Just as data science is applied to drive major success in business, it’s also being applied to improve or resolve major social issues. That’s the mission and focus of DataKind, a non-profit that organizes data science volunteers, so that they can lend expertise to social organizations, civic groups, and non-governmental organizations as part of efforts to create a better, data-driven future for all humanity. With six chapters on three continents, DataKind has been doing lots of good for lots of people.

You may be familiar with the business intelligence company, Teradata. Well, what you might not have known is that Teradata also has a volunteer program of its own, called Teradata Cares. Teradata Cares recently partnered with DataKind to kick off Teradata’s annual conference that was held in Nashville, Tennessee from October 19th to 23rd, 2014. As part of this partnership, the two days before the conference were dedicated to hosting a DataDive (or, a volunteer work session between DataKind and Teradata Cares volunteers) to help projects by iCouldBe, the Cultural Data Project, HURIDOCS, and GlobalGiving. As part of a small Data-Mania blog series featuring this work, today’s post covers the work that was done for iCouldBe and Cultural Data Project.

iCouldBe: The metrics of mentoring

iCouldBe is a non-profit organization that acts as an online mentoring program to e-mentor at-risk youth students, in order to help them stay in school and succeed in their coursework. Since the year 2000, they’ve successfully helped over 19,000 young people, while at the same time collecting a large dataset of student-mentor interactions. People at iCouldBe were wondering if they could use this data to predict whether a student-mentor relationship would be successful so that they could use this information to adjust their curriculum and/or train mentors to better help and keep students.

The first, vital task was to define a metric for success; After looking at the data, DataDive volunteers determined that the criteria for success should be set as the successful completion of three learning modules in three months. When volunteers analyzed student-mentor interactions, they found that the more verbose the mentor was, the more likely the student was to leave the program. In other words, mentors need to keep it simple. By contrast, mentors who used encouraging, supportive phrases were more likely to receive back appreciative and positive student responses. This information has provided the organization a data-driven framework to ensure they help even more students, and a programming infrastructure to analyze future student-mentor interactions.

Cultural Data Project: helping the arts succeed

The Cultural Data Project collects and distributes data about more than 11,000 American arts and culture organizations. In this portion of the DataDive event, volunteers were asked to analyze the arts and culture data to determine what factors make for project success. The team used machine learning techniques to cluster the organizations based on factors like size, number of funding streams, and interest area. Volunteers then identified what clusters were more financially successful, and developed a taxonomy to allow arts and cultural organizations to determine in what part of the model they fit. Interestingly, volunteers found that the cluster with the least chance of financial success was also the hardest cluster to describe in a simple taxonomy – but more investigation is needed to determine whether the commonalities between organizations in this cluster are related to the overall financial underperformance of the cluster at-large.