A Data Scientist’s Take on Electability and the Democratic Candidates

Benamy Yashar
Dec 25, 2019 · 17 min read

UPDATED 1/15/2020: scores, graphs, and rankings using latest data (surveys through 1/13). Added Michael Bloomberg. Adjusted some text to reflect new data.

In last week’s Democratic debate, most of the candidates made a direct appeal to their audience to think about which of them can beat Trump in the general election. Here are a few snippets from the closing statements.

Amy Klobuchar:

“This primary comes down to some simple questions. Who has the best ideas, the best experience? Mostly, who can beat Donald Trump, and how will she do it?”

Pete Buttigieg:

“So the nominee is going to have to do two things, defeat Donald Trump, and unite the country as president. It’s a tall order.”

From Bernie Sanders:

“…we have over a million volunteers. We have some of the strongest grassroots organizations. We have raised more individual contributions than any candidate in American history. Please join the political revolution…Let’s defeat Trump. Let’s transform this country.”

From Joe Biden:

“…we have to ask ourselves…straight up and honestly…Who has the best chance, most likely chance of defeating Donald Trump? Who is the one who’s most likely to do that?”

You get the idea. And it makes sense that they would address this question, because every poll I have seen has said that beating Trump is the top priority of an overwhelming majority of Democratic voters.

Unfortunately, it is impossible for anyone to know which candidate is truly the best foil for Trump. Some people argue that moderate policy positions are the only way to win over a majority of Americans. Others say bolder, more progressive vision will get more people to the ballot boxes in 2020. Some make the case that Midwestern roots will help flip key swing states. Many contend that age, gender, experience, honesty, and a host of other candidate attributes are all factors in “electability”. All of these are probably different pieces of the larger puzzle, but trying to analyze candidates through all of these lenses makes it hard to see who is really in the strongest position.


Measuring “Electability”

Being a data scientist, I tried to put a number on “electability”. Two lessons I took away from 2016 were that (1) the polls of head-to-head matchups between Clinton and Trump were misleading, and (2) Clinton’s favorability ratings were a harbinger of her defeat. So I have been using detailed weekly polls of candidates’ favorability ratings and a simple statistical model to rate and track their ability to win the general election.

Without getting into the grimy details of the model, here is an outline of how I analyze “electability”.

First, the ability to win the general presidential election is the ability to win swing states. The ability to win swing states is the ability to appeal to three distinct groups of voters.

  1. People who voted for Hillary Clinton in 2016.
  2. People who voted for Donald Trump in 2016.
  3. People who sat on the sidelines in 2016.

These are the three critical, distinct populations that will elect the next president. Here is how I view each of them and combine their opinions to measure electability.

Clinton Voters

There are two key, closely related facts about Clinton voters. First, almost all of them despise Donald Trump. His favorability rating among Clinton voters is usually around just 5%.

Second, likely because of Trump’s unfavorability, the overwhelming majority of them will vote for any Democrat over Trump. Tellingly, exit polls¹ showed that about 13% of 2016 Clinton voters had an unfavorable opinion of her and voted for her, against Trump, anyway. So, while my model takes favorability ratings of this group into account, it also assumes that this group will turn out for any Democratic nominee in very large numbers.

Trump Voters

There is no question that the overwhelming majority of 2016 Trump voters continue to support him now and will vote for him again in 2020. The key fact for Trump voters is that Trump is a historically unpopular president, and this does extend to some of the people who voted for him.

While his supporters are usually portrayed as adoring, devoted, unshakable fans, this picture forgets about the many reluctant voters who carried him to very narrow victories in several swing states. Trump needed the votes of thousands of Democrats to win. About 13% of his voters supported Obama in 2012, and the pre-election surveys and exit polls¹ showed that 15–20% of Trump voters had an unfavorable opinion of him when they voted for him in 2016. This looks to be down to around 10% today, but that still leaves the door open for a Democratic candidate to win the election by swinging a chunk of Trump voters back.

Disaffected Voters

This third group of voters, whom I’m calling “Disaffected Voters”, is a mixed bag of people who rarely vote, who were unsatisfied with the candidates, who voted for third party candidates, who were too young to vote in 2016, and many others. It is hard to make many generalizations about this group or to predict exactly how they will vote (if they will vote) in 2020. I think the best we can do is simply listen to what they are saying in the polls about who they like.

There are four key facts for this group of voters.

First, there are a lot of them out there. The 2016 election had the lowest voter turnout for a presidential election in 20 years, with just 55% of eligible voters participating. At the same time, 5% of voters voted for third party or write-in candidates, the highest amount in 20 years. So this group represents about 48% of all eligible voters.

Second, there are reasons to believe many of them will show up to vote in 2020. The 2018 midterm elections had the highest voter turnout for a midterm since 1966. On top of that, every election will bring in some number of first-time voters. In 2016, supposedly a low-turnout year, 10% of voters had never voted before. So I am predicting that these “Disaffected Voters” will be 10–20% of all voters in 2020.

Third, most of these people still don’t like Trump today. They didn’t like him enough to vote for him in 2016, and polls show that most of them still have an unfavorable opinion of him². This means a Democrat who gains favor with this crowd will not have to convince them to abandon Trump.

Finally, many of these people are very picky. They aren’t inclined to pick between the lesser of two evils. When they didn’t like Clinton or Trump, they just stayed home. Today, when asked if they would vote for the Democratic nominee or Trump, 15–20% of them say, “It depends”², a much higher percentage than any other group.

So, even if this group ends up being only 10–20% of the voters, they may have the greatest potential to carry a candidate across the finish line, especially in the states where the margins were smallest in 2016.

Favorability Scores

For each of the groups described above, I have developed a method for calculating a “favorability score” from the weekly polls administered by The Economist and YouGov². In the surveys, respondents are asked if their opinion of a candidate is “Very Favorable”, “Somewhat Favorable”, “Somewhat Unfavorable”, or “Very Unfavorable”. (“Don’t know” is also an allowed response.) I combine the different responses for a candidate into one score per group.

The scores were scaled to correspond roughly to the percent of people in a group who might vote for a candidate in the general election against Donald Trump, but the scores should not be treated as true predictions.

The overall scores and candidate rankings presented below are based on a weighted average of the three favorability scores calculated with the last six weeks of surveys. The weights were intended to model voters in critical swing states, which have slightly more 2016 Trump voters than Clinton voters as well as a smaller chunk of voters who were on the sidelines in 2016. Scores are not calculated for all Democrats running for the nomination³.

Overall Score = (0.42 × Clinton) + (0.43 × Trump) + (0.15 × Disaffected)


1. Andrew Yang

Clinton Voters: 90

Trump Voters: 13

Disaffected Voters: 52

My electability model gives the highest overall score to dark horse Democratic candidate Andrew Yang. It also suggests that, should Yang win, his coalition would look the most different from Clinton’s. His favorability among Clinton voters is the lowest of these candidates, but his favorability scores for Trump and disaffected voters were the highest by a notable margin.

Interpretation

Yang’s strength among Trump and disaffected voters may be explained by this status as a nontraditional Washington outsider with heterodox policy positions, whose campaign slogan is “Not left, not right, forward.” Many Obama voters went to the polls in 2008 and 2012 looking for “hope and change”. In 2016, many of them stayed home or turned to Donald Trump when they didn’t find that in Clinton. 39% of voters said that ability to bring change was the most important quality for the president, and 82% of those voters chose Trump¹. So Yang may be appealing to many voters who are, again, looking for a major change agent in the White House.

It is worth noting that he is still unfamiliar to many voters, so his numbers are likely to move over time. Notably, while his scores among Trump supporters have been the highest of the field every week over the last few months, they dipped fairly low in December. If that support disappears again, his position at the top of this list would certainly be in jeopardy.


2. Elizabeth Warren

Clinton Voters: 94

Trump Voters: 5

Disaffected Voters: 45

If Warren wins the presidency, my model suggests it will be by turning out the liberal vote. She is both the highest scoring candidate among Clinton voters and the lowest scoring candidate among Trump voters.

It is worth noting here that Warren was at the top of my list a couple months ago, when she had a favorability score approaching 97% among the Clinton voters. Her favorability with that group has dropped dramatically since then. While it is still the highest by a good margin, it no longer fully makes up for her weakness with Trump voters. If Warren falls further, she could easily end up in the lower tier of candidates. On the other hand, if she can regain the favor she recently lost, she might rise back to the top of this list.

Interpretation

Warren is a liberal firebrand who has the potential to both reassemble Clinton’s coalition and to draw in some of the voters Clinton failed to reach. The similarities with Clinton are easy to see — Hillary Clinton was also a Democratic, female senator from a northeastern state who had history in the South. That said, they are running very different campaigns and have different public perceptions, and those differences may allow Warren to win some of the close races that Clinton lost.


3. Bernie Sanders

Clinton Voters: 91

Trump Voters: 7

Disaffected Voters: 50

Sanders’ greatest strength is his potential to draw people off the sidelines, and his greatest handicap is the dislike among Trump voters. His favorability among people who voted for his 2016 competitor is about average for the field, but his favorability among Trump voters is almost as low as Warren’s.

Interpretation

Sanders is arguably the best known candidate in the field. He rose to national prominence in his 2016 campaign, stayed in the public eye for three years, and is running on essentially the same platform now. With the exception of the weeks after his heart attack, his favorability scores have changed little week-over-week. (Hence the lack of a time series graph for him here.)

