Inviting and gathering pre-selected participants to one place for interview is a common practice in sample surveys. Selecting the participants is, however, a different matter. In your case, I do believe that there is no way to get a statistically sufficient and randomly selected number of participants from your population.

Your situation is not uncommon that is why non-probability sampling methods have been designed, because situations like yours do happen.

Involving only samples that are available is equivalent to convenience sampling.

On the other hand, you will not end up in this situation if you "go an extra mile", that is you exert additional efforts (and resources) getting the full list of residents and reaching out to your random samples.

Thank you, Eddie. Sorry for not replying you soon since the question was raised long time ago without any reply, so I haven't checked this frequently.

You are right. Actually we had to compromise that survey with convenience sampling last year. We had to increase the sample size to achieve higher representative level.

Nam Ha -

For future consideration, you might try "area frame" designs for survey sampling. The USDA uses that a lot, so there is information available if you search the Internet. I think it sounds like it might be useful for you. You have to divide your geographic area by regions for sampling.

As for purposive sampling, I got a great deal done with that when there are regressor data available, and when you can stratify well, but other than that, purposive sampling can give you biased results without a good way to evaluate accuracy.

You might be interested in reading what was said in the a attached link to a question on "representative" sampling.

Best wishes. - Jim

https://www.researchgate.net/post/What_is_a_representative_sampling

Thank you Jim! Your post really helps me to learn new knowledge. It is actually difficult to conduct a "pure" random sampling in my country, particularly in rural areas because of the lack of population statistic information.  

What do you think when comparing the importance of 'representativeness' among quantitative vs qualitative vs mixed-methods approaches?

For randomized sampling, perhaps you might try a "multistage sample," perhaps two stages where you might pick randomly from basically equal size geographic regions, or by probability proportional to size (PPS) sampling, perhaps stratified by rural vs city, and then in the final stage you can determine the human population in just those randomly selected geographic regions, and either census or random sample from those humans in each randomly selected region. 

Actually, if you randomly select from among geographic regions, and then census humans in the selected regions, I guess that is basically just a one stage cluster sample. 

You might look up "cluster sampling."

Regarding qualitative data, and mixed methods, I do not even have a guess there, except to say that, as in quantitative sampling, I think you are right to consider representativeness.  Stratifying your population by important categories and looking at each (say, city, and perhaps rural can be further broken down).  So maybe stratified cluster sampling will do for all your needs. 

I'm just guessing though.  I do not recall participating in a project very much like yours. 

Best wishes.  

Hi Jim, you do shed more light on my question. In my current study, I'm also applying a similar sampling strategy like what you mentioned, starting with stratification and cluster then randomness. However, for this study I hold the view of mixed-methods so I just wonder how different each research paradigm may value the importance of representativeness. I have read some papers on methodologies and still being unsure so I need to find more "right" materials.

Thanks again for sharing your time and wisdom!