US Presidential Election: Poll numbers notwithstanding, Trump remain favorite to register an upset win??

We are being daily bombarded for months with the claim that Joe Biden is leading ​Donald Trump in the US national polls for the presidential election. A poll of poll suggests that Biden leads Trump by a margin of 7%+ which is suffice to whip up a thumping landslide victory.

Despite such attempts of mental conditioning, Gallop Poll found that regardless of whom they personally support, a whopping 56% of Americans expect Trump to prevail in next month's election, while just 40% think Biden will win. 90% of Republicans believe Trump would romp home comfortably while only 76% of Democrats think Biden will make past the finish line. Fifty-six percent of independents predict that Trump will win.

There appears simply little takers for the numbers being churned out by pollsters after their 2016 debacle. Polls were widely maligned because Trump won the election when the polling industry almost unanimously said he would not. Polls have a Margin of Error (MoE) of (+)  3-5%. So even if the numbers they project are within this MoE, polls maybe publicly perceived as flopped for failing to call right though they for all purpose, satisfy the criteria of statistical accuracy.

Hillary Clinton had a clear lead over Trump in the polls for almost the entire 2016 campaign. Clinton won 65,844,610 votes, or 48 percent, with Trump winning only 62,979,636 votes, or 46 percent and yet lost the election. Confidence in pollsters nosedived and appears have not fully recovered even after four years. 

States are allocated delegates roughly proportional to their populations, with 538 delegates in total. Hillary lost because the US presidential ​voting system assigns each state a number of winner-take-all Electoral College Votes (ECV), which​ go to the state’s victor regardless of the​ margin of victory. Typically, the electoral outcomes of 30-35 states are predictable even before the first vote is cast. They are categorized as either blue states (wherein the Democratic Party commands massive popularity) or red states (wherein the Republican Party commands massive majority). This leaves about 15-20 odd states known as Battleground States, within which a handful of Swing States usually decide the election and be targeted heavily by campaigners.  

The 2016 presidential election was very close though Donald Trump’s 306 to 232 Electoral College victory over Hillary Clinton obscures just how competitive the race was. Just roughly 80,000 voters in Michigan, Pennsylvania, and Wisconsin collectively swung the election in favour of Trump. If Clinton had won these states, this would have flipped the election. Clinton would have then won 278 ECV to Trump’s 260.

Soon after winning the election, Donald Trump tweeted:

"Campaigning to win the Electoral College is much more difficult & sophisticated than the popular vote. Hillary focused on the wrong states! I would have done even better in the election, if that is possible, if the winner was based on popular vote - but would campaign differently".  

Trump hit the nail on its head in analyzing the 2016 outcome. The vital difference was campaign strategy. Trump won states by razor-thin margins in some cases, converting 46% of popular votes cast into 56.5% of EMV. Conversely, Clinton’s huge popular vote tally was concentrated in big states such as California and New York, a sizeable proportion, wasted votes.

Hillary strategically focused on the wrong key success variable and paid a heavy price in the husting while Trump made history because he had intuitively a better sense of strategy. Those results make Clinton the fifth presidential candidate in US history – and the second this century – to win the popular vote and lose the Electoral College - the other being Al Gore, another Democrat.

With this as introduction, this blog post tries to fathom who is currently ahead and what their respective pathways to victory are? The election map of 2016 will no doubt be different than the landscape this November and may not offer the best template for forecasting. Accordingly, the analysis will take into consideration, state wise, the latest available opinion poll instead of average of polls.

Statistical accuracy is not what is usually sought by political pundits and other consumers of opinion polls, such as the media and politicians. They want to know who is likely to win the election. But it gets much worse. The (+) 3% Margin of Error (MoE) is theoretically the maximum accuracy such a poll can hope for. If there are errors in the polling methodology, such as unrepresentative samples, the MoE will be correspondingly greater. Nonetheless, this blog treats the projected numbers of these polling at their face values. For example if poll numbers projected are Biden 47% and Trump 45%, this would mean that state is given to Biden even though statistically it should be treated as a dead heat or includes a possibility of Trump also being ahead.

Accordingly, here are the data for Trump and Biden respectively.

