We take advantage of the fact that for the Austrian SILC 2008--2011, two data sources are available in parallel for the same households: register-based and survey-based income data. Thus, we aim to explain which households tend to under- or over-report their household income by estimating multinomial logit and OLS models with covariates referring to the interview situation, employment status and socio-demographic household characteristics. Furthermore, we analyze source-specific differences in the distribution of household income and how these differences affect aggregate poverty indicators based on household income. The analysis reveals an increase in the cross-sectional poverty rates for 2008--2011 and the longitudinal poverty rate if register data rather than survey data are used. These changes in the poverty rate are mainly driven by differences in employment income rather than sampling weights and other income components. Regression results show a pattern of mean-reverting errors when comparing household income between the two data sources. Furthermore, differences between data sources for both under-reporting and over-reporting slightly decrease with the number of panel waves in which a household participated. Among the other variables analyzed that are related to the interview situation (mode, proxy, interview month), only the number of proxy interviews was (weakly) positively correlated with the difference between data sources, although this outcome was not robust over different model specifications.
%0 Journal Article
%1 angel2017differences
%A Angel, Stefan
%A Heuberger, Richard
%A Lamei, Nadja
%D 2017
%J Social Indicators Research
%K socdes
%R 10.1007/s11205-017-1672-7
%T Differences Between Household Income from Surveys and Registers and How These Affect the Poverty Headcount: Evidence from the Austrian SILC
%U https://doi.org/10.1007/s11205-017-1672-7
%X We take advantage of the fact that for the Austrian SILC 2008--2011, two data sources are available in parallel for the same households: register-based and survey-based income data. Thus, we aim to explain which households tend to under- or over-report their household income by estimating multinomial logit and OLS models with covariates referring to the interview situation, employment status and socio-demographic household characteristics. Furthermore, we analyze source-specific differences in the distribution of household income and how these differences affect aggregate poverty indicators based on household income. The analysis reveals an increase in the cross-sectional poverty rates for 2008--2011 and the longitudinal poverty rate if register data rather than survey data are used. These changes in the poverty rate are mainly driven by differences in employment income rather than sampling weights and other income components. Regression results show a pattern of mean-reverting errors when comparing household income between the two data sources. Furthermore, differences between data sources for both under-reporting and over-reporting slightly decrease with the number of panel waves in which a household participated. Among the other variables analyzed that are related to the interview situation (mode, proxy, interview month), only the number of proxy interviews was (weakly) positively correlated with the difference between data sources, although this outcome was not robust over different model specifications.
@article{angel2017differences,
abstract = {We take advantage of the fact that for the Austrian SILC 2008--2011, two data sources are available in parallel for the same households: register-based and survey-based income data. Thus, we aim to explain which households tend to under- or over-report their household income by estimating multinomial logit and OLS models with covariates referring to the interview situation, employment status and socio-demographic household characteristics. Furthermore, we analyze source-specific differences in the distribution of household income and how these differences affect aggregate poverty indicators based on household income. The analysis reveals an increase in the cross-sectional poverty rates for 2008--2011 and the longitudinal poverty rate if register data rather than survey data are used. These changes in the poverty rate are mainly driven by differences in employment income rather than sampling weights and other income components. Regression results show a pattern of mean-reverting errors when comparing household income between the two data sources. Furthermore, differences between data sources for both under-reporting and over-reporting slightly decrease with the number of panel waves in which a household participated. Among the other variables analyzed that are related to the interview situation (mode, proxy, interview month), only the number of proxy interviews was (weakly) positively correlated with the difference between data sources, although this outcome was not robust over different model specifications.},
added-at = {2017-11-27T16:48:50.000+0100},
author = {Angel, Stefan and Heuberger, Richard and Lamei, Nadja},
biburl = {https://www.bibsonomy.org/bibtex/26564785b5bc012289f16c54daa391933/dirtyhawk},
day = 12,
doi = {10.1007/s11205-017-1672-7},
interhash = {1b89e097682e70d7f8d135114dbe2dab},
intrahash = {6564785b5bc012289f16c54daa391933},
issn = {1573-0921},
journal = {Social Indicators Research},
keywords = {socdes},
timestamp = {2018-11-01T00:59:05.000+0100},
title = {Differences Between Household Income from Surveys and Registers and How These Affect the Poverty Headcount: Evidence from the {A}ustrian {SILC}},
url = {https://doi.org/10.1007/s11205-017-1672-7},
urldate = {2017-11-27},
year = 2017
}