With the 2018 midterm elections behind us, and the field of Democrats seeking to challenge Republican President Donald Trump in 2020 taking shape, let us turn to the 2020 presidential elections, scheduled to take place (technically, to conclude) on November 3, 2020.
- Calculating each candidate’s polling average (WAPA; weighted by time to election and pollster rating), both nationally and within any state with available 2019 polling.
- Calculating a single average (NSW-WAPA) for each candidate by weighting their WAPA in early primary/caucus states (Iowa, New Hampshire, Nevada, South Carolina) more than in later states, in turn weighted more than national polls.
I similarly approach the 2020 presidential election: mixing “fundamentals” with public polls released since January 1, 2019 asking voters to choose between a named Democratic candidate and Trump.” “Fundamentals” are simply expected margins between the Democratic and Republican presidential nominees in a state (the Electoral College [EC] elects the president of the United States) based upon my 3W-RDM measure (weighted average arithmetic difference between state and national Democratic margins in the last three presidential elections). This measure is not perfect: it has “missed” (in either direction) by an average of 5.4 percentage points (“points”) over the last six presidential elections, with 67 misses (22%) higher than 5.4%.
Nonetheless, “fundamentals” (3W-RDM plus hypothesized 2020 Democratic national popular vote margin) provide a useful baseline against which to assess wicked early polling data. This allows me to avoid such “macro-level” indicators as the state of the economy or Trump’s approval ratings advocated by political scientists in (solid) volumes like this:
Let me begin with the metaphorical (literal?) elephant in the room: incumbent presidents who seek reelection win about two-thirds of the time, according to Yale political science professor David R. Mayhew.
All else being equal, then, Democrats have an uphill battle to defeat Trump in 2020. However, all is not equal, as both “fundamentals” and polling suggest. To calculate the “fundamentals” for each state (and District of Columbia [DC]), one needs
- That state’s 3W-RDM and
- The Democratic margin in the national popular vote.
While #2 is difficult to estimate more than 18 months in advance, there may be historic guidance.
Just bear with me while I review 60 years of presidential election results. As usual, election data come Dave Liep’s indispensable Atlas of U.S. Presidential Elections.
Over the last 10 presidential elections (1980-2016), the Democratic presidential nominee has lost the national popular vote by an average of 1.0 point. However, because Democrats have fared better in presidential elections since 1992, perhaps more weight should be given to more recent elections. One way to do this is to weigh the 1980 margin “1,” the 1984 margin “2” and so forth through a weight of “10” for 2016; this yields a Democratic edge of 1.7 points. This is broadly similar to taking the mean of the 10- , 5- and 3-election averages (D+1.9). Even better for Democrats is a weighted average of the the last five presidential elections (D+2.9), not surprising given that Democrats have won the national popular vote in four of those elections (while only winning the EC twice).
Based on recent electoral history, then, a Democratic national popular vote margin of between 1.5 and 3.0 points is certainly plausible; the 2.1 points by which Democrat Hillary Clinton beat Trump in 2016—an election with no incumbent seeking reelection, which Mayhew found were 50-50—falls squarely in this range.
But incumbency does need to be taken into account. In the last 10 presidential elections in which an incumbent president election sought reelection (including 1964 and 1976, when elevated Vice Presidents Lyndon Johnson and Gerald Ford, respectively, first sought election in their own right), that incumbent won by an average of 7.7 points. Once you exclude the landslide reelections of Dwight Eisenhower (1956, 15.4 points), Johnson (22.6 points) and Richard Nixon (1972, 23.2 points), however, the average drops to just 2.2 points. And that latter figure drops to a loss of 0.5 points if you exclude Ronald Reagan’s 1984 reelection landslide (18.2 points).
