Using LASSO Regression to Analyze School Performance

While interning with the NC Department of Public Instruction in 2017, I designed and executed a study to determine the impact of teacher-held master’s degrees on a school’s performance score (a composite of 20% student growth and 80% academic achievement). The hope was that my work would convince NC Legislators to restore master’s pay for teachers. Unfortunately, this study found a weak correlation between the proportion of teachers with master’s degrees and the school’s performance scores, accounting for only 3% of the variance in 2014. This meant that, on average, for each 10-percentage point increase in teachers with master’s degrees, there was a 3.5-point increase in the school performance score.  

After working as a high school math teacher, I learned firsthand why having a master’s degree may not impact student performance when faced with the other challenges schools, teachers, and students experience inside and outside of the classroom. While learning about LASSO regression in machine learning at the Institute for Advanced Analytics, I wanted to see if better tools and new skills expanded my perceptions. By implementing decision trees and LASSO regression in Python, I determined that while the proportion of teachers with master’s degrees did not significantly affect 2017 school performance scores, the following factors did:

  • Proportion of economically disadvantaged youth
  • Number of short-term suspensions
  • Race
  • Graduation rates
  • Teacher experience

Modeling School Growth and School Performance Scores

Since growth is a component of the school performance score, I created a decision tree to predict school growth (met growth, did not meet growth, and exceeded growth). Some schools that exceeded growth had teachers with more years of experience, fewer teachers with graduate degrees, and a lower proportion of economically disadvantaged students. Ultimately, it was challenging to determine which features contributed to school growth. 

Examining the factors that contribute to the school performance grade (A-F) with a decision tree revealed that a lower proportion of economically disadvantaged youth resulted in higher grades. Some major differences between C and D schools included federal funding and the number of short-term suspensions per 100 students. More alarming, however, was the finding that a smaller proportion of white students (indicating a larger proportion of minority students) was associated with a lower school performance grade. Typically, schools with a population in which less than 21% of the students are white received a D or F grade.

To identify the variables that contribute to school performance scores, I built three separate LASSO regression models using the significant variables from the growth decision tree, the significant variables from the school grade decision tree, and a selection of all variables to ensure convergence. Figure 1 compares the coefficients for each of the three LASSO models.

Figure 1: Variable Significance for 3 LASSO Regression Models for School Performance Grades

Overall, this figure indicates that while the teacher credentials are important, features of the students’ environment are more important. For example, school performance is affected by more significant variables, like economics, race, and school funding. Many of these variables may show correlations.

Results for the Impact of Master’s Pay

The proportion of teachers with advanced degrees, master’s degrees, and national board certification was not significant in determining school performance. This indicates that teacher credentials may not matter, and reinstituting master’s pay may not be necessary. Instead, teacher experience is a crucial component of school performance. However, this result may be influenced by the fact that lower-performing schools generally have more difficulty employing experienced teachers (NC DPI).

Regardless, the importance of teacher experience is concerning for North Carolina public schools, given that approximately 11.3% of teachers with three years’ experience or less leave the profession each year. It is challenging to build an experienced population of teachers when many leave before they clear beginning teacher status. 

Impact of School and Environmental Factors

Based on these results, it seems that larger problems exist within NC public schools. When taken individually, the proportion of teachers with less than 3 years of experience explains only 12% of variance in school performance. The number of short-term suspensions per 100 students explains 17% of variance in school performance, and the proportion of economically disadvantaged students explains 47% of the variance. Figure 2 shows the relationship between school performance and the proportion of economically disadvantaged students.

Figure 2: Relationship between school performance scores and the proportion of economically disadvantaged students

Figure 2 indicates that for every 10-proportion point increase in the number of economically disadvantaged students in a school, there is a 5-point average drop in the school’s performance scores.

Furthermore, it is extremely concerning that the proportion of white youth explains 33% of variance in school performance scores. This may be influenced by the relationship between race and economic disadvantage. The proportion of white students could explain 27% of variance in the proportion of economically disadvantaged students. Schools that have a larger population of white students tend to have a smaller population of economically disadvantaged students.

Recommendations for NC Public Schools

After revisiting this project from the NC Department of Public Instruction, these findings indicate that we should focus our efforts on incentivizing teachers to remain in the profession for at least 4 to 10 years. It may be necessary to improve teacher working conditions and increase teacher pay to retain teachers. Furthermore, these findings indicate that schools with a greater proportion of economically disadvantaged students need greater support. Merely implementing a pay increase to incentivize teachers to work in schools with difficult circumstances will not be enough. Some resources that would have helped me as a beginning teacher support students with high-needs include:

  • Provide opportunities to co-teach with another certified content teacher to learn the school culture and extend opportunities to learn from veteran teachers.
  • Include a “floating” teacher to support students who may need 1-on-1 support or opportunities for advanced learning.
  • Implement smaller class sizes to allow for more individualized attention and small group support.
  • Integrate counselors in the classroom at least twice a month to help students develop emotional intelligence to manage stress, anger, and complex emotions.
  • Increase support for implementing accommodations for students with special needs that require a separate classroom, such as small-group testing and read-aloud for classes without an EC teacher.
  • Improve support for adapting assignments for students with special needs and implementing Tier 2 and Tier 3 supports for neurotypical students.
  • Increase opportunities for students to learn about potential careers to motivate them to work hard in the classroom.
  • Increase opportunities to visit museums, meet with professionals, and travel to trade schools, military organizations, community colleges, and universities to increase awareness.
  • Implement parent and student meetings to learn when to ask a teacher for help, how to contact the school for support, and what parents can do at home to support student learning.

As a former teacher, I learned that concentrating large numbers of students with higher needs in one room or one school is overwhelming for students and teachers. It makes it far more difficult for any student to get the necessary attention or support needed to thrive when the teacher is spread so thin by the increased needs of economically disadvantaged students. Implementing classroom size reductions in these schools by hiring more teachers may better support these students. Furthermore, if it is possible to hire “floating” teachers who can provide one-on-one or small group support, it may lessen the load on the primary classroom teacher and allow them to meet individual students’ needs.

While reinstituting master’s pay may be crucial to increase teacher retention and therefore increase teacher experience, master’s pay is not the root of the problem. The school and surrounding community’s economic, racial, financial, and administrative characteristics seem to be the underlying relationship to school performance scores. This means we have significantly more work to do to increase equality in education across NC. Increased opportunity may disrupt the current cycles.

Ashley Avis


Keung Hui, T. (2019, February 7). Will NC teachers once again get extra pay for master’s degrees? Plan has bipartisan support. The News & Observer.

Keung Hui, T. (2018, June 1). High poverty schools lack experienced teachers. NC says its options are limited. The News & Observer. 

North Carolina Department of Public Instruction (2020, February 5). 2018-2019 State of the teaching profession in North Carolina. 

Data source

All data compiled from NC Department of Public Instruction.