6月27日星期四，乔·拜登总统进行了近年来在任总统中最糟糕的一场辩论。根据一份538/益普索民意调查辩论结束后，使用益普索的知识面板，一般辩论观众认为总统表现“很差”(从“糟糕”到“优秀”)，只有21%的人认为他在辩论中表现最好。相比之下，尽管前总统唐纳德·特朗普的表现“一般”，包括无数的谎言，60%的辩论观察员告诉538/益普索，他们认为他表现最好。并且根据538的平均值自辩论以来，特朗普在全国民调中的领先优势增加了2个百分点。
在拜登表现之后，总统面临着一连串要求退出总统竞选的呼声，以便为更有可能击败特朗普的民主党人让路。例如，在其专栏作家写了一系列文章之后，《纽约时报》编辑部写道“拜登先生现在能做的最大的公共服务就是宣布他不会继续竞选连任。”甚至一些当选的民主党人也采取了非同寻常的步骤，公开建议他应该下台众议员劳埃德·道吉特周二写道，“特朗普获胜的风险太大了。”
如果拜登退出，副总统卡玛拉·哈里斯是取代他成为民主党提名人的最有可能的选择——不一定是因为她是最佳人选(这是不可能检验的)，而纯粹是因为她是总统候选人的第一人选。那么，最重要的问题是，民主党人选择哈里斯作为提名人是否会比选择拜登更好。
这最终是一个无法回答的问题，考虑到这种转变前所未有的假设性质，但是我们能根据我们掌握的(有限的、不完善的)信息，尝试量化哈里斯获胜的几率。所以，作为一个思想实验，我们运行了两个不同版本的538总统预测哈里斯取代拜登成为民主党提名人。
538的选举预测，哈里斯-特朗普版
让我们先提出一个警告:我们没有那么多民意调查测试哈里斯反对特朗普。从4月1日到7月2日，只有十几项民意调查询问了这种替代性的比赛。但是我们有来自所有主要摇摆州的民意调查，这主要归功于从晨间咨询开始跟踪，我们有足够的全国调查来计算哈里斯对特朗普的全国民调平均值，从而预测她在没有任何民调的州的表现。
在很大程度上，全国民调显示，哈里斯在与特朗普的面对面民调中表现与拜登大致相同。在一个三月福克斯新闻频道民意测验比如，特朗普领先哈里斯6个百分点，领先拜登5个百分点(完全在调查误差范围内)。就在最近的6月28日，一个进度调查数据显示总统和副总统各输给特朗普3个百分点(也在误差范围内)。也就是说，一个6月28-30日美国有线电视新闻网/SSRS民意调查发现哈里斯仅以2分之差输给特朗普，而拜登落后6分。这也在误差范围之内，但仍然是一个更大的差距，可能标志着哈里斯转变的开始。
当我们将所有这些民意调查纳入538预测模型的仅民意调查版本时——该版本抛弃了我们完整模型使用的经济和政治先验，为我们提供了候选人之间的比较——哈里斯有一个轻微地比拜登赢得选举团的机会更高，但这不是一个显著的差异:38%对35%。从各州的情况来看，拜登在宾夕法尼亚州和威斯康星州看起来比哈里斯更强，而哈里斯在内华达州的胜算高于拜登。
在我们对全国普选的预测中，哈里斯也比拜登略胜一筹。该模型预测，特朗普将在全国范围内超过哈里斯1.5个百分点，而他将超过拜登2.1个百分点。然而，这可能是我们的模型没有包括独立候选人小罗伯特·F·肯尼迪在内的任何哈里斯对川普的民意调查的一个假象，当被纳入民意调查时，他往往比共和党人略多一点从民主党人手中夺走选票。
然而，哈里斯的普选优势几乎完全被更大的选举团对她的偏见所抵消。在我们对拜登和特朗普的单独民调预测中，民主党候选人只需赢得1.1个百分点的普选就可以赢得总统大选。这要归功于拜登在宾夕法尼亚州做得更好，这是我们模型中最有可能出现转折点的州。相比之下，哈里斯需要赢得3.5-4个百分点的普选才能赢得宾夕法尼亚州，并随之赢得选举人团。
然而，哈里斯是否真的是比拜登更强的候选人，还取决于民调以外的信息。在我们的完整预测模型中，包括各种非投票的经济和政治变量，我们称之为“基本面”——哈里斯在各方面都比拜登做得差得多。拜登有48%的机会赢得选举人团，而哈里斯只有31%的机会。
这在很大程度上归功于提升我们的模型关于拜登作为现任总统这使得拜登在我们的全国普选基本面预测中比哈里斯多得了一分。然而，我们的模型没有考虑的一个因素是，总统的支持率和经济增长对竞选连任的现任者的影响是否大于来自同一政党的非现任者，这实际上可能会推动哈里斯的数字超过拜登。换句话说，你的里程数可能会有所不同，这取决于你认为拜登应该得到提升的程度，因为他是现任总统。这里没有客观正确的答案;选举预测很难的原因之一是它需要像这样的判断。
有时候，押注于不确定性是有意义的
然而，拜登的核心问题可能在当前的民意调查中没有被抓住。正如他的批评者在辩论后所说，对年龄和能力的担忧不会随着时间而消失;事实上，它们往往会变得更糟。我们今天对拜登的预测取决于许多已知的已知和未知，不幸的是，对民主党人来说，拜登已知的不利因素目前超过了已知的有利因素。另一种说法是，拜登的主要风险是“硬”风险(相对于软风险)，这是选民不太可能在11月忽视的不可改变的特征的产物。
相比之下，用哈里斯取代拜登会给选举带来更多的不确定性;换句话说，民主党人将打赌她的“软”风险没有拜登的那么糟糕。例如，哈里斯有一个最左倾的投票记录这可能会伤害她在温和派选民中的地位。然而，拜登在温和派中已经表现不佳——在目前的政治两极分化水平下，像拜登年龄这样的交叉问题可能会比任何狭隘的政策分歧对摇摆选民产生更大的影响。
归根结底，民主党提名谁为总统候选人的任何决定都将——或者应该——不仅仅考虑民调。现在，拜登的选择有很多不确定性。例如，我们对哈里斯的民意调查很少，其他候选人不太为人所知，也没有经过总统或副总统的审查(问问南达科他州州长克里斯蒂·诺姆就知道了).直到最近，这种不确定性足以阻止要求新的民主党提名人的呼声。但正如不确定性会给一个政党带来更多的负面影响，它也会带来更多的正面影响。鉴于民主党人现在对拜登的认识，他们可能愿意冒这个风险。
Would Kamala Harris be a stronger candidate than Biden?
