| Home| About BEAR| Events| Research Projects| Publications| Archives|Measurement| Contact BEAR |

 http://bear.berkeley.edu

B  E  A  R   C e n t e r

 
          BEAR Center Logo

Berkeley Evaluation & Assessment Research Center

Director: 

 

Mark Wilson

Convenor:

 

Carolyn Hofstetter

Coordinator: 

 

Mariella Ruiz

 

| 3 Sep | 24 Sep | 1 Oct| 21 Oct | 22 Oct | 29 Oct | 19 Nov | 26 Nov

|Archive of Past Events |

 


BEAR Events
Fall 2002

Unless otherwise specified, events take place from 2-4 PM at:
UC Berkeley
Graduate School of Education
Tolman Hall
Room 2515

  

3 Sep

(Tue)

 - 

Research Challenges and Opportunities in Test Publishing

Dr Richard Patz, UC Berkeley & CTB/McGraw-Hill

Interested in learning what it is like to conduct research in business and industry? Visiting Scholar Rich Patz, on sabbatical from his role as director of research for CTB/McGraw-Hill, will discuss the business of educational testing and assessment, interesting and active areas of research in the current assessment climate, and career paths in research organizations. Rich will discuss some of the research he will be engaged in during the tenure of his Berkeley visit.

 

 

 

24 Sep

(Tue)

 - 

The Hierarchical IRT Model for Criterion-referenced Measurement & the Problem of Standard Setting

Rianne Janssen, Catholic University of Leuven (Belgium)

In the hierarchical IRT model for criterion-referenced measurement (Janssen, Tuerlinckx, Meulders, & De Boeck, 2000) items are grouped under criteria (e.g., "standards") and for each criterion a set of hyperparameters is introduced. This allows one to estimate a success probability on a criterion itself, rather than on individual items, as is standard in IRT. It will be argued that this model can be relevant for the problem of standard setting in two interrelated ways. First, the success probability on a criterion can be interpreted as a domain score. Bock (1997) has already discussed the use of domain scores for standard setting. Second, the hierarchical IRT model can be supplemented with item judgments according to the method of Jaeger (1978). This would lead to a weighted success probability on the criterion, with each item weight expressing the importance of the item according to the judges.

 

 

 

1 Oct

(Tue)

 

 - 

Multilevel item-response and structural equation modeling

Dr Sophia Rabe-Hesketh, Institute of Psychiatry, Kingís College London, UK

In item-response models, ability is typically treated as a latent (unobserved) variable, assumed to be independently normally distributed with constant mean and variance. A useful extension of this model is to allow mean ability to depend on observed covariates.

In multilevel or hierarchical settings such as schools, where students are nested in classes and classes in schools, important predictors of students' abilities are likely to include class and school level covariates such as teacher to student ratio. However, in practice not all these higher level covariates will be observed, leading to unexplained differences in mean abilities between schools and between classes within schools, even after adjusting for the observed covariates. As a result, the abilities of students in the same class or school can no longer be assumed to be independent (conditional on the covariates). We can model this dependence by including random effects of class and school in the model for ability. The model can be further extended by including latent covariates, for instance teacher's attitude, leading to multilevel structural equation models.

These kinds of models can all be formulated as special cases of a general model framework, GLLAMM (Generalized Linear Latent And Mixed Models), and estimated using the Stata program, gllamm.

 

 

 

21 Oct

(Mon)

 - 

Workshop: Bayesian IRT Modeling with WinBUGS

Dr Werner Wothke, CTB/McGraw-Hill

WinBUGS is a software which aims to make practical MCMC methods available to applied statisticians and quantitative researchers.Ý The workshop will cover setting up a 1PL IRT model, diagnostic with MCMC simulation and other IRT models (2PL, 3PL).

 

ÝÝÝÝÝÝ

 

22 Oct

(Tue)

 - 

Paradigms, Models, and Methods for Statistical Inference in Educational Measurement

Dr Richard Patz, UC Berkeley & CTB/McGraw-Hill

The application of Bayesian statistical methodology in physical and social sciences has seen a dramatic increase in recent years.Ý Powered on the one hand by an attractive formal framework for inference and on the other by increasingly powerful set of statistical computing tools, Bayesian inference is affecting educational research in significant and sometimes controversial ways.Ý

ÝÝÝÝÝÝÝÝ In this talk we will first examine several interesting issues in the foundations of statistical inference that have implications for the way we interpret information collected in educational measurement and assessment contexts.Ý Second, we consider student proficiency, and the data and models used in its assessment.Ý We explore the suitability of Bayesian analysis and the applicability of Bayesian statistical computing methods, including Markov chain Monte Carlo, for answering important questions regarding student achievement.Ý Finally, we examine some recent and emerging examples of the application of Bayesian statistical inference in educational measurement and assessment settings.

 

 

 

29 Oct

(Tue)

 - 

SAT Coaching, Bias and Causal Inference

Dr Derek Briggs, UC Berkeley

This presentation considers the extent to which unbiased causal inferences can be drawn about the effect of coaching on SAT performance.Ý Following a review of the literature, I present the linear regression model and the Heckman Model as two statistical approaches that might be used control for bias in an estimated coaching effect.Ý The assumptions necessary before an estimated effect can be given a causal interpretation are described in some detail.Ý I estimate coaching effects for both sections of the SAT using data from the National Education Longitudinal Study of 1988 (NELS).Ý There is some indication that the linear regression model successfully reduces bias due to omitted variables.Ý It appears that commercial coaching programs have an effect of about 3 to 20 points on the verbal section of the SAT, and an effect of about 10 to 28 points on the math section of the SAT.Ý These effects may be somewhat bigger or smaller if coaching is defined more broadly.Ý There is some evidence that coaching is more effective for certain types of students.Ý I demonstrate the sensitivity of the Heckman Model to the choice of variables included in the selection function.Ý Small changes in the selection function are shown to have a big impact on estimated coaching effects.

 

 

 

19 Nov

(Tue)

 - 

Dr Jason Ravitz, Buck Institute for Education

Title to be announced.

 

 

 

26 Nov

Ý(Tue)

 

Paul Warren, Deputy Superintendent, Accountability Branch, California Department of Education

 

Title to be Announced

 

 

 ^