Estimating micro area behavioural risk factor prevalence from large population-based surveys: a full Bayesian approach

Table 2 Design- and Model-Based Prevalence Estimates for Behavioural Risk Factors in the Erie-St. Clair Region, by County

	Lambton		Chatham-Kent		Essex		Erie-St. Clair Region
	Males % (95 % CI)	Females % (95 % CI)	Males % (95 % CI)	Females % (95 % CI)	Males % (95 % CI)	Females % (95 % CI)	Males % (95 % CI)	Females % (95 % CI)
Current smoking
Design-based^a	27.5 (25.2–29.9)	22.3 (20.2–24.4)	29.2 (26.7–31.6)	23.0 (21.0–25.1)	23.5 (21.8–25.2)	21.8 (20.1–23.4)	25.1 (23.8–26.4)	22.1 (20.9–23.3)
Model 1^b	28.5 (27.8–29.2)	20.6 (19.9–21.3)	29.2 (28.7–29.7)	21.9 (21.4–22.4)	25.8 (25.4–26.2)	21.3 (21.0–21.7)	27.1 (26.7–27.4)	21.3 (21.0–21.6)
Model 2^b	27.7 (26.8–28.6)	21.8 (20.8–22.8)	28.9 (28.6–29.2)	23.4 (23.1–23.7)	26.2 (26.0–26.4)	22.7 (22.5–22.9)	27.1 (26.9–27.2)	22.6 (22.5–22.8)
Excess bodyweight
Design-based^a	58.4 (56.0–60.9)	43.5 (41.0–45.9)	61.4 (59.3–63.6)	46.4 (43.9–48.9)	61.3 (59.2–63.4)	46.6 (44.5–48.6)	60.8 (59.3–62.3)	45.9 (44.4–47.4)
Model 1^b	57.9 (57.6–58.2)	44.6 (44.3–45.0)	60.3 (60.0–60.6)	46.6 (46.2–47.0)	61.5 (61.3–61.8)	45.7 (45.5–46.0)	60.5 (56.8–64.1)	45.6 (45.5–45.8)
Model 2^b	58.0 (57.8–58.2)	44.5 (44.3–44.7)	60.2 (60.0–60.4)	46.7 (46.5–47.0)	61.6 (61.4–61.7)	45.7 (45.5–45.8)	60.5 (60.4–60.6)	45.6 (45.5–45.7)

To obtain a design-based pooled prevalence estimate across all 5 CCHS cycles, the cycle-specific survey weights for each respondent were re-scaled based on the proportion that cycle contributed to the total weighted pooled population using Statistics Canada’s Bootvar program (1). The re-scaled weights were used to calculate the design-based prevalence estimates. The model-based prevalence estimates were aggregated to the micro area using post-stratification methods (see Additional file 1). The mean risk factor prevalence was calculated for each county, as well as the entire region (LHIN) for both models. Model 1 results are comparable to the design-based estimates, which do not adjust for income
a The interval of variation of the design-based prevalence estimate is expressed as a 95 % confidence interval (CI). bThe interval of variation of the Bayesian model-based prevalence estimate is expressed as a 95 % credible interval (CI).
Abbreviations: CCHS Canadian Community Health Survey, LHIN Local Health Integration Network
(1) Statistics Canada. Bootvar: User Guide (Bootvar 3.1 – SAS Version). Ottawa. 2005

ISSN: 1471-2458