Nate Silver's Blog, page 48

Trump’s Reelection May Hinge On The Economy — And Coronavirus

President Trump’s approval rating has improved slightly amid the coronavirus pandemic. But the short-term gains, reflecting a possible rally-around-the-flag effect at the time of national emergency, may not hold. On the contrary, the strong likelihood of a potentially very deep recession triggered by coronavirus puts Trump’s reelection chances in jeopardy.

Aggressive measures to slow the spread of the disease, and economic stimulus packages that soften the economic blow, could allow the United States to rev back up to speed in the second half of the year and improve Trump’s position. Without that — even though there isn’t any perfect analogy to coronavirus in recent electoral history — the bulk of what we do know suggests that he could be in trouble.

We’ve written a lot about the effect of the economy on presidential elections here at FiveThirtyEight, especially in the run-up to the 2012 general election, when the economic recovery was a major focal point in the contest between Barack Obama and Mitt Romney. For that reason, I’m going to compress a very complicated discussion into under 3,000 words here:

First, I’ll explain my basic view on how to implement an election forecast based on economic conditions.
Next, we’ll see what a series of relatively simple economic models would say about Trump’s potential performance in the fall.
Finally, we’ll run through some possible objections — do econometric models really apply for a president like Trump in the case of something like coronavirus?

To be clear: This is meant to be an opening bid rather than a comprehensive evaluation of these questions, many of which will be worth exploring at greater length between now and November.

A bad economy makes life harder for the incumbent party. Beyond that, we don’t know much.

I’ve often found myself trying to carve out a middle ground between people who think that presidential elections are strictly predicted by economic conditions — and people who think it all boils down to idiosyncratic factors such as candidate charisma and campaign strategy.

It might surprise you to learn that I’m agnostic in this debate, rather than siding with the seemingly more data-driven economic “fundamentalist” approach. The problem with the data-driven approach is that … there just isn’t all that much data to work with. There have been only 18 presidential elections since World War II. (Before that U.S. economic data isn’t very robust.) To account for all of the various ways that the economy can impact people’s lives (for example, joblessness, inflation, take-home income, etc.) — plus all of the other factors that influence elections (such as incumbency, wars, pandemics, scandals, etc.) — is not easy to do with only 18 data points.

In particular, we don’t have enough data to make overly specific claims about the economy. That is, any time you see what you might call a “magic bullet” claim, such as that second-quarter GDP is crucial or that per-capita disposable income is the key economic variable, you should be wary.

In fact, magic-bullet models such as these don’t perform very well out of sample. That’s because they’re prone to suffer from p-hacking and overfitting because of the small sample size of post-WWII elections and the large number of ways to design a model. In practice, what this means is that people designing these models can twist enough knobs so that they fit the past data well, but they don’t actually predict future election outcomes effectively when the modelers don’t know the outcome in advance.

At the same time, it’s very likely that there is some type of meaningful correlation between economic performance and the performance of the incumbent party’s presidential candidate. If you look at a wide range of economic variables, you’ll find that most of the more obvious ones (GDP or employment numbers) indicate that a stronger economy predicts a better performance for the incumbent party.

The best approach in our view — and the one we’ve used in FiveThirtyEight’s presidential forecasts — is to combine various major economic variables into an overall index of economic conditions. To avoid overfitting, we choose variables that reflect a cross-section of economic activity rather than picking ones that happen to fit the results from a small number of presidential elections. In addition, FiveThirtyEight’s Economic Index averages how these numbers have changed at various points, instead of focusing on one particular time frame such as the second quarter.

Overall, the data shows a reasonably clear — although far from perfect — correlation between the economy and incumbent-party performance. Since 1968, the worst years for economic performance as of Election Day were 1980 and 2008, each of which were associated with steep losses for the incumbent party. The next-worst economic year was 1992, in which incumbent George H.W. Bush lost to Bill Clinton. The best years for the economy, 1984 and 1972, resulted in incumbent-party landslides, although another good year, 1968, produced a narrow loss for the incumbent Democrats.

A recession would probably make Trump an underdog — although not hopelessly so

OK, let’s plug in some numbers to our Economic Index and see what it says about Trump. The goal here is to predict Trump’s margin of victory — or defeat — in the popular vote; the Electoral College is another issue and a potential advantage for Trump; we’ll talk about that more in a moment. I’m going to run four versions of the model:

Model 1: Predict the performance of the incumbent party based on the FiveThirtyEight Economic Index as of Election Day. We’ve calculated the Economic Index back to 1968, so that’s the data we’ll use.
Model 2: The same as above, but using the data for incumbents only. That means we’d include 1972 (Richard Nixon), 1980 (Jimmy Carter), 1984 (Ronald Reagan), 1992 (George H.W. Bush), 1996 (Bill Clinton), 2004 (George W. Bush), and 2012 (Barack Obama) but exclude 1968, 1976 (Gerald Ford was an unelected incumbent), 1988, 2000, 2008 and 2016.
Model 3: Predict the performance of the incumbent party based on the FiveThirtyEight Economic Index as of Election Day and the incumbent president’s approval rating as of this point in the election year. For Trump’s approval rating, that’d be 45 percent — which is his approval in polls of likely or registered voters as of March 24 in the FiveThirtyEight presidential approval tracker.
Model 4: The same as above, but for incumbent presidents only.

