I am using sampling to estimate characteristics of a population. My sampling ratio is 32. I take N/32 samples, calculate their sum and multiply by 32. Then I compare it to actual value to get error. It is 3.67%.
Then I take sampling ratio as 64 and do the same and get error as 3.62%. Generally one would expect a higher sampling-error.
Is that possible. Thanks for your reply.
It is certainly possible. You could be luckier in your sampling of 1 in 64 and have no error at all. Imagine a population of 640, with 600 of value 0, 20 of value -1 and 20 of value 1. The average is 0. If your 10 samples are all 0, the average will be exactly right. If your 20 samples (for the 1/32 case) include a couple of 1's and no -1's there will be an error.