Mid-Career Methodological Fellowships

Bayesian Model Evaluation and Prediction of Economic Time Series

9811983

Susan A. Murphy
University of Michigan
Total Award Duration: 12 months
FY98 Amount: $64,218

Abstract and Additional Information

Social scientists use event history models in order the understand the causes of variation in the duration and timing of many life events such as the duration of poverty spells, timing of retirement, timing of premarital birth, timing of drug abuse initiation, etc. An endogenous time-dependent covariate such as amount of governmental support and person-tailored intervention programs may be used to explain this variation. Since the value of the endogenous time-dependent covariate may be partially determined or selected by the subject, it is important to adjust for confounding. Confounding may occur when both the time-dependent covariate and the duration or timing are outcomes of a common cause. In this case the time-dependent covariate may be associated with the variation in the duration or timing, yet may not cause the variation. Yet in designing intervention programs or social programs it is important to understand the causes of the variation in duration/timing. In order to ascertain the proportion of variation in duration/timing caused by time-dependent covariate, confounding must be controlled. However, proper adjustment for confounding of the effect of a time-dependent covariate on the outcome needs very careful thought. Indeed, the traditional approach of controlling for confounding by including the confounder in the event history analysis model will often only introduce more confounding. This project will (1) apply an experimental perspective to questions concerning the effect of a time-dependent covariate on duration/timing, (2) illustrate the confounding issues inherent in measuring the effects of time-dependent covariates, (3) illustrate how confounding may be eliminated, and (4) develop research methodology to eliminate confounding of the effect of time-dependent contextual covariates on hazard rates in multilevel models. This research is supported by the Methodology, Measurement, and Statistics Program and the Statistics and Probability Program under the Mid-Career Methodological Opportunities Fellowship Announcement.

FY 1999

Learning and Using Combinatorial Structure in Language

9812169

Mark Johnson
Brown University
Total Award Duration: 9 months
FY99 Amount: $72,400

Abstract and Additional Information

Statistical language models are of considerable scientific importance, where they provide insight into the importance of various information sources for language understanding, and they have significant technological applications such as speech recognition and machine translation. To date most statistical language models have concentrated on simple phrase-structure and lexical relationships. This research investigates probabilistic models capable of describing richer combinatorial structures also found in human languages. The goal of this research is to understand how humans learn a language and use it to communicate. Because these linguistic processes seem to be inherently computational in nature---they crucially involve the transformation and manipulation of information-conveying representations---this research focuses on developing explicit computational models. Besides the primary scientific goal of understanding the human language learning and processing mechanisms, the research also has important technological applications in areas such as automatic speech recognition, information extraction, summarization and indexing of on-line texts, translation of texts from one language to another by computer, etc. This research is supported by the Methodology, Measurement, and Statistics Program and the Statistics and Probability Program under the Mid-Career Methodological Opportunities Fellowship Announcement.

Statistical Models for Monitoring Educational Progress

9907447

Brian W. Junker
Carnegie Mellon University
Total Award Duration: 12 months
FY99 Amount: $64,870

Abstract and Additional Information

This award supports the investigator's work at the University of Pittsburgh's Learning Research and Development Center (LRDC). Projects to be initiated include: (1) Analyzing school district data archives with an eye toward evaluating educational progress and monitoring the outcomes of educational innovations; (2) Exploring social judgement in education, in particular in the development of an institutional portfolio rating system for classrooms and schools, based on the "Principles for Learning" of LRDC's Institute for Learning; and (3) Laying technical groundwork for a bank of linked topical tests instantiating a purely standards-referenced testing program. All three are connected to ongoing research programs at LRDC. These projects are designed to contribute to the development of data collection systems for adequate school accountability systems and for educational policy evaluation. Research conducted through the Institute for Learning and elsewhere suggests that sustained improvement in student achievement is most reliably attained through institutional change. Yet most currently implemented accountability systems focus instead on individual student outcomes, and often are confounded with high-stakes decisions for individual students. The first project will explore whether existing school district data archives can be exploited to limit additional individual student testing when student achievement data is called for. The second project will apply methodology developed over the past ten years for student portfolio assessment to the development and rating of institutional portfolios intended to show that local institutions (e.g., classrooms, schools and districts) are engaged in a process of professional development that ensures long term gains for students. The banked tests in the third project would each cover fairly narrow topics, such as integer arithmetic, fractions, etc., and could be used for example to assess the distribution of student achievement within a district, school, or classroom relative to specific learning standards. This research is supported by the Methodology, Measurement, and Statistics Program and the Statistics and Probability Program under the Mid-Career Methodological Opportunities Fellowship Announcement.

