Interesting. For my data mining class I took NYC Open data - merging temporal Crime Data and Capital Expenditure data by geographic district/precinct, to see if there was evidence that spending money in a region had a positive impact on crime reduction. Then to see if I could use that model to see if I could predict how much crime would be reduced given a specified funding injection into the region. We assumed there was some lag (e.g. 6 mos. or 1 year) between funded project completion and community impact (reduced crime).
The model was only a little over 70% accurate, but the data was sloppy because we had to get creative mapping precincts to districts since their boundaries don't overlap exactly. However, I think this could be significantly improved since crime data is geo-tagged, so you can get much more specific about where crimes occur with respect to funded projects.
I think that model can be made to help inform where funding could be most beneficial from a social perspective. I realize there are other factors in infrastructure investments than crime - I was focussed on community improvement projects (e.g. blight removal, green space development, school construction/repair, etc.)