I am designing a survey research for environmental practices in SMEs. However, finding the definitive number of SMEs with that particular criterion is not feasible. Hence, I may have to select non-probability sampling. I could not find definitive answer or source for this. I appreciate if anyone could help me on this, please?
The basic principles to generalize the results are to have random sample, heterogeneous, representative, and large.
If you go for convenience sampling, you can generalize if it is large and you can compare the participants characteristics with the target population.
Calculating the sample size depends on the analyses that you will follow: if you are doing inferential statistics, then you need to do power sample calculation. Otherwise, there are free online programs to calculate the survey sample size.
Abdul -
If you are looking at continuous data, y, where you have regressor data, x, for a given data element (question), then even if you do not know all the members of the population, if you do know (x,y) for a sample, and you know the total of x, say X, for the population, then you can estimate total Y if you use a classical ratio estimator.
An example is
https://www.eia.gov/petroleum/production/pdf/eia914methodology.pdf
where you can see that you do not need to know the individual x_i for the missing i=n+1,N cases, just a total.
This is feasible if x is information for which you know the total (say, industrial output of some kind, for example), you know the x_i for the sample, i=1,n, and y and x are approximately linearly related, and we expect y will be zero if x is zero. These conditions are very often met in business surveys.
Stratification can be very important!
Not knowing your application, I'll just say that the following paper by Ken Brewer might give you some insight (note that the model-based example he gives uses the classical ratio estimator), and if you must use something less rigorous, you may find an idea in the paper below that by Mike Brick.
I hope there might be something here which can be of use to you.
Cheers - Jim
Ken Brewer's Waksberg Award article:
Brewer, K.R.W. (2014), “Three controversies in the history of survey sampling,” Survey Methodology,
(December 2013/January 2014), Vol 39, No 2, pp. 249-262. Statistics Canada, Catalogue No. 12-001-X.
http://www.statcan.gc.ca/pub/12-001-x/2013002/article/11883-eng.htm
"Beyond traditional survey taking: adapting to a changing world
Explorations in Non-Probability Sampling Using the Web," J. Michael Brick
Proceedings of Statistics Canada Symposium 2014
http://www.statcan.gc.ca/sites/default/files/media/14252-eng.pdf
The basic principles to generalize the results are to have random sample, heterogeneous, representative, and large.
If you go for convenience sampling, you can generalize if it is large and you can compare the participants characteristics with the target population.
Calculating the sample size depends on the analyses that you will follow: if you are doing inferential statistics, then you need to do power sample calculation. Otherwise, there are free online programs to calculate the survey sample size.
Hi
One important observation is, "What determines a sample is not the size of the population, but the homogeneity of the population. The more homogeneous the population is, the smaller the sample size can be, while the more heterogeneous a population is, the larger is the size of a sample".
Dear Abdul
If your population can be stratified to represent different categories of respondents, you may distribute your convenient sample accordingly to add more validity.
Hello Abdul Ali,
As many of my colleagues stated previously, generally, probability - based samples (random, systematic, stratified) are preferred over convenience samples. In fact, many of the assumptions of parametric statistics requires a certain number of participants selected by probabilty. However, having big numbers and trying to add diversity in your sample might help "fixing" your design.
Often, the use of non parametric statistics is required for non - probability selected samples.
Dear Abdul
Logically you can not generalise result based on purposive samples. However you can go for systematic sampling, if it is feasible to your case.
Moreover, you can generate no of samples with replacement after getting a sample (known as bootstrap). You may seek the possibility of this also.
Regards
Rajiv
- Badges
- Science topic
- Similar topics
- Methodology
- Research Methodology