My book, Build or Die: How America Suffocates Its Cities and What to Do About It, will be published by Princeton University Press on December 8, 2026. Preorder the book now from Bookshop.org.
Way back in March 2023, the Urban Institute released a paper that broke new ground in the study of how pro-housing land use reforms affect housing supply. For the first time ever, the paper’s authors wrote, they would “use a machine-learning approach to identify a diversity of reforms, and then examine their effects in multiple cities simultaneously.” Finally, we would get something like a generally applicable formula describing how zoning changes relieve—or fail to relieve—housing supply shortages.
The results were disappointing. Based on their analysis of 180 different pro-housing reforms, they found that upzoning was associated with a mere 0.8 percent increase in housing supply within three to nine years. Worse, “this increase occurs predominantly for units at the higher end of the rent price distribution; we find no statistically significant evidence that additional lower-cost units became available or became less expensive in the years following reforms.” In other words, YIMBY land use reforms don’t benefit low-income households.
This finding was, of course, catnip to left-NIMBYs. In the three years since the study’s publication, numerous supply skeptic pundits and journalists have held it up as Exhibit A demonstrating that YIMBY reforms are, at best, futile. Supply skeptics Brian Callaci and Sandeep Vaheesan (who I later debated in Dissent) cited it in their anti-YIMBY Harvard Business Review essay. Michael Friedrich relied on the study in his New Republic essay, “The Case Against YIMBYism.” And it was featured in not one but two pieces from the Washington Monthly: one on “the meager agenda of abundance liberals” and another on why policymakers should “ignore the YIMBY playbook” (which, strangely, endorsed strategies that are very much part of the YIMBY playbook).
I always found the paper’s methodology to be suspect at best. Here’s what I wrote about it in my response to Friedrich, “The Case Against the Case Against YIMBYism”:
The challenges start with the paper’s scope of observation: the authors aim to observe the effect of land use reforms over no fewer than eight metropolitan areas, including the regions around Los Angeles, Chicago, Dallas, Philadelphia, and Miami. These are all very different regions, with wildly disparate histories, demographics, housing markets, regulatory regimes, geographies, and built environments.
Land use is simply going to mean very different things in different contexts: a city that already has very permissive zoning might not see much change in housing supply if you upzone further, whereas a city with very strict zoning might see an immediate supply response following even modest tweaks to the zoning code. Similarly, developers might not build much new housing in an area with a collapsing population even if you make it easier for them to do so. Trying to average out the effect of land use reforms over these jurisdictions can only tell us so much.
Then there’s the paper’s variable of interest: land use reform, broadly defined. In order to get a statistically significant result, the authors lumped together a number of different reform types, ranging from raising height limits to encouraging accessory dwelling unit development. As they themselves write: “We also do not differentiate between relative impacts of different changes, and we do not have the power to assess the varying impacts of reform types, like ADU or height-limit policy.”
So a change in the height limit of, say, ten feet gets treated the same as a twenty foot increase. Similarly, a bill legalizing mixed-use development in certain areas gets treated the same whether or not it gets loaded down with poison pills that render the policy unworkable. At such a high level of abstraction, it’s simply impossible to say much about the impact of land use reform, except the following: there is evidence that land use reform, broadly speaking, is associated with more homebuilding, broadly speaking.
As it turns out, I was being too kind. The Urban Institute study doesn’t just fail to capture the nuances of land use reform; it’s also rife with serious factual errors.
Recently, researchers at the center-right American Enterprise Institute went back and checked the Urban Institute’s work by examining each data source and zoning reform in the study. The results of AEI’s own research are damning; so damning, in fact, that I think it’s fair to say I’ve never seen anything like it in the annals of academic debates over housing policy.
Here’s what the AEI researchers found after looking at all 180 of the Urban Institute’s data points [emphasis mine]:
Our independent review reveals that 60 of these articles should be disqualified outright due to duplication, incorrect geographic attribution, or policies that only affected commercial or industrial areas or properties. Among the remaining 120 entries, 118 are either incorrectly classified in direction (more vs. less restrictive), are either not major municipality-wide reforms, or have insufficient information to accurately determine policy direction. After a thorough and painstaking review of all 180 cases, we found only two that plausibly qualify as “major” reforms—and even these warrant caution given uncertainty about whether their supply impact was measurable using Stacy et al.’s methodology.
No less than a fifth of the 180 land use changes were “attributed to the wrong city.” For example, the Urban Institute researches said that a height limit increase in Aventura, Florida had actually taken effect in Miami. Why? Because the article about this increase appeared in the Miami Herald. (The article’s lead sentence is: “Developers who had plans to build in Aventura on hold can now proceed, but face stricter zoning laws.”)
As mentioned, I’ve never seen anything quite like this. There’s a whole cottage industry for anti-YIMBY academic papers with shoddy methodology, but this specific combination of poor methodology and pervasive data entry error is a novelty. As best as I can tell, it’s because the Urban Institute researchers were so confident in their machine learning tool that they didn’t bother to check the computer’s work. There’s a lesson here for anyone using large language models or machine learning for social science research: always check the computer’s work.
There’s another lesson, too. The academic consensus is that eliminating restrictive land use rules will generally lead to more homebuilding in high-demand areas, which will consequently drive down housing costs. There’s nothing wrong with publishing research that challenges an academic consensus; but in the case of this particular consensus, very few, if any, of the countervailing studies hold up to serious scrutiny. Because, as it so happens, YIMBYism does actually work.