Skip to main content

Table 1 Three methods of fitting used to model the physicians’ acceptable differences of rates of events

From: A Bayesian non-inferiority approach using experts’ margin elicitation – application to the monitoring of safety events

1 Optionbetamixfunction: For each pair (j,k), application of the betamix function with 3 as maximal number of components of the finite mixture.
2 Option manual mixture of 2betaregfunction: The levels of the observed values of dj,k,e were dichotomized. Then, we fit 2 Beta distribution by applying the betareg function (or the equivalent betamixfunction with 1 as the number of components) on each level of dichotomisation. All levels of dichotomisation were compared, from that separating the two left values from the others, to that separating the two right values from the others. The two distributions were then mixed by applying the weights w1,j and w2,j=1−w1,j to each distribution. The weights w1,j(0,0.05,0.10,0.15,…,0.95,1) were tested. The models obtained with the different levels of dichotomisation and with the different weights were compared using the criteria for goodness of fit described in Section 1 of the Additional file 7. The fit with the lowest criteria was retained for the comparison with the other 2 methods.
3 Option manual mixture of abetamixfunction and abetaregfunction: A mixture of betamix function and manual mixture: We mixed: (i) a first Beta distribution obtained on the left level of dichotomisation (the one obtained with method 2), (ii) a mixture of a second and a third distribution, obtained by applying to the right level of dichotomisation the betamix function with 2 as the number of components. The weights given to those distributions were: (i) for the first distribution the w1,j was obtained through method 2, (ii) for the second and third distribution, the weights w2,j and w3,j were obtained through the ’betamix’ procedure, multiplied by (1−w1,j).