1985?
6) The musical My Fair Lady is based on which Bernard Shaw play?
7) Despite their extensive discography, the Backstreet Boys have only managed to top the UK singles charts once with which 1999 song? A parody titled 'eBay' was released by Weird Al Yankovich several years later.
8) The Qualcomm Stadium is home to the NFL team the San Diego Chargers. California is home to a further 14 major sports teams (those that play American football, ice hockey, basketball or baseball in the NFL, NHL, NBA and MLB respectively), name 5 of them.
The answers
1) Bioshock
2) Indonesia
3) Sainsbury's
4) Tunis
5) Rainbow Warrior
6) Pygmalion
7) I Want It That Way
8) NFL: Oakland Raiders, San Francisco 49ers (and the aforementioned Chargers)
NHL: Anaheim Ducks, Los Angeles Kings, San Jose Sharks
NBA: Golden State Warriors, Los Angeles Clippers, Los Angeles Lakers, Sacramento Kings
MLB: Los Angeles Angels of Anaheim, Los Angeles Dodgers, Oakland Athletics, San Diego Padres, San Francisco Giants.
Hi Michael! I noticed you wrote a post about comparing teams on University Challenge earlier this year; having suddenly become incredibly invested owing to my own appearance on the show, I've been trying to do the same. Would you be interested in comparing notes, perchance?
Hi there. I'm afraid I don't have many notes to compare, as it's something I've mainly been thinking about from a theoretical/modelling approach, but haven't been organized enough to collect the data. Because of the starter/bonus arrangement it creates a potentially fun points distribution, as you have bonuses which are independent of the other team, but starters that aren't. I'd imagine you could do some cool stuff where bonus conversion gives one indicator of a team's quality, and then add in number of starters adjusted for the opponent's ability (perhaps by using *their* bonus conversion rate), but I'd want to give it slightly more thought than I have here...
Looking forward to seeing what you put together :)
Oh, I meant notes of a theoretical nature. :D I've created a fairly straightforward measure of a team's performance in any one match, which when averaged ought to give an idea of how good they "really" are. It goes 0.5CS * (BC + PS), where CS is the combined score for the match, BC the team's bonus conversion, and PS the proportion of correctly answered starters that team got. I don't have much in the way of statistical training, though; my goal was to reflect the elements which I reckon tell the most about a team. The idea is that in order to do well you need to both be able to answer bonuses (more or less pure knowledge recall) and get starters (a combination of knowledge recall and things like speed, risk assessment, and so on), which are quite different skills, so they're both included in the measurement. Using the combined score is supposed to take into account the ability of the other team.
Anyway, I've applied this to the last two series (not counting the current one), and I think it holds up fairly well. In neither case is the winner the team with the highest score, but then I don't think that's particularly anomalous - after all, the first seed in any given competition is hardly guaranteed to win. Would be interested to hear your thoughts!
That certainly seems like a reasonable start (although the statistical pedant in me feels obliged to point out that the 0.5 at the front doesn't really matter as it's the same for everyone). Certainly you want bonus conversion and starter conversion to factor in, but it gets fiddly with the latter because it would be good to take into account the strength of the opposition. Multiplying everything by CS presumably goes some way to doing that (since this is ultimately a proxy for the opposing team's conversion rate), but I suppose you can't really distinguish a team that's good on the buzzer from an average one that's playing a team that's bad on the buzzer.
Depending on how many data you have, you could perhaps try and build in a team's history to indicate how strong they are. So if team A thrashes team B in round 2, when team B looked very good on paper from their round 1 result, that tells you that team A might be a bit better. (You could perhaps multiply everything by 'opponent strength' which for every team starts at 1 and then deviates depending on their performances, a bit like the FIFA coefficients or similar.)
Another question is how to assess how well the model is working, for which you need lots of 'outcomes'. Perhaps rather than just looking at whether it predicts the series winner, you could see what percentage of individual matches it successfully predicts the winner from. (Maybe you're already doing this, of course.)
Sorry, bit of a ramble, these are just my unedited thoughts on the problem!
The 0.5 is there because it makes the score I give the team comparable to the score they actually got, which I find useful to see in what situations teams appear to do better or worse than I think they did.
