Sim 538 - Best site for US Election Predictions 2020

This is the only site with all 2020 US Election Predictions

EVALUATION OF ELECTION FORECASTS

Status of Elections

Nov 13: CNN has projected winners for all state, resulting in 306 electoral votes for Joe Biden and 232 electoral votes for Donald Trump. Some news networks may hold back on Georgia and North Carolina, as the difference between the candidates is less than 0.5%. Also, Georgia is starting a recount, which is expected to be complete by midnight on November 18. Georgia will certify the results on November 20. The recount is done as a precaution, and not in response to any allegation of fraud or improper ballots being counted.

Click the map to create your own at 270toWin.com

Is it over?

This was an incredible election! Congratulations to all election forecasters. Many warned us that this election would come down to the wire. As early results came in, Trump had multiple pathways to his re-election. Florida quickly added 29 electoral votes to Trump's total, and the "Blue Wall" wasn't there to save Biden. But the map was slowly changing. it looked like Biden might win in Michigan and possibly in Wisconsin although the margin was very narrow. Trump's lead in North Carolina held. So, Biden could win with either Arizona and Nevada or Pennsylvania.

Many forecasters that we track made great predictions. Mark Zandi of Moody's Analytics gave us a great heads up on the October 27 update podcast that Pennsylvania, Wisconsin and Michigan were critical to Biden's victory. Dr. Jim Campbell warned us a month earlier to get our stomach acid medication ready in his predictions of September 17, 2020. I remember a network interview with Larry Sabato around September where he was reluctant to say who was going to win. His website at that time had about 6 states labeled as toss-ups.

Back in January 2020 (yes, pre-impeachment, pre-Covid-19, with the primaries still going on), Skyler Dale used a "learning model" and correctly predicted this was going to be a very close election. His one forecast was incredibly accurate. One strong basis of his forecast is the swing states of 2016 were still swing states.

I think Dr. Campbell was so prophetic as he first gave a brief review of the history of election forecasting, then wrote, "There is history, and then there is 2020." Sage words from a veteran forecaster. Dr. Jim Campbell candor and humor came through as in his September 17 presentation.

With such a close election, our overheated political climate and the controversies sure to follow the additional adoptions of mail-in balloting as well as the many highly anticipated campaign events to come, the whole nation may be on blood pressure medication before this is over. .. Do they make Maalox in red and blue?

Note: Maalox is popular stomach acid medicine. Dr. Campbell's prediction is given under National forecasts. Just to give a little hint - his estimate on popular vote is closest to the current 50.6% for Biden.

Mark Zandi and Bernard Yaros of Moody's Analytics also made excellent predictions, particularly with their particular focus on Pennsylvania as the tipping point state. Even those who did not accept their conclusions, they did an excellent job of explaining everything including where some of us could find all the "gory details."

Nate Silver and his team at Five Thirty Eight provided "tipping point" simulations to determine the most likely tipping point state, and surprise, Pennsylvania came out on top with a 36% chance. Trump campaign team may have scoff at the forecasts along with the media reports, but they were listening as both Biden and Trump hurried to Pennsylvania on the last days before election.

The polling errors made in 2016 gave the impression that somehow, despite their best intentions, election forecasters were biased to the Democratic party. This time they correctly predicted Joe Biden would win, but they were overly optimistic on the margins of victory in almost every swing state. Trump early on appeared to take the lead in many states, and this changed as mail-in ballots were very slowly counted. This spawned additional rumors that somehow, our manual vote counters were doing something suspicious or fraudulent. Pollsters have two more years before the midterm elections to improve their survey methods.

Thus, red mirage was real and we saw it live. And what's more, the networks did show remarkable restraint in not announcing a winner on election night. As Nate Silver, of Five Thirty Eight was quick to point out, there was red shift also, as some states will count the received mail-in ballots prior to election day, giving Democrats a leg up. A quote from the Philadelphia Inquirer in January 2020, "'This is the new normal,' Edward B. Foley, the Ohio State University law professor who first identified the blue shift in a 2013 paper, said in an interview last week. 'This is how we conduct elections now, in a world with absentee voting on demand. … We just have to be patient, let the system work, let the ballots be counted, and whoever wins the authentic votes is the winner.'" From research seven years ago coincides with what just happen - now that fits Hamming's insight. See link at the bottom of page.

Evaluation of Forecasts

Our collection of forecasts suddenly grew about 30% on October 15, 2020 as the American Political Science and Politics held their 2020 Election Symposium which added 11 more forecasts by university professors to our list of approximately 17 forecasts at the time. We are grateful to Dr. Jared Farley an associate professor of political science at the New Mexico Military Institute in Roswell, NM, who provided links to many of the quantitative models and the APSA Symposium on the 2020 election predictions.

I also note the free access to these papers will end by November 30, 2020. These articles show there is much more to be gained from forecasting than just coming up with a few numbers, and in particular, how do candidates succeed or fail in becoming elected. The APSA is being very generous to share the papers for free.

I've subdivided the forecasts into these four groups

State-by-State Predictions (quantitative and polling-based methods)

National Forecasts

Non-numercal method (Professor Lichtman)

Betting/futures websites.

