Modular Questionnaire Designs for Social Surveys: Statistical Modelling of Designed Missingness

Research question/goal: 

Surveys have become an indispensable source of information on social and political circumstances in modern societies. Quantitative social research based on survey data requires ever larger datasets containing ever more complex structures. Together with decreasing response rates and increasing fieldwork efforts, the heightened expectations regarding data quality lead to surging survey costs.

Fortunately, the developments in statistical modelling and associated computing power have seen large developments in the past twenty years, enabling us to rethink traditional survey data collection methods. In particular, two developments seem promising: modular (or split) questionnaires and imputation methods.

The project aims to investigate whether these methods can be combined and further developed to replace large-scale face-to-face surveys by shorter online surveys while preserving the same degree of population coverage and quality. This project is a first step in developing and evaluating the necessary statistical tools to complement data structures collected by modular questionnaire designs. The main interest lies in assessing the estimation efficiency and bias of imputation methods. Further considerations concern the potential for cost savings and usability.

In the first phase of the project, datasets of the waves of the German Internet Panel are used to evaluate the approaches. In the second phase, we will analyse and impute datasets from modular questionnaire designs, implemented in the European Value Survey. Resulting datasets are imputed and analysed regarding the aim of the project.

Current stage: 

A research grant was submitted to the DFG and approved. Now, we are focusing on the recruitment of the corresponding project staff. A major part of the project is the development and application of appropriate imputation methods to compensate for missing values generated by the modular questionnaire design. To this end, preliminary imputation routines were applied to the datasets of the European Value Survey and were evaluated with respect to computational times.

Fact sheet

2017 to 2021
in preparation
Data Sources: 
German Internet Panel, European Value Study
Geographic Space: