How do humans learn about the reliability of automation?

Cognitive Research: Principles and Implications

Table 2 Group BIC values for each model for each experimental condition

Model	One		Two			Three
Model	High	Low	Start-high	Start-low	Constant	Large drop	Medium drop	Small drop
Experiment
Two-kernel Delta	0	0	0	515	0	0	0	0
Delta	342	226	449	728	387	335	375	87
Sampling (proportional to delta weights)	990	1147	2171	0	219	1549	1950	97
Sampling (last/average)	679	728	1790	931	318	1542	1548	171
IIAB	1267	982	346	1436	1416	809	1013	502
Bayesian	1016	873	2089	1956	1122	2645	1764	376
Contingent Sampling	4034	5249	3695	3942	4648	3527	4218	3175
No updating	1214	1124	2276	2023	981	2978	2100	274

We report BIC values after subtracting the BIC for the most supported model for each experiment condition (Hence, the most supported model for each experimental condition has a value of 0). We report BIC in this manner because it is the differences between BICs that matter for the purposes of model comparison (Kass & Raftery, 1995) and it is easier to see which model is best fitting for each condition, and the relative performance of other models to that best fitting model