Scaling level of responses, heaping and censoring in factorial surveys: Expectations and evidence in view of a simple cognitive model

DSpace Repository


Dateien:

URI: http://hdl.handle.net/10900/109298
http://nbn-resolving.de/urn:nbn:de:bsz:21-dspace-1092981
http://dx.doi.org/10.15496/publikation-50675
Dokumentart: Aufsatz
Date: 2020-11-09
Source: Frame paper of the cumulative dissertation: Lang, Volker. 2020. Response behavior in factorial survey experiments: Challenges and innovative solutions. Tübingen: University of Tübingen. Link: https://rds-tue.ibs-bw.de/opac/RDSIndexrecord/1725134705
Language: English
Faculty: 6 Wirtschafts- und Sozialwissenschaftliche Fakultät
Department: Soziologie
DDC Classifikation: 300 - Social sciences, sociology and anthropology
Keywords: Soziologie , Experiment , Erhebung , Skala , Skalenniveau , Kognition , Datenqualität
Other Keywords: Vignettenstudie
Skalierung
vignette design
scaling
sociology
scale level
data quality
cognition
License: Publishing license including print on demand
Order a printed copy: Print-on-Demand
Show full item record

Inhaltszusammenfassung:

Die zugehörige Dissertation ist in der Unibibliothek Tübingen unter folgenden Signaturen bestell- und einsehbar: Us 120. 352.4:1 und Us 120. 352.4:2.

Abstract:

The conceptual literature on factorial survey experiments assumes that ratings are continuous and interval scaled, and that response behavior in factorial surveys can be adequately described by an additive model. Alternatively, I hypothesize that response behavior in factorial surveys is guided by simple cognitive heuristics, and the structures of these heuristics lead to ratings which are not interval scaled and heaped at salient values of response scales. In this frame paper I introduce these two different conceptualizations of response behavior in factorial surveys and summarize findings to assess my hypothesis. In line with my expectations the studies in my dissertation show that non-interval scaled, heaped and censored ratings are common in factorial surveys. My results also show that respondents likely evaluate vignettes in a stepwise manner, and that they start off their evaluations with a focus on salient aspects of the experiments. Furthermore, I find that methods of analysis which do take non-interval scaled ratings and a stepwise evaluation process into account lead to more efficient parameter estimates.

This item appears in the following Collection(s)