There’s been a map floating around the internet claiming to show the election outcome if only millennials had voted; but it’s based on October polls, not on actual exit polling. “Alas” moderator and commentator Charles, understandably (if pedantically)1 annoyed by this, decided to make a couple of maps based on exit polls, and kindly said I could post them here.
- As a bit of a pendant myself, I relate to this. [↩]
My annoyance was actually more at Shaun King for getting pissed at people for posting the incorrect image. In Shaun King’s defense however, white millennials voted 49-44 for Trump, so any white people using these images to talk about the hope for the future generation needs to remember that the future generation isn’t just young white people. The hope for the future generation is partly that it is less white Anglos than the older generations.
Some minor methodological notes:
for states lacking exit polls (the entire plains except MO and IO and the mountain west and much of the south, plus some of New England), I left them with the candidate who won over all.
for states (OR and WA) where there wasn’t a large enough sample under 30, I used 18-44. For states where there wasn’t a large enough sample under 25, I reused 18-29.
You think that millenials’ politics won’t change as they get older? Mine did. Lots of people that I was young with and now am older with did.
Ron’s right. My own politics have shifted significantly over the decades — although I don’t know how typical my experience is, given that I’ve moved from being a standard-issue Democrat to, by American standards anyway, a foamy-mouthed socialist.
I have a hard time believing that the 25-29 voters would have flipped IA red and simultaneously flipped MN and WI blue. I suspect that these are based on two different data sources, and those three states should properly all be tossups in both maps.
Jameson,
The data source is the same for all three states and for both age groups:
The CNN web site display of Edison Research’s exit polls:
Minnesota exit polls:
exit polls
25-29 year olds show a suspiciously high third party vote compared to 18-24 year olds. The number of respondents in each age group is about 160 people, so the margin of error is pretty large.
Iowa exit polls:
exit polls
The jump in Iowa is huge, and doesn’t seem to be driven by third party voting. Also, twice as many people were surveyed in Iowa, so about 300 people in each age group. The 18-29 group is a 6 pt margin, which should probably be called too close to call.
In Wisconsin:
exit polls
The Wisconsin shows 30-29 as the most pro-Clinton, with 25-29 less pro-Clinton, and 18-24 barely pro-Trump. I’d agree that both 18-24 and 18-29 should be marked as too close to call.
I may make up a version of the map using a cutoff for too-close-to-call of a 10 pt margin. The raw results for 18-29 year olds from that are:
Clinton wins: AZ, CA, CO, FL, GA, IL, MI, NV, OR, NJ, NM, NY, NC, TX, VI, WA
To close to call: IN, IA, ME, MN, NH, OH (Clinton wins 18-24 y.o.), PA, SC, UT, WI
Trump wins: KY, MO
The only change from those results for 18-24 year olds is that Ohio swings over to Clinton.
Just popped in out of curiosity to see what Amp was writing these days. Random comment: You should probably use a county specific map.
County ones are way more helpful because they demonstrate the Brexit-like nature of things: http://www-personal.umich.edu/~mejn/election/2016/countymaprb1024.png
but if you want more detail you may want to use a color-scaled one:
http://www-personal.umich.edu/~mejn/election/2016/countymappurple1024.png
(for the full page and some interesting though odd maps see http://www-personal.umich.edu/~mejn/election/2016/)
Those maps are “all voters” as opposed to “18-29 voters,” but still help to illustrate the problem with state-level maps. Flipping from red to blue is less helpful than seeing the reality underneath.
Well, the advantage of state-level maps is that the electoral college is state-level. Counties give you a better sense of where different voters are distributed, but not of what election results would have panned out to be if only certain demographics voted.
Also, exit polls only exist at a state-level. Trying to translate state level exit polls into county by county maps would be way beyond what I have time for and would require access to the full exit poll cross-tabs at a minimum.
However, those purple county-by-county maps are my go-to election maps. Amp and I were talking a few nights ago about making a post around the differences between the 2012 map (I prefer the population scaled cladograms, as acreage doesn’t vote):
http://www-personal.umich.edu/~mejn/election/2016/countycartpurple1024.png
and the 2016 cladogram:
http://www-personal.umich.edu/~mejn/election/2012/countycartpurple1024.png
(Frustratingly, the shape of those images is slightly different between 2012 and 2016. I also think there may be an error in some counties in Kentucky in 2016).
They do show the very striking shift in rural Appalachia and the Mid-west, with those areas shifting from being purple to being as red as the rural Plains and rural white (that is, excluding the black belt) Deep South.
Even with the full cross-tabs, I wouldn’t think the exit polls on a county-by-county basis would be correct. They’re trying to hit a representative sample in some states individually, and a representative sample nationally, but cutting that down to county-by-county would probably have weird artifacts.
(At first I thought the “doesn’t care about representative sample of non-swing states individually” might cause your weird Midwestern thing, but MN, IA, and WI are all states they actually polled, so *shrug*)
I mean, you said there were about 300 people in your Iowa age groups. There are 99 counties in Iowa, so 3 people per county per age group on average. That’s just not enough to tell you anything.
(Why are there 99 counties in Iowa? So there’s nowhere you’re more than a day’s horse ride from the county seat, so you never have to spend a night on the prairie even if you have business in town.)
And lots of those counties are tiny, so even if they’d distributed the surveyed sites evenly across counties (which they don’t), it wouldn’t be 3 per county. I was thinking more that you could try to related the exit poll results to county demographics, but I agree that when even at the state level we are getting dodgy results like 18-24 year olds being way more conservative in MN and WI but way more liberal in IA, that trying to pull any more detail out of the exit polls would be a fool’s errand. Even at just the level of rural 18-29 year olds, we’d still be down to less than 100 respondents.
Andrew Gelman has a methodology for imputing a national survey down to county-level results using demographics; he calls it “Mr. P”, for “multilevel regression with poststratification”. And yes, it does require raw data, or at the very least extensive crosstabs.
As far as I know, there aren’t any tools yet that do that kind of thing automatically. So really if you’re not a statistician, state level results are the best you can do.
I wouldn’t trust exit polls. In fact, the opinion pollsters are coming around to something that I’ve been saying for a couple of years now – conservative voters tend to not respond to opinion polls, exit polls and media questioning more than liberal ones do due to a general distrust of them. They don’t see the media as reporting on the election. They saw the media’s actions over the last 16 or so months as campaigning for their candidate.
National polls missed by about 2% this year. They missed by about 1% in the opposite direction in 2012. At the state level, the median poll in the last two weeks in Michigan missed by 2% (Michigan wasn’t heavily polled). The median poll in the last week in Pennsylvania missed by 3.5%. Etc. None of those are large misses.
Exit polls have their own large problems, but non-response from older Trump voters would make younger voters seem more pro-Trump in the exit polls (since the polls are reweighted to the vote totals, and the Trump votes have to have come from somewhere).