Table 3. Robustness check in study of the percentage of local tax revenue coming from TCT revenue and participation of local governments (n = 163) in an anti-smoking programme, Republic of Korea, 2002–2003.

Model estimation		Binomial logit				Multinomial logit
Dependent variable		y = 1 if participated				Anti-smoking [versus NP]	Othera [versus NP]
		y = 0 if otherwise
Explanatory variableb		Coefficient (95% CI)	Coefficient (95% CI)	Coefficient (95% CI)		Coefficient (95% CI)	Coefficient (95% CI)
Percentage of local tax revenue derived from the TCT		–0.98**	–0.58**	–		–1.04***	–0.30
		(–1.73 to –0.22)	(–1.14 to –0.04)			(–1.76 to –0.33)	(–1.23 to 0.62)
Per capita local tax revenue (in wons of the Republic of Korea × 100 000)		–0.48	–	0.20		–0.45	0.06
		(–1.09 to 0.13)		(–0.29 to 0.68)		(–0.92 to 0.02)	(–0.66 to 0.78)
Pseudo R squaredc		0.0950	0.0884	0.0706		0.0646
Log pseudolikelihood		–85.12	–85.74	–87.41		–153.07
Likelihood ratio testd χ² (df = 7)		–	–	–		3.27

*P < 0.10; **P < 0.05; ***P < 0.01.
CI, confidence interval; df, degrees of freedom; NP, no participation; TCT, tobacco consumption tax.
^a The other category collectively includes the promotion of physical activity, good nutrition and moderation in drinking.
^b Other explanatory variables include population size, public health centre staff size and being a county or not.
^c A measure of goodness-of-fit for non-linear models, including logistic models.
^d The null hypothesis is that the other category can be pooled into the NP category. For this test, robust standard errors were not used.