The diagram shows 9 scatter plots of scaled bias in a 3-by-3 array. Each scatter plot shows the scaled bias for the discrete likelihood method on the x-axis versus the scaled bias for logit-transformed data on the y-axis, for a number of points corresponding to different true proportions (.001, .005, .01, .05, .1) and sample-size scenarios (small, medium, large, and mixed). The rows of the array, from top to bottom, correspond to 5, 15, and 30 studies, respectively. The columns of the array, from left to right, correspond to the following ranges of expected counts: less than 1, between 1 and 5, and greater than 5.

Figure 19Comparison of proportion bias: approximate method (logit transformation) versus discrete likelihood

Rows correspond to number of studies. Columns correspond to ranges of expected count. Points on the dashed line have equal proportion bias for both methods. Points indicate simulation scenarios, and are coded with two characters, a letter (indicating sample size scenarios; S=small, M=medium, L=large, X=mixed); and a number (indicating the true proportions; 1=0.001, 2=0.005, 3=0.01, 4=0.05, 5=0.1). Black color indicates scenarios with small heterogeneity, and red scenarios with large heterogeneity. Only a representative subset of scenarios is plotted. The solid black reference lines indicate 0 bias. Note the change in scale across columns.

From: Results

Cover of Simulation-Based Comparison of Methods for Meta-Analysis of Proportions and Rates
Simulation-Based Comparison of Methods for Meta-Analysis of Proportions and Rates [Internet].
Trikalinos TA, Trow P, Schmid CH.

NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.