Abstract:
This paper assesses the performance of a large nation’s education system, combining results from a long-term field experiment with a structural-estimation framework developed to analyze how certification policies affect educational outcomes. The empirical motivation of this paper comes from the rates of high school mathematics certification–a prerequisite to training in technical and science-related occupations–that vary across nations: Tanzania, 17 percent; Kenya and Uganda, 50 and 54; the UK and India, 71 and 93.

A large-scale, long-term field experiment in Tanzania tested (1) financial incentives, (2) inputs, combining free solar-energy access, bilingual textbooks, and videos that showed how to study; and (3) both of the above. At the end of the third year, results on the incentivized mock test showed strong and significant impact of both (0.28σ in test score and 5.3 p.p. in pass rate), and a weak and insignificant impact of incentives or inputs alone. On the real certification test, however, no gains in allocated grades appeared, even though the latter test covered equivalent curriculum subtopics and took place only a month after the mock test. Of the difference, 40 percent was attributable to lower resolution of the grade brackets; 40 percent to deterioration in performance; 20 percent to control students catching up. Altogether, knowledge gains induced by these interventions could explain little of Tanzania’s “pass-rate gap.”

Motivated by the question of optimal certification thresholds, as well as heterogeneous treatment responses, I propose and estimate a model of knowledge production and “rational (knowledge) satisficing.” Preliminary results suggest that maximal average effort is elicited when the threshold is lowered to pass 57 percent, based on survey data matching study habits to outcomes. The magnitudes of predicted gains in average effort, knowledge, and welfare from that policy change are small, however, because of highly inelastic production-cost curves. A salient difference that the lower threshold would make is on the share of students who meet the status-quo threshold in expectation but fail because of tough luck: this share would reduce from 16 percent of those currently passing to less than 0.1.