Skip to main content
. 2022 Apr 21;5:733163. doi: 10.3389/frai.2022.733163

Table 3.

Distribution analysis and inter-rater agreement.

Human responses Distribution
Resp/Part P1 P2 P3 P4 AVG %
−1 2,065 995 645 1,185 1,223 34.0%
0 149 1120 1895 1,270 1,109 30.8%
1 1,386 1485 1060 1,145 1,269 35.3%
TOT 3,600 3600 3600 3,600 3,600 100%
Participant Agreement analysis
P1 P2 P3 P4 Average %
P1 0 1726 1308 1650 1561 43%
P2 1726 0 1944 1758 1809 50%
P3 1308 1944 0 1741 1664 46%
P4 1650 1758 1741 0 1716 48%
TOTAL 6,751
AVG xPART 1,688
Average Particip match each other 47%
Fleiss-Kappa Error Confidence Interval Agreement Z p-value
0.202 0.0048153 0.19955 0.20446 “Fair” 41.951 0

The top part shows human judgement distribution for the three possible questionnaire responses “less” (−1), “neutral” (0), and “more” (1). The bottom part shows percent agreement for the four raters; Fleiss' Kappa analysis revealed that the agreement between raters is better than chance with a p < 0.05. The task was difficult and the responses noisy. Thus, only the most reliable questions were used to compare to the CEREBRA model.