And for each model, I’ll run through five economic scenarios:

If the economy looks like 1984, meaning booming growth.
1996, meaning above-average growth.
2012, meaning sluggish but positive growth.
1992, meaning a mild recession, although on the brink of recovery.1
Finally, 2008, meaning a severe recession.

All right, here’s what each economic scenario looks like under each model:

Even a mild recession could make Trump an underdog

Projected Trump margin of victory or defeat in the popular vote

Model 1
Model 2
Model 3
Model 4


IF THE ECONOMY LOOKS LIKE …
538 Economic Index
538 Index With incumbents only
538 Index + Pres. Approval
538 Index + Pres. Approval with incumbents only




1984
R+12
R+18
R+9
R+11


1996
R+5
R+9
R+4
R+5


2012
R+2
R+4
R+2
R+2


1992*
D+5
D+6
D+4
D+5


2008*
D+11
D+15
D+9
D+11

* 1992 and 2008 indicate mild and severe recessionary scenarios, respectively.

Note that the models are reasonably similar to one another. However, there are some differences. Namely, the predictions are more sensitive to economic performance when we use data for elected incumbents only (Models 2 and 4). That means incumbents like Trump get a larger share of credit or blame when they’re managing the economy, rather than when the torch is being passed, such as between Clinton and Al Gore in 2000 or George W. Bush and John McCain in 2008.

The economy also matters a bit less once you account for a president’s approval ratings (Models 3 and 4) since approval ratings gives us some sense for how popular the president actually is in practice, rather than how popular he “should” be based on the economy. In the pre-coronavirus economy, Trump was less popular than you might have expected based on economic conditions. That could limit his upside in the event the economy recovers or somehow manages to avoid recession. At the same time, he’s popular enough — and partisanship is strong enough — to potentially limit his downside in the case of a recession.

Nonetheless, in the event of a mild recession — with economic conditions tantamount to 1992 — all four models predict that Trump would lose the popular vote by a solid amount, by margins ranging from 4.0 percentage points (Model 3) to 5.7 points (Model 2). And in the event of a severe, 2008-style recession, they predict a potential landslide loss, by amounts ranging from 9.1 percentage points (Model 3) to 14.6 points (Model 2). However, Trump has two potential saving graces:

First, the Electoral College. In 2016, there was roughly a 3-point gap between Trump’s performance in the popular vote, which he lost by 2 percentage points to Hillary Clinton, and in the tipping-point state, Wisconsin, which he won by about 1 point. Losing the popular vote by 4 to 6 points would probably not be enough to save Trump in the Electoral College, but it would at least be an open question.
Second, these models have fairly high margins of error, which ranges between roughly 5 points and 10 points depending on which version you use. Thus, a model showing Trump losing the popular vote by 4 to 6 points wouldn’t have to be that far off for Trump to win the Electoral College (and perhaps even the popular vote) in the event of a mild recession. In the case of a severe recession, a popular vote win would be quite unlikely, but he’d retain some outside chances at drawing an inside straight in the Electoral College.

Will models like these really work in 2020?

Do models built on ordinary business cycle crests and slumps work in the midst of a global pandemic — something that the U.S. hasn’t experienced in any recent election year?

I’m reserving my right to change my mind on this subject upon deeper philosophical reflection — but the truth is, we can’t really know for sure. When we release the full-fledged FiveThirtyEight 2020 election model later this year, we may have it use a combination of several different priors, some of which use our Economic Index and some of which do not. All of this is still in the whiteboard stage at the moment.

At the same time, there are some objections that I don’t necessarily find compelling. Each of these could make for its own article, and we may cover some of them at more length later. But let’s run through them quickly in a lighting round:

Could voters give Trump a pass because coronavirus is the cause of the recession? Maybe. But even in the case of ordinary economic booms and busts, it’s never entirely clear how much credit or blame the president actually deserves — and the answer is, probably less than he typically gets from the public.

Does Trump deserve more blame for a coronavirus-triggered recession than Bush did for the financial crisis in 2007 and 2008, or Carter did for the rampant inflation and the oil crisis of 1979 and 1980? Well, the Democrats will say yes — especially given Trump’s slow-footed, erratic response on coronavirus — and the White House will say no. But the more a recession brings hardship to families and communities, the more the Republican side of the argument will be pushing uphill.

But what about Trump’s approval rating improving since the coronavirus crisis began? Indeed, Trump’s approval rating has improved in recent days so that it’s among the highest ratings of his presidency. As I mentioned, his approval rating among voters is now roughly 45 percent, which is up from 43 or 44 percent since early March, while his disapproval rating has fallen from 52 to 53 percent to 51 percent.