0083418

Janet M. Box-Steffensmeier
Ohio State University
Total Award Duration: 12 months FY00 Amount: $53,745

Abstract and Additional Information

Scholars have long known that multiple events data, which occur when subjects experience more than one event, cause a problem when analyzed without taking into consideration the correlation among the events. In particular, there has not been a solution about the best way to model the common occurrence of repeated events, where the subject experiences the same type of event more than once. This research project will result in an assessment of whether one of the two main approaches for the study of repeated events, variance corrected or frailty, is better able to account for within-subject correlation. Monte Carlo evidence will help determine whether and under what conditions alternative modeling strategies for repeated events are appropriate. Next, the project will compare frailty and multi-level frailty models by examining the results of a standard hazard model with no correction for clustering, three single frailty effect models to allow for clustering, and finally a model based on a cross-classified frailty model that allows for clustering by all three levels. Finally, the project will investigate the treatment of missing data in analyses of heterogeneity. Simulations will be rerun comparing variance corrected and frailty models with the complication of missing data. The statistical work resulting from this project will help resolve debates about political dynamics, such as the liberal peace, by commenting on the reliability of the different modeling strategies used to test those theories and applying the models discussed. The fellowship period will allow me to deepen my understanding of event history models and acquire new skills in the areas of Monte Carlo simulations, Bayesian analysis, and computer programming. This research is supported by the Methodology, Measurement, and Statistics Program and the Statistics and Probability Program under the Mid-Career Methodological Opportunities Fellowship Announcement.

FY 2001

Social Security and Statistical Prediction Problems

0095343

Edward Frees
University of Wisconsin, Madison Total Award Duration: 12 months FY01 Amount: $69,816

Abstract and Additional Information

The objective of this project is to investigate the use of longitudinal/panel data methods to combine forecasts for different cohorts and series used in Social Security forecasting. Forecasts of the Social Security system rely on forecasts of two components, demographic variables and economic variables. The demographic variables, for example, include mortality, fertility, and immigration; the economic variables include labor productivity, labor force participation, and inflation. Because they are from different functional areas, the forecasts of each component tend to be done in isolation of the other. Moreover, series often are decomposed into cohorts, generally by age and sex and sometimes marital status, and then forecast individually, that is, each cohort is forecast in isolation of the others. The project will use Bayesian and empirical Bayesian methods to combine, in a disciplined manner, information about neighboring cohorts with expert opinion of the future and currently available data. This modeling strategy, together with data from selected series, will be used to construct and validate the proposed forecasting techniques.

As with almost every developed country, the U.S. government maintains a large financial security system that provides partial protection for its constituents in the event of adverse contingencies. The largest portion of the U.S. Social Security system is the Old-Age, Survivors and Disability Insurance (OASDI) program, which provides protection against loss of earnings due to retirement, death, or disability. The 2000 Social Security Trustees' Report forecasts that by the year 2037 revenues to fund this program will be exhausted. The magnitude and timing of this predicted shortfall heavily influences numerous academic and public policy debates concerning program reform. This project will supplement the forecasting methods used by the Social Security Administration and provide insights regarding the reliability of the predicted shortfalls. By combining information from different sources, the research will achieve more reliable (efficient) forecasts and forecasts of components that are integrated and consistent with one another. Moreover, the stochastic prediction methods will allow us to quantify the uncertainty in the forecasts in a probabilistic manner. This is not possible with the current deterministic forecasting system employed by the Social Security Administration. This research is supported by the Methodology, Measurement, and Statistics Program under the Mid-Career Methodological Opportunities Fellowship Announcement.

FY 2004

Mid-Career Research Fellowship for Study in Law and Statistics

0423077

Nancy Staudt
Washington University Total Award Duration: 12 months FY04 Amount: $64,869

Abstract and Additional Information

This award supports mid-career training in statistical methodology for application to legal questions. The training involves organizing two workshops on empirical research in the law, as well as a year of coursework in applied statistics. The workshops and the individual study will foster progress on a series of scholarly projects investigating legal decision making in the United States Supreme Court in the economic context. The underlying goal of the research is to understand the similarities and differences between court decisions in civil rights cases and in cases addressing economic issues. Formal training in statistics will advance the investigator's individual research on empirical legal questions. It also will enable her to join other legal scholars in bringing scientific and statistical methodology to the law school curriculum. Lawyers, law professors, and judges frequently face difficult empirical questions, but, for the most part, they have no formal training in quantitative research methodology. This mid-career training opportunity will help foster a sub-field in "law and statistics," thereby promoting empirical legal scholarship as well as law school methods courses. These activities are supported by the Methodology, Measurement, and Statistics Program and the Law and Social Sciences Program under the Mid-Career Fellowship component of the MMS Program Solicitation.