Date of Award

Fall 2015

Project Type


Program or Major


Degree Name

Doctor of Philosophy

First Advisor

Karen Conway

Second Advisor

Robert Mohr

Third Advisor

Damian Betebenner


Education production and, more generally, human capital production, is a significant driver of economic well-being both locally and nationally. This research informs policy decisions regarding how educational inputs are utilized and how those inputs can be influenced by the private sector. This dissertation, “Education Production: Teacher Evaluation Models and Weekend Feeding Programs,” is comprised of three chapters. The first addresses the rising issue in education policy of the measurement of student-test-based teacher evaluation. The second examines the scholastic effects of the relatively recent emergence of weekend feeding programs. The third uses the emergence and spread of the same weekend feeding programs to understand the factors affecting the adoption of new charitable programs.

Chapter 1 identifies and explains key differences between the two leading models used in evaluating teachers based on student standardized test scores. Historically, teacher evaluation has been based heavily on subjective measures of assessment. While few policy makers suggest these subjective measures of teacher quality should be discarded, all states now at least recommend the use of objective measures of teacher performance based on student test scores. The two most prominent statistical models of teacher effects, Value Added Models (VAM) and Student Growth Percentile (SGP) models, while conceptually similar, differ in estimation method and can lead to sizable differences in estimated teacher effects. Using a simulation, I evaluate the relationship under clean and controllable conditions. I then verify that the results persist in observed student-teacher data from North Carolina. The two models disagree regarding estimated teacher effects when the classroom distribution of test scores conditional on prior achievement is skewed (i.e. when a teacher serves a disproportionate number of high- or low- growth students). Moreover, the magnitude of skewness needed to drive these models apart by three or more deciles is within the ranges observed in the data. As such, a teacher who appears weak in one model may appear strong when evaluated by the other, yet typically only a single model is used. To further the effective use of these objective evaluation methods in high-stakes decisions such as payment, retention, and tenure, a complete understanding of their differences is vital.

Chapter 2 examines how weekend feeding programs affect both educational and behavioral outcomes. Weekend feeding programs provide weekend food to students who otherwise may have limited access to nutritious food between their free school lunch on Friday and free breakfast at school on Monday. In particular, I focus on the BackPack program (a national program through the Feeding America food bank) in Northwestern North Carolina. This is the first economic study to examine the effects of weekend feeding programs on standardized test scores, attendance, or behavioral incidents. I compile a unique data set of program enrollment at each participating school over time, which I then match to restricted student outcome and demographic data. Although the data do not identify which students actually receive the food, the intent-to-treat effect of these programs can be estimated by comparing the outcomes for students who are and are not eligible for the program via a quasi-experimental, panel difference-in-difference methodology. The potentially treated group is the free/reduced lunch eligible student population at schools with an active BackPack program. Ineligible students at BackPack schools and eligible students at non-participating schools act as control groups. Simple descriptive difference-in-difference models suggest these programs benefit students. However, more complete models reverse this finding, and I explore several alternative explanations. A false-start-date analysis (examining the effects of a two-year lead of program adoption) suggests that negative selection could be driving the negative association between student outcomes and program participation. Conditions two years prior to program adoption are associated with both poorer student outcomes as well as the decision to adopt a BackPack program. This work serves to provide a comprehensive first look at these programs and to highlight the importance of addressing non-random selection in identifying a causal link between program participation and student outcomes.

Chapter 3 uses the same BackPack programs from Chapter 2 in Northwestern North Carolina to understand key factors associated with the decision and timing of program adoption. Many studies have examined the causes and determinants of charitable giving, but not often are we able to observe the emergence, expansion, and adoption of a new and wide-spread private charity. The BackPack programs provide a unique microcosm of the donor-recipient relationship because they are largely locally funded and staffed and require substantial community buy-in. I use the same unique data set of program enrollment collected for Chapter 2, along with school data from the North Carolina Department of Public Instruction to understand recipient characteristics, as well as census tract data from the US Census and American Community Survey to understand donor (community) characteristics and how they changed over time. I find mild support for the common finding that donor racial diversity is negatively associated with charitable giving; however, I also show that donor income diversity is positively associated with program adoption (requiring donations of both time and money). I further find strong evidence that schools are more likely to adopt a program the more prevalent BackPack programs are in their county, as well as evidence that deteriorating local economic conditions (measured by the change in the local unemployment rate) play a significant role in adoption.

Taken together, Chapters 1-3 provide an examination of educational inputs: the measurement of teacher performance, the effects of weekend nutrition on scholastic outcomes, and the adoption process for student-centric community intervention. The following pages provide the details of these analyses.