of the components in the model, contrastswere defined to estimate the
strength of the BOLD response in each trial type relative to baseline. The
resulting weight maps were used to compute whole brain percent signal
change maps, which were used in the follow-up region of interest
based group level analysis described below.
A whole brain group analysis integrated data collected during three
runs of the sametask fromeach of the participants. This fixed effects approachwas
chosen because the sample sizewas too small to yield sufficient
statistical power in a random effects analysis. Due to the fixed
effects approach the conclusions drawn from this group analysis have
to be limited to the sample of participants studied and cannot be generalized
to the population from which the sample was drawn. However,
this does not affect the conclusions that can be drawn from the study,
since the purpose of the group analyses was merely to provide a
frame of reference for interpreting the multivariate pattern analysis results,
which are single-subject results. After estimating the betaweights
associated with each of the variablesmodelled, contrasts of each condition
of interest relative to baseline were computed, as well as specific
contrasts between these conditions. All these contrasts ignored the
time and dispersion derivative regressors associated with the different
conditions. The resulting statistical t-maps were evaluated at the single
voxel significance level of 0.05, corrected for multiple comparisons
based on Gaussian field theory (i.e., family-wise error correction)
(Friston et al., 1991; Worsley et al., 1992).
Group-level region of interest (ROI) analyseswere performed on the
participant and condition specific percent signal change data that resulted
from the first level analyses described above. These analyses
were conducted as post-hoc follow-up analyses intended to reveal the
specific nature of the BOLD signal modulations at selected activation
sites. ROIs were defined by local maxima in the statistical maps created
in the integrated group-level analyses, thatwere described above. A position
of a ROI was defined by the MNI coordinates of the local maximum
and comprised all the voxels within a 6 mm range of these
coordinates. Eleven ROIs in the left hemisphere were examined. For
each participant the average percent signal change was computed for
the 20% voxels within the ROI with the strongest percent signal change
(positive or negative) in a condition. This voxel selection procedure
leads to an optimization of the ROI to the specific activations of each
participant. The analysis was focused on identifying ROIs whose
power for discriminating long term from immediate overt recall trials
did not replicate in the monitoring task, which statistically translates
into a significant interaction between task and trial type. Despite the
small number of participants, a random effects analysis revealed six regions
for which this interaction was significant.We therefore choose to
report these random effects results.