Refining election prediction models

“Based on past elections and economic factors, two professors at the University
of Oregon predict that Senator Barack Obama will win the presidential election
by a 52 to 48 margin.” Newswise

This blog has bored readers rigid since at least mid 2006 on election prediction models. That was when we became excited about the predictor based on simple household economic indices - the unemployment rate, inflation and interest rates; basically the rise and fall of consumer/houshold confidence over an election cycle.

From my early excitement it soon became obvious that this was a very general, sledge hammer approach; and increasingly difficult to track as governments became increasingly cagey about releasing hard figures. In fact it also became evident there were numerous factors skewing the economic prediction argument. The US presidential election presented big problems for a non-economist, the variables across states under that electoral model are daunting.

Well now a pair of academics from Oregon think they’ve cracked that code. Their paper can be downloaded at: A Disaggregate Approach to Economic Models of Voting in U.S. Presidential Elections. I was interested in the way they broke Breaking the states into five groups of 10, ranging from the states with the lowest average income to those with the highest average income. From there they were able to factor in the additional voting behaviors.

I’m not convince that some being touted are consistently valid. The Bradley effect and the reverse Bradley effect are dependent on social attitudes which are changeable. The tendency for a late surge for the challenger, or decline of the front-runner are not in themselves predictable events.

Public opinion polls and election betting (prediction markets), depending on the country and system, can be good to outstanding, but often rely on too many historic weightings. This is the case in the US now where pollsters must speculate on a range of issues to come up with a formula to predict voter turnout and its likely mix.

Australian federal elections should be about the easiest to predict on the basic economic model, it rarely fail. Aussie states are more problematic, with many side issues, including which party controls the federal government, determining the result. Perhaps this is a reflection of the dominant federal economic control.

The same can be said for Canada, except that opposition conservatives tend to splinter and the dominant centre left is permanently splintered. According to the economic prediction model the non-conservatives in Canada should have won easily, and in fact did, but the largest single party (Conservative) forms a minority government.

I had suspected that emotional factors overrode economic factors in the US presidential election. In many ways, being sold on America’s booming economy it must cause a major guilt trip to be among the losers. In fact, according to the study above, it comes back purely to economic distribution through the country.

Another aspect I’ve noticed, no doubt a product of the economic prediction theory, is that economies tend to deteriorate under conservative governments. Often the progressive parties are left to clean up the mess. This tendency might in fact work backwards, as some researchers believe voters lean towards conservatives when consumer/household confidence is high, and swing to progressives when confidence wavers.

As a progressive I should perhaps be selling my preferred parties to ensure poor economic outcomes, or at least reduced consumer confidence, in key voter catchments. Fortunately our politicians aren’t that bright, or the knowledge could become the basis of entrenched governments.