Exploring New York City Airbnb Data: Neighborhoods and Trends

Analyst: Raine Huebner Gloege

Date: September 30th, 2025

Introduction

The purpose of this study was to better understand a subset of the Airbnbs available for rent in New York City using descriptive and inferential statistics. Airbnb provides an online platform for homeowners to rent out their properties to guests while collecting a partial commission from the bookings. This model offers flexibility for both hosts and guests, and provides travel opportunities not available in traditional hotels. This study focused on exploring how different metrics like reviews, policies, and costs varied and were distributed across the five boroughs of New York City included in the data: Brooklyn, Queens, Manhattan, The Bronx, and Staten Island. Additionally, we investigated whether there was a relationship between the construction year of Airbnb listing properties and the rental price per night.

Methodology

The data used for this study was sourced from Kaggle, an online platform for data scientists, which can be found at the following link: https://www.kaggle.com/datasets/arianazmoudeh/airbnbopendata. We used a subset of this data that includes 399 listings out of the 102599 available. Although our subset of data was relatively clean, we still examined it for potential inconsistencies, outliers, and missing values, to ensure appropriate handling prior to analysis. The variables of interest in this study had limited missing values.

This exploratory analysis was conducted in R and used the following packages:

• readxl to read in the data from a .xlsx file

• tidyverse to optimize data manipulation, analysis, and visualization processes

• coin to conduct permutation tests

• ggpubr to extends the functionalities of the ggplot2 package (within the tidyverse library) to assist in making publication ready visualizations

Results

We were first interested in whether the distribution of the cost per night varies by neighborhood. An initial graph of the cost per night for all listings did not show any skewness and suggested a relatively uniform distribution based on its density curve (See Figure 1).

Figure 1. Histogram of Airbnb listing prices per night.

The boxplot of cost per night by borough indicate a similar spread of prices — from the lowest to the highest typical values — for Brooklyn, Manhattan, Queens, and The Bronx (See Figure 2). In contrast, the price range for Staten Island appears narrower, which may be attributed to the limited number of listings in our dataset, with only four from that borough. The median values suggest that listings in The Bronx tend to be more expensive on average; however, since this observation is based on only 11 listings, it may not be representative of the broader market. The position of the median lines also indicates that the distributions of nightly prices in Brooklyn, Manhattan, and Staten Island are relatively symmetric, while Queens shows a right-skewed distribution and The Bronx shows a left-skewed one.

Figure 2. Boxplot showing the distributions of price per night by borough.

Next we looked into how many Airbnbs were instantly bookable for each borough. One listing with a missing value was remove from our stacked bar chart (See Figure 3). We found that a greater proportion of Airbnbs were instantly bookable in Queens (67%) than Brooklyn and Manhattan (49% and 44% respectively). Proportion labels were excluded from Staten Island and The Bronx due to their low contribution to the overall number of listings in our dataset. More data from these boroughs would be needed to provide meaningful proportion values.

Figure 3. Stacked bar chart showing the proportions of listing that are instantly bookable by borough.

We were also interested in visualizing the distribution of cancellation polices by borough (See Figure 4). We found that all boroughs have more listings with strict cancellation policies than flexible or moderate ones. Brooklyn and Manhattan have a similar proportions of listings for each policy category, while Queens had a higher proportion (73%) of listings with strict cancellation policies.

Figure 4. Stacked bar chart of the number of listings by the two cancellation policy types with proportions.

The process needed to conduct an inferential test to determine if the average review score (from a 1-5 rating system) varies by borough was complicated by the fact that our data violates the assumption of normality needed for ANOVA (See Figure 5).

Figure 5. QQ-Plot that violates the assumptions of normality by having residuals that do not generally fall on the diagonal line.

The Kruskal-Willis test can be used with data that violates the assumption of normality. However, it requires the assumption of equal variance, which we have already observed does not hold with our data due to a lack of listings from Staten Island and The Bronx. We opted for the non-parametric permutation test to try to determine if the average review score varies significantly by borough. While permutation tests generally can deal with differences in group sizes in can be sensitive to very large differences. Because of this, we ran the permutation test with all five boroughs, all boroughs except Staten Island, and all boroughs except Staten Island and The Bronx. The results of these tests all have p-values > 0.05, which does not suggest a significant difference between review scores by borough. However, since our dataset is a very small subset of the Airbnb data available it would be worth repeating these tests with more listing for a potentially more reliable result.

## 
##  Approximative K-Sample Fisher-Pitman Permutation Test
## 
## data:  review rate number by
##   neighbourhood group (Brooklyn, Manhattan, Queens, Staten Island, The Bronx)
## chi-squared = 5.6405, p-value = 0.2312

## 
##  Approximative K-Sample Fisher-Pitman Permutation Test
## 
## data:  review rate number by
##   neighbourhood group (Brooklyn, Manhattan, Queens, The Bronx)
## chi-squared = 5.4158, p-value = 0.1461

## 
##  Approximative K-Sample Fisher-Pitman Permutation Test
## 
## data:  review rate number by
##   neighbourhood group (Brooklyn, Manhattan, Queens)
## chi-squared = 2.9221, p-value = 0.2405

The final question we wanted to answer was whether there was a statistically significant relationship between the construction year of the listing’s property and price per night. The scatterplot of these two variables does not suggest a relationship as the regression line is nearly flat (See Figure 6).

Figure 6. Scatterplot of listing’s construction year versus price per night with confidence intervals on the regressio line.

The Pearson’s correlation coefficient of -0.068, the p-value of 0.1724, and a confidence interval of the true correlation between construction year and price per night provides strong evidence that there is not a statistically significant relationship between these two variables.

## 
##  Pearson's product-moment correlation
## 
## data:  AirBnb_data$`Construction year` and AirBnb_data$price
## t = -1.367, df = 397, p-value = 0.1724
## alternative hypothesis: true correlation is not equal to 0
## 95 percent confidence interval:
##  -0.16551115  0.02992703
## sample estimates:
##         cor 
## -0.06844872

Conclusion

This study highlighted the importance of considering data completeness when evaluating the results of descriptive and inferential analyses. We found that our dataset lacked sufficient coverage of The Bronx and Staten Island, which made drawing definitive conclusions of variations of listing metrics by borough difficult. Based on population density and geographic area alone, it is reasonable to assume that the amount of listings per borough could vary widely. However, to better understand how Airbnb listing by boroughs differ going forward, it would be wise to re-run these analyses with as much data as is computationally reasonable to hopefully get a better representative sample of each borough.