A better estimate of the internal consistency reliability of frontal EEG asymmetry scores

Abstract

Frontal alpha asymmetry is typically computed using alpha power averaged across many overlapping epochs. Previous reports have estimated the internal consistency reliability of asymmetry by dividing resting EEG sessions into segments of equal duration (e.g., 1 min) and treating asymmetry scores for each segment as "items" to estimate internal consistency reliability using Cronbach's alpha. Cronbach's alpha partly depends on the number of items, such that this approach may underestimate reliability by using less than the number of distinct items available. Reliability estimates for resting EEG data in the present study (204 subjects, 8 sessions) were obtained using mean split-half correlations with epoch alpha power as treated as separate items. Estimates at all scalp sites and reference schemes approached .90 with as few as 100 epochs, suggesting the internal consistency of frontal asymmetry is greater than that previously reported.

Figure 1

Estimated internal consistency reliability (r_TT) of asymmetry scores for epoch set sizes n ranging from 20 to 400, across average (black), online (gray) and linked-mastoids (dashed) reference derivations and all homologous electrode pairs. Graph markers and table insets indicate the epoch set size n at which the estimated internal consistency reliability coefficient for each reference derivation was greater than or equal to 0.90.

Figure 2

Percentage of homologous electrode pairs in which estimates of internal consistency reliability (r_TT) of asymmetry scores were greater than or equal to 0.70 (white), 0.80 (light gray) and 0.90 (dark gray) as a function of epoch set size n and reference derivation.

Figure 3

Estimated internal consistency reliability (r_TT) of asymmetry scores for epoch set sizes of 120 and 200, with light gray numbers indicating .85 ≤ r_TT < .90, and bold numbers indicating r_TT ≥ .95 (the pair CB2-CB1 was omitted).

Figure 4

Mean, minimum (▲) and maximum (▼) epoch set sizes n in which estimates of internal consistency reliability of asymmetry scores exceeded 0.90 for all homologous electrodes. Each row represents epochs sampled according to resting minute type (all, eyes-closed only, and eyes-open only). The first column of graphs represent epochs remaining after ocular artifact rejection (removing epochs in which vertical EOG amplitude deviates more than 75μV from the epoch mean); the second column represents no ocular artifact rejection (though in both cases epochs overlapping manually rejected segments of the EEG data are omitted).

Figure 5

Percentage of resting EEG datasets (of a total of 832, aggregated across participants, recording session, and within-day resting task) as a function of percentage of artifact-free epochs (with resting task having a maximum of 952 epochs).