Thanks for posting your paper which I will print out and read properly!
Section 3.2 and Table 1: does the 'continuous urban' category cover e.g. a fairly large park in the middle of a city? I can't see a category for 'parkland'.
As a concrete example have a look at Cannon Hill Park in Birmingham [1]. It struck me that it is possible to be surrounded by green semi-natural parkland within the inner ring of a city of 1 million.
The dataset used is the UK Land Cover Map 2000 [1]. It's at 25m resolution, and urban parkland should show up as the appropriate land cover types: grassland, deciduous forest, etc.
> If you want to know how people feel, I’m afraid self-reports are the gold standard
"How people feel" can't directly be measured scientifically, but it is a function of many things, some of which can be measured reliably (weight, blood pressure, money, etc). I'd rather have a reliable measurement of quantity that I believe contributes to well-being than a direct measurement of well-being that I think is unreliable.
This is brought into sharp relief when you consider the applications of this research. Suppose intervention A is know to promote subjective well being, whereas intervention B is known to promote healthy blood pressure levels. What would you rather spend money on?
> Some other studies have large sample sizes
This is a fair point. The 2019 Nature study also has large-n. I also like that the authors of that one write "Future studies would ideally collate as much data via non self-report measures as possible."
But as far as I can tell, all the large-n studies are observational. As other commenters have pointed out, when the treatment is so general it's really hard to control for confounding factors. The study you've linked (and the 2019 Nature study) both implemented controls, all of which look reasonable. But when we find an effect, we are still left the question: "What if we just missed a major confounder? What if the cumulative impact of small confounders for which we can't control for is responsible for the signal?"
> I'd rather have a reliable measurement of quantity that I believe contributes to well-being than a direct measurement of well-being that I think is unreliable.
Sometimes there's a difficult trade-off here, especially since (un)reliability isn't binary.
But a heavy prioritisation of reliability over direct measurement (which seems to be your position) is probably why governments have spent decades obsessing over GDP, which is a very poor wellbeing measure for all sorts of reasons.
It's not obvious to me how blood pressure affects happiness, or how you would even prove a link here without asking people how they feel. I also feel like it is important to point out that asking people how they feel is a scientific measurement, you just have to be careful on how to use that data.
(1) If you want to know how people feel, I’m afraid self-reports are the gold standard
(2) Some other studies have large sample sizes (e.g. this one, which is based on my PhD: https://eprints.lse.ac.uk/49376/1/Mourato_Happiness_greater_...)