4 Result and Conclusion
This section is separated in two parts. First, we will first discuss the finding from the model we selected. Secondly, we develop on our conclusion finish with our main learning point.
4.1 Interpretation of the model
4.1.1 Prediction of airbnb prices
In this section, we will discuss the model predicted the general price of an airbnb based on its characteristics.
| Observations | 43764 |
| Dependent variable | price |
| Type | OLS linear regression |
| F(20,43743) | 924.41 |
| R² | 0.30 |
| Adj. R² | 0.30 |
| Est. | S.E. | t val. | p | |
|---|---|---|---|---|
| (Intercept) | -143.21 | 5.58 | -25.68 | 0.00 |
| property_typeCondominium | -32.80 | 7.91 | -4.14 | 0.00 |
| property_typeGuest suite | 30.15 | 10.30 | 2.93 | 0.00 |
| property_typeGuesthouse | 45.62 | 8.48 | 5.38 | 0.00 |
| property_typeHouse | -12.74 | 4.89 | -2.60 | 0.01 |
| property_typeTownhouse | -85.98 | 10.47 | -8.21 | 0.00 |
| property_typeOther | 79.02 | 6.07 | 13.03 | 0.00 |
| room_typePrivate room | -46.67 | 4.59 | -10.16 | 0.00 |
| room_typeShared room | -128.78 | 9.78 | -13.17 | 0.00 |
| accommodates | 16.55 | 1.32 | 12.54 | 0.00 |
| bathrooms | 178.59 | 3.09 | 57.74 | 0.00 |
| bedrooms | 74.29 | 3.11 | 23.89 | 0.00 |
| beds | -34.14 | 1.89 | -18.11 | 0.00 |
| pool1 | 69.89 | 5.50 | 12.72 | 0.00 |
| hot_tub1 | 37.65 | 5.80 | 6.49 | 0.00 |
| gym1 | -23.54 | 6.38 | -3.69 | 0.00 |
| fireplace1 | 31.36 | 4.65 | 6.75 | 0.00 |
| garden1 | -11.38 | 5.32 | -2.14 | 0.03 |
| AC1 | -22.91 | 4.23 | -5.42 | 0.00 |
| sauna1 | 374.00 | 105.63 | 3.54 | 0.00 |
| balcony_patio1 | -28.71 | 4.84 | -5.94 | 0.00 |
| Standard errors: OLS |
It is interesting to notice that the Intercept is negative. This might look like an issue but, in fact, it isn’t as a property has to have a minimum of characteristic. For example, An airbnb must have a bedroom and a bathroom.Secondly, we see that an apartment (the default value in our model) has more value than a house, townhouse or condomium given the same characterstics. We can also note that a private room rented will have a lower price than a whole apartment and a shared room will be even lower.
Regarding the bathroom variable, we see that a really huger impact on price while beds have a negative effect. In our opinion, the number of bathrooms better reflects the size of the good and therefore has a bigger impact on the price. We couldn’t fully figure why the beds have a negative impact though.
A few other non intuitive coefficients are the one of gym, garden, AC and balcony/patio.
