I think that's a good idea -- or just post as yourself (?)

(Ofc I think I and others understand that things are in flux and this is all NBD)

👋 Looks interesting! What do you think about having the title reflect its origins, e.g. "linkpost: Climate Change Is Worse Than Factory Farming", or "suggested reading: [X]" or something like that?

At a glance right now, the UX here looks like the EA Forum Team is itself endorsing this pretty radical position. (FWIW I appreciate the drive to cross-post interesting material/the broader drive to improve the forum experience, I have been thinking about your other post a bit lately and hope to respond soon)

Hi Ruben, I am not expert on that strand of research, but here a few papers that may be of interest (lead author/year/title):

My implicit  knowledge on the topic of knowledge production (rather than of veganuary) is that rosy results like the one you are citing often do not stand up to scrutiny. Maya raised one very salient objection to a gap between the headline interpretation and the data of a past iteration of this work here.

I believe that if I dig into it, I’ll find other, similar issues. 

Sorry for such a meta answer…

No meaningful relationship! (see code below.) However, big caveat here that we had to guess on some of the samples because many studies do not report how many subjects or meals were treated (e.g. they report how many restaurants or days were assigned to treatment and control but didn't count how many people participated)

> summary(lm(d ~ total_sample, data = dat))

Call:
lm(formula = d ~ total_sample, data = dat)

Residuals:
     Min       1Q   Median       3Q      Max 
-0.59897 -0.13702 -0.01868  0.12322  0.75767 

Coefficients:
                Estimate  Std. Error t value Pr(>|t|)  
(Intercept)   0.06330835  0.02664964   2.376   0.0193 *
total_sample -0.00002876  0.00004690  -0.613   0.5410  
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 0.2474 on 110 degrees of freedom
Multiple R-squared:  0.003407,	Adjusted R-squared:  -0.005653 
F-statistic: 0.376 on 1 and 110 DF,  p-value: 0.541

Delay indicates the number of days that have elapsed between the beginning of treatment and the final outcome measure. How outcomes are measured varies from study to study, so in some cases it's a 24 hour food recall X number of days after treatment is administered (the last part of it), in others it's a continuous outcome measurement in a cafeteria (the entire period of delay). 

I don't know this, sorry, and not every study reports enough location data to begin to estimate this (e.g. studies that recruit an online sample from multiple countries)

This I can say more about!

The median delay, in days, is 14, and the mean is 52 (we have a few studies with long delays, the longest is 3 years (Jalil et al. 2023). 

So I'd say, think "about 2 weeks on average with some lengthy outliers". Also there's basically no relationship between delay and effect size.

to replicate in R (from the root directory of our project):

> source('./scripts/libraries.R')
> source('./scripts/load-data.R')
> summary(dat$delay)
   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
   4.00   11.50   14.00   52.05   60.00 1095.00 
   
> source('./functions/sum-lm.R') # this is a little function we wrote that puts summary(lm()) into a dplyr-friendly pipe
> dat |> sum_lm(y = d, x = delay)
            Estimate Std. Error t value Pr(>|t|)
(Intercept)  0.05312    0.02552 2.08181  0.03968
delay        0.00005    0.00019 0.23166  0.81723

Hi Vasco, I'm afraid not, sorry. The diversity of outcome measures makes this all but impossible, e..g one study measures "servings of meat per week", others it by the gram, others count how many meals are served in a given time period, etc. 

Thank you David! We will post any updates to https://doi.org/10.31219/osf.io/q6xyr

The paper is currently under submission at a journal and we likely won’t modify it until we get some feedback.