Abortion and Politics
States with Anti-Abortion Trigger Laws Score Higher on Measure of Hostile Sexism
But the connection between the two is more complicated than you’d might think
On June 24th, the United States Supreme Court in ruled 5–4 in Dobbs v Jackson that Americans do not have the constitutional right to an abortion. The ruling was historic not only in that it overturned Roe v. Wade, Planned Parenthood v. Casey and, consequently, 50 years of legal precedent–but represents one of the few times where the court ruled to revoke individual rights that it had previously extended.
A number of states were ready for this development. Not only had many created laws explicitly intended to challenge or circumvent Roe (which now effectively have the blessings of the Court), several had laws in place that were explicitly set to activate or be enforced in the event that Roe was overturned. Conservative politicians, activists, and religious groups across the nation praised the decision and broadly heralded the enforcement of these laws as a step in the right direction. But they are decidedly in the minority in that view.
These laws and the Court’s ruling is not a popular move–even within the states with the newly enacted restrictions. In a recent article for the Washington Post’s Monkey Cage blog, Jake Grumbach and Christopher Warshaw used a statistical technique called multilevel regression and poststratification (also called MRP — or Mr. P, when among friends) to estimate support for abortion in each individual state.
They found that, unsurprisingly, abortion attitudes vary quite a bit across US states — but attitudes were still broadly supportive of keeping abortion legal in all but a few. Yet, despite that, 13 states have laws severely restricting abortions that were ready to activate or be enforced in the event that Roe was repealed — and others had bans on the books prior to Roe that are likely to go into effect.
In a follow-up analysis, Sharif Amlani found that many of these states supported abortion access and that no state had a majority of people endorse the view that abortions ought to be illegal in all circumstances.
Inspired by these works, I wanted to investigate an attitudinal structure that I suspected correlated well with the presence of these laws: hostile sexism.
On the individual level, hostile sexism has been shown to be associated with stronger support for President Trump (in 2016 as well as 2020), and more restrictive views on the acceptability, and permissibility, of abortion. Recent work has also shown how interwoven abortion attitudes are with views on traditional gender roles. I’ll discuss potential mechanisms at the end, but it’s possible that state-level hostile sexism could similarly correspond with more restrictive abortion policies.
I used the 2020 American National Elections Survey (ANES) to construct an MRP model that estimated each state’s hostile sexism score and looked to see if it was associated with whether or not the state had an anti-abortion law that would go into effect after the federal protections were repealed.
I found that, of the states with the 15 highest scores, 14 had such a law. Of the states with the bottom 15 scores, in contrast, only 1 (Utah) had a trigger law. Indeed, the mean hostile sexism score was significantly higher in states with trigger laws than in states without it (Though the substantive difference on this point is modest).
In short, states with higher levels of hostile sexism appear to be more likely to have enacted an anti-Roe trigger law which severely curtails access to abortion.
Measuring “hostile sexism” across the US
The concept of “hostile sexism” was originally advanced in 1996 by Peter Glick and Susan Fiske and was conceived and validated as part of a broader “ambivalent sexism battery” (so-called because some of the questions instead tap into the mirroring concept of “benevolent sexism”). Hostile sexism seeks to tap into sexist attitudes that conceptualize women in a cynical, antipathetic light. Different surveys contain different numbers of items to tap into it, but the 2020 ANES does so with the following four:
- Women seek to gain power by getting control over men.
- Many women interpret innocent remarks or acts as being sexist.
- When women demand equality, how often are they actually seeking special favors?
- When women complain about discrimination, how often do they cause more problems than they solve?
With answers to each being a 1–5 scale with answers agreeing (or feeling like they happen often) taking the lowest values, answers disagreeing (or never happening) taking the highest values, with a neutral option in the middle.
Following the convention in studies that use the concept, I took the mean of these four items and rescaled the resulting score so that it lay on a “continuous” 0–1 scale. The image below shows the distribution of hostile sexism scores in the sample. (Or at least the 4,000 respondents who didn’t skip any of those 4 questions or the demographics I used for the model). Areas with greater peaks symbolize scores that were more common. We can see that the vast majority of respondents score below 0.50 on the scale — with most of those being between 0.20 and 0.50.
