More Trump Deterioration

Another four weeks or so has gone by since I posted so it seems like it is time for another update. Once we are hot and heavy in the depths of 2024, there will probably be more frequent updates. For now, every four weeks seems sufficent. It may even be too frequent. State level polling is still relatively sparse, and changes to the various metrics we track here come slowly.

In the last four weeks there have been 28 new data points added to Election Graphs. There were 13 polls for Biden vs DeSantis (GAx2, NVx2, AZ, VA, CO, NM, MN, PA, TN, TX, WI), 12 polls for Biden vs Trump (GAx2, NVx2, AZ, VA, CO, NM, MN, TN, TX, WI), 2 polls for Biden vs Pence (AZ, GA), and 1 poll for Kennedy vs DeSantis (PA).

If you want updates on each and every poll as it comes out, as well as daily summaries of the status for the best polled matchup (currently Biden vs Trump), follow Election Graphs on Mastodon.

Let's start again with the "spectrum of close states" for Biden vs Trump.

Last time on 2023-04-24 it looked like this:

And now on 2023-05-21:

So here are the changes:

  • Colorado: Biden by 7.5% -> Biden by 6.8% (Trump +0.7%)
  • Virginia: Biden 5.4% -> Biden by 5.6% (Biden +0.2%)
  • Texas: Trump by 6.5% -> Trump by 6.0% (Biden +0.5%)
  • Wisconsin: Trump by 0.7% -> Trump by 0.2% (Biden +0.5%)
  • Minnesota: Biden by 6.0% -> Biden by 6.9% (Biden +0.9%)
  • Georgia: Trump by 3.3% -> Trump by 0.7% (Biden +2.6%)
  • Nevada: Trump by 4.3% -> Biden by 0.4% (Biden +4.7%)

6 out of 7 of the close states with changes moved toward Biden.

Now out of these, only one jumped categories in our categorization view: Nevada.

Here on Election Graphs, we usually use a five poll average (to understand when there are exceptions, read the FAQ). You can see that what happened here is that what looks like an outlier leaning toward Trump was replaced in the "last five polls" by an outlier that leans toward Biden, thus moving the average significantly toward Biden. But this leaves us JUST BARELY on the Biden side of the fence. The next poll in Nevada could easily flip the state back to the red side. Or it could show that this was no outlier, but a harbinger of a new trend. Too soon to tell.

Also relevant, based on the historical performance of Election Graphs averages since 2008, the 0.4% Democratic lead here still only translates into a 41.1% chance of a Biden win in the state. More often than not since 2008, when Democrats have led a state with this small an Election Graphs average, the Republican has actually ended up winning.

But this does still flip Nevada to Biden's side on the Categorization view, and this plus the movement on all the other states improve Biden's odds on the two probabilistic views. Here are the overall summaries as of now:

Lets look at some comparisons with last time:

Biden Win Odds:

  • 2023-03-25: Between 17.8% and 25.8%
  • 2023-04-24: Between 24.9% and 32.3%
  • 2023-05-21: Between 34.9% and 35.9%

Here are the two odds charts representing the extremes of how correlated the states might be:

The win odds show Trump continues to have an advantage. But it has been slipping away since November, and while Trump having approximately a 2/3rds chance of winning is decently better than a coin toss, it is very much still in the range where either side has a reasonable shot, and anything could happen.

Categorization Trump Best / Expected / Biden Best:

  • 2023-03-25: Trump+96 / Trump+18 / Biden+162
  • 2023-04-24: Trump+96 / Trump+18 / Biden+162
  • 2023-05-21: Trump+96 / Trump+6 / Biden+162

Here is the trend chart for the categorization view:

There has been no change since four weeks ago to the inventory of states where the margin is less than 5%, which the categorization view imagines as being able to potentially go either way, so the two best cases remain the same. Nevada flipping to Biden moves Trump's "expected" case here from winning by 18, to winning by 6. But the huge range between the best cases shows this is anybody's game.

Tipping Point:

  • 2023-03-25: Trump by 0.7% in Wisconsin
  • 2023-04-24: Trump by 0.5% in North Carolina
  • 2023-05-21: Trump by 0.2% in Wisconsin

So like the others, the tipping point (the margin in the state that puts the winning candidate over the top) shows Trump's lead continuing to deteriorate.

