Table 2.

Simulation example with random missing, for varying observation probability p, signal strength $w_k^{*}$, and fusion setting f₀.

p	$w_k^{*}$	f 0	method	Error of ${\mathcal{B}}^{*}$	Error of ${\mathit{\beta }}_{k,j}^{*}$	TPR	FPR
0.5	30	0.3	No-fusion	0.091 (0.001)	0.059 (0.001)	1.000 (0.000)	0.121 (0.001)
			POSTER	0.055 (0.001)	0.037 (0.001)	1.000 (0.000)	0.021 (0.004)
		0.7	No-fusion	0.088 (0.001)	0.056 (0.001)	1.000 (0.000)	0.099 (0.026)
			POSTER	0.079 (0.002)	0.051 (0.001)	1.000 (0.000)	0.079 (0.024)
	40	0.3	No-fusion	0.068 (0.001)	0.044 (0.001)	1.000 (0.000)	0.120 (0.000)
			POSTER	0.042 (0.001)	0.029 (0.001)	1.000 (0.000)	0.019 (0.003)
		0.7	No-fusion	0.066 (0.001)	0.043 (0.001)	1.000 (0.000)	0.072 (0.000)
			POSTER	0.059 (0.001)	0.039 (0.001)	1.000 (0.000)	0.056 (0.003)
0.3	30	0.3	No-fusion	0.119 (0.002)	0.078 (0.002)	0.998 (0.001)	0.148 (0.023)
			POSTER	0.077 (0.002)	0.054 (0.002)	1.000 (0.000)	0.052 (0.016)
		0.7	No-fusion	0.113 (0.002)	0.074 (0.002)	0.998 (0.001)	0.104 (0.026)
			POSTER	0.103 (0.002)	0.066 (0.002)	0.998 (0.001)	0.086 (0.024)
	40	0.3	No-fusion	0.092 (0.020)	0.060 (0.001)	1.000 (0.000)	0.120 (0.000)
			POSTER	0.058 (0.001)	0.042 (0.001)	1.000 (0.000)	0.025 (0.005)
		0.7	No-fusion	0.084 (0.001)	0.054 (0.001)	0.999 (0.000)	0.074 (0.001)
			POSTER	0.075 (0.001)	0.049 (0.001)	1.000 (0.000)	0.054 (0.030)

NOTES: Reported are the average estimation errors of ${\mathcal{B}}^{*}$ and of ${\mathit{\beta }}_{k,j}^{*}$, and the true and false positive rates of selection based on 30 data replications (the standard errors in the parentheses). Two methods are compared: our method without the fusion constraint (No-fusion), and our proposed method (POSTER).