Donors Choose is a website designed to help teachers request donations for technologies, textbooks, and other resources. Donors Choose provides access to data from all of their projects. One of the data sources is a list of all of the requested resources and their cost. Another contains information about donors to projects. There are also essays in which teachers describe their need and how their benefits their students.
In order to examine differences between resources requested from different groups of teachers, I analyzed the need statements from teachers in high poverty and low poverty classrooms. In order to reduce some of the variability in their responses, I examined only science teachers. Also, to maximize the potential differences from an analytic perspective, I analyzed teachers in what Donors Choose determined were the highest poverty schools (n = 1,315) with those in low poverty schools. I then randomly sampled for 1,315 requests from students in low poverty schools, so the analysis was based on 2,615 total need statements.
To analyze the data, I adapted Sherin’s (2013) approach: sum(table(kfit$cluster)) 1. Process documents (in this case, the need statements) 2. Transform text into a term-document matrix 3. Cluster documents the documents (using hierarchical and k-means clustering) 4. Compare the groups using cosine similarity
Here are the number of tweets and the top 10 words in each cluster (or topic):
##
## 1 2 3 4 5 6
## 1110 150 569 113 347 326
## Cluster 1 Cluster 2 Cluster 3 Cluster 4 Cluster 5 Cluster 6
## 1 need printer science include kit their
## 2 student makerbot material cost set learn
## 3 ipad print math shipping machine project
## 4 computer filament activate targetnew build own
## 5 camera design book price knex them
## 6 case ink center fulfill electric enhance
## 7 laptop bundle handson science class work
## 8 projector color hand proposal circuit ipad
## 9 mini academic experience delta robot skill
## 10 digital create lab educate simple help
Here is a plot of the cosine similarity between the high poverty and low poverty groups and the clusters / topics:
High-poverty teachers seem more likely to request science laboratory / investigation-related materials (topic 3) and to make cost-sensitive requests (topic 4). Low-poverty teachers seem more likely to request digital technologies (topic 1), 3D printing / “Maker” technologies (topic 2), and materials for personalized learning (topic 6). It appears that robotics / K’nex kits (topic 5) are requested a bit more in high-poverty than low-poverty classrooms.
Donors Choose. (2015). Open data. Retrieved from http://data.donorschoose.org/docs/overview/
Sherin, B. (2013). A computational study of commonsense science: An exploration in the automated analysis of clinical interview data. Journal of the Learning Sciences, 22(4), 600-638.