Explanation of the estimates: In 'practice' the estimates can be made in a variety of ways: old census data, knowledge from local populations, etc. In my code they are made with the make.estimates() function. This function takes the vector of actual populations and specifications of precision. These specifications are an upper and lower number (u and l), and the estimate for a cluster is a random integer between p_n-u*p_n and p_n+l*p_n (u or l can be negative to make sure that the estimate is always higher or lower than the actual). In the example I put up, u and p are both equal to .5.

I had originally gotten estimates by generating point patterns in R with lines marking where the clusters were and asking friends to make visual estimates of the population in each cluster, but with this method I only had a few test populations.

Yes, the things I'm interested in are seeing the validity and estimating variance. I'm also interested in looking at how estimation strategies affect the final population estimate. This is a lower priority and probably something I'll be able to work on outside of this tutorial, but its something I'd like to look at in this tutorial if there's time.

I see your definitions and the R code, and will take a closer look.

I don't understand from your definitions where the estimates in parens come from - and this seems pretty basic to the technique.

Nor am I sure what is done with the n (n

I'll look at the code later and see if I can understand what's going on, but some motivation or a more complete example (including where the estimates come from).

It sounds like your questions are

• Is this method valid?
• How should the variance be estimated?
  • And in particular, is the method you're using underestimating it?

Oct 13

So I've had a change this week. The part of my paper that was originally going to have the quadrat method is likely out, so I've started working on a different project, Scaled Estimation Sampling (SES), a sampling method I'm developing that uses ratio estimation to make estimates.

There are a couple of things that I'm hoping to do related to SES. The first, and more pressing one, is testing the validity of the method. I'm not sure what the best way to go about this is. A possibly related problem is variance estimation. So far I've been using the variance estimation from a method that uses the same estimator, but this is seeming to underestimate the true variance (found by simulating the method many times and taking the variance of the results).

The other project that I'm hoping to do is related, but more specific. The method involves making estimates of population groups and I am interested in seeing what constitutes "good" or "bad" estimating. Particularly I'm interested in seeing how precision (correlation between estimates and actuals) and accuracy (closeness of the estimates to the actuals) affect the final estimation.

I've put up my code for working with SES and a file defining the method and all variables/equations used for estimation.

http://cs.marlboro.edu/ courses/ fall2011/jims_tutorials/ dylan/ Oct_13
last modified Tuesday October 18 2011 9:42 pm EDT