Skip to main content

Table 2 Differences in coverage of the treatment effect on the outcome of interest by the 95% confidence interval and difference in mean of standardised absolute biases of the treatment effect on the outcome of interest. Default empirical calibration was performed with five negative controls and using the all-controls systematic error model

From: Assessing the effectiveness of empirical calibration under different bias scenarios

Difference Coverage

Difference in Mean of Standardised Absolute Bias

 

Ideal—Suitable negative controls

Random— Suitable negative controls

Random—Unsuitable negative controls

Ideal—Suitable negative controls

Random—Suitable negative controls

Random—Unsuitable negative controls

No calibration vs default empirical calibration

 Unmeasured confounder

0.48

0.60

0.05

−0.46

−0.23

0.00

 Model misspecification Quadratic term

0.24

0.27

0.08

−0.05

0.03

−0.01

 Model misspecification Interaction term

−0.09

0.04

0.02

−0.19

−0.09

− 0.02

 Non-positivity

0.02

0.00

0.01

−0.02

0.07

−0.01

 Measurement error

0.01

0.01

NA

−0.02

−0.02

NA

Default empirical calibration vs empirical calibration with the NULL systematic error model

 Unmeasured confounder

0.09

−0.06

−0.04

0.55

−1.56

0.00

 Model misspecification Quadratic term

−0.09

−0.13

− 0.08

0.20

0.35

0.02

 Model misspecification Interaction term

−0.52

−0.06

− 0.03

0.29

0.00

0.01

 Non-positivity

−0.02

−0.01

− 0.03

−0.01

0.09

−0.01

 Measurement error

−0.04

−0.02

NA

0.02

0.01

NA

Default empirical calibration (5 negative controls) vs empirical calibration with 30 negative controls

 Unmeasured confounder

0.05

0.06

0.01

0.85

−8.31

−0.36

 Model misspecification Quadratic term

0.00

0.00

−0.05

0.04

−0.11

0.08

 Model misspecification Interaction term

0.02

0.02

0.02

0.06

0.06

−0.60

 Non-positivity

0.02

0.03

0.00

−0.02

0.43

0.12

 Measurement error

0.01

−0.01

NA

−0.11

−0.06

NA