An even better way to assess the effect of incumbency, though, is to compare initial and reelection margins. Starting with Eisenhower in 1956, there have been eight such election pairs:
Table 1. Presidential margins, initial and reelection, 1952-2012
|Year||% Margin||Year||% Margin||R – I|
|George H. W. Bush||1988||7.7%||1992||-5.6%||-13.3%|
|George W. Bush||2000||-0.5%||2004||2.5%||3.0%|
|Average, GOP only||5.7%||10.7%||5.0%|
Since 1956 (Table 1), elected presidents won reelection by an average 1.6 points more than they won initially; this would put Trump at just 0.5 points down in 2020. However, when you examine only the five Republican presidents who sought reelection over this time span (Eisenhower, Nixon, Reagan, Bush 41, Bush 43), the margin increase jumps to 5.0 points; this would put Trump 2.9 points ahead in 2020. That said, the last three elected presidents to seek reelection actually saw an average drop of 0.8 points in their margins; the decline is similar (-0.6 points) for the last three Republican presidents to seek reelection (Reagan, Bush 41, Bush 43).
In other words, a longer analytic time frame implies Trump would win the 2020 national popular vote by 1.0 to 7.7 points, while a shorter analytic time frame implies a Democratic win of 1.7 to 2.9 points—a range of R+7.7 to D+2.9.
Well, that was not very helpful, was it?
Actually, this range provides a baseline against which to assess 2020 presidential election polls asking respondents to choose between a named Democrat and Trump.
Table 2: “Fundamentals”-based 2020 state-level Democratic margins
|TOTAL DEM EV||182||191||245||259||279||279|
1 DC, HI, VT, CA, MD, MA, NY, RI, IL, CT, DE, WA, NJ
2 TX, SC, MO, IN, MS, MT, AK, LA, KS, NE, SD, TN, AR, AL, KY, ND, UT, ID, WV, OK, WY
Table 2 shows how EV would hypothetically be distributed across states under scenarios ranging from Trump winning the national popular vote by 7.7 points to the Democratic nominee winning it by 2.2 points. Boldfaced results are expected Democratic wins, with italicized results reflecting uncertainty based on the average 3W-RD miss of +/-5.4 points.
What jumps out are the eight states comprising 78 EV with 3W-RDM between D+2.2 and D-0.4 (MI, CO, NV, MN, VA, WI, NH, PA). This range includes the states that effectively cost Clinton the presidency in 2016: MI, WI, PA. All but Pennsylvania lean slightly toward the Democrats, but even a relatively narrow 2.3-point Trump win in the national popular vote (well within the historic range) could easily put all 78 EV in the Republican column, giving Trump a solid 347-191 EC win. However, because there is a 3.0-point jump from Pennsylvania (truly “The Keystone State”) to Florida in 3W-RDM, a similar 2.2-point national popular vote win for the 2020 Democratic presidential nominee would be expected to yield only a narrow 279-259 EC win. In fact, the 2020 Democratic presidential nominee winning Florida (and the EC 308-230) would seem to require winning the national popular vote by at least 3.5 points. That same nominee winning 347 EV would similarly require winning the national popular vote by at least 6.0 points, adding Ohio, Iowa and North Carolina to her/his column.
In other words, there is clear asymmetry in the distribution of EV that works to the Republicans’ advantage: they can win solid EC victories with smaller national popular vote margins.
Of course, this analysis is purely hypothetical, even before re-emphasizing every state-level Democratic margin in Table 2 is the center of a 10.8-point range of possible outcomes (with roughly 80% certainty). Thus, a Trump win by 2.9 points nationally could result in anything from 259 to 356 EV for Trump, while that same margin for the Democratic nominee could yield anything from 191 to 347 EV.
This broad uncertainty about the outcome in 2020, with perhaps a slight historical edge to Trump (though nothing like being a 2-1 favorite, per Mayhew), is why we now turn to 2020 presidential polling.
