You basically have collected ordinal data, but if you have several such items that each measure the same thing, then you can use coefficient alpha to assess the possibility of summing those items into an interval-level scale. I would not recommend using a cut-point to convert the data to a dichotomy because that "throws away information." In the words, the variance in your scores is meaningful, and you should examine it rather than collapsing it away.

First of all: WHY would you do that? It's a waste of information. You can do regression on your original data, not if you reduce them to just 2 classes.