Fascinating write-up Will, thanks for putting it together. Here in Australia, our polling industry is much smaller and we don't have quite the range of questions to measure voter opinion of a govt. We occasionally get right track/wrong track questions and the odd economic performance etc but they tend to be infrequent entries.
The only equivalent question for which we have frequent polling is Prime Minister satisfaction and whether people would prefer the current PM or the current Opposition Leader as Prime Minister (usually referred to here as the "Better PM"/"Preferred PM" score). Interestingly, an Aussie pseph (Kevin Bonham) found that Preferred PM scores show a much weaker correlation with the final result in Australia (https://kevinbonham.blogspot.com/2020/04/why-better-prime-ministerpremier-scores.html) - we're talking R values of 0.5. The other takeaway from that piece is that the voting-intention polling correlated much more strongly with the final result than the Better PM score.
The other question I would have is how well the Best PM scores in your sample correlate with the election result when they're taken this far out (it's approximately 2 years to the next GE, right?). I'd expect the correlation to be a bit weaker but still reasonably predictive. Interestingly I've found that in Australia, Better PM scores are better predictors further out than voting-intention figures (https://armariuminterreta.com/2021/08/23/approval-ratings-and-better-pm/ - hope you don't mind me linking to my own website?)
I wouldn't go that far in that leader ratings are often composite measures of other factors and also suffer from problems of endogeneity to vote choice (in an earlier draft of this post I had a section on endogeneity and partisan reasoning)... That said, they do seem to be less noisy than other predictors - so consumer sentiment (as we saw in 1997 and 2019) isn't always a reliable indicator of election outcomes, but quite often it is. The takeaway message for me is to be cautious and avoid focusing on any single indicator to form predictions.
Fascinating write-up Will, thanks for putting it together. Here in Australia, our polling industry is much smaller and we don't have quite the range of questions to measure voter opinion of a govt. We occasionally get right track/wrong track questions and the odd economic performance etc but they tend to be infrequent entries.
The only equivalent question for which we have frequent polling is Prime Minister satisfaction and whether people would prefer the current PM or the current Opposition Leader as Prime Minister (usually referred to here as the "Better PM"/"Preferred PM" score). Interestingly, an Aussie pseph (Kevin Bonham) found that Preferred PM scores show a much weaker correlation with the final result in Australia (https://kevinbonham.blogspot.com/2020/04/why-better-prime-ministerpremier-scores.html) - we're talking R values of 0.5. The other takeaway from that piece is that the voting-intention polling correlated much more strongly with the final result than the Better PM score.
The other question I would have is how well the Best PM scores in your sample correlate with the election result when they're taken this far out (it's approximately 2 years to the next GE, right?). I'd expect the correlation to be a bit weaker but still reasonably predictive. Interestingly I've found that in Australia, Better PM scores are better predictors further out than voting-intention figures (https://armariuminterreta.com/2021/08/23/approval-ratings-and-better-pm/ - hope you don't mind me linking to my own website?)
Great piece Will - ultimately shows leader ratings are the only one that has any real predictive value?
I wouldn't go that far in that leader ratings are often composite measures of other factors and also suffer from problems of endogeneity to vote choice (in an earlier draft of this post I had a section on endogeneity and partisan reasoning)... That said, they do seem to be less noisy than other predictors - so consumer sentiment (as we saw in 1997 and 2019) isn't always a reliable indicator of election outcomes, but quite often it is. The takeaway message for me is to be cautious and avoid focusing on any single indicator to form predictions.
Great piece thanks
Great piece cheers