Table 1 ranks forecasts by based on the electoral votes absolute difference for Joe Biden. This seems more appropriate than to compare forecasts on popular vote, because the popular vote does not determine the outcome of our elections. The far left hand column shows the sum of EV's incorrect. Due to cancelling out of errors, Plural Vote with an absolute error of 7, has a sum of errors equal to 76, the highest sum of errors in the list. This shows the inadequacy of examining just the overall score for ranking forecasts.

Table 1: Comparison of Forecasts based on total EV

Rank	Model	Biden's EV	Abs Diff	States Incorrect	Sum of EV Incorrect
1	Predictit (Betting site)	305	1	GA, NC	31
2	Election Projection	307	1	AZ, GA	44
3	Bruno Jérôme et al, Tough Victory for Biden, University of Paris, Montreal University	308	2	AZ, FL,GA	56
4	Plural Vote	313	7	AZ,FL,GA,PA	76
5	Alan Abramowitz U of Virginia ** Paper	319	13	National forecast
6	Sabato's Crystal Ball	321	15	NC	15
7	Enns and Lagodny, Presidential Approval/ State Economy Model, Cornell University *	290	16	GA	16
8	Polly Vote Macromedia University * Symposium Paper	330	24	AZ, GA,FL,PA	76
9	Moody's Model	279	27	AZ, GA, PA	47
10	Skyler Dale, Medium website	279	27	AZ, GA	27

The second table shown below is based on the sum of incorrect EV's. Some poll-based forecasters categorize states as toss-ups, due to the lack of definitive poll data. My scoring penalize these partial forecasts, by adding to their score, a penalty equal to one half of their toss up vote count. This was done to make partial and full forecasts.comparable. I am presuming that half of their toss-up votes would be wrong. Also, on the comparison list, I do not list any forecast with toss-ups greater than 40 electoral votes. Minor discrepancies may be apparent because I did not include the two Congressional Districts in Maine and Nebraska which voted opposite party of the state.

Table 2: Comparison of State-by-State Forecasts

Won by Democrats: AZ(11), GA(16), PA(20) and won by Republicans NC(15), FL(29), IA(6), OH(18), TX(38)

Rank	Score	Model	EV	Toss-ups	EV Error	Errors in States that went Democrat	Errors in States that went Republican	Model Category
1	15	Sabato's Crystal Ball	321	0	15		NC	Poll-based
2	16	Enns and Lagodny, Presidential Approval/ State Economy Model, Cornell University *	290	0	(16)	GA		Quantitative
3	27	Election Projection	307	0	1	AZ , GA		Poll based
4	27	Skyler Dale, Medium website	279	0	(27)	AZ, GA		Quantitative
5	31	Predictit (Betting site)	305	0	(1)	GA	NC	Betting
6	44	Five Thirty Eight	348	0	42		FL, NC	Simulation/Poll-based
7	44	Desart, Long range state level forecast, Utah Valley Univ *	350	0	44		FL, NC	Quantitative
8,9,10	44	Real Clear Politics (no toss-up map) & JHK Forecasts & Politico	350	0	44		FL, NC	Poll-based
11	47	Moody's Model 1: Average Dem turnout (Base Case)	279	0	(27)	AZ , GA , PA		Quantitative
12	52	Princeton Election Consortium	335	16	44		FL, NC	Simulation/Poll-based
13	56	Bruno Jérôme et al, Tough Victory for Biden, University of Paris, Montreal University	308	0	2	AZ,GA	FL	Quantitative
14	59	Electoral-vote	356	18	50		FL , NC , IA	Poll-based
15	60	Sim 538	334	0	28	GA	FL , NC	Poll Based
16	64	Economist	334	40	28		FL , NC	Poll-based
17	64	Polly Vote Macromedia University * Symposium Paper	330	0	24	AZ,GA	FL, NC	Combined Models
18	76	Plural Vote	313	0	7	AZ , GA , PA	FL	Poll-based
19	82	Our Progress	389	0	83		FL, NC, TX	Poll-based
20	106	Leantossup	384	0	78		FL , IA, NC , OH, TX	Poll-based

Comments

A quick scan shows forecasting errors occurred in mostly four states: Arizona, Georgia, Florida and North Carolina won. In addition, two models misjudged Pennsylvania as a perspective Republican win. The two sites that predicted Biden could win over 380 votes, included Texas as a Democratic win. Texas has 38 electoral votes.

Out of 20 forecasts, North Carolina and Florida won by Trump, were each misjudged in 12 out of 20 times, Georgia won by Biden was misjudged 9 out of 20 times and Arizona was misjudged 7 out of 20 times.

Some forecasters dodged the tough calls. These forecasters might claim they called every state correctly, but they they did so, by categorizing the hard to predict states as toss-ups. By my way of my scoring, they paid a penalty of half their toss-up votes.

