Abstract

1. This paper systematically studies all kinds of election survey data in the United States, such as poll (independent, non-independent, comprehensive, state) data, public disapproval rate, amount of funds raised, bet odds and so on, and points out which are more reliable. to help us predict the election situation in the United States.

2. The first kind of data, in the US opinion poll data, the reliability order is: comprehensive poll > independent poll > non-independent poll. At present, the two most important comprehensive polling institutions are Real ClearPolitics and FiveThreeEight. State-level polls are relatively inaccurate and have no reference significance.

3. The second kind of data, "public disapproval rate" is a relatively important but often ignored data, which has high accuracy and is worthy of attention.

4. The third type of data, "vote with money", bipartisan campaign donations and political betting websites imply the probability of a candidate winning:

1) generally speaking, the more campaign donations raised, the more likely the party will win the election. Bipartisan campaign fundraising can represent public opinion on the one hand and can also be used for propaganda on the other. Only three candidates with low fundraising have won in 10 general elections since 1980.

2) at present, there is no final conclusion as to whether the results predicted by political betting websites or opinion polls are more accurate. It is worth noting that the people involved in political betting are not a representative sample of American voters.

5. In the 2020 election, the current data are very good for Biden: (1) the comprehensive poll data Biden is 7.5 points higher than Trump; (2) Trump has a high public disapproval rate, which is comparable to that of Bush senior and Carter when they sought re-election. Both failed re-election; (3) Biden raised far more money than Trump in August.

6. Will there be another situation of "winning the universal suffrage but losing the general election", that is, the 2016 Clinton election? We don't think the probability is high.

Text

The election in the United States is extremely lively, and in order to "judge" who will be elected, the market, the media, and think tanks all show their special powers and make "predictions" through various opinion polls, odds, and the amount of money raised. As the saying goes, "experts watch the way, laymen watch the hustle and bustle." how can we grasp the election situation in this numerous and complicated election materials?

Our report, from the perspective of popular science, systematically studies various election surveys in the United States and points out how we should use these election data.

We have summarized all kinds of data on the US general election, including comprehensive opinion polls, independent polls, non-independent polls, public disapproval rate, campaign fundraising, political betting and so on. The highest accuracy and reference value is still the national comprehensive opinion poll, followed by independent opinion polls, political betting and public disapproval rate, while the accuracy and reference value of state comprehensive opinion polls and campaign fund-raising are average, and non-independent polls are the lowest.

We believe that various types of data only provide reference for investors and help investors establish a "coordinate system", and the final election result will also be affected by various factors such as unexpected events.

1. Three kinds of reliability order of poll data: comprehensive poll > independent poll > non-independent poll.

1.1. Global opinion polls are relatively reliable, comprehensive polls are the most accurate, and the accuracy is improving.

In a statistical sense, the accuracy of comprehensive opinion polls is higher than that of independent polls and non-independent polls. At present, the two most important comprehensive polling institutions are Real Clear Politics and FiveThreeEight. Academically, the accuracy of pollsters refers to the gap between the predicted results and the actual results, not who wins or loses the final result. The structure of American opinion polls is mainly divided into three categories: independent polls, non-independent polls and comprehensive polls. Independent pollsters are opinion polls conducted independently of parties, independent of party positions, rather than the opposite. Generally speaking, non-independent pollsters are more willing to exaggerate their party's support rate by means of "questioning guidance" and so on. Because of getting rid of party influence, independent polls do not presuppose positions in advance, and the results of independent polls are more accurate and credible than those of non-independent polls.

Integrated polling institutions generally do not really conduct field polls, but further process the data of various polling institutions to get the final polling data. Comprehensive pollsters further process the data of each pollster, for example, Real Clear Politics calculates the arithmetic average of each pollster, while the FiveThreeEight website gives different weights according to the historical performance, party attributes, polling methods and other factors of each pollster, and then carries on the weighted average. In a statistical sense (the law of large numbers), the arithmetic average or weighted average of the polling institutions can offset or eliminate the "noise" in the polling institutions to a certain extent, making the results more accurate.

