2.5 Statistical Analysis
For SORT performance, we utilized repeated measures general linear
models (GLM; IBM SPSS Statistics 28.0) with group as a between-subject
factor (two group comparisons: YA vs HA, or HA vs MCI) and conditions as
a within subject factor (retrieval vs non-retrieval). RT and accuracy
data were analyzed separately. Post-hoc comparisons were adjusted with
Bonferroni correction. Significant results were reported when p< 0.05.
EEG data were analyzed for theta (4– 8 Hz), alpha (8–12 Hz), low beta
(12–20 Hz), and high beta (20-30 Hz) frequency bands with mean values
calculated from each frequency range, based on canonical classification
from prior EEG studies (Weiss & Mueller, 2012; Poulisse et al., 2020;
Markiewicz et al., 2021). In order to examine differences in individual
alpha frequency (IAF), given different age groups, we performed peak
frequency extraction within the frequency range 7-14 Hz from epochs
between -750 to 2500 ms to stimulus onset, for each individual across
conditions (Klimesch, 1999). We ran a repeated measures GLM and did not
find any significant effects (no effects of group, condition, or group
by condition interaction), thus ruling out that individual alpha
differences affected our analysis, and supporting the use of the
standard window.
Permutation tests (N = 2000) were performed to compare group data, for
difference between retrieval and non-retrieval conditions, based on
two-tailed independent t-tests. These tests were performed to assess (1)
normal aging effects by comparing YA to HA groups and (2) pathological
aging effects by comparing HA to MCI groups, for all electrodes (N = 62)
and all windows (N = 26) spanning from 188 to 1105 ms post stimulus
onset. We chose this time window given that prior studies have shown
effects within this time window on the SORT task (Chiang et al., 2014,
2015, 2016, 2020). A False Discovery Rate (FDR) < 0.05 from
permutation tests was considered significant, which is a robust
statistical approach commonly adopted in functional imaging studies to
avoid excessive false negative results, so further corrections for
multiple comparisons were not applied. Additionally, we tested
conditional effects (retrieval vs non-retrieval conditions) within each
group separately, based on permutation tests (N = 2000) using two-tailed
paired-t tests, thresholded at FDR < 0.05. These results are
presented in the supplementary materials since condition effects are not
related to our primary research questions.
Post hoc analysis was performed based on the mean power scores
extracted within the time window at the electrodes showing significant
results, using 2-tailed paired-t tests to compare conditions (retrieval
vs non-retrievals). Significant results were reported if p< 0.05.