03 November 2020

Predictions and Probabilities

Who's going to win in the 2020 election?

How Accurate Is Polling?

The 2016 polls were seriously off, although barely within normal margins of error. This has led to lots of uncertainty which is one factor that has been driving up voter turnout.

Polling inaccuracies are strongly correlated with:

* the strength of third-party candidates (high in 2016 due to Sanders supporters favoring a third-party candidate over Clinton, low in 2020), 

* the proportion of likely voters who are undecided (high in 2016, an open race with Trump having no political history, low in 2020 in which Trump and Biden are known quantities), "A week before the 2016 election, around 14% of respondents said they were undecided or intended to vote third party -- and the vast majority of late deciders voted for Trump. This year, there are much fewer.", and

* the volatility of pre-election polling (high in 2016, very low in 2020). Instead, there has been a steady trend of declining support for Trump with him having 31% odds of winning on August 30 that have fallen steadily to 10% in the final forecast 65 days later. "Clinton's lead was smaller throughout, and more unstable. Biden's has never been < 6.6 points."

Pollsters have also corrected their polls where possible to address methodology flaws unearthed in 2016.

These points (and the fact that there has been lots of early voting) are recapped here (the source of the quotes above).

So, I think that, on average, polls are going to be much more accurate in 2020 than they were in 2016.

Another issue that polling struggles with in turnout models. Most polls adjust their raw results based upon what they think the electorate will look like in 2020 which is always an educated guess.

This year, particularly in Southern states and red Western states with a history of voter suppression efforts, COVID-19 facilitated early voting has caused turnout to be very high, for example, in North Carolina, Georgia, Florida, Texas, Arizona and Montana. Turnout models in most polls probably don't fully reflect this reality. So, polls in those states are going to tend to err in the Republican direction since higher turnout almost always favors Democrats.

538 has the most thoughtful and data driven prediction, although it probably overestimates tail probabilities somewhat (while other predictions probably underestimate them).

The Presidential race

The final forecast from 538 is that Biden has an 89% chance of winning and that Trump has a 10% chance of winning (Trump was given a 35% chance in 2016 using essentially the same approach). The Economist gives Trump even weaker chances of winning (about 4%).

538 predicts that Pennsylvania is most likely to be the marginal state with a forecast vote share of 52.0% for Biden and 47.3% for Trump (a 4.7 percentage point margin of victory).

Montana still shows a strong polling lead for Trump (6.3 percentage points) that even high turnout probably won't flip.  So do South Carolina (7.5), Alaska (8.5), Missouri (9.5), Indiana (10.8), and Nebraska's First Congressional District (12.2).

Three states are leaning in favor of Trump but by very thin margins that insufficiently optimistic turnout models, trend lines, and just random error could easily flip to Biden: Texas (1.5), Iowa (1.5) and Ohio (0.6).

Four more states and two Congressional Districts that aren't the marginal states lean towards Biden: Georgia (0.9), Maine's Second District (1.6), North Carolina (1.7), Florida (2.5), Arizona (2.6), and Nebraska's Second Congressional District  (3.2).

These seven states and two districts before one gets to the marginal state, leaves a lot of hurdles for Trump to overcome to win the electoral college, and as I explained above, I think that Texas, Georgia, North Carolina, Florida and Arizona are probably slightly erring in favor of Trump due to higher turnout than the turnout models in some of the polls have predicted.

The hurdles involved in Trump taking states with less favorable polling than Pennsylvania are quite dim in Nevada (4.1), Michigan (8.0), Wisconsin (8.3), Minnesota (8.9), New Hampshire (10.6), Colorado (11.7), Maine (11.8) and Virginia (12.5), and the polls probably lean inaccurately towards Trump due to turnout models predicting lower voter turnout in Nevada. 

States beyond those are pretty much safe without question.

The odds of Trump winning the popular vote according to 538 are about 3%. So, 70% of the odds of Trump winning are with a loss in the popular vote and a win in the Electoral College.

The odds of no one getting 270 electoral votes, sending the election to the House of Representatives is less than 1%.

The odds that the race comes down to less than half of a percent in a decisive state, likely triggering a recount, is about 4%.

Pennsylvania worries me the most, since pre-election turnout there is only 38% of the 2106 vote, but the polling is very consistent in showing a lead for Biden of at least five percentage points, and there are lots of high quality polls to support that.

The Real Clear Politics site (which leans slightly conservative) has a "no toss up" map based upon state polling averages with 319 electoral votes for Biden and 219 electoral votes for Trump. This leaves Biden with a 49 electoral vote margin of error.

With toss ups, it puts 216 electoral votes in the Biden column, 125 in the Trump column, and 197 electoral votes as toss ups, with Biden needing 54 electoral votes from "toss up" states to win. 

