🚀 go-pugleaf

Dear Readers;

I would greatly appreciate your feedback/comments/remarks etc
as well as any reference sources on the following issue.

I am recently involved in a simulation study to investigate the
degree of sampling variability and possible bias of estimates.
I am using a multivariate-normal random number generator to obtain
sample data with a specified population covariance matrix as an input.

I am considering to choose one of the following sampling methods to
generate a set of sample data (say 2000 samples, each having n=100)
and subsequently calculate sample statistics from each sample.

A) assume an infinite population size and;
  1) generate a sample size of 100 and repeat this 2000 times, or
  2) generate a sample size of say 200000 and randomly select
     a subsample of 100 for 2000 times without replacement, or
  3) the same as 2), but with replacement.

B) assume a finite population size and ;
  4) generate a population (1000000 values) and select a random
     sample of size 100, 2000 times without replacement, or
  5) the same as 4), but with replacement, or
  6) generate a population (say 200000 values) and select a random
     sample of size 100, 2000 time without replacement (thus,
     every value in the population would be selected exactly once), or
  7) the same as 6), but with replacement.

If I calculate a mean and variance, and confidence interval for the
population value based on 2000 replications, would there be substantial
differeces among the 7 options?  If so, how one could justify the 
use of a particular method in a simulation study?

  Many thanks for your time in advance.

Hanjoo Eom (e-mail: umsk@unixg.ubc.ca)
Department of Education
Univ. of British Columbia
Vanc. B.C., Canada.