Background: Two strategies to resolve the problem of under- or overreporting of tobacco use among adolescents have been utilized: (a) objective measures for validating self-reports and (b) procedures for improving validity of self-reports, such as the pipeline procedure. The objectives of this article are to investigate the hypothesis that reporting biases may be related to intervention status and to examine what effect such biases would have on interpretation of treatment effects.
Method: A two-by-two factorial design was used, with the first factor a pipeline manipulation consisting of pipeline versus control condition, and the second factor treatment status, consisting of treatment versus reference schools. Within each of the schools, half of the 9th-grade classrooms were randomly assigned to a pipeline condition and half served as controls. Analysis was conducted with school as the unit of analysis.
Results: The main effect for pipeline condition and the significant interaction between treatment and pipeline conditions were not significant. However, the pipeline manipulation did have an effect on the difference detected between treatment and reference schools; 4.3% difference between treatment and reference schools in the control condition versus 9.9% difference in the pipeline condition, both in the direction of a treatment effect. Using saliva thiocyanate as an objective measure of smoking status suggested differential false negative reporting where students in the reference community falsely claimed to be nonsmokers more frequently than in the treatment community (10.04% versus 5.96%).
Conclusions: The reporting bias assessed by the pipeline procedure alone appears to have masked treatment outcome effects. Adjusting the smoking-dependent variable for false negatives seems to have increased the treatment outcome effects even further. This result is contrary to the expectation that the treatment community would experience greater demand pressures to underreport their smoking behavior. Further investigation to address response biases in intervention studies is warranted.