On Thursday, June 27, President Joe Biden had one of the worst debates for an incumbent president in recent memory. According to a538/Ipsos pollconducted after the debate using Ipsos’sKnowledgePanel, the average debate watcher thought the president did “poor” (on a scale from “terrible” to “excellent”), and only 21 percent thought he performed best in the debate. By contrast, although former President Donald Trump turned in just an “about average” performance that includednumerous lies, 60 percent of debate watchers told 538/Ipsos they thought he performed the best. And according to538’s average, Trump’s margin in national polls has increased by 2 percentage points since the debate.
In the aftermath of Biden’s performance, the president has faced a barrage of calls to drop out of the presidential race to make way for a Democrat with a better chance of defeating Trump. Following a slew of articles written by its op-ed columnists, for example,The New York Times editorial board wrote, “The greatest public service Mr. Biden can now perform is to announce that he will not continue to run for re-election.” Even some elected Democrats have taken the extraordinary step of publicly suggesting he should step aside, withRep. Lloyd Doggett writing on Tuesday, “Too much is at stake to risk a Trump victory.”
If Biden were to step aside, Vice President Kamala Harris is the most likely choice to replace him as the Democratic nominee — not necessarily because she is the best pick (this is impossible to test) but purely by virtue of her being first in line to the presidency. The million-dollar question, then, is whether Democrats would be better off with Harris as their nominee than with Biden.
That’s ultimately an unanswerable question, given the unprecedented and hypothetical nature of such a switcheroo, but wecanattempt to quantify Harris’s odds of winning based on the (limited, imperfect) information we do have. So, as a thought experiment, we ran two different versions of the538 presidential forecastwith Harris as the Democratic nominee instead of Biden.
538’s election forecast, Harris-Trump edition
Let’s get one caveat out of the way: We don’t have that many public polls testing Harris against Trump. From April 1 through July 2, just over a dozen polls asked about this alternative matchup. But we do have polls from all the major swing states, thanks largely totracking from Morning Consult, and we have enough national surveys to calculate a Harris-versus-Trump national polling average — and thus to forecast how she would perform in states without any polls.