What this data says

Given that 270 ECV are needed to win US Presidency, with respect to 2016, Trump can afford a net loss of 36 ECV while Biden needs to net gain 38 ECV. If elections are held now, Trump leads in 26 states with 244 ECV (a net loss of 62 ECV from 2016) while Biden leads in 24 states with 283 ECV ( a net gain of 51 ECV from 2016). Biden not only leads Trump by 39 ECV, but also wins the Presidency by crossing the 270 ECV victory line.  

The Blue (Democrats) and Red (Republican) states each could further be sub categorized into two - those strongly leaning respectively towards them (viz those over 10% lead margin) and those weakly in their favour (viz less than 10% margin lead). In other words, categorization into those states having greater probability of being flipped and those with lesser probability.

Such a statistical disaggregation enables a better insight into the comparative vulnerability of the two candidates. Trump is found strongly leading in 13 states that adds up to only 78 ECV while Biden strongly leads in 18 states that account for 206 ECV. Trump is clearly in the back foot according to these polls.

Biden evidently has wider range of paths to victory and to consolidate gains projected by opinion polls. If one goes by the average of polls, then Biden is poised to win handsomely whereas Trump is doomed to be soundly defeated.

But what if the 2020 opinion polls are wrong as they were in 2016?

The American Association for Public Opinion Research (AAPOR) postmortem analysis of 2016 polling debacle suggest the following factors:

1. Undecided/Independents

According to AAPOR  "The number of “undecideds” in 2016 was more than double that in prior elections. Of these, a disproportionate number voted for Trump. Pollsters were foiled by an unusual wave of undecided voters breaking for Trump at the last minute; there was too little polling in key swing states to really know what was going on; conclusions extrapolated from that paucity of data were broadcast with far too much certainty."

But the 2020 polling todate reveals far fewer undecided voters, suggesting this source of poll error will not be as large in this year’s election. In 2016 elections, independents and other parties cornered 6% of votes. Polling data this year currently hovers around 6% of undecided/independents. This blog  expects the share of votes by independents and other parties to tumble to below 4% levels in actual voting this year  - the lowest in history. The 2% loss in this category is expected to mostly break in the favour of Trump as in 2016.

2. Oversampling of Democrats

It is argued that gold-standard, nonpartisan surveys have found for decades that more U.S. voters identify with or lean toward the Democratic Party than the Republican Party – whether these surveys take place under Republican or Democratic presidential administrations. Consequently, most polling agencies adopt stratified sampling methods that oversamples Democrats.

There are three main problems with such justifications. Firstly, nonvoters are now much more Republican leaning than regular voters. Secondly, Democrats also are more clustered geographically than Republicans, a factor that dilutes their electoral strength. The conversion efficiency of their popular votes into ECV is relatively poor. Thirdly, the US is getting more and more bi-polarized with the percentage of independents and those affiliated with other parties reducing with Republicans benefiting from this contraction trend.

The well-known and widely followed Franklin & Marshall College poll based in Lancaster, Pennsylvania, had Clinton ahead by double digits in the state in its last poll just before the 2016 election. Yet Trump won Pennsylvania spectacularly with a 0.7% margin, virtually coming from no where. This poll entered up with mud on their faces because they oversampled Democrats.

Similarly the internals of last month's ABC survey show Republicans are clearly under-sampled as compared to Democrats. The only consolation is that this sampling bias was much worse in 2016. Besides in 2016, only around 4% voted independents and minor parties. However, the poll inexplicably samples them five times their strength.

Besides the internals of the survey indicates that their MoE ranges from (+) 4.5-5%. So this poll projects 54% Biden 45% Trump. At  (+) 5% MoE, it is statistically possible the result to be also 49% Biden 50% Trump - a repeat of 2016 with Trump pipping the Democratic candidate by a whisker! This raw fact is not revealed to readers by mainstream media.

3. The Lying Factor

The question of whether "shy Trump” voters were undercounted in polls before the 2016 election was part of an exhaustive postmortem published by AAPOR. “Some Trump voters who participated in pre-election polls did not reveal themselves as Trump voters until after the election, and they outnumbered late-revealing Clinton voters...those who admit changing their minds more or less wash out, breaking about evenly between the Republican candidate and the Democratic candidate.  Not in 2016, though: People who changed their answers when called back after the election had voted for Trump by a 16-percentage-point margin."

