A Panel Selection

We examine data from the following 11 real-world sortition panel selection instances, generously provided to us by several groups that specialize in organizing citizens' assemblies. Table 1 shows the instance short-names we use throughout the paper, and which organization was responsible for each panel. The final two columns compare the values of our theoretical upper bounds on the marginal discrepancy, illustrating that in all instances except "obf", the bound from Theorem 3.3 is tighter. Finally, we give some metadata about each instance, which is required for calculating the values of our theoretical upper bounds. In particular, n = number of pool members, k = number panel members, C = set of distinct realized feature-vectors in the pool.