How do we know that these 4 items are actually measuring hostile sexism–or at least something we’re comfortable labeling “hostile sexism”?
Well, the ANES contains a few other questions that we could expect to be either positively or negatively associated with the concept, such as: How important is it for more women to be elected to political office(5 point scale from 1. Extremely important to 5. Not at all), the degree to which women face discrimination in the US (5 point scale from 1. A great deal to 5. None at all), whether it is better for men to work and women to take care of the home (7 point scale from 1. Much better to 7. Much worse), and respondents’ feelings towards the #MeToo movement and feminists (0–100 with 0 being cold and 100 being warm).
I ran a set of statistical models to test that the hostile sexism score is, in fact, associated with these things (even when controlling for age, gender, race, education, income, party ID, and political ideology).
And it is in all five cases.
All 5 times, higher degrees of hostile sexism was associated with less sympathetic attitudes towards the #MeToo movement and feminists (the top 2 line charts) and with a higher likelihood of giving an answer that was less supportive of women (i.e., better for women to take care of the home, having only “a little” discrimination or “none at all”, and seeing less import in more women being elected.
The chart below shows the estimated effect of hostile sexism on the 4 theoretically related concepts. The top two are the feeling thermometers and show that increases in hostile sexism are related to cooler ratings towards feminists and the #MeToo movement. The bottom three model the probabilities of respondents providing specific responses to the questions, conditional on their hostile sexism score. Larger areas indicate that an answer is more likely and changes to the area suggest that hostile sexism influences the probability of selecting that particular answer.
As one moves up the hostile sexism scale, holding the remaining variables in the model constant at their means, respondents were more likely to give answers that reinforced traditional gender norms, downplayed the importance of electing more women as well as the discrimination women face in general, and tended to rate the #MeToo movement and feminists less favorably. No scale is perfect, but that it’s consistently correlated with measures one would expect such a scale to correlate with (albeit not perfectly; there’s a lot going on with all of these questions) shows that it’s tapping into what we want it to. Higher scores on the scale correspond to more sexist beliefs — or at least beliefs that are less supportive of women and movements seeking to advance their interests.
Estimating each state’s hostile sexism score
As I mentioned earlier, I used an MRP model to estimate each state’s level of hostile sexism. Broadly, MRPs take survey data collected at a superordinate level (like a country) to estimate the contours of attitudes and behaviors at a lower geographic level (like a state, congressional district or county). It does this by leveraging a kind of statistical model that simultaneously considers the characteristics of both the individual respondents and the geographies they come from, allowing places with fewer observations to gain information from places with more observations. The resulting model is then paired with reliable population data to extrapolate results up to every geography of interest. It sounds (and feels) a bit like magic, but the technique has been pretty rigorously studied and vetted; it’s become a fairly prominent method in fields as diverse as political science and epidemiology.
In order to estimate the MRP, I used age (18–24, 25–34, 35–44, 45–64, 65+), sex (male/female), education (less than high school, high school or equivalent, some college, bachelor’s degree or higher), and race/ethnicity (non-Hispanic White, non-Hispanic Black, Hispanic, and other). Since a great deal of importance lies in the state-level variables, I used state-level measures whose individual-level analogues have been shown to correlate with hostile sexism. These include 2016 Trump vote, the proportion of the state that is Evangelical (as of 2014), the gender pay gap between men and women’s salaries (as of 2019), the number of abortion and family planning clinics in the state (per 100,000 people–as of 2017), the proportion of the state legislature that is female, and the number of rapes in each state (per 100,000 people–as of 2020).
Finally, to complete the critical poststratification step, I gathered microdata from the 2016–2020 five year American Community Survey — which is a project of the Census Bureau — from the University of Minnesota’s IPUMS site. I conducted the analysis in R version 4.04. I used the brms package to run my MRP model in a Bayesian framework. Which is really just a stats-y way of saying that I chose a modeling technique that would allow me to clearly understand and convey the (inherent) uncertainty in the model estimates. The code and most of the data used to do this analysis is located in an R project on my GitHub.