Since January 1, 2019, pollsters have tested national match-ups between Trump and 16 announced (or soon-to-announce) candidates for the 2020 Democratic presidential nomination:
- Former Vice President Joe Biden (15 times)
- California Senator Kamala Harris (14)
- Vermont Senator Bernie Sanders (14)
- Massachusetts Senator Elizabeth Warren (13)
- Former Texas U.S. Representative Beto O’Rourke (10)
- New Jersey Senator Cory Booker (7)
- South Bend, Indiana Mayor Pete Buttigieg (4)
- New York Senator Kirsten Gillibrand (4)
- Minnesota Senator Amy Klobuchar (3)
- Former Secretary of Housing and Urban Development Julián Castro (2)
- Maryland U.S. Representative John Delaney (2)
- Hawaii U.S. Representative Tulsi Gabbard (2)
- Former Colorado Governor John Hickenlooper (2)
- Washington Governor Jay Inslee (2)
- Author Marianne Williamson (2) and
- Entrepreneur Andrew Yang (2)
Five other national polls matched Trump against former New York City Mayor Michael Bloomberg, Ohio Senator Sherrod Brown, New York U.S. Representative Alexandria Ocasio-Cortez, House Speaker Nancy Pelosi (D-CA) and Senate Minority Leader Chuck Schumer (D-NY), none of whom are running for president in 2020; Representative Ocasio-Cortez, who was born in 1989, would actually be too young to serve as president on January 20, 2021.
Not surprisingly, the number of polls conducted with each candidate closely matches her/his current standing for the Democratic nomination, based on NSW-WAPA; Table 3 below shows these values as of April 20, 2019. To calculate how each candidate currently fares against Trump in the 2020 national popular vote, I used the procedure detailed here. Essentially, I weighted “unbiased” polling margins by date-to-election and pollster rating two ways: 1) treating all polls (even from the same pollster) as statistically independent and 2) treating polls from the same pollster as a single “average.” WAPA is the average value from the two methods.
Table 3: Current WAPA margins for hypothetical 2020 matchups between Donald Trump and selected Democratic presidential nominees
|Democratic nominee||Nomination polls1||Margin vs. Trump|
|Margin weighted by nomination polls||3.1%|
|Margin weighted by nomination polls (w/o Biden)||1.3%|
1 Not listed are Ohio U.S. Representative Tim Ryan (0.1%), California U.S. Representative Eric Swalwell (0.1%) and Miramar, Florida Mayor Wayne Messam (0.01%)
2 The weighted average of the five non-candidate matchups
As of April 2019, Biden—with near-universal name recognition—fares best against Trump in 2020, beating him by nearly eight points, which translates to around 347 EV (and perhaps more if he wins Georgia’s and Arizona’s 27 combined EV as well). More surprising, though, is Sanders’ solid performance against Trump—beating him by 3.7 points in the national popular vote (and likely winning Florida for 308 EV).
Moreover, both Biden’s and Sanders’ prospective margins are well outside the historic range computed above. The next tier of candidates—Harris, Warren, O’Rourke, Booker and Buttigieg—essentially tie with Trump, averaging a -0.1% loss in the national popular vote; Gillibrand also would essentially tie Trump (-0.2%). Of this group, Harris, O’Rourke and Booker fare best, beating Trump in the national popular vote by between 1.0-1.6 points and eking out an EC win, more likely than not. The other candidate with at least three public head-to-head matchups, Klobuchar, for now fares as poorly as fellow Midwesterner Buttigieg, losing the national popular by just over five points.
I would not read too much into polls showing Yang, Gabbard, Castro, Hickenlooper, Delaney, Inslee and Williamson losing the 2020 national popular vote to Trump by between 7.5 and 11.6 points. Both sets of polls were conducted by HarrisX, a firm rated “C+” by FiveThirtyEight.com and for whom no “skew” has yet been calculated. They may be affiliated with Harris Interactive, another C+-rated pollster with a significant pro-Republican (mathematical) bias of 1.5 points.
Nonetheless, it is noteworthy (and likely reflecting the strong Democratic desire to defeat Trump in 2020) that the better a candidate is doing in the nomination polls (i.e., NSW-WAPA), the better that candidate currently fares against Trump; the correlation between the two values is 0.84.
But most interesting of all is that a possible proxy for “generic Democrat” (weighted-adjusted average of the five candidates not running for president) would theoretically beat Trump in 2020 by just over four points (putting Florida, Iowa, Ohio and North Carolina in play).
