Table 4

Adjusted ORs and 95% confidence limits (CLs)* for the impact of survey design elements on the 2000 Behavioral Risk Factor Surveillance System (BRFSS) underestimating versus overestimating the Census for factors included and not included in the BRFSS adjustment weight for non-response/non-coverage (NR/NC)

Factors included in adjustment weightFactors not included in adjustment weight
Survey design element% Women AOR (95% CL)% White AOR (95% CL)% Age 18–34 y AOR (95% CL)% Hispanic AOR (95% CL)% Income <$15000 AOR (95% CL)% Married AOR (95% CL)
County level
BRFSS response rate (ref≥60%)
 <40%2.11 (1.10 to 4.09)0.31 (0.11 to 0.88)2.16 (1.09 to 4.27)0.97 (0.30 to 3.19)2.94 (0.51 to 17.13)0.41 (0.13 to 1.25)
 40–60%1.04 (0.90 to 1.20)0.76 (0.62 to 0.94)0.92 (0.81 to 1.04)1.32 (1.07, to .64)1.23 (0.92 to 1.65)0.84 (0.69 to 1.01)
Data collection method (reference contracted out)
 In-house1.24 (0.93 to 1.67)0.49 (0.23 to 1.05)1.57 (1.00 to 2.47)1.02 (0.31 to 3.36)0.57 (0.18 to 1.81)1.00 (0.57 to 1.76)
Sampling fraction (ref≤median)
 >Median (0.14)1.02 (0.80 to 1.31)0.76 (0.32 to 1.82)1.00 (0.65 to 1.55)0.70 (0.26 to 1.90)6.10 (1.68 to 22.19)0.70 (0.38 to 1.28)
Number of call attempts (ref≤5 call attempts)
 >5 call attempts0.88 (0.63 to 1.23)1.00 (0.46 to 2.17)1.59 (1.02 to 2.49)0.82 (0.29 to 1.90)1.44 (0.53 to 3.89)1.29 (0.68 to 2.44)
  • * Robust polytomous logistic regression used controlling for Census region and telephone coverage (results not reported). The outcome variable was BRFSS overestimates Census (BRFSS prevalence adjusted for NR/NC >2.5% points higher than Census– reference category) versus BRFSS underestimates Census (BRFSS prevalence adjusted for NR/NC >2.5% points lower than Census) versus BRFSS approximates Census (BRFSS prevalence adjusted for NR/NC within 2.5% points of Census). Analyses comparing underestimates to overestimates shown.

  • Number of counties available for county-level analyses varied due to missing data and minimum sample size requirements (gender: n=732; race: n=730; age: n=732; ethnicity: n=729; income: n=638; marital status: n=730).

  • Interpreted as relative to counties with BRFSS response rates ≥60%, counties with BRFSS response rates <40% had twice the odds of under versus overestimating % women.