評論
打印
Joe biden's job-approval rating stands at 39%, putting him roughly in a tie for lowest of any president at this point in his term in the history of American polling.
喬·拜登的支持率為39%，這一數字是美國民調歷史上總統任期內的最低水平之一。
In all six states that could prove decisive in November he trails by between one and six percentage points.
在11月可能決定選舉結果的六個州，他的支持率都落后對手1到6個百分點。
In the two where he is closest, Wisconsin and Michigan, Democratic candidates' margins have underperformed the final polls by an average of six points in the past two elections.
在威斯康星州和密歇根州這兩個最具競爭性的州，民主黨候選人在過去兩次選舉中的支持率均低于實際得票率6個百分點。
Even if he wins both, Mr Biden would still need one more swing state to secure the 270 electoral votes necessary for re-election.
即使拜登贏得這兩個州，他仍需要一個搖擺州來獲得連任所需的270張選舉人票。
These numbers suggest that the race is hardly a "toss-up". True, the five months before the vote give Mr Biden time to make up ground, and the polls may underestimate his true support.
這些數據表明，這場競選幾乎不能被視為“勢均力敵”。的確，距離投票還有五個月的時間，拜登有機會縮小差距，民調可能低估了他的實際支持率。
But it is also possible that the candidate to benefit from any polling error could be Donald Trump.
但也有可能，從民調錯誤中受益的候選人可能是唐納德·特朗普。
In 2016 most pundits found it unfathomable that a manifestly unqualified candidate like Mr Trump could win the presidency.
2016年，大多數專家都認為，像特朗普這樣明顯不合格的候選人能夠贏得總統大選簡直不可思議。
This bias was reinforced by polls that consistently put Hillary Clinton in the lead.
這種觀點被持續領先的希拉里·克林頓的民調所強化。
Now, after a presidency that yielded two impeachments and a riot at the Capitol, the prospect that voters might willingly return to office a man recently convicted of 34 felonies seems nearly as outlandish. Yet surveys suggest it is more likely than not.
現在，經歷了兩次彈劾和國會騷亂之后，選民可能愿意讓一位最近被判34項重罪的人重返總統職位，這同樣不可思議。然而，調查顯示，這種情況更有可能發生。
The Economist's statistical model of the election-which relies solely on polls, past results and economic data, and knows nothing of Mr Trump's record in office or in the courts-gives Mr Biden a one-in-three chance of re-election.
《經濟學人》的選舉統計模型僅根據民調、過去的選舉結果和經濟數據進行分析，對特朗普在任或法庭上的記錄一無所知，最終給出拜登連任的幾率為三分之一。
That means a victory for Mr Biden would count as only a mild surprise, somewhat more likely than the 30% share of days on which it rains in London.
這意味著拜登勝選只是一個小小的意外，大致相當于倫敦某一天下雨的概率（30%）。
Four years ago this week this model gave Mr Biden an 83% chance.
四年前的今天，這個模型給出拜登的連任幾率為83%。
Our model combines two types of data: horse-race polls and "fundamentals", or expectations based on historical precedents.
我們的模型結合了兩類數據：選情民調和“基本面”，即基于歷史先例的預期。
Its starting-point, drawing on work by Alan Abramowitz of Emory University, is a national "fundamentals forecast"
它起源于埃默里大學艾倫·阿布拉莫維茨的研究成果，是一個全國性的“基本面預測”，
that seeks to predict the incumbent party's share of the vote (excluding third-party candidates) based on three factors: the president's approval rating, incumbency and the economy.
旨在根據總統的支持率、現任優勢和經濟狀況三個因素預測現任政黨（不包括第三方候選人）的得票率。
We have reinterpreted Mr Abramowitz's incumbency advantage as the absence of a "term penalty" suffered by parties seeking to hold the White House for more than eight years, a feat accomplished only once since 1950.
我們對阿布拉莫維茨的“現任優勢”進行了重新解釋，認為當一個黨派試圖連續掌權超過八年時，會遭遇“任期懲罰”，自1950年以來，這種情況只成功過一次。
We also account for the weakening of the link between economic performance and the vote shares of incumbents seeking re-election, due to growing polarisation of the electorate into camps of committed partisans.
我們還考慮到，由于選民的極化導致他們堅定地站在黨派一邊，經濟表現與尋求連任的現任黨派得票率之間的聯系已經減弱。
As anyone who followed American politics in 2000 or 2016 knows, winning the popular vote is no guarantee of prevailing in the state-by-state electoral college.
任何關注過2000年或2016年美國政治的人都知道，贏得全國普選并不能保證在各州選舉人團中獲勝。
Our model also includes a fundamentals forecast for each state, based largely on how much more Democratic or Republican its past results have been than America as a whole.
我們的模型還包括對每個州的基本面預測，主要基于它們過去的表現相對于全國的黨派傾向。
For example, in 2020 Mr Trump won Florida by 3.3 percentage points while losing nationwide by 4.5, putting Florida 7.8 points to the right of the country.
例如，2020年特朗普在全國落后4.5個百分點的情況下，卻在佛羅里達州贏得了3.3個百分點，使佛羅里達州相對于全國右傾了7.8個百分點。
Our model then hunts for the true voting intentions in each state, on each day, taking into account opinion polls published so far.
我們的模型每天在每個州搜索真實的投票意圖，考慮到目前發布的民調，
It adjusts for the impact on survey results of methods, sample sizes, pollster biases and the use of likely-voter screens.
并調整方法、樣本大小、民調偏差和可能的選民篩選影響。
Every day, the model generates 10,001 different possible scenarios for the election. The most common ones land close to poll results and its fundamentals forecasts.
每天，該模型都會生成10001種不同的選舉可能情景。最常見的情景接近民調結果及其基本面預測。
But it also includes a healthy number of big surprises, to allow for the risk of significant polling errors.
但它也包括大量的意外，以應對顯著的民調誤差風險。
In 2012 the overall picture painted by state-level surveys turned out to be more accurate than nationwide polls; four years later, the reverse was true.
2012 年，州級民調描繪的總體情況比全國民調更為準確；四年后的情況則正好相反。
To avoid overweighting either type of survey, the model treats the election as a giant jigsaw puzzle, in which vote shares in each state have to add up to the national total.
為了避免過度依賴任何一種調查，該模型將選舉視為一個巨大的拼圖，其中每個州的得票率必須加總到全國總數。
An unusually strong national poll for a candidate will yield higher expected vote shares in every state.
某候選人的一項全國性強勢民調將提高其在每個州的預期得票率。
Conversely, a surprisingly poor state-level poll for a candidate will lower the model's expectations not just in that state but nationwide, especially in places that are nearby and demographically similar.
相反，某州的一項意外糟糕的民調將降低模型對該候選人不僅在該州而且在全國、特別是在相鄰和人口結構相似地區的預期。
