MISSPECIFIED WEIGHTS IN WEIGHT-SMOOTHING METHODS

Simple item page
dc.contributor.advisorSlud, Eric Ven_US
dc.contributor.authorLi, Xiaen_US
dc.contributor.departmentApplied Mathematics and Scientific Computationen_US
dc.contributor.publisherDigital Repository at the University of Marylanden_US
dc.contributor.publisherUniversity of Maryland (College Park, Md.)en_US
dc.date.accessioned2018-07-17T05:49:19Z
dc.date.available2018-07-17T05:49:19Z
dc.date.issued2018en_US
dc.description.abstractMisspecification happens for various reasons in weight adjustment procedures in survey data analysis. To study the consequences of weight misspecifications, we study the effects of using a multiplicative biasing factor to describe the weight adjustments and reflect the distributional change from design/initial weights to final weights. The necessary and sufficient condition of the Horvitz-Thompson (HT) estimator of a population total being consistent is then given in a superpopulation setting. When HT is consistent, we first investigate the bias in other estimators for population totals. We show the necessary condition for bias in Generalized Regression (GREG) estimator and the resulting bias formula in the superpopulation limiting sense. We also link the bias in a model-based estimator of Zheng and Little to the failure of extrapolated model-fitting outside the sample. Both findings are validated in simulation studies. Next we find that the biasing factor affects estimators so that one particular estimator may have the smallest variance under design weights but not under misspecified weights due to variance inflation. A preliminary analysis on simulated samples drawn from a population of real American Community Survey (ACS) data illustrates the quality of fit of the biasing factor model we proposed to the ACS data with weights modified by a few calibration/raking steps.en_US
dc.identifierhttps://doi.org/10.13016/M2DB7VT2J
dc.identifier.urihttp://hdl.handle.net/1903/20839
dc.language.isoenen_US
dc.subject.pqcontrolledStatisticsen_US
dc.subject.pqcontrolledApplied mathematicsen_US
dc.subject.pquncontrolledAnalysis of survey dataen_US
dc.subject.pquncontrolledHorvitz-Thompson estimatoren_US
dc.subject.pquncontrolledMisspecified weightsen_US
dc.subject.pquncontrolledSuperpopulationen_US
dc.subject.pquncontrolledSurvey methodologyen_US
dc.subject.pquncontrolledWeight-smoothingen_US
dc.titleMISSPECIFIED WEIGHTS IN WEIGHT-SMOOTHING METHODSen_US
dc.typeDissertationen_US

Files

Original bundle
Now showing 1 - 1 of 1
Thumbnail Image
Name:
Li_umd_0117E_18809.pdf
Size:
1.08 MB
Format:
Adobe Portable Document Format
Download

Collections

UMD Theses and Dissertations
Computer Science Theses and Dissertations
Mathematics Theses and Dissertations