OK, with the comparisons with four weeks ago done, let me introduce something I added to the site since my update four weeks ago. Namely comparisons with 4 and 8 years ago. Here is one of the two new charts:

This compares the "expected case" of the categorization view from this cycle, to where the final two candidates were in 2016 and 2020 the same number of days before election.

What we see is that Trump is running significantly stronger now than he was either in 2020 when he lost to Biden, or in 2016 when he beat Clinton.

There is a lot of talk among Democrats of how in a Biden vs Trump rematch they "won in 2020 and know how to win again". I would simply caution that this is not 2020. Trump is polling better than he was then, and Biden is polling worse.

Trump may be doing worse in my models now than he was in November, but he is still doing better than he EVER did here on Election Graphs in either the 2016 or 2020 cycles. Election Graphs never had Trump in the lead in either of those election cycles.

I have him ahead right now.

Trump should not be underestimated.

I'm still not spending much time on any combination besides Biden vs Trump, because that is still the only combination with enough state level polling to be able to feel confident about the national picture.

Biden vs DeSantis is getting close though, and we have a view where you can compare candidate combinations. On that view you can see that in the last few weeks for the first time since 2021 DeSantis looks like he fares better against Biden than Trump does in the Independent States Probabilistic View.

But the data for everything other than Biden vs Trump is still sparse.

Biden vs Trump has 15 states with 5 or more 2024 polls, including ALL the close states, meaning that we have poll averages that do not rely on "filling out" the average with old election results.

By comparison Biden vs DeSantis only has 5 or more polls in Georgia, Arizona, and Florida. Three states. That's it. So treat that combination with a big grain of salt until there is more polling.

If you want to look at that matchup yourself anyway, here is the Biden vs DeSantis summary.

As for ANY other combination besides Biden vs Trump and Biden vs DeSantis, there is so little data it is not worth looking at, unless you are specifically curious about the handful of states that polled a specific combination, rather than trying to get any insights about the national picture.

But you can explore all the combinations you want on the 2024 Electoral College page, and see some charts I haven't highlighted on the blog yet, as well as being able to click through to individual states to see individual polls in each state for each combination, etc. I encourage you to explore.

In the mean time, I'll close with the updated national map:

534.0 days until polls start to close on Election Day 2024.

Slow Trump Weakening

It has been been just over four weeks since my last blog post.

Since then on a state level there have been 6 Biden vs Trump polls (NC, IA, PAx2, AZ, MI) along with 5 Biden vs DeSantis polls (NC, IA, PA, AZ, MI), and 1 Biden vs Pence poll (NC). If you want updates on each and every poll as it comes out, as well as daily summaries, follow Election Graphs on Mastodon. Right now the polling volume is still pretty low. It will accelerate as the election gets closer. Eventually there will be tons of new polls every single day. But not for a long while yet.

Honestly the situation isn't all that different than it was four weeks ago. There has been some movement toward Biden, but not a really dramatic change. Just Trump slowly getting a bit weaker. But it has been four weeks, so it seems worth looking at things again.

Let's start by looking at how the "spectrum of close states" has changed:



What changes do we see?

  • Michigan: Biden by 2.4% -> Biden by 2.2% (Trump+0.2%)
  • Pennsylvania: Biden by 1.6% -> Biden by 1.8% (Biden+0.2%)
  • North Carolina: Trump by 1.9% -> Trump by 0.5% (Biden+1.4%)
  • Arizona: Trump by 4.1% -> Trump by 1.8% (Biden+2.3%)
  • Iowa: Trump by 11.0% -> Trump by 10.0% (Biden+1.0%)

So 4 out of 5 states with Biden vs Trump polls moved toward Biden.

None of these changed the picture in my categorization view, but  this does adjust the probabilistic views a bit. Here is the new overall status summary:

Since ties would almost certainly end up going for Trump in the end once it got thrown to the House voting by state delegation, rather than confuse things by having to add those up, I'll look at the odds for Biden:

  • 2023-03-25: Biden win odds between 17.8% and 25.8%
  • 2023-04-24: Biden win odds between 24.9% and 32.3%

As always, these are "if the election was held today", which it isn't. This still shows Biden as an underdog in this matchup. But doing better than a month ago.

The odds charts:

As usual, the Independent States chart is more dynamic, but these represent extremes, and the "truth" is somewhere in between. And both views still show Trump peaking right around the 2022 midterms, and slowly deteriorating ever since.

