Table 6.

Tests for inconsistency for full design-by-treatment interaction model and various Lu and Ades models for smoking data set. Model ‘ABCD’ follows the algorithm as specified in Section 3.3. Other models change the order of the treatments as they appear in the algorithm. For this particular dataset, the models grouped in rows turn out to be identical models

	Inconsistency d.f.	Chi-squared	p value	Heterogeneity standard deviation (τ)
Full design-by-treatment interaction model	7	5.11	0.65	0.74
Lu–Ades models for different treatment orderings
ABCD, ABDC, BACD, BADC	3	0.67	0.88	0.72
ACBD, ACDB	3	1.30	0.73	0.70
ADBC, ADCB	3	0.75	0.86	0.71
BCAD, BCDA	3	3.52	0.32	0.69
BDAC, BDCA	3	0.76	0.86	0.69
CABD, CADB, CBDA, CBAD, CDAB, CDBA	3	3.89	0.27	0.68
DABC, DACB, DBCA, DBAC, DCAB, DCBA	3	0.60	0.90	0.73