Response Rates Trends in Household Surveys in Turkey: An Assessment through DHS and AAPOR Approaches


Saraç M., Adalı T.

AAPOR 74th Annual Conference, Toronto, Canada, 16 - 19 May 2019

  • Publication Type: Conference Paper / Unpublished
  • City: Toronto
  • Country: Canada
  • Hacettepe University Affiliated: Yes

Abstract

Achieving high response rates, conversely reducing nonresponse rates, has always been one of the main interest of surveys all around the world in terms of data quality. Although major efforts are spent by survey organizations in order to reduce nonresponse, the trend is inevitably an increasing one. However, survey literature tends to ignore the topic for countries where high response rates are still possible, like Turkey; where household response rates for face to face surveys can be over 80%. Moreover, we acknowledge that the calculation of the response rate directly depends on how it is defined, and the classification of interview result codes.

Nationwide demographic surveys have been conducted in Turkey since 1968 in a quinquennial manner; under the Demographic and Health Surveys Program since 1993. These surveys are large scale surveys that are conducted face-to-face with households. DHS response rate definitions were adopted when presenting Turkey DHS response rates for the survey years between 1993 and 2013. In this study, response rates based on DHS and AAPOR approaches have been computed and compared for all TDHS surveys. Furthermore, Turkey DHS nonresponse trend will be assessed according to other available factors that contribute non-response (re-visiting, survey time, training and etc.) under the following analyses. All the analyses have been conducted based on residential and regional differences in Turkey.

The study findings suggest a decreasing trend of household response rates in Turkey based on the DHS definition; yet no trend is observed when using the AAPOR definition. We argue that the choice of response rate definition is dependent on the result codes available; less detailed result codes lead to stronger assumptions. Moreover, we have also seen that it is very difficult to assess non-response trend over time when aims have been adopted to keep response rates over a certain level.