Sanders’s brand of democratic socialism is alienating to some Democrats and completely repulsive to almost all Republicans. This earns him his average and below-average scores among Clinton and Trump voters. What Sanders offers, however, is an ability to draw in a large pool of disaffected voters. As was seen in 2016, many of his devoted supporters are not going to turn out to vote for anyone but him, even if he endorses the Democratic nominee. Still, it is not clear if that devoted following can make up for the voters he alienates, so my model has him just a hair above his main rival, Biden.


4. Amy Klobuchar

Clinton Voters: 90

Trump Voters: 10

Disaffected Voters: 41

Klobuchar’s main strength is her relatively strong appeal among Trump voters, but her low favorability scores among Clinton voters and disaffected voters put her in this bottom tier of candidates. Excepting Bloomberg, she is lowest among both Clinton voters and far below everyone else among the disaffected voters.

Interpretation

Klobuchar is running for president on the idea that she can translate her history of strong wins in Minnesota to strong wins in other Midwestern swing states and, thereby, assure a victory for the Democratic Party. While her moderate positions have gained her the favor of Trump voters, the same positions may be costing her the favor of many Clinton voters. My model suggests that she is just breaking even between these two groups.

Her appeal among Clinton voters seemed to be on the rise going into October. Since then, however, it seems to have flatlined. She still has a lot of room to improve with this group. So, if this has been just a temporary slowdown in her rise, she could break out of this lower tier in time. If, however, she has already hit her ceiling with Clinton voters, it seems likely that she will be stuck in this tier. She is still unfamiliar to many voters, so she has a good amount of room to move both up and down.

Klobuchar’s biggest liability is her lack of appeal among disaffected voters, with whom she has a damagingly low favorability score, higher only than Bloomberg’s. The 2020 election is sure to attract many voters who were sitting on the sidelines in 2016. These voters did not vote for Clinton — a female, a moderate Democrat, and a former senator — and it does not seem like they are currently excited about Klobuchar either. If she cannot find a way to start appealing to this group, she would be hard-pressed to win the general election.


5. Joe Biden

Clinton Voters: 92

Trump Voters: 8

Disaffected Voters: 41

Biden does not show up on the top or bottom of any of the three lists. His strength is with Clinton voters, among whom he is second only to Warren. His clearest weakness is with the disaffected voters, where he shares the bottom half of the rankings with Klobuchar and Bloomberg.

Interpretation

Often considered the safest bet against Trump, my view of the data does not support that conclusion. Biden has been campaigning on both the successes of the Obama-Biden administration and the promise of drawing moderates back into the Democratic party. However, neither of those look like a winning argument here.

In 2016, Hillary Clinton got the full endorsement of Barack and Michelle Obama, however, in the exit polls, only 28% of voters said they thought the next president should continue Obama’s policies, and Clinton won 91% of those votes¹. With a favorability score of 91 among Clinton voters, Biden seems well positioned to get most of those votes again, but, clearly, that would not be enough to win.

As for getting Trump-voting moderates on his side, Biden was looking stronger in that regard some months ago. As his campaign has worn on, his favorability among Trump voters has declined steadily. It seems as though many Trump voters had favorable memories from his time as VP but feel differently after being reintroduced to him years later. Now his moderate competitors, Klobuchar, Buttigieg, and Bloomberg, are all scoring higher than him, and he is just one point above Sanders, a self-described democratic socialist.


6. Pete Buttigieg

Clinton Voters: 90

Trump Voters: 8

Disaffected Voters: 41

Buttigieg’s overall score has fallen to the bottom, because he scores fourth or fifth in every group. While he does not have any damagingly low scores, he doesn’t have any particularly high scores to lift him up.

Interpretation

Buttigieg seems to have been perceived as an agreeable, liberal candidate without much baggage. His recent appeals to the moderate wing of the Democratic party seem to have landed him between Klobuchar and Biden in favorability among Trump voters. He is not attracting as many of the disaffected voters as Yang or Sanders, but he does not seem to be turning them away either. His broad appeal is making him a serious contender, and he still has room to grow further.

His position on this list a few weeks ago was much higher. Since then, his favorability with Clinton voters dropped very dramatically. It seems to have bounced back somewhat, but, if this score falls further, Buttigieg’s position in this ranking would likely fall to the very bottom of the list.


7. Michael Bloomberg

(Added 1/14/20)

Clinton Voters: 81

Trump Voters: 8

Disaffected Voters: 30

According to my model, the differences in electability between the other six candidates are minuscule in comparison to the difference between them and Bloomberg. His favorability among the Clinton and disaffected voters is so much lower than the others that it is hard to see how he would muster enough votes in the swing states to win the general election. Either he is, by a wide margin, the least electable candidate of these seven, or there is something fundamentally wrong about the assumptions I have made.

That said, too little time has passed since he entered the race to know if and how his favorability scores are trending. So it may be the case that he is currently closing the gap with his competitors.

