2

I am working on a problem that requires me to measure groupings of people. I have the location of every individual in my sample at every point in time. It's therefore trivial to calculate the distances between each individual for a certain point in time.

Is there a way I can think about groupings using these pairs? Currently I am defining a group as two individuals $\le$ a certain distance away from one another. This is the way I'd like to define my groupings. Basically, I'm trying to find a way to measure "groupishness" of all these characters over a set time span. I know there are n choose 2 pairings, and therefore I could calculate for every possible pair the average distance between them, but I'm trying to get a measure for the entire group. Any ideas would be helpful. Thanks.

user48944
  • 121
  • This is called "clustering" and there are many clustering algorithms available. In R you should check out the package cluster. If you provide a minimal working example we will be able to show you code for how to cluster based off a distance matrix. – bdeonovic Jun 24 '14 at 21:12
  • Thank you for the link. I have a sense of how this is done with a single timepoint data set (ie form dendograms based on euclidean distance), but do you know how one could think about this over time? – user48944 Jun 24 '14 at 21:55
  • you get get everyone's group assignment at each time point and in the end you could have the relative time a person was in each group. Your last paragraph is still a bit unclear; not quite sure what you mean by groupishness. Could you elaborate? – bdeonovic Jun 24 '14 at 22:34
  • I think the relative time is what I meant as groupishness. Basically I want a way to measure how much time people spent in groups, but I suppose that depends on how I define my groups. Using my crude method (ie. distance to <= n) it looks like I can define my groups using my creteria and do that for each point in time like you said. Then I can evaluate either percentage of time each person was in a group over the course of our experiment or percentage of the experiment time a group existed. Do you think that will be a good approach? – user48944 Jun 24 '14 at 22:41
  • To be more clear, our experiment consists of periods, each lasting a certain number of seconds. I want to look at groupings both over the entire experiment and per period. Thanks for the help thus far. – user48944 Jun 24 '14 at 22:45

0 Answers0