Errors in social surveys
Errors in social surveys
The central theme of this thesis is that knowledge is needed about all forms of error in the social survey process to enable survey results to be properly assessed and methodological advances to be made. The thesis divides into two sections, one concerned with sampling and the other with non-sampling errors. Most large-scale social surveys employ stratified multi-stage sample designs. Practical methods of sampling error estimation appropriate for such designs are described, and are employed - mainly for percentages - in two national attitude surveys and the multi-purpose General Household Survey. Kish's design effect model for means and percentages is shown to be useful both for imputing sampling errors, and for helping in the planning of future sample designs. The ultimate cluster sample approximation for a multi-stage epsem design is shown to provide simple sampling error estimators and a justification for the design effect model. After illustrating the importance of non-sampling error considerations in survey design by an appraisal of the use of population survey methods in the field of health and morbidity, some specific aspects of non-sampling errors are considered. A theoretical examination of the effect of using a grouped scale to measure a continuous variable (as with opinion items) indicates that the resulting errors are generally small providing about five or more scale points are used. The literature on the sensitivity of responses to opinion questions to the precise wording and format of the questions is reviewed, and the results of some new experiments are reported; appreciable effects occurred in several of the experiments, and in some cases the size of effect varied between different subgroups of the population. The results of an experiment on the variability between coders in making judgemental codings of responses to open-ended questions demonstrate that substantial errors can occur in processing such responses for statistical analysis.
University of Southampton
Kalton, Gordon Graham William
1978
Kalton, Gordon Graham William
Kalton, Gordon Graham William
(1978)
Errors in social surveys.
University of Southampton, Doctoral Thesis.
Record type:
Thesis
(Doctoral)
Abstract
The central theme of this thesis is that knowledge is needed about all forms of error in the social survey process to enable survey results to be properly assessed and methodological advances to be made. The thesis divides into two sections, one concerned with sampling and the other with non-sampling errors. Most large-scale social surveys employ stratified multi-stage sample designs. Practical methods of sampling error estimation appropriate for such designs are described, and are employed - mainly for percentages - in two national attitude surveys and the multi-purpose General Household Survey. Kish's design effect model for means and percentages is shown to be useful both for imputing sampling errors, and for helping in the planning of future sample designs. The ultimate cluster sample approximation for a multi-stage epsem design is shown to provide simple sampling error estimators and a justification for the design effect model. After illustrating the importance of non-sampling error considerations in survey design by an appraisal of the use of population survey methods in the field of health and morbidity, some specific aspects of non-sampling errors are considered. A theoretical examination of the effect of using a grouped scale to measure a continuous variable (as with opinion items) indicates that the resulting errors are generally small providing about five or more scale points are used. The literature on the sensitivity of responses to opinion questions to the precise wording and format of the questions is reviewed, and the results of some new experiments are reported; appreciable effects occurred in several of the experiments, and in some cases the size of effect varied between different subgroups of the population. The results of an experiment on the variability between coders in making judgemental codings of responses to open-ended questions demonstrate that substantial errors can occur in processing such responses for statistical analysis.
This record has no associated files available for download.
More information
Published date: 1978
Identifiers
Local EPrints ID: 458512
URI: http://eprints.soton.ac.uk/id/eprint/458512
PURE UUID: 32e45b82-1b77-40a3-b324-6dec343e2f6f
Catalogue record
Date deposited: 04 Jul 2022 16:50
Last modified: 04 Jul 2022 16:50
Export record
Contributors
Author:
Gordon Graham William Kalton
Download statistics
Downloads from ePrints over the past year. Other digital versions may also be available to download e.g. from the publisher's website.
View more statistics