2016 Presidential Election Thread: TRUMP vs. Hillary SMACKDOWN
10-31-2016
, 07:48 PM
10-31-2016
, 07:53 PM
10-31-2016
, 07:55 PM
10-31-2016
, 07:59 PM
10-31-2016
, 08:05 PM
10-31-2016
, 08:08 PM
10-31-2016
, 08:08 PM
10-31-2016
, 08:09 PM
10-31-2016
, 08:33 PM
10-31-2016
, 08:33 PM
Quote:
I thought I'd make a poast explaining why us politics nerds are so sure Clinton will win. Getting into the final week before election we're probably going to get a bunch of people in here who don't know this stuff.
The Clinton Firewall
Clinton has a set of states that are solidly in her column, which Trump will need to make inroads into to win. These states are CO, PA, NH, WI, MI and MN. Traditionally VA was included as well but that is now safe Clinton to an extent that takes it out of the conversation (RCP: C+8.0, 538: C+7.2).
Clinton's worst-case map with the Clinton Firewall holding looks like this:
So that's Trump's basic problem. He needs to somehow prevent that map from happening.
Why are we so sure those states are going Clinton?
Since July, in the RCP poll aggregator, Trump has virtually not led in any polls in any of these states. There was one in Michigan in August, two in Colorado in September, a few outliers basically. Nothing in any of the rest. Clinton has racked up a ton of polls showing her ahead. Here's how a couple of poll aggregators have the leads in the Firewall States:
These are serious leads this late in the campaign. Most voters have made up their minds by this point.
Basically we're looking at serious polling error, or some huge swing to Trump late in the campaign, or it looks like the Firewall will hold, tightening race or not. The problem with expecting the polls to be wrong is that they look to be correct based on early voting in OH, which is a good test for whether polls will hold across the rest of the Midwest.
The NV+NC insurance policy
Clinton is also showing leads in NV (RCP: C+1.5, 538: C+1.4) and in NC, which has been surprisingly strong for her (RCP: C+3.3, 538:C+1.3). Importantly, early voting has been going on there for some time, and the early voting numbers also show Clinton leading. This means a lot of her advantage there is baked in and provides an insurance policy against losing one of the Firewall States.
NC by itself will take care of losing anything except PA, adding NV to that also handles PA. One other scenario worth mentioning is that NV plus either NE-2 or ME-2 (the congressional districts worth 1 EV each) will handle the loss of CO without needing NC.
What about OH and FL?
On election night the networks will make a big deal out of these big traditional battleground states. They're really just icing on the cake for Clinton. If she wins either of them, the election is instantly over, but she doesn't need them if the Firewall is holding. OH looks to be going Trump, FL will be close.
The early states to watch are PA and NH. If they look to be holding comfortably for Clinton, then she's not going to have any worries.
Why does 538 give Trump such a big chance to win then?
538 use a methodology that builds a lot of uncertainty into polling numbers. Nate's justification for this is that we only have a small sample size of Presidential elections, which means that maybe we've just gotten lucky so far in terms of polling accuracy and haven't seen the tail of the distribution. In other words he's saying the chances of systematic polling error may be higher than we think. I don't buy this for a number of reasons, the most important of which is that early voting numbers are in line with polling.
The Clinton Firewall
Clinton has a set of states that are solidly in her column, which Trump will need to make inroads into to win. These states are CO, PA, NH, WI, MI and MN. Traditionally VA was included as well but that is now safe Clinton to an extent that takes it out of the conversation (RCP: C+8.0, 538: C+7.2).
Clinton's worst-case map with the Clinton Firewall holding looks like this:
So that's Trump's basic problem. He needs to somehow prevent that map from happening.
Why are we so sure those states are going Clinton?
Since July, in the RCP poll aggregator, Trump has virtually not led in any polls in any of these states. There was one in Michigan in August, two in Colorado in September, a few outliers basically. Nothing in any of the rest. Clinton has racked up a ton of polls showing her ahead. Here's how a couple of poll aggregators have the leads in the Firewall States:
| RCP | 538 | |
|---|---|---|
| CO | +5.0% | +4.6% |
| PA | +5.0% | +5.0% |
| NH | +5.4% | +5.4% |
| WI | +5.7% | +5.6% |
| MI | +7.0% | +5.3% |
| MN | +4.6% | +6.2% |
These are serious leads this late in the campaign. Most voters have made up their minds by this point.
