Next: Methods
Up: tr03mw2
Previous: Results - signal
In this section we analyse some null/rest data using the full
spatio-temporal model. The intention is to do a test that is
similar to the investigation of false positive rates (FPR) in
frequentist statistics.
To investigate FPRs in a frequentist framework, a statistic is
calculated a number of times from data conforming to the null
hypothesis, to build up an experimentally obtained null
distribution. The resulting null distribution should correspond to
the known theoretical distribution under the null hypothesis. In
Bayesian statistics we do not work with theoretical null
hypothesis distributions for the statistic of a parameter, but
instead we have a probability distribution over a parameter of
interest. Nevertheless, we can still use artificial null data to ask:
`If we threshold
to assign voxels as ``positive'' when
, then
at what rate do we
produce ``positives'' when we know
?'
Unlike frequentist statistics we do not
require this rate of ``false positives'' to approximate
, however, it is
still interesting to investigate its properties.
Subsections
Next: Methods
Up: tr03mw2
Previous: Results - signal