This paper estimates a private school learning premium in Tanzania by implementing a flexible value-added model with unique administrative data on exam scores. The dataset covers 635,000 secondary school students with information on both their primary and lower secondary school exam records, allowing three out of four assumptions imposed in standard value-added models to be relaxed.

The preferred coefficient estimate suggests private secondary schools, on average, increase student exam scores at national exams by 0.45 standard deviations after two years of secondary schooling. Standard value-added models are found to bias the learning premium upward.

An instrumental variable model confirms a positive causal learning premium. Subjectspecific learning premiums for Kiswahili, English, and mathematics are 0.28, 0.39, and 0.50 standard deviations, respectively.