School Reviews

Analyzing how parents talk about US public schools online

Parents often select schools by relying on subjective assessments shared by other parents—which are increasingly becoming available on school ratings websites in the form of written reviews. We apply recent advances in natural language processing to analyze nearly half a million reviews posted by parents for over 50,000 publicly-funded US K-12 schools on a popular ratings website.
We find: i) schools in urban areas and those serving affluent families are more likely to receive reviews; ii) review language correlates with standardized test scores—which are known to closely track race and family income—but not school effectiveness, measured by how much students improve in their test scores at the school over time; and iii) the linguistics of reviews reveal several racial and income-based disparities in K-12 education. These findings suggest that parents who reference school reviews may be accessing, and making decisions based on, biased perspectives that reinforce achievement gaps.

Date: December 2019 - March 2021
Type: Research Lab
< Back to projects

Motivating Context

School ratings websites like,, and others have gained popularity in recent years as resources to help parents research and identify schools, and are now an important part of the landscape in facilitating parental choice. They have the potential to offer parents who may be less likely to be tapped into privileged networks the opportunity to identify higher-quality schooling options for their children. Given this context, it is important to further explore the information presented on these platforms to understand the role they could play in reducing, maintaining, or exacerbating inequalities in education.


We collect nearly 1 million reviews from school pages on and link them to school performance measures and demographics available through the Stanford Educational Data Archive, resulting in a dataset of nearly 500,000 reviews posted for over 50,000 publicly-funded US K-12 schools. We then build upon recent advances in natural language processing and interpreting black-box machine learning models to identify correlations between review text and these different school-level characteristics.

For more information on the approach and findings, please check out our publication below.