Interpretation

Bloomberg entered the race just two months ago, when he decided that the candidates already in the race were looking weaker than he expected and would not fare well in the general election. Of his now-competitors, he said, Trump would “eat them up.” He believes he would fare better.

One of the assumptions I make in my model is that, while most Clinton voters will vote for the Democratic nominee, some can be lost. With extremely high unfavorability ratings among Clinton voters, Bloomberg is currently testing the limits of that assumption. If that assumption is wrong, he may be a more viable candidate than my model suggests.

However, he also does not seem to have enough traction with the group of voters who sat out the 2016 election. It makes sense that this group, which favors Yang and Sanders the most, would have such distaste for a multibillionaire former Republican from Wall Street. It is hard to imagine that there could be much overlap between the Sanders supporters who refused to vote for Clinton and potential Bloomberg supporters.

He may have been figuring that, as a former Republican and Independent, he could be a unifying compromise candidate. It seems, though, that the Republicans already have their candidate, and, unless he turns things around, he would just be the losing candidate.


Summary

The most recent survey from The Economist/YouGov suggest that people currently believe Biden and Sanders are the most electable Democratic candidates². It is easy to understand their reasons for believing this. Joe Biden was a popular, two-term vice president advocating center-left policies and promising a return to normalcy. Bernie Sanders has spent years building a national movement of millions of people who have waited several years to vote for him in the general election. However, my analysis of candidate favorability across different voter groups tells a different story. And, in that story, Biden and Sanders are two of the candidates least likely to win the presidency in the general election.

Among 2016 Clinton voters, Elizabeth Warren set the high water mark for favorability a few months ago, when she looked like the consensus pick of the liberals and the strongest candidate overall. However, consensus among former Clinton voters no longer looks possible for any candidate. Warren is still the leader, but every candidate looks like they will alienate some portion of the group. Candidates running as moderates — Biden, Buttigieg, and Klobuchar — seem to be alienating almost as many Clinton voters as Sanders and Yang, who are running on much more radical policy ideas. While almost all Clinton voters agree that Trump needs to be replaced, there are voters in this group who are going to be picky about whom they support. In close races, those voters could prove to be crucial or costly.

If the general election becomes a game of inches, former Trump voters could prove to be similarly crucial or costly. Naturally, these voters are sending a different message from Clinton voters — the most “progressive” candidates, Warren and Sanders, are going to find very few supporters in this crowd. The polls support the electability arguments of the more moderate candidates, who are doing much better with this group of voters, Amy Klobuchar most of all. However, the Democratic candidate doing the best with this group may come as a surprise to many. Week over week, Andrew Yang’s unique brand of populist, outsider politics with heterodox policy proposals earns the highest scores with this group.

Finally, I give the oft-ignored group of disaffected voters the last word. And they seem to have the strongest opinions about the Democratic candidates. As much as the Trump voters may be willing to support the moderate candidates, the disaffected voters may be just as willing to punish them; Klobuchar, Biden, and Bloomberg earn the lowest scores of the bunch. Warren and Buttigieg do passibly with this group, but Sanders and Yang are the very clear favorites. While most of these voters do not support Trump, they have little appetite for either “politics-as-usual” or another New York billionaire. Depending on who the Democratic nominee is, this group, which declined to show up for Hillary, could either sit on the sidelines again or carry the nominee all the way to the White House.

My overarching theory in this analysis is that the competitiveness of the Democratic candidates against a historically unpopular incumbent can be predicted by their current favorability across different voter groups. This analysis suggests that there are different viable paths to the White House but that the candidate best suited to take those paths may not be the ones currently leading in the polls. Most illustrative of this is how Andrew Yang and Elizabeth Warren both do well in my electability model. Yang’s path to victory relies on his appeal among disaffected and disengaged voters, and Warren’s path mainly relies on energizing the Democratic base. What they have in common is a substantial mix of key people who like them. So, if Democratic voters are interested in strategically nominating a candidate who can beat Trump in the general election, I recommend simply listening, carefully, to who the voters say they like most.


Footnotes

[1] https://www.cnn.com/election/2016/results/exit-polls

[2] https://today.yougov.com/topics/economist/survey-results

[3] I am only doing this analysis for candidates who seem to have a realistic path to the nomination. I have not included Michael Bloomberg in this analysis yet, because he entered the race relatively recently and YouGov does not yet have six weeks of polling data available for him. I plan to update this analysis when enough polling data is available.

Benamy Yashar

Written by

I am a data scientist living in Brooklyn. I am starting to write and share my political thoughts.

Welcome to a place where words matter. On Medium, smart voices and original ideas take center stage - with no ads in sight. Watch
Follow all the topics you care about, and we’ll deliver the best stories for you to your homepage and inbox. Explore
Get unlimited access to the best stories on Medium — and support writers while you’re at it. Just $5/month. Upgrade