The organization I belong to studies indigenous populations through primarily survey research. One study for example aims to identify the relationship between culture and the evaluation of odors. We are looking for a statistical procedure that can be used to compare a larger amount of individuals. Some records were measured directly on individuals, while others were measured by a group spokesman. In the case of groups, however, individuals should be "downscaled" in order to be able to compare these data records with the data sets of individuals measured. The group speakers of a group are not always the same people from a particular group, but different people can respond on behalf of the group. The group size can vary within the measurement. A mean number of group size, ratio between man and woman and age is estimated per questionnaire/measurement (data set). Due to the difficulties collecting data in indigenous contexts, this is the only way we found to be possible. We are looking for a statistical method that can compare individual survey responses, if some were measured individually, and others were measured as a group (from which we need to extract individual responses).

  • 2
    $\begingroup$ This is on-topic here, but could also find a home at Cross Validated in case it does not fair well here. In the meantime, could you perhaps structure the question a bit more, i.e., break up the text in paragraphs? $\endgroup$
    – Steven Jeuris
    Jun 5, 2018 at 13:08

1 Answer 1


Sounds like you're looking for a hierarchical model. You might find some joy in BDA, http://www.stat.columbia.edu/~gelman/book/ (in particular, chapter 5)

To quickly get some idea if this sort of thing is right for you, you could check out this blog-post on one particular example here http://andrewgelman.com/2014/01/21/everything-need-know-bayesian-statistics-learned-eight-schools/

Whether this will work for you also depends quite a bit on what you want to ask (what does 'identify the relationship' mean?) and the target audience. If you do head over to Cross Validated (strongly recommended!) make those things as clear as possible in your question.

  • $\begingroup$ Thank you so much. Do you have any suggestions as to apply a hierarchical model in this case? Specifically, we have a survey questionnaire that we give to individuals and they rate their evaluation of a specific odor: the degree to which they perceive it as familiar, intense, likeable, etc. on a likert scale from 1 to 5. Fill in the blank questions are also a part of it. With this we are trying to analyze how odors/odor perception may differ cross-culturally. $\endgroup$ Jun 7, 2018 at 11:44
  • $\begingroup$ Um, that's a bit broad. A hierarchical model specifies a relationship between individuals, groups, and the whole population, with individuals drawn from a group distribution and groups drawn from a population distribution. The advantage of doing that is that it lets you say something about the people represented by a spokesperson, and lets the whole population inform what's plausible for groups where you have very sparse data. That's as specific as I can get in an SO comment, you need a stats person onboard. But I hope this helps you "find the question" that you want to ask them. $\endgroup$ Jun 7, 2018 at 16:02

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.