This report examines treatment group members in randomized experiments who may choose not to participate in program services despite being offered access to these services. In this paper, a comparison of alternative computations of the standard error of the treatment-on-the-treated (TOT) impact is conducted using three approaches (the Bloom correction, Instrumental Variables (IV), and the bootstrap, all of which are understood as "truth" from simulations) and both simulated data (where the "true" standard error is known) and real-world data from the GJ-HC Impact Evaluation.

Affiliation: Abt Associates; Pennsylvania State University

Authors: Peck, Laura R. (Abt Associates); Litwok, Daniel (Abt Associates); Keele, Luke (Pennsylvania State University)

Key words: no-shows; Bloom correction; Instrumental Variables (IV); the bootstrap; simulations; simulated data; real world data; Green Job-Health Care Impact Evaluation