Fig. 2
![Fig. 2](http://media.springernature.com/full/springer-static/image/art%3A10.1186%2Fs13063-018-2491-0/MediaObjects/13063_2018_2491_Fig2_HTML.gif)
Distribution of \(\hat {\beta }_{T\times B}\) for scenarios with Σ=Σ3, β k = β v , and low censoring (a) or high censoring (b) for no biomarker–treatment interaction (βT×B= ln(1.0)=0, top rows) or qualitative biomarker–treatment interaction (βT×B= ln(1.33)=0.285, bottom rows). Scenarios for different numbers of potential prognostic variables are shown in different columns. The dashed red lines indicate the true value of βT×B, the blue triangles represent the observed confidence interval coverages and the green dots the observed probability for a type I error (a) or estimated power (b). AIC Akaike’s information criterion, qual. qualitative, Sig significance