iiinova — where your digital self keeps evolving

1938 is the end of a dramatic realignment period. The years 1966, 1982, and 2006 were previously discussed in this thread:

010

These mismatches then show up as outlying data points on the regression of *change* in House popular vote versus *change* in national primary margin. But all the other years fall nicely along one tight line.

Regression of change in house popular vote versus change in national primary margin from 1939-2022.

110

But because four of the epochs have very similar slopes, we can transform the chart by looking at the *change* in both variables. And when we do that, the epoch divisions transform into mismatches between the two time series.

Graph showing the *change* in margin for both House popular vote and national primary between 1918 and 2022. From 1942 onwards, they closely follow each other, except at epoch changes (1938, 1966, 1982, 2006) and one more potential outlier - 1970.

120

Some of these divisions are obvious on the original chart, and some are not.

Graph showing the national primary margin and the House popular vote margin from 1918 to 2022. The are right on top of each other from 2006 to 2022, but before that they have a significant but varying gap.

100

... this plot hides within it four epochs with nearly the same slope (and one epoch that is a mess, 1918-1938). It's kind of like a Simpson's paradox (not sure if it counts since it doesn't reverse the sign of the slope).

111

The Economist @economist.com · 24/09/2026

There's still more to show in the relationship between primary votes and the House popular vote! You can plot the data in graph below as a scatterplot. But a) it's obvious from the original graph this is a bad idea and b) it's not too helpful if you want to predict things. But...

A scatterplot of House popular vote margin versus national primary margin in midterm elections from 1918 to 2022. While there is a decent regression for the data, they fall in a wide band, such that any prediction you try to make will not be helpful. For instance, this year we would expect a House popular vote margin between about R +15 or so and D +12 based on the primary margin of D +11. Not useful at all!

244

Reposted by Daniel Donner

Deaths now outnumber births in reliably Republican counties for the first time since at least 1982. We explain how this may already be affecting voting patterns in America bit.ly/471apDL

85881289

This is why diesel prices - which selectively affect Trump voters and rural areas in particular - matter for the election. www.kff.org/public-opini...

Chart showing that 59% of Republican rural voters who are dissatisfied with Trump on the economy are absolutely certain to vote, compared 79% of Republican rural voters who approve of Trump on the economy.

085

From the NYT/Press Herald/Siena Maine poll crosstabs: Maine likely voters prefer Democratic control of the Senate by 55-43. Including... 13% of those who are voting for Susan Collins. www.nytimes.com/interactive/...

8366

If we instead look at the data available in July of election year, both for primaries and generic ballot polls, the predictions based on primaries are much better than the generic ballot polls.

0141

We can see both measures do a very good job since 2010; in the 20th century, the generic ballot polls were a little worse, especially mid-century. However, this uses the *final* generic ballot polls, which aren't available until election eve. What about earlier?

Graph showing house popular vote, November generic ballot polls, and the predicted house margin from primary votes.

1120

The biggest misses can be reasonably explained by this thread from yesterday:

1160

Carl Quintanilla @carlquintanilla.bsky.social · 24/09/2026

If we could go back in time to use the national primary margin to predict the midterm House popular vote with the data available at that time, how well would it have done? Pretty well overall, with some significant misses like 2006. (Note we see no warning signs of a miss like 2006 for this year.)

Graph showing the actual House Popular vote margin compared to that predicted by the primary margin. Major trends match, and about half the time the prediction is very close.

211036

Reposted by Daniel Donner

WSJ: The AI build-out is on track to become the biggest economic bet in US history, dwarfing the investments made to fund other huge US infrastructure projects — such as the railroads, the highway system and the plumbing for the internet.” (via Jefferies)

37503181

It was defined by the voting behavior of the states - basically split into this group and the Upland South (next in the thread), leaving out Florida, which likes to do its own thing.

160

Finally, in the Northeast, Democrats have long outvoted Republicans in primaries, but never so much as they have done this year. The 1930s surge is still prominent but subsequently subsided as in the Midwest.

Primary votes from 1918 to 2022 in ME, NH, VT, MA, RI, CT, NY, NJ, PA, MD, and DE.

