Fudging cluster sampling in a survey research class

Home Forums Methodspace discussion Fudging cluster sampling in a survey research class

Viewing 2 posts - 1 through 2 (of 2 total)
  • Author
  • #1235
    Simon Kiss

    I’m teaching a third year survey research class. This year I totally revamped the course to make it hands-on versus lecture based. The assignment is to develop a research question, ground a hypothesis In theory, identify the concepts in the hypothesis and theory, develop measures to measure the concepts, put them in a survey, develop a sampling strategy, execute the survey, gather the data and present findings at the end of the term. It’s ambitious, but so far the students have been pretty engaged. Much more so than last year. here is my dilemma. most of the survey assignments they are developing involve surveying their students. There is a way to get a simple random sample by distributing a link to an online survey in an email that goes out to every student.

    But I also pointed out to them the possibility of doing a clustered random sample by randomly selecting classes from the publicly-available list of classes and visiting each class. two questions : should I weight the probability of being selected into the sample by class size (i.e. Engage in Probability sampling proportionate to size). I’m inclined to think not as most students will be in more than one class. it’s not like like I’m cluster sampling US states where people only reside in one state and the options are so imbalanced. Two, I understand that the statistical measures used in a clustered random samples (e.g.means and variances ) are different than for Simple random samples. teaching the students how to use software to calculate those statistics is really going to be beyond these students. should I continue? is it sufficient in this context if I tell them that in such a survey design, they would need to use specific types of statistical analyses? am I doing more harm than good?

    Love to hear your thoughts.


    I think it is a problem that students belong to multiple classes. One option would be to use a completely different example where you do not have this problem. Alternatively, I would say that it is okay to tell the students that, in a strict view, they should take the problems into account. But for your purpose, this can be ignored. At least, they have learned what the problem at hand is and that they would need to take care of this if it was “real” empirical research.
    Basically, I think this happens all the time; e.g., one uses survey data in class and ignores non-responses without checking whether listwise deletion is appropriate.

Viewing 2 posts - 1 through 2 (of 2 total)
  • You must be logged in to reply to this topic.