Evaluation of the performance of tests for spatial randomness on prostate cancer data

Table 4 Global clustering test results with different adjustments.

Data/Model	Cuzick-Edwards k-NN		Moran's I		Tango's MEET
	k = 1%		k= 1%
	Test Statistic	p-value	Test Statistic	p-value	Test Statistic	p-value
Later Stage
1:Unadjusted data	173544	0.001	0.779	0.001	<10^-15	0.0001
2:Individual-level adjustments; no area-level random effects	170735	0.001	0.672	0.001	3.89 × 10^-15	0.0001
3:Individual-level adjustments; area-level random effects	170899	0.001	0.675	0.001	3.44 × 10^-15	0.0001
4:Individual- and area-level adjustments; no area-level random effects	168227	0.001	0.483	0.001	2.18 × 10^-08	0.0001
5:Individual- and area-level adjustments; area-level random effects	168684	0.001	0.489	0.001	7.17 × 10^-10	0.0001
Higher Grade
1:Unadjusted data	197019	0.001	0.478	0.001	1.36 × 10^-13	0.0001
2:Individual-level adjustments; no area-level random effects	195707	0.001	0.350	0.001	4.11 × 10^-08	0.0001
3:Individual-level adjustments; area-level random effects	195727	0.001	0.351	0.001	4.31 × 10^-08	0.0001
4:Individual- and area-level adjustments; no area-level random effects	195085	0.001	0.355	0.001	2.18 × 10^-08	0.0001
5:Individual- and area-level adjustments; area-level random effects	197598	0.001	0.701	0.001	<10^-15	0.0001

Higher values of the test statistic indicate more clustering for Cuzick-Edward's k-NN and Moran's I and less clustering for Tango's MEET. Test statistics can only be compared between models, not between the three methods.

ISSN: 1476-072X