OBJECTIVE: This paper promotes rigorous methods and designs currently underutilized in obesity research, informed by a recent systematic review of the methods and risks of bias in studies of policies, programs, and built environment changes for obesity prevention and control. METHODS: To determine the current state of the field, relevant databases from 2000 to 2017 were searched to identify studies that fit the inclusion criteria. Study design, analytic approach, and other details of study methods were abstracted. These findings inform recommendations for obesity researchers and the field as a whole. RESULTS: Previously identified were 156 natural experiment studies. Most were cross-sectional (35%), pre-post single group comparison (31%), or difference-in-differences designs (29%). Few used rigorous causal designs such as interrupted time series with more than two time points, propensity score methods, or instrumental variables. The potential relevance for obesity research is discussed, and recommendations for obesity researchers are provided. CONCLUSIONS: To strengthen natural experiment study designs and enhance the validity of results, researchers should carefully consider and control for confounding and selection of comparison groups and consider study designs that address these biases.
OBJECTIVE: This paper promotes rigorous methods and designs currently underutilized in obesity research, informed by a recent systematic review of the methods and risks of bias in studies of policies, programs, and built environment changes for obesity prevention and control. METHODS: To determine the current state of the field, relevant databases from 2000 to 2017 were searched to identify studies that fit the inclusion criteria. Study design, analytic approach, and other details of study methods were abstracted. These findings inform recommendations for obesity researchers and the field as a whole. RESULTS: Previously identified were 156 natural experiment studies. Most were cross-sectional (35%), pre-post single group comparison (31%), or difference-in-differences designs (29%). Few used rigorous causal designs such as interrupted time series with more than two time points, propensity score methods, or instrumental variables. The potential relevance for obesity research is discussed, and recommendations for obesity researchers are provided. CONCLUSIONS: To strengthen natural experiment study designs and enhance the validity of results, researchers should carefully consider and control for confounding and selection of comparison groups and consider study designs that address these biases.