Simulating worst case sampling

Sampling strats and stats
1

Hi again,

Two posts in 5 minutes!

My next question is, if I am running some sample simulations, what would be the best way to simulate biased sampling?

Paul Hewitt has this feature in his EASS tool, and my understanding is that you can select samples from simulated exposures where the GM is in the top 25%, or 50% of exposures. The other alternative is to just filter all the simulated data for the top 25% and 50% of exposures, and then sample from that.

It seems like the GM approach is more realistic. Any other ideas?

Thanks!
Mike

2

This depend on your simulating software and associated flexibility.

I would simulate biased sampling through drawing random sample from a truncated lognormal distribution : e.h. drawing samples from the part of the distirbution above the 50th percentile, or above the 70th percentile. So you keep the original true distribution but simulate selection bias.

It would probably worse googling “how to simulate biased sampling”, which I haven’t done :slight_smile:

3

Thanks Jerome, I should be able to model both and compare!
Mike