i would like to use cluster sampling in my study on the effects of fiscal decentralization on household income inequalities in Kenya
Munyua -
It seems to me that it depends upon what your sample units of interest are. You want your primary units (clusters) to be made up of secondary units (the level at which the data are collected) such that those secondary units provide the information needed, individually. I don't know your survey. If you need different information from each household member, then it seems to me that the secondary unit is not individual household members. If you need a whole household to provide information in each case, then that is your secondary unit, and the primary unit level is some "cluster" of households, such as those living in a city block, or an apartment building, or say 10 square kilometers in a rural area. The secondary units are those households which live there. The only way that a household would be a primary unit is if the level of the data to be collected is the same for each household member, and they are not being compared. That is, you are not looking for discrepancies between them; they all are identical in the sense of being members of the same pool of possible respondents. But, not knowing your subject matter, I can only say that I get the impression that you are in need of information from each member of a household, at the lowest, i.e., secondary unit, level. If that is the case, then your household is your secondary unit, and your primary units are apartment buildings, or city blocks, or some other geographic or other clusterings of these households. You could pick a randomized sample of primary units, say a simple random sample or an unequal probability sample, or even stratified random sample of primary units (though that could get complicated), and then select a randomized sample of secondary units (households), or even a census, within each primary unit selected.
You might randomly select households from your population, but if you need data from each member of the household selected, that is just randomly selecting households. I would not call that cluster sampling.
But if I am wrong, and each member of a household is an independent unit such that you do not have to obtain data from each member if you chose not to do so, then maybe a household could be a cluster (primary unit).
I am thinking it might be best for you to use simple random sampling of primary units, and then within those primary units either take a simple random sample or a census of secondary units, and I suggest you get a textbook such as Cochran(1977), 3rd ed, Sampling Techniques, Wiley, or Lohr(2010), 2nd ed, Sampling: Design and Analysis, Brooks/Cole, or Thompson(2012), 3rd ed, Sampling, Wiley, to help you.
I am not certain if you mean inequality between different households, or between members of the same household, or both, and what data/questions you are collecting/asking, but whatever you are doing, you need to decide the lowest level at which you are collecting individual data, the secondary unit level, and how you are going to cluster them into primary units.
Best wishes - Jim
Google advantages of different sampling methods for some useful information. Strongly recommend @James Knaub reply
You can clusters of more than 10 members
Cluster of 5-10
cluster of 2-5
Cluster of 2
Cluster of 1
- Badges
- Science topic
- Similar topics
- Correction