For the most part, national polls have shown Harris doing about the same as Biden in head-to-head polls against Trump. In aMarch Fox News pollfor example, Trump led Harris by 6 points and Biden by 5 points (well within the survey’s margin of error). And as recently as June 28, aData for Progress pollshowed the president and vice president each losing to Trump by 3 points (also within the margin of error). That said, aJune 28-30 CNN/SSRS pollfound Harris losing to Trump by only 2 points while Biden was trailing by 6. This was also within the margin of error but was nonetheless a bigger gap and could mark the beginning of a shift for Harris.
When we plug all these polls into a polls-only version of the 538 forecasting model — which jettisons the economic and political priors our full model uses, giving us an apples-to-apples comparison between candidates — Harris has aslightlyhigher chance of winning the Electoral College than Biden, but it’s not a significant difference: 38-in-100 versus 35-in-100. On a state-by-state level, Biden looks stronger than Harris in Pennsylvania and Wisconsin, while Harris’s odds are higher than Biden’s in Nevada.
Harris also does slightly better than Biden in our forecast of the national popular vote. The model forecasts that Trump would outpace Harris nationally by 1.5 points, while he would outrun Biden by 2.1 points. However, this could be an artifact of our model not having any Harris-versus-Trump polls that include independent candidate Robert F. Kennedy Jr., who tends to take slightly more votes away from Democrats than Republicans when included in a poll.
However, Harris’s popular-vote edge is almost entirely negated by the bigger Electoral College bias against her. In our polls-only forecast pairing Biden against Trump, the Democratic candidate needs to win the popular vote by just 1.1 points to win the presidency. That’s thanks to Biden doing better in Pennsylvania, the likeliest tipping-point state in our model. Harris, by contrast, would need to win the popular vote by 3.5-4 points to win Pennsylvania and, with it, the Electoral College.
However, whether Harris would truly be a stronger candidate than Biden also depends on information besides the polls. In our full forecast model — which includes avariety of non-polling economic and political variables, which we call the “fundamentals” — Harris does much worse than Biden across the board. Whereas Biden has a 48-in-100 chance to win the Electoral College, Harris has only a 31-in-100 chance.
This is thanks in large part to theboost our model conferson Biden as theincumbent president, which is worth an extra point for Biden over Harris in our fundamentals-only forecast of the national popular vote. However, one factor our model does not consider is whether presidents’ approval rating and economic growth impact incumbents running for reelection more than non-incumbents running from the same party, and that may actually push Harris’s numbers over Biden’s. In other words, your mileage may vary depending on how much you believe that Biden should get a boost because he’s the sitting president. There is no objectively correct answer here; one of the reasons election forecasting is hard is that it requires judgment calls like these.
Sometimes, it makes sense to bet on uncertainty
Biden’s core problem may not be captured in current polls, however. As his critics said after the debate, concerns about age and competency do not go away with time; in fact, they tend to get worse. Our forecast for Biden today depends on many known knowns and known unknowns, and, unfortunately for Democrats, the known downsides for Biden currently outweigh the known upsides. Another way of saying this is that Biden’s chief risk is a “hard” risk (as opposed to a soft one) — a product of an immutable trait that voters are unlikely to overlook by November.
Replacing Biden with Harris, by contrast, would introduce more uncertainty into the election; in other words, Democrats would be betting that her “soft” risks aren’t as bad as Biden’s. For example, Harris had one of themost left-leaning voting recordsduring her time in the U.S. Senate, which could hurt her among moderate voters. Yet Biden already does poorly with moderates — and at current levels of political polarization, a cross-cutting issue like Biden’s age may make more of a difference with swing voters, anyway, than any narrow disagreements about policy.
At the end of the day, any decisions about whom the Democratic Party nominates for president will — or should — take more than just the polls into account. Right now, there is a lot of uncertainty around the alternatives for Biden. We have few polls for Harris, for example, and other alternatives are less well known and have not been vetted at the level a president or vice president is scrutinized at (just ask South Dakota Gov. Kristi Noem). Until recently, that uncertainty was enough to hold calls for a new Democratic nominee at bay. But just as uncertainty creates more downside for a party, it also creates more upside. Given what Democrats have now realized about Biden, they may be willing to take that risk.
