I have a dataset generated from a survey. Each respondents mentioned several friends and how they interacted with these friends. Below is a simplified example. enter image description here

I am interested in knowing how friendship level is affected by both the respondent's characteristics and the relational variables regarding how the respondent communicates with a friend.

A friend of mine suggested that I could stack the data by making the friends of a respondent's individual cases, while duplicating the respondent's characteristic variables.

My question is, would I be able to apply regression (OLS, logistic, etc.) to this stacked data? The data stacking example I found online mostly do not involve the respondent's characteristics. Would mixing up variables at two levels violate any assumption? If so, what would be a good statistical method to analyze the data to achieve what I intended?

Note that this is for an academic paper (social science), so it has to pass reviewer #2's scrutiny. It cannot be "good enough for practice".

Thank you!


I don't think you've provided enough detail for me to point you toward anything specific. The main question I still have is, "What is the goal of your analysis?" What are you trying to predict/measure with respect to these relationships? I don't even see a clear dependent variable in this small example.

My gut instinct is, no, this is not an appropriate way to analyze this data regardless of the response. You have clear dependence between the subjects in your study. In fact the type of dependence is pushing me toward telling you to investigate network models.

I'd be happy to help more if you provided a bit more detail.

  • $\begingroup$ The dependent variable, in the stacked data, is the "friendship level". Independent variables: demographic variables of the respondent, whether the respondent communicates with the friend via phone or email. $\endgroup$ – PickledXu Jan 7 at 18:42
  • $\begingroup$ Okay, I’m confused about one other thing. This is really important. Are the respondents the same people as the people identified as “friends of the respondents” in friendship1 and friendship2? Or, are they simply just multiple instances of friends of those respondents? $\endgroup$ – MentatOfDune Jan 8 at 6:52
  • $\begingroup$ Each respondents generated a series of friends. These respondents do not know each other. And the friends they generated are just their own friends thus have no relation to anyone else in the data. Upon reading a few materials, I believe I should use a generalized linear mixed model? $\endgroup$ – PickledXu Jan 11 at 22:04

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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