Notes on "An optimizing neuroeconomic model of Discrete Choice" (Michael Woodford)

An Optimizing neuroeconomic model of Discrete Choice

Michael Woodford 

Columbia University, Feb 2,2014

The stochastiaty of choice

• reasons to prefer the interpretation of stochastic choice as representing random errors in cognition processing, Alain to random errors in perception:
• when random preference can be hypothesised that would account for observed data as trial- by- trial optimal choices, the kind of variation in preference required is not equally plausible
• a longstanding literature in experimental psychology and neuroscience has documented the randomness of the responses that subjects given when asked to make perceptual judgement, in a variety of sensory domains.
• in economic choices just as in the care of perceptual taxes, there is observed to be  a sybemat relation between the time required for experimental subject to make a decision and the characteristic of the alternations, specifically the average response time is shorter in the care of" easier" choices, which are also the ones for which repeated trials yield the same response a greater fraction of the time 
• the hypnosis that choice are based on errors, however, rather than on DMs' true preference , has some potentially unappealing features, in the absence of a more specific theory about the picture of the errors

drift- diffusion model

• a continons-valued subjective state variable, that one may think of as an evolving perception of the weight of sensory evidence in favor one response relative to the other, follows a random walk with drift on a bounded internal, a decision is made in favor of one or the other of the two responses when the bound corresponding to that decision is reached

As information - constrained dynamic model of discrete choice

• poison

The predictions of the OICM and the DM are not identical 

a common feature of experimental data on response times in binary perceptual classification tasks as well and is a famous empirical falling of the DDM. The OICM instead correctly predicts that correct choices should be made more quickly and by roughly the amount by which the mean response times are different in the data.

The OICM's predictions regarding response times are also more accurate than those of the basic (unconstrained) RI model, which is as successful as the DDM in explaining the data on choice frequencies alone.

predictions of RI are obtained under the assumption that response time has no consequences for reward

this model weakness:

        it generally over- predicts the variability of response times.