PhD graduand digs deeper into survey error

03 June 2013

Reza DanielsA chapter in graduand Reza Daniels' PhD thesis has won him second prize in the Cochran-Hansen competition for best paper on survey research methods submitted by a young statistician from a developing or transition country.

The chapter, Questionnaire design and response propensities for employee income microdata, is part of his PhD thesis on income distribution with multiple sources of survey error. This concerns the sources of error that can affect income data arising out of the survey process.

Daniels is a lecturer in the School of Economics and a research associate in the Southern Africa Labour & Developmental Research Unit.

The Cochran-Hansen Prize is awarded to eligible individuals under the age of 40 every two years by the International Association of Survey Statistics, a division of the International Statistical Institute.

Household income and expenditure surveys provide government and organisations with a powerful tool to monitor development and inform policymaking. However, all surveys contain sources of error that can impact upon the reliable analysis of collected data. Daniels' work attempts to provide a framework through which researchers can better understand which sources of error are most significant and how to deal with these sources of error.

"The design of household income questions in household surveys usually includes response options for actual income ranges, but also has 'Don't Know' and 'Refuse' options. The data of those who tick either of these options can be imputed if sufficient information about the characteristics of these respondents is available in the survey," said Daniels.

Using sequential response models, a type of statistical model, Daniels analysed these responses to the employee income question in Statistics South Africa's October Household Surveys between 1997 and 1999 and Labour Force Surveys from 2000 to 2003.

"An analysis of this sort sheds light on the underlying response process," Daniels explained in his abstract. This is useful as an exercise to see if those who do not respond to the income question are systematically different to those who do respond.

Daniels said that the presentation of a second follow-up income question that asked for a bracketed income range after the initial exact income question was not answered helped to overturn respondents' refusals – and that these respondents tended to be higher-income earners.

"In the sequential response models, initial non-response was therefore clearly correlated with predictors of income, but after the presentation of the bracket showcards, this correlate of income effect was removed. This suggested that final non-response was no longer a function of income," added Daniels.

This finding differs from earlier international research on this theme and has important implications for survey organisations and researchers interested in imputing plausible information for those who state that they don't know their income or refuse to provide an answer.

Daniels graduated from geography to developmental studies (his BSc and MA degrees in the fields are from the University of Auckland, New Zealand), and then survey methodology (University of Michigan, Ann Arbor, USA) to economics at UCT.

Creative Commons License This work is licensed under a Creative Commons Attribution-NoDerivatives 4.0 International License.

Please view the republishing articles page for more information.