Skip to main content
Fig. 2 | BMC Medical Research Methodology

Fig. 2

From: Accounting for confounding by time, early intervention adoption, and time-varying effect modification in the design and analysis of stepped-wedge designs: application to a proposed study design to reduce opioid-related mortality

Fig. 2

Performance of models with and without fixed and random intervention-by-time interactions. Models are compared across scenarios listed in Table 2. First row compares intervention effect estimates ± empirical standard error. Horizontal gray line: true intervention effect θ= log(0.6). Second, third, and fourth rows compare models on coverage rate of 95% confidence intervals, Type 1 error rate, and power, respectively. Horizontal gray lines indicate 95% coverage rate, 0.05 Type 1 error rate, and a power of 0.80 in the second, third, and fourth rows, respectively. Covariance structure for cluster-by-time random effects: Hooper/Girling labeled by circles, unstructured labeled by triangles, and linear labeled by diamonds. Models which incorporate a discrete term for the main effect for time (fixed) are labeled by blue shapes; models which incorporate a linear term are labeled by red shapes. Models without intervention-by-time interactions displayed in left column, where blue shapes correspond to models 2 through 4 in “Methods” section. Models with intervention-by-time interactions displayed in right column, and correspond to models 5 through 10 in “Methods” section. Models 5 through 10 include a linear time effect for the fixed intervention-by-time interaction term

Back to article page