Some of that downward movement is after his indictment in NY, but there was even more before that. So at this point you can't really say that event was an inflection point. The already in progress decline just continued.

Oh, and I didn't mention Biden vs DeSantis, or any other combination other than the 2020 rematch, because that is still the only combination with enough state level polling to be able to feel confident about the national picture. Biden vs DeSantis is getting there. But isn't there yet. If you want to look at that combination yourself, here is the Biden vs DeSantis summary.

Not much else to say for the moment. So lets end with the new map:

561.8 days until polls start to close on Election 2024.

Here We Are: Narrow Trump Lead Over Biden

Since the last post a week ago where I laid out how things stood using just the last five presidential election results as a starting point, I've gone ahead and ingested all 214 state level presidential data points that are already out there into Election Graphs .

The main thing this tells me? Yes, it is still "super early" to look at 2024 match ups. All kinds of things will change between now and November 2024. But even last year in 2022 there was a lot going on in the polling. So maybe start earlier for the 2028 cycle. 🙂

Let me start with how things look now, then go back to how things seemed to evolve over the last year.

I will be looking specifically at the Biden vs Trump matchup.

That is by far the most polled matchup, and the only one with enough polling to even remotely be able to claim there is a good picture of what all the close states look like. Second is Biden vs DeSantis, but that lags FAR behind in polling volume. You might be able to see a few things by carefully looking at select states where there is polling, but it is hard to get a meaningful national picture. Beyond those two, other matchups only have a smattering of isolated polls, to the point it isn't even useful to look at them at all until or unless there is a lot more polling.

So where does Biden vs Trump stand right now?

Here come the map, the summary stats, and the spectrum of the closest states:

Bottom line, as of March 25th 2023, the state by state polling shows Trump with an advantage.

In terms of the straight up categorization view it is a slight advantage. Only one state (Wisconsin) needs to flip to put Biden in the lead, and Trump leads there by only 0.7% in my average.

Look more carefully at Wisconsin, and you'll see that the 5 poll average contains 4 actual 2024 polls but still includes the 2020 result (Dem+0.6%), one of the 4 polls looks like it may be an outlier to the Republican side, showing a full 10% Trump lead where all the others show a close race.

So this Trump lead looks pretty tenuous.

At the moment though, our two probabilistic views show this translating into pretty good odds for Trump. 72.2% chance of winning with the Uniform Swing view, and up to 80.6% with the Independent States view. The "true" odds are somewhere between these two depending on how strongly polling errors end up correlating between states.

Why is this the case if the straight up polling gives us a very slight Trump lead?

Mainly because of what I outlined in my January post about the math behind how I calculate those probabilities. To summarize, looking at all the final Election Graphs averages from 2008 to 2020 in aggregate as they compared to the actual election results, for the Democratic candidate to have more than a 50/50 shot at winning, being ahead in the polls is not good enough. They have to have more than a 1.2% lead.

So some of those states where Biden is showing a 1%-4% lead are really pretty uncertain, where some of the states Trump is leading by the same margin are more safe than they seem.

Well, at least that is what you get by averaging out the poll errors covering the last four presidential election cycles. As covered in that previous post, each of those four cycles actually looked pretty different in terms of how the errors played out. We have no idea what the 2024 errors will look like. But starting with the range of errors over the last few cycles and the variability between those cycles is as good a guess as we have.

Also, everything on Election Graphs is "if the election was today". The election is not today, and lots and lots will happen between now and then. EG does not provide a forecast. We show how things look right now.

With the status above, you would say that team Biden needs to work hard at shoring up Pennsylvania, Michigan, and New Hampshire where he barely leads, and start pushing hard on the barely red states like Wisconsin. And not to take for granted Georgia, Arizona, and Nevada, which even though Biden won them in 2020, right now are looking red. And Florida may not be as far out of reach as it seems, so maybe don't ignore it.

Looking from the opposite direction, Trump should be playing some defense in those barely red states because they could easily slip away if he doesn't, and he should be going hard after Pennsylvania and Michigan. Etc.

Some of this is all pretty obvious. We have most of the same "swing states" that we have had for the last few election cycles. And both candidates will end up concentrating their efforts there and pretty much ignoring the rest of the country. Because that is how races look under an electoral college system.

