One of the polling difficulties is the fact that the US has become a majority cell phone only nation. Pollsters have made considerable efforts to adapt to this evolution but have yet to effectively solve the problem. While under counting cell phone only households can be approximately solved by weighting by demographics (low income families are more likely to be cell phone only), there is one problem that is much harder to fix: Mobility.
Cell phone only families are also much more likely to move to other states in search of jobs and opportunities. They bring with them the cell phones and the phone numbers along with the area code. This makes it nearly impossible for pollsters to track this group of people. These people are the most likely to vote Democratic because of their economic status, education, and experience, and are also often unlikely voters — many of them unable to keep up with politics within the first few years of a new job. In other words, by missing these people in the polls, they are likely under-estimating support for the Democratic Party. (This would not contradict poll results in midterm elections when they tend to under-estimate Republican support, because people who recently moved to a new place are unlike to vote in local elections).
So how much are the polls likely wrong in a presidential election? We can make a rough estimate by looking at the population change in each state, and assuming 50% of the people who move in are cell phone only households, and neglecting population growth by birth and by foreign immigration, which is around 0.77% in 2016. Update: I had neglected voter turnout which should not be 100%. If we assume the national average of about 60%, then the estimated poll error should be reduced accordingly. The table is revised using 60% turnout number
Let’s look at the key battleground states.
State |
Population change 2010-2015 |
potential poll error |
texas |
8.1% |
2.4% |
Colorado |
7.5% |
2.2% |
Florida |
7% |
2.1% |
nevada |
6.5% |
1.9% |
Arizona |
6.1% |
1.8% |
georgia |
4.9% |
1.5% |
north carolina |
4.8% |
1.4% |
Nationally, about 12% of US population move to another state each year. Again if about half of these people are cell phone only household, there are potentially 6% of all population that is difficult to capture by pollsters, and the potential poll error on the national level is about 3.6%. This might also explain why a highly respected pollster Selzer, who had been impeccible in polling Iowa, were coming out with what would seem to be outlier poll results on both national level and in states such as Florida. This is much more than a likely voter screen being too strict. Iowa never had much influx of population!
While this theory might be too simplistic and overstate the potential polling error, another data point(s) supporting my hypothesis is that in both 2008 and 2012 Obama outperformed the poll averages by about 2-3%. Update: There is also an argument that can be made that the national polls should not have this error if the pollsters do not poll state by state and add the numbers. We don’t know how each pollster conducts their polls.
If the above analysis holds true, based on the current polling numbers one could reasonably expect Hillary to easily win Florida, Navada, North Carolina, and to have decent shots in Arizona, Georgia, and Texas. If we also believe that the superior GOTV can add another 1-2%, Hillary’s national vote margin could well be in the high single digits, about 4-5% above where the polling averages are now.
Of course the validity of this analysis hinges on these unlikely voters actually voting. So the bottom line is, GOTV and we can make this a landslide!