Historically, the overall accuracy of comprehensive national polls has improved (figure 2). According to a report released by the American Public opinion Research Association (AAPOR), the accuracy of comprehensive national opinion poll data has generally improved since 1936, and the comprehensive poll data in recent elections are very close to the actual support rate of the final candidate. The overall national poll data for the 2016 presidential election are already so accurate by historical standards that the difference between Hillary Clinton and Trump in the final vote is only 2.2% lower than the comprehensive poll data. Since the advent of modern opinion polls in 1936, the level of error in 2016 has been less than half of the average error of the national comprehensive poll data (4.4%), and lower than the average error since 1992 (2.7%).

1.2. The accuracy of state polls is relatively low and there is no significant improvement.

Compared with national polls, the accuracy of state polls is relatively low and has not improved significantly in recent elections, and there is a large error in state polls in 2016 (figure 3), which explains why Hillary Clinton won the universal suffrage but lost the general election. According to a report released by AAPOR, the margin of error of state-level polls in the United States in 2016 was 5.1%, more than any presidential election since 2000. Although the margin of error in swing states is only 3.6% (figure 4), it is still higher than in any presidential election since 2004, and the accuracy of swing states has not improved significantly, which partly explains why pollsters generally predict that Mrs Clinton will win a majority of electoral college votes.

1.3. The accuracy of swing state polls is not significantly improved, but the overall error is controllable.

Swing states are the key to the presidential campaign. "those who have swing states win the world". In the last ten US elections, the party that won the majority of votes in swing states won the general election. For a long time, some states in the United States, affected by their own industrial structure, population structure and other factors, will benefit from the policies of the Democratic Party or the Republican Party for a long time, so they will also vote for their party candidates in the general election. These states are also known as bipartisan iron voting positions. In addition to the iron vote warehouse, there are some states in the United States that do not have a long-term bipartisan preference and will vote according to the specific policy propositions of the two candidates during the general election. These states are called swing states.

There are mainly six major swing states in the United States, with a total of 101 electoral college votes (the presidential candidate wins the general election by 270 votes), which has become a key force in the US general election. According to Real Clear Politics statistics, in nine of the last 10 US elections, the party that won the majority of votes in swing states won the election, lost only once, and it was the controversial 2000 general election between George W. Bush and Al Gore. It can be said that "he who has to swing the state wins the world".

The accuracy of opinion polls in swing states has not improved significantly, and there is even a downward trend (figures 5, 6), which is consistent with the AAPOR report. However, overall, the error of the swing state poll data is still controllable, and the average error of the three presidential elections from 2008 to 2016 is all less than 3%, which still has important reference value. According to RCP poll data, the average poll error of candidates in the six swing states has been increasing in the three presidential elections from 2008 to 2016, overestimating Republican candidates by 0.93% in 2008, overestimating Republican candidates by 1.93% in 2012, and overestimating Democratic candidates by 2.78% in 2016. The error in opinion polls on swing states has also directly led to a far cry from the actual election results in 2016, which show that Hillary Clinton has a big lead in Wisconsin, Michigan and Pennsylvania. But in the end, all of them were narrowly reversed by Trump, and Hillary Clinton was completely defeated in the six major swing states.

two。 An important but often overlooked indicator of whether a president can be re-elected: popular disapproval rate

In addition to the approval rating, there is another important but often ignored poll data when the incumbent president is seeking re-election-the public disapproval rate. When there are only two candidates, if you are particularly opposed to one of them, the other will be elected because of opposition. It can be said that part of the reason for supporting Biden's running for president is that "he is not Trump".

According to the FiveThreeEight website (figure 7), Trump's disapproval rate has been highly consistent with that of George W. Bush and Carter, who both failed to win re-election. The disapproval rates of the other six re-elected presidents were lower than those of Trump. According to Pew Research Center's mid-August poll data (figures 8 and 9), the main reason for supporting Biden for president is that "he is not Trump", accounting for 56%. By contrast, the main reason for supporting Trump is his "leadership and performance", with "he is not Biden" accounting for only 19%.