However, compared with typical rally-around-the-flag effects that follow national crises, these gains are fairly meager. For instance, Bush’s approval rating improved from 51 percent to 86 percent following the September 11 attacks, and Carter’s approval rating nearly doubled in 1979 in the immediate wake of the Iran hostage crisis. (Granted, both of their ratings declined sharply from there.) But Trump is also not seeing nearly as much of an approval rating bounce as other leaders in Western countries, such as Italy’s Giuseppe Conte, France’s Emmanuel Macron, and the UK’s Boris Johnson. So it’s not clear that a small approval rating gain is a bullish sign for Trump.

Do “the fundamentals” even apply anymore? Isn’t everything different in the age of Trump? Sorry, but this is dumb. The 2016 election result — a narrow popular vote win for the incumbent Democrats given the mediocre economy — was actually fairly well-predicted by economic models. (These models don’t say anything about the Electoral College.) The 2018 midterms also went pretty much exactly how the fundamentals predicted given Trump’s middling approval rating and the typical midterm backlash against the incumbent party. Heck, even this year’s Democratic primary, in which Joe Biden is the very likely winner, has been good for the fundamentals-driven “Party Decides” view of the primaries in which the party establishment has a lot of influence.

Could partisanship dull the response to a recession? This is a better objection. With more polarization in the electorate and fewer swing voters, it stands to reason that Trump’s approval ratings will be less responsive to different news events than an earlier president’s might have been. And indeed, Trump’s approval ratings have trended within a narrow range so far throughout the course of his presidency, despite tumultuous events such as the Ukraine scandal and the impeachment proceeding against him.

There is an important catch, however. If the range of possible outcomes is narrower for Trump, that also means the margin of error is lower since the outcome is more predictable. Suppose that in less polarized times, a sharp recession would result in Trump being projected to lose by 12 points, plus or minus 10 percentage points. Given polarization, however, he might only be projected to lose by half as much, or 6 points — but the margin of error would also be half as much, or 5 points.

It’s also worth noting that Trump currently trails Biden in most general election polls by a wide enough margin that the Electoral College probably wouldn’t save him. So if higher partisanship means the outcome is more “locked in” and less likely to change, that isn’t great news for Trump.

Could an economic recovery in the second half of the year help Trump? Yes, it could. If there’s a sharp decline in economic activity over the next few months, and then a steep rebound, I wouldn’t want to have a model that only used second-quarter GDP and pretended everything that happened afterward didn’t matter.

At the same time, there is not a lot of certainty in how long the coronavirus crisis will last, nor how long the economic recovery would take. Moreover, a lot of epidemiologists worry about a potential second wave of the coronavirus in the fall, as occurred in the flu pandemic of 1918.

If I were Trump, I’d want to think six months ahead to the fall. That means I’d want a broad-based stimulus plan that helps ordinary Americans and small businesses to stay afloat during the weeks — or months-long shutdown. I’d want to stamp out the disease as much as possible — even if that means social distancing is in effect for a bit longer. And I’d want to have a Manhattan Project on treatments, testing and surveillance so that the coronavirus is more manageable until a vaccine is developed, which is unlikely until well after Election Day.

Frankly, this isn’t that complicated, and Trump’s incentives are well-aligned. The better off America is by November, the more likely he is to be re-elected.

Could A New Therapy Provide Immunity To COVID-19?

https://abcnews.go.com/video/69715860

 

Politics Podcast: How COVID-19 Is — And Isn’t — Changing Politics

By Galen Druke, Clare Malone, Micah Cohen and Nate Silver, Galen Druke, Clare Malone, Micah Cohen and Nate Silver, Galen Druke, Clare Malone, Micah Cohen and Nate Silver and Galen Druke, Clare Malone, Micah Cohen and Nate Silver












 












More: Apple Podcasts |
ESPN App |
RSS
| Embed


Embed Code

According to the latest polling, the spread of COVID-19 has disrupted the daily lives of the vast majority of Americans. In this installment of the FiveThirtyEight Politics podcast, the crew discusses how that broad effect has — and has not — changed partisan politics. They also ask how the pandemic has shaken up the 2020 general election.

You can listen to the episode by clicking the “play” button in the audio player above or by downloading it in iTunes , the ESPN App or your favorite podcast platform. If you are new to podcasts, learn how to listen .

The FiveThirtyEight Politics podcast publishes Mondays and Thursdays. Help new listeners discover the show by leaving us a rating and review on iTunes . Have a comment, question or suggestion for “good polling vs. bad polling”? Get in touch by email, on Twitter or in the comments.

