All time references are in CEST
Methods for questionnaire development: SQP, Cognitive Interviewing, and others 1
|Session Organisers|| Dr Cornelia Neuert (GESIS - Leibniz Institute for the Social Sciences)
Dr Lydia Repke (GESIS - Leibniz Institute for the Social Sciences)
|Time||Thursday 20 July, 14:00 - 15:30|
There are various methods and procedures available for evaluating survey items during questionnaire development. They all share the goal to help researchers to identify questions that are flawed, difficult to comprehend, and might lead to measurement error. However, the available methods differ in several aspects, such as timing, the outcome, the amount of data collection needed, and whether respondents, experts, or computer systems are involved.
In practice, considerations like the available time and budget, the target population, and the mode of data collection affect the choice.
As there is still no clear consensus about best practices for question evaluation, this session aims to discuss current questionnaire evaluation practices using different methods and tools. We invite papers that use different types of evaluation methods, such as expert reviews and cognitive interviewing, or the Survey Quality Predictor (SQP; https://sqp.gesis.org).
We invite papers that
(1) exemplify which evaluating method might best be used during questionnaire development;
(2) assess the different evaluation methods with respect to their advantages and disadvantages;
(3) provide information on how methods can best be used in combination.
Keywords: questionnaire development, question evaluation methods, pretesting, cognitive interviewing, SQP, expert reviews, web probing, behavior coding
Dr Paul Scanlon (National Center for Health Statistics) - Presenting Author
Complex social and psychological concepts, such as emotional well-being (EWB) and discrimination, are oftentimes measured using lengthy, multi-construct scales and question sets that have been designed and refined using psychometric approaches. Particularly in health and wellness-related domains, these scales are developed primarily for administration in clinical settings, and not with national, omnibus, household surveys in mind. In the United States, the National Center for Health Statistics and its partners are working to ensure that the coverage of concepts, including EWB and discrimination, is expanded on the federal government’s national health surveys.
However, transitioning measures designed primarily for a clinical or small-scale settings to that of a household survey requires additional question evaluation—one cannot just assume that the psychometric or measurement properties of a scale will be same across these environments. This presentation will describe the mixed- and multi-method question evaluation approach used to accomplish this transition in the cases of the Everyday Discrimination Scale (EDS) and a large set of items used to measure EWB. In both examples: 1) iterative cognitive interviews and open-ended web probes were used to understand what constructs each individual survey item captured, as well as respondents’ overall conceptualizations of the question sets; 2) close-ended web probing was used to confirm these findings and to explore differences in interpretation across population subgroups; 3) experimental designs were used to compare and contrast formatting styles that are more or less appropriate for large-scale survey interviews; and 4) confirmatory factor analyses were performed so that any changes in the psychometric properties could be evaluated. The strengths and weaknesses of each individual method will be reviewed, and the presentation will conclude with a discussion of how this mixed-method, multi-phase approach performed and considerations for best practices going forward.
Ms Charlotte Hunsicker (MEA-SHARE and SHARE Berlin Institute) - Presenting Author
Dr Arne Bethmann (MEA-SHARE and SHARE Berlin Institute)
Professor Christina Buschle (IU International University of Applied Sciences)
Dr Herwig Reiter (German Youth Institute (DJI))
The pretesting of draft questionnaires is usually outsourced to survey agencies and supervised by research institutes. The paper introduces the rationale behind the alternative of establishing an in-house team of embedded interviewers for doing Qualitative Pretest Interviews (QPI) and illustrates the first steps of its establishment in the frame of the Survey of Health, Aging and Retirement in Europe (SHARE). SHARE is a multi-disciplinary and cross-national panel research infrastructure for studying the effects of health, social, economic and environmental policies over the life-course in 28 European countries and Israel.
The QPI is a novel method of pretest interviewing for improving the quality of standardised research. By advocating a comprehensive consideration of qualitative-interpretive methodology, the QPI transfers the idea of intersubjective understanding in everyday communication to the situation of pretest interviewing. Following the requirements of qualitative research the QPI relies heavily on the familiarity of researchers and interviewers with all aspects of the research topic as well as the purpose of single items and item batteries. Therefore, the establishment and training of an in-house team for pretest-interviewing combines the advantages of substantive expertise with the possibility of immediate and stepwise revision of standardised instruments. This alternative can improve, accelerate and focus the process of pretesting. The paper concludes with a discussion of advantages an limits of this alternative.
Dr Marina Aleksandrova (NRU HSE) - Presenting Author
A variety of methods aimed at assessing and improving the quality of questionnaires continue to be developed by sociologists. One of the promising areas for improving questionnaires is to work on reducing the frequency of item nonresponse in surveys. While most studies are focused on dealing with item nonresponse error (INE) when survey data has already been collected, almost no attention was given to investigate how this type of error can be predicted prior to data collection process. Different characteristics, which can influence respondent's willingness to answer questions in surveys, were studied in scientific literature, and among them, two approaches can be distinguished: research of characteristics related to respondents, and research of characteristics related to the questionnaire. These two approaches were never investigated together, despite existing assumptions that respondents with different sociodemographic characteristics respond differently to different wordings of survey questions. In this paper, we investigate how gender, age and financial situation as sociodemographic characteristics of respondents, and characteristics of questionnaires (question wording) simultaneously affect the occurrence of INE in the sociological data. We achieve the research goal using natural language processing (NLP) and machine learning methods. We develop a methodology to organize and preprocess the data with survey questions, and sociodemographic characteristics of respondents. Three machine learning algorithms were compared (Binary Logistic Regression, Gradient Boosting, Random Forest) to choose best models based on the quality metrics and interpret results. The database of the study is ESS (European Social Survey), which were conducted in Great Britain in English language, which was used to train models of prediction the INE.
Results of the conducted research include description of specifics of response behavior of respondents depending on question wording and sociodemographic characteristics, and the model of prediction of probability of item nonresponse occurrence depending on the formulation of question.