Many of you have read Slime Mold Time Mold’s series on the hypothesis that environmental contaminants are driving weight gain. I haven’t done a deep dive on their work, but their lit review is certainly suggestive. 

I think this is as good a place as any to point out that the SMTM authors have been repeatedly misleading about their evidence and unwilling to correct their mistakes, both on A Chemical Hunger and elsewhere. Here are a few examples that come to my mind at the moment: 

1. They claim that Texas “tends to be more obese along its border with Lousiana [sic], which is also where the highest levels of lithium were reported,” but their own source says that lower levels of lithium, not higher, are found along Texas's border with Louisiana. A commenter on their post has pointed out that error, as have I on a Twitter thread, but the authors have not edited their post or addressed this in any other way. (Incidentally, the correlation between drinking water lithium levels and obesity rates across Texas counties is negative).
2. They claimed on Twitter that geospatial associations between drinking water contaminants and obesity rates in the US are probably not confounded by SES, because “SES isn't really associated with obesity rates.” However, the correlation between obesity and ln(income) across n = 3110 U.S. counties was −0.486 in 2013,  and my own analysis of 2019 data suggests the correlation was -0.65 in that year (using median household income data). ^{[1] [2] [3]} (Their response to the 2013 data ^[4] is pretty much just “this correlation didn’t exist 30 years ago,” but I don’t see how that supports their statement that “SES isn't really associated with obesity rates,” since that's a statement in the present tense rather than the past tense.)
3. They claim that hypoxia probably cannot explain the effect of altitude on obesity, saying that “exercise in a low-oxygen environment does seem to reduce weight more than exercise in normal atmospheric conditions, but not by much.” However, when you read the abstract they linked to, you see that what they are calling “not by much” is a 60% increase in weight loss. 
4. In several posts, they claim that wild animals have been getting more obese, citing Klimentidis et al. (2010). However, that paper does not make that claim; it doesn’t even examine body weight data from wild animals at all. When confronted about this on Twitter, they provided evidence that some white-tailed deer populations under increased predation from humans have been getting heavier over the past several decades, but there’s archeological evidence that they are simply returning to their normal historic body size after being smaller than normal for a while due to a temporary decrease in predation by humans (which increased their population density and thus competition for food). For sources and more details, see this Twitter thread. 
5. (Unrelated to obesity) there's a post in which they claim that “Sicilian lemons really ARE more like polar bear meat than they are like West Indian limes, at least for the purposes of treating scurvy” (implying that Sicilian lemons have lower vitamin C content than West Indian limes and polar bear meat). I investigated this and found that West Indian limes have ~60% of the vitamin C concentration of lemons, and that polar bear meat has much less vitamin C than either (but that all three of those can still prevent scurvy if eaten regularly at not-extremely-large portions, and lemons and limes both have enough to treat it).^[5] They have been contacted about this, and their response was that we don't know whether historical Sicilian limes had enough vitamin C to treat scurvy or not. Clearly, that is different from asserting (as they do in the post) that we know they don't have enough vitamin C. But they have not edited their post. 

ETA: What I initially said on point 5 was wrong (specifically, I embarrassingly confused lemons with limes at some point), and I have now fixed it. 

1. ^{^}

   Using the dataset found here for median household income and the one found here for obesity rates, this is the association between the two that I found (each datapoint is a US county or county equivalent in the 50 states plus DC):

   

    

2. ^{^}

   Individual-level data yield a much weaker correlation (in my own analysis of NHANES 2017-2020 data, the correlation is -0.14 for white women in their 30s and 40s, and -0.05 for white men of the same age). But NHANES only records income levels in multiples of the poverty line up to 5, and individual-level data is known to be noisier than county-level data, so that probably explains the discrepancy. Moreover, in that specific context (figuring out whether geospatial associations between drinking water contaminants and obesity rates are confounded by SES or not) county-level data are more relevant than the individual-level data.

3. ^{^}

   For comparison, my own analysis suggests that the correlation between altitude and obesity rates across US counties (which the SMTM authors think is a big deal) is -0.35. The altitude value I used for each county in my analysis was the average altitude of the centroids of its census tracts, which gives you the closest thing to a population-weighted average altitude by county that you can get with cheap and fast computation. I haven't published the details of this analysis yet, but you can ask me for the Google Colab notebook and I'll share it with you.

4. ^{^}

   They address the 2013 data in the paragraph starting with “The studies that do find a relationship between income and obesity tend to qualify it pretty heavily.”

5. ^{^}

   Livers tend to be more vitamin C-rich than other tissues in the animal body, so I looked for data for them too, and found that, for several animals, their vitamin C content ranged from lower than that of West Indian limes to higher than that of lemons. So West Indian limes did not stand out in my data as being unusually lacking in vitamin C.

Since it looks like you're using Pandas, I'd recommend adding Seaborn for simple statistical plots, to avoid the saturation effect from having so many points on a scatterplot. It's a lovely toolkit for producing specific kinds of useful plots quickly and easily, with minimal customisation.

(specifically for these plots, I'd reach for a joint distribution plot with kind="hex" or kde or reg)

None of those metrics would catch if a contaminant makes some people very fat while making others thin ( SMTM thinks paradoxical effects are a big deal, so this is a major gap for testing their model).

FWIW, it does not seem to be the case that, at a population level, very low BMIs are more common now than they were before 1980. The opposite is true: when you compare data from the first NHANES to the last one, you see that the BMI distribution is entirely shifted to the right, with the thinnest people in NHANES nowadays being substantially heavier than the thinnest people back then. 

So, contra what the SMTM authors argue, whatever is causing the obesity epidemic does not seem to be spreading out the BMI distribution so much that being extremely thin is more common now.

Your obesity source is using BRFSS. You should go directly to BRFSS, where you can get mean BMI, rather than %obesity. By county.

But BRFSS does not really have county data. It samples 400k people per year. If spread uniformly by county, that would be pretty good coverage of the 3k counties, but it weights by population and thus only lightly samples rural counties. But people want maps, which are dominated by rural counties, so it smooths out its patchy data. This produces the maps at SMTM and SSC with their sharp lines at state borders.