Skip to main content
. 2018 May 18;18:41. doi: 10.1186/s12874-018-0492-z

Fig. 2.

Fig. 2

Simulation-based power estimates (based on 10,000 replications) for the planned IPD random effects meta-analysis* of 24 trials for detecting a treatment-BMI interaction when the true effect was − 0.1, conditional on a range of values of the between-study standard deviation (τλ) of the interaction effect, when either correctly analysing BMI as continuous or when wrongly analysing BMI as binary (≥ 30 versus < 30). *Based on using change score model (4) in each trial followed by random effects meta-analysis model (9). Standard = DerSimonian and Laird estimation, with p-values and CIs derived using standard normal-based method; HKSJ = DerSimonian and Laird estimation, with p-values and CIs derived using approach of Hartung-Knapp Sidik-Jonkman