For starters, neither groups is representative of the homeless demographic.

And when you take a group of people who have the best chance of success through your selection and sample biases, you are engineering a positive result.

Well designed studies don’t work that way.