1335

The Midwest has flipped back and forth for a century, with no party dominant for decades at a time. The Progressive Era end looms large, with Democrats quadrupling their primary vote total from 1934 to 1938, but then falling back again. Recent increases in vote totals are more muted but still clear.

Primary votes from 1918 to 2022 for IA, MN, WI, IL, IN, and OH.

1282

The rest of the West also saw the Democratic primary vote rise from nearly nothing in 1934, but then it continued to keep pace with population. Dem numbers have skyrocketed since 2014, outpacing increases in Republicans and this year yielding the highest collective Democratic margin since 1974.

Primary votes from 1918 to 2022 from HI, AK, WA, OR, CA, NV, AZ, NM, and CO.

1262

Outside of the South, only one other subregion shows Democratic fortunes fading to any extent in recent years, and that is in the Northern Plains and Northern Rocky Mountains. In these states, 1934 jump-started the fortunes of the Democratic party, followed by a plateau.

Primary votes from 1918 to 2022 in UT, ID, MT, ND, and SD. Republican votes have surged in recent years, while Democratic votes have remained at roughly the same levels since the 1940s.

1274

The Upland South, on the other hand, does not show a recent Democratic renaissance. The decrease in primary margin from 2018 may be an artifact due to chance, as there are only two states in this group which elect their governors in midterm years, yielding a less consistent dataset.

Primary votes in the Upland South: WV, KY, TN, AR, MO, and OK from 1918 to 2022

2333

Note the complete absence of Republican primaries in the Core South until mid-century. We can see the surge in voters in 1966 and the backlash following in 1970. Then we see the gradual bleeding of Democratic voters to the Republican side of the ledger for a generation as foretold.

1502

Yesterday we showed how important primary turnout can be on a national level. But there's interesting things going on at the regional level, too. Most notably, in the core of the South, Democrats *outvoted Republicans* for the first time in decades in the primaries! (thread)

Total primary midterm votes in the core of the South from 1918 to 2022 showing a recent Democratic surge. States include VA, NC, SC, GA, AL, MS, LA, and TX.

434299

These graphs are for midterms only though.

100

Keep reading the thread to find out!

110

Interestingly, the two states with the next-best track records in this regard are... NV and SC. Is this a coincidence? I don't know! Maybe being an early presidential primary state makes voters more attuned to national politics.

150

And the regression here is also super tight... but it predicts a 2026 House popular vote of D +1. Either both of these predictions will be pretty wrong, or one will be very, very wrong. This illustrates how correlations in politics can fall apart, even if you have beautiful r-squared values.

Regression with 9 data points and R2 of 0.96

151

But Iowa has an even longer track record of the primary margins nearly matching the behavior of the House popular vote!

A graph showing the changes in primary margin in midterm years match the changes in House popular vote margin for Iowa. While the magnitude differs, the direction and relative change is nearly identical.

150

This gives us a *very* tight regression which predicts a midterm result this year of D +9.

Regression with 6 data points and R2 of 0.97. The 2026 value is beyond the edge of the regression to the right.

140

G Elliott Morris @gelliottmorris.com · 23/09/2026

Yesterday we showed you how the sum of primary votes in all the states can predict midterm election outcomes surprisingly well. But that's not all! There's also some individual state primaries that appear to be bellwethers. Here is New Hampshire since 2002. Note the two different axes.

A graph showing the changes in primary margin in midterm years match the changes in House popular vote margin for New Hampshire. While the magnitude differs, the direction and relative change is nearly identical.

2213

Reposted by Daniel Donner

Yes. here you go. Trump’s approval is significantly below where it was in 2018

31811172

There just happen to be some pretty good explanations! bsky.app/profile/donn...

000

Outside of the South, however, the two measures, on average, follow each other a little better, with the exception of the most recent midterms.

080

And then again in 2006, we see what may be the final stages of ideological realignment, with Texas and Florida shouldering a significant share of the blame, perhaps related to the connections of George W. Bush to those states.

290

The Reagan midterms - 1982 and 1986 - moved quite a few voters out of the Democratic primaries or into the Republican primaries (and not just in the South).

170