Hi Merry Try in the document section in the link below: http://www.brixtonhealth.com/ Rogers

Thanks for this Rogers. I did have these, but they're summary/introductory and I'm trying to get something a bit more detailed. Do you know of anything?

Try: [url]http://www.brixtonhealth.com/CSASCoverageMethodSimple.pdf[/url] That is a simple guide. Try: [url]http://www.brixtonhealth.com/CRCaseFinding.pdf[/url] That deals with the validation of the case-finding method by capture-recapture. Try: [url]http://www.brixtonhealth.com/SampleCSAS.pdf[/url] That deals with sample size requirements. The links: [url]http://www.brixtonhealth.com/opencsas.html[/url] [url]http://www.brixtonhealth.com/CSASCoverEmpty.xls[/url] [url]http://www.brixtonhealth.com/CSASCoverExample.xls[/url] Provide estimation and mapping software. The peer-reviewed reference is: [url]http://www.who.int/bulletin/volumes/83/1/20.pdf[/url] I hope this helps. VALID may have something. I know that Yohanna and Lio both had manuals / guides that they developed for training NGO staff.

Thanks Mark. I've looked at most of these but will look at the new ones. One additional question about CSAS. In the Valid article "Notes on required sample sizes for CSAS..." it says that calculating a sample size for each quadrat may not be feasible since prevalence and coverage might vary so much between them. So how do I figure how many communities (total population) need to be surveyed within each quadrat? And if I survey different numbers of people in different quadrats, doesn't that kill the random systematic thing that prevents me from having to use a design effect mulitplier? Do I assume homogenous coverage for the sake of planning even though we know that prevalence is rarely homogenous? Or do the enumerators keep adding communities until they have a minimum number of cases in a particular quadrat? Thanks

I can tell you what I do ... I have a rule that says we sample the number of villages that a single survey team can do in one day. Since the sample is "centric" the sampled villages will tend to be close to each other. This means that it is usually feasible to sample between three and six villages from each quadrat when using active and adaptive case finding (fewer for house-to-house screening). Teams just continue sampling until they run out of time. I am mostly interested in the per-quadrat estimates (e.g.identfying holes in coverage). The overall estimate only means something real when coverage is homogenous. When we first used CSAS we did weight results by the populations of the villages sampled (individually or by quadrat) and used a weighted estimator and a bootstrap estimator but found (after several surveys) that this made remarkably little difference to the estimate or width or the confidence interval from when we treated the overall sample as a simple random sample. I suppose we could have expected this from: Milne A, The centric systematic area sample treated as a random sample, Biometrics, 1959;15:270-297 The method does not assume homogenous coverage over the wide area only at the scale of the quadrat. So keep the quadrats small. Sample size calculation is complicated by the low population size (i.e. there will usually only be a small number of SAM cases). I don't do many "CSAS" surveys now. I tend to use SQUEAC os SLEAC. Both use CSAS / grid sampling but in different analytical framework (SQUEAC uses a Bayesian estimator, SLEAC uses an LQAS classifier). Sample size requirements for these methods are much lower (i.e. typically n = 40 or lower). I have done some work on improving CSAS using a rectangular grid and triangular estimation / classification areas with extensive re-use of data from neighbouring PSUs. This requires fewer PSUs but delivers finer maps and better local estimates. This method will be piloted over very wide areas in October this year. CSAS type samples have also been used in the informant method for mortality estimation. Is this any help?

That does help. Thanks alot. Merry