But the polling puts Michigan, Wisconsin and Minnesota, which have a combined 36 electoral votes, all as safely in the Biden column as it puts South Carolina in Trump column and RCP puts South Carolina in the Trump column. This would put 252 electoral votes in the Biden column and 125 in the Trump column with 161 electoral votes as toss ups, with Biden needing only 18 toss up electoral votes to win. Pennsylvania or Florida or Texas would cinch the 270 electoral votes on their own. Trump, in contrast, needs 145 of the 161 toss up electoral votes.

Without Pennsylvania or Florida or Texas, there are still los of combinations of states that could win it for Biden. For example, North Carolina and Georgia, North Carolina and Iowa, Georgia and Iowa, Nevada, Arizona and Iowa, North Carolina and Nevada, North Carolina and Arizona, North Carolina and Ohio, Georgia and Nevada, Georgia and Arizona, Georgia and Ohio, Arizona and Ohio,

In a tie vote for President with 269 electoral votes for Biden and 269 electoral votes for Trump, the President is selected by U.S. state delegates in the House voting as single states with one vote each. In this calculus, Democrats get 21-23 votes and Republicans get 27-29 votes, so Trump is re-elected.

So, not only is Biden up against an electoral college that is shifted about 3 percentage points of the popular vote against him, but Biden also needs 270 electoral votes to win, while Trump needs only 269 electoral votes to win.

Likewise, a conservative 6-3 majority on the U.S. Supreme Court gives Trump a clear edge over Biden if an outcome dependent decision like the one made in Bush v. Gore in 2000 comes before it. But that unlikely to happen in this election.

Basically, Biden needs a clear win to win, while Trump is favored if there is any gray in the outcome of the election.

CNN's poll of polls gives Biden a 10 percentage point lead in the national polls going into the election, compared to the 3 percentage point polling lead that Clinton had over Trump. The shift from 2016 to 2020 looks to be roughly a 8 percentage point decline in Trump support reasonably uniformly nationwide (compared to the final results where Clinton led by 2 percentage points in national polling, it is a 6-7 percentage points lead compared to the final polls). 

For now though, Biden is leading in the national polls by about 10 points. That’s 8 points better than Hillary Clinton won the national popular vote by in 2016. 
And remember, Biden’s lead is also significantly wider than where the final national polls put Clinton’s lead in 2016. Those national polls had Clinton up 3 to 4 points in the national popular vote, which turned out to be quite accurate.

Now take a look at the current average of polls in the states Trump won by 10 points or less in 2016. At the same time (in parentheses), we’ll examine what we’d expect those averages to be by applying an 8 point uniform swing from the 2016 results. A uniform swing is simply shifting all the results a certain amount (e.g. 8 points in Biden’s direction). We’re shifting these states 8 points because Biden’s winning nationally by 10 points, and Clinton won nationally by 2 points.

* Michigan: Biden +8 points (Biden +8 points)
* Wisconsin: Biden +8 points (Biden +7 points)
* Pennsylvania: Biden +7 points (Biden +7 points)
* Arizona: Biden +4 points (Biden +5 points)
* Florida: Biden +4 points (Biden +7 points)
* North Carolina: Biden +3 points (Biden +4 points)
* Georgia: Biden +2 points (Biden +3 points)
* Iowa: Biden +1 point (Trump up 1 point)
* Ohio: Tied (Tied)
* Texas: Trump +2 points (Trump +1 point) 
What should be quite apparent is the state polls look almost identical to what you’d expect given a uniform shift of 8 points across states. The average difference is just a point. Moreover, there is no bias with Biden doing 8 points better than Clinton did in the average of state polls, as you’d expect with the national polls where they are.
In early voting, more than 20% of voters have been first time voters. Given polling of young voters and older voters in the past, probably more than half of the shift is due simply to older voters who disproportionately support Trump dying and younger voters who disproportionately support Biden entering the pool of voters. A few percent of Americans who voted in 2016 have shifted away from Trump over the last four years, but that percentage is surprisingly small for such a catastrophe of a Presidency.

Control of the U.S. Senate

Democrats currently hold 47 seats in the U.S. Senate.

538 puts the odds of Democrats retaking the U.S. Senate at 75% in its final forecast. The most probable outcome is 51 Democrat U.S. Senate seats, the mean is 51.5 U.S. Senate seats. There is a 10.6% chance that Democrats come up one short with 49 U.S. Senate seats, a 7.0% chance of coming up two short with 48 U.S. Senate seats, and a 6.3% chance of doing worse.

Real Clear Politics in its "no toss up" analysis predicts a 50-50 split in the U.S. Senate, which is Democratic control if Biden wins the Presidency (although not until January 20, 2021, before then, Vice President Pence casts the tie vote).

This implies Democratic pickups in Arizona, Colorado, Maine and North Carolina, and a Democratic loss in Arkansas.