It is generally believed that when the opposition rate of the incumbent presidential candidate is relatively high, the voter turnout rate will also be higher and will vote for his competitor, because voters who oppose the incumbent president may have long been dissatisfied with the current policy and have long held a grudge. Therefore, he will "vote" the incumbent president in the general election. This was clearly reflected in the Brexit referendum. For example, British fishermen have long been dissatisfied with the fact that other EU countries can fish in the UK's exclusive economic zone, so they had a very high turnout in the Brexit referendum and supported Brexit.

3. Will "vote with money" (that is, campaign fundraising and betting odds) be more accurate than opinion polls?

3.1. The overall performance of bipartisan campaign fundraising is good.

Generally speaking, the more campaign donations raised, the more likely the party will win the election, but there will be exceptions. The campaign fundraising of the two parties, on the one hand, represents that the party is more supported by the public; on the other hand, it can also use political donations to carry out publicity activities such as television advertisements, rallies and speeches to expand the popularity and influence of candidates and call on voters who support their own party to vote. So generally speaking, the more campaign donations are raised, the more likely the party will be to win. On the other hand, political donations do not completely coincide with the wishes of voters, and do not fully represent the wishes of voters, so it is possible that parties with little campaign fund-raising will win the general election. Judging from the 10 general elections in the United States from 1980 to 2016, seven were won by parties with high fund-raising, with three exceptions in 1984, 1996 and 2016 (figure 10).

In this election, Biden raised money all of a sudden, raising $364.5 million in August, breaking the single-month fundraising record for the US presidential campaign and 73.5 per cent higher than that raised by Trump (figure 11).

3.2. The odds of political betting websites are relatively controversial.

The probability of a candidate winning based on the odds of a political betting website is also seen as a prediction of the election result, but it is worth noting that the people who participate in political betting are not a representative sample of American voters. there are differences with ordinary voters in terms of risk preference.

At present, there is no final conclusion in academic circles as to who is more accurate in political betting or opinion polls. Political bets on American elections have appeared since the founding of the United States. According to the research results of Paul W. Rhode,2004 and others, among the 15 presidential elections from 1884 to 1940, only 1916 political bets were wrong, the remaining 11 were successful, and the odds of candidates in the remaining three general elections from 1884 to 1892 were almost the same, which did not constitute a prediction. After 1940, the US government stepped up its crackdown on the gambling industry, and political betting disappeared from public view until 1988. According to the research results of Robert S. Erikson,2008 and others, the accuracy of political betting in the prediction of the five general elections from 1988 to 2004 was higher than that of opinion polls, but then some scholars questioned the research methods of Eriksson and others. After adjusting their research methods, they found that the polls from 1988 to 2004 were more accurate, and this conclusion was also applicable to the general elections from 2004 to 2016.[1]

In the four general elections from 2004 to 2016, political betting and opinion polls successfully predicted the first three elections, but failed to predict the general election in 2016. But the 2016 poll results are closer to the actual voting results, and Hillary Clinton's popular vote rate is indeed 2.1% higher than Trump's, while political betting websites had predicted that Clinton's victory rate was as high as 88% (figure 12).

4. Will the story of "winning universal suffrage but losing the general election" be repeated in 2020?

After discussing the value of these poll data, for now, Biden does have a better chance of winning, but will he win the general election and lose the general election again? On the whole, we don't think it's possible this time.

Historically, there have been only five cases in which a US presidential candidate won the general election but lost the general election (figure 14), accounting for 11.1% of all previous presidential elections in the United States. The five events are Trump's victory over Hillary Clinton in 2016, George W. Bush's victory over Gore in 2000, Harrison's victory over Cleveland in 1888, Hayes's victory over Tilden in 1876, and lower Adams's victory over Jackson in 1824.

The main reason for winning the universal suffrage but losing the general election is the "electoral college system" in the presidential election of the United States. First of all, the presidential electors of the state will be elected by the voters of each state, and the winner-take-all will be implemented, that is, the winner of the state will win all the electoral college votes of the state (except individual states), and then the president will be elected by the electoral college of the states, and the candidates who get more than half of the votes (270 votes) will win the election. However, because the electoral college votes of the states do not exactly match the actual population of the states, it is possible that the presidential candidate wins the universal suffrage but loses the general election, but the probability is not high from the previous presidential elections in the United States.

