Hello Bira,

Yes. Cluster sampling (with same overall N Ccases) is less precise than simple random sampling. Hence, there is greater error associated with the population estimate, and CIs will be wider. The trade-off is that cluster sampling is usually easier to implement than SRS.

Some good resources:

Cochran, W. G. (1977). Sampling techniques (3rd ed.). Wiley.

Henry, G. T. (2016). Practical sampling. Applied Social Research Methods Series, Vol. 21. SAGE Publications.

Youtube links:

https://www.youtube.com/watch?v=be9e-Q-jC-0

https://www.youtube.com/watch?v=Rf-fIpB4D50

Good luck with your work!

Hello Bira,

Yes. Cluster sampling (with same overall N Ccases) is less precise than simple random sampling. Hence, there is greater error associated with the population estimate, and CIs will be wider. The trade-off is that cluster sampling is usually easier to implement than SRS.

Some good resources:

Cochran, W. G. (1977). Sampling techniques (3rd ed.). Wiley.

Henry, G. T. (2016). Practical sampling. Applied Social Research Methods Series, Vol. 21. SAGE Publications.

Youtube links:

https://www.youtube.com/watch?v=be9e-Q-jC-0

https://www.youtube.com/watch?v=Rf-fIpB4D50

Good luck with your work!

No, confidence interval is an artificial criteria set during the analysis of data

Confidence interval depends upon the the of sampling used in drawing of sample , the size of sample drawn and the pattern of probability distribution of the population from which the sample is drawn.

Therefore, it is obvious that the confidence interval of a sample drawn by simple random sampling changes if the sample design is changed to the cluster sampling (though with the same mean and the sample size).

NO the confidence interval has nothing to do with the method of sampling. Sampling method is the means by which the sample is taken from the population, i.e. how to pick sample.

Confidence interval is the level of significance fixed by the researcher---he/she may fix it arbitrarily at 0.90, 0.95 or 0.99 depending on the demands of the circumstances, field of studies, etc. In engineering or genetics, for instance, confidence interval may be specified at 0.9999999 as required for DNA testing.

These two concepts are not related: (i) sampling method; and (ii) confidence interval. Sampling method = how to select sample. Confidence interval = at what % error level to define pValue as significance, i.e. Confidence interval = 1 - CDF.

@ Paul Louangrath: You forget that to construct the confidence interval, one needs 1) a distributionnal assumption (to have quantiles) and 2) an estimate of the estimation variability (to have an estimation of the standard devation of the point estimate). Both of them DO depend on the experimendal design/sampling protocol. So the concepts are indeed different, but the practical obtention of a confidence interval needs a correct experimental design.

As an extreme case to illustrate this, imagine you want to estimate the sex ratio of a given population. No, imagine your sample design is 1) select randomly a single patient and 2) just repeat the observation on this patient *brothers*. Then, by construction, the estimate is either 1 or 1 - 1/n, n being the total sample size. Obviously, you will not be able to compute a confidence interval, because the only valuable observation is the first one (or you can give the [0;1] interval or the [0.025 ; 0.95] interval, but not best, even if the patient has 6 brothers). No do the same experiment with the sample design 1) select randomly n patients and 2) do the observations. You can apply the usual formulas for probability confidence intervals.

That's extreme, I agree, but that prooves that the confidence interval has a result does depend on the sampling method.