I thought I'd take a look at margins in abortion referendums since Roe was overturned to see where they might fail.
Ohio: Trump +8, Abortion +13 (21 point gap)
Michigan: Biden +3, Abortion +13 (10 point gap)
California: Biden +29, Abortion +34 (5 point gap)
Kentucky* : Trump +26, Abortion +5 (31 point gap)
Kansas* : Trump +15, Abortion +18 (33 point gap)
*Referendum was to declare NO right to abortion, meaning that pro-life was the YES vote.
The average gap was 20 points.
Based on a crude average gap, the pro-life side could win a referendum in the following states:
Now, if we remove the NO right to abortion referendums (and only include the OH-style ones), the average difference drops to Abortion +12. That map might look like this:
That map certainly looks off in places. Obviously, Montana and Mississippi wouldn't vote virtually identically in an abortion referendum for example.
One other trend that seems to be apparent is that the referendum runs closer to partisanship in blue states. That actually makes sense if you assume that all Democrats are voting for abortion on these. If the same percentage of Trump voters defect in all states, they will look closer to partisanship in blue states. So, let's look at the ratio of the pro-life percentage to Trump's percentage.
California: 96.5%
Michigan: 90.7%
Ohio: 81.6%
Kentucky* : 76.8%
Kansas* : 72.9%
Even factoring in the number of Republicans, there still seems to be a wide variation. But, it seems clear that it matters which side is YES and which is NO. The overall average percentage is 83.7%. In order for that to project a victory for the pro-life side, Trump would need to have gotten over 59.7% of the vote in 2020. That map would look as follows:
I'm still not really satisfied with any of these maps as predictive tools because of differences in religiosity combined with a stark trend towards these underperforming in red states. But, that wasn't really the case if we look at 2018 and 2020 referendums. The pro-life side outran Kay Ivey in Alabama in the 2018 midterms and outran Donald Trump in Louisiana in the 2020 general election. And those were even NO right to abortion referendums where the pro-life side was the YES vote.
I think another thing that could be looked at would be the percentage of Democratic votes coming from non-white voters. One thing that California, Alabama, and Louisiana (where the pro-life side has held up relatively well to partisanship) all have in common is majority non-white Democratic electorates. High numbers of white Democratic votes is often an imperfect sign of social and cultural liberalism, so that might be a good starting point.
Turnout clearly matters too. The biggest overperformance for abortion was in an August 2022 off-cycle election in Kansas. The counter-examples above were all simultaneous with midterm or presidential elections. One further difference in Kentucky's (which is ultimately the one I still am having the hardest time making sense of) was that it was the only one that took place in a place where abortion was illegal as the status quo at the time of the election. As a NO vote in the referendum did not actually legalize abortion, some pro-life voters may have not seen the point to adding it to the state Constitution. Likewise, pro-choice voters may have been even more energized because of the fact that abortion was already illegal in the state. (The opposite effect was probably true in California- threats that abortion could be banned in the state may have seen unrealistic).
Ultimately, here's my guess of how a right to abortion referendum (with YES being the pro-abortion side) would play out in a high turnout presidential election: