the polls did better than you expected in the 2016 election, but not state polls, they sucked
Remember when everybody said that the polls completely got the 2016 presidential election wrong? Now we have final numbers on the popular vote count, and guess what? The national polls were on target:
- In the Real Clear Politics rolling average, the final estimate was HRC up by 3.3%. In the final popular vote count, the Cook Report found that the final difference was 2.1%. Being 1.2% off on the margin is pretty flipping good.
- In terms of the percent per candidate, the polls did worse because people over reported support for 3rd parties. Stein and Johnson together got 3% more in the polls than the results. This is evidence for the “parking lot theory of third parties.”
However, the state polls sucked. Not too hard, but they did suck a little bit, except Wisconsin and Minnesota, which totally sucked:
- Wisconsin – off by over 7%.
- Michigan – off by 3.4%
- Ohio – off by 4.6%
- Pennsylvania – off by 2.6%, which is not bad. HRC losing Pennsylvania was definitely within the margin of error here.
- Minnesota – off by 4.7% (My average, 6.2% vs. 1.5% final)
This is consistent with conventional wisdom about state polls, which is that they are less reliable because it is hard to pinpoint people in states, hard to identify likely voters, and have smaller electorates that can fluctuate (e.g., voter registration laws or bad weather).
Still, in retrospect, looking at state polls did suggest that a popular vote/electoral vote split was possible. A Trump victory was within the margin of error of the polling average in a number of states such as New Hampshire and Pennsylvania. This observation about state polls is also consistent with the finding that the HRC lead was due to urban centers.
Bottom line: The conventional social science about polls held up. National polls do decently, states polls a bit worse and in some cases badly. However, they was plenty of evidence that Trump might get an electoral college victory, but you had to really read the state polls carefully.
50+ chapters of grad skool advice goodness: Grad Skool Rulz ($2!!!!)/Theory for the Working Sociologist/From Black Power/Party in the Street
They actually sicked less if we consider that out of 50 states, 2 or 3 predictions should be beyond the 95‰ confidence interval. so maybe 1 error (Michigan) can be explained by difficulty in getting the right sample.
LikeLike
sebastianguzman
January 17, 2017 at 5:50 am
Fair point.
LikeLike
fabiorojas
January 17, 2017 at 5:15 pm