# Sampling Procedure

Home Forums Methodspace discussion Sampling Procedure

Viewing 6 posts - 1 through 6 (of 6 total)
• Author
Posts
• #4445
Asif Ranjha
Participant

I am PhD student at Dundee University working on research methodology and sampling for my research. I need some guidance on sampling procedure and sample size.
Total population size is 6369 which lies in 116 geographical areas (Tehsils) and these 116 Tehsils lie in 36 geographical areas (Districts).
What is suitable sampling methods to take sample?
What should be the sample size out of 6369?

Waiting

Asif

#4450

Hi,

Try cluster sampling: http://en.wikipedia.org/wiki/Cluster_sampling. They also describe two-stage cluster sampling, which can be extended to three-stage sampling.

If I was you, I would sample all 36 districts and 116 tehsils, and then, within tehsils randomly sample around 10-20 individuals per tehsil. The reason for this is that statistical efficiency (precision) raises, when you have many higher-level units, almost regardless of the how many individuals you have per tehsil. This is because you have to calculate an “effective sample size”, which is a weighted average of the number of cases at the different levels (multiplied with a kind of correlation coefficient). This book explains it in easily accessible terms: http://www.itsjustregression.com/.

Another option is to sample the full population. It depends on your resources and the expected amount of nonresponse.

All the best,
Kristian

#4449
Member

i agree that cluster sampling might be a good option and i advice that you seek a specialized statistical advice to calculate your sample size based on the information you had about the population and based on the questions that you are asking and the statistical test that you are planning to use during data analysis. good luck

#4448
Asif Ranjha
Participant

Dear Kristian Karlson,

Thank you very much for your response on my question.
What is your opinion about proportionate random sampling (10% respondents from all 116 tehsils)?

Another question is how much sample size and how to take sample when number and adresses of respondent population(NGOs) is unknown?
Should it be through snow ball sampling? and what should be sample size if total population is unknown?

Kindly guide.

Many Regards

#4447
Asif Ranjha
Participant

Thank you very much for your response on my question.
What is your opinion about proportionate random sampling (10% respondents from all 116 tehsils)?

Another question is how much sample size and how to take sample when number and adresses of respondent population(NGOs) is unknown?
Should it be through snow ball sampling? and what should be sample size if total population is unknown?

Kindly guide.

Best Regards

#4446

Depending on the average number of individuals in each tehsil, 10 percent per tehsil seems to be to few. You have, on average, 6369/116 = 55 individuals per tehsil. 10 percent of 55 is 5,5 individual on average. Since you probably will encounter non-response, you may run into the problem of having less than 5 per tehsil for quite a few tehsils, which should be avoided.

Two options exist:
A. Chose 20 percent.
B. Do not use proportional sampling, but sample a fixed number, for example 10 or 12, in each tehsil.

Regarding your second question: Drawing samples when the population is unknown is a big problem that no method can solve effectively. The reason for this is that, logically, you do not know to which population you want to infer your results in the sample. However, snowball sampling is a possibility, but I have no experience with the reliability and validity of that procedure. You could argue that, since you know you are inferring to NGO’s, you have an informed idea about which population, you are interested in (NGO’s). The next problem is that you have no clue as to whether the snowball sample represents that population. Who is included, who is excluded? The question is, then, whether you have genuine interest in generalizing to the population of NGO’s or rather will focus on the sample you collect.

/Kristian

Viewing 6 posts - 1 through 6 (of 6 total)
• You must be logged in to reply to this topic.