I've gone and checked how well my method worked predicting the results of the two series I have data on. In both cases, it failed 9/23 times (counting those 23 matches which involved teams which had appeared before). That isn't terribly much better that random chance, but looking at it in a bit more detail it's more promising. Of the matches which my score (which I call the FQI, for reasons I've already forgotten) got wrong, 9 had a difference of less than 25 between the two teams, 5 of 25-50, and 4 of 50+. Given that the FQI score is roughly comparable to the actual score, we can think of the result in 14 of the 18 matches I missed as being decided by two full starter/bonus sets or fewer.
Furthermore, all but one of the 25+ upsets occurred in the second round, when there's only one data point per team. Of the quarter final or later upsets, the largest overcome difference in FQI is 33.
Overall I think that's pretty decent. In any given competition, you expect one or two upsets, and you also expect close matches to go either way.
Interesting (although I'm not sure where the 25+ upsets figure comes from; is that different to the 18 matches where your method didn't predict the winner?). Certainly encouraging on the QF and later rounds.
Out of interest (and assuming you can be bothered), how does it compare to simply working with total score?
Oh, I meant "upsets in which the difference was 25+". I've also gone and checked how many of the matches in each category, (difference 0-25, 25-50, 50+) I got right; in round two it's a bit dodgy, but for QFs+ it gets half the 0-25 matches right, 7/9 of the 25-50, and 1/1 of the 50+.
I'll see if I have some time later to go digging back for the total scores.
Aha, gotcha. You could also try simply the winning margin too (not that I particularly expect such crude methods to work, but it's always good to identify the simplest models to compare more complex ones too).
Also, I have only just joined the dots on who you are (my move to Canada interrupted my UC watching so I was a bit out of the loop). Looking forward to your next round; I'm a fellow Trinitarian so the prospect of seeing the college go anywhere in the contest is pretty exciting :)
I was watching your last episode with a room full of assassins (I haven't been converted to that august institution yet, I fear) and there was sort of a collective "wait, what?" when your introduction mentioned the guild. But I didn't realise you'd been at Trinity - that's really cool! I hope our performance lives up to your expectations.:)
Ha, excellent. Fun fact: all three of the Board Gamers have played assassins (it was also one of several team names we either weren't allowed to use, or decided against ourselves). Also, join the guild! It is the best thing ever! etc. etc.
Aha, an impressive memory :) Yes, I'll be putting up something short about that as a 'preview' to our next episode (as it could be our last). I just hope it lives up to expectations...
And that sucks, although hopefully if they are merely 'technical' you'll be able to reapply.
Hi Michael! I noticed you wrote a post about comparing teams on University Challenge earlier this year; having suddenly become incredibly invested owing to my own appearance on the show, I've been trying to do the same. Would you be interested in comparing notes, perchance?
ReplyDeletePS. looking forward to Only Connect tonight...
Hi there. I'm afraid I don't have many notes to compare, as it's something I've mainly been thinking about from a theoretical/modelling approach, but haven't been organized enough to collect the data. Because of the starter/bonus arrangement it creates a potentially fun points distribution, as you have bonuses which are independent of the other team, but starters that aren't. I'd imagine you could do some cool stuff where bonus conversion gives one indicator of a team's quality, and then add in number of starters adjusted for the opponent's ability (perhaps by using *their* bonus conversion rate), but I'd want to give it slightly more thought than I have here...
DeleteLooking forward to seeing what you put together :)
Oh, I meant notes of a theoretical nature. :D I've created a fairly straightforward measure of a team's performance in any one match, which when averaged ought to give an idea of how good they "really" are. It goes 0.5CS * (BC + PS), where CS is the combined score for the match, BC the team's bonus conversion, and PS the proportion of correctly answered starters that team got. I don't have much in the way of statistical training, though; my goal was to reflect the elements which I reckon tell the most about a team. The idea is that in order to do well you need to both be able to answer bonuses (more or less pure knowledge recall) and get starters (a combination of knowledge recall and things like speed, risk assessment, and so on), which are quite different skills, so they're both included in the measurement. Using the combined score is supposed to take into account the ability of the other team.
DeleteAnyway, I've applied this to the last two series (not counting the current one), and I think it holds up fairly well. In neither case is the winner the team with the highest score, but then I don't think that's particularly anomalous - after all, the first seed in any given competition is hardly guaranteed to win. Would be interested to hear your thoughts!
