Poll Accuracy
Posted: Sat Jan 12, 2019 7:41 pm
I did something for another discussion I was having and got some interesting results. I just took the RealClearPolitics polling averages for the past 10 national races and compared them to the actual results to get an idea as to how much error there is. When I did that I found that, actually, the idea that the polls of overall national sentiment for the 2016 Presidential race actually did very well. I looked at the 2018, 2016, 2014, 2012, 2010, 2008 and 2006 House generic ballot polls as well as the 2016, 2012, and 2010 Presidential polls.
The average absolute value for the difference between the actual result and the polling average is 2.01 percentage points. The absolute value for the difference in the 2016 Presidential race is 1.2 percentage points. In terms of "closeness" the polling average for the 2016 Presidential race was 3rd closest among the 10 cases.
The extremes were the 2006 House race, where the polling average under estimated Republican performance by 3.6 percentage points, and the 2012 Presidential race, where the polling average under estimated Democrat (Obama) performance by 3.2 percentage points.
The overall average difference between the actual Democrat - Republican difference and the polling Democrat - Republican difference is -0.01. What that means is that there is very little bias towards one party or the other over time. The value you're looking for to see "unbiased" is "0" and "-0.01" is about as close as you can reasonably expect to get to that.
Yes I know that's arcane. But the bottom lines are that the polls were not way off during 2016 and that, on average, you can figure that when you look at the body of polling of national sentiment it might be "wrong" with respect to which position is in the majority as long as one side or the other isn't "up" by more than about 3 1/2 percentage points or more. When you get into the 2 percentage point margin area it gets really uncertain.
The average absolute value for the difference between the actual result and the polling average is 2.01 percentage points. The absolute value for the difference in the 2016 Presidential race is 1.2 percentage points. In terms of "closeness" the polling average for the 2016 Presidential race was 3rd closest among the 10 cases.
The extremes were the 2006 House race, where the polling average under estimated Republican performance by 3.6 percentage points, and the 2012 Presidential race, where the polling average under estimated Democrat (Obama) performance by 3.2 percentage points.
The overall average difference between the actual Democrat - Republican difference and the polling Democrat - Republican difference is -0.01. What that means is that there is very little bias towards one party or the other over time. The value you're looking for to see "unbiased" is "0" and "-0.01" is about as close as you can reasonably expect to get to that.
Yes I know that's arcane. But the bottom lines are that the polls were not way off during 2016 and that, on average, you can figure that when you look at the body of polling of national sentiment it might be "wrong" with respect to which position is in the majority as long as one side or the other isn't "up" by more than about 3 1/2 percentage points or more. When you get into the 2 percentage point margin area it gets really uncertain.
