Using the snowball sampling, first I make a comprehensive list of the respondents and then from the list, I choose the whole population (around 90).
Can we use data generated from total population sampling (also known as non-probability sampling) for Binary logistic regression?
Thank you.
Yes you can because the probability of each member is 1/N where N is the size of the population assuming a simple random sampling from the population. See Cochran, Sampling Techniques, 3rd edition, for the equations should you need them. Best wishes, David Booth
Yes, but even with ‘all’ the
’population’ you have to think about who these results are representative of.
Kelvin Jones use of the word representative is the everyday use. If you use the Cochran reference I have above Cochran uses the term representative to mean a STRATIFIED sample not a simple random sample as many sampling statistcians do. DON'T be confused by the difference between the two uses. However if as the question says the observations really make up the entire population of interest then Professor Jones'suggestion is redundant. PLEASE READ CAREFULLY. Best wishes to All, D. Booth
Sorry , I should have been clearer ; I was trying to warn about making claims about the results given the well known problems of snowball sampling - that is not collecting data on respondents who are not 'connected to the network'. These might be quite different in their characteristics.
We all know that the "population" must have properties that define the scope of our study. If we do not know them, we describe some relationships between variables in a research object that we could not identify and would be valid only for an unrecognizable subset of a community. The results thus derived may be of little interest. Snowball samples easily have this problem. If Khondokar Humayun Kabir can clearly describe the population coverage of his collected data, beyond saying that it is the data he collected, he will be able to defend his analyzes, including logistic regression.
Data generated from total population can be used for Binary logistic regression. But if you are planning to use whole population as participant for the study, it is no more called as sampling, rather called as census method. In this case the probability of each member is 1/N where N is the size of the population, hence it is not non-probability sampling. Furthermore it is unlikely you can have the list of total population by using the snowball sampling method, because of persons ‘not connected to the network’.
- Badges
- Science topic
- Similar topics
- Social Psychology
- Attitude