Here is one resource, available on-line, with a relevant section on sampling (including when accurate maps are not available): Gorstein J, Sullivan KM, Parvanta I, Begin F. Indicators and Methods for Cross-Sectional Surveys of Vitamin and Mineral Status of Populations. The Micronutrient Initiative (Ottawa) and the Centers for Disease Control and Prevention (Atlanta), May 2007. Though the title indicates micronutrient status surveys, the section on sampling is relevant for other surveys as well. Please Google the title, you should find it on-line.

Not sure that you really need random sampling. Some form of systematic sampling might do. Take a look at: [url]http://www.brixtonhealth.com/ASTRA.mov[/url] If this suits then I can send the paper documentation. You may find this useful: [url]http://www.brixtonhealth.com/urbanSampling.pdf[/url] I hope this helps.

Dear Mark, dear Mary, thank you so much for your input! Mark, the epidemiologist I discussed with heavily advised against systematic sampling. She said it is too much prone to bias. So can you tell me why you think it would be ok to to it? I very much liked the video and this method seems very feasible in the given setting. Could you send it to me in written form, too? I just think that perhaps the sample size is too big for this approach. We have a population of about (!!) 30.000 inhabitants / 8.000 households and I have calculated a sample size of 1065 children in the age 0-24 months. And from the satellite pictures not more than 10 "subdivisions" make sense. Do you still think the method in the video is feasible? I am sorry if my questions are terribly naive. Kind regards Jenny

From Brad Woodruff: Dear Ms. Dörnemann: I would recommend the survey manual published by WFP. It is available at [url]http://www.unhcr.org/45f6abc92.html[/url]. Although there are many good survey manuals available from many organizations, they frequently recommend some variant of the "find the center/spin a pen" household selection method popularized by the Expanded Programme on Immunization back in the early 70s. Both theory and evidence indicate that this method potentially injects sampling bias into household selection. Moreover, this method also allows the survey teams doing the household selection too much leeway to decide which households to select, and this always increases the risk of sampling bias. For these reasons, I recommend against it. In your situation, it sounds like you might be able to use Google Earth to delineate segments of the population using boundaries which can be identified on the ground, for example, roads, streams, fences, etc. The segments should be as small as possible. You then estimate the relative size of the segments, as measured by the number of households, so you can randomly select 25-30 of them using probability proportional to size. In each selected segment, you ask local informants to help you construct a list of all households currently living in that segment. Then you can use systematic random sampling or simple random sampling to select the required number of households from the list. If you would like additional information, you can consult a web-based course on epidemiologic methods in emergencies set up by the London School of Hygiene and Tropical Medicine: [url]http://conflict.lshtm.ac.uk/[/url]. I hope this is useful. Regards, Brad Woodruff

Jenny, Sorry for the delay ... I'm on leave. You can use systematic sampling provided that you avoid the "gotcha" of a coincidence of your sampling interval with an aspect of population structure. If you take an even spatial sample (e.g. sample on a 10 km grid) you might find that you sample falls only in valleys or on hilltops. Take very second name in a list and you might find you select on males or only females. You can see that this sort of mistake can lead to a biased sample. Systematic samples are quite common when you don't already have good lists on which to base a simple random sample. Be careful and you should be OK. The ASTRA sampling guide is at: [url]http://www.brixtonhealth.com/astraSG.pdf[/url] Your sample size looks very large ... you must want high precision. I think that you may need to calculate your sample size allowing for a finite population. With a population of 30,000 you'd expect there to be about 8% between 0 and 2 years (2400). To estimate 50% with +/- 5% precision you'd need about n = 330 which would be less than about 500 households.