The RCP list shows more states as toss ups, but again, some of the toss ups lean strongly Democratic.

Control of the U.S. House of Representatives

The final U.S. House forecast from 538 is that Democrats have a 97% chance of retaining control, with an average of 239.3 seats. A majority is 218 seats.

Currently there are 232 Democrats, 197 Republicans, 1 Libertarian and 5 vacancies.

The current mix of House seats was set in the 2018 midterm elections with similarly high turnout for a midterm election and with a similar political climate, so less of a shift is expected in these races. The odds of the Democrats losing 18 seats that they hold and all five 5 vacant seats, and not picking up a single Republican held seat, is low.


Colorado is a safe state for both Biden-Harris and its Democratic U.S. Senate candidate, John Hickenlooper, our former two term Governor and former Mayor of Denver.

The only federal race that is close in Colorado is the Third Congressional District consisting of Western and Southern Colorado, in which several polls have shown Diane Mitsch Bush, the Democrat, leading by 1 percentage point, over primary upset winner, Lauren Boebert, a far right conspiracy theory adherent with a checkered past and no prior political experience.

Democrats are strongly favored to retain control of the Colorado State House, and favored, although less strongly to retain control of the Colorado State Senate. High voter turnout in the state and longer term trends to the left for Colorado due to demographics, migration into the state, increased urbanization and changing attitudes, also favors this outcome.

There are some important ballot issues at the state level in Colorado that have been the subject of only a little polling.

The repeal of the Gallagher Amendment is favored to win, but with large numbers of undecided voters who tend to vote against tax increases.

The National Popular Vote ballot issue 113 which would add Colorado to a National Popular Vote interstate compact, is favored to win but only by a thin margin.

Issue 115 to ban late term abortion is favored to lose, but only by a thin margin.

Issue 118 to create a government financed paid medical and family leave system paid for with a 0.9% payroll tax (half employer and half employee) is favored to win by a safe margin.

I have seen no polling on the many other measures facing Colorado voters and have only win intuitions myself about how they will come out.

Other States

While coverage of no federal elections is scarce, the likelihood is, going into redistricting following the 2020 Census, that Democrats will shift more states to total control and more states from total Republican control to mixed control. This should reduce gerrymandering against Democrats as will the shift of many states towards more neutral redistricting commissions.

Virginia is voting on moving to a neutral redistricting commission today (potentially weakening pro-Democratic gerrymandering).

Many states are leaning towards legalizing marijuana, in some cases only medical marijuana by ballot measures today, and in some cases recreational use as well. These measures could help tip the balance towards national legalization in a Biden administration.

What it would do: Allow adults 21 years and older to possess, consume or transfer up to one ounce of cannabis and create a regulatory system for the products' cultivation and sale. 
What it would do: Allow physicians to recommend medical cannabis for patients with any of 22 qualifying conditions such as cancer, multiple sclerosis and post-traumatic stress disorder. 
What it would do: Limit the smoking of medical cannabis to people who are terminally ill, and would leave the future regulatory framework up to the legislature. 
What it would do: Amend the state's constitution to establish 21 as the legal age to purchase, possess and consume cannabis. 
What it would do: Allow adults in the state to possess, buy and use cannabis for recreational use and defined a 20% tax on recreational cannabis. It would also allow people serving a sentence for certain cannabis-related acts to apply for resentencing or records expungement. 
What it would do: Amend the state constitution to legalize cannabis for personal, non-medical use by adults 21 and older. 
What it would do: Establish a medical cannabis program and registration system for people with qualifying conditions. 

What it would do: Legalize cannabis for all adults and require state legislators to adopt medical cannabis and hemp laws.

Puerto Rico has a clean up or down vote on statehood today that is likely to pass. If it passes, it could speed up its admission as a U.S. state if Biden wins and Democrats win control of the U.S. Senate.

Mississippi voters will vote on a new flag that is not longer based on the Confederate Battle Flag. It's old flag has been retired but the new one has not yet been adopted (the new one still has the "In God We Trust" dig).

There is an anti-abortion measure in Louisiana (that would end all state level protections for abortion rights limiting those protections to the U.S. Constitutional minimum), a measure about the employment status of gig workers like Uber drivers in California, measures to adopt ranked choice voting in Alaska and Massachusetts, a measure to drop "and Providence Plantations" from Rhode Island's official name, a measure to prohibit local government concealed carry gun control laws in Montana, a measure to restore affirmative action in California (banned in 1996), a measure to allow 17 year olds to vote in California, a measure to end cash bail (under a 2018 law), and in Mississippi:

Mississippi's Ballot Measure 2 asks voters whether to remove a Jim Crow era voting provision that requires a candidate to secure a majority in the state's popular vote and to win a majority of Mississippi's 122 state house districts.

No comments: