Data sources and analytical techniques

Data from the 2004 Canadian Community Health Survey (CCHS): Nutrition were used to produce overweight and obesity prevalence rates for adults aged 18 or older by selected demographic, lifestyle and socio-economic factors (see /concepts/hs-es/index-eng.htm#content ). The 2004 CCHS was designed to gather information at the provincial level on the nutritional status of the Canadian population (see /concepts/hs-es/index-eng.htm#content ). It does not include residents of the three Territories, Indian reserves and some remote areas, and regular members of the Canadian Armed Forces. The response rate was 76.5%. The height and weight of 57.5% of adults (18 or older) who responded to the survey were directly measured (see Limitations).

Overweight and obesity rates for American adults were estimated from the 1999-2002 National Health and Nutrition Examination Survey (NHANES). The NHANES obtained measured height and weight data for 9,488 respondents aged 18 or older.

Historical estimates of Canadian obesity rates, based on directly measured height and weight, are from the 1978/79 Canada Health Survey and the Canadian Heart Health Surveys that took place in different provinces during the 1986 to 1992 period. Rates based on self-reported data are from the 1985 and 1990 Health Promotion Survey; the 1994/95, 1996/97 and 1998/99 National Population Health Survey (NPHS); and the 2000/01 and 2003 CCHS.

The American and Canadian historical estimates in this analysis are based on weighted data.

Descriptive statistics were used to estimate the proportion of adults who were obese in relation to selected characteristics (Tables A, B and C). Directly measured height and weight data were obtained for 12,428 CCHS respondents aged 18 or older. Because they represented just 57.5% of adults who responded to the 2004 CCHS, an adjustment was made to minimize non-response bias. A special sampling weight was created by redistributing the sampling weights of the non-respondents to the respondents using response propensity classes. Variables such as province, age, sex, household income, race, education, physical activity, fruit and vegetable consumption and chronic conditions were used to create the classes. The classes were created with the CHAID (Chi-Square Automatic Interaction Detector) algorithm available in Knowledge Seeker¹⁷ to identify the characteristics that best split the sample into groups that were dissimilar with respect to response/non-response. This adjusted weight was used to produce all estimates in this analysis. Standard errors and coefficients of variation were estimated using the bootstrap technique, which accounts for the survey design effects.^{18, 19, 20}

	Table A. Overweight and obesity rates, by selected characteristics, male household population aged 18 or older, Canada excluding territories, 2004
	Table B. Overweight and obesity rates, by sex and selected characteristics, female household population aged 18 or older, Canada excluding territories, 2004
	Table C. Overweight and obesity rates, by selected characteristics, household population aged 18 or older, Canada excluding territories, 2004

The body mass index (BMI) distribution (Chart 2) was smoothed by calculating three-point averages. For example, the percentage of the population with a BMI of 23 was calculated by summing the percentage of people with a BMI of 22, the percentage with a BMI of 23 and the percentage with a BMI of 24, and then dividing the result by 3.

Chart 2. Percentage distribution of body mass index (BMI), household population aged 18 or older, Canada excluding territories, 1978/79 and 2004

Standard errors and coefficients of variation for estimates from the 1978/79 Canada Health Survey and the 1999-2002 National Health and Nutrition Examination Survey (NHANES) were estimated with SUDAAN, which uses a Taylor series linearization method to account for the complex survey sample design.

To compare obesity rates between surveys, the data were age-standardized to the 2004 CCHS using the direct method. The following six age groups were used:

Logistic regression models were used to determine if associations between obesity and fruit and vegetable consumption and leisure-time physical activity remained when age, marital status, education and household income were taken into account.

Separate logistic regressions for each sex were used to model having high blood pressure, diabetes, and heart disease in relation to BMI. The model included the following control variables: age, marital status, education, household income, smoking status and leisure-time physical activity. Respondents who were underweight or had missing information for education, smoking status and leisure-time physical activity were excluded from the models.