1

This is a follow-up to a question of what to do if you get a sample that is clearly not representative [What if your random sample is clearly not representative? ]

I have a population of 800 in 25 cities in 5 geographic areas. I want to avoid the possibility of getting a random sample that omits a geographic area. I expect no differences between geographic areas or cities.

What are the pros and cons of using stratified random sample (to assure that some people are sampled from each geographic area) as compared with a purely random sample. (Having a sample without people from each geographic area would be politically problematic.)

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
Joel W.
  • 3,096
  • 3
  • 31
  • 45
  • I’m unaware of any substantial cons to stratified sampling. But what is the purpose of your sampling... what will you do with your sample? Do you actually want to do some kind of cross validation, or something else? – Wayne Jul 14 '19 at 03:23
  • I want to determine the proportion for a yes-not variable. – Joel W. Jul 14 '19 at 10:01
  • OK, I'd been misreading your statement and thinking you had 800 samples and were doing some kind of analysis. But you literally have a population of 800 and can only sample N of them. Is this a poll, where you may have to worry about non-responses? – Wayne Jul 14 '19 at 12:34
  • There will be no non-responses. Motivation is very high, and there is strong follow-up planned. – Joel W. Jul 15 '19 at 00:54
  • 1
    I see no cons for this. Stratified or block sampling is used in such cases, precisely to get a representative sample. – user2974951 Jul 15 '19 at 08:03
  • @user2974951 There is no analog to losing one df for each restraint? – Joel W. Jul 16 '19 at 11:10

0 Answers0