Now, let's look at how things have evolved over the last year. you can see a variety of different views of this evolution on the 2024 Electoral College page. For now though, I'll look at the Electoral College Margin chart for the Probabilistic Independent States view. It is one of the two extremes I show for the probabilistic views, specifically one that assumes the polling errors between states are completely independent, which to be clear, we know they are not. But this view is the most dynamic in terms of how it changes with new polls, so is nice for showing how the overall race trends over time:

First just an explanation of the graph. The vertical axis is the Trump minus Biden margin in the electoral college. So if it is positive, that means Trump wins, if it is negative that means Biden wins. The dark green line is the median result in Monte Carlo simulations based on the state averages and the probabilities we described in January. The three bands of color are the ranges of outcomes in the 1σ (68.27%), 2σ (95.45%), and 3σ (99.73%) bands. So, roughly, the closer to the center line of this you are, the more likely that outcome, and as the green gets lighter, the less likely that outcome.

At the far left of the chart the ranges shown are just representing the "last five elections" average I posted about last week. Over the course of 2021 and 2022 the averages in the close states slowly filled in with real 2024 polls, pushing the old elections out of the average. As such, it is hard to tell how much of the "movement" seen above prior to 2023 is actually driven by changes in public opinion versus just the averages filling out. By late 2022 though, most of the close states had five or more actual 2024 polls, and we can start to interpret movement as actual changes in the status of the race.

Ignore the jagged ups and down at small scale, and at a very broad level this tells a story of either Biden weakening and Trump strengthening over the course of 2022, OR just the polling catching up to that position after showing a bluer version of things by starting with the 2004 to 2020 averages. Either way though, the trend toward Trump peaked toward the end of 2022. In fact the peak occurs very close to the 2020 midterms in November.

For the moment anyway, that seems to be the moment where things started to go the other way directionally, and started heading in Biden's direction.

To be clear though, Biden is still behind. All three of the views provided by Election Graphs show Trump in the lead at the moment. But more narrowly than he was right before the midterms.

At the moment things point to another close nail biter of a race, just like 2016 and 2020. The picture we see right now shows a race that could very easily go either way if the election was today, let alone allowing for all the things that could happen over the course of more than a year and a half.

Biden hasn't even officially announced his candidacy yet. We don't know if Trump will be indicted, and if he is, if that will help him or hurt him. We don't know if others like DeSantis will officially run, and how they will do. And who knows, one or both of the front runners could end up having health issues that impact their ability to run. You never know. Plus all the normal back and forth of national and international events.

We have a long way to go.

Specifically, we have 591.0 days until the first polls start to close on Election Day 2024.

Keep checking back here for updates on how things evolve!

Oh, and if you want to see all the notifications for new polls as they get entered, and the status changes those trigger, as well as daily summary stats of the overall situation, follow Election Graphs on Mastodon too:

[Edit 2023-03-25 23:42 – Slight wording change.]

2024 Begins: Five Election Average

Welp, here is the big moment. I have just opened up the 2024 Electoral College section of the site. No actual 2024 polls are reflected yet. That will come over the next few days.

It may still be 598.2 days until polls start closing on the Election 2024 cycle, but believe it or not there have already been quite a few state level 2024 polls. I'll start getting those into the system shortly.

But first, Election Graphs has always used an average of the last five election results as the "baseline" when polls aren't available in a state. So the traditional "so it begins" marker for Election Graphs has been the electoral map using the average of the last five elections for all 50 states, DC, and the ME/NE congressional districts.

So here we go…

Some folks might argue that a better baseline would just be the 2020 results. After all, five elections means averaging out the results from 2004 to 2020. The world of 2004 does not really resemble the world of 2024. That was almost 20 years ago!

There is definitely a point to that. But we're basically just looking for a starting baseline here.

All the states that are remotely close will get real 2024 polls relatively quickly, certainly before we really get into the thick of the 2024 campaigns. So we will get a sense of if states that have seemed to be trending in one direction or the other over those 20 years will continue that trend, or revert to the mean a bit.

Meanwhile, the states that have booked 20 straight years of massive wins for one party or the other may get polled less and take longer to fill out with real 2024 polls, but they are also much less likely to end up being competitive. It would take an upheaval of almost unimaginable proportions for Wyoming and Oklahoma to turn blue, or for DC and Hawaii to turn red.

Based on the five election average, the "battleground" as we start this cycle is made up of all the usual suspects:

These are all the states (and ME/NE congressional districts) where the five election average gives a margin of less than 10% for the leading party.

