association of binary data with multidimensional scaling?

#1
Hi everyone,
I have 10 binary variables, each coding the presence or absence of an event for each of 2000 participants in the study. I would like to find out, how these events co-occur, so that I may be able to group them together. I have been looking into different ways to do this and was wondering, if I could apply multidimensional scaling. For that I would then aggregate the co-occurrence data over persons. Would this be a good approach or is there another method I should be considering? For you this is probably a pretty basic question, but I don't deal with binary data very often and would really appreciate a pointer in the right direction.
Thanks for the advice
 

bugman

Super Moderator
#2
The quick answer is yes you can apply MDS to binary data. Though without further details of your varaibles I can't say much else.
 
#3
Thanks for the reply. Let me try to give a little more information. The variables are life events that were either experienced (1) or not experienced (0) by the participant. It stands to reason that some of the variables, such as 'moving' and 'separation' are more likely to co-occur. I'm not interested in testing any hypothesis, but to explore and describe the relationship between these events before I decide on how to agglomerate them for analysis. The actual hypothesis is the negative effect of these life events on mental health. As a first step I looked at the frequency of events. Then I was trying to get a sense of which events co-occur with each other with the distances similarity measure in SPSS. Since I'm not only interested in similarities (co-occurrence in my case) among pairs of variables, but more to see how they group together, I was thinking about MDS.
Some of them, like 'death in the family' are very rare. 104 out of 1884 participants experienced this event. Would these rare events be a problem?
I think, I will just try it and see what happens.
I would still appreciate suggestions for other methods, if there is something more appropriate. I work with SPSS and R, if that is of interest.
Thank you for your input.
 

bugman

Super Moderator
#4
I think your approach is right, MDS may not be as useful to you as cluster analysis though; and be sure to select a distance measure that includes joint absences, such as Euclidean or simple matching.