Table 3.

Multilevel Regressions: Effect of No Obesity-Prevention Service in Jurisdiction (q67f) on Obesity

	Pooled Sample				Men				Women
Dependent Variable (DV)	CE	p-Value	SE	PM	CE	p-Value	SE	PM	CE	p-Value	SE	PM
All-income levels
	N = 415,348				N = 166,264				N = 249,084
DV: Obesity	0.0149	<.0001	0.0017	0.0027	−0.0457	<.0001	0.0024	−0.0079	0.0686	<.0001	0.0025	0.0118
DV: Morbid obesity	0.2267	<.0001	0.0044	0.0049	0.0855	<.0001	0.0071	0.0015	0.2690	<.0001	0.0055	0.0064
Low-income sample (income <$35,000)
	N = 118,972				N = 40,689				N = 78,283
DV: Obesity	0.1051	<.0001	0.0034	0.0232	0.1110	<.0001	0.0054	0.0218	0.1133	<.0001	0.0046	0.0244
DV: Morbid obesity	0.4986	<.0001	0.0074	0.0161	−0.0082	0.4785	0.0115	−0.0082	0.7705	<.0001	0.0087	0.0352

Control variables include the following: at individual-level, household size and its squared term, number of children and its squared term, 5-year interval age indicators, gender (when pooled), marital status, education, employment, income, race, interview month; at county/local health department level, health department organization, county urbanization code, county-level median household income variable; and an indicator of year 2005 (baseline: year 2004).

The model used is a multilevel logistic regression with three levels: individual, county, and state. State specific intercepts are assumed to be fixed.

CE, coefficient estimate; DV, dependent variable; PM, predicted marginal (calculated as the difference in the changes of the predicted probabilities of [morbid] obesity from 2004 to 2005 between those residing in counties with a positive response to q67a and those residing in counties with a negative response).