Remedying Education with Personalized Homework : Evidence from a Randomized Field Experiment in India

One of the main reasons for poor state of education in developing countries is an inability to meet the heterogeneous learning needs of a large student population with constrained educational resources. We provide experimental evidence that computer-generated personalized homework can effectively meet the diverse learning needs of middle school students in India and significantly improve their learning outcomes. We randomly assigned personalized homework to a group of students and non-personalized paper homework to the rest. Our algorithm dynamically generated homework questions in topics and in difficulty levels as per the learning needs of students. We found that students obtained lower homework scores in personalized homework because they solved questions in topics that they did not understood well due to personalization. In this process, they achieved higher learning, and hence obtained higher exam scores. We examined their performance in two standardized exams over an academic year and found that they obtained 4.16 percent (0.16σ) higher scores than that of their paper (non-personalized) counterparts in exams. We provide empirical evidence of higher learning from personalized homework by showing that: (1) students who obtained higher levels of personalization in homework obtained higher exam scores and (2) personalization in topics with high weight in the exam has a bigger impact on exam scores compared to personalization in topics with low weight in the exam. We also show that personalized homework significantly improved the learning outcomes of low and medium ability students. Thus, we contribute to the literature on personalization of education by establishing the efficacy of personalized learning by doing through homework assignments. Our work has important practical implications because implementing computer generated personalized homework is inexpensive. Further, this type of homework can substitute for regular paper based homework with minimal disruption in the regular classroom pedagogy and imposes no requirements of additional time and effort from teachers or students. Therefore, it is possible to implement personalization in homework relatively easily and this change is likely to be stable without need for external monitoring