Poster Abstracts, SAE 2001 USPS

Details for Poster Presentations

The poster session will be Wednesday evening in the reception area. Refreshments will be served there and hopefully most of the attendees will be there. Preseneters will have a flip chart stand that will hold roughly a 3' x 4' poster for the presentation. This reception will run from 6 to 7:30 approximately.

Poster Presenters

Julianna Berg, University of Nebraska -- Lincoln (with Karen Nylund) abstract
Marie V. Bousfield, City of Chicago, Department of Planning, City Hall abstract
Margaret D. Carroll, Centers for Disease Control (CDC) abstract
Michele Crescenzi, University La Sapienze Of Rome, Italy abstract
Michael R. Elliott, University of Michigan, School of Public Health abstract
Andrea Hicks, University of Nebraska -- Lincoln abstract
Xiaoming Liang, Louisiana State University abstract
Karen Nylund, University of Nebraska -- Lincoln (with Julianna Berg) abstract
Meri Raggi, University of Bologna, Italy abstract
Felix Seijas, Research Student Social Science, University of Southampton. abstract
Amang S. Sukasih, Department of Statistics, Texas A&M University abstract
Michail Sverchkov, Hebrew University abstract
Fujun Wang, University of Minnesota, Department of Biostatistics abstract

Poster Abstracts

 AUTHORS AND PRESENTERS: Julianna Berg and Karen Nylund,
University of Nebraksa -- Lincoln

TITLE: A look at the temperature display inconsistency using small amount of data.

ABSTRACT: Is it 40 degrees or 47 degrees? Within one block of each other in downtown Lincoln, two thermometers both give current temperatures that are consistently at least two degrees off from each other. How can there be that much inconsistency if they're measuring the same thing? This inconsistency isn't only true within the same block of downtown Lincoln; it's true all over the city. For our project we have decided to look at the thermometers around the city of Lincoln, Nebraska and estimate the error with which the temperature is measured.

We sampled utilizing multistage stratum design and estimated the lack of accuracy of the thermometers around Lincoln by calculating deviations from a consistent measurement. We're interested in both the magnitude of the difference and it's direction (i.e. above vs. below true temperature). We also considered if the average magnitude of the deviations are the same throughout stratums, different weather (cold vs. warm) as well as throughout the day.

Return to top
AUTHOR AND PRESENTER: Marie V. Bousfield, City of Chicago, Department of Planning, City Hall

TITLE: POPULATION ESTIMATION BY RACE FOR CENSUS TRACTS USING DYNAMIC MODELS

ABSTRACT: In this study, we describe how empirical Bayesian techniques and dynamic models can be used to prepare postcensal population estimates by race for small areas such as census tracts. Traditionally, these are obtained as fixed quantities called vital rates estimates and are derived using postcensal counts of vital events and assumed vital rates. In this study, a stochastic approach is developed by assuming that vital events are generated by Poisson processes, by borrowing strength from other areas to estimate vital rates, and by assuming a linear growth model for population change. This leads to a new dynamic population estimator obtained by using this year's vital rates estimate to update the extrapolation of last year's dynamic population estimator. This new dynamic population estimator is an improvement over the vital rates estimate especially for very small or rapidly changing populations and for the early years of the decade. Using the 1980 census counts and annual counts of births for the 1980-90 period, the theory is illustrated by deriving annual estimates of the number of African American women age 15 to 44 for the census tracts in Chicago for the 1980-90 period. A comparison between the 1990 estimates and the 1990 census counts is made.

Return to top
AUTHOR and PRESENTER: Margaret D. Carroll, Centers for Disease Control

TITLE: Model-Based Small Area Estimates of the Prevalence of High Serum Total Cholesterol Using Sample Selection Adjustment

ABSTRACT: Using data from the third National Health and Nutrition Examination Survey (NHANES III) and from the Area Resource File we apply Markov Chain Monte Carlo simulation techniques to estimate the prevalence of high serum total cholesterol of adults by states. To account for geographic variability and to model possible overdispersion of estimates, a two stage hierarchical model was used. We compare our model based estimates with design based estimates at the national level and obtain excellent agreement.

 Return to top
AUTHOR AND PRESENTER: Michele Crescenzi, Department of Statistics, Probability And Applied Statistics University La Sapienze Of Rome, Italy

COAUTHOR: Giovanni Maria Giorgi, Department of Statistics, Probability And Applied Statistics University La Sapienze Of Rome, Italy

TITLE: Poverty Measures Based On The Bonferroni Inequality Index

ABSTRACT: Within the most widely used poverty measures, the income inequality component is described by the famous Gini concentration ratio (R). We prove that the Bonferroni inequality index (B), in addition to the properties of R, is particularly sensitive to lower levels of income distribution and gives more weight to transfers among the poor. So, to improve poverty measurement, we propose three new summary indices based on B.

