One way is a simple random sample (SRS) which basically involves putting everyone's name into a hat and drawing out a list of names at random. Or the computer is the one doing the random selections.
Another way is to use cluster sampling. Each cluster would be the various stores. So if you had say 20 stores nationwide, then the company picks a few people at random from each cluster (store) and forms the sample that way. This method ensures that no store is left out, and could possibly be more representative compared to pure SRS which will pick any name at random. With SRS it is possible that some stores are over-represented while others are under-represented or simply represented not at all.
--------------
Side note:
An example of "over-representation" would be having a group of 90% males in a committee when the population at large is roughly 50% male, 50% female.
Under-representation is the complete opposite, so that example would be the 10% of females on the committee (when it should be closer to 50%)
