Table 1. Comparison of different models’ abilities in capturing subjects’ patterns of decoy efficacies based on three goodness-of-fit measures.

Model #	Number of parameters	Model parameters	Cross-validation prediction error	AIC	BIC
1	7	fp,fm,b1,b2,b3,b4,σ	0.3716	285.84	305.91
2	5	b1,b2,b3,b4,σ	0.0456	211.77	226.10
3	7	fp12,fm12,fp34,fm34,b12,b34,σ	0.0371	128.28	148.35
4	5	fp,fm,b12,b34,σ	0.0344	127.53	141.86
5	5	fp,fm,b13,b24,σ	0.0433	205.53	219.86
6	5	fp,fm,b14,b23,σ	0.0476	223.96	238.29
7	6	fp,fm,b12,b34,b0,σ	0.0359	128.06	145.26
8	6	fp,fm,b13,b24,b0,σ	0.0472	223.12	240.32
9	6	fp,fm,b14,b23,b0,σ	0.0493	230.05	247.25
10	5	b0,f0,b12,b34,σ	0.0521	231.08	245.41
11	4	f0,b12,b34,σ	0.0538	237.63	249.10
12	4	b0,fp,fm,σ	0.4007	303.31	314.78
13	4	b0,b12,b34,σ	0.0402	181.30	192.77
14	3	fp,fm,σ	0.0382	169.93	178.53
15	3	b12,b34,σ	0.0459	214.97	223.57
16	3	b0,f0,σ	0.0502	227.08	235.68
17	2	b0,σ	0.0461	213.70	219.43

Reported are cross-validation prediction error (i.e., the absolute difference between the predicted and actual), AIC, and BIC values for each model and its corresponding sets of parameters. The green shading indicates the best overall model, and blue shading shows the best model with only low-level components. As parameters, σ measures the stochasticity in choice, f₀ is the single neural representation factor, and f_p and f_m are independent neural representation factors for probability and magnitude, respectively. f_pij (respectively, f_mij) indicates location-dependent representation factors for probability (respectively, magnitude) with similar values for decoys at locations i and j. High-level parameters b₀ and b_k (k = {1,2,3,4}) indicate the constant and location-dependent biases, respectively, and determine the weights of different attributes on final choice (b_ij indicates the case in which location-dependent biases b_i and b_j are assumed to have the same value, b_i = b_j = b_ij).