KEY WORDS: Bonferroni inequality index; Gini concentration ratio; Sen and Dagum poverty measures.

Return to top
AUTHOR and PRESENTER: Michael R. Elliott, University of Pennsylvania School of Medicine

COAUTHOR: Roderick J.A. Little, University of Michigan School of Public Health

TITLE: Estimating Undercount in the 1990 US Census by Combining Census, Coverage Measure Survey, and Demographic Data

ABSTRACT: Demographic analysis of data on births, deaths, and migration and coverage measurement surveys that use capture-recapture methods have both been used to assess US Census counts. These approaches have established that unadjusted Census counts are seriously flawed for groups such as young and middle-aged African-American men. Adjusting Census counts using only recapture data from follow-up surveys assumes that the probabilities of capture and recapture are independent and equal across individuals: failure of this assumption leads to "correlation bias," and combining Census, recapture, and demographic data allows identification of and adjustment for correlation bias under a variety of models of human behavior. Previous efforts to combine these data sources have yielded models that imply certain undesirable features, including implicitly negative cell counts and overly variable or outlying subnational estimates of population counts and sex ratios. We describe a Bayesian approach that reduces the problem of outlying adjustments for small sub-populations, tests for presence of bias in Census and recapture data, and can be extended to provide estimates of precision that incorporate uncertainty in the estimates from demographic analysis and other sources. The model is applied to data from the 1990 Census, and results compared with those from existing methods. Application to the 2000 Census will be discussed as well.

WEB PAGE: http://cceb.med.upenn.edu/elliott/mikehome2.htm

Return to top
AUTHOR AND PRESENTER: Andrea Hicks, University of Nebraska -- Lincoln.

COAUTHOR: Jennifer Joseph, University of Nebraska -- Lincoln.

TITLE: Renovating a Measure of Performance

ABSTRACT: A small cinnamon roll and snack company utilizes a traffic count to measure the performance of its franchises. The traffic count is performed by counting the customers that enter the store and the people that walk by the store. This project centers on perfecting this system. It addresses concerns in the current system such as sample size, measuring variability, and missing data. Then solutions are explained that resolve these concerns including stratification, minimum sample size, and other suggested changes. A system of Excel spreadsheets utilizes the collected data to create relevant graphs for the company and each franchise. The project is a good example of balancing statistical techniques and real world practice.

Return to top
AUTHOR AND PRESENTER: Xiaoming Liang, Louisiana State University, Department of Experimental Statistics

COAUTHORS: E. Barry Moser and Kenneth W. Paxton,
Louisiana State University, Department of Experimental Statistics

TITLE: Multidimensional preference analysis for Louisiana Cotton Farmer Survey

ABSTRACT: Multidimensional preference analysis is frequently used in consumer research to explore the preference pattern of subjects over services. However, the influences of subject attributes are usually not shown. A Louisiana Cotton Farmer Survey (LCFS) was conducted to estimate the importance of fourteen services. There were some missing values and subject mistakes involved in the survey data. Missing values were treated as the least preferred rankings while the RANK procedure was used to correct the subject mistakes. To evaluate the precision of this model, SURVEYSELECT was used to select bootstrap samples from the preference data set. Multidimensional preference mapping was constructed through PROC PRINQUAL and %PLOTIT macro to identify the most preferrred services. GPLOT procedure was used to produce a variety of graphics using the space. PROC CORR and PROC GLM, and two-way contingency table analysis were performed to judge if scores differ among sizes of farms and areas of Louisiana that those farmers are from. A one dimensional solution analysis was obtained to force a single ordering of preferences.

Return to top
AUTHOR AND PRESENTER: Meri Raggi, University of Bologna, Italy.

TITLE: A Hierarchical Generalized Linear Mixed Model for Estimating the Abundance of Animal Populations.

Return to top
AUTHOR AND PRESENTER:Felix Seijas, Research Student Social Science, University of Southampton.