2020 (Simulated) March Madness Predictions

FiveThirtyEight

2020 (Simulated) March Madness Predictions

Politics Podcast: How Uncertainty Around The Coronavirus Affects Our Life And Politics

By Galen Druke, Nate Silver and Micah Cohen, Galen Druke, Nate Silver and Micah Cohen and Galen Druke, Nate Silver and Micah Cohen












 












More: Apple Podcasts |
ESPN App |
RSS
| Embed


Embed Code

The spread of the new coronavirus has upended American life overnight. In this installment of the FiveThirtyEight Politics podcast, the crew discusses how much we don’t know about the crisis when trying to assess its impact on life and politics. They also discuss how the virus is affecting plans in the states that were planning on holding primary elections on Tuesday.

UPDATE (March 16, 2020, 8:00 p.m.): A judge denied the Ohio governor’s request to postpone the primary election, so, pending an appeal, voting in Ohio is back on for Tuesday. We have removed a disclaimer in the podcast that said Ohio would not vote in person on Tuesday.

You can listen to the episode by clicking the “play” button in the audio player above or by downloading it in iTunes , the ESPN App or your favorite podcast platform. If you are new to podcasts, learn how to listen .

The FiveThirtyEight Politics podcast publishes Mondays and Thursdays. Help new listeners discover the show by leaving us a rating and review on iTunes . Have a comment, question or suggestion for “good polling vs. bad polling”? Get in touch by email, on Twitter or in the comments.

Politics Podcast: Biden And Sanders Face Off Amid Coronavirus Crisis

By Galen Druke, Nate Silver and Micah Cohen, Galen Druke, Nate Silver and Micah Cohen and Galen Druke, Nate Silver and Micah Cohen












 












More: Apple Podcasts |
ESPN App |
RSS
| Embed


Embed Code

In a late-night edition of the FiveThirtyEight Politics podcast, the crew reacts to the first one-on-one Democratic primary debate between former Vice President Joe Biden and Sen. Bernie Sanders. The candidates were sober in discussing the novel coronavirus pandemic, but became testy as they challenged each other’s records in Congress on other issues.

https://abcnews.go.com/video/69618395

You can listen to the episode by clicking the “play” button in the audio player above or by downloading it in iTunes , the ESPN App or your favorite podcast platform. If you are new to podcasts, learn how to listen .

The FiveThirtyEight Politics podcast publishes Mondays and Thursdays. Help new listeners discover the show by leaving us a rating and review on iTunes . Have a comment, question or suggestion for “good polling vs. bad polling”? Get in touch by email, on Twitter or in the comments.

Election Update: Sanders Needs Something Extraordinary To Happen

The FiveThirtyEight primary model is back on after incorporating the results from “Triple M Tuesday”, in which Michigan, Missouri, Mississippi, Washington, Idaho, North Dakota and Democrats Abroad voted. But you may have hardly noticed the change.

That’s because the model was already pretty certain that former Vice President Joe Biden would win a majority of pledged delegates based on his lead so far, as well as polls in those states in addition to states set to vote later this month. Because the results from last night largely matched our forecast, the forecast is essentially unchanged.

But to back up a bit: If Super Tuesday was a good day for Biden, everything since then — including the results from last night — have been great for him. The model expected Biden to get a bounce from Super Tuesday, but the bounce has actually been a bit larger than the model expected, with Biden having claimed a 17-point lead in national polls while also dominating the polls in upcoming states. The results of Tuesday’s contests were largely consistent with the polls, too. As of Wednesday afternoon — with results not yet officially finalized — Biden leads in Michigan by 17 percentage points, in Missouri by 26 points and in Mississippi by a whopping 66 points, where Sanders looks as though he may fall just below the 15 percent threshold required to win statewide delegates.

Biden also won Idaho, although he lost North Dakota, whose party-run primary, with a limited number of voting sites and short voting hours, resulted in very low turnout, which helped Sen. Bernie Sanders as he overperforms in caucuses relative to primaries. So going forward, the model will treat party-run primaries — which are also set to take place in Alaska, Hawaii and Kansas — as being essentially halfway in between primaries and caucuses, boostingSanders’s odds in those states.

In two other states, results are still really incomplete. First, Washington, where Sanders is probably in trouble. He led by a meager 0.2 percentage point margin in the initial vote reported out of Washington on Tuesday night, which consists of ballots that were mailed in some time ago. However, it is likely that Biden will gain ground in Washington from this point forward. That’s because ballots returned later on will more fully reflect Biden’s surge over the past few weeks.

It’s a similar situation in California, another vote-by-mail state, where Sanders’s lead has fallen to 6.7 percentage points after having been 8.7 points in initial returns on election night. But based on the pattern of returns in California, we came up with a formula to estimate what the remaining ballots will look like in Washington. Specifically, the formula is based on the age of each candidate’s voters, and when his voters said they made up their minds, according to the Washington exit poll. Usually, candidates with younger supporters (such as Sanders) tend to see their support grow as additional returns are counted because younger voters are more likely to wait to send in their ballot. However, the exit polls also show Biden having done much better with voters who made up their minds in the last few days, and this consideration will likely outweigh the fact that Biden’s supporters are older.

