What happened in Michigan?
Crossover voters, early voters, or money? Plus: the censored DNC autopsy chapter, hallmarks of AI writing, who's #1 in AI use, the youth gender gap widens again
No. 411 | August 7, 2026
🗳️ Michigan’s Senate primary was a win for El-Sayed but a disaster for pollsters
The highly-anticipated Michigan Democratic primary ended up in a photo finish between Abdul El-Sayed and Haley Stevens. The pre-election polls showed El-Sayed with a solid double-digit lead. What happened?
Yes, primaries are harder to poll than general elections. Because there are smaller ideological differences between candidates, primary voters are much more up for grabs than general election voters. That makes the races themselves more volatile, with last-minute shifts the rule rather than the exception. It’s also harder to weight primary polls. We don’t have reliable targets for the things most correlated with primary vote choice — ideology or what kinds of candidates the voter has voted for in the past (recalled primary vote is notoriously unreliable, unlike general election vote).
While it’s true that the Michigan polling error was well within range of a normal primary polling error, Michigan was also heavily polled. Most primary races are polled once or twice, if at all. Every recent poll agreed on El-Sayed’s margin.
People will also say that pollsters just polled “the wrong electorate,” but that’s just hand-waving when you don’t get specific about how exactly the electorate was wrong.
One possibility is that some Republicans simply crossed over into the Democratic contest under Michigan’s open primary rules, something that could easily be missed if pollsters were targeting self-identified Democrats. When partisans cross over like this, they tend to vote for the candidate ideologically closest to them, not the more extreme candidate who’d be easier to beat in a general election.
There’s modest evidence of this in the results. Once you control for 2024 vote — county-level Trump vote is highly correlated with the Stevens vote — counties with a bigger shift towards Democrats in primary vote composition tended to vote for Stevens more, but it’s not that statistically significant.
One thing the charts clearly do show is robust suburban turnout — especially in Stevens’ home base of Oakland County, but also in Macomb County, where Stevens unexpectedly won. Turnout gains relative to past Democratic presidential vote (and 2018) were as strong if not stronger in the pro-Stevens suburbs than in El-Sayed’s base in Ann Arbor, Lansing and Grand Rapids. Turnout was quite strong across the board — whether this was Republicans or lower-information Democrats. And in a place like Michigan, this benefited the moderate.
On his new Substack, Jon Cohen does what you should do when evaluating polling error: look at crosstabs you can verify against real results. Actual votes don’t come with a note about the voter’s age, race, or gender. What we do know is where a vote was cast and by which vote method. Where one poll ended up missing was in the mail and early vote, which it showed tied but which actual returns showed went handily for Stevens. An earlier Glengariff Research poll (showing a Stevens win) showed her winning handily among Black voters, a result that many ridiculed, but which was spot-on when you look at precinct results.
Or, maybe we should just look at Haley Stevens’s performance and say to ourselves, “That’s what the money is for!” An AdImpact analysis found this to be one of the most expensive Senate primaries in history, with a 9-1 spending edge for Stevens.
🔵 A new window into the Democrats’ 2024 defeat
The New York Times published the disputed chapter omitted from the Democratic Party’s official 2024 post-election review. Two things that stand out:
First, Biden’s favorability in internal polling was no better than any public polling.
And Democrats saw massive turnout declines among base groups that went hand in hand with declines in support. These were concentrated in non-battlegrounds, but hint at a bigger persuasion problem.
🤖 It’s not just “It’s not X, it’s Y”
What are the hallmarks of AI writing? We know all about “It’s not X, it’s Y.” But it’s other things too, like longer words, scientific vocabulary, and lists of three items.
Meanwhile, Washington, D.C. outranks California and the rest of the country in Claude usage.
👫 The youth gender gap is wider than ever
Dan Cox is out with a new look at the youth gender gap — which has grown wider since 2024. Young women have dropped from 44% to 27% Republican just in the past 2 years while young men only declined by 4 points.
