Impact of estimation techniques on regression analysis: an application to survey data on child nutritional status in five African countries
Impact of estimation techniques on regression analysis: an application to survey data on child nutritional status in five African countries
This paper illustrates the impact of ignoring survey design and hierarchical structure of survey data when fitting regression models. Data on child nutritional status from Ghana, Malawi, Tanzania, Zambia, and Zimbabwe are analysed using four techniques: ordinary least squares; weighted regression using standard statistical software; regression using specialist software that accounts for the survey design; and multilevel modelling. The impact of ignoring survey design on logistic and linear regression models is examined. The results show bias in estimates averaging between five and 17 per cent in linear models and between five and 22 per cent in logistic regression models. The standard errors are also under-estimated by up to 49 per cent in some countries. Socio-economic variables and service utilisation variables are poorly estimated when the survey design is ignored.
Southampton Statistical Sciences Research Institute, University of Southampton
Madise, Nyovani
2ea2fbcc-50da-4696-a0a5-2fe01db63d8c
Stephenson, Rob
79610976-1582-4cd2-bb9d-3966ebbc9ed8
Holmes, David
acb9dc00-6021-4eee-8219-2c5032d62ce7
Matthews, Zoë
ebaee878-8cb8-415f-8aa1-3af2c3856f55
2003
Madise, Nyovani
2ea2fbcc-50da-4696-a0a5-2fe01db63d8c
Stephenson, Rob
79610976-1582-4cd2-bb9d-3966ebbc9ed8
Holmes, David
acb9dc00-6021-4eee-8219-2c5032d62ce7
Matthews, Zoë
ebaee878-8cb8-415f-8aa1-3af2c3856f55
Madise, Nyovani, Stephenson, Rob, Holmes, David and Matthews, Zoë
(2003)
Impact of estimation techniques on regression analysis: an application to survey data on child nutritional status in five African countries
(S3RI Applications and Policy Working Papers, A03/07)
Southampton, UK.
Southampton Statistical Sciences Research Institute, University of Southampton
26pp.
Record type:
Monograph
(Working Paper)
Abstract
This paper illustrates the impact of ignoring survey design and hierarchical structure of survey data when fitting regression models. Data on child nutritional status from Ghana, Malawi, Tanzania, Zambia, and Zimbabwe are analysed using four techniques: ordinary least squares; weighted regression using standard statistical software; regression using specialist software that accounts for the survey design; and multilevel modelling. The impact of ignoring survey design on logistic and linear regression models is examined. The results show bias in estimates averaging between five and 17 per cent in linear models and between five and 22 per cent in logistic regression models. The standard errors are also under-estimated by up to 49 per cent in some countries. Socio-economic variables and service utilisation variables are poorly estimated when the survey design is ignored.
More information
Published date: 2003
Identifiers
Local EPrints ID: 8142
URI: http://eprints.soton.ac.uk/id/eprint/8142
PURE UUID: e93cc41c-562e-4408-ab3a-0ab9d1754c16
Catalogue record
Date deposited: 11 Jul 2004
Last modified: 16 Mar 2024 02:47
Export record
Contributors
Author:
Nyovani Madise
Author:
Rob Stephenson
Author:
David Holmes
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