Election Graphs

Since the last general election update on February 29th, there have been new state-level polls in Texas (x3), North Carolina (x3), Colorado, Florida, California (x2), Virginia, Georgia, Mississippi, Maine, Arizona, Utah, Michigan, Pennsylvania, and Wisconsin.

Although, as always, not all the polls individually go the same direction, in aggregate, this was a very good set of polling for the Democrats. Or to be more precise, the new results tended to be better for the Democrats than the older polls they displaced from the Election Graphs averages.

In this time frame, quite a few Democrats dropped out as well.

I would typically just go ahead and remove them from the charts and graphs I present here and leave us showing only Biden and Sanders. But some notable things happened in this last batch of polls for some of the others. So I will include them one last time.

In the next update, they will be gone.

But for now, here we go. This time lets start with the old fashioned chart of just how we would end up if every candidate won every state where they lead the Election Graphs averages:

First of all, Biden moved from just barely winning Georgia (16 EV) and Arizona (11 EV) to losing them, while Sanders went from just barely losing Florida (29 EV) to winning it.

With these changes, Biden and Sanders end up winning the same states and end up with the same 124 electoral vote margin over Trump.

Either way, the Democrat gets 331 electoral votes, and Trump gets 207.

Now, their margins in the close states are different, which will impact all of the other metrics we track here at Election Graphs, including the odds of winning. Every other metric still shows Biden in a stronger position than Sanders.

Nevertheless, it is striking that in terms of who leads states, you have the two leading Democrats with identical maps. At least for the moment.

Now, I could have left out the other candidates and still noted the above.

But there was also a comeback for Buttigieg in the weeks before he dropped out of the race.

On February 21st, the "Expected Case" showed Buttigieg losing by 84 electoral votes.

On February 22nd, the average for Nevada (6 EV) flipped, and he was only down by 72 electoral votes.

On February 23rd, the average for Michigan (16 EV) flipped and he was only down by 40 electoral votes.

On March 7th, the average for Pennsylvania (20 EV) flipped and he was now TIED with Trump in the Electoral College.

That's right, an exact 269 to 269 tie in the Electoral College.

If that were to happen, the election would go to the House, voting by state delegations, and Trump would almost certainly win.

But still, that was a big movement in this metric in a short time, resulting in the all so exciting and rarely seen tie scenario.

This showed him performing better than Warren against Trump as well.

But Buttigieg and Warren are both out now, so that doesn't matter anymore. Bloomberg's improvement doesn't either.

Now the tipping point metric:

The tipping point is how much polls have to move uniformly to flip the outcome. In other words, you can look at it as a measure of how easy it would be to change the outcome.

By this measure, every Democrat except Warren improved.

Despite being at a tie, to actually WIN Buttigieg would still need polls to move another 1.4%.

But as I have pointed out before, the main thing to note with the tipping point is that all of these numbers are small. The largest is Biden at 2.8%, and that is TINY. Polls can move 3% or even 5% in a week or two easily. They can also have systematic errors that cause them off by that much.

The shift from Buttigieg losing by substantial margins to rally back to a tie in a matter of weeks is a perfect example of this, and another reason to note these other candidates one last time.

The structure of the Electoral College means small changes in the polls can result in a massive change to the Electoral College margin.

The small tipping point is the warning flag that whatever the Electoral College margins look like, either in the simplistic categorization model or in the probabilistic model we'll look at shortly, that it is still a close race, and things could change very very quickly.

Now the median margins in the probabilistic simulation:

In this view, all of the Democrats except Warren improve. In addition to the Buttigieg surge, there was a Bloomberg surge here too. But both Biden and Sanders improve nicely as well.

Notably, since the median case and the earlier expected case usually don't match, Buttigieg still ends up in a tie in this view too.

I guess that is a fine way for him to close things out.

Finally, let's look at the odds:

Again, both Buttigieg and Bloomberg made huge gains.