Hostile sexism across the US
The map below shows the resulting estimated degree of hostile sexism for each state (and DC). DC is estimated to have the lowest score at 0.23 and West Virginia has the highest score at 0.41.
These values are far more compressed than the full 0–1 scale and so may not seem like that large of a difference on its face. However, that gap actually represents about a full standard deviation’s difference in the distribution of scores. If DC was a person, they’d be substantially more likely than West Virginia to say that it’s “extremely” or “very” important for more women to be elected (55% vs 42%) — in contrast, West Virginia would be more likely to say that women only face “a little” discrimination or “none” compared to DC (39% vs 30%).
So, we are talking about a fairly substantively interesting difference between our highest and lowest values. Plus, the fact that the state estimates are more compressed than individual respondents’ makes sense: states are political amalgams of large numbers of people, some with high values of the scale and others with low values. It would be weird to find states that were ideologically homogeneous enough, with residents ideologically consistent enough, that the average position is closer to either endpoint. It also makes sense that the values are all pretty low on the scale. Even if you don’t share my outlook that the vast majority of people are fundamentally decent (albeit ambivalent and malleable on many sociomoral ills — certainly more so than I’d otherwise hope), we still ought to expect that large groups of people will tend to voice the more socially acceptable position and will at least outwardly repudiate such flagrantly sexist statements as “Women seek to gain power by getting control over men”.
All that’s to say: while we see scores that are generally low in all of the states, the gaps between them are substantive.
And said gaps do appear to correlate to which states do, and do not, have trigger laws. The figure below shows the estimated scores by state again. This time, though, they show the (un)certainty in each state’s estimate via the 95% credible intervals (there’s a 95% chance that the true value for that state’s hostile sexism score is somewhere within that distribution) with each distribution colored based on whether the state has a trigger law in place. The points show the means for the distributions and correspond to the values visualized on the map. As the means get higher, and the distributions shift more to the right, the likelier it is that the state has an anti-abortion trigger law. Indeed, the average score for states with these laws is 0.32 whereas it’s 0.36 for those without it. That may not seem like much, but we can be very confident that these differences are non-zero.
So does this mean that we can say that state-level hostile sexism led to the existence of the trigger laws? Not exactly. I’ve been very careful throughout this analysis to not use words that explicitly assert a direct causal linkage between these scores and the enactment of the trigger laws. That’s intentional. It’s not that I don’t think the attitudinal aspects aren’t a contributing factor, I just don’t think it’s directly from a to b, from attitudes to policy.
There are a few reasons for this. First off, a direct link is complicated by the fact that, as Amlani shows, the majority of people in state with trigger laws still want abortion to be legal. It can’t just be that the hostile sexism translates into the laws. Hostile sexism may be associated with less permissive attitudes towards abortion but that doesn’t mean that it guarantees adherents are maximally restrictive towards abortion. To put it another way, just because a state has higher hostile sexism, it doesn’t mean that everyone, or even most people, want abortion banned.
A direct attitudinal explanation also neglects the role of structural factors. For starters, the connection between prevailing attitudes and policy has always been imperfect — but it’s especially exacerbated by state-level gerrymanders that translate small popular vote wins (or even popular vote losses) into outsized, relatively stable GOP majorities. And as Grumbach points out in a Politico article on the subject:
Republican state legislative majorities in states like Wisconsin, where partisan gerrymandering empowers conservative rural voters over more liberal urban voters, will be electorally insulated from a backlash to an abortion ban…Partisan lawmakers occupy highly secure seats, rather than having to forge compromise positions that appeal to a majority of state residents…As state governments start to play an increasingly influential role in the lives of Americans, this imbalance will become only more important, not just on abortion but on issues like taxes and state services, access to guns or organizing labor unions.
Again, it’s not that hostile sexism doesn’t matter. You may have heard about a pregnant 10 year old (a description that fills me with fury and revulsion just typing it) who would’ve been forced to bring the child to term if she did not travel out of her home state to one that allowed abortions. That’s in Ohio, which ranks in the top 15. I’d hazard a guess that it definitely plays a part there.
