Item-Based Non-Response in Pre-Election Surveys: A Case Study of Georgia

This paper examines item-based non-response in pre-election surveys and exit polls conducted in Georgia over the past decade, highlighting a widespread issue prevalent in many countries around the world. High item-based non-response rates, particularly regarding questions about party voting, pose significant challenges to understanding voter behavior. To address this, GORBI has implemented a range of methodological innovations aimed at enhancing response rates. These strategies include shifting from traditional verbal prompts and show cards to more interactive formats, such as tablet-based self-reporting and the use of images of actual ballots.

Despite these efforts, a substantial portion of respondents continue to abstain from answering the voting question. In response to this ongoing challenge, GORBI has developed sophisticated voting allocation models to estimate and address item-based non-response. This paper will explore these models in detail, evaluate their effectiveness in mitigating non-response bias, and compare their outcomes with actual election results in Georgia. Through this analysis, we aim to provide deeper insights into the dynamics of voter response and the broader methodological challenges faced in electoral research globally.

Most of the studies in question were conducted using CAPI methodology, except for a portion conducted using a probability based CATI Panel during Covid-19 period. The reason for this is that CAWI and CATI RDD methodologies are problematic due to insufficient (and due to mountainous terrain – uneven) penetration rates of internet and telecommunications in the country. Therefore, the paper concentrates on tackling the questions posed using experience with available modes.