Grant Open

Statistical Innovations for Clustered Observational Studies

NCER

Program: Statistical and Research Methodology in Education

Program topic(s): Core

Award amount: $899,023

Principal investigator: Luke Keele

Awardee:

University of Pennsylvania

Year: 2021

Project type:

Methodological Innovation

Award number: R305D210014

Purpose

In many observational research settings in education, the treatment is allocated to entire clusters of students, instead of to individual students. These clustered observational studies (COSs) arise when the treatment is applied at the group level (e.g., teachers, classes, or schools) with outcomes of interest measured at the student level. The intervention is not randomized, but it is clustered such that all students within a given cluster are treatment or control students. There is very little research on the best way to design COSs and on statistical methods for the analysis of data from a COS. The purpose of this research is to develop causal estimands for COSs, guidelines for their use, and software for computing the estimands.

Project Activities

The development and guidelines will involve theoretical work and Monte Carlo simulation studies.

People and institutions involved

IES program contact(s)

Charles Laurin

Education Research Analyst

NCER

Products and publications

Products: The software will be an open-source R package for which the research team will also create vignettes, a user's guide, and other training materials. The theoretical work and the software package will be published via journal manuscripts, conference presentations, and seminars.

Publications:

Ben-Michael, E., Page, L., & Keele, L. (2024). Approximate balancing weights for clustered observational study designs. Statistics in Medicine.

Ye, T., Westling, T., Page, L., & Keele, L. (2024). Nonparametric identification of causal effects in clustered observational studies with differential selection. Journal of the Royal Statistical Society Series A: Statistics in Society, qnae018.