Methods

View the most recent version.

Data source
Analytical techniques

Data source

This analysis is based on data from five cycles (cycles 2 to 6) of the National Population Health Survey (NPHS), conducted by Statistics Canada from 1996/1997 to 2004/2005.  Every two years since 1994/1995, the NPHS has collected data about health status, health behaviours and other determinants of health.  The survey is representative of the household residents in all provinces in 1994/1995.  It excludes the territories, Indian reserves, Crown Lands, health care institutions, and residents of Canadian Forces bases and some remote areas in Ontario and Quebec.  Although the NPHS also has an institutional component covering residents of long-term care institutions such as nursing homes, that sample was not analyzed in this article.

In 1994/1995, 20,095 household residents were selected to be members of the NPHS longitudinal panel.  Of these, 86.0% agreed to participate, yielding a sample of 17,276.  Response rates in subsequent cycles were 92.8% in 1996/1997; 88.3% in 1998/1999; 84.8% in 2000/2002; 80.5% in 2002/2003; and 77.4% in 2004/2005.  More detailed descriptions of the NPHS design, sample and interview procedures are available in other papers and reports.⁶ 

Data were collected primarily through computer-assisted personal interviews in 1994/1995, and primarily through computer assisted-telephone interviews thereafter.  To rule out the potential impact of collection mode on the results, only data from 1996/1997 to 2004/2005 were analyzed.  Telephone interviews comprised over 96% of all interviews in 1996/1997 and 1998/99; over 98% in 2000/2001 and 2002/2003; and more than 99% in 2004/2005.

Haut de la page

Analytical techniques

The sample for this analysis consisted of people who were aged 10 to 60 in 1994/1995.  Individuals were included in this analysis if they had reached age 18 by the beginning of a given interval (starting in 1996/1997), and were excluded if they were 65 or older at the end of a given interval.  For instance, a respondent who was 62 in 1996 would have been included in the 1996/1997-to-1998/1999 interval, but excluded thereafter.  Records for women who were pregnant at the beginning or the end of an interval were excluded for that interval.

For the first interval (1996/1997 to 1998/1999), there were 9,387 respondents aged 18 to 64 at the beginning and end; 318 cases were missing data on weight; and 203 women were excluded because they were pregnant.  Thus, the sample size for the first interval was 8,866 cases.  For the second, third and fourth intervals, there were 8,689, 8,098, and 7,517 respondents aged 18 to 64 at the beginning and end of the interval.  Of these, 220, 160, and 139 cases, respectively, were missing data on weight, and 156, 152, and 139 women were excluded because of pregnancy.  The resulting samples were 8,313, 7,786, and 7,239.  Additionally, for analyses using body mass index (BMI), cases with missing height were excluded:  7, 6, 12 and 8 cases for the first, second, third and fourth intervals, respectively.

Average weight change for the four two-year intervals from 1996/1997 to 2004/2005 was calculated for men and women.  To reduce the effect of outliers, individuals who gained or lost more than 25 kg were given a value of 25 kg.  For the first interval, 28 cases of weight loss and 23 cases of weight gain were truncated at 25 kg; for the second interval, 32 and 31 cases; for the third interval, 27 cases of weight loss and 42 cases; and for the fourth interval, 30 cases each of weight loss and weight gain were truncated.

To determine whether the rate of change in body weight was increasing or decreasing over time, multiple linear regression was conducted using a person-period dataset, predicting two-year weight differences from time, time squared, sex, age group and BMI category at the beginning of the interval.  Age group and BMI category were time-varying covariates.  Records for individuals missing data on any variable for a given interval were excluded, but this accounted for less than 4% of records for any interval.  As well, if respondents had complete data for another interval, the record for that interval was retained in the analysis.  Thus, for the regression model, the sample was 8,866 records for the 1996/1997 to 1998/1999 interval; 8,313 for the 1998/1999 to 2000/2001 interval; 7,786 for the 2000/2001 to 2002/2003 interval; and 7,239 records for the 2002/2003 to 2004/2005 interval.

To clarify patterns underlying observed differences in weight change, the proportion of people gaining weight, losing weight or remaining stable (no change in reported weight), as well as mean gain among those who gained weight and mean loss among those who lost weight, were calculated for each two-year interval.  Confidence intervals for the sex-specific prevalences of weight gain, weight loss and weight stability were calculated and prevalences were compared to determine if they differed significantly over time.  To test for the association between time and weight gain or weight loss, a linear regression was performed only on records where an individual experienced a gain or loss.

In order to take the complex survey design of the NPHS into account, the bootstrap method was used to generate confidence intervals of the estimates for all analyses.  Bootstrap weights for individuals were applied to each record for an individual.7  Significance was set at p <0.05, and the weights for the longitudinal square file were used to weight the records to reflect the Canadian household population in 1994/1995.  All analyses were conducted in SAS 9.