Friday 4 May 2012

Playing games at SCOR 2012.

The 3rd Student Conference on Operational Research (SCOR 2012) took place in Nottingham in the UK from the 20th to the 22nd of April. (Photos available on the chair of the organising committee's G+ profile).

Louise Orpin (Education officer of the OR society: http://learnaboutor.com/) and I were invited to give a plenary on the last day on the subject of outreach events. The slides we delivered are available here. The talk was in two parts, in the first part I discussed a game theory outreach event (that I have already blogged about here) and in the second part Louise discussed how members of the audience could get involved.

One of the "highlights" (for me anyway) was getting all the students to play the "2/3rds of the average game". We played it twice: once without discussing the rational behaviour and a second time afterwards. 

I got all the audience (about 80) to write down their answers and hand them down to some of our Cardiff University postgraduates who had agreed to fill in a gdoc spreadsheet with the results so that by the end of my talk I could show the results (and announce the winner!).

The first graph (i.e. before we all discussed rational behaviour) shows the distribution of guesses:

Interestingly the lowest guess was actually 5 but also you can see that all the guesses were relatively uniform... The winning guess was 18.

The second chart shows the results after we all discussed the rational behaviour (i.e. that all strategies other then 0 are in fact dominated).



Straight away we see the overall strategies are beginning to "converge" (the maximum guess was 40 this time as opposed to 63). The winning guess was 10 (a 4 way tie).

None of this is surprising but I thought it was interesting to see that we do indeed get similar behaviour to the behaviour discuss in this previous blog post where this game was played with high school kids as opposed to PhD students.

I've been more or less (as and when time allows) taking the game theory course offered by coursera and I know that they are collecting similar data. They have obviously got a ridiculously large sample set in comparison to mine so I like forward to seeing the same graphs...

EDIT: Marc Harper on G+ pointed me to one of his papers that "somewhat captures the result of your experiment that even knowing that a guess of anything other than 0 is dominated, 0 is not the winning guess." Here's the link: http://arxiv.org/pdf/1005.1311.pdf