Recent Developments in Small Area Estimation and Non-probability Sampling Techniques

Sanjay Chaudhuri Chair
University of Nebraska-Lincoln
 
Sanjay Chaudhuri Organizer
University of Nebraska-Lincoln
 
Thursday, Aug 7: 10:30 AM - 12:20 PM
0835 
Topic-Contributed Paper Session 
Music City Center 
Room: CC-106C 

Applied

Yes

Main Sponsor

Survey Research Methods Section

Co Sponsors

Government Statistics Section
Social Statistics Section

Presentations

Estimation from Combined Probability and Non-probability Samples under Uncertain Sampling Overlap

Under a quasi-randomization approach to inferences from a non-probability sample, the unknown sample participation probabilities are estimated by combining the non-probability set with a reference probability sample. There could be a substantial overlap between the two sets; however, the identity of units belonging to both the non-probability and reference samples are usually unknown to the analyst. If records are not linked, approximate methods (such as, record linkage) could be applied. We study feasibility of using statistically linked sets under several scenarios of an uncertain non-probability and probability samples overlap. 

Keywords

data combining

non-probability sample

participation probabilities

reference sample 

Speaker

Julie Gershunskaya, US Bureau of Labor Statistics

Privacy and fidelity trade-off in small area estimation

We consider the problem of area-level small area modeling, and consider the efficiency and accuracy of small area predictions when privacy guarantees are desired. We consider a Bayesian framework for this problem. Here, an adversary has some information about the data and has prior opinions about the rest of the data. The parameter estimates and the small area predictions are rendered confidential in such a way such that a balance is achieved between privacy guarantees and quality of the inferences.  

Speaker

Snigdhansu Chatterjee, University of Maryland, Baltimore County

Quantile Processes and their Applications in Finite Populations

In this talk, I shall discuss the weak convergence of the quantile processes, which are constructed based on different estimators of the finite population quantiles under various well-known sampling designs. The results related to the weak convergence of these quantile processes are applied to find asymptotic distributions of the smooth L-estimators and the estimators of smooth functions of finite population quantiles. Based on these asymptotic distributions, confidence intervals are constructed for several finite population parameters like the median, the α-trimmed means, the interquartile range and the quantile based measure of skewness.  

Keywords

High entropy sampling designs

Ratio estimator

Regression estimator

Stratified multistage cluster sampling designs

Skorohod metric

Hadamard differentiability 

Co-Author

Probal Chaudhuri, Indian Statistical Institute

Speaker

Anurag Dey, University of Maryland, College Park

RESPONSE MODEL SELECTION IN SMALL AREA ESTIMATION IN CASE OF NOT MISSING AT RANDOM NONRESPONSE AND ONLY TOTALS AVAILABLE FOR AUXILIARY VARIABLES

Sverchkov and Pfeffermann (S-P, 2023) suggested Information Criteria, similar to AIC and BIC for response model selection for Small Area Estimation under informative probability sampling of areas and within the sampled areas and not missing at random (NMAR) nonresponse. The approach considered by S-P requires knowledge of auxiliary data for complete sample before response occurs. In this paper we generalize this approach to the case where auxiliaries are not known for the non-respondents, but their totals are known.
REFERENCES
Sverchkov and Pfeffermann (2023), Response Model Selection in Small Area Estimation Under not Missing at Random Nonresponse. Calcutta Statistical Association Bulletin, pp. 1 -11
 

Keywords

information criteria

likelihood ratio tests

missing information principle

non-probability sampling 

Speaker

Michael Sverchkov, US Bureau of Labor Statistics

PresentationWW

Speaker

Takumi Saegusa, University of Maryland