Power-matching
Subjects were sub-selected from each age group, such that the
distribution of change in band power within the groups were matched. We
first binned the subjects within each age group with respect to the
power and computed the greatest common distribution. Within each power
bin, subjects were randomly selected from the group with more subjects,
to match the subjects in the other group. Ultimately, the subjects in
each bin were adjusted to arrive at the common distribution for the two
age groups. This procedure eliminates the dependence of power on the
effect under study, i.e., the FC (Churchland et al. , 2010).
Since each iteration produced a different subset of subjects, the
resultant p-value (obtained using Kruskal-Wallis Test) varied with each
iteration. We executed this random sampling 50 times and chose a
representative iteration (Figure 3) that represented the overall
distribution of results (p-values). Specifically, we used the iteration
with p-value near the median of the distribution for both left and right
electrode groups.
Power matching for MCI vs. healthy comparison was done on a single
subject basis. The subject among the age (within ±1 yr gap) and gender
matched controls with minimal separation in change in power from the
case subject was selected for each case.