UC Riverside

This dissertation consists of three essays that study higher education effectiveness and budget decisions. In the first chapter, I evaluate whether instructional spending in colleges, the main input in the human capital production process, is effective in promoting student earnings after graduation. Using a nationally representative dataset, I find a positive elasticity between instructional spending and student earnings of around 2 percent which is robust to various specifications and potential confounding factors. The effects are mostly driven by private institutions and four-year institutions. The effects are slightly lower for well-established public institutions. Cost-benefit analysis reveals that a student has to work for over 40 years until her increased earnings can cover the cost of the spending, indicating the cost-ineffectiveness of increasing instructional spending from an investment perspective.

In the second chapter, we analyze one driving factor for college budget decisions - the competition for good students. We construct a model to show that colleges choose their share of spending on educational inputs and on amenities to attract good students. Their optimal decision depends on their competitors' choices. Simulation results reveal that over the range where budget choices are commonly made, colleges respond positively to their competitors' decisions. In addition, when the competitive pressure from competitors weakens, such responsiveness also declines. Empirical evidence generally supports the predictions from the simulation exercise.

In the third chapter, we propose a method that bounds the treatment effect for the general population of interest under randomized controlled trials. Our bounds depend on a set of mild assumptions that are different from existing methods in the literature. Hence, we view our method as an alternative for applied researchers to choose from. Applying our method to analyze the effectiveness of a first-year learning program on retention rates, our bounds suggest a possible negative population effect and restrict the possibility of a large positive population effect, despite the estimated positive effect for the experimental sample. We also show that assumptions required for other methods in the literature to be applicable are unlikely to hold in this context, highlighting the importance of the availability of our new method.