Data for Diplomas

Inspiration

We combined our experience in school data, machine learning and mapping visualization to put together our website that highlights school districts who are graduating more students than expected.

What it does

Provides scores for school districts to understand how their graduation rate compares to the peers. We also score schools on factors that we found to be predictive of improving graduation rates.

How we built it

We used various datasets, R statistical packages, and QuantumGIS to create our data. We then used Meteor to create a webapp.

Challenges we ran into

It was difficult to clean and join the various data source files. We also had to determine the appropriate statistical methods and perform feature engineering to get more meaning out of the raw data.

Model Building

Our two-step model building process first involved building a PCA factor model with the census demographics, student demographics and school district state and federal funding amount datasets and then performing our cross-validated lasso regression on the resulting factors. The PCA model excluded the school district specific interventions we wanted to test such as, teachers per pupil. This model gave us an in-sample R-squared of 55.05% and out-of-sample R-squared of 48.89%. We next tested our interventions using a cross-validated lasso linear regression to see which "interventions" could predict the residuals from our first model. Our signficant variables increased our in-sample R-squared by 5.37% (for 60.42% total in-sample R-squared) and out-of-sample by 4.24% (for 53.14% out-of-sample R-squared).

Scoring

We ranked all the schools who outperformed their modeled expectation on the following metrics and calculate percentiles to achieve scores from 0 - 100. These metrics, excluding graduation rate, were found to improve or hurt (-) graduation rate.

Graduation Rate

Compensation: % of expenses on teacher salary, % of expenses on teacher benefits (-), total employee benefits per pupil, total spending per pupil on school

Schools: schools per pupil (-), high schools per pupil (-), charter schools per pupil (-), magnet schools per

Staff: teachers per pupil, guidance counselors per pupil