We believe that there is little to be gained by arguing with Roland Auer and colleagues about whether diet or physical inactivity is the most important variable leading to obesity in Canadian children: together these factors determine caloric balance and therefore both are important. We advocated both healthy nutrition and physical activity throughout our paper.1 The “massive caloric intake we ‘enjoy’” is a problem only if we do not counter it with a proportional increase in physical activity. The comment by Auer and colleagues that “most data suggest that energy intake has increased over the past several decades” is perhaps based on selective information.2,3 Also, increasing energy expenditure produces multiple physiological and psychological benefits beyond maintaining caloric balance4,5 and these effects should not be ignored. Finally, significant problems exist in assessing physical activity and energy intake, and current techniques are clearly inadequate. The leap of faith required to accept “energy availability”6 as a legitimate surrogate for energy intake is large.
In response to Murray Finkelstein, in our study we used data derived from stratified random samples of the Canadian population.1 The sample designs, which are typical of surveys conducted by Statistics Canada, oversampled respondents in the smaller provinces, such that reasonably accurate estimates of provincial statistics can be obtained. Our analyses used the design weights provided by Statistics Canada, which take into account the stratified sampling design as well as potential bias due to nonresponse.7 Finkelstein's point regarding sampling variances may be valid, however, in that the sample of children for the National Longitudinal Study of Children and Youth was clustered within families, with up to 2 children sampled within each family. In our subsample, for example, about 40% of the children were members of a sibling pair. We estimated the sampling variances using hierarchical linear models to achieve more accurate estimates of the standard errors and found that they increased by only about 5% when within-family clustering was taken into account. We agree that this more complex approach is preferable, but note that its use has no appreciable effect on our results or conclusions. The bootstrap method suggested by Finkelstein is computationally intensive and has some undesirable properties. For surveys such as these, which are derived from stratified samples and where individuals are clustered within higher level units such as families or schools, approaches based on balanced repeated replications provide a simple, robust approach to estimate sampling variances8 and are generally preferable to bootstrap techniques.9
We estimated secular changes in the prevalence of overweight and obesity using well-established guidelines.10,11,12 Peter Katzmarzyk suggests using new guidelines to define overweight and obesity13 that were published after our paper had been submitted to CMAJ. We agree with Katzmarzyk about the value of these new definitions. We believe the important finding in our study is that the prevalence of childhood overweight or obesity, however defined, is increasing rapidly. Katzmarzyk points out that when using the method proposed by Cole and colleagues,13 the magnitude of the problem may be smaller than we reported, but the rate of change of the problem may in fact be larger than we reported. Difficulties in establishing acceptable definitions for childhood overweight and obesity are not new.14 The findings in Katzmarzyk's letter will facilitate future research in this area.