|
Meta
|
k
|
I
2
|
\(\hat {\tau }^{2}\)
|
Equal- α
|
W-optimal
|
\(\alpha ^{*}_{2}\)
|
Width
|
|---|
|
Analysis
| | | |
CI
|
interval
| |
Ratio
|
|---|
|
CERVIX3
|
5
|
56 %
|
0.087
|
(0, 1.660)
|
(0, 1.100)
|
0.050
|
0.662
|
|
NSCLC4
|
11
|
75 %
|
0.132
|
(0.052, 0.787)
|
(0.021, 0.638)
|
0.048
|
0.839
|
|
NSCLC1
|
17
|
45 %
|
0.024
|
(0.000, 0.181)
|
(0, 0.147)
|
0.050
|
0.815
|
|
CERVIX1
|
18
|
62 %
|
0.112
|
(0.041, 0.500)
|
(0.017, 0.427)
|
0.046
|
0.892
|
- I2 is the heterogeneity statistic of Higgins and Thompson [28] and \(\hat {\tau }^{2}\) is the DerSimonian and Laird estimate. In each case we show the equal tailed (α1=α2=0.025) 95 % confidence interval, the W-optimal interval, the value of \(\alpha ^{*}_{2}\) that provides the W-optimal interval and the ratio of the width of the W-optimal interval and the equal tailed confidence interval. In each case we see that there is substantial reduction in the interval width by adopting α2>>α1
