Cluster Analysis

We applied a well known clustering algorithm called K-Means to explore the project data at both the national and regional levels. The premise was to find interesting groupings of projects that shared common features. We used converted all the categorical variables into indicator variables so the Euclidean distance measure used by K-Means would be meaningful. In other words, the K-Means algorithm essentially counted the number of differences between each project to determine which were similar to each other. It should be noted that the cluster analysis was done on the April 2011 data.

The resulting clusters highlighted some regional differences on the national level and some patterns of unusual giving within each region. For instance, there was a dichotomy between some projects in the Western United State and those in the South. Consider the following clusters:

West vs South

These two clusters are largely the same except they are from different regions. Both are mostly requesting technology, however the success rate for the South is much lower. Furthermore, consider these two clusters from the Northeast and South:

East vs South

Again, the South appears to be not as successful at attaining funding. The cluster from the Northeast has a higher success rate when requesting non-essential items over twice as much in value as the South. However, a cluster from the South requesting cheap technology is one of the two most successful projects:

Most Successful clusters

National Results

Western States

Southwest States

South

Northeast

Midwest