ACS Data Users Group

Hi,

I am creating 1-year estimates for American Indians for each state. Is there a sample size that is considered too small to create estimates from? My smallest frequency that I am creating estimates from is 941.

Thank you!

KIMBERLY: I work extensively with AI/AN data. Can't answer your specific question, but Census confidentiality parameters may be involved. What I do know is that sampling error is really important in looking at any ACS AI/AN data. I always compute the coefficients of variation for any counts or characteristics data I use from the ACS. NORM

Thank you, Norm!

For the most part, it is for you to decide -- since you are in academia, you need to figure out what is "passable" with reviewers and editors. In more practical terms, you would need to see what kind of impact the precision of your estimates and/or the power of your tests have on the stability / defensibility of your findings.

For most practical reasons, the sample of 1000 is plenty. Most political polls end up with that sample size, and are entirely content with that.

Are you sure 941 is the *actual* frequency, not the weighted total (which would then be based on a much smaller number like 20 or 30)?