Statistical Engineering
Experimental Design and Data Analysis
Statistical Modeling for Information Technology
Modeling for Ultra Cold Neutron Lifetime Experiment
Bayesian Statistics for Metrology
Key Comparisons for Interlaboratory Standards Evaluation
Division Contact: Kamie Roberts
We collaborate with other NIST researchers and also with industrial partners to address measurement and standards aspects of physical science, engineering, and information technology. Together with subject matter experts, we develop techniques for evaluating measurement processes, tying measurement processes to accepted standards, and for ensuring the quality of measurements. We develop probabilistic models for physical processes, designs for experiments and strategies for data collection, methods for graphical and numerical data analysis, and procedures for evaluating experimental uncertainties. Our current collaborative research projects involve the development of:
 |
a statistical process control paradigm for security of computer networks that flags intruders as process anomalies through a visualization interface; |
|
a state-of-the-art collection of algorithms for testing the integrity of random binary generators that are key components of cryptosecurity systems; |
|
methods for assessing the spatial dimensions of physical scenes reconstructed by laser-based remote sensing systems together with evaluation of the uncertainties in the assessment; |
|
statistical methodology for comparing biometric algorithms for identification of an individual by gait; |
|
stochastic models for high-speed communications using optical fibers; |
|
new measurement methods for characterizing the complex permittivity of dielectric materials, widely used throughout electronics, microwave, communication, and aerospace industries; |
|
statistical models for polymer temperature and pressure measurement during fabrication; |
|
characterization of high-speed oscilloscopes for use in optoelectronic device metrology, non-linear device metrology and in high speed digital circuit design; and |
|
development and application of computational methods and algorithms for feature extraction and pattern recognition in images. |
Contact: James Filliben
Experimental Design and Data Analysis
Measurement is the backbone for advancing scientific research and creating new technologies. Expertise in the design of experiments, process characterization, modeling, estimation of components of variance, interlaboratory studies, quality control, and uncertainty analysis is coupled with a strong focus on applied research to bring NIST statisticians into contact with leading researchers in measurement science.
Here are examples of our work:
|
methods to characterize measurement uncertainty and establish traceability to NIST for highly accurate measurements made by coordinate measuring machines; |
|
a statistical measure of sharpness of scanning electron microscope images, now being used by industry; |
|
a statistical approach to measuring and comparing the propensity of different types of cigarettes to ignite upholstery; |
|
test methods for assessing the field performance of instruments that measure lead concentration in painted surfaces; |
|
statistical experimental design and analysis for testing police body armor; |
|
statistical methods for certification and calibration of electro-deposited coating thickness measured by magnetic induction; |
|
a generic test method for electromigration reliability testing of magneto-resistive heads used in computer disk drives; |
|
statistical procedures to reduce the number of tests on electronic devices by building a high-dimensional empirical linear model of a production process to predict results at unmeasured points; and |
|
methods for electrically based measurement of test structure linewidths on integrated circuits. |
Contact: James Filliben
Statistical Modeling for Information Technology
Commerce at the beginning of the 21st century is seeing an information technology revolution driven by software innovation, network systems, and hardware advances. Assessing quality in each of these areas draws on statistical models that are developed expressly for this purpose. At NIST, statistical modeling and research addresses these areas with major collaborative efforts in software testing, information retrieval, network performance assessment, and ongoing work on standards for electronics.
We have developed Statistical Reference Datsets, a web service for assessing the accuracy of statistical computations, that provides a collection of statistical reference data sets with certified results of computations using those data sets.
Contact: Nien Fan Zhang
Modeling for Ultra Cold Neutron Lifetime Experiment
An international team of researchers from NIST, Harvard University, Los Alamos National Laboratory, and the University of Berlin developed a new method measuring the mean lifetime of the neutron. This measurement allows scientists to test their current understanding of electroweak interactions and to estimate an important parameter for astrophysical theories and models. At the heart of this achievement was a new magnetic trapping technique and a two-stage experiment. The first stage neutrons are guided into a superfluid 4He bath where some dissipate almost all their energy by inelastic scattering. Then the neutron beam is blocked for the second stage, which is a decay stage during which decay events as well as background events are recorded. Designing this experiment required the development of new statistical theory. Based on a birth-death stochastic model of the neutron trapping process, developed an algorithm that determines the optimal time for each of the two stages. Then algorithms for drawing correct statistical inferences were developed from the likelihood models for two-stage neutron lifetime experiments. This new methodology has now been published; as a consequence, second-generation experiments are now being refined on the basis of this theory with incorporation of background correction at the second stage as well.