Based on this formula, our best guess is that Biden will eventually win Washington by around 3 percentage points, which would closely match pre-election polls there. The Washington results used by the model reflect this adjustment, and assume (consistent with betting markets) that there’s an 80 percent chance that Biden eventually wins Washington.

For the Democrats Abroad primary, no results have yet been officially reported, and Democrats Abroad has not promised to release any until March 23, even though the voting window closed yesterday. For the time being, we are relying on scattered reports of precinct-level returns, which show Sanders well ahead, and Sen. Elizabeth Warren actually in second place ahead of Biden. (Keep in mind that some global locations held their primaries before Warren dropped out.) These may or may not be representative of the final results, but since Sanders overwhelmingly won Democrats Abroad in 2016, we think it’s reasonably likely that he does so again.

Overall, based on results as of late Wednesday morning in Michigan, Missouri, Mississippi, Idaho and North Dakota, our adjusted results in Washington, and the unofficial, partial results in Democrats Abroad, we show Biden having won 215 delegates last night, Sanders 145 delegates and other candidates 5 delegates. That’s very close to our projections before the evening began.

Biden likely gained 70 more delegates than Sanders

Estimated delegate counts based on results as of 11 a.m. on March 11

State
Biden
Sanders
other




Michigan
73
52
0


Washington*
45
42
2


Missouri
44
24
0


Mississippi
34
2
0


Idaho
11
9
0


North Dakota
6
8
0


Democrats Abroad†
2
8
3


Total
215
145
5

* Washington results reflect an adjustment as Biden is expected to gain ground as more ballots are returned.

† Democrats Abroad results are based on a limited sample of partial, unofficial returns.

In all other states, we use results “as is” based on reports through the late morning of March 11.

And when combined with states that voted before Tuesday, Biden has an overall lead of around 150 pledged delegates.

Biden leads by about 150 delegates

Estimated delegate counts based on results as of 11 a.m. on March 11

Candidate
Before March 10
After March 10
Current

estimated total




Biden
720
215
935


Sanders
640
145
785


Warren
61
2
63


Bloomberg
53
3
56


Buttigieg
18
0
18


Klobuchar
5
0
5


Gabbard
2
0
2

Washington results reflect an adjustment as Biden is expected to gain ground as more ballots are returned. Democrats Abroad results are based on a limited sample of partial, unofficial returns. In all other states, we use results “as is” based on reports through the late morning of March 11.

One hundred and fifty delegates might not sound like much, but this is a nontrivial lead for Biden. But as Nate Cohn of The New York Times’s The Upshot points out, the delegate math is not actually Sanders’s biggest problem. If the national race were tied going-forward, Sanders would have his work cut out for him, but he would still have a puncher’s chance at the nomination despite trailing in the delegate count so far.

Rather, Sanders’s biggest problem is that he’s down 15 to 20 points nationally, a result reflected not only in the national polls but also in the results from states such as Michigan. That means Sanders tends to lose more ground every time a new set of states votes.

That’s especially likely to be a problem for him over the next two weeks, with Florida, Ohio, Illinois and Arizona set to vote on March 17, and Georgia on the schedule for March 24. Even under the best of circumstances, these would not be a great set of states for Sanders. He’s almost certain to lose badly in Florida and Georgia, as he has elsewhere in the South. Ohio and Illinois look like longshots for Sanders, too, given the results in Michigan. Arizona might theoretically be a good state for Sanders, but he’s well behind in the only recent poll there.

And by the time these states have finished voting, the delegate math will be a major problem for Sanders, and even a massive late surge would probably not be enough to help him win the nomination.

Put another way, Sanders needs something like a 20-point surge within the next week just to remain competitive for the nomination, and even then it would still be an uphill battle for him. And he needs it at a time when Biden potentially stands to gain more ground because of his strong results last night; states such as Michigan could potentially give Biden a further bounce in the polls. Thus, even a strong debate on Sunday for Sanders might not be enough and just merely offset further momentum Biden gained from Tuesday.

The model does not account for any possibility that Biden drops out. (It assumes that candidates will not drop out so long as they remain either the delegate leader or the leader in national polls.) Actuarially speaking, there is some small chance that Biden could have to leave the race because of a scandal, a health problem, and so on; we don’t try to estimate this possibility. And I don’t mean to keep dwelling on these possibilities. But we’re at the point now where if Biden were to lose the nomination, it would likely require something highly out of the ordinary to happen.

Politics Podcast: Is The Democratic Primary Over?

By Galen Druke, Clare Malone, Nate Silver and Micah Cohen, Galen Druke, Clare Malone, Nate Silver and Micah Cohen, Galen Druke, Clare Malone, Nate Silver and Micah Cohen and Galen Druke, Clare Malone, Nate Silver and Micah Cohen












 












More: Apple Podcasts |
ESPN App |
RSS
| Embed


Embed Code

In a late-night edition of the FiveThirtyEight Politics podcast, the crew reacts to the results from Tuesday night’s elections in Michigan, Missouri, Mississippi, North Dakota, Idaho and Washington. They also ask if the Democratic primary is over, given former Vice President Joe Biden’s good night.

https://abcnews.go.com/video/69537848

You can listen to the episode by clicking the “play” button in the audio player above or by downloading it in iTunes , the ESPN App or your favorite podcast platform. If you are new to podcasts, learn how to listen .