Immediate standouts are Arizona and Georgia, which over five elections, still are tinted pretty strongly red. One of the questions new polling will answer is if the Democratic wins in these states in 2020 were flukes, or if these should truly be closer to the center of the block above as true swing states for 2024.

Similarly, in the five election average Ohio and Florida are red but not by a lot. They look like they could be in contention. Democrats won both states in two of the last five elections. But in both 2016 and 2020 Ohio actually turned out much redder. And Florida seems to be on a trend toward the red as well, although not quite as dramatically as Ohio in 2020. Will these states go even further red? Or go back to being swingy states? Iowa is in the same boat. Just smaller.

Election Graphs shows you charts for individual states. Just as an example, here is Florida since 1980.

We usually won't look at that much history as actual 2024 polls start to come in. But you can see the individual election results, and how the five election average has evolved over time.

OK, so what does all this mean in terms of who has the advantage as we start this. Well, we are dealing with a five election average, and Democrats have won three of those five elections, so you would expect them to start with a little advantage given how we defined the starting line, and you would be right:

This shows my traditional "Categorization View" which just takes the averages at face value and gives an expected view where every state falls with the average, and best cases for each party assuming they win all of the "weak" states (where the margin is under 5%).

And what do we have? Democrats with a very narrow lead, with the range of outcomes based on the close states falling one way or the other including victories by both parties.

Also above are the two probabilistic views representing the extremes of how correlated poll errors are between states (completely independent to completely in lock step). These are based on looking at how far off the FINAL Election Graphs averages were from the actual election results from 2008 to 2020. See my last blog post for details.

Anyway, both of these also show a close race with a slight Democratic lead that could easily go either way, with Democratic win odds somewhere between 55.6% and 73.3%. Which leaves Republican win odds between 26.7% and 42.1%. Odds of a 269/269 tie are somewhere under 2.3%.

Keep in mind that Election Graphs does a "nowcast". That's "if the election was held today", which obviously it isn't. And we don't even have real 2024 polls in there yet. So this is even less of an actual prediction for 2024. This is just a starting point. A default look at the playing field before the game actually starts.

I'll be catching up with the actual state level 2024 polls we have over the next few days.

So here we go… Election Graphs 2024 has begun!

598.2 days until polls start to close.

Buckle up.

[Edited 2023-03-19 17:53 to correct the number of days until the election as of the time of the post, which I'd made a math error on.]

[Edited 2023-03-20 01:03 to uncorrect the number of days until the election, because I was actually right the first time, plus to correct the numbers in the probabilistic numbers, because I found an error there as well. Specifically, the Monte Carlo simulation was still using the 2020 electoral college distribution which because of the census was changed for 2024. Fixed now.]

[Edited 2023-03-21 05:55 UTC to correct a typo in the last note above.]

Prepping the Math Stuff for 2024

OK, in my last post I mentioned that the next thing to do was:

Finish up the calculations to use all four elections from 2008 to 2020 as the baseline "how well did Election Graphs averages do compared to the final results" data I use to try to generate odds from the polling averages and do a blog post about that.

So I guess it is time to do that.

OK, actually, it is way past time for that. I had hoped to have all this done by midterms, but I ended up spending most of the time I would have spent on that helping to do things like put out campaign signs for my wife's campaign. She won. Other things took up my time too. Anyway. I was delayed. But lets get this done…

Let's start with a simple scatterplot showing every state (and the DC and ME/NE congressional districts) from 2008 when I started doing this through to the 2020 results. Each data point will have the FINAL Election Graphs average on the X axis, and the ACTUAL election results on the Y axis:

As you would hope, these are nicely correlated at this scale. If EG's averages were always exactly right, every point would be along the black diagonal line. Of course, polls don't work like that, and even poll averages don't work like that. There is a vertical spread due to the inherent randomness of polling. A pretty wide spread actually.

But also if you look carefully, you can see that on the right side of the graph there are more points above the line, and on the left there are more points below the line. This means that there is also some bias here. Specifically a bias where the Election Graphs polling average tends to UNDERSTATE the magnitude of the winner's margin.

