This paper is a summary of analysis conducted of an international database of 31,310 polls from 473 elections and voting events across 40 countries around the world from 1936 through to 2017 compiled by Kantar.
When examined at a global level polls are generally very accurate, the average error of polls conducted within seven days before an election is +/-2.5%.
The aim of this paper, though, is to help you understand why there can be differences in the size of errors seen in different types of election in different countries around the world and to educate on the things you should be looking out for when evaluating any political poll.
Polling companies around the world are judged, fairly or not, by how accurately their polling projections predict the outcome of election results. Polling companies can be easily criticised for not predicting accurately enough the end results of elections and vote, but how do we judge what accurate actually is? The polling industry has been subject to some severe criticism recently for what were seen as mistakes in not predicting the outcome in three recent prominent voting events: the 2015 UK general election, the Brexit referendum and the 2016 US general election. But how big were these "mistakes"? Where they indeed mistakes at all, were they bigger or smaller than you would expect from similar elections?