ASA/NSF/BLS Fellowship Possible Research Topics in Statistical Methods and Computing


    General Information

    See the main Fellowship page for more information on eligibility and application requirements. Applicants may also be interested in topics in Economics or Behavioral Science.

    Statistical Methodology and Computing

    Proposals in this area should be for research projects generally applicable to the collection, processing, or analysis of BLS data.

    Statistical Quality Control

    BLS is continuing to strengthen our quality control program. Of special interest is the development of additional techniques for assessing the quality of data obtained from establishment surveys. Some areas for research include:

    • Integrated overall error models, with provisions for estimating magnitudes and interrelationships of various error sources
    • Methods of adapting evolutionary operations techniques to BLS operations
    • Effective patterns of quality control feedback, training, and incentives applicable to establishment surveys
    • The redesign of management information systems to incorporate process performance and data quality monitors
    • Development and measurement of proxy variables for estimating components of nonsampling error

    Item Imputation

    Diversity in the types of BLS surveys provides a wide variety of missing-data problems. Some examples are:

    • Longitudinal imputation techniques for establishment surveys where units remain in the sample for 4 to 5 years
    • Multivariate imputation techniques for surveys like the Consumer Expenditure Survey, which collects data on all sources of income, all types of expenditures, and household demographics.
    • Multiple imputation techniques for computing employment and wages from sampled establishments for various BLS surveys.

    Small Domain Estimation

    Demand is continuously growing in survey programs for estimates for domains where the corresponding sample size is insufficient to produce standard design-based estimates of sufficient accuracy. Consequently, model-based or model-assisted estimators are needed. BLS programs such as the National Compensation Survey program are interested in producing small domain estimates.

    Time Series Methods

    Recent years have seen broad advances in the field of time series modeling, but BLS is interested in further development of these methods and their application. BLS has a broad array of monthly and quarterly time series for study and testing. Topics of particular interest include:

    • Application of model-based seasonal adjustment methods
    • Estimation of standard errors for seasonally adjusted estimates, particularly from complex surveys
    • Treatment of outliers in time series modeling and seasonal adjustment
    • Multivariate time series modeling
    • Survey estimation incorporating time series methods

    Statistical Methods for Data Analysis

    Inference based upon data from complex sample surveys is a broad area of research. Examples are:

    • Analysis of longitudinally linked data sets from continuing surveys
    • Graphical analysis techniques
    • Analysis of sample and administrative data sets
    • Complex survey applications of nonparaametric regression and other computationally intensive statistical methods
    • Area-based statistics and spatial data analysis
    • Statistical analysis of unstructured text in survey records
    • Linking survey data with administrative records

    Statistical Graphics and Data Visualization

    BLS publishes large volumes of economic data on employment, wages, and prices. Research is needed in the following areas:

    • Improved ways of graphically presenting cross-sectional and time series data that are normally published in tables
    • Graphical methods to detect outliers in high-dimensional and massive data sets
    • Approaches that visually convey the uncertainty in predictions and estimates
    • Graphical techniques to dynamically explore BLS data spatially
    • Tools that allow users to do interactive analyses through the BLS public Internet site as well as our Intranet.

    Disclosure Limitation / Confidentiality Protection

    BLS is also interested in a wide range methods for disclosure limitation/confidentiality protection. Examples include:

    • Masking of microdata in public-use datasets
    • Construction of partially or fully synthetic datasets
    • General evaluation of disclosure risk and/or data utility in disclosure-limitation settings

    Information Dissemination

    BLS provides public access to micro-, macro-, and metadata via the Internet. Research is needed on:

    • Improving the ‘user friendliness’ of BLS data-dissemination web sites
    • Explaining simple and complex statistical concepts to the diverse audiences that visit BLS data-dissemination web sites
    • Improving the presentation of statistical information through the use of graphics
    • Providing tools that facilitate data analysis

     

    Last Modified Date: November 21, 2014