Very general cluster analysis question

[Collin James]

Hi,

I am interested in thinking about a cluster analysis (or some other analysis, open to suggestions) to get to an interesting question. I have a group of people who were either homeless or not (indicated by a 0 or 1 variable). For those who are homeless, I wanted to identify some typologies or clusters that might exist among observations. For instance, I have a bunch of data on previous substance abuse, mental health problems, domestic violence, income, etc, and I wanted to identify some typologies/pathways of homelessness. (i.e. maybe there is one type of person that ends up homeless because of income and another type that ends up homeless due to substance abuse and mental health). I thought a cluster analysis might be nice to see what factors are closest to one anotehr and closest to being homeless at the same time. I know this is a broad question but I am just startying to think about it. Any thoughts? Thanks--CJ

[Carlo Lazzaro]

Collin:
I'm not an expert, but -cluster- may sound reasonable here.
That said, I'd probably be more interested in a -logistic- regression aimed at regressing the probability of being homeless (or not) on a set of predictors, like the ones that you propose in your post.