Skip to main content

Table 2 Simulation Study Results: Relative bias (Bias), RMSE, and 95% coverage (Cov.) of the risk difference under various spatial variances (rows) and estimation methods (columns)

From: Propensity score matching for multilevel spatial data: accounting for geographic confounding in health disparity studies

 

Non-spatial

Spatial

Unadjusted

Adjusted

Unadjusted

Adjusted

Bias

RMSE

Cov.

Bias

RMSE

Cov.

Bias

RMSE

Cov.

Bias

RMSE

Cov.

\(\sigma ^2_{u}=0\)

0.043

0.005

96

0.040

0.005

95

0.035

0.004

95

0.050

0.006

94

\(\sigma ^2_{u}=2\)

0.084

0.011

77

0.085

0.011

75

0.064

0.009

90

0.052

0.007

92

\(\sigma ^2_{u}=4\)

0.096

0.012

70

0.100

0.012

66

0.064

0.008

93

0.049

0.006

95

  1. \(\sigma ^2_{u}=0\) represents a non-spatial scenario that excludes county-level covariates \(V_{i}\) and \(U_{i}\). \(\sigma ^2_{u}=2\) and \(\sigma ^2_{u}=4\) represent spatial scenarios that include county-level covariates \(V_{i}\) and \(U_{i}\) where the spatial variance of \(U_{i}\) is either 2 or 4, respectively