Andew Ho (Applied Stats Workshop)
Date and Time
Location
Speaker & Title
Andrew Ho, "Two Watches: Measurement Error Models for Estimating Educational Progress from Discrepant Test Score Trends"
Abstract
Understanding large-scale educational progress often requires reconciling information from multiple testing programs that differ in their purpose, precision, and periodicity. Like two watches that disagree about the time, two tests may report different trends for the same populations, subjects, and time periods. We develop a precision-adjusted multilevel measurement error model of the relationship between score trends from the National Assessment of Educational Progress (NAEP) and state testing programs in the United States. The model jointly estimates the true variance in NAEP trends, the true variance in state trends, their respective reliabilities, their true correlation, and systematic bias in state trends. We find that NAEP trends have reliabilities around 0.5, whereas state test trends have reliabilities near unity due to census testing. The true correlation between NAEP and state test trends is around 0.6, and state tests show upward bias around 0.04 standard deviation units per two-year period from 2009 to 2024. This modeling framework accounts for both measurement error and negative serial correlation in consecutive trends, enabling optimal estimation of past trends and prediction of future trends that neither test alone provides.
The Applied Statistics Workshop (Gov 3009) meets all academic year on Wednesdays. This workshop is a forum for advanced graduate students, faculty, and visiting scholars to present and discuss methodological or empirical work in progress in an interdisciplinary setting. The workshop features a tour of Harvard's statistical innovations and applications with weekly stops in different fields and disciplines and includes occasional presentations by invited speakers.
All interested Harvard affiliates are invited to attend. Lunch will be provided.