Quality Management in Statistical Agencies - Maria João Zilhão, Instituto Nacional de Estatistica, Portugal and Mats Bergdahl, Statistics Sweden

The course will provide an overall picture of quality management approaches and how they have been implemented in statistical agencies. The aim is for the participants to be more able to implement quality management within their area of responsibility or to improve already existing approaches. Examples from National Statistical Institutes, mainly in Europe, will be provided and discussed.

The course has its roots in the work of the Leadership Group (LEG) on Quality, which presented its work at the conference 'Q2001 - The International Conference on Quality in Official Statistics'. Though the course content has been significantly updated to reflect advances that have been made within the LEG Implementation Group, that succeeded the original LEG, as well as within National Statistical Institutes and other organisations.

The course will cover Quality with a broad perspective, recognising that the users perception of quality is to a large extent based on the statistics and services that they utilise. The quality of the processes in place within an agency and how well they function in a system, will determine the capability of the agency to produce high quality output, that satisfies the needs of the users at a low cost. Areas that will be covered are: Data Quality, Quality Management Models, Process Orientation, Leadership and participation of Staff and Quality Assessment.

Model-based Small Area Estimation - Danny Pfeffermann, University of Southampton, UK

This course will review and discuss some model-dependent methods for small area estimation in common use, with emphasis on point prediction and MSE (Mean Square Error) estimation. We will distinguish between the case where the survey variable of interest is continuous and the case where the survey variable is discrete or a count variable. A further distinction made is between the use of the frequentist approach and the use of the Bayesian paradigm.

Provisional Program

1. Cross-Sectional Models for Continuous Measurements, I

• Motivation, drawbacks of design-based methods
• The Area Level Random Effects model, Best Linear Unbiased Predictor (BLUP), Empirical Best Linear Unbiased Predictor (EBLUP)

2. Cross-Sectional Models for Continuous Measurements, II

• The Nested Error Unit Level Regression model, BLUP, EBLUP
• Estimation of MSE of the EBLUP

3. Bayesian Small Area Estimation with Discrete Measurements

• Approximation of the posterior distribution by the Normal distribution
• Application to estimation of small area proportions
• Generation of observations from arbitrary posteriors, the acceptance/rejection method

4. Markov Chain Monte Carlo Simulations (MCMC)

• The Gibbs sampler, basic steps
• Implementation of the Gibbs sampler
• Application to estimation of proportions and counts

Sampling and Estimation in Business Surveys - Mike Hidiroglou, Office for National Statistics, UK

Business surveys are routinely conducted by statistics agencies. They are typically characterised by highly skewed universes that change quite rapidly over time. The challenge is to design surveys that will stand the test of time, and yet produce reliable data.

The course presents key issues in statistical and measurement design of business. It will describe the techniques for designing business surveys, introducing the participant to ad hoc and periodic (ongoing) business surveys. It will cover list frame construction and maintenance, as well as sampling procedures. Issues in building and maintaining a list frame (Business Register); sampling procedures, including sample size determination and allocation, and sampling methods; and weighting and estimation.

Introduction to Survey Quality - Lars Lyberg, Statistics Sweden and Paul Biemer, Research Triangle Institute, USAsampling

This course will span a range of topics dealing with the quality of data collected through the survey process. The course begins with a discussion of total survey error and its relationship to survey costs, and provides a number of measures of quality that will be used throughout the course. Then the major sources of survey error are discussed in some detail. In particular, we examine a) the origins of each error source (i.e., its root causes), b) the most successful methods that have been proposed for reducing the error emanating from these error sources, and c) methods that are most often used in practice for evaluating the effects of the source on total survey error. The course is not designed to provide an in-depth study of any topic but rather as an introduction to the field of survey data quality.

The purpose of the course is threefold:

• Provide an overview of the basic principles and concepts of survey quality, with particular emphasis on the components of sampling and nonsampling error
• Develop the background for the continued study of survey measurement quality through readings in the literature on survey methodology
• Identify issues related to the improvement of the survey quality that are encountered in survey work and provide a basic foundation for resolving them.

Practical Tools for Non-response Bias Studies - Bob Groves, University of Michigan, USA and Mike Brick, Westat, USA

This course discusses methods for conducting non-response bias studies to help reveal whether estimates are biased from non-response. Practical tools are described and examples are used to illustrate methods that can be used to conduct these studies. The advantages and disadvantages of these methods are presented, and the value of having multiple approaches is highlighted. The need to devise strategies for nonresponse and for its analysis in the planning stage, prior to completing the survey are emphasized.

The course is aimed at individuals in government, universities, business and nonprofit organizations who are involved in the development, implementation or evaluation of surveys. The course will assume a working knowledge of data collection methods in survey research. Examples will be presented and only rudimentary statistical knowledge of concepts such as bias and variance of the estimates is required of participants.