Divide the population for whom you are estimating proportions into clusters, based on geographic location such as neighborhoods, schools or census blocks. Make sure that every member of the population is a member of exactly one cluster.
Select a random distribution of clusters from which to take samples. Select a higher percentage of clusters if you want a more accurate approximation of the proportions, but take into account logistical issues such as number of interviewers, transportation costs and time constraints.
Choose a random distribution of interview or survey sites within each cluster, the same number of sites in each cluster. As with the clusters, choosing more interview sites will increase the accuracy of the approximation, but to a lesser extent.
Take samples through interviews or surveys until you have reached the predetermined quota for your sampling.
Combine the proportion of responses in each category across all interview sites within all clusters to obtain an estimate of the proportions in the entire population. In addition to the number of clusters sampled, the accuracy of this estimate depends on the percentage of the population in each cluster that responded and the randomness of each cluster and site.