CloudResearch conducted a similar survey but online that found that Republicans and Independents are twice as likely as Democrats to say they would not give their true opinion in a telephone poll question about their preference for president in the 2020 election. That raises the possibility that polls significantly understate support for President Donald Trump.

4. Turnout

In 2012, voting rates for non-Hispanic blacks (66.6 percent) were higher than non-Hispanic whites (64.1 percent) for the first time because of Obama. In 2016, turnout increased to 65.3 percent for non-Hispanic whites, but decreased to 59.6 percent for non-Hispanic blacks. This change in trend proved devastating for Democrats particularly in key swing states.

Trump successfully mobilized white voters who ordinarily have low rates of voter turnout. These were largely non-urban voters and those with lower levels of education who polling samples under-weighed. However, in 2020, most polling appeared to have corrected this deficiency in their sampling methodology. 

Surveys also indicate that Democrats are much more likely to rely on vote by mail compared to Republicans (72% to 22%). Trump and his Republican supporters have raised doubts about the validity and security of vote by mail. This issue is soon to be decided by the US Supreme Court to rule on the validity of the results after the election, on the basis mail ballots have been either improperly included or excluded in official tallies. If mail ballots are disallowed then this would be a huge devastation to Democrats as Republicans are much more likely to physically turn up at booths to cast their ballots, particularly to show solidarity with Trump who contracted COVID.

The turnout, particularly because of the COVID impact is expected to be the most important X Factor to the November elections.

Main Takeaways

On one hand, this year’s election seems to have historically low levels of undecided voters, a factor that increases the accuracy of the polls. However oversampling of Democrats though lesser than 2016 is still a factor that can distort findings of polls.There are also exists tremendous uncertainty about turnout and whose votes will be cast and counted.

Trump's support appears still understated though perhaps less so relative to 2016. Whether he wins or loses, his chances of re-election are most likely higher than suggested by the polling as illustrated in the above table.

All this suggests considerable caution be exercised in relying on the polls to forecast the election. These forecasts are almost surely overconfident.

Biden on the face of polling seems cruising to a comfortable win, carrying with him 24 states that accounts for 283 EMV - 13 more EMV than required. His pathways to victory are multiple. He needs to only defend his current above leads in the above 6 swing states and raid the group of 18 states that currently leans weakly for Trump to either offset any defeat in these 6 states or increase his EMV tally even further.

The pathways for Trump on the other hand are much more narrow. He currently leads with 244 EMW that includes 13 states accounting for 166 EMV that are weakly leaning Red with wafer thin majority averaging around 3%. In other words, he appears highly vulnerable. He needs to not only defend his leads in this group of 13 states but also needs to win another 26 EMV to snatch victory from the jaws of death by flipping at least 3 of these 6 swing states projected as weakly leaning Biden.

This may look an uphill task but Trump could rise to the challenge and make it look a facile win a la 2016! An ABC-Langer Research Associates study found Trump with a 20-point advantage in strong enthusiasm among his supporters, 53 percent approval for handling the economy, solid backing in his base and a clear lead among voters who plan to cast their ballot in person on Election Day. Compared to 2016, he has increased his support among Blacks, Hispanics/Latinos and Asians including Indians. In particular, his support among Hispanics/Latinos are surging as evident from his 4% lead in Florida. If in 2016, the turnout among Blacks declined, it is expected to be much worse this year. These all are trends that do not augur well for Democrats. On the other hand, Democrats could take heart that Trump's base with the less educated blue collared white voters particularly in the Rust Belt states has eroded somewhat significantly. However the Catholic Vote appears to be the central key in 2020. Biden is a practicing Catholic but the anti Catholic utterances of his running mate, Kamala Harris may offset some of this advantage. The confirmation of Amy Coney Barrett, also a Catholic, as the Supreme Court judge on the other hand may work to Trump's advantage. 

In short, the US Presidential Elections is far from being done and dusted. Watch this space for further updates.