[agallant]

Again, not vouching for the study as a whole - and agreed that scientists can get a bit "creative" when trying to actually describe and motivate causal mechanisms (in their defense, a very hard problem).

But I'm just talking about the statistics here, and specifically that saying "correlation is not causation" is a bit overused. Researchers know about it too, those four words don't magically dismiss all statistical studies. Most modern statistical approaches are explicitly built to try and help address these sorts of concerns.

There could well be other flaws with their statistics, and even if there is causation they could be failing at theoretically motivating or connecting it to their overall narrative. But it takes more than four words to make that case.

EDIT - just acknowledging that you've since edited your comment to add concerns about p-hacking and reproducibility. And that may be the case - but it wasn't what I was responding to in my initial comment.

[disabled]

> But I'm just talking about the statistics here, and specifically that saying "correlation is not causation" is a bit overused.

I think that term has erupted into popularity with the widespread adoption of AI, which is intellectually bankrupt. With AI you can find correlation between things, and draw a very basic rudimentary conclusion, but never actually know why this happens (the causation), in this day and age.

For example, let's apply an unethical use of AI. Let's say an individual goes to the grocery store weekly and buys a dozen eggs and 1 container of dry shampoo (for washing your hair without water), every single week for the past 2 months. With AI and the hoarding of data, it can be found that this individual is going to die in the next 6 months to a 95% confidence interval.

The individual gets harassing ads during this, even though they are not going to die. The ads, of course, in this day and age, play into everyone’s hopes and fears anyways, which is abusive.

[in_cahoots]

It’s an overused phrase because it’s so often true. The analysis acknowledges the possibility of confounding variables but only makes a weak attempt to address it, using demographic info, income, and homeownership rates. This is the definition of a correlational approach. And to make matters worse they throw in some hypothesizing about Airbnb eroding the ‘local social dynamic’.

‘Causal linkage’ is great in theory. In reality it often shows directionality but not causality. This is a prime example.

[agallant]

So, you're saying because past research has been correlated with causality fallacies, we should just assume it's the case when we see claims of this sort? ;)

More seriously - I'm well aware of the difficulties of causality, and also use causal direction as a great illustration of them. As I've said in pretty much every comment here - I'm not championing the study, I simply haven't done a deep enough pass to have a strong opinion, and it may have any number of subtle flaws (off the cuff my biggest concern is that they're focused on one city, and I'd like to see similar results elsewhere, preferably in different geographic areas and cultures).

In other words, yes - more controls like you said. But I am responding to the overuse of a simple statistical argument in the face of studies that, whatever flaws they have, are not cases of "the researcher forgot the controls." Demographics, income, and homeownership are actually not bad features to have I'd say, and again it seems like most of the large claims bothering people are from the coverage and not the research. To refute research you generally need to dig into the details of the research itself.