Let's do a transformation on the graph to try to look into the patterns a bit more deeply though:

That just looks like a messy colored blob initially. But what did I do here? I just transformed things by subtracting out the diagonal. Instead of looking at the actual election results vs the Election Graphs final margin, I look at the Delta… how far off the election results were from the Election Graphs final margin, vs the Election Graphs final margin. So, for instance, if on the top graph we had a point where the EG average was a 10% Republican lead, but the actual result was that the Republican won by 15%, that would show up as x=10%, y=5%.

OK, but can we say anything at all about this blob? Is this just two things that show no useful relationship at all?

Well, in this post from 2019 I actually looked at this before. So let me just quote a bit from there:

Before going further, let's talk a bit about what this chart shows, and how to interpret it. Here are some shapes this distribution could have taken:

Pattern A would indicate the errors did not favor either Republicans or Democrats, and the amount of error we should expect did not change depending on who was leading in the poll average or how much.

Pattern B would show that Republicans consistently beat the poll averages… so the poll averages showed Democrats doing better than they really were, and the error didn't change substantially based on who was ahead or by how much.

Pattern C would show the opposite, that Democrats consistently beat the poll averages, or the poll averages were biased toward the Republicans. The error once again didn't depend on who was ahead or by how much.

Pattern D shows no systematic bias in the poll averages toward either Republicans or Democrats, but the polls were better (more likely to be close to the actual result) in the close races, and more likely to be wildly off the mark in races that weren't close anyway.

Pattern E would show that when Democrats were leading in the polls, Republicans did better than expected, and when Republicans were leading in the polls, Democrats did better than expected. In other words, whoever was leading, the race was CLOSER than the polls would have you believe.

Finally, Pattern F would show that when the polls show the Democrats ahead, they are actually even further ahead than the polls indicate, and when the Republicans are ahead, they are also further ahead than the polls indicate. In other words, whoever is leading, the race is NOT AS CLOSE as the polls would indicate.

In all of these cases the WIDTH of the band the points fall in also matters. If you have a really wide band, the impact of the shape may be less, because the variance overwhelms it. But as long as the band isn't TOO wide the shape matters.

Now, back in 2019, at this point I jumped directly into looking at the pattern based on the combination of all the data from 2008 to 2016. Rather than doing the same thing now, but just adding in 2020, I think it is actually instructive to take a bit of a detour to look at each of the four election cycles separately.

First lets look at just the 2008 data points:

OK, the circles are just the individual 2008 data points, but what is the rest, what have I done here?

I've constructed "envelopes" using windowed averages and standard deviations.

Specifically, I am looking at windows with a 5% radius. At every value for the polling average at 0.1% increments, I look 5% in either direction (so a window 10% wide) and find all the data points within that window, then if there are at least five points, I calculate the average and standard deviation of those points. The 5% is of course just an arbitrary round number, as is the 5 data point minimum.

The bold line in the center is the mean, the next lines out are 1 standard deviation from the mean (about 68.3% of the data points should be inside these lines), and the next lines are 2 standard deviations from the mean (which should contain about 95.4% of the data points).

So the final chart here shows both the general trend in how far off final election results were from the Election Graphs average given where the average was, but also just how variable those results are.

For 2008, we can now see pretty clearly that this looks like "Pattern F".

When Election Graphs showed a tie, on average the actual election results were a Republican win by 0.2%. That's pretty close to a tie too. So not a lot of bias in one direction or another at the center. There is a BIG window though. If the EG average was a tie, the 95% confidence interval goes all the way from Democrats winning by 6.8% to Republicans winning by 7.3%.

But there is also definitely a trend where in cases where when Republicans were ahead in the EG average, they actually tended to win by MORE than the EG average, and if the Democrats were ahead, THEY tended to win by more than the EG average.

OK, you get the idea, so lets look at the rest of the election cycles Election Graphs has covered.

I've made the scales the same on all of these to make it easier to compare the cycles. While these all have the same high level diagonal pattern, the detailed shape of the curves is very different cycle to cycle, both in terms of the central average, and pattern of the variances.

Lets concentrate on those central curves and put them all on one chart…

You can immediately see the two elections Obama was in (the red and goldish ones) clump together, and the two that Trump was in (the light and dark greenish ones) clump together too.

They all have the general diagonal shape (Pattern F), but it stands out how, at least near the "Election Graphs average near zero" area which by definition are the states that matter in close races, how much more the poll averages underestimated Trump in 2016 and 2020, compared to a more neutral 2008, and an underestimation of Obama in 2012. Maybe these all have the Pattern F shape, but the slopes are different, and even more importantly, they are shifted vertically. They don't all just go through the origin.