To get a sense of how a typical Democrat nominee would fare against Trump in 2020, I calculated a weighted average of the 17 margins listed in Table 3 using each candidate’s NSW-WAPA (a rough proxy for “likelihood of winning the nomination”). By this measure, the typical 2020 Democratic presidential nominee would beat Trump nationally by 3.1 points, just outside the upper end of the range based upon recent electoral history. However, if you remove Biden’s projected 7.9-point win, the margin drops to 1.3 points…roughly how Harris, Warren, O’Rourke and Booker would theoretically fare against Trump. They would still be slightly favored to win, though arguably 164 EV from 14 states (NM to NC in Table 2) would be in play.
But what about the states?
Since January 1, 2019, head-to-head 2020 matchups between a named Democrat and Trump have been conducted in 12 states (AZ, IA, MA, MI, MN, NV, NH, NC, PA, SC, TX, WI; mean 3W-RDM = D-1.9). Table 4 shows how specific Democrats (as well as “typical Democratic nominee”) currently fare against Trump (minimum 2 matchup polls in state), sorted from most- to least-Democratic by 3W-RDM.
1 Emerson College, 4/4-4/7/2019, 761 registered voters
2 O’Rourke: Firehouse Strategies/Optimus, 3/19-3/21/2019, 540 likely voters; Klobuchar: Emerson College, 3/7-3/10/2019, 743 registered voters
3 Emerson College, 3/28-3/30/2019, 719 registered voters; Oth=Buttigieg
4 DFM Research, 2/26-3/3/2019, 719 registered voters; Oth=Klobuchar
5 Emerson College, 3/15-3/17/2019, 775 registered voters; Oth=Klobuchar
6 Emerson College, 2/21-2/22/2019, 910 registered voters
7 Emerson College, 3/28-3/30/2019, 719 registered voters
8 Emerson College, 1/30-2/2/2019, 831 registered voters; Other=Gillibrand; mean of Brown, Pelosi
9 PPP, 1/4-1/7/2019, 750 registered voters
10 OH Predictive Insights, 2/12-2/13/2019, 600 likely voters
11 Quinnipiac University, 2/20-2/25/2019, 1,222 registered voters; Other=Castro
12 Emerson College, 2/28-3/2/2019, 755 registered voters; Other=Klobuchar
Most public 2019 state-level polls have been conducted by Emerson College (B+, D+0.1), who occasionally test “trial heats” when they conduct Democratic (and, less frequently, Republican) nomination polls in a state. While Emerson is a high-quality pollster, it would help to see other high-quality pollsters test these matchups. Also bear in mind that these are primarily polls of “registered voters,” not voters deemed likely to vote; the latter polls often tilt slightly Republican. And, of course, the 2020 U.S. presidential election is still more than 18 months away.
Nonetheless, Table 4 reveals some interesting things. One, it is clear pollsters believe that Biden, Sanders, Harris and Warren (≥14 head-to-head matchups)—in that order—are the likeliest 2020 Democratic presidential nominees, followed by O’Rourke (10 matchups); this perfectly matches the top five using NSW-WAPA. Pollsters asked about Booker, Gillibrand and Klobuchar earlier in the year, but have ceased doing so, reflecting their stagnation in the race.
As they do nationally, both Biden and Sanders perform very well in the states—especially in the key swing states of Michigan, Wisconsin, New Hampshire (one poll has Biden up 10 points), Pennsylvania and, most importantly, Iowa. Removing Biden from the analysis, in fact, lowers hypothetical margins a mean 2.2 points across these states, albeit without changing the state “winner.” Harris, Warren and O’Rourke do not perform as well, but neither do they fare poorly.
Overall, Democrats (using the same weighting method as national polls) are collectively ahead by solid margins in Michigan (3.6), Wisconsin (6.1) and Pennsylvania (5.9)—the states that cost Clinton an EC victory in 2016; strong 2018 Democratic performances in these three states may not have been a fluke. They are winning New Hampshire (D+0.1) by 8.1 points. And they are holding their own in Iowa, a state trending sharply Republican—from D+2.0 after 2012 to D-4.7 after 2016; being behind only 1.4 points could be a terrific sign for Democrats in 2020.
