Chapter 12: Other nonresponse correction techniques

Slides:



Advertisements
Similar presentations
Aaker, Kumar, Day Ninth Edition Instructor’s Presentation Slides
Advertisements

Chapter 7 Selecting Samples
Stratified Simple Random Sampling (Chapter 5, Textbook, Barnett, V
Marketing Research Aaker, Kumar, Day Seventh Edition Instructor’s Presentation Slides.
Sample Design.
SPECA Regional Wrokshop on Disability Statistics, Dec 13-15, 2006 Issues Related to Disability Measurement: Cognitive testing and mode Jennifer Madans.
Sampling Techniques LEARNING OBJECTIVES : After studying this module, participants will be able to : 1. Identify and define the population to be studied.
Scot Exec Course Nov/Dec 04 Survey design overview Gillian Raab Professor of Applied Statistics Napier University.
The Challenge of Non- Response in Surveys. The Overall Response Rate The number of complete interviews divided by the number of eligible units in the.
Population vs. Sample. Population: a set which includes all measurements of interest to the researcher (The collection of all responses, measurements,
Implementing cawi into the data collection process Kees van Berkel Mariëtte Vosmer Jerusalem, July 2013.
Sampling: Design and Procedures
Sampling From Populations
Marketing Research Aaker, Kumar, Leone and Day Eleventh Edition
Part III – Gathering Data
CHAPTER 4 Designing Studies
SAMPLE DESIGN.
Sampling: Design and Procedures
Sampling: Theory and Methods
Sampling: Design and Procedures
Mixed Mode Effects of Web and Telephone Surveys Using Coarsened Exact Matching to Explore the Results on Employment Status Joachim Schork, Cesare A. F.
The second wave of the new design of the Dutch EU-SILC: Possibilities and challenges Judit Arends.
STRATIFIED SAMPLING.
Keller: Stats for Mgmt & Econ, 7th Ed Data Collection and Sampling
Chapter 2: The nonresponse problem
The European Statistical Training Programme (ESTP)
Sampling: Design and Procedures
Data Collection and Sampling
The European Statistical Training Programme (ESTP)
AP Stats – 4.1 Sampling and Surveys.
EVALUATING STATISTICAL REPORTS
Dependent interviewing in the Swedish LFS
Chapter 7: Reducing nonresponse
The European Statistical Training Programme (ESTP)
The European Statistical Training Programme (ESTP)
Chapter 14: Mixed-mode datacollection
The European Statistical Training Programme (ESTP)
Chapter 8: Weighting adjustment
The European Statistical Training Programme (ESTP)
WARM – UP Use LINE 5 of the random digit table. 30. The World Series.
Chapter 1: Basic concepts of surveys
Chapter 10: Selection of auxiliary variables
National needs for AES Purpose - describe participation in learning during a 12 months period. The main parameters are; Participation rates in different.
The European Statistical Training Programme (ESTP)
MATH 2311 Section 6.1.
Chapter: 9: Propensity scores
Chapter 3: Response models
Inference for Sampling
SURVEY RESEARCH (re: Zikmund, Chapter 7).
Keller: Stats for Mgmt & Econ, 7th Ed Data Collection and Sampling
Who are the Subjects? Intro to Sampling
New Techniques and Technologies for Statistics 2017  Estimation of Response Propensities and Indicators of Representative Response Using Population-Level.
Keller: Stats for Mgmt & Econ, 7th Ed Data Collection and Sampling
Sampling and estimation
CONSUMER SURVEY RESEARCH
The European Statistical Training Programme (ESTP)
SURVEY RESEARCH.
What do Samples Tell Us Variability and Bias.
Collecting the Data Tim Vizard, Office for National Statistics.
The European Statistical Training Programme (ESTP)
Collecting time use data
Chapter 6: Measures of representativity
The European Statistical Training Programme (ESTP)
Chapter 13: Item nonresponse
Chapter 2: The nonresponse problem
Adaptive mixed-mode design WP1
Chapter 5: The analysis of nonresponse
Determining Subsampling Rates for Nonrespondents
Stratification, calibration and reducing attrition rate in the Dutch EU-SILC Judit Arends.
Keller: Stats for Mgmt & Econ, 7th Ed Data Collection and Sampling
Presentation transcript:

Chapter 12: Other nonresponse correction techniques Handbook: chapter 10 Call-back Approach (Hansen & Hurvitz) Basic Question Approach Estimating contact probabilities (Politz & Simmons)

Call-back Approach Hansen & Hurvitz (1946) A probability sample is selected from the nonrespondents and then re-approached In a different data collection mode, or By specially trained interviewers, Offering (more) incentives. Under the Fixed Response Model, the Call-back approach is: Draw a simple random sample from the nonresponse stratum IF everybody responds OR if the nonresponse in the call-back is ignorable it is possible to compute unbiased estimates

Call-back approach Select a sample of size m for the call-back. Denote the number of respondents in the call-back approach by mR. Main survey Follow-up survey Response Nonresponse

Call-back Approach Estimation Let denote the mean of the mR values of the responding elements. Then is an unbiased estimator of the mean of the population mean of Y. However, we do not know NR / N and NNR / N so we have to replace them by their unbiased estimates nR / n and nNR / n: Estimator is unbiased in case of complete response, or if nonresponse is MCAR or MAR.