And of these four years, 2020 is the only race yet where the average curve is above the zero line for its full length. For 2008 and 2012 if the Democrats were ahead in the average, on average they would do even better in the election than the polls indicated. In 2016, that was only true if the Democratic poll advantage was more than a 12.8% margin. But that didn't happen at ANY part of the range in 2020. At every range for the Election Graphs average, on average the Republicans did better than the average indicated, and the Democrats did worse. (Note that is not true for every single data point, just for the trend line.)

Does this mean anything for 2024? Quite probably not. As they say "past performance is not a guarantee of future results". It is unclear what all the reasons are for the shape of these error curves, and if pollsters are actively working on "correcting" things in the next cycle. Were the two Obama curves more "normal" and there is something unusual that just makes it harder to poll races with Trump in them? Or were the Obama races unusual too, and "normal" is somewhere in between? Or was there a systematic error that pollsters have a handle on now, and 2024 will just be a nice flat line? Or none of the above?

It would be nice if I had this kind of data from before 2008, but since that is when I started doing this, I don't. Also of course, you could argue that the world of 2024 is so different to 2004 that it wouldn't really be meaningful to look at that anyway. For other things on Election Graphs I use five election cycles (20 years) as a baseline, so I might still use older data if I had it. At least 2004. In general, I'd love to see how these kinds of curves have varied over even longer time periods. Oh well.

As it is though, since we don't know which way the errors will go in 2024, the best option I have available is to create this same sort of envelope using all four available election cycles.

This is what that looks like:

Doing this you lose the specific distinctiveness of the four presidential election cycles going into this. Instead, you essentially fold in some information about just how much polling accuracy has varied cycle to cycle.

Now, the "core" of Election Graphs has always been the dead simple method of taking the polling average at face value and classifying any states where the margin is less than 5% as states that could go either way, and presenting that range of possible results.

But starting in 2020 I experimented with producing probabilistic results too. Those probabilities were based on the 2008-2016 version of the chart above, and for 2024 I'll be using this 2008-2020 version.

The key is that for every value of the Election Graphs average, we have numbers for the mean actual election result (the EG average plus the delta), as well as an associated standard deviation. From this, we can construct a chart showing for each value of the Election Graphs average, the chances of a Democratic win and the chances of a Republican win based on the historical data. (As usual, since Election Graphs operates off the margin and not raw support numbers, if a 3rd party is ever in contention, this method falls apart.)

So what does that look like?

OK, that's nice at the same scale as all the other charts, but lets zoom into the critical central part here.

OK, given my methodology, this is a bit bumpy. I should probably smooth it out a bit or do a logistic regression or something. But given the levels of uncertainty we are talking about, I probably won't bother. I like a little bumpiness.

Anyway, this is the critical graph. And yes, this does smell a little of "unskewing" the polls. Don't worry though, ElectionGraphs will continue to show the straight up averages.

But what we see here is that over the last four election cycles, there HAS been a tendency to underestimate Republicans. Yes, as discussed above, all four cycles look different, and 2008 and 2012 look distinctly different than 2016 and 2020. But taken as a whole, the Election Graphs poll averages have underestimated Republicans.

So lets look specifically at a few data points on this chart.

Now, technically speaking because the way Election Graphs calculates averages (see FAQ, which I'm just realizing I need to update as well) there can't be exact ties in the Election Graphs average. But if there was an exact tie according to the curve above based on historical data, the Republican would actually have a 62.4% chance of winning, and the Democrat only a 37.6% chance. In order to have a better than 50% chance of winning, the Democrat would need to lead by at least 1.3%.

Looking at this another way, for the Democrat to have a better than 95% chance of winning, they need to be leading by 7.1% or more. For the Republican to have a better than 95% chance of winning, they only have to lead by 3.8%.

Again, this is based on comparing the Final Election Graphs averages to actual election results from 2008 to 2020. There is a chance 2024 looks nothing like the last four elections. Polling may be better. Polling may be worse. Polling may underestimate the Democrats this time rather than underestimating the Republicans. We just don't know. So looking at the four cycles of data I have so far is the best I can do…

Anyway, that is the curve I will be using to make my "probabilistic" views for 2024. This will not impact the traditional "categorization" views at all. They will remain as they always have been, classifying any state with a margin less than 5% as a "weak" state that could go either way. No "unskewing" there.

