Welcome to regression to the meat!
Hi! I’m Seth Green, I’m a research scientist at the Humane and Sustainable Food Lab at Stanford. I did PhD coursework in political science and ultimately specialized in meta-analysis, e.g. here, here and here.
My personal website is setharielgreen.com.
About the lab I work at
The Humane and Sustainable Food Lab (HSFL) researches ways to accelerate society's transition away from factory farming. It is directed by Maya Mathur. For the most part, we do behavioral science, meaning we try to alter how people eat, think, and feel about meat and animal products (MAP). We’re definitely interested in clean/cultivated/lab-grown meat, but we’re not material scientists or chemists.
The health and environmental reasons for cutting back on MAP are well-known. However, we do not want this to come at the expense of animal welfare. This means, for instance, that we wouldn’t encourage people to switch from beef to chicken or fish because that means more animals suffering and dying. We would encourage people to adopt plant-based diets instead.
I wrote more about my circuitous journey to joining the lab here.
What is this blog about?
I read a lot of stuff about reducing MAP consumption. Sometimes I want to publicize especially good papers, and sometimes I have observations/questions about research that don’t merit a standalone paper. This blog is going to be on those topics, e.g.
What’s a reasonable impact to expect for a nudge-type intervention?
Most Americans eat about 3X as much MAP as dietary guidelines suggest. What does the population distribution look like, e.g. how many people do eat the recommended amount or less? Is the recommendation too low to be useful?
What are some neglected animal welfare policy interventions that we might push for, e.g. windows in long barns for broiler chickens or requiring anesthesia for pig castration? (this is probably as painful for a pig as it would be for you.)
What can we learn from other social science literatures, like political persuasion, edutainment, and norm perception, as we design interventions?
If that sounds interesting to you, I’d be glad to have your readership. (I plan to keep this blog free/open.)
Also, I plan to keep it positive. The literature definitely has issues, but it has unusual strengths too,1 and I will generally avoid singling out particular papers for criticism. I think it’s a lot more helpful to observe a tendency in the literature and be concrete about how it can be improved, and to highlight how strong papers have dealt with the same underlying challenges.2
Finally, I’m assuming that my readers are interested in research but not professional researchers. I’ll try to explain technical concepts via intuitive primitives as much as possible.
Do I know what I’m talking about?
Yes, but only because I spent a lot of 2024 leading a meta-analysis of interventions intended to reduce MAP consumption. Unlike the ~160 previous reviews I surveyed, our paper set reasonably strict criteria around design (randomized controlled trials), statistical power (≥ 25 subjects in both treatment and control, or ≥ 10 clusters for cluster-assigned trials), and measurement strategy (had to measure MAP consumption directly ≥ 1 day after treatment began). I read (or skimmed) about 1000 papers to find studies meeting our criteria, and though I only know about 25-50 papers down pat (the ones I coded for inclusion ), I am probably about as current on the literature as anyone.3
What’s in a name?
Regression to the mean is a statistical term for the idea that if you take a random sample of something and get an extreme result, your next sample is more likely to be closer to the true average. In behavioral science land, this means, first, that if you take a random draw and get result that match your pet theory, you should probably redraw to validate; and second, that systems tend to return to baseline after perturbations, including after deliberate interventions.
Regression to the meat, in my opinion, is the key unanswered question of the MAP reduction literature. If you intervene to make meat less salient in a dining hall at lunch, what happens at dinner? If you show people a video of factory farming conditions, how long before most of them return to their normal diets? For the most part, we don’t know. But I think we can find out.
Happy reading!
For instance, a lot of advocacy groups in the space are really good about publishing null results where they find them.
Tyler Cowen writes: “However pressing a social or economic issue may be, there is almost always a positive and constructive way to reframe your potential contribution. This also will force you to keep on thinking harder, because it is easier to take apparently justified negative slaps at the wrongdoers.”
By the way, setting some inclusion criteria that really shouldn’t be a very high bar but in reality filter out like 95% of the literature is fine approach to writing meta-analyses, I think. Meta-analyzing thousands of papers gives you a nice overview of what the entire body of research says, but if you average a strong design and a weak one together, with vastly different measurement strategies and interventions to boot, it’s not clear the resulting average means anything. Also, a small dataset is a lot easier to reason about, add new rows to, etc. Finally, I’m pretty darn sure there are approximately zero errors in my dataset. There’s no way I could be sure of that with hundreds of studies.