In sum, homeowners & renters are “the policy adjacent”—indirect winners & losers of affordable housing. We identify similar groups for other policies (eg, SNAP) & set an agenda for studying the spillover effects of policy on long-term support.
Replication: dataverse.harvard.edu/dataset.xhtm... 9/9
Our findings present a trade-off. Renters want local priority, but the FHA limits such prioritization due to entrenching segregation. NYC was recently forced to lower local prioritization via court order, but Jersey City is pursuing 100% local priority by using only local funds. 8/9
But the ability of affordable housing to ⬆️ home values may threaten renters, who are unlikely to receive a unit in the project due to long wait lists. The renter backlash to new affordable housing is driven by gentrifying neighborhoods, where housing instability is greatest. 7/9
Why do homeowners & renters respond differently? New affordable housing often replaces blight, increasing nearby home values. Homeowners like this. To quote a homeowner who had written to his mayor to try to block a LIHTC development: "If they wanted to build another, more power to them." 6/9
We also find null results using a placebo of the “near-far” design. Results consistently polarize by homeownership status, not other block-level traits (e.g., race). We also conduct a bounding exercise to estimate how much of our effect may be driven by residential sorting rather than persuasion 5/9
But affordable housing may be targeted at the micro-level. So, we compare blocks near housing built from '03-'06 to blocks near housing built from '07-'10. These blocks are equally targetable for housing, but only some are treated prior to the '06 bond. Results are comparable. 4/9
Comparing blocks near new affordable housing to blocks slightly farther away, majority homeowner blocks ⬆️ support for the '06 bond by 2 to 3 p.p. In contrast, “renter blocks” ⬇️ support by 1 to 2 p.p. Results are the same if treatment is defined by proximity or # of units. 3/9
In '02 & '06, Californians voted on nearly identical $2+ billion housing bonds. We measure how affordable housing built from '03 through '06 affected support for these bonds at the block-level. Thus, we have pre- & post-treatment behavioral data on how policy implementation ➡️ policy support 2/9
Building affordable housing ➡️ support for funding housing.
Nearby homeowners ⬆️ support; renters ⬇️ support. Both are “the policy adjacent”—secondhand recipients & drivers of policy feedback.
Forthcoming @ajpseditor.bsky.social ( doi.org/10.1111/ajps...) w/ A. Magazinnik & @msands.bsky.social 1/9
📢 Call for Hidden Papers & Data on Ingroup Favoritism in Dictator Games
We look for unpublished and hard-to-access experimental studies for our meta-analysis!
Inclusion criteria: no deception, adults, manipulation of group membership of dictator game recipients.
All tips and data very welcome!
Early career researchers (especially in Europe / UK, but all are welcome):
Apply to present your work on the urban/rural divide (very broadly defined!) @ LSE
@lseinequalities.bsky.social @lsegovernment.bsky.social
We've also created an interactive dashboard where you can explore our data in detail.
You can look at specific street view images, and see the distributions of LMM and human ratings across all four measures, split by gender if desired.
Of course, these models may (potentially) improve. But while the specific relationships may change, the broad patterns reflect hard-to-resolve biases.
To that end, the entire workflow is public and reproducible, so that future models can be tested with the same framework and data.
We then study these relationships by gender. LMMs easily perform worst in recovering the views of Detroit men.
Why? This difficult-to-reach population is likely poorly represented in training data/reinforcement feedback. It may be that this problem is hard to resolve even with better models.
We then asked the same questions of two human samples: a U.S. nationally representative sample and Detroit-representative sample (through DMACS at @umfordschool.bsky.social).
We then compare these responses, two ways. In general, LMMs better recover national evaluations than local ones.
We pulled tens of thousands of open-source street view images from Detroit. We processed the images, and then sampled 85 for closer analysis.
For each, we queried five LMMs (propietary and open-source) 30 times for ratings on wealth, safety (day and night), and disorder.
Can Large Multimodal Models (LMMs) extract features of urban neighborhoods from street-view images?
New with Paige Bollen (OSU) and @joehigton.bsky.social (NYU): Sometimes, but the models better recover national assessments that local ones, even w/additional prompting (which can make things worse!)
🚨Job News: Postdoc positions in LSE Government Department 🚨
Apply to work with the brilliant and wonderful Carl Muller-Crepon @carlmc.bsky.social on his BORDERS ERC project at LSE @lsegovernment.bsky.social
jobs.lse.ac.uk/Vacancies/W/...
I’m with you. I don’t even own a blow dryer 🫣.
📢 Call for Submissions
UK-based early career researchers in political science, political economy & public policy: apply to our January Early Career Workshop at LSE.
🗓️ Deadline: 28 Dec 2025
🔗 forms.gle/HXU4DnBq9HmX...
the case for slower, better/more careful research papers in poli sci gets another data point in its favor
2. Popper tells us that we should be our own best skeptics. Unfortunately our professional incentives often make us the opposite.
As we conduct research, we should all be self-skeptical and self-questioning and self-doubting. We should expose our own thinking and decisions to close examination.
As I reflect on this, three thoughts:
1. This case is only symptomatic of a bigger problem; increasingly broken incentive structures in our disciplines and industry.
Our job is *not* to publish for the sake of publishing. It is to produce and advance *knowledge*. This process is hard and slow.
Certainly comports w/ my personal experience
🙌🏼 happy to chip in to buy a billboard
Biased as I am, this is a great paper on the limits of current generative AI models for learning about fundamentally subjective assessments (in this case, evaluating features of neighborhoods from photographs).
Logo of Political Analysis on a yellow and red background with the hashtag OpenAccess at the bottom.
#OpenAccess from @polanalysis.bsky.social -
Nationally Representative, Locally Misaligned: The Biases of Generative Artificial Intelligence in Neighborhood Perception - https://cup.org/3Kw6VkD
- Paige Bollen, @joehigton.bsky.social & @msands.bsky.social
#FirstView
