I'm trying to find a representative sample from a population of about 3100. I checked out sample-size calculators and a sample of about 1,000 seems to be the size I want for confidence/margin of error.

What I need to do is find a random sample that is representative, balancing a few categorical variables for sub-groups and also mean/distribution of a continuous variable that matches the population.

Are there any tools/techniques for doing this?

Am I over-thinking this and just a straight random sampling will work?

  • $\begingroup$ Your approach is inconsistent with the assumptions made by the sample size calculators, which (typically) are that (1) the sample is taken with replacement and (most importantly) (2) the sample is random. By not randomizing, you no longer have a valid basis to perform any of the usual statistical tests. Your approach reads a lot like quota sampling, a method that was popular in the early to mid 20th century but passed out of favor after several spectacular failures. $\endgroup$ – whuber Jun 27 at 12:36
  • $\begingroup$ Okay, so I shouldn't worry about needing to quota or stratify sample. You've said what I shouldn't do, but I'd like to clarify on what I *should& do: Are you saying I should instead just use normal random sampling and rely on having chosen an appropriate sample size to produce a valid result? Thanks! $\endgroup$ – Jim the fourth Jun 29 at 18:58
  • $\begingroup$ Stratified sampling is a legitimate technique and is useful for reducing sampling costs. It's used all the time, especially in large complex studied. Regardless, it's a little unusual for sample size calculators to give accurate estimates when you're taking a large sample from a small population, so make sure the calculator is doing what you think it is. $\endgroup$ – whuber Jun 29 at 19:26

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.