Policy and Complex Systems
ters . We can see that there is a sharp turn when the number of cluster is equal to three . Therefore , three appeared to be a reasonable number of clusters to divide into the agent types .
Calinski Criterion
Another popular method for this purpose is the Calinski Criterion ( also known as the Pseudo F statistics ). Figure 6 shows the results of applying Calinski Criterion to our data . The Calinski Criterion suggests that we should also use three clusters .
Majority Rule
Third , we applied 26 other indices on the same problem and used the majority rule to select the number of clusters . We consider up to 10 clusters as the possible number of clusters into which we could group agents . As shown in Figure 7 , the Y-axis means the frequency that a number is selected as the best number of clusters chosen by the indices , and the X-axis is the possible best number of clusters . Eleven out of the 26 indices selected three as the best number of clusters . Therefore , according to the majority rule , we will assume three is
Figure 5 . Within groups sum of squares versus the number of clusters .
Figure 6 . Calinski Criterion results .
175