Opinion poll data are affected by many factors, such as concealing true ideas, swinging voters, sample selection, convergence effectiveness and many other factors, and subsequent political events may change the balance of power between the two sides of the campaign in a short time. After the "Black Swan incident" broke in the 2016 election in the United States, polling institutions and research institutions began to rethink the accuracy and shortcomings of the poll data.

The concealment of true thoughts by respondents is considered to be an important reason for the distortion of the 2016 election poll data, and this factor may also affect the accuracy of the 2020 election poll. Under the pressure of "political correctness" such as immigration and racial equality, many Trump supporters have a "shame phenomenon of supporting Trump" (Shy Trump Effect). According to the Pew Research Center (figure 15), 4.4% of Trump's final voters came from those who had previously expressed support for other candidates or refused to take a stand, compared with 2.6% for Hillary Clinton. Many respondents did not express their true wishes when they were surveyed, resulting in the final distortion of the poll data. The shameful phenomenon of supporting Trump is likely to continue in the 2020 election. According to the Cato Institute, 77% of conservatives, 53% of liberals and 64% of moderates say they are no longer free to express their political views. This will have a greater impact on the polls.

Swing voters are also an important factor in the failure of opinion polls in 2016 (figure 16), but the proportion of swing voters in 2020 is much lower than it was four years ago. There are still a large number of swing voters who do not know who to vote for in the two months before the 2016 election, a phenomenon that is more pronounced in the three swing states of Michigan, Wisconsin and Pennsylvania, with about 13% to 15% of voters deciding who to vote for in the last week. Post-election polls show that a majority of swing voters in these states voted for Trump. According to a 2020 report by the US media Newsweek, the proportion of swing voters in August 2016 was about 20 per cent (inconsistent with the proportion of swing voters in figure 16, possibly due to different standards), but it fell to 10 per cent in 2020.[2]

The choice of poll samples will also affect the results of the polls. In 2016, the samples of opinion polls are underrepresented to those with low education level, and in 2020, the representativeness of the samples of opinion polls has been enhanced. The main problem in the 2016 poll was to underestimate Trump's support among poorly educated whites, and the problem was more pronounced in state polls, leading to a big deviation in the final state polls. According to a report by the New York Times in May 2020, only about 20% of swing state polls adjusted the proportion of people with different educational levels in the sample in 2016, rising to about 50% in 2020.[3]

Some of the polling institutions will be affected by the media and public opinion, and there is a certain degree of convergence effect. When the media and most other pollsters believe that Hillary Clinton's election as president is a high-probability event, few pollsters dare to take the risk of being an enemy of the whole market to think that Trump is likely to be elected president. Therefore, in the sample selection, question setting and other aspects may be more inclined to Hillary Clinton, resulting in the final poll results convergence. After the turnaround in 2016, both the media and pollsters have become more cautious in this regard in 2020.

Political emergencies could also change voters' attitudes towards the two candidates in a short period of time, such as when the FBI reopened its investigation into Mrs Clinton's "email scandal" about two weeks before the 2016 election. Mrs Clinton's lead has shrunk sharply since then (figure 17), from 4.6 per cent on October 28th to 1.3 per cent on November 3. With reference to the experience of previous US general elections, we expect that the two weeks or so before the 2020 election will be the fiercest time for candidates of the two parties, and the two parties will then reveal all the black material they have about each other. Even do not hesitate to resort to fraud, rumors and slander and other means, "the best is to come."

We believe that after a series of improvements to the pollsters, the general election poll data for 2020 are still credible. Overall, the accuracy of national comprehensive opinion polls is still improving, and the biggest risk of general election polls in 2020 comes from state polls, especially swing state polls. Although the "shame phenomenon of supporting Trump" may still exist compared with 2016, it is moving in a direction conducive to improving the accuracy of opinion polls in terms of swinging voters, poll samples and poll convergence. It will help to improve the accuracy of national and state polls in 2020, and the poll data are still credible. But at the same time, we also need to guard against the impact of sudden political events on opinion polls and election results.

[1]Xia Chun: is "voting with money" more accurate than "voting by mouth"? "

[2][3]Haiguo Tuzhi Research Institute: "are the general election polls that overturned four years ago still trustworthy?" "