Table 1.

Performance comparisons of CCLasso and SparCC based on simulation results

n	Method	d1	dF	$\text{AUC}$
Random Model
200	$\text{CCLasso}$	0.033(0.001)	2.954(0.049)	0.791(0.015)
	$\text{SparCC}$	0.057(0.001)	3.528(0.080)	0.823(0.014)
300	$\text{CCLasso}$	0.028(0.001)	2.409(0.057)	0.885(0.012)
	$\text{SparCC}$	0.047(0.001)	2.901(0.059)	0.891(0.011)
500	$\text{CCLasso}$	0.023(0.001)	1.994(0.053)	0.953(0.007)
	$\text{SparCC}$	0.038(0.001)	2.332(0.056)	0.951(0.006)
Neighbor Model
200	$\text{CCLasso}$	0.039(0.003)	3.355(0.206)	0.948(0.015)
	$\text{SparCC}$	0.076(0.001)	4.606(0.081)	0.888(0.014)
300	$\text{CCLasso}$	0.033(0.002)	2.675(0.151)	0.986(0.006)
	$\text{SparCC}$	0.070(0.001)	4.176(0.060)	0.931(0.009)
500	$\text{CCLasso}$	0.026(0.002)	2.064(0.121)	0.999(0.001)
	$\text{SparCC}$	0.065(0.001)	3.800(0.041)	0.967(0.006)
AR(4) Model
200	$\text{CCLasso}$	0.021(0.001)	2.444(0.134)	0.885(0.021)
	$\text{SparCC}$	0.061(0.001)	3.766(0.087)	0.858(0.019)
300	$\text{CCLasso}$	0.018(0.001)	1.994(0.133)	0.922(0.017)
	$\text{SparCC}$	0.052(0.001)	3.210(0.078)	0.890(0.017)
500	$\text{CCLasso}$	0.015(0.001)	1.549(0.087)	0.958(0.011)
	$\text{SparCC}$	0.044(0.001)	2.693(0.059)	0.918(0.011)
Hub Model
200	$\text{CCLasso}$	0.037(0.001)	3.453(0.037)	0.749(0.021)
	$\text{SparCC}$	0.067(0.001)	4.194(0.070)	0.690(0.014)
300	$\text{CCLasso}$	0.036(0.001)	3.133(0.047)	0.768(0.021)
	$\text{SparCC}$	0.059(0.001)	3.686(0.049)	0.735(0.012)
500	$\text{CCLasso}$	0.032(0.001)	2.918(0.048)	0.828(0.018)
	$\text{SparCC}$	0.051(0.001)	3.248(0.043)	0.788(0.010)
Block Model
200	$\text{CCLasso}$	0.039(0.001)	3.307(0.113)	0.782(0.014)
	$\text{SparCC}$	0.070(0.001)	4.268(0.072)	0.734(0.010)
300	$\text{CCLasso}$	0.035(0.001)	2.773(0.079)	0.854(0.014)
	$\text{SparCC}$	0.062(0.001)	3.788(0.052)	0.765(0.011)
500	$\text{CCLasso}$	0.029(0.001)	2.258(0.076)	0.924(0.011)
	$\text{SparCC}$	0.057(0.001)	3.374(0.038)	0.796(0.012)

d₁ and d_F are the two distances between the estimated correlation matrix and the true one defined in the text. AUC is the area under the receiver operation characteristics curve. The results are the averages over 100 simulation runs with standard deviations in brackets.