Skip to main content

Table 3 The proportion of large ORs in each estimation procedure under different models

From: Laplace approximation, penalized quasi-likelihood, and adaptive Gauss–Hermite quadrature for generalized linear mixed models: towards meta-analysis of binary outcome with sparse data

Models

Model 1

Model 2

Model 3

Model 4

Model 5

OR = 1 (τ=0.2)

 LA

0.22%

0.32%

0.00%

0.29%

0.3%

 PQL

1.88%

83.74%

0.09%

84.41%

80.5%

 AGHQ

1.25%

14.79%

1.17%

14.39%

10.22%

OR = 2 (τ=0.2)

 LA

0.03%

0.43%

0.00%

0.43%

0.43%

 PQL

1.03%

83.01%

0.15%

83.76%

79.63%

 AGHQ

0.03%

0.50%

0.03%

1.28%

0.27%

OR = 3 (τ=0.2)

 LA

0.00%

0.68%

0.00%

0.66%

0.70%

 PQL

0.96%

82.55%

0.31%

83.03%

79.01%

 AGHQ

0.72%

4.98%

0.69%

5.39%

4.52%

OR = 4 (τ=0.2)

 LA

0.00%

1.00%

0.00%

0.94%

1.02%

 PQL

0.74%

81.15%

0.41%

82.06%

77.49%

 AGHQ

0.48%

4.90%

0.59%

4.78%

4.07%

OR = 5 (τ=0.2)

 LA

0.00%

0.74%

0.00%

0.74%

0.77%

 PQL

0.50%

80.35%

0.21%

81.38%

76.24%

 AGHQ

0.01%

0.37%

0.02%

0.71%

0.21%

  1. All the results were based on 20,000 iterations