How it works
In Applications Quest, holistic review and diversity are achieved by creating distinctive groupings or clusters of applications which share a high degree of similarity. Because each grouping/cluster represents a distinctive pattern of similarities across a set of attributes (relative to other clusters), clusters form the basis of diversity in Applications Quest.
Imagine if two applications could be compared in such a manner that their relative difference, or similarity, could be quantified on a 100 point percentage scale. Using this notion, two identical applications would be 0% different and 100% similar. All other application comparisons would fall between 0% and 100% on this relative scale.
How are applications compared?
The difference/similarity between any two applications is determined by measuring the difference between all relevant attributes, and then arriving at a cumulative difference score for the pair. To determine the cumulative difference, institutions may compare applicants across a broad range of attributes, in line with their admissions objectives. These may include both numeric and nominal attributes.
The difference matrix
If the difference between every application pair could be determined, we could arrive at a difference matrix holding the cumulative difference score for every possible pairing of applications in the pool. For an application pool consisting of n applications, the number of application pairs - and hence the number of application comparisons that have to be performed - is given by the following formula:
Based on the formula above, for an application pool consisting of 1000 applicants, the number of application comparisons required is equal to 499,500. For many degree programs, it is not uncommon for the number of applications to number in the tens of thousands, with the result that a staggering number of application comparisons would have to be performed at this stage. This is one of the primary reasons why such an objective approach needs to be automated with the help of software.
Application Clustering
Once the difference matrix for all application pairs in the pool has been calculated, the applications are clustered on the basis of their difference/similarity scores. This is done by searching the difference matrix for the application pair showing the greatest cumulative difference score, and then using the two applications in the pair as clustering points for all other applications in the pool.
In Figure 1 below, it is assumed that Applications A and D show the greatest cumulative score in the pool (as determined from the difference matrix), and hence will be used as the clustering points for the initial pool. This means that they will be used to divide the application pool into two different clusters. All other applications will be assigned to either cluster based on their respective differences/similarities to these two applications - for example, Application B is placed in the same cluster as Application A, because it is more similar to Application A than to Application D; in the same manner, Application F is placed in the same cluster as Application D, because it is more similar to Application D than to Application A, etc:
Clustering is repeated until the number of clusters is equal to the number of desired admits.
The clustering process described above is repeated until the number of clusters is equal to the number of desired admits. For example, if an institution has a target admissions goal of 4 admits, the clustering process is repeated until there are four clusters. At this point, each cluster represents a diverse, distinctive grouping of applications from which a single application is chosen as the recommended application.
Recommended Applications
Once the number of clusters is equal to the number of desired admits, a single application is chosen from each cluster as the recommended application. The recommended application is the one that is holistically most unique - i.e. shows the greatest overall difference, as determined by the difference matrix - relative to all other applications in the same cluster. The recommended application(s) from each cluster collectively make up the recommended application pool
It is important to note that in this final stage, the clusters represent groupings of applications whose diversity (differences) are not determined solely on the basis of any single factor such as race/ethnicity. Choosing a single application from each cluster therefore ensures the diversity of the recommended application pool in a manner that excludes race/ethnicity as the determining factor in admissions (such that one race is given preference over another), but which still allows race/ethnicity to be one of many factors contributing to the diversity of the recommended application pool.
Measurable Results
Applying the methodology used by Applications Quest creates measurable, consistent gains in diversity. In one case study involving the graduate psychology program at a major research university, the pool of applications recommended by Applications Quest showed a 12% gain in diversity compared to the accepted application pool identified through the university's regular evaluation methods; this gain in diversity was realized in a manner which not only ensured all recommended applications met the university's minimum requirements, but that the use of racial preference was precluded as a possibility.