The most striking positives for Democrats in Table 1, however, are being down “only” 3.3 points in Texas and 7.7 points in South Carolina (albeit in a single set of late February polls). First, these averages imply an 8-12-point lead in the national popular vote, which has not happened since Clinton won reelection in 1996 by 8.5 points. Moreover, this could confirm the southeast and southwest are trending Democratic…although the averages from Arizona (D-6.2) and North Carolina (D-5.6) suggest otherwise.
Being collectively 5.6 points down in North Carolina, in fact, is not ideal for Democrats, as it implies essentially a tied race nationally. The worst news for Democrats in Table 4, though, is a set of polls conducted by Emerson College in Nevada in late March. Nevada has been trending steadily more Democratic, so to be collectively down roughly one point there is worrisome, because it implies a national environment where Trump wins by 3.0 points. Across the other 11 states, on average, the Democratic nominee would beat Trump by 6.6 points!
Still, the overall message from Table 4 is that a hypothetical Democratic nominee
- is faring well precisely where (s)he needs to (MI, WI, NH, PA—even IA) and
- leads the 2020 national popular vote by 5.8 points–roughly the margin by which Democrats now lead in the “generic ballot” (e., some variant of “If the election for U.S. House were held today, would you vote for the Democratic or Republican candidate?”).
I cannot emphasize enough the following caveats to this analysis:
- The 2020 presidential election is more than 18 months away.
- National polls only reveal so much.
- State-level polling is limited and mostly conducted by a single (respectable) pollster.
- Most polls are of registered voters, not likely voters, introducing some pro-Democratic (mathematical) bias
- Hypothetical matchups reflect name recognition as much as considered choice.
- I have not controlled for a possible Independent run by Schultz.
That said, signs are good for Democrats right now. The best-known Democratic candidates (and current adjusted polling leaders)—Biden and Sanders—are beating Trump both nationally and in the key states. The next tier of candidates—Harris, Warren and O’Rourke—while less well-known, are also faring relatively well, beating Trump by just over one point on average. As they get better known, that margin could increase (just as the Biden and Sanders margins could shrink).
And while a wider historical frame shows Trump as much as a 2-1 favorite (winning the national popular vote by as much as 7.7 points), recent history favors the Democrats to win the national popular vote by between 1.5 and 2.9 points—though even then their nominee would not be guaranteed an Electoral College victory.
Moreover, current polling suggests Democrats are ahead of even those positive numbers. Using national polls only, a typical Democratic nominee is ahead by 3.3 points in the national popular vote, while state-level polls imply a lead nearly double that: 5.8 points! Removing Biden from the analysis drops these values to 1.8 and 3.6 points, respectively, precisely in line with recent electoral history.
All in all, then, I would rather be the eventual Democratic presidential nominee than Trump in 2020.
In April 2019, anyway.
Until next time…
 Weighted-adjusted polling average
 National-and-state-weighted WAPA
 For now, I ignore a March 8-10, 2019 poll by Change Research which substituted Vice President Mike Pence for Trump. I also ignore (for now) five national polls (and Emerson College polls from Iowa, Michigan, New Hampshire, South Carolina and Wisconsin) that also include former Starbucks CEO Howard Schultz, who may run for president as an Independent.
 Democratic percentage of total vote minus Republican percentage of total vote. Weights for most recent 3W-RDM are 2008=1, 2012=2, 2016=3
 This drives our eldest daughter—a Democrat like her parents—crazy because she LOVES elephants.
 Mayhew, David R. 2008. “Incumbency Advantage in U.S. Presidential Elections: The Historical Record.” Political Science Quarterly 123:2.
 Using the FiveThirtyEight.com “Mean-Reverted Bias” listed in their Pollster Ratings.
 Starting with the three presidential elections from 1984-1992, Nevada’s 3W-RDM are D-8.5, D-6.9, D-5.0, D-2.5, D+2.0, D+3.2, D+2.0