Table 4.

Comparison of AUC of the proposed P-value-based method, the LASSO method and the LRT method with 1, 5 and 10 anchor items, respectively.

			Small DIF			Large DIF
			High	Medium	Low	High	Medium	Low
$N=500$	Small $d_j$	proposed	0.936	0.933	0.942	0.996	0.997	0.998
		LASSO	0.802	0.805	0.789	0.992	0.991	0.987
		LRT 1	0.861	0.853	0.867	0.982	0.984	0.982
		LRT 5	0.915	0.917	0.920	0.992	0.991	0.988
		LRT 10	0.929	0.919	0.922	0.989	0.995	0.989
	Large $d_j$	proposed	0.910	0.915	0.917	0.986	0.988	0.990
		LASSO	0.685	0.672	0.670	0.920	0.938	0.936
		LRT 1	0.823	0.800	0.826	0.966	0.966	0.969
		LRT 5	0.884	0.878	0.881	0.980	0.980	0.978
		LRT 10	0.897	0.875	0.884	0.983	0.975	0.977
$N=1000$	Small $d_j$	proposed	0.984	0.986	0.987	1.000	1.000	1.000
		LASSO	0.815	0.818	0.817	0.995	0.995	0.993
		LRT 1	0.965	0.968	0.960	0.997	0.997	0.994
		LRT 5	0.979	0.975	0.976	0.990	0.990	0.990
		LRT 10	0.985	0.966	0.977	0.995	0.984	0.988
	Large $d_j$	proposed	0.964	0.964	0.965	0.997	0.998	0.998
		LASSO	0.685	0.673	0.667	0.937	0.953	0.947
		LRT 1	0.944	0.942	0.941	0.989	0.995	0.992
		LRT 5	0.962	0.961	0.962	0.990	0.993	0.992
		LRT 10	0.972	0.953	0.962	1.000	0.998	0.992