Bias and tracking accuracy in voting projections using the New Zealand attitudes and values study

Abstract

ABSTRACTWe introduce a method for assessing the overall level of bias and relative tracking accuracy in time series analysis of polling data on voter intentions by comparing two data sets: the New Zealand Attitudes and Values Study (NZAVS; N = 13,936) and the One News Colmar Brunton election polls (N = 10,210). The NZAVS is a national probability longitudinal postal panel survey started in 2009. It was set up to track stability and change in values, without any intention to predict election results. The One News Colmar Brunton Poll is a nationally representative random-digit dial telephone survey conducted by Colmar Brunton specifically to measure voter sentiment. Our comparison focused on responses to both surveys from February 2014 to September 2014, in the run up to the New Zealand general election. The NZAVS and One News Colmar Brunton Poll detected nearly identical changes in party vote over time, and both effectively predicted the outcome of the 2014 General Election. This shows the unexpected power of national-scale probability postal studies to track change in democratic decisions that affect all members of a society. These results further support the validity of the NZAVS as an independent and impartial source that might be ‘repurposed’ to create a continuous monitor tracking political attitudes and detecting citizens’ reaction to, and relative change in, opinions toward different social policies, values, and attitudes, over time.