Table 4. Standardized Beta Values for Linear Regressions Predicting Individual Differences in Treatment Effect Sizes Following Two Different Types of Intervention, Normalization and Compensation, in Simulated Networks With a Connectivity Over-Pruning Disorder (Davis, 2017).

	Intervention type
	Normalization		Compensation
Parameter	Training set performance	Generalization performance	Training set performance	Generalization performance
Note. N = 790 networks (only those from the population showing a behaviorally assessed performance deficit). Separate regressions were carried out for performance on the training set and generalization set. The shaded area shows parameters related to the pathological process, elevated values of the pruning threshold, permitting larger connections to be removed following the onset of pruning. Bold shows significant at p < .05.
a This parameter was set to atypical values to produce the developmental disorder.
Number of hidden units	−.016	.012	.011	.023
Sigmoid temperature	−.040	−.001	−.098	−.127
Processing noise	.028	.032	.007	−.012
Learning rate	−.065	−.086	−.053	−.016
Momentum	−.014	−.011	−.013	−.011
Initial weight variance	−.015	−.002	−.031	−.023
Architecture	−.110	−.101	−.112	−.092
Learning algorithm	−.006	−.059	−.011	.010
Response threshold	−.055	−.063	.000	.036
Pruning onset	.022	−.007	.057	.045
Pruning rate	−.006	.006	−.062	−.075
Pruning thresholda	.014	.082	.039	−.047
Weight decay rate	.021	.007	.036	.025
Sparseness of connectivity	.027	.065	.049	.052
Richness of environment	−.030	−.036	−.028	−.028