Skip to main content
. Author manuscript; available in PMC: 2009 Aug 18.
Published in final edited form as: J Am Stat Assoc. 2008 Dec 1;103(484):1648–1658. doi: 10.1198/016214508000001057

Table 1.

Comparison of ranks of the true model M1 from various model selection criteria for MAR covariates in linear models

(I)
 (II)
n = 100
 n = 300
 n = 500
 n = 100
 n = 300
 n = 500
Rank 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4
ICH,Q with 2d
1 331 25 4 0 329 17 1 0 325 14 1 0 301 51 13 0 325 37 4 0 327 32 6 0
2 1 48 8 0 5 57 12 0 4 63 10 0 5 31 21 4 6 35 17 5 3 35 15 3
3 0 1 49 3 1 3 53 2 0 3 50 3 0 9 35 8 0 2 36 9 0 7 36 8
4 0 0 1 29 0 0 0 20 0 0 2 25 0 0 1 21 0 0 2 22 0 1 1 26
IC H̃ (0),Q with 2d
1 329 23 5 0 328 16 3 0 322 14 0 0 302 49 10 0 323 35 5 0 320 31 10 0
2 3 50 8 1 6 57 12 0 7 59 10 1 4 34 25 5 8 36 15 5 10 34 16 1
3 0 1 48 1 1 4 51 2 0 7 50 3 0 8 34 9 0 3 37 9 0 8 31 10
4 0 0 1 30 0 0 0 20 0 0 3 24 0 0 1 19 0 0 2 22 0 2 1 26
IC H̃(1), Q with 2d
1 325 25 5 0 323 19 2 0 310 21 1 0 298 49 11 0 314 33 6 0 301 27 6 1
2 7 46 9 1 9 54 13 0 18 49 11 1 8 34 22 3 15 37 16 5 23 37 17 3
3 0 3 47 2 3 3 50 3 1 10 49 4 0 7 36 8 2 4 35 10 6 9 33 6
4 0 0 1 29 0 1 1 19 0 0 2 23 0 1 1 22 0 0 2 21 0 2 2 27

NOTE: Two cases of missing fractions for xi were included. Three different sample sizes, n = 100, 300, and 500 simulated data sets, were used for each case. The columns represent the results from AICQ.