Sanders and Biden gained too, but the higher up you are, the harder it is to make further gains.

And Warren slipped a bit further, ending out her run at only about a one-in-three shot at beating Trump.

That would have required winning the nomination though.

But Bloomberg, Buttigieg, and Warren have all dropped out of the presidential race.

So now there are two.

239.7 days until polls start to close.

For more information:

This post is an update based on the data on the Election Graphs Electoral College 2020 page. Election Graphs tracks a poll-based estimate of the Electoral College. The charts, graphs, and maps in the post above are all as of the time of this post. Click through on any image for current interactive versions of the chart, along with additional details.

Follow @ElectionGraphs on Twitter or Election Graphs on Facebook to see announcements of updates. For those interested in individual poll updates, follow @ElecCollPolls on Twitter for all the polls as I add them. If you find the information in these posts informative or useful, please consider visiting the donation page.

It has only been eight days since the last update. Still, there have been new polls in Nevada, Michigan, Pennsylvania (x2), Wisconsin (x2), Virginia, New York, Missouri (x2), Maryland, Montana, New Hampshire, Arkansas, California (x2), Texas (x2), and South Carolina.

So we might as well get in another update before South Carolina primary results start coming in, and Super Tuesday results three days later.

There are lots of reasons not to pay attention to current head-to-head polls against Trump when making decisions about primary choices. Most pointedly, things change and change quickly, so where things are at the end of February do not necessarily correspond to where they will be at the beginning of November. And of course, things like policy and character should also play a role.

But for those for whom "How might the general election go?" is an important decision making factor, here is the latest from Election Graphs, based on state poll averages.

Let's start with the "odds of winning the electoral college" based on the state level head-to-head poll averages, and a Monte Carlo model using the historical accuracy of the final Election Graphs poll averages to determine how far off the polls tend to be. Keep in mind this is "if the election was today." Which it is not.

The last week of polling has improved Sanders, Warren, and Buttigieg's prospects against Trump.

Meanwhile, Biden and Bloomberg have both slipped.

The order has not changed, though.

• Biden still is the strongest against Trump by far.
• Sanders is more of a gamble but still significantly favored.
• Bloomberg is a little better than a coin toss, but not much.
• Warren is a bit worse than a coin toss but still has a decent chance.
• Buttigieg would be a long shot. About the same as Trump in 2016.

Now would be a good time to talk a little about a Twitter thread by Johnathan Mummolo, a political scientist at Princeton. The thread summarizes a paper by Westwood, Messing, and Lelkes titled "Projecting Confidence: How the probabilistic horserace confused and demobilizes the public."

The bottom line is that the vast majority of people do not understand probabilities.

I have repeatedly ranted both here and on my Curmudgeon's Corner podcast ever since the 2016 election about people looking at a 14% chance of Trump winning (the median odds from all the sites I could find that gave odds) and acting as if it was 0%. That 14% is approximately the same as rolling a one on a six-sided die. And while people might be disappointed in that result if they wanted a six, nobody would be surprised by getting a one. Ones happen all the time.

This paper gets at a different but related problem. When looking at a probability of a candidate winning, vs. an equivalent percentage margin in the polls, people looking at the margins will interpret the situation as being a closer race and be more likely to vote than the people looking at probabilities even though the underlying truth is precisely the same.

Here at Election Graphs, we used the historical performance of the final, right before election day, Election Graph poll averages for every state vs. actual election results in 2008, 2012, and 2016 to estimate given a particular margin, how often would each party win? The detailed methodology is in this post from January 2019.

This analysis gives us numbers like if a Democrat is leading a state by 3.0% entering election day, they have a 73.8% chance of winning the state.

But it seems if people see 73.8%, they think it is a sure thing, so why should they bother voting? Whereas if they see a 3.0% lead, they believe it is a close race, and maybe they should vote.

Of course, 73.8% is not a sure thing at all! There is more than a one in four chance things will go the other way!

But human psychology and probabilistic innumeracy win the day!

