Stratified random samples - Your goal is to match relevant subgroup proportions. Then, choose randomly in the subgroup. I've seen survey firms go a little nuts with this technique. A client wants a representative sample and the survey company talks the client into using a stratified random sample. However, due to expense, only 50 folks are sampled. Given the demographics of the area - this resulted in only three African Americans in the sample. Can you really generalize about the opinions of African Americans with a sample of three?

With this technique,you also have to be careful about the variables that you use for the stratification. You might be interested in the role of race in jury selection issues. However, some studies found that income was just as important. Did you also stratify on that?