E-Thesis 387 views 180 downloads
Towards a New Ontology of Polling Inaccuracy: The Benefits of Conceiving of Elections as Heterogeneous Phenomena for the Study of Pre-election Polling Error / JACK TUDOR
Swansea University Author: JACK TUDOR
-
PDF | E-Thesis
Copyright: The Author, Jack Tudor, 2023. Distributed under the terms of a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License (CC BY-NC-ND 4.0).
Download (5.6MB)
DOI (Published version): 10.23889/SUthesis.64642
Abstract
A puzzle exists at the heart of pre-election polling. Despite continual methodological improvement and repeated attempts to identify and correct issues laid bare by misprediction, average polling accuracy has not notably improved since the conclusion of the Second World War. In this thesis, I conten...
| Published: |
Swansea, Wales, UK
2023
|
|---|---|
| Institution: | Swansea University |
| Degree level: | Doctoral |
| Degree name: | Ph.D |
| Supervisor: | Wall, Matthew and Fahey, Kevin |
| URI: | https://cronfa.swan.ac.uk/Record/cronfa64642 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Abstract: |
A puzzle exists at the heart of pre-election polling. Despite continual methodological improvement and repeated attempts to identify and correct issues laid bare by misprediction, average polling accuracy has not notably improved since the conclusion of the Second World War. In this thesis, I contend that this is the result of a poll-level focus within the study of polling error that is both incommensurate with its evolution over time and the nature of the elections that polls seek to predict. I hold that differences between elections stand as a plausible source of polling error and situate them within a novel four-level model of sources of polling error. By establishing the heterogenous nature of elections as phenomena and its expected impact on polling error, I propose a new election-level ontology through which the inaccuracy of polls can be understood. I test the empirical validity of this new ontology by using a novel multi-level model to analyse error across the most expansive polling dataset assembled to date, encompassing 11,832 in-campaign polls conducted in 497 elections across 83 countries, finding that membership within different elections meaningfully impacts polling error variation. With the empirical validity of my proposed ontology established, I engage in an exploratory analysis of its benefits, finding electoral characteristics to be useful in the prediction of polling error. Ultimately, I conclude that the adoption of a new, multi-level ontology of polling error centred on the importance of electoral heterogeneity not only offers a more comprehensive theoretical account of its sources than current understandings, but is also more specifically tailored to the reality of pre-election polling than existing alternatives. I also contend that it offers pronounced practical benefits, illuminating those circumstances in which polling error is likely to vary. |
|---|---|
| Keywords: |
Polling error, elections, multi-level modelling |
| College: |
Faculty of Humanities and Social Sciences |