Is the race for the White House closer than it appears?

Is the race for the White House closer than it appears?


If you’re reading this blog post, there’s a good chance that you know a bit about Embee, so you can skip to the second paragraph. For those of you who need some background on Embee, here goes: We build and manage unique mobile-based insight communities that help our clients—firms like Kantar and Nielsen—gain rapid, accurate insights into the behaviors and experiences of mobile device users worldwide. Our scalable, seamless technologies and processes create long-term relationships with panelists that enable us to gain in-depth, longitudinal insights that you can’t get with traditional panels.

Like a lot of you, we’re interested in politics—and we’ve been watching the latest election cycle with great interest. With the Democratic convention last week and the Republicans this week, election season is kicking into high gear. To satisfy our political curiosity, about a month ago, we started a daily poll of our panel, and we recently built a dashboard that displays the results on a 7-day rolling basis.

Before we get into some of the results, a quick note on our methodology: Every day, Embee surveys our panel, getting approximately 300 responses per day, or 2000 responses per week. As the responses come in, on an hourly basis, Embee processes all the responses, and recalculates and re-weights all the data for the past 168 hours (7 days * 24 hours), and then automatically updates our dashboard accordingly (see link below). From a sampling perspective, Embee is sampling equally by age and gender, and population-representative by state, using our DIY research tool, ResearchDesk. The responses are weighted by age, gender, race and ethnicity to census. Additionally, the responses are weighted by who they voted for in 2016. To sum, what you see in this dashboard is a real-time, 7-day rolling average, weighted by census and other factors, updated constantly throughout the day, providing the most current view possible on the state of the election.

So, what do we see?

The race is closer than it appears, and there are more undecideds.

To start, let’s look at a couple of representative polls as of Friday, August 21, 2020. FiveThirtyEight’s poll of polls, which is useful because it rolls up a raft of polls, shows Biden at 51.1% and Trump back at 42.6%, or an 8.5 point spread. Meanwhile, RealClearPolitics (“RCP”), also a poll of polls, shows Biden leading on August 21 50.0% to 42.4%, or a 7.6 point spread. Many other sources show very similar margins.

By comparison, Embee’s results show a tighter race. As of August 21, we show Biden at 43.5% and Trump at 37.2%, a difference of only 6.3 points. More significantly, however, is that we are seeing a higher proportion of undecided voters in our results, with 15.1% yet to make up their minds.


This difference in undecideds, comparing our results to other polls, is significant because if Biden is at or above 50% and there are few undecideds remaining (as the other polls appear to show), then Trump has a much larger hill to climb: Trump can’t simply convert late-breaking undecideds; he has to actually convince people to defect from Biden. However, if our results are closer to the truth, then Trump has an easier task: He doesn’t have to convince as many Biden supporters to defect. Rather, Trump need only win over the undecideds. Granted, it’s an uphill climb for Trump either way, but our results show an easier climb, nonetheless.

Why would this be?

If you drill down into the data further, a number of other interesting data points from our panel reveal themselves. One of them may provide a clue. Namely, who is winning “The Turnover Battle”?

The Turnover Battle

In 2016, the results were a surprise to many. One common explanation is that, in 2016, there was a late-breaking shift among undecided voters toward Trump in the final days and weeks leading up to the election. It was the “Let’s give the new guy a chance” vote: a portion of voters, who might ordinarily have leaned Democratic, but who never warmed up to Clinton, decided to just “give the new guy a chance.”

Correspondingly, there is a narrative that makes an implicit assumption that has largely gone unquestioned: in 2020, that those “new guy a chance” voters will generally return to their natural political home turf, and that Trump is mostly on defense to keep those voters.

Put another way: there appears to be a fairly broad assumption that Biden will naturally win the “Turnover Battle” in 2020.

What is the “Turnover Battle”? In American football, a “turnover” is when the team on defense takes the ball away from the team on offense, typically because the team on offense fumbled the ball or threw a poor pass that the other team was able to intercept. These turnovers are fairly rare—a team is lucky to benefit from more than one or two such turnovers a game – and, they provide an immense advantage to the team that gets the turnover, because the team that gets the most turnovers, in turn, gets the most chances to score, and more often than not, wins the game. Unsurprisingly, the “Turnover Battle” stat (which team got the most net turnovers) is one of the most closely watched stats in football.

The same is true in a tight political race – the candidate that gets the most net “turnovers” greatly improves the odds of winning.

So, who is winning the “Turnover Battle” in 2020?

As of August 21, our data says: Too close to call.

Looking at the data, what we see is that Trump is doing a little better than Biden at “holding the ball”: 81.8% of Trump 2016 voters intend to vote for Trump, compared to 81.5% of Clinton 2016 voters that intend to vote for Biden. On the other hand, Biden is doing better in terms of “getting turnovers”: 7.1% of Trump 2016 voters intend to vote for Biden, while 5.9% of Clinton 2016 voters intend to vote for Trump. So, on balance, we would rate the Turnover Battle to slightly lean in Biden’s favor, but it is definitely too close to call at this point.

In any case, one reason Trump may be doing better in our results, as compared to other polls, is that Trump is doing a better job at holding onto his 2016 supporters than many people seem to be assuming. If other polls are building those assumptions regarding Biden and the Turnover Battle into their turnout models, and are weighting their results by those models, they are likely underestimating Trump’s chances.

What does it all mean?

It means, at this point, the race is more similar to 2016 than most people seem to realize. If Biden had a substantially larger lead at this point than Trump, then, unavoidably, Biden would have to be doing better on turnovers – after all, if both sides experience the same level of loyalty and switching relative to 2016, then on balance, the outcome should be largely the same as well.

Does our panel naturally skew towards Trump? As noted above, our results are weighted based on census and 2016 voting results – both of which should, if anything, benefit Biden as compared to other polls. Other polls often use a turnout model, which assumes some demographics – typically demographics favorable to Trump – vote in higher numbers. By instead weighting against census, we are correspondingly assuming all demographics – including demographics more favorable to Biden – vote at roughly the same levels. Lastly, as we are also weighting by 2016 results, that too would tend to overweight Biden relative to Trump (since Clinton got more votes than Trump in 2016). We have additional data, also, which shows there may be, if anything, a slight Biden lean in our panel (data we will be sharing in upcoming posts!). So, at this point, no, we do not believe there is a Trump skew to our results; and arguably, even our results may be understating Trump’s chances.

Are Trump-leaning voters, who might normally be less willing to respond to other polls and surveys, more willing to respond to Embee, due to the nature of the relationships we create with our panelists? Maybe! One other stat that can be seen in our dashboard is that the percentage of voters that intend to vote (73.6% intend to vote as of August 21) is higher than do typically vote (usually in the 60% range). So, perhaps we are reaching a more engaged group of voters, including Trump-leaning voters who are more willing to share their views with us than with other pollsters. This is an area we will look at more closely in upcoming blog posts.

Are other polls and surveys getting a less-than-accurate picture because of their sampling methods? There is definitely a possibility of that!

These are some of the many questions we’ll continue to ponder in the coming weeks as we share additional data. We haven’t drawn any conclusions, but our overwhelming sense is that the race may be closer than it appears.

Track the 2020 election for yourself here: 


** Edit of August 25: Since we originally wrote this post, using the data from Aug 21, we are seeing a larger lead for Biden today. By all appearances, Biden is getting a strong post-convention bounce, particularly in the days following Biden’s speech. With the Republican convention this week, we can expect Trump to recover somewhat. How much? Only time, and our real-time tracker, will soon tell! (Also, since the original post, we also updated the methodology section a little bit to provide a little more detail on how it all works.)