Murray Finkelstein raises issues similar to those raised by peer reviewers for Human Resources Development Canada and CMAJ. We persuaded them that weighting was inappropriate for the comparison of hungry subgroups and for the comparison of hungry and nonhungry groups because we were not presenting population estimates.1 We did originally run all of our results from the 1994 National Longitudinal Survey of Children and Youth (NLSCY) using weighted estimates, and many more analyses were significant. Weighting would have overstated the power of our tests, whose results were mainly determined by the groups with the smallest degrees of freedom.
The issue of weighting data when comparing subgroups that together comprise the whole sample is contentious. Weighting is designed to correct for inequities in the whole sample, not segments of it. We could have weighted each segment separately so as not to exceed the actual sample size, but the gain would have been marginal. When we compare subgroups that together are only a portion of the whole sample (e.g., occasional v. frequent hunger), there is much less call for weighting. Given the disturbing nature of our results, we felt it prudent to err on the side of underreporting associations rather than to report all positive results using weighted analyses.
In the second cycle of the NLSCY, collected in 1996, circumstances required a large proportion of the sample to be dropped; because of this change, we employed both sample and longitudinal weights for the analyses in this cycle. However, only cross-sectional and longitudinal weights are calculated in the NLSCY file; bootstrap methodology is not used. In a forthcoming follow-up study of hunger in NLSCY families we compared adjusted prevalence rates of hunger in 1994 and 1996. The population estimates of child hunger in Canada using weights were 1.4% (53 995 children) and 1.6% (75 615 children) for 1994 and 1996 respectively.
Bruce Leistikow's argument is based on the presumption that the associations observed are causal; of course, such a relationship cannot be inferred from cross-sectional data. We offer another analysis: smoking in the primary caregiver is associated with hunger in the primary caregiver. Smoking is a coping mechanism for the physical and psychological stress of hunger.2 That being said, high tobacco taxes can be viewed as regressive taxes that target the tobacco-addicted poor,3 reducing family resources for food and other essentials. We can all agree on one thing: sensitive supports for tobacco cessation must be offered to low-income caregivers who smoke. Cessation will improve health and reduce the financial stress on households.