Dear Asadullah,
Hopefully this article will beneficial for you.
https://pdfs.semanticscholar.org/79a2/c4a4111275b3efbfa0522284ccd0fecc556a.pdf
Thank you.
Thank you all.... could you give me a short example with justification
such as your study is conducting on IT implemention in an organisation and you want to know about Are there accept among employees for implementing IT or not So, you have to given out a survey to IT staff only not all staffs who are working on finance , administration...etc .
IT STAFF IS CALLED PURPOSIVE SAMPLING METHOD.
Good luck
There are two distinct ways to do inference from a sample to a population: (1) randomized/design-based methods and (2) regression/model-based methods. You can also combine them. See the following:
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
For strictly model-based inference, we can use any sampling deemed appropriate, as long as we have good regressor data and all parts of a population are represented by a model, likely in subpopulation/strata/groups - so then, a different model for each group. Some use "balanced sampling," where the mean of an auxiliary/regressor/independent variable known for the population, or subpopulation being covered by a particular model, is the same for the sample as for the population or subpopulation. For very skewed establishment survey data (i.e., a few large establishments, more medium sized, and relatively many more small establishments) a cutoff or quasi-cutoff sample (accommodating multiple dependent variables/survey questions) can have much lower total survey error than other alternatives. Examples are given here:
https://www.researchgate.net/publication/319914742_Quasi-Cutoff_Sampling_and_the_Classical_Ratio_Estimator_-_Application_to_Establishment_Surveys_for_Official_Statistics_at_the_US_Energy_Information_Administration_-_Historical_Development
But purposive sampling without regressor data on the entire population is not very rigorous. There is no way to assess the accuracy without more information. In particular, bias may be large and unknown.
I noticed some points in
https://www.researchgate.net/deref/https%3A%2F%2Fpdfs.semanticscholar.org%2F79a2%2Fc4a4111275b3efbfa0522284ccd0fecc556a.pdf, see Muhammad above, which are good, but not all. For example, in the conclusions they note that purposive sampling cannot be used in a quantitative study, which is incorrect when you have regressor data for prediction, and thus use model-based inference. In the case of a cutoff sample, you could call this imputation. Estimates are obtainable for the variance of the prediction error for each total. Also, contrary to what is stated at the end of page 1 about observing the entire population, a census is not always preferred, even when only considering accuracy, and not cost, because nonsampling error can be larger for a census than sampling and nonsampling error combined for some samples. I'm just saying that some points can be argued. I had not really considered a convenience sample as not being purposive, but I get their point. Interesting.
PS -
Oops. One more important point: Whether purposive or randomized sampling is used, substantial improvement is often possible when you stratify your population. This requires knowledge of your population, even in stratified random sampling. You need to know the best ways to group your data into more homogeneous groups from which to sample. This could also reduce inaccuracy from a purposive sample.
- Badges
- Science topic
- Similar topics
- Engineering
- Chemical Engineering