BAYESIAN FRAMEWORK AND STATISTICS

The analyses were conducted with WINBUGS (free software) and MATLAB

During my Masters my main project featured an investigation within a Bayesian framework (i.e. a particular theoretical approach) also using actual Bayesian statistics (i.e. the practical side of the approach).

The experiment required participants to perform rapid pointing movements toward a large touchscreen. Given the specific experimental set-up, an "ideal" performance was calculated, which takes into consideration the variability in human movement and the experimental gains and losses (participants were rewarded with points if they hit targets, and penalized when they hit penalties).

Data Visualization is achieved through Matlab code.

Above is a model based on performance of subject RD. That subject was very precise, so in the most upper condition, when the penalty was small and the chance of hitting the correct target was high, RD's gain distribution (the red hill) shows a large red area and the possibility of losing points is almost non-existent. This is because the model is based on RD's precise error rate and variability of movement, and also calculates the possible gains and losses.

In the bottom illustration its possible to see a huge dip, corresponding to the relatively large possibility that RD would hit a large penalty. However, given that RD is also precise with the pointing movement, there is still a realistic chance the correct target will be hit, shown by the still present red area at the top of the gain distribution.

The complete project was written up as my main Dissertation project at the University of Glasgow. The highest grade was obtained.

An abstract of the project was also sent to a small online journal called i-Perception

Lages, Smith & Puntiroli (2012)

BAYESIAN STATISTICS

Data Analysis conducted with WINBUGS

Bayesian statistics used to be too computationally heavy for most computers. Now that computers can handle the computations several software programs have emerged that can take care of Bayesian estimations and analyses. One of these programs is called Winbugs and it is possible to download it for free. In Glasgow University during the Masters several lessons focused on how to get the most out of Winbugs. Below is an example of what the Winbugs interface looks like.