Sigh. This appears to be one of those cases of media exaggeration for eyeballs. The study says: > the effect on violence was only consistent visible for the measure of Airbnb penetration–or the extent to which buildings in the neighborhood have one or more listings (and for the measure of density, or the listings per household in the two-year lags). It was never present for overall usage, or the estimated quantity of…
>They even acknowledge that the correlation could be due to a factor that isn't being accounted for.
Related to this, they're performing a regression between Airbnb prevalence and crime, but they only control for a single variable: income[0]. They look at others in a robustness check, but a single control variable practically screams p-hacking.
They also don't address the fact that both Airbnb prevalence and crime are nonstationary[1]. Regressing two nonstationary time series results in a nonsense coefficient[2]. Two totally unrelated time series will have a high coefficient if both exhibit consistent trends.
[0] "We report the results based on using income as the main tract-level control variable"
[1] They are consistent trends over time, see https://www.investopedia.com/articles/trading/07/stationary....
[2] https://stats.stackexchange.com/questions/94723/using-non-st...