Table B1 Frequency coverage of the true values by MP judged in simulation
Diallel effect50%90%95%99%
a42.5 [5.1]87.4 [12.5]93.7 [15]99.6 [19.7]
ϕa67.3 [2.8]98.4 [6.8]99.2 [8.1]100 [10.6]
m19.7 [4.8]68.5 [11.8]85.8 [14.1]98.4 [18.6]
βfemale99.6 [0.1]100 [0.6]100 [0.8]100 [1.3]
ϕm0.8 [0.4]96.9 [1.1]99.6 [1.3]100 [1.9]
b99.2 [0.7]100 [1.9]100 [2.5]100 [4.1]
ϕb100 [0.3]100 [0.9]100 [1.2]100 [2.1]
v94.1 [1.3]100 [2.7]100 [3.3]100 [4.5]
ϕv55.9 [0.7]100 [1.5]100 [1.8]100 [2.5]
w86.6 [1.5]100 [3.1]100 [3.8]100 [5.2]
ϕw0 [0.4]76.4 [1.1]96.1 [1.4]99.6 [2]
σ219.3 [1.6]58.7 [4]70.9 [4.8]88.2 [6.4]
  • Realized frequency coverage of the truth, and average width [in brackets], is given for estimated credibility intervals of TReVarP in 400 simulations. Coverage was examined for HPD intervals containing 50%, 90%, 95%, and 99% of the posterior. In this simulation, only a, m, and ϕa had nonzero contributions to treatment effect variability. Credibility intervals appear to have realistic coverage of the truth, although they tend to undercover the noise of σ2 contribution—likely due to the true model being a subset of the full model.