Skip to main content

Table 2 Comparison of model fit and sensitivity of detecting areas with an elevated risk

From: Evaluating the effect of neighbourhood weight matrices on smoothing properties of Conditional Autoregressive (CAR) models

Neighbourhood type

Kappa

Fraction

DIC

AUC*

Sensitivity of detecting SLAs with elevated risks (PP cut-off = 0.33)

Specificity

Queen-1

0.05

22%

2283

0.61

77%

41%

Queen-2

0.05

51%

2282

0.62

78%

43%

Rook-1

0.05

14%

2282

0.62

76%

41%

Rook-2

0.05

41%

2282

0.62

78%

39%

Distance

0.07

41%

2274

0.62

80%

30%

Weight-1 (1/distance)

0.07

41%

2213

0.62

80%

30%

Weight-2 (1/distance2)

0.12

98%

2204

0.61

75%

44%

Weight-3 (1/distance3)

0.08

93%

2218

0.59

75%

46%

Weight 4 (Gravity)

0.15

99%

2202

0.60

74%

39%

Weight 5 (Entropy)

0.08

91%

2227

0.60

74%

47%

Weight 6 (Density)

0.14

97%

2208

0.60

74%

37%

Weight 7 (Covariate)

0.10

95%

2210

0.62

74%

47%

  1. * Area under curve from ROC analysis