Basically we're looking at serious polling error, or some huge swing to Trump late in the campaign, or it looks like the Firewall will hold, tightening race or not. The problem with expecting the polls to be wrong is that they look to be correct based on early voting in OH, which is a good test for whether polls will hold across the rest of the Midwest.
The NV+NC insurance policy
Clinton is also showing leads in NV (RCP: C+1.5, 538: C+1.4) and in NC, which has been surprisingly strong for her (RCP: C+3.3, 538:C+1.3). Importantly, early voting has been going on there for some time, and the early voting numbers also show Clinton leading. This means a lot of her advantage there is baked in and provides an insurance policy against losing one of the Firewall States.
NC by itself will take care of losing anything except PA, adding NV to that also handles PA. One other scenario worth mentioning is that NV plus either NE-2 or ME-2 (the congressional districts worth 1 EV each) will handle the loss of CO without needing NC.
What about OH and FL?
On election night the networks will make a big deal out of these big traditional battleground states. They're really just icing on the cake for Clinton. If she wins either of them, the election is instantly over, but she doesn't need them if the Firewall is holding. OH looks to be going Trump, FL will be close.
The early states to watch are PA and NH. If they look to be holding comfortably for Clinton, then she's not going to have any worries.
Why does 538 give Trump such a big chance to win then?
538 use a methodology that builds a lot of uncertainty into polling numbers. Nate's justification for this is that we only have a small sample size of Presidential elections, which means that maybe we've just gotten lucky so far in terms of polling accuracy and haven't seen the tail of the distribution. In other words he's saying the chances of systematic polling error may be higher than we think. I don't buy this for a number of reasons, the most important of which is that early voting numbers are in line with polling.
+1 to all of it
10-31-2016
, 08:33 PM
10-31-2016
, 08:34 PM
Quote:
Thank you for this post. A couple of quick questions: as far as her firewall states where her lead seems to be 5-6%, how can we translate that to a percentage of chance to win that individual state? High than 90%
It's scary though to think that in a 6 state firewall, even if her odds in each one was 90%, she's only a 53% to go 6 for 6 (.9^6). Or am I thinking about this the wrong way?
Also, would you mind elaborating if you have the time on the other reasons you don't buy Nate's valuing of uncertainty?
It's scary though to think that in a 6 state firewall, even if her odds in each one was 90%, she's only a 53% to go 6 for 6 (.9^6). Or am I thinking about this the wrong way?
Also, would you mind elaborating if you have the time on the other reasons you don't buy Nate's valuing of uncertainty?
You can't just multiply the percentages together like that because they're not even close to independent events, they're highly correlated. If you look at the models, a good rule of thumb is that the win% in the most vulnerable Firewall State is roughly the same as the model's overall election win%. There's no reason that has to be the case, that's just how the maths seems to work out.
Other reasons, um, so here's an excerpt from this article discussing their methodology:
Quote:
Assumption No. 2: The FiveThirtyEight model is calibrated based on general elections since 1972.
Why use 1972 as the starting point?... That’s when we begin to see a significant number of state polls crop up in our database...
What if we changed this assumption? If we calibrated the model based on presidential elections since 2000 only — which have featured largely accurate polling — Clinton’s chances would rise to 95 percent, and Trump’s would fall to 5 percent.
Why use 1972 as the starting point?... That’s when we begin to see a significant number of state polls crop up in our database...
What if we changed this assumption? If we calibrated the model based on presidential elections since 2000 only — which have featured largely accurate polling — Clinton’s chances would rise to 95 percent, and Trump’s would fall to 5 percent.
This is where context matters imo and you can't just look at raw numbers. 1992 and 1996 were big misses. But 1992 had Perot, which made polling more difficult. And probably more importantly, both elections were polling as enormous wins for Bill Clinton. It's quite possible that some voters didn't bother to show up on election day because victory was so inevitable. I'm not sure these blowout elections are good data points.
10-31-2016
, 08:36 PM