That certainly seems like a reasonable start (although the statistical pedant in me feels obliged to point out that the 0.5 at the front doesn't really matter as it's the same for everyone). Certainly you want bonus conversion and starter conversion to factor in, but it gets fiddly with the latter because it would be good to take into account the strength of the opposition. Multiplying everything by CS presumably goes some way to doing that (since this is ultimately a proxy for the opposing team's conversion rate), but I suppose you can't really distinguish a team that's good on the buzzer from an average one that's playing a team that's bad on the buzzer.
DeleteDepending on how many data you have, you could perhaps try and build in a team's history to indicate how strong they are. So if team A thrashes team B in round 2, when team B looked very good on paper from their round 1 result, that tells you that team A might be a bit better. (You could perhaps multiply everything by 'opponent strength' which for every team starts at 1 and then deviates depending on their performances, a bit like the FIFA coefficients or similar.)
Another question is how to assess how well the model is working, for which you need lots of 'outcomes'. Perhaps rather than just looking at whether it predicts the series winner, you could see what percentage of individual matches it successfully predicts the winner from. (Maybe you're already doing this, of course.)
Sorry, bit of a ramble, these are just my unedited thoughts on the problem!
The 0.5 is there because it makes the score I give the team comparable to the score they actually got, which I find useful to see in what situations teams appear to do better or worse than I think they did.
DeleteI've gone and checked how well my method worked predicting the results of the two series I have data on. In both cases, it failed 9/23 times (counting those 23 matches which involved teams which had appeared before). That isn't terribly much better that random chance, but looking at it in a bit more detail it's more promising. Of the matches which my score (which I call the FQI, for reasons I've already forgotten) got wrong, 9 had a difference of less than 25 between the two teams, 5 of 25-50, and 4 of 50+. Given that the FQI score is roughly comparable to the actual score, we can think of the result in 14 of the 18 matches I missed as being decided by two full starter/bonus sets or fewer.
Furthermore, all but one of the 25+ upsets occurred in the second round, when there's only one data point per team. Of the quarter final or later upsets, the largest overcome difference in FQI is 33.
Overall I think that's pretty decent. In any given competition, you expect one or two upsets, and you also expect close matches to go either way.
Interesting (although I'm not sure where the 25+ upsets figure comes from; is that different to the 18 matches where your method didn't predict the winner?). Certainly encouraging on the QF and later rounds.
DeleteOut of interest (and assuming you can be bothered), how does it compare to simply working with total score?
Oh, I meant "upsets in which the difference was 25+". I've also gone and checked how many of the matches in each category, (difference 0-25, 25-50, 50+) I got right; in round two it's a bit dodgy, but for QFs+ it gets half the 0-25 matches right, 7/9 of the 25-50, and 1/1 of the 50+.
DeleteI'll see if I have some time later to go digging back for the total scores.
Aha, gotcha. You could also try simply the winning margin too (not that I particularly expect such crude methods to work, but it's always good to identify the simplest models to compare more complex ones too).
DeleteAlso, I have only just joined the dots on who you are (my move to Canada interrupted my UC watching so I was a bit out of the loop). Looking forward to your next round; I'm a fellow Trinitarian so the prospect of seeing the college go anywhere in the contest is pretty exciting :)
I was watching your last episode with a room full of assassins (I haven't been converted to that august institution yet, I fear) and there was sort of a collective "wait, what?" when your introduction mentioned the guild. But I didn't realise you'd been at Trinity - that's really cool! I hope our performance lives up to your expectations.:)
DeleteHa, excellent. Fun fact: all three of the Board Gamers have played assassins (it was also one of several team names we either weren't allowed to use, or decided against ourselves). Also, join the guild! It is the best thing ever! etc. etc.
DeleteSpeaking of, I believe you promised several more of those rejected team names? Including some open-source related puns, if memory serves? :D
Delete(My own team went with the fairly banal but acceptable "Zelda Fans" - alas, we didn't make it on due to technical reasons.)
Aha, an impressive memory :) Yes, I'll be putting up something short about that as a 'preview' to our next episode (as it could be our last). I just hope it lives up to expectations...
DeleteAnd that sucks, although hopefully if they are merely 'technical' you'll be able to reapply.