7 ms·

Digital Discrimination: The Case of Airbnb.com [pdf]

11 points by 001sky 13y ago

kazagistar 13y ago
This is somewhat off topic, but I found the typography of this paper to be really really horrible. Maybe I am just used to the nice latex templates that journals use, but something about this (maybe the extra spacing between lines?) makes it really hard to read.
ryan_j_naughton 13y ago
I'm glad that someone is doing this analysis. That being said -- their methods are not quite sound. 1. Their sample has 3752 observations, of which black hosts account for 7% (~260 obs). Given that the group of interest is a relatively rare event or outlier, it would be better to not compare them to the entire population using standard OLS regression models. Instead, a matching solution (e.g. propensity score matching) would be a better solution so a more representative group would be constructed that matches the black hosts. Despite the regression attempting to control for this to some degree by including the other observable variables, the regression model is attempting to fit to all the observed prices. Given that the majority of the listings aren't black hosts not likely to be suitable matches to the black hosts, the fit isn't most representative of that relationship. 2. They don't include the adjusted R-squared value for their regressions; thus, we don't know how much of the price variance is really being accounted for. 3. They conclude that by including other observables available to customers on Airbnb (like the location rating), that they have removed omitted variable bias. Let's take the 'location rating' as an example of how this is problematic. When I book in Airbnb, I may initially sort by Airbnb's metrics; however, once I find a short list of matching options, I drill down into them to try and determine how well they fit my needs. This additional consumer analysis isn't accounted for by the high level Airbnb metrics, yet I bet it is a primary driver in users booking with one listing over another. Despite this, I still think it is a good paper to start this analysis. They are likely correct that there is discrimination in Airbnb, and their logic underpinning this is sound.

11 points by 001sky 13y ago

This is somewhat off topic, but I found the typography of this paper to be really really horrible. Maybe I am just used to the nice latex templates that journals use, but something about this (maybe the extra spacing between lines?) makes it really hard to read.

ryan_j_naughton 13y ago

I'm glad that someone is doing this analysis. That being said -- their methods are not quite sound. 1. Their sample has 3752 observations, of which black hosts account for 7% (~260 obs). Given that the group of interest is a relatively rare event or outlier, it would be better to not compare them to the entire population using standard OLS regression models. Instead, a matching solution (e.g. propensity score matching) would be a better solution so a more representative group would be constructed that matches the black hosts. Despite the regression attempting to control for this to some degree by including the other observable variables, the regression model is attempting to fit to all the observed prices. Given that the majority of the listings aren't black hosts not likely to be suitable matches to the black hosts, the fit isn't most representative of that relationship. 2. They don't include the adjusted R-squared value for their regressions; thus, we don't know how much of the price variance is really being accounted for. 3. They conclude that by including other observables available to customers on Airbnb (like the location rating), that they have removed omitted variable bias. Let's take the 'location rating' as an example of how this is problematic. When I book in Airbnb, I may initially sort by Airbnb's metrics; however, once I find a short list of matching options, I drill down into them to try and determine how well they fit my needs. This additional consumer analysis isn't accounted for by the high level Airbnb metrics, yet I bet it is a primary driver in users booking with one listing over another. Despite this, I still think it is a good paper to start this analysis. They are likely correct that there is discrimination in Airbnb, and their logic underpinning this is sound.