Skip to main content
. Author manuscript; available in PMC: 2014 Jan 13.
Published in final edited form as: Biometrics. 2012 May 2;68(4):1055–1063. doi: 10.1111/j.1541-0420.2012.01766.x

Table 3.

Number of times (out of 200) the PPL and DICO (observed data likelihood DIC) criterion choose the true model when fitting one of the following three models: MAR Selection model (SM0), Mixture model 1 (MM1), and Mixture Model 2 (MM2) for four choices of T(r, ry): T1(r, ry) = rJyJ − r1y1, T2(r, ry) = rJ(rJyJ − r1y1), T6(r,ry)=j=1J{I(rj=1,rj+1=0)rjyj}I(r2=1)r1y1 and T8(r,ry)=[j=1J{I(rj=1,rj+1=0)rjyj}I(r2=1)r1y1]2.

True Model Size Model T 1 T 2 T 6 T 8 DICo
SM0 50 SM0 193 198 192 194 200
MM1 7 0 2 3 0
MM2 0 2 6 3 0

SM0 100 SM0 161 197 158 168 199
MM1 39 3 2 20 1
MM2 0 0 40 12 0

SM0 2000 SM0 16 165 47 113 200
MM1 184 35 1 74 0
MM2 0 0 152 13 0

MM1 50 SM0 117 4 21 21 0
MM1 83 196 179 179 200
MM2 0 0 0 0 0

MM1 100 SM0 111 0 2 0 0
MM1 89 200 198 200 200
MM2 0 0 0 0 0

MM1 2000 SM0 79 0 0 0 0
MM1 121 200 200 200 200
MM2 0 0 0 0 0

MM2 50 SM0 29 0 5 7 0
MM1 91 98 98 78 40
MM2 80 102 97 115 160

MM2 100 SM0 5 0 0 0 0
MM1 87 90 100 72 46
MM2 108 110 100 128 154

MM2 2000 SM0 0 0 0 0 0
MM1 101 110 106 103 57
MM2 99 90 94 97 143