Biomedical Engineering Reference
In-Depth Information
TABLE 6.1: Estimation Results for the HIV Data: Posterior Means and 95%
Credible Intervals for the Regression Coecients under Different Models.
1
2
3
2
1
3
2
Probit 1.1720
1.9508
2.3836
0.7788
0.4327
(0.903, 1.446)
(1.633, 2.275)
( 2.021, 2.745)
(0.479, 1.078 )
(0.098, 0.762)
PO{M1 2.1010
3.4302
4.1466
1.3292
0.7164
(1.633, 2.596)
(2.864, 4.032)
(3.490, 4.861)
(0.834, 1.829)
(0.169, 1.276)
PO{M2 2.1237
3.4771
4.2223
1.3571
0.7440
(1.644, 2.604)
(2.909, 4.055)
(3.550, 4.877)
(0.896, 1.871)
(0.220, 1.297)
PO{M3 2.0089
3.3383
4.0863
1.3294
0.7480
(1.555, 2.473)
(2.807, 3.876)
(3.478, 4.692)
(0.828, 1.840)
(0.204, 1.311)
PH 1.3863
2.5619
2.9692
1.1756
0.4073
(1.030 1.745)
(2.213, 2.959)
(2.584, 3.368)
(0.832, 1.495)
(0.055, 0.747)
results from the three different methods under the PO model are indeed very
similar, as expected. From Table 6.1, the results under the different models
all indicate the same conclusion because all the posterior means are positive
and the corresponding 95% credible intervals located to the right of 0. These
results suggest that there is a significant dose effect between each dose group
and the non-dose group, and also a significant difference between the adjacent
dose groups.
We also observed that using different number of knots (say 10, 20, 30) and
different degree (1, 2, 3, 4) did not produce very different estimates for this
data set. Part of the comparison results can be found in Lin and Wang (2010)
and Wang and Lin (2011), among our other work using monotone splines.
Search WWH ::
Custom Search