To understand bankers, you have to first think like one. That is what the Asbury Park Press tried to do when it set out to study banking behavior in New Jersey. Bankers are fond of telling us that they build branches where there is the most business. Race and other basic demographics don't matter at all, they say, as long as there is money to be made. That isn't what we found. In order to compare apples with apples, I overlaid bank location data with commercial business data borrowed from our marketing department. We used commercial business data because it is a closer proxy for banking business than population or income data. That data came in a geographic layer that could be imported into mapping software. I used Atlas GIS for Windows in combination with ArcInfo, a more powerful mapping program, for some of the analysis.

We then overlaid the business data with bank branch locations for all commercial banks. We had to reconcile Federal Reserve data with our own survey because neither alone gave us a complete picture.

What we found was startling. When we divided the state into census tracts and analyzed branch-to-business ratios, minority communities had significantly fewer bank branches relative to the number of businesses than white communities. Race turned out to be a significant predictor of banking service — even after controlling for income, commercial activity and population density.

This was accomplished by creating a regression model with the help of a Rutgers University professor who teaches statistical analysis. The model showed that race was independently associated with branch location, meaning it wasn't just a proxy for income or the density of commercial activity.

We also took a more traditional approach to redlining, overlaying branch data with income maps and looking at lending data from the Home Mortgage Disclosure Act. Again, the picture that emerged was one of underservice in minority communities.

We built all of our data into a geographic information system so readers could see these disparities on maps published with the stories. The maps made the numbers vivid in a way that tables and charts could not.

The whole package, titled, "No checks or balances," ran Sept. 10-13. It can also be found on Nexis.