The Call Back approach Example: Labour Force Survey 2005 After 1 week, non-contacts, refusals and unprocessed cases were selected for a follow-up survey. Specially selected interviewers, additional training. Same questionnaire, same mode (face-to-face). Interviewers could offer incentive. Also interviewers could get a bonus. Longer fieldwork period (2 months Response Percentage Main survey 62 % Follow-up survey 43 % Overall 78 %

The Basic Question Approach Method of Basic Questions, Kersten and Bethlehem (1984) Ask non-respondents to answer small number of questions Use key survey questions, so-called basic questions. Basic idea Non-respondents inclined to give quick response Additional response can be used to improve estimates Examples Housing Demand Survey: Do you intend to move within two years? Labour Force Survey: How many people in this household have a paid job? Holiday Survey: Have you been on holiday during the last 12 months? Family Planning Survey: Taking into account your present circumstances, and your expectations of the future, how many children do you think to get from this moment on?

The Basic Question Approach From literature it is known that wording and design of questionnaire may affect answers to questions. Use wording of original questionnaire Select questions from start of questionnaire Test and compare answers Example Dutch Housing Demand Survey 1981 Call-back survey among all respondents, including those who did the Basic Question Approach. Basic question Call-back Intends to move Does not intend to move 8.6 % 5.1 % 12.1 % 74.1 %

The Basic Question approach Issues Amount of additional response Number of basic questions. Questionnaire effects Strategy Different interviewer Different interview mode Timing Multiple sets of basic questions

The Basic Question approach Example, LFS 2005 Non-contacts, refusals and unprocessed cases received basic question approach after one week using telephone, paper and web. Overall response rate: 0.62 + 0.35  (0.60  0.55 + 0.40  0.23) = 0.77 LFS Response R = 65% Nonresponse R = 35% Telephone n = 573 Paper/web n = 280 Response R = 55% Nonresponse R = 45% Response R = 23% Nonresponse R = 77%

The Basic Question approach Example, LFS 2005 CATI if registered fixed-line phone available PAPI/CAWI if registered fixed-line phone not available One member of household (next birthday selection method) Two short questionnaires, CATI and PAPI/CAWI Proxi interviewing Basic questions about employment status (employed, unemployed or non-labour force)

The Basic Question approach Estimation under the Fixed Response Model Differentiate nonresponse stratum with respect to nonresponse types Assume strong relation between basic question variables and other target variables in the survey Assume response to basic question approach to be a random subsample of nonresponse within nonresponse strata Notation under the Fixed Response Model Response stratum with elements Nonresponse strata with elements Target variable Y and basic question variable Z

The Basic Question approach Estimator for basic question variable Z Estimators for target variable Y Use Z as auxiliary variable in the estimation for Y In case Z is qualitative use for instance post-stratification. Say Z has M categories, is the estimated proportion of persons in category m and is the mean of Y for respondents that answered in category m. In case Z is quantitative, use for instance regression estimator

The Basic Question approach Example Proportion of paid jobs in response: 50.2% Proportion of people looking for paid job in response: 12.5% Estimate for proportion of paid jobs: Estimate for proportion of people looking for paid job: Result Proportion of sample Has a paid job Response 65 % 50.2 % Non-contact 10 % 59.7 % Refusal 20 % 38.7 % Unprocessed 5 % 51.0 % Response Looking for a job Not looking for a job Paid job 5 % 95 % No paid job 20 % 80 %

Call-back and Basic Question Approach compared Response rates and response composition R-indicator Maximum bias Root Mean Square Error

Call-back and Basic Question Approach compared LFS 2005 Response rates and response composition – full population Response rates and response composition – registered phones Response Response rate R-indicator Max bias Max RMSE LFS 62% 79% 8,4% 8,5% LFS + BQ 76% 77% 7,6% LFS + CB 85% 4,9% Response Response rate R-indicator Max bias Max RMSE LFS 68% 85% 5,5% LFS + BQ 83% 87% 3,9% 4,0%

Estimating contact probabilities Method of Politz & Simmons (1949) Correct for bias due to not-at-home. Other types of nonresponse are ignored. Useful if relation between not-at-home and target variable. Basic idea Estimate probability of being at home. Apply post-stratification by these probabilities. No bias within strata, because response probabilities are (approximately) the same. Estimation of at-home probability Simply ask (also for previous evenings): Would you mind telling me whether or not you happened to be at home last night at just this time? Or (less reliable): How many nights out of the last five where you at home at just this time?

Estimating contact probabilities Period of L days, including the day of the interview. Let tk be the number of times someone was home at previous L-1 days. Estimate of at-home probability Possible values: 1/L, 2/L, …, L/L Post-stratification by at-home probability requires stratum sizes N1, N2, …, NL to be known. They are unknown, so the have to be estimated. Let nh be the number persons in the response with at-home probability h/L. Estimate for total number of people with at-home probability h/L:

Estimating contact probabilities This techniques is restricted to people with positive at-home probability. Those with probability 0 are excluded. Required: total size of population. This is estimated by: Post-stratification estimator: Complications At-home probabilities in household surveys difficult to estimate Assumption: interviewers make contact at randomly chosen moment. This may be violated in case of appointments or announcement letters.

Estimating contact probabilities Example: 1981 Labour Force Survey. One town, sample of 105 households from 3500 households. Question: On how many of the past 6 days was somebody at home around this time? Results: N’ =4200. Estimates of at-home probabilities are wrong. Number of times someone was at home during the last 6 days Frequency At-home probability 1 1/7 2 2/7 3/7 3 6 4/7 4 9 5/7 5 10 6/7 77 7/7 Total 105