Random Effects and Latent Class Models for Nominal Data

Professor Sophia Rabe-Hesketh, UC Berkeley

The most common type of nominal data are polytomous responses arising for instance when subjects choose one of a set of  alternatives, such as which course to enroll in. A less well-known kind of nominal response is a ranking of alternatives, such as a preference ordering of candidates for admission to university.  Both types of nominal responses can be modeled by assuming that  subjects have unobserved, continuous 'utilities' for the alternatives.  When the subjects are grouped into clusters, for instance schools, between-cluster heterogeneity can be accommodated by including cluster-level random effects in the model for the utilities.  Alternatively, clusters can be assumed to be of different types yielding latent class models.

These ideas will be illustrated by discussing random effects models for longitudinal election data and latent class models for rankings of political values. 

Individual and Contextual Determinants of Job Turnover:Evaluating the effectiveness of child care retention incentive programs

Danny Huang, Ph.D. Quantitative Methods in Education 

Principled Staff turnover is an issue of considerable interest within the child-care industry.  Despite extraordinary efforts in field research in the past decade, the process underlying turnover among child-care providers remains baffling. In 2000, Alameda and San Francisco County implemented Child-Care Retention-Incentive (CRI) programs which sought to retain qualified child care providers through offering salary supplements and training incentives in the form of stipends.  Using data collected by the California Children and Families Commission for the evaluation of CRI program effectiveness, this study examines the turnover intentions of child-care teaching staff (assistant teachers, teachers, and teacher-directors), as well as the effectiveness of retention-incentive programs aimed at using stipend awards to reduce turnover among child-care providers.  

Principled Assessment Design for Inquiry (PADI):
A Framework for Developing Assessment Tasks for Science Inquiry

Cathleen Kennedy, Ph.D.

Designing assessment tasks that elicit reliable evidence about student knowledge beyond factual recall can be challenging. When we want to obtain evidence about a student's ability to apply scientific knowledge to practical problems we find that ancillary proficiencies such as knowledge of mathematics, knowing how to formulate explanations, and understanding how to interpret data of different forms may obscure the scientific knowledge we intended to assess. In many cases, it may be desirable to capture, rather than ignore, these other competencies (dimensions). To do so requires consideration of correaltional relationships among the competencies involved and also conditional dependencies among any responses that are elicited from common stimuli. Together these contradict the standard assumptions of unidimensionality and local independence common to classical test theory and the most common item response modelsóhence we apply the Multidimensional Random Coefficients Multinomial Logistic (MRCML) model, which addresses such complexities.

The PADI system is being developed to help assessment developers construct reusable task blueprints, or templates, to model such complex tasks in a robust manner that facilitates reliable interpretation of student work. The system is currently being used to model an existing paper-and-pencil summative assessment for the BioKids curriculum developed at the University of Michigan, and a new interactive online assessment that will provide both formative and summative feedback to students and teachers for a FOSS curriculum under development at the Lawrence Hall of Science. This presentation will describe the underlying MRCML psychometric model that is incorporated into the system, with an emphasis on within-item multidimensional bundling, and demonstrate how a number of these assessment tasks can be modeled, scored, and interpreted.

The Political Construction of Measuring School Growth: Federal and State Accountability Reforms

Bruce Fuller

Abstract:

Fuller will discuss how technical measurement issues -- tracking schools' progress and student achievement levels -- becomes shaped by political and institutional pressures. For background reading, go to PACE website

[http://pace.berkeley.edu/pace_publications.html] and read "Penalizing Diverse Schools?"

Real technical folk could also read David Rogosa's 'dialogue' over our analysis of NCLB distortions [http://www-stat.stanford.edu/~rag/nclb/pacerejoind.pdf

Mar30

(Tue)

Pre AERA Presentations

Aprl 13

(Tue)

BEAR Seminar cancelled due to AERA Conference

Quality Assessment Systems

Trond Fevolden- Secretary General

Danish Ministry of Education and Research

Mr. Fevolden will be talking about the testing system in Norway. The presentation will include  the system design, their national high school exam and Norwayís participation on international tests and how these have changed their belief in the systems ability to create equality.

Mr. Fevolden has been touring the world; Africa, Europe and the United States investigating issues of Quality Assessment Systems.  His U.S. visit has included The World Bank, the University of Minnesota and now the University of California, at Berkeley. He will be a Visiting Scholar at the School of Education starting May 1st.