Stratified sampling is the process ofsampling from subgroups rather than the whole group.
For instance, you could sample 1000 adults from the USA. But, you could also first subdivide people into four geographical groups 1) Northeast, 2)Midwest, 3) South, and 4) West and then sample 250 people from each of those four groups.
Or, if you’re a brand manager, you might want to stratify your sample your total sample based on specific criteria. For instance, you might want tocreate two product usage groups and sample 250 people each from those who 1) have previously tried the product and 2) currently use the product.
But why would we do this instead of simply generating a random sample of everyone? Surely a random sample would give us a good idea of everyone’s opinion?
There are two important reasons why stratified sampling might be preferred.
1) Random sampling doesn’t ensure every group is represented. For all the good things that random sampling does, it cannot guarantee that every subgroup will be represented in similar proportions as they are in real life. You might get 10% more men, or 20% more blue-eyed people, or no business owners at all. Overall, random sampling does a fabulous job but sometimes, those rare blue moon events, you might get a completely unusable sample.
2) Sometimes subgroups are very small, perhaps only 10% of your entire population. First, that can make it very difficult for random sampling to even pick up members of the subgroup. Second, it is very difficult to make generalizations about a group if they comprise just 50 people out of your sample of 500 people. Using stratified sampling means that you canbe certain to collect a good sample of 200 completes from one subgroup and a good sample of 300 completes from another. Just don’t forget to weight the small group back to 10% of the total sample so that you can generalize the overall results!