Skip to main content
. 2015 Jan 15;5(1):e006721. doi: 10.1136/bmjopen-2014-006721

Table 2.

Type I error for two typical PEM in different scenarios

Sample size 50 100 200 300 500 1000
Pathway length K=2 Inline graphic Inline graphic
Multivariate Δ* 0.027 0.039 0.045 0.032 0.040 0.051
Exact* 0.019 0.035 0.042 0.031 0.040 0.050
Unbiased* 0.033 0.043 0.045 0.033 0.040 0.051
Bootstrap* 0.040 0.052 0.057 0.042 0.049 0.059
Percentile bootstrap† 0.046 0.057 0.059 0.047 0.057 0.059
Bias-corrected bootstrap† 0.056 0.068 0.058 0.048 0.057 0.058
Pathway length K=3 Inline graphic Inline graphic
Multivariate Δ* 0.005 0.021 0.037 0.035 0.048 0.045
Exact* 0.003 0.015 0.033 0.035 0.047 0.044
Unbiased* 0.015 0.026 0.042 0.040 0.050 0.047
Bootstrap* 0.011 0.020 0.044 0.037 0.053 0.048
Percentile bootstrap† 0.029 0.034 0.049 0.047 0.054 0.050
Bias-corrected bootstrap† 0.059 0.057 0.059 0.053 0.058 0.057
Pathway length K=4 Inline graphic Inline graphic
Multivariate Δ* 0.004 0.012 0.022 0.030 0.035 0.047
Exact* 0.001 0.009 0.021 0.026 0.032 0.047
Unbiased* 0.010 0.019 0.032 0.035 0.037 0.047
Bootstrap* 0.014 0.020 0.034 0.037 0.039 0.054
Percentile bootstrap† 0.033 0.034 0.042 0.045 0.048 0.058
Bias-corrected bootstrap† 0.065 0.060 0.055 0.052 0.054 0.056
Pathway length K=5 Inline graphic Inline graphic
Multivariate Δ* 0.000 0.004 0.018 0.034 0.035 0.032
Exact* 0.000 0.003 0.014 0.031 0.033 0.029
Unbiased* 0.001 0.013 0.023 0.039 0.038 0.033
Bootstrap* 0.004 0.007 0.026 0.037 0.037 0.041
Percentile bootstrap† 0.022 0.035 0.043 0.049 0.049 0.043
Bias-corrected bootstrap† 0.066 0.061 0.059 0.057 0.056 0.046

*For PEM-UD with different methods, estimate the variance.

†For PEM-D with different methods, estimate the CI.