The answers to these questions depend on two conditions. One is the arithmetic involved. Obviously, the final number of 150 is going to be a product of the number of primary and secondary clusters chosen and the number of households selected within each secondary cluster.   In any real situation, many combinations of these numbers will produce the desired number for the final sample. Fortunately, several guidelines help in making these decisions. First, select as many clusters at each level - primary, secondary, and, if used, tertiary (third level down) as your resources will allow. We will make this point clear shortly. This will spread the sample over the largest possible area. The reason behind this rule is simple: As smaller areas are selected, populations generally become more similar. Households or whatever is being sampled in the last set of clusters generally will be more alike and differ somewhat from households in more distant clusters.To ensure that the final sample best represents the population from which it was selected, we want to get as much variation or heterogeneity as possible. Therefore, it is advisable to spread the sample as widely as possible at the beginning, by selecting as many primary clusters as possible and minimizing the number of sampling units selected in the final clusters. Figure 8.2 shows how this can be done.

Figure 8.2. Illustrative cluster sample design