AH 7614 Average daily energy expenditure EE in kilocalories

Average daily energy expenditure (EE) in kilocalories per kilogram of body weight was the indicator of physical activity level. The CCHS asked respondents about their participation in various activities (e.g., sports, hobbies, exercise) in terms of frequency (within a given time period) and duration (average duration). A metabolic equivalent (MET) value was assigned to each activity as an indicator of intensity. To calculate daily energy expenditure, the MET value of each activity was multiplied by N (the number of times a respondent engaged in the activity in the past 12 months) and D (the average duration of the activity in hours), then divided by 365. Due to the highly right-skewed distribution of the daily EE variable, the data were ranked in ascending order and grouped into 10 deciles for the analysis. Sense of belonging to the local community on a 4-point scale was selected as the indicator of an individual\'s level of social cohesion, conceptualized as connectiveness to his or her communty. Respondents were asked “How would you describe your sense of belonging to your local community? Would you say it is…?” The response options were very strong, somewhat strong, somewhat weak, or very weak, coded as 4, 3, 2, and 1, respectively. In the analysis, sense of belonging was treated as a continuous variable to avoid the loss of variation in data associated with the categorization of variables (Lovasi et al., 2012). In the multilevel models, community-level social AH 7614 was defined as the average score for sense of belonging within a community. Resultantly, communities with a high level of social cohesion are those in which a large proportion of residents reported having a strong sense of belonging to the local community. Community was defined as what some refer to as a neighbourhood, a geographical unit in which the circumstances are shared by residents (Chaskin, 1997). Using neighbourhoods with pre-determined geographical boundaries is advantageous in public health research because it allows for the analysis of health data from secondary sources such as the CCHS that also include data pertaining to areas within these boundaries (Weiss, Ompad, Galea & Vlahov, 2007). Communities were represented by Forward Sortation Areas (FSAs), geographical units defined by the first three characters of a postal code in Canada (Statistics Canada, 2008a). Using FSAs as the geographical unit was appropriate because they are larger than full Postal Codes which often include only one street block, but smaller than Census Subdivisions, the next largest geographical areas, which frequently include entire municipalities and therefore may be too large to represent communities in urban settings (Statistics Canada, 2015b). There was some concern that estimate of community-level social cohesion may not be meaningful in FSAs with very few respondents, so only FSAs with at least 5 respondents were incldued in the analysis. All analyses were performed in SAS version 9.3. In addition to accounting for the idea that an individual tends to be more similar to persons in the same neighbourhood than to those from other neighbourhoods, mulitlevel regression models allow for the testing of hypotheses that are multilevel in nature (Brauer & Mikkelsen, 2010). The first multilevel regression model (Model 1) was used to compute an intraclass correlation coefficient (ICC), which describes the extent to which data within a cluster are correlated (Park & Lake, 2005). The ICC in this analysis describes the proportion of variance in physical activity level that is attributable to communites. In the second multilevel regression model (Model 2), both the intercept and individual-level social cohesion were defined as random effects to allow for their effects to vary across communities (Bell, Ene, Smiley & Schoeneberger, 2013; Hayes, 2006) while the community-level social cohesion was entered as a fixed effect. Age, sex, household income, education and urban-rural status were included as control variables. All descriptive statistics and regression models were calculated using sampling weights provided in the CCHS, which was necessary to allow for estimates to be calculated from survey data that is representative of the population in Canada. In the CCHS, a survey weight is provided for each respondent, and corresponds to the number of individuals the respondent represents in the covered population (Statistics Canada, 2011). Missing data were filled in using multiple imputation, and 10 imputations in total were performed using the PROC MI procedure. Data were imputed for variables in order from those that have the lowest proportion of missing data to those that have a highest proportion of missing data (sense of belonging, education, household income). Frequency tables were produced for each imputation to verify that the imputed data are plausible in that all intervals were appropriate and that the imputed data fell between the minimum and maximum values for each variable. For each multilevel model, the relevant statistical model was fitted to each of the 10 imputed data sets, and the results were pooled using the PROC MIANALYZE procedure to obtain results that take into account the range of estimates from all 10 imputations.