Skip to main content
. Author manuscript; available in PMC: 2011 Jan 25.
Published in final edited form as: Ann Appl Stat. 2010 Sep;4(3):1476–1497. doi: 10.1214/10-AOAS326

Table 2.

Observed and predicted counts (based on learning sample data)

Observed Total
> 200 < 200
Predicted > 200 1932 215 2147
< 200 34 184 218
Total 1966 399 2365
Observed Total
> 350 < 350
Predicted > 350 1194 289 1483
< 350 137 745 882
Total 1331 1034 2365
(a) GLMM approach with a “naive” 0.50 probability cutoff
Observed Total Observed Total
> 200 < 200 > 200 < 200
Predicted* > 200 826 18 844 1206 37 1243
< 200 1140 381 1521 760 362 1122
Total 1966 399 2365 1966 399 2365
Observed Total Observed Total
> 350 < 350 > 350 < 350
Predicted* > 350 669 50 719 880 103 983
< 350 Total 662 1331 984 1034 1646 2365 451 1331 931 1034 1382 2365
(b) GLMM approach
Observed Total Observed Total
> 200 < 200 > 200 < 200
Predicted* > 200 1291 19 1319 1558 38 1596
< 200 675 380 1055 408 361 769
Total 1966 399 2365 1966 399 2365
Observed Total Observed Total
> 350 < 350 > 350 < 350
Predicted* > 350 760 51 811 940 103 1043
< 350 571 983 1554 391 931 1322
Total 1331 1034 2365 1331 1034 2365
(c) LMM approach
*

Predicted counts are based on rules with resubstitution FP rate estimates of approximately (but not greater than) 5% (left panels) and 10% (right panels).