Many of the names and organizations on this list are very familiar to academics and professionals in political forecasting. However, Skyler Dale, a data scientist, writing for Medium.com is not. He also writes blogs about the Panda software, which I find very interesting. What is even more amazing, is he posted his forecast in January 2020. Yes, this was before the trial of impeachment of Donald Trump, pre-Covid-19, and before we knew who was running against Donald Trump. His forecast is ranked fourth, as he missed on only two states, Arizona and Georgia. His "training model" apparently worked well in identifying the most important variables. Like Moody's model, a critical variable is the voting share of the the parties won in the prior election year on a statewide basis.

Larry Sabato ranks #1 with his team, J. Miles Coleman and Kyle Kondik with only North Carolina being forecasted wrong. But listening to their video their aim is to interpret the data so they get estimates of the state's popular vote on all states. They talk about getting more granular, to understand pretty much into the heads of the voters in various communities. This is very much aligned to the presentations from Moody's. Excellent documentation of the forecasts are provided on many of the websites, including JHK Forecasts, Economist and Moody's.

National Forecasts

Dr. Jim Campbell's forecast of Joe Biden gaining from 50.3 to 50.8% of the popular vote came closest to Joe Biden's 50.6%. He is a professor of political science at the University of Buffalo, author of four books and active in forecasting elections with the APSA for years. See link: Political Economic Model and Dr. Campbell biography from University of Buffalo. .Also see the 2020 APSA review of his forecasts on their symposium site: . APSA Symposium website with free access until the end of November 2020.

Based on a regression analysis of electoral votes for an incumbent president verses their net approval rating, Dr. Abramowitz calculated Biden would win 319 electoral votes. See links Alan Abramowitz U of Virginia Paper

Allan Lichtman, 13 Keys to the White House

The only non-numerical model "The Keys to the White House" worked again. He explained well exactly his reasons for each of the 7 negative answers, and predicted victory for Joe Biden. See Allan Litchman, Keys to the White House . Also, I liked Wikipedia's summary which confirms the success of the method since 1982.

Betting Sites

The betting site, Predictit website is ranked #5, missing just Georgia and North Carolina in Table 2. Just prior to the election, the website was constantly changing as Georgia, North Carolina and Arizona were very even bets. Their prediction went from 305 to 290 to 279 electoral votes in the three days before election time. This was just a difference of 3 states, so the betters as a group were very good forecasters.

See: https://www.270towin.com/maps/predictit-2020-presidential-election-odds .

The Real Clear Politics summarized the odds presented from overseas betting sites. The odds of Joe Biden becoming elected was nearly equal on September 3. From this point forward, Biden odds of being elected increased. By election day, the odds of being elected was 63.8%. This probability is lower than many of the forecasters were predicting at this time, so in general the overseas betters and the Predictit participants were less hopeful of a Biden win than those who simulated the results, such as Five Thirty Eight or Princeton Election Consortium.

Final Comments

I tried to organize forecasts as "poll based" and "non poll based" and this labeling really belies the integrated approach taken by many of the websites. I hope to find a more appropriate means of organizing forecasts before our next presidential election.

The documentation on many forecasts was excellent. The question of what ultimately governs our election is unresolved. Incumbents have an above average chance of winning re-election. Approval polls for Trump showed little change from the start of his administration to the end. Party loyalty was, I believe, a large part of the unchanging ratings. Our economic downturn is a result of Covid-19. The extent to which this was a factor in the election will be subject to extensive review.

There is much to be learned from 2020. I wonder if the solid blue and red designations will hold firm in future elections. There seems little changed from 2016 in this regard. I also wonder if the particular list of swing states will hold through to the 2022 midterms. There is much less participation of the electorate in congressional elections.

I also wonder if the extremely tight elections in seven states is the future of elections to come. Each candidate is looking for vulnerabilities in their opponent in the most closely divided states. I see elections like a tug of war, where each side rushes to add more people to their side with the slighest shift in the rope. As a result, the flag in the center hardly budges. It's a little bit more complicated, but the efforts by each side to increase support in reaction to weakness seen on the other side may be netting out. The tug of war mechanism was very evident in the last minute travel of both candidates within a few days of the election.

There is a swing state list, which by now has become very familar. Even though Florida was won by 3.4% by Donald Trump, I am certain with its 29 electoral votes, it will continue to be on the list. Obviously Georgia, Arizona, North Carolina, Pennsylvania, Wisconsin and Michigan are on the list and will be thoroughly studied by Democrat and Republican strategists in preparation for the 2022 midterm elections.

I'm from Miami, and perhaps we need a sign as people drive into Florida, "Welcome to Florida, where nobody can figure out what we're thinking."

David Lord

Updated Nov 14, 2020

Links:

There are a number of very good links, which I've overlooked. Larry Sabato, Director of the Center for Politics at the University of Virginia, has put together a series of Youtube videos, with the latest listed below. I particularly recommend #13.

Sabato's Crystal Ball American Votes

Comments on Quantitative Methods

Comments on Citizen Model (Oct 25, 2020)

Abramowitz equation: Could it be this simple?

Lichtman's 13 Keys to the White House

Moody's Analytics Forecasts

Primary Model

Comments on Don Luskin's Model (October 6, 2020)

US 2020 Presidential Election Predictions

This is the only site with all 2020 US Election Predictions