But the part I think it plays in enacting these laws is more complicated. Basically, I’d venture that hostile sexism is part of a constellation of factors that contribute to states electing more socially conservative state legislatures on average. In effect, if a state has a higher level of hostile sexism, that may translate to them voting for state legislative candidates who are more socially conservative (and, thus, more likely to enact anti-abortion trigger laws even if it does not fall fully in line with the broader preferences of the state).
Rigorously testing that hypothesis would require a whole other analysis — but one way to see if the idea at least passes the sniff test would be to use legislature ideology data from the American Legislatures Project. The project uses roll-call votes to estimate the average ideological leaning of state legislatures from 1993–2018. Higher values indicate more conservative legislatures and lower values reflect more liberal legislatures. Thanks to the way that the scores are calculated, we’re able to compare legislatures across states as well as across time. Looking at scores from 2016–2018, a state’s estimated hostile sexism score is strongly associated with its legislatures’ ideological leanings. This relationship persists even when controlling for Trump’s 2016 vote margin in that state. This suggests that hostile sexism may play a role in how conservative a state legislature is apart from its support for the then-and-current leader of the nation’s conservative party. Which, in turn, may have led to a higher likelihood of enacting an anti-abortion trigger law once Roe was repealed.
Final Thoughts
There are a few things I’d be remiss in not noting and/or clarifying.
First, I don’t want readers coming to the conclusion that people supporting abortion restrictions are all raving sexists. As noted earlier, even most “pro-life” people support limited access to abortion; many of these laws are disjunctures from their preferences too. Indeed, this analysis isn’t really about the association between hostile sexism and abortion attitudes at the individual level anyways: it’s about the tendencies of the average prevailing level of hostile sexism in the states and how that translates to these trigger laws. And what we see is that states that have passed these laws have stronger average tendencies towards hostile sexism.
On that note, though, I don’t want to convey that the connection between a state’s level of hostile sexism and whether or not it passed (or started enforcing) an anti-abortion law is direct. While I’d personally bet that hostile sexism plays a role, these data aren’t structured to make an iron-clad causal argument. And when thinking about the evidence available (such as the prior MRPs and the analysis leveraging the American Legislatures Project data above), the most likely case is that state-level hostile sexism is but part of the story, playing an indirect role, with other factors mediating its influence on policy.
Finally, I don’t want this to somehow inspire the idea that the issue is settled. Dobbs didn’t settle abortion in the US. It blew it up — and cast the radioactive debris all across the country. Rather than being decided at the federal level, it’s now going to be fiercely litigated and protested at the level of every individual state. It wouldn’t surprise me if hostile sexism continues to play a role in the development of this new patchwork status quo — such as with the laws some states are contemplating that would criminalize women freely traveling between states to get abortions.
There is a pretty famous (though apocryphal) curse often attributed to the Chinese: “May you live in interesting times.” Things are doubtlessly going to get much more “interesting” from here out.
Peter Licari is a social data scientist specializing in American political behavior. He received his PhD in American Politics and Political Methodology from the University of Florida and is currently a Director of Commercial Data Science for Morning Consult. The views expressed here are his own. He can also be found on YouTube and on Twitter(@PRLPoliSci). What little spare time remains is dedicated to long-distance running, binging games and media with his wonderful wife, Stephanie, playing with his daughter, Rosalina, walking his dog, Dude, and holding oddly productive one-sided conversations with his cat, Asia.
This is version 1.1.0 of this analysis. This post’s changelog is maintained on Github.
I believe all work benefits from readers’ constructive feedback as well as the writer’s own revisiting and reflection. But not all work fits the mold of academic publication. To that end, for the sake of transparency, I have decided to visibly index what point my non-academic projects are at once they go live in a format where others will (hopefully) come into contact with it. Minor revisions (such as grammar or minor image formatting issues) lead to an increase in the third digit. Major revisions inspired by my own revisiting and minor revisions inspired by the suggestion of readers leads to an increase in the second digit. Major revisions driven by the suggestion of readers or by future reflection and revisiting the project leads to the an increase in the first digit. After 6 months of no updates, a version should be considered “final.” The original version was published 7/8/2022. The current version was published 7/9/2022.
