6. Competition Rules

The tournament includes three independent competitions. Each competition focuses on the prediction of one of the three experimental conditions (Description, E-sampling, and E-repeated).

To participate in each competition, researchers will be asked to submit a model that predicts behavior in the relevant experimental condition.

The model should be implemented in a computer program (in MATLAB, SAS or Visual Basic) that reads the parameters of gambles (High, p, Low and Medium).

The parameters have to be read as input from a specific data file (named "Compar.dat")¹ that will have the same structure as the following file Estdat.dat.²

The program should predict choice proportion of the risky option as an output. It should include three parts (see specification in the examples below). Part 1 (Input) and Part 3 (output) should be basically copied from the examples.

The model should be implemented in Part 2. Notice that this part should derive quantitative prediction. That is, if the model includes free parameters they should be estimated before the submission and their estimated values should be used in part 2.

All models should be submitted via email to: dmcompete@gmail.com and have to be accompanied with the estimated values of its parameters according to the estimation set.

The deadline for submissions to the Computational Model Competition is: 1st September 2008

Framework and Examples of the competition programs:

One shot Decisions under Risk (Condition Description):

SAS program of Cumulative Prospect Theory: Download

MATLAB program of Cumulative Prospect Theory: Download

One shot Decisions from Experience (Condition E-sampling):

SAS program of Primed Sampler: Download

Repeated Decisions from Experience (Condition E-repeated):

SAS program of the Normative Reinforcement Learning: Download

SAS program of Explorative Sampler with Recency: Download

The competition set will be run in May 2008. It will focus on a set of 60 problems that will be drawn (independently of the first set) from the population of choice problems described above.

The winner in each competition will be the model that provides the best prediction (minimizes mean square deviation) between the observed and predicted choice proportion of the 60 problems in the relevant tasks.

The designers of the winning models (up to two authors) will be invited to be co-authors of a paper that is to be submitted to the JBDM Special Issue on Experience-Based Decision Making. In addition, the first author of the winning submissions will be the invited speakers in a workshop that will focus on the outcomes of the competition.

¹ The file Compar.dat will be uploaded to the site after the submission deadline (at 12.01AM, EST, September 2, 2008). It will be used by the organizers to rank the submissions.

² The file Estdat.dat includes the gamble parameters (High, p, Low and Medium) of the 60 problems of the estimation set (each raw is a different problem). The competition will be run with the same file after the parameters will be replaced with the parameters of the competition set.