The thread and paper also mention that at the moment, Democrats are more likely to frequent sites (like this one!) that give probabilistic forecasts. So presenting this sort of information ends up serving as a form of voter suppression for Democrats (if the Democrat is in the lead anyway).

I'll also note that when looking at a national election based on the Electoral College instead, people are going to be confused too. A significant Electoral College margin can rely on a small number of states being just barely on one side of the line or the other. A lead there can disappear in a flash with a slight movement in those states.

But of course, looking at the popular vote isn't a solution either, since as both 2000 and 2016 illustrated nicely, we don't pick presidents by the popular vote.

Here at Election Graphs, we are going to continue to present the probabilistic views anyway, of course. But if you are paying attention to them, you do need to understand what they mean, as well as pay attention to the various caveats about how quickly things change that I repeat endlessly. It is important.

But to get a full view of what is going on, we also present the national picture in three other ways regularly in the blog, and there are even more available on the blog. We let you dig into what is happening in all of the states and see all the individual polls too if you want to get granular.

This stuff is complicated. Dig in. Understand the details.

Anyway, we now turn to the median of the Electoral College margin simulations. Roughly speaking half the time, the Democrat will do better than this, and half the time, they will do worse than this. Maybe that is a little less confusing than the probability of winning?

The changes here parallel the odds, of course. But does presenting it at a margin make you FEEL differently about the results? Maybe.

Also, of course, the median margin in the model does not alone tell you how about the distributions, and how easily it would be for things to change. That is what the probability helps to understand. Two candidates might show the same median margin, but be in very different situations depending on the margins in the individual states.

Simplifying this even further to look at the margins if each candidate wins exactly the states they lead, you get this:

Interestingly, in this view, Bloomberg improves, even though his odds of winning and his median margin got worse. This divergence is because Bloomberg improved his margin by 6 EV by taking the lead in Virginia (13 EV) and losing the edge in Wisconsin (10 EV). But meanwhile, he weakened in other states enough to lower his overall chance of winning, even though the straight-up list of places he is ahead improved.

Finally, the tipping point, representing how much of a national shift in polls would change the outcome:

If I could only keep two of these charts, it would be the probability of winning and the tipping point. The likelihood of winning tells you what might happen if the election was today. But the tipping point tells you how easy it is for those probabilities to change.

There is not a single one of these five candidate combinations that are further than 3% from the centerline. That means that if there is a systematic bias of 3% in the polls toward the other side, the outcome will change. Similarly, any news event that can move the margins by 3% can change the results.

To kill a 3% margin, only 1.5% of the public needs to change their minds. People deciding to stay home and not vote can also eliminate a 3% margin in an instant.

So yes, the odds here show that if a Biden vs. Trump election were today, Biden would have a 97.7% chance of winning. But a tipping point of only 2.6% tells you that Biden's entire advantage could disappear virtually overnight with the right bit of negative news hitting the headlines, or with a pretty slight polling error in the critical states.

Which brings us back to the importance of correctly interpreting the numbers we share here on Election Graphs.

There is a big difference between "would probably win if the election was today" and "will probably win in November."

And even if the election was tomorrow, 97.7% is not the same as 100%. And 82.0% is certainly not the same as 100%.

No matter which candidate pair you look at, this is still a close and highly contested election.

The results of the South Carolina primary start coming in just a few hours.

248.2 days until polls start to close on the general election.

We have a long way to go.

For more information:

This post is an update based on the data on the Election Graphs Electoral College 2020 page. Election Graphs tracks a poll-based estimate of the Electoral College. The charts, graphs, and maps in the post above are all as of the time of this post. Click through on any image for current interactive versions of the chart, along with additional details.

Follow @ElectionGraphs on Twitter or Election Graphs on Facebook to see announcements of updates. For those interested in individual poll updates, follow @ElecCollPolls on Twitter for all the polls as I add them. If you find the information in these posts informative or useful, please consider visiting the donation page.