Measurement of outcomes, mediators, and moderators in behavioral obesity prevention research

Measurement enables intervention scientists to determine whether their interventions had the intended outcome effects and the expected pathways of effects across mediating variables. Low reliability of measurement (i.e., substantial random error) attenuates the relationships of these measures to other variables, including treatment effects. This attenuation may indicate that interventions were not effective, when in truth they were. There has been little assessment of the quality of measurement in obesity prevention trials.

A brief review is provided of measurement issues within classical test theory. Seventeen obesity prevention trials were then identified in the literature. The extent to which the reliability and validity of the measures were reported in the articles was assessed.

With few exceptions reliability coefficients of measures were not reported in the obesity prevention literature. When they were reported, there was evidence that low reliability attenuated reported intervention outcome relationships.

The quality of measurement is important in intervention science and consequently should be clearly presented in scientific reports of outcomes. Better measures are needed in obesity prevention to provide appropriate tests of state of the art interventions. A brief overview is provided of each of the articles in this special issue on measurement in the Girls health Enrichment Multisite Studies (GEMS).