I am in the process of planning an anthropometric survey and looking at a number of options. I would be interested in hearing your thoughts as to how many out of 12 hardest hit district should be surveyed to assess the severity of malnutrition among children under 5? The districts vary in children under 5 age population size (2000 to 34000). What criteria to use for randomly selecting the districts?
Hi
Based on the sampling principle of representativeness of the process. I suggest that there are only 12 subjects which can go through a simple random selection and account for probability proportionate to size....here you can simply lotter and randomly select them base on time and resources availability. I recommend the design of simple random sampling for severity and how agent is the situation for immediate with the budget consideration.....otherwise consider systematic and simple random techniques( other things remain factor in;time and money).
Thank you in endavours
Answered:
8 years agoThe number of districts to select depends entirely on what population you want to generalize your results to. You should not select districts randomly because if you do, the district defines the cluster, and you will have fewer than 12 very large clusters. This will make your design effect very, very large, resulting in substantial imprecision in your overall estimate of the prevalence of whatever form of malnutrition you are interested in. The resulting level of imprecision will make your results useless to make program decisions, which of course is the reason you are doing a survey. For example, you may end up with an estimate of the prevalence of acute malnutrition of 11.4% with confidence intervals of 1.8 - 23.1%. An estimate like this is utterly useless and means you wasted all the resources and effort used to do the survey.
You should first decide what population you are interested in, then select 25-30 clusters from within this population without regard to districts boundaries. If you want district-specific estimates, you will have to apply a calculated sample size to each district.
However, in order to do appropriate sampling, you should consult some guidelines, or better yet, get the assistance of someone with experience in surveys and sampling. There are many more potential pitfalls in sampling and survey implementation than can be addressed in this short post.
Answered:
8 years agoIt is important to start with your objective of this survey. I am not certain if you are interested in results for each district, the 12 districts as one large Region or Both.
First, it is important to determine the smallest geographical units (whose population estimates are either available or can be obtained) in existence to use as your clusters. Given your scenario, 2 stage cluster sampling would apply based on probability proportion to size.
If the only available population estimates are at district level and you need to obtain estimates representative at Regional level (12 Districts), then you will consider each district as a cluster. Based on the number of clusters determined by your sample size, you will select clusters using PPS and more than 1 cluster are likely to be assigned to each district given the population sizes. Apply Segmentation based on PPS for each district followed by the second stage sampling method identified (Simple Random Sampling, Systematic Random Sampling) in each of the selected segments/clusters.
Further information on how to go about this is available at SMART METHODOLOGY
Answered:
8 years agoThank you for your valuable comments to my question.
Answered:
8 years agoHi everyone,
Many thanks for your valuable comments.
Designing a sampling solely depends on the objectives and other prior information. But first you have to consider few things:
1. Are these 12 districts homogenous?
2. Do you have any prior information on child malnutrition?
3. What is the ultimate objective? You want to interpret result for each district or want to have a global scenario?
For any of the above cases it’s better to follow multilevel cluster sampling. If you have good secondary information you can go Stratified sampling. At the bottom you can do SRS (simple random sampling) or Systematic random sampling depending on the availability of sampling frame (listing of population). You can find the steps well designed in ENA software. Suggestion is to do with a larger sample also it is important to consider time and budget constraints.
Answered:
8 years ago