2016-Survey Methodology 1


Survey Methodology 1 
Chair: John L. Eltinge (U.S. Bureau of Labor Statistics) 
[PDF]

JESSICA ANDREWS, Statistics Canada
Small Domains and Sparse Variables; Challenges with Creating an Active Collection Strategy  [PDF]
 
Statistics Canada has recently put into place the Integrated Business Statistics Program (IBSP) as a common framework to process its business surveys. One aspect of this new system has been to introduce an active collection strategy to improve efficiency, data quality and lower costs. The Quality Indicator and Measure of Impact (QIMI) collection tool was built to facilitate this. There have, however, been a number of challenges in turning a theoretical tool into a practical system. In this paper we explore some of the challenges faced including the impact of misreported values, small domains and sparsely reported variables. 
 
SUSAN DEMEDASH, Statistics Canada
Job Vacancy and Wage Survey Paradata Log Files: A New Realm of Respondent Behavioural Analysis [PDF]
 
The Job Vacancy and Wage Survey collects information from Canadian employers on a number of labour issues, such as job vacancies and hourly wages by occupation. Started in February 2015, this survey is Statistics Canada's largest business survey with 100,000 business locations surveyed quarterly. The survey is conducted via electronic questionnaire and underlying paradata information on the communication from the respondent's computer to an internal host server was deciphered and analysed. This paper presents paradata results of the early quarters of this survey, which can help improve our questionnaire and reduce response burden. 
 
MICHAEL FREIMAN, U.S. Census Bureau
Two Approaches to Disclosure Avoidance for a Microdata Analysis System at the U.S. Census Bureau  [PDF]
 
The U.S. Census Bureau is developing an online Microdata Analysis System where users request a table or other analysis of Census Bureau data and receive the results without seeing the underlying microdata. We compare two approaches to maintaining the confidentiality of the data: output suppression, where results are withheld if they are deemed too risky, and pre-tabulation perturbation, where the underlying microdata are modified before the analysis is performed. Our research shows severe limitations to a suppression approach, with adequate protection requiring an unreasonably large number of tables to be suppressed. Ongoing research on pre-tabulation methods is discussed. 
 
ZIXIN NIE, International Justice Mission
Methodological Challenges Faced in the 2015-2016 Kolkata/Mumbai Commercial Sex Worker Study [PDF]
 
International Justice Mission (IJM) conducted a study in 2015-2016 to measure the prevalence of minors working in the commercial sex industry of Kolkata and Mumbai, India. To sell minors for sex is illegal, and therefore the population is hidden and transient, resulting in several methodological challenges. We present the methods used to address and account for such issues, along with the results based on the resulting inference procedure, and we discuss how to address these limitations for future studies. 
 
JAVIER OYARZUN, Statistics Canada
Needle in a Haystack – How to Detect Break in Series: The Capital Expenditures Survey Integrated Business Statistics Program Experience [PDF]
 
Statistics Canada's Capital Expenditures Survey (CES), which collects data on capital and repair expenditures, was integrated into the Integrated Business Statistics Program (IBSP) for 2013. The transition from a stand-alone survey to the CES-IBSP survey introduced several methodological changes that could have potentially introduced a break in annual estimates between 2012 (CES-based) and 2013 (IBSP-based). Using SAS High-Performance Forecasting and along with the Time Series Research and Analysis Centre, a modeling approach was developed to determine whether a break in the series exists. An overview of the methodology used and results using the published 2013 estimates will be presented. 
 
QIAN WEI, Carleton University
Statistical Inference Based on Survey Weighted Composite Likelihood  [PDF]
 
Survey weighted composite likelihood (WCL) method (Rao et al. 2013) has been proposed to estimate multilevel model parameters when data are from a complex survey with a multistage sample design. In this paper, we study inferential properties of the WCL method, such as the asymptotic normality of the maximum WCL estimators and likelihood ratio tests. Simulation studies are conducted to evaluate the performance of the proposed test statistics.

Key Words: Composite Likelihood, Multilevel Models, Multi-stage Survey Data