The FiveThirtyEight Politics podcast publishes Mondays and Thursdays. Help new listeners discover the show by leaving us a rating and review on iTunes . Have a comment, question or suggestion for “good polling vs. bad polling”? Get in touch by email, on Twitter or in the comments.

3 Ways The Next 2 Weeks Could Go: Even The Best Case Scenario For Sanders Doesn’t Look So Great

Our primary model is now extremely confident that former Vice President Joe Biden will win a majority of pledged delegates. There are some contingencies that the model doesn’t account for — for instance, Biden unexpectedly dropping out for health reasons (the model assumes the delegate leader never drops out) or the primary calendar being radically restructured because of precautions that states are taking around coronavirus. But our model is saying that so long as anything remotely resembling normal operations continues, Biden is very likely to win.

If that confidence seems misplaced — and keep in mind, our forecasts are usually more conservative than other statistical algorithms — it may be because Biden’s ascent has been so rapid that it’s hard to entirely grasp its scope. Biden’s rise in national polls over the past two weeks — from 15.4 percent on Feb. 24 to 51.7 percent today — has been perhaps the fastest in the history of the primaries. He has massive polling leads in states all over the map that were once expected to be competitive.

But it’s also worth performing a stress test on the model. What if Sen. Bernie Sanders substantially beats our forecasts? Could the race look more competitive then?1

As a baseline, let’s start with what happens if the model’s projections are exactly right in every state and territory up through Georgia, which votes in two weeks from now on March 24. (Note that this scenario will look slightly different from our final forecast, because it’s deterministic — it assumes our projections are exactly right to the decimal place — while the full-fledged model accounts for uncertainty.)

Scenario 1: FiveThirtyEight’s projections are spot-on

Delegate projections for March 10 through 24, according to FiveThirtyEight’s forecast as of 9:00 a.m. Eastern on March 10

Date
State
Biden
Sanders
others




3/10
Michigan
77
48
0


3/10
Washington
48
41
0


3/10
Missouri
45
23
0


3/10
Mississippi
28
8
0


3/10
Idaho
11
9
0


3/10
North Dakota
9
5
0


3/10
Democrats Abroad
7
6
0


3/14
Northern Mariana Islands
4
2
0


3/17
Florida
152
67
0


3/17
Illinois
100
55
0


3/17
Ohio
85
51
0


3/17
Arizona
42
25
0


3/24
Georgia
74
31
0





Period
Biden
Sanders
others





Pre-March 10 total
720
640
139



March 10 total
225
140
0



3/14 – 3/24 total
457
231
0



Grand delegate total through 3/24
1402
1011
139

Unless something big changes, things will get very out of hand for Sanders very soon. In states that have already voted, he currently trails Biden by around 80 delegates based on results as reported as of this morning. (That deficit that may expand as California continues counting its vote — late-counted ballots have been more favorable for Biden than the ones counted on election night — but leave that aside for now.) Then, if our projections are exactly right tonight, Biden would net another 85 delegates, putting him 165 delegates ahead overall.

But still bigger problems await Sanders in the March 17 states and in Georgia on March 24. These states have huge numbers of delegates and some of them — Florida and Georgia in particular — are projected as big Biden wins. The model currently has Biden gaining a net of 226 delegates on Sanders in the states and territories that vote from March 14 through March 24.

That would give Biden an advantage of 391 delegates — which would be essentially insurmountable barring extraordinary contingencies. Sanders would have to win around 69 percent of the remaining delegates from that point forward to claim a majority of pledged delegates, which would equate to beating Biden by 35 or 40 percentage points in the remaining states. A delegate plurality would be almost-as-daunting, requiring Sanders win 64 percent of the remaining delegate haul.

Unlike the past couple of times we’ve run these scenarios, I’m not going to bother running any upside cases for Biden because the base case is already so good for him.

But it is worth looking at some Sanders upside cases. Suppose that Sanders beats our projections by a net of 10 percentage points in all contests through March 24, including tonight. So, for example, instead of losing Washington by 3 percentage points, as our forecast shows, he wins it by 7 points. Here’s how the delegate math would work out under this scenario:

Scenario 2: Sanders beats our projections by 10 points

Delegate projections with a net of 10 percentage points added to Sanders’s projected vote share in all contests, according to FiveThirtyEight’s forecast as of 9:00 a.m. Eastern on March 10

Date
State
Biden
Sanders
others




3/10
Michigan
71
54
0


3/10
Washington
40
49
0


3/10
Missouri
40
28
0


3/10
Mississippi
23
13
0


3/10
Idaho
11
9
0


3/10
North Dakota
8
6
0


3/10
Democrats Abroad
6
7
0


3/14
Northern Marianas
3
3
0


3/17
Florida
145
74
0


3/17
Illinois
92
63
0


3/17
Ohio
74
62
0


3/17
Arizona
38
29
0


3/24
Georgia
70
35
0





Period
Biden
Sanders
others





Pre-March 10 total
720
640
139



March 10 total
199
166
0



3/14 – 3/24 total
422
266
0



Grand delegate total through 3/24
1341
1072
139

That is actually … not all that much better for Sanders. He’d still be 269 delegates behind Biden after Georgia finishes voting, and would need to win 64 percent of delegates in the remaining states to claim a majority, or about 60 percent of them for a plurality. So even if polls and projections are 10 points too high on Biden, Sanders would still be a major underdog in the race.

