Roundtable: [DATA] When Is a Result “Significant”?
(Tools of Analysis: Methods, Data, Informatics and Empirical Design)

Thursday, November 6, 2014: 1:00 PM-2:30 PM
Laguna (Convention Center)

*Names in bold indicate Presenter

Roundtable Organizers:  Laura Peck, Abt Associates
Moderators:  Molly Irwin, Administration for Children and Families
Speakers:  Melissa Kovacs, Maricopa County, AZ, Winston Lin, Abt Associates, Austin Nichols, The Urban Institute and Jeffrey Smith, University of Michigan

Many factors influence which evaluation results are deemed “important” for policy decisions and program design. At the design phase, these factors comprise the inputs to power analyses used to determine desired sample sizes and anticipate a project’s detectable effect sizes. At the reporting phase, these factors determine which results are elevated to a study’s main conclusions, to inform both scholarly literature and recommendations for policy makers and practitioners. These factors include the power to detect program effects, the threshold of statistical significance used to judge whether a finding arises not just by chance, and whether to use a one- or two-tailed hypothesis test. Also relevant is the designation of research questions as providing “confirmatory” or “exploratory” evidence, and the related situation of having many questions, outcomes or subgroups that imply adjusting for the multiplicity of tests. This roundtable brings together diverse perspectives to shed light on how we know when a given evaluative result is “significant.” Presenters include (alphabetically): Melissa Kovacs, Ph.D. (Public Policy, UMaryland), is the Director of Maricopa County’s Justice System Planning and Information program, where she regularly makes decisions regarding whether something is "significant enough" to pass along to policy makers to inform their decisions about policy change. She has held faculty positions at Arizona State University and Allegheny College, and a Fulbright professorship at Universitat Duisburg-Essen, Germany. She founded a statistics and evaluation consulting firm, FirstEval, and holds board positions with Human Services Campus, Phoenix; Maricopa Area of Government Continuum of Care Committee; and Arizona Fulbright Association. Winston Lin, Ph.D. (Statistics, Berkeley), is a Scientist at Abt Associates. His publications include “Agnostic Notes on Regression Adjustments to Experimental Data: Reexamining Freedman’s Critique” (Annals of Applied Statistics, 2013) and the co-authored book Does Training for the Disadvantaged Work? Evidence from the National JTPA Study. To this roundtable, Dr. Lin brings both familiarity with the scholarly literature on these questions and experience in the practical world of program evaluation. Austin Nichols (MPP, Harvard; Economics Ph.D., Michigan), Senior Research Associate in The Urban Institute's Income and Benefits Policy Center, specializes in applied econometrics, labor economics, and public finance. His research focuses on the well-being of families and varied social insurance programs. He also studies education, health, and labor market interventions, and determinants of poverty and inequality. To this roundtable, Dr. Nichols brings a rich discussion of the misuse of cluster-robust standard error estimators in that context of determining “when a result is ‘significant’.” Jeffrey Smith, Ph.D. (Economics, Chicago), is Professor of Economics and Public Policy at the University of Michigan. His research centers on experimental and non-experimental methods for the evaluation of interventions, with particular application to social and educational programs. He also has examined the use of statistical treatment rules to assign persons to government programs. He is well published on these topics and has consulted to governments in the United States, Canada, the United Kingdom and Australia on evaluation issues. He brings this vast scholarly work as foundation for addressing the question posed in this proposed roundtable.