DATA 467 | Linear Regression
A regression project asking which listing characteristics most strongly explain Airbnb nightly prices across New York City's five boroughs.
raw listings
cleaned listings
best adjusted R2
median nightly price
Airbnb prices reflect more than a host's chosen number. In New York City, price varies with room type, borough demand, review activity, host scale, and yearly availability. This project models nightly price using the 2019 NYC Airbnb open dataset.
Because raw price is strongly right-skewed, the main response is log(price). The final analysis uses ordinary least squares regression, interaction terms, diagnostics, and a logistic high-price check.
| Model | Specification | Purpose |
|---|---|---|
| Model 1 | log(price) using room type and borough | Test whether the strongest structural factors explain price patterns. |
| Model 2 | Add minimum nights, reviews, host listing count, and availability | Check whether listing activity variables change the main story. |
| Model 3 | Add room type by borough interaction and log-transform skewed predictors | Improve predictive flexibility while keeping the model interpretable. |
| Logistic GLM | Predict whether price is above the cleaned-sample median | Confirm whether the same predictors separate high-price listings. |
| Model | Adjusted R2 | AIC | BIC | RMSE |
|---|---|---|---|---|
| Model 1 | 0.480 | 53000.55 | 53060.52 | 0.4787 |
| Model 2 | 0.498 | 51591.63 | 51694.43 | 0.4700 |
| Model 3 | 0.507 | 50910.83 | 51082.17 | 0.4658 |
Model 3 had the best overall fit, but the improvement over Model 2 was moderate. The main interpretation stayed stable: location and room type drove most of the signal.
In Model 2, entire-home listings were about 118% higher than private rooms on the expected price scale.
Manhattan listings were about 35.6% higher than Brooklyn listings after controls.
The notebook predicted about $207 for a representative Manhattan entire-home listing, with a wide interval.