Sanders wouldn’t necessarily win many more states under this scenario either. A 10-point swing toward Sanders would only be enough to flip Washington and Democrats Abroad and tie the projected delegate count in the Northern Mariana Islands. (As an aside, our model doesn’t really know what to do with the territories and is probably underrating Sanders in Democrats Abroad — I’d bet on him winning it even though our model says otherwise. But other than Puerto Rico, the territories don’t have enough delegates to really make a difference.)

So let’s get more creative in helping Sanders. Let’s say Sanders beats his projections by 10 points tonight … except in Michigan, where he beats the forecast by 24 points and narrowly wins the state! (Hey, it happened in 2016!) As a result, Sanders gets a big bounce in the remaining contests over the next two weeks, beating our projections by 20 percentage points in all of them, except in Florida and Arizona where the presence of a large amount of early voting curbs his gains to 15 points. Here’s what that scenario looks like:

Scenario 3: Sanders wins Michigan (!) and surges (!)

Delegate projections with anywhere from a net of 10 to 24 percentage points* added to Sanders’s projected vote share, according to FiveThirtyEight’s forecast as of 9:00 a.m. Eastern on March 10

Date
State
Biden
Sanders
others




3/10
Michigan
59
66
0


3/10
Washington
40
49
0


3/10
Missouri
40
28
0


3/10
Mississippi
23
13
0


3/10
Idaho
11
9
0


3/10
North Dakota
8
6
0


3/10
Democrats Abroad
6
7
0


3/14
Northern Mariana Islands
3
3
0


3/17
Florida
137
82
0


3/17
Illinois
84
71
0


3/17
Ohio
69
67
0


3/17
Arizona
37
30
0


3/24
Georgia
61
44
0





Period
Biden
Sanders
others





Pre-March 10 total
720
640
139



March 10 total
187
178
0



3/14 – 3/24 total
391
297
0



Grand delegate total through 3/24
1298
1115
139

*A net of 24 percentage points added to Sanders’s projected vote share in Michigan, 10 points added in other March 10 contests, 15 points in Florida and Arizona, and 20 points in other March 14 through March 24 contests.

Even with this massive overperformance, Sanders would … still be losing ground to Biden! He’d lose a net of nine delegates tonight and then another 94 in the rest of the states through Georgia, putting him 183 delegates behind Biden after March 24.

Sanders would then need to win 61 percent of the remaining delegates after Georgia to eventually claim a majority of pledged delegates, or 57 percent of them for a plurality. Also, note that even in this massive surge scenario, Sanders might not actually win all that many states, because Biden’s lead exceeds 20 points in states like Florida, Georgia and Ohio, which would give him enough of a buffer to withstand a 20-point Sanders surge.

So while such a race would certainly be interesting — especially with a more favorable set of states for Sanders like Wisconsin set to vote in early April — it might not actually be all that competitive. Instead, it might resemble the 2008 nomination race between Barack Obama and Hillary Clinton, with Biden in Obama’s position. In that race, the candidates traded wins throughout the second half of the primary calendar, leading to massive shifts in the media narrative. But Clinton never came especially close to overcoming the delegate advantage that Obama had built up in February.

How Has RAPTOR Done So Far?

FiveThirtyEight












 












More: Apple Podcasts |
ESPN App |
RSS
| Embed


Embed Code

Hot Takedown recorded a live show last week at the MIT Sloan Sports Analytics Conference. FiveThirtyEight editor-in-chief Nate Silver joined Neil and Sara on stage to discuss how our player rating metric, RAPTOR, has performed so far this NBA season.

Our Rabbit Hole of the Week is inspired by an oddity in NBA rosters.

What we’re looking at this week:

RAPTOR!!
Todd Whitehead of Nylon Calculus investigated optimizing an NBA lineup.
The New York Times puzzled over how Keith Van Horn made $4.3 million long after “retiring” from the game.

Michigan Might Not Be As Sanders-Friendly This Time

“Anybody else getting deja vu?” tweeted Monmouth University polling director Patrick Murray on Monday. He, of course, was referring to the fact that Sen. Bernie Sanders currently trails in Michigan polls by about the same margin as he did in 2016 — before he went on to beat Hillary Clinton there in one of the greatest upsets in polling history.