TITLE: Labour Force Estimates for Venezuelan States

ABSTRACT: We base our study on the needs for labour force indicators shown by Latin American users, i.e. rates describing the labour force structure by sex-age groups for states. Labour Force Surveys (LFS) sample sizes usually guarantee reliable estimates only for large aggregated areas or sub-population. Although methods that provide us with options to tackle this kind of situations have been developed, most of them are based on the availability of auxiliary information such as administrative registers of unemployed. Unfortunately, a common feature in Latin American countries (LAC) is the lack of auxiliary information available with the minimum requirements to be used for statistics estimation purposes. As the main source (and the only one in many cases) of auxiliary information in LAC are censuses and their population projections, they are the only auxiliary data that any small area estimation procedure can take into account.

In this study we use the Venezuelan case as a model for the common situation faced by most LAC. We attempt to obtain simultaneous estimates for Venezuela of the four rates (Employment, Unemployment, Activity and Non-activity Rates) related to three basic counts (employment, unemployment and non-activity) that define the basic structure of the labour force. These indicators have to be obtained for sub-populations that are a combination of both spatial (states) and demographic (sex-age groups) dimensions. A requirement is that these counts have to agree with the aggregate counts at state levels as well as at the national sex-age groups.

As the sub-population groups can be seen as a cross-classification of the people by sex, age group and state, the method of Structure Preserving Estimation (SPREE) for categorical variables offers a possible answer to the situation described above. The problem is that LAC usually have unstable economies that do not guarantee the preservation over time of structures such as the labour force, especially at small area levels. We use product multinomial logistic regression in this first stage of the study to obtain indirect estimates, which are then equivalent to an application of the SPREE algorithm without the preservation of any marginal count from the census. A simulation study using samples withdrawn from the Venezuelan 1990 Census replicating as close as possible the Venezuelan LFS sample design is used in order to assess different logistic models and to evaluate the properties of the estimators.

Return to top
AUTHOR AND PRESENTER: Amang S. Sukasih, Department of Statistics, Texas A&M University

COAUTHORS: Eltinge, John L., Bureau of Labor Statistics, and
Weber, Wolf, Bureau of Labor Statistics.

TITLE: Diagnostics to Evaluate Mean Estimators Based on Consumer Expenditure Interview and Diary Data

ABSTRACT: The U.S. Consumer Expenditure Survey collects data through two instruments known as the diary and interview. For 78 expenditure items (recorded at the six-digit Universal Classification Code (UCC) level), data currently are collected through both the diary and interview, but published estimates are based only on the diary data. This paper develops methods for evaluation of the extent to which it may be feasible to use the interview data as auxiliary information to produce improved estimators of mean expenditures. The principal issues considered are: (a) measurement biases in the interview or diary data; (b) stability of relative bias ratios over time; (c) construction of a generalized least squares estimator that combines the diary and interview data; (d) approximation to the idealized estimator in (c); and (e) empirical assessment of the bias and efficiency properties of estimators in (d), relative to current diary-based estimators. Results for some UCCs are reported. Special emphasis is placed on the importance of checking the consistency of assumed conditions with observed data; and on direct evaluation of the bias and mean squared error of combined-data estimators.

Return to top
AUTHOR AND PRESENTER: Michail Sverchkov, Hebrew University

COAUTHOR: Danny Pfeffermann, Hebrew University

TTITLE: On Small Area Estimation Under Informative Sampling

ABSTRACT: Classical small area estimation techniques assume either that all the areas are represented in the sample or that the selection of the areas is noninformative. When the areas are sampled with unequal selection probabilities that are related to the values of the response variable, the classical estimators are biased; the magnitude of the bias depends on the sampling fraction and the covariances between the sampling weights and the response variable values. We illustrate this point using a very simple model employing the notions of the sampe and sample - complement distributions defined in the paper.

Return to top
AUTHOR AND PRESENTER: Fujun Wang, University of Minnesota, Department of Biostatistics

COAUTHOR: Melanie Wall, Division of Biostatistics, University of Minnesota

TITLE: Modeling Factors underlying Multivariate Spatial Data

ABSTRACT: In this paper, I will analyze spatially correlated multivariate data where the correlations within and across variables are hypothesized to be a function of one underlying spatially correlated factor. A spatial factor model is developed under normal assumption of the data and the parameters are estimated with the maximum likelihood method. Since the data set is just one sample of the population spatial random field, conditions for consistency and asymptotic normality of the estimators are discussed. Our final goal is to predict the underlying factor at observed locations as well as locations where no data are observed and provide a map which may be used to identify the spatial trends or clusters of high and low values of the underlying factor. Different spatial factor predictors are proposed and compared. The model is applied to a vector of several different disease specific mortality rates in Minnesota by county.

 Return to top
Michael D. Larsen
Last modified: Tue Apr 17 17:24:33 CDT 2001