What is the most statistically reliable sampling method when you are dealing with a highly fragmented group of individuals coming from a relatively unknown total sample size? For context, I am preparing to conduct a study on the entrepreneurial orientation, success factors and risk aversion of foreign SME entrepreneurs doing business in Korea. There is no centralized list of business people and the total number of entrepreneurs is unknown to even government immigration and taxation offices. Study participants will be gathered through the various chambers of commerce, foreign business associations, embassies, former associates and referrals from the mentioned groups. I was hoping to apply a little more academic rigor to the study than simply using the snowball method.
Recommendations?
In this circumstance, I don't see anything wrong with the 'snowball' method, provided you (a) keep a complete record of the progress of sampling, and (b) start with a large initial sample which is gathered from a variety of databases (pools). Snowball sampling is unreliable if you start with a small sample (such as one), or with a sample drawn from a single pool within a more diverse set. It is often impossible to obtain a random sample, and so you are trying to obtain a sample that is representative of your entire set and large enough to be statistically meaningful. Of course, when you report your results you also report your methods and the limitations of your methods, one of which is that the sample, while selected as much as possible to be representative, may not be perfectly so.
Of course, if your snowball method creates a large pool, then you can sample within the pool randomly.
Hi Brandon
The problem that could arise if you use a snowball sampling methods is that you could inadvertently miss a self-contained part of the potential sample, because your snowball only roll around in a different part of the sample. For instance the entrepreneurs who speak English only associate with other English speakers – because we are so bad at learning other languages! So if you start with one of those, you might miss out on other non-English speakers.
But it sounds as if your sampling frame has probably been set up to enable you to sample across the different constituent groupings. In effect, what you need to do is a kind of ‘theoretical’ sampling, making sure that your sampling frame is constructed acooring to a good theory of who the entrepreneurs are, and uses methods to get to all parts of the foreign entrepreneurial community.
So the answer is to use all the information you can get, plus key informants who know about the area, to make sure you have a very good idea of the entire population of entrepreneurs, and then make sure you are sampling from all parts of that population, and that you get a reasonable sample size from each part, rather than lots form one part and only one or two from another. So long as you are covering all parts, you can use a snowball method in each part to recruit.
Does this help?
Nick
Hi Brandon
One approach I have used for more selective/purposive sampling has been to use a questionnaire which provides a database of people to select from for interviews. In past research I have used this to identify people with particularly high or low scores on an attitude scale:
https://www.researchgate.net/publication/230664474_Man_Points_Masculine_Capital_and_Young_Men%27s_Health?ev=prf_pub
or to select people reporting particular behaviours: https://www.researchgate.net/publication/258101405_Identifying_and_understanding_barriers_to_sexually_transmissible_infection_testing_among_young_people?ev=prf_pub
This does not necessarily address the concern Nick raised, but it does allow you to be a little more selective
I hope that helps
R de V
Article "Man Points": Masculine Capital and Young Men's Health
Article Identifying and understanding barriers to sexually transmiss...
Brandon, from your story, I believe that your "most statistically reliable...." expression would be limited to descriptive, non-inferential statistics. ( I wish you would not find this a negative opinion). From your target population, it would be very difficult to apply any of the probability sampling techniques (simple random, stratified random, systematic random, cluster random or even multi-stage random sampling). Similarly, snowball sampling is a non-probability sampling design. Your alternatives to snowball sampling would be convenient, judgmental, and quota sampling. But non-probability sampling methods will not allow you to draw an inference from the sample to the population. Ed
- Badges
- Science topic