Dear Asadullah,

It is a discussion that whether we can use non-random sampling techniques in quantitative studies or not. Some scholars rigidly reject this alternative. On the other hand, according to some scholars, using purposive sampling require non-parametric statistical analysis. When you obey this rule, there will be no problem remainedfor them.Good luck in your studies!

Best regards,

Mehmet

Dear Asadullah,

In addition these two groups in my previous answer, an other group scholars don't see any problem with use of purposive sampling in surveys (In contrast to the second group, they don't also demand non-parametric statistical analysis).

Some studies below illustrating use of purposive sampling with survey may be helpful:

http://www.ijbs.unimas.my/repository/pdf/Vol15-no3-paper4.pdf

http://dergipark.gov.tr/download/article-file/364886

Best regards,

Mehmet

Hello Asadullah,

We've all seen survey studies wherein there really was no sampling procedure except a solicitation to participate. Those are known as convenience samples, and are often the least defensible for claiming representativeness of the target population.

You are free to use whatever sampling process you wish. However, your ability to generalize confidently to the target population depends heavily upon your choice. Short of census sampling (getting every member of the population), probability sampling offers the greatest protection against wildly unrepresentative samples (and, as well, good efficiency for population parameter estimates within target tolerance and adhering to target confidence levels).

That said, purposive samples can be very helpful for trying to achieve a representation of the target population. To the degree that your method captures most or all of the salient membership characteristics that could account for differences in the survey responses, you have a better chance of obtaining a reasonable population representation.

As well, purposive sampling is often used in qualitative research (though from your description, that doesn't seem to be your direction of inquiry).

Good luck with your work.

As you introduce the issue of purpose sampling in quantitative survey research, you need to be aware of the quantitative - Qualitative debate. What is quantitative in your research? perception oriented data? or hard figures like sales, production or a combination.

Purposive sampling requiring defining the characteristics of the participants you want to include in your sample(s). In this sense, it is just like quota sampling -- which sometimes has a bad reputation from its early history, which people tried to use it as a short-cut to obtain representative samples.

Simply put, if the researcher has the full list of the target population (e.g., full list of employees in fast food industry of London or full list of SMEs working in a country etc); which in most cases is impossible; only then a probability sampling technique can be used, otherwise a non probability sampling technique is appropriate (such as Purposive Sampling) to get a representative sample.

Please refer to the reference given below:

Rowley, J. (2014). Designing and using research questionnaires. Management Research Review, 37(3), 308-330.

Kindly read below reference for the issue of Generalization.

Seddon, P. B., & Scheepers, R. (2012). Towards the improved treatment of generalization of knowledge claims in IS research: drawing general conclusions from samples. European Journal of Information Systems, 21(1), 6-21.

Dear Mr. Asadullah

Choosing a appropriate sampling technique depends on many criteria specifically knowing the elements of population, sampling frame etc. Certainly, Purposive sampling technique is one of the most adopted sampling technique in quantitative research, however you should be very careful while determining the criteria before selecting the sample element.

I agree with Mr. Muhammad Zia Aslam. Please read the above mentioned articles. These are very useful articles.

With Thanks

If you want to be able to do inference from a purposive sample, then the prediction approach may be available to you, if you have the necessary auxiliary (predictor) data, and your sample can be used for one or more models adequately covering your population. Probably the most accurate such situation is when you have Official Statistics where a census (for predictor data) occurs occasionally, say annually, and sampling occurs (perhaps monthly or weekly) which collects data on the same data elements. Then if you have sufficient data at hand, you can test your models, and adjust them to work well. You would still need to be aware of changing circumstances. As Waksberg Award winner (Survey Statistics) Ken Brewer told me, his mentor, Ken Foreman, said there is "no substitute" for being familiar with your data.

In particular, for Official Statistics, especially when you have repeated establishment surveys, many years of research showed me that the following model-based/prediction approach works very well under these circumstances:

"Application of Efficient Sampling with Prediction for Skewed Data," JSM 2022: 

https://www.researchgate.net/publication/362370770_Application_of_Efficient_Sampling_with_Prediction_for_Skewed_Data 

This has a long history of successful inference for large numbers of samples for the US Energy Information Administration (EIA), and should work well under similar circumstances for other subject matter applications, as noted in the paper.

Other than that, there has been work by some on "pseudo-random sampling," and other I think more complex modeling for cases I would consider more 'wobbly' (i.e., more likely to fail without a huge amount of work and other resources).

So to be clear, you cannot collect any sample you want and infer to a population from it without other information and/or planning. You can use a probability-of-selection-based (design-based) sampling and estimation approach, a model-assisted design-based approach, or perhaps a model-based approach if checked carefully, but you have to have some basis for inference.

I generally agree with statements above by David Morse, though I also include the model-assisted and model-based approaches. Also, please note that it was stated that other than a census, "...probability sampling offers the greatest protection against wildly unrepresentative samples...." In the model-based approach you can use 'balanced sampling.' Perhaps the simplest case would be to include in your sample, y-values such that their corresponding x-values have a mean about equal to the population mean of x. This basically guarantees a sample like the one you hope for but routinely fail to achieve with a simple random sample.

Some references:

Särndal, C.-E., Swensson, B., and Wretman, J.(1992), Model Assisted Survey Sampling, Springer-Verlang.

Brewer, K.R.W.(2002), Combined Survey Sampling Inference: Weighing Basu's Elephants, Arnold: London and Oxford University Press.

Survey Sampling: Theory and Methods, First Edition, 1992,

Chaudhuri, A., Stenger, H.,

Marcel Dekker, lnc., New York, Basel, Hong Kong.

Chambers, R, and Clark, R(2012), An Introduction to Model-Based Survey Sampling with Applications, Oxford Statistical Science Series.  

Valliant, R, Dorfman, A.H., and Royall, R.M.(2000), Finite Population Sampling and Inference: A Prediction Approach, Wiley Series in Probability and Statistics.

Olukunle Saheed Oludeyi noted that "...if you are studying how a village was attacked by lions, you need only the members of the same village who witness or experience the attack. You can purposively choose them and not all the villagers."  However, in such a case, I would consider that as defining the population, not a sampling technique.