A lot of people were shocked to discover that the Wisconsin polls were so wrong. While some polls correctly had Walker winning by 7 points, others had the race a dead heat. The challenge was with the assumptions used by the people taking the poll. For one thing, there is a difference between surveying likely voters and surveying registered voters. Republicans are more likely to vote than Democrats. But the problem is more than just this one issue. Pollsters cannot interview everyone, only a random sample of people. Let’s assume that a poll surveys 600 registered voters. This is not necessarily a truly random sample, so they adjust the results based on the demographics they expect. Let me explain that. Let’s assume that a pollster believes that 40% of people are Democrats, 40% are Republicans and 20% are independents. They would expect the survey to draw 240 Democrats and 240 Republicans. If they have too many Democrats or too many Republicans they will adjust the results. To keep this short, I will assume they do not adjust because of the percentage of independents.
Most of the assumptions with regard to political party are based on the 2008 Presidential election. But analysts know that things have changed dramatically since 2008. Remember the 2010 election? There was a significant movement from Democrat to Republican or Independent. No one really knows the correct percentage of people registered in either party.
In many cases the pollsters were anticipating a huge discrepancy between Democrats and Republicans. I have seen reports that this could be 9 or even 10 points. That really impacts the results. Let’s go back to that 600 voter sample. Let assume the pollster expected 45% Democrat and 35% Republican, because of the 2008 election results. That means they expected 270 Democrats and 210 Republicans. If they interviewed 600 voters and they got 240 Democrats and 240 Republicans, they would make an adjustment. Assume that the survey showed a tie between Mitt Romney and Barrack Obama. The pollster would assume that Democrats were under represented by 30 voters and Republicans were over represented by 30 voters. So the results of the poll would be changed from 50% – 50% to 55% – 45% in favor of Obama. The actual way results are adjusted varies between pollsters, but I think this demonstrates my point.
This is not necessarily a devious plot to make Obama look better. Pollsters may strongly believe the results they are reporting. This is just the predictable result of using flawed assumptions. It is exactly this type of reasoning that resulted in the prediction of a 50 50 tie in Wisconsin. When the votes were counted, Walker won 53 to 47, a bigger margin than he won by in 2010.
This is one reason the Wisconsin election really matters. In addition to the obvious momentum for Republicans and the public rebuke of unions, it is going to force pollsters to re-think their basic assumptions. Regardless of which party they favor, no pollster wants to be that wrong. They get paid for getting it right. Watch for polls in the new few days and weeks to start showing a significant difference in the surveys. They are likely to show major improvement for Mitt Romney. Of course Democrats will respond by saying the polls are flawed, but they will be paying close attention.
The reality is that while Obama has had a couple of really rotten weeks and his polls numbers will probably suffer as a result, the real change will be the result of more accuracy in the polls themselves. I consider this chump change. A lot of people realized they were taken for chumps, so they’ve changed parties.
TDM