Contact: Dominic
Vecchia
Bayesian Statistics for Metrology
Bayesian statistical methods have become powerful and efficient tools in diverse areas of statistics, particularly with the increasing computing power of the last decade. The Bayesian approach to statistics provides a unified framework for optimally combining information from multiple sources and for incorporating previous experimental results and/or expert opinion into current experimentation and modeling. This results in simpler, highly efficient experimental designs and statistical analyses. Bayesian methodology can also be applied where a conventional (frequentist) analysis is either technically impossible or extraordinarily difficult. Both NIST researchers and industry have much to gain by taking advantage of this methodology.
We have undertaken the expansion of fundamental statistical theory necessary to the specialized development and implementation of Bayesian methods for metrology. The first topics considered are: traceability of measurements and the propagation of uncertainty; interlaboratory comparisons; calibration; and part inspection plans. The incorporation of prior information in each case is considered for three different circumstances: virtually no prior information available; prior information is available in the form of data gathered in past similar experiments or on past, similar occasions; prior information is available in the form of expertise elicited from scientists based on their past experiences relevant to the task at hand.
The dependence of the Bayesian experimental design on the particular prior information is quantified; and the dependence of the Bayesian inference from the data analysis is also determined to assess its robustness. Developmental work on Bayesian methodology includes comparison, both theoretical and by simulation of the relative advantages and disadvantages of Bayesian compared to classical frequentist methods. Graphical representations of Bayesian inferences are also being developed, implemented, and tested by applying them to NIST data in conventional experiments.
Contact: James Filliben
Key Comparisons for Interlaboratory Standards Evaluation
Interlaboratory studies establish and ensure measurement capability for commerce since accurate measurements are necessary for assessing product specifications. We have been responsible for the statistical design and analysis of interlaboratory studies for many years, but now we need to address new national and international needs for understanding comparability in a fundamental way. Recently, a new form of interlaboratory study, "a key comparison," has taken an important place in our mission. These key comparisons, international interlaboratory studies for comparing measurement results among the leading national metrology institutes, now serve as the technical basis for international agreements about standards. The International Committee on Weights and Measures has developed a Mutual Recognition Arrangement (MRA) which requires a measurement comparison mechanism that reflects accurately the true relationships between measurement systems maintained by its member laboratories. The results of these key comparisons must extend to members of regional metrology organizations to maximize recognition of measurement capabilities of other metrology laboratories around the world. Our first contributions to key comparisons focused on data analysis to ensure that reported uncertainties from comparisons are given at the 95 percent confidence level, per MRA policy.
Currently, we are designing experimental paradigms to assure efficiency as well as accuracy in these international interlaboratory comparisons. Statistical research in the optimal design construction for international comparisons, both among leading laboratories and within regional groups of laboratories, is being coupled with development of graphical and numeric presentations of results both for mean laboratory values and for their uncertainties. The use of key comparisons is increasing both in scope and in depth. Our experience with key comparisons in many different areas makes our group a natural hub where design templates and analytic methodology for key comparisons can help ensure the success of the MRA to streamline measurement issues in international trade and measurement science.
Contact: Nien
Fan Zhang
NIST/SEMATECH Engineering Statistics Internet Handbook
NIST statisticians together with the Statistical Methods Group of SEMATECH have developed a hyperlinked web document for constructing experiments and analyzing data in order to improve and document measurement and production processes. This electronic handbook is unique because readers access statistical software from within its pages to analyze either case studies in the handbook or to reproduce analyses on other data.
The handbook updates the NBS Handbook 91 Experimental Statistics and utilizes public domain software, Dataplot (developed at NIST), coupled with the case studies in the handbook. Individuals can access the handbook for instructions to construct a calibration curve from their own data. Alternatively, the handbook can be used in design of experiments, analysis of data using linear models, or other topics. The electronic handbook also interfaces with various commercial software packages.
Contact: James Filliben
Date
created:October
22, 2001
Last modified:
Aug. 07, 2007
July 17, 2006
Contact: inquiries@nist.gov