Sanders’s polling deficit is eerily similar to 2016. Sanders trailed Clinton by 21.3 percentage points in our final polling average of 2016 in Michigan. As of late Monday evening, he’s behind former Vice President Joe Biden by 23.3 percentage points in our polling average.

There’s one big difference, though. In 2016, Sanders’s big polling deficit never made that much sense. Sanders was strong that year with white working-class voters, who are plentiful in Michigan, and contributed to Donald Trump’s victory there in the general election.

And if you look at the states that border Michigan, Sanders performed relatively well in them in 2016, too. He won two of Michigan’s three neighbors, in fact — Wisconsin and Indiana — while losing Ohio. And although the Ohio loss was fairly significant (by 13 percentage points) it was still fairly modest as compared to the margin that polls were showing for Clinton in Michigan four years ago.True, all of those states voted after Michigan in 2016. But Sanders performed well in other Midwestern states that voted before Michigan, including Kansas, Minnesota and Nebraska which he won, and Iowa, where he essentially tied Clinton. (One important caveat: all of those states had held caucuses in 2016 in contrast to Michigan’s primary, and Sanders generally did better in caucus-states in 2016.)

This year, though, it’s not so clear that working-class white voters are a strength for Sanders. Instead, as Nate Cohn of The New York Times’s The Upshot points out, they might be more of a Biden strength, or at least this was true on Super Tuesday. Biden won Minnesota, the only Midwestern state on the docket last week (it switched from a caucus to a primary this year). He also won states in the inland South, such as Tennessee and Oklahoma, that have demographics that somewhat resemble Michigan’s (i.e., a mix of working-class whites and some African Americans). Biden also won Maine, which has one of the highest percentages in the country of whites without a college degree.

And Biden generally performed well in the rural parts of these states. He won every county in Oklahoma and all but a few of them in Minnesota. In Maine, Biden’s upset win came because of rural areas; he lost to Sanders in Portland’s Cumberland County and in some of the more well-to-do areas along the coast.

Here’s another way to look at things. The FiveThirtyEight primary model calculates a “fundamentals” forecast in each state based on a combination of demographics and geography. The demographic calculation is based on both polls of other states and previous results,1 accounting for factors like the racial composition of each state, how liberal or conservative it is, the socioeconomic status of its voters, and how rural it is. By geography, I mean accounting for home-state and home-region effects. These home-region factors aren’t terribly relevant in Michigan since neither Sanders or Biden are really2 from the Midwest.

Here is Biden’s demographic forecast in each remaining state as of late Monday evening. For reference sake, I’ve also listed Biden’s current margin in our national polling average.

Michigan is demographically favorable for Biden

Biden’s projected margin over Sanders based on demographic and geographic “fundamentals,” per FiveThirtyEight primary model as of 8:30 p.m. EST on March 9

State
Projected Biden margin of victory




Delaware
59.5%

–



Mississippi
47.4

–



Louisiana
40.1

–



Georgia
35.2

–



Maryland
34.6

–



Florida
31.1

–



Pennsylvania
30.3

–



Michigan
22.9

–



New Jersey
22.6

–



Missouri
22.6

–



Illinois
21.4

–



Ohio
21.0

–



Montana
19.7

–



Kentucky
19.5

–



West Virginia
17.9

–



South Dakota
17.7

–



Indiana
17.1

–



New York
16.5

–



Connecticut
14.8

–



Arizona
14.6

–



Nebraska
14.2

–



North Dakota
14.2

–



Wisconsin
13.5

–



Kansas
13.4

–



Alaska
11.6

–



New Mexico
10.4

–



Idaho
10.2

–



Wyoming
9.3

–



Oregon
9.0

–



Hawaii
8.6

–



Washington
5.5

–



Rhode Island
4.5

–

The demographic forecast puts Biden up by 22.9 percentage points in Michigan, which is noteworthy for two reasons:

It’s very similar to the polling average there (23.3 percentage points).
It suggests that Michigan is, if anything, a slightly above-average state for Biden. His projected margin of victory in Michigan is slightly wider than his lead in national polls (18.3 points), and Michigan ranks as the 8th-best state for Biden based on demographics of the 32 states that have yet to vote.

Does that mean Biden is completely out of the woods in Michigan? Well, who knows. Even a 2 percent chance (what our model gives to Sanders as of late Monday) is something; we take our 2 percent chances seriously here at FiveThirtyEight. For a more complete rundown of the polling issues in Michigan and what pollsters say has changed this year, I’d recommend this article by Steven Shepard at Politico.

But nothing about the polling in Michigan is terribly out of line with what we’ve seen in national polls or in polls of other states (which have been uniformly quite bad for Sanders over the past few days). Nor is it out of line with the results on Super Tuesday. Sanders’s strengths this year have been concentrated among younger progressives and among Hispanics, neither of who are plentiful in great numbers in Michigan. And the places that most closely resemble Michigan were strong areas for Biden last week, so tomorrow might not be the start of the comeback Sanders needs.