| Zusammenfassung: | Surveys, or survey designs still represent the silver bullet in quantitative-oriented social sciences. However, surveys often suffer from two major problems: first, the systematic under-representation of individual population groups (Coverage Error) and second low response rates (Nonresponse Error). Since these two phenomena are usually not coincidental but systematic, data quality often suffers f... Surveys, or survey designs still represent the silver bullet in quantitative-oriented social sciences. However, surveys often suffer from two major problems: first, the systematic under-representation of individual population groups (Coverage Error) and second low response rates (Nonresponse Error). Since these two phenomena are usually not coincidental but systematic, data quality often suffers from their occurrence. Research on overcoming these problems, which often cause systematic biasing of surveys, is still lacking. Hence, the overall research goal of this doctoral thesis is to address these gaps by investigating the possibilities of decreasing the coverage error as well as measuring the nonresponse error. Based on data collected from the PPSM Access Panel in the years 2012 and 2014, I developed a heuristic model, which consists of four parts: (A) Register based sampling and paradata, (B) mixed-mode design, (C) postal advance notification of survey request and (D) incentives. Applying the propensity score method and the random response model, findings show several insights to the measuring of the effect of nonresponse on the given answers in the survey. Regarding the use of various incentives, this study demonstrates that the education bias can be reduced by the use of monetary incentives, which are assigned independently of participation. Surveys carried out via the internet are more likely to be affected by distortions caused by non-coverage, since there is no framework for interviewing the general population over the internet. As a result, the recruitment of respondents often takes place "offline" by telephone or by postal mail. The findings of this dissertation show that pre notification letters have a positive effect on the response rate. The additional information of the address based sampling frame could be used to examine the differences between respondents and nonrespondents via the random response model and possible effects of these differences on the given answers. All members of the sample received a pre notification letter, regardless of having access to the internet or a landline telephone. In connection with a mixed mode design (CAWI, CATI) the negative effect of the coverage error on data quality could be minimized. The best method for the recruitment of members for an access panel, this study concludes, is a pre notification letter in connection with a telephone call and monetary incentives of five euros, which are assigned independently of participation. Finally, besides discussing the resulting empirical implications for research, the findings also provide actionable recommendations for research practice. |
