Keywords: Methodological challenges and improvements, including in the areas of sampling, measurement, survey design and survey response or non-response

Turkey has a heterogeneous ethnic structure, and although most of the groups speak Turkish, official language of the country, each of the groups has its own language such as Kurdish and Arabic. Turks are the majority group in the population of Turkey, approximately 18 percent of the population is Kurds, and 3 percent is the Arabs. The household and women questionnaires that are used as tools of data collection are designed in Turkish only in 2013 Turkey Demographic and Health Survey (TDHS-2013). Therefore, the on-spot translation process that committed by interviewers or interpreters is in use during the data collection process of the demographic surveys, instead of using translated questionnaires into Kurdish and Arabic. In many social surveys, it is well documented that the translation process of the questionnaires into different languages or translation process during fieldwork may produce errors on item-level non-response and accuracy of estimates, mostly produces non-response errors. Within this context, the proposed paper looks at the impact of on-spot translation from questionnaires designed in Turkish to Kurdish and to Arabic on the quality of birth date data collected by the TDHS-2013. We mainly focus on the date type variables, namely date of birth of the respondents and their children which are required to produce demographic indicators, with the aim of understanding the impact of on-spot translation. In this study, we mainly focus on the imputed cases as a reflection of the data quality. Five different groups are created for the analyses on the basis of three variables, the mother tongue of the respondents, the language used during the interview and anyone used for translation. The groups are as follows: Turkish-Turkish (no translation); Kurdish-Kurdish (on-spot translation by interviewer); Kurdish-Turkish (on-spot translation by the interpreter); Arabic-Arabic (on-spot translation by interviewer) and Arabic-Turkish (on-spot translation by interpreter). The preliminary results show that the proportion of imputed cases is just 2.4 percent for the first group, while the proportion increases significantly for the other groups, 9.3 percent, 16.5 percent, 7.0 percent and 18.8 percent respectively. In line with the descriptive analyses logistic regression analyses put forward that apart from the interviews conducted in Turkish-Turkish groups, the data quality lowers significantly when other covariates are not under the control. However, when all other covariates such as age, education, working status and welfare status of women, and survival status, birth cohort, sex and birth order of the children, the data quality problems emerge in only Kurdish-Turkish (on-spot translation by the translator) and Arabic-Turkish (on-spot translation by translator) groups. This actually shows that to decrease item level non-response errors during the on-spot translation should be done by the interviewers who knows the Kurdish and Arabic, instead by the local interpreter. In other words, the composition of the field teams should be designed properly by taking the language ability of interviewers into consideration.