Tuesday 16th July
Wednesday 17th July
Thursday 18th July
Friday 19th July
Download the conference book
Investigating Non Respondents: How to Get Reliable Data and How to Use Them |
|
| Chair | Dr Michele Ernst Staehli (FORS) |
Immigrants often show particularly low response rates. Consequently, they may be under-represented and object to response bias. Since many surveys lack comprehensive information on non-participants, detailed analyses of non-response are still scarce. This holds particularly true for non-response of immigrants.
Here, data provided by the National Educational Panel Study (NEPS) in Germany offer a unique chance for more direct analyses: Among others, NEPS provides a school-based sample of German 9th-graders. In addition to PAPI assessments of students, teachers and headmasters parents were interviewed using CATI. This allows me to analyze parental non-response using data gathered from students and schools as contextual information. Thereby, I can identify more precisely what characterizes non-participating parents.
To this end, I apply logistic random intercept models to estimate probabilities of contact, of non-cooperation and of total non-response for immigrant parents from Turkey, the former Soviet Union and Eastern and Southern Europe. With about 39 percent the response rate of immigrant parents is 15 percentage points lower than for native parents. Furthermore, analyses reveal huge differences between immigrant groups. Additional information like seniority of interviewers, students' language skills, parental socio-economic status and ethnic school composition can explain initial differences widely. However, even controlling for all these aspects net effects of "country-of-origin" remain.
Results may help to tailor future studies to better meet the particular requirements of immigrants and thus to reduce response bias. Therefore, I discuss potential implications for fieldwork and sampling procedures.
With the move to address-based sampling, survey firms can now purchase data from commercial marketing companies to supplement information about sampled households. Because such ancillary data can provide information about non-respondents as well as respondents, these data could aid nonresponse adjustment, targeted sampling, and sampling frame corrections. For these potentials to be realized, it is important to document both accuracy and missingness in ancillary data and to assess the use of ancillary data as a corrective tool. The current study uses a unique dataset linking survey self-reports derived from an address-based sample with demographic ancillary data collected on all sampled individuals (not just respondents). Exploring correspondence and discrepancies between the data sources, this study identifies patterns of systematic inaccuracy and nonignorable missingness threatening conclusions derived using consumer file data. Multiple imputation is also used to generate demographic information for all individuals in the sampling frame based on the ancillary data. Comparing imputations with benchmark estimates reveals that matching strategies only partially address biases due to unit non-response. Imputed demographics varied in their accuracy, indicating that the imputation strategy was insufficient. Overall the results suggest important limitations in our ability to use consumer file ancillary data to improve population inference.
Additional Author: Charles DiSogra (Abt SRBI)
Please note that the presenting author cannot make July 15 or the morning of the 16th due to flights
All panel surveys have to deal with problems such as unit non-response and item non-response. Researchers address this problem using ex-post approaches (e.g. imputation schemes or weighting), but also using ex-ante strategies, analyzing the determinants of survey response for reducing non-response during fieldwork (i.e. applying fieldwork monitoring or responsive designs).
This paper adopts the ex-ante approach and proposes an analysis of survey participation with particular focus on item non-response and paradata for the longitudinal sample of the Survey of Health, Ageing and Retirement in Europe (SHARE). We borrow from Nicoletti and Peracchi (2005) to model contact and cooperation as sequential events. Therefore, conditional on eligibility, the response process is completely described by two elements: the probability of contact between the interviewer and the interviewee and the probability of co-operation given contact (i.e. interviewee completes the interview).
Compared to other studies, we consider a broader indicator of item non-response, i.e. the percentage of missing information by questionnaire section that could be informative for operational issues. Moreover we exploit a larger set of variables to integrate the analysis with health information and paradata (e.g. interviewer characteristics) and assess their role in predicting survey response.
Reference: Nicoletti C. and Peracchi F. (2005) "Survey response and survey characteristics: microlevel evidence from the European Community Household Panel," Journal of the Royal Statistical Society Series A, Royal Statistical Society, vol. 168(4), pp.763-781.