A common challenge of scientific research is an inability to randomly assign subjects to a treatment that varies on only one variable (e.g., assignment of subjects to groups that differ only in a drug provided). The inability to randomly assign variation on a single treatment variable limits the strength of causal inferences made from the data—a common goal of scientific analysis. Dr. Greg Pavela, postdoctoral trainee at the Nutrition Obesity Research Center at the University of Alabama at Birmingham, and Dr. David B. Allison, distinguished professor and director of UAB’s Nutrition Obesity Research Center—in collaboration with UAB colleagues Dr. Howard Wiener, statistician II in department of epidemiology, and Dr. Kevin R. Fontaine, professor and chair in the department of health behavior—recently introduced the concept of packet randomized experiments (PREs), an experimental design that improves the internal validity of studies by eliminating a class of observed and unobserved potential confounders.
Drs. Pavela and Allison along with their colleagues introduced this design as an intermediary between purely observational studies and classic randomized trials. They then explored the inferential properties of PRE design and explained its application to a wide range of phenomena, including obesity, genetic linkage, and criminal sentencing disparities. Finally, the researchers described a situation in which confounders uncontrolled by packet randomization can be asymptotically controlled for by conditioning on certain covariates.
The team concluded that PREs can improve public health decisions insofar as well-designed PRE’s offer better estimates of causal effects than do ordinary association studies.
“Packet Randomized Experiments for Eliminating Classes of Confounders” is published in the December issue of the European Journal of Clinical Investigation.
Journal article: http://onlinelibrary.wiley.com/doi/10.1111/eci.12378/abstract