Table 5.

Reliable parameters from Model 3 using the BIC model selection technique. Smaller and more negative values indicate that a letter was less perceivable, biased against, or that a letter combination was discriminated less well than was expected by the perceivability and bias alone. Baseline log-odds accuracy was 1.306 which is equivalent to a percent accuracy of 78.7, which was restricted to be the same across both experiments.

Letter	Experiment 1 (“@”)		Experiment 2 (“#”)		Global Dissimilarity (δ)			Experiment-specific Dissimilarity (δ)
	Perceivability (λ)	Bias (γ)	Perceivability (λ)	Bias (γ)
Intercept	1.306		1.306
A	−0.571				B	G	−0.388	Experiment 1
B		0.227			C	G	−0.737	B	Z	0.621
C	−0.174		0.602	−0.206	C	O	−0.686	C	D	−0.638
D			0.208	0.155	C	Q	−0.540	D	G	−0.927
E					C	U	−0.364	D	O	−0.597
F			−0.153		D	Z	0.349	D	Q	−0.849
G				−0.230	E	F	−0.648	D	U	−0.618
H			−0.612		H	N	−0.457	E	I	0.475
I		−0.273	−0.948	−0.346	I	L	−0.335	G	Q	−0.560
J	−0.338		−0.242	−0.146	K	M	−0.345	G	U	−0.599
K					K	X	−0.355	O	U	−0.522
L		−0.164	−0.573	−0.191	M	N	−0.424	R	Y	−0.600
M	0.246				M	W	−0.468
N				−0.118	N	W	−0.408	Experiment 2
O			0.672	−0.161	O	Q	−1.152	A	V	−0.551
P					Q	U	−0.646	D	L	0.653
Q					X	Y	−0.342	E	K	−0.635
R		0.171		0.148				F	T	−0.497
S	0.258			−0.126				G	L	0.482
T			−0.382	−0.199				G	X	0.473
U					Differential Dissimilarity			H	W	−0.494
V	0.422				A	Q	−0.472	L	P	0.663
W			−0.251		E	N	0.360	L	Q	0.621
X	−0.25		−0.851	0.161	G	P	−0.413	Q	R	0.470
Y	0.204		−0.626		I	Z	0.379
Z					U	X	−0.415