But I will allow the probabilistic views to take into account that based on this historical data, the Democrats have to be further ahead to have an even shot, etc.

Also important to note, is that these are the odds on a PER STATE basis (and CD for ME and NE). You can't apply these numbers to national polls. It just doesn't work that way. For that, I'll once again be doing Monte Carlo simulations using the state poll averages and the odds on the chart above.

One thing none of the above takes into account though is trying to estimate how correlated errors are between states. If errors were just completely uncorrelated, then when you run your simulation, you just roll the dice for each state. Distributions end up a bit narrower. But if things are completely correlated, so that if one state underestimates the Republican, they all do, then you essentially roll once to see how far off ALL the states are, and you end up with a much wider distribution with higher odds for the tails.

The reality is somewhere in between. I haven't had a good method for modeling "somewhere in between". So in the 2020 cycle, after a brief time mentioning only the fully independent version (which was a mistake, for which I was indirectly called stupid by Nate Silver), I ended up just showing both extremes. This wasn't entirely satisfactory either though. But at the moment I don't have a better idea.

If you are a stats and modeling person who wants to help me properly model the right degree of correlation between errors here, please get in touch. I'd love to learn more and do better.

That applies to everything else in the analysis above as well. I am well aware I am doing a few things that may not be exactly the right way to do things. I don't think anything is outright "bad", but I recognize there may be better ways of accomplishing what I am trying to accomplish, and maybe I'm wrong and some things ARE just bad. If so, I'd love to hear about it and learn… as long as you can give me that feedback nicely and gently, rather than being mean about it. Thanks.

OK. I guess that is it for the "preliminary math stuff".

Next up is standing up the actual 2024 page with only the previous actual election data (no 2024 polls yet) to define a "starting point". And updating the FAQ and things like that.

I'll post again once that is done, and before I start feeding it actual state polls for 2024 (of which there have already been quite a few).

Knock Knock… Is this thing on?

Well well. I haven't posted here in a long long time. The last thing was an update to my November 3rd 2020 Live Election Graphs Results Updates post on January 7th 2021 with the final official electoral college results after the craziness of the January 6th United States Capitol attack.

I had intended to post some sort of detailed 2020 post mortem like I did for the 2016 cycle. This site did pretty well again, in the same ballpark as some of the big guys. Roughly. I didn't ever sit down and look through all that in detail. So that post never happened. Frankly I was tired and exhausted with the election stuff, and aside from a few very minor things I would occasionally pick up and look at, I immediately fell into the mode of letting Election Graphs lie dormant until it was time to start thinking about 2024.

Welp, we are now only 78 days until the mid-term election, which is traditionally the starting gun for the next Presidential election, and we have at least one candidate making noise that they intend to make their run official even before then. So I think it is probably time.

There is no Election 2024 content live on this site yet. I've generally tried to start updating the site in earnest right around the mid-terms, although sometimes it has ended up being a few months later.

I have a to do list which I am starting to work through though, so here are some of the highlights:

  1. This post waking up the site, and announcing the preparations for 2024 are under way.
  2. Finish up the calculations to use all four elections from 2008 to 2020 as the baseline "how well did Election Graphs averages do compared to the final results" data I use to try to generate odds from the polling averages and do a blog post about that.
  3. Stand up the basic Election 2024 national summary, state pages, national comparison, and state comparison pages using only the averages of the 2004 to 2020 election results as the "starting averages" and do a blog post about that.
  4. Get all the state level 2024 presidential polls that have already been done (yes, there are quite a lot of them) entered into my system and see where things are as of now and do a blog post on that.
  5. Start regularly scanning for new state level 2024 presidential polls and adding them as they are released and doing periodic blog posts when there are any interesting changes.
  6. Set up the delegate race part of the site as well, although there is still quite a long time until the first delegates get allocated.
  7. If I have time, start adding a couple of enhancements to a few of the graphs I have been thinking about, and maybe some new maps or charts that were not there for the 2020 cycle.

I don't want to hold up anything for that last one. I have a few ideas in my head, but nothing solid, the ideas may or may not work in real life, and time is at a premium, so I may or may not have time to do them anyway. But one thing at a time.

And the one thing right now is to post this, put my stake in the ground as to what I am doing next, and get to it…

Welcome to the start of Election 2024 coverage here on Election Graphs!