New working paper with Julian Schuessler on Graphical Causal Models for Survey Inference
New working paper with Julian Schuessler on Graphical Causal Models for Survey Inference out on SocArXiv!
We demonstrate the usefulness of graphical causal models to communicate theoretical assumptions about the collection of survey data, to determine whether typical population parameters of interest to survey researchers can be recovered from a survey sample, and to support the choice of suitable adjustment strategies. Starting from graphical representations of prototypical selection scenarios, we provide an explicit justification for the use of standard weighted regression estimators, which is missing in the literature. We then introduce multiple selection nodes to represent the various stages of the survey data collection process in line with the Total Survey Error approach as the dominant conceptual foundation for studying survey errors. Finally, we discuss areas for future survey methodological research that can benefit from recent advances in the graph literature in computer science and epidemiology.