4.1.2 prediction of the impact of the criminality
| Observations | 43764 |
| Dependent variable | price |
| Type | OLS linear regression |
| F(44,43720) | 984.30 |
| R² | 0.50 |
| Adj. R² | 0.50 |
| Est. | S.E. | t val. | p | |
|---|---|---|---|---|
| pred_price | 0.75 | 0.03 | 24.19 | 0.00 |
| score_violent_COMPSTATvery low | 5.11 | 9.33 | 0.55 | 0.58 |
| score_violent_COMPSTATlow | 43.51 | 21.43 | 2.03 | 0.04 |
| score_violent_COMPSTATmedium | 24.99 | 22.34 | 1.12 | 0.26 |
| score_violent_COMPSTAThigh | 197.64 | 44.51 | 4.44 | 0.00 |
| score_violent_COMPSTATvery high | 209.57 | 46.53 | 4.50 | 0.00 |
| score_property_crimelow | 4.01 | 13.14 | 0.31 | 0.76 |
| score_property_crimemedium | -56.49 | 17.93 | -3.15 | 0.00 |
| score_property_crimehigh | -71.35 | 32.85 | -2.17 | 0.03 |
| score_property_crimevery high | -64.46 | 33.95 | -1.90 | 0.06 |
| score_crime_murder_rapelow | 67.95 | 15.63 | 4.35 | 0.00 |
| score_crime_murder_rapemedium | -44.32 | 25.58 | -1.73 | 0.08 |
| score_crime_murder_rapehigh | -94.20 | 26.70 | -3.53 | 0.00 |
| score_robberylow | 33.82 | 18.68 | 1.81 | 0.07 |
| score_robberymedium | 61.92 | 27.20 | 2.28 | 0.02 |
| score_robberyhigh | 112.23 | 29.13 | 3.85 | 0.00 |
| score_burglarylow | -186.70 | 16.01 | -11.66 | 0.00 |
| score_burglarymedium | -187.05 | 20.23 | -9.25 | 0.00 |
| score_burglaryhigh | -211.64 | 20.82 | -10.16 | 0.00 |
| score_vehiclelow | 134.73 | 22.21 | 6.07 | 0.00 |
| score_vehiclemedium | 83.23 | 15.48 | 5.38 | 0.00 |
| score_vehiclehigh | 86.45 | 25.22 | 3.43 | 0.00 |
| score_vehiclevery high | 105.03 | 26.19 | 4.01 | 0.00 |
| pred_price:score_violent_COMPSTATlow | -0.10 | 0.07 | -1.40 | 0.16 |
| pred_price:score_violent_COMPSTATmedium | -0.09 | 0.07 | -1.40 | 0.16 |
| pred_price:score_violent_COMPSTAThigh | -0.90 | 0.13 | -7.01 | 0.00 |
| pred_price:score_violent_COMPSTATvery high | -0.98 | 0.14 | -7.28 | 0.00 |
| pred_price:score_property_crimelow | -0.15 | 0.05 | -3.32 | 0.00 |
| pred_price:score_property_crimemedium | 0.73 | 0.05 | 13.83 | 0.00 |
| pred_price:score_property_crimehigh | 0.80 | 0.10 | 8.38 | 0.00 |
| pred_price:score_property_crimevery high | 0.92 | 0.10 | 9.16 | 0.00 |
| pred_price:score_crime_murder_rapelow | -0.51 | 0.05 | -9.51 | 0.00 |
| pred_price:score_crime_murder_rapemedium | -0.02 | 0.08 | -0.28 | 0.78 |
| pred_price:score_crime_murder_rapehigh | 0.38 | 0.08 | 4.68 | 0.00 |
| pred_price:score_robberylow | -0.14 | 0.07 | -2.16 | 0.03 |
| pred_price:score_robberymedium | -0.39 | 0.08 | -4.76 | 0.00 |
| pred_price:score_robberyhigh | -0.78 | 0.09 | -8.85 | 0.00 |
| pred_price:score_burglarylow | 1.10 | 0.04 | 24.83 | 0.00 |
| pred_price:score_burglarymedium | 1.07 | 0.07 | 15.62 | 0.00 |
| pred_price:score_burglaryhigh | 1.37 | 0.07 | 19.42 | 0.00 |
| pred_price:score_vehiclelow | -0.64 | 0.08 | -7.98 | 0.00 |
| pred_price:score_vehiclemedium | -0.52 | 0.05 | -9.97 | 0.00 |
| pred_price:score_vehiclehigh | -0.81 | 0.09 | -9.42 | 0.00 |
| pred_price:score_vehiclevery high | -0.96 | 0.09 | -10.57 | 0.00 |
| Standard errors: OLS |
As we can see, the the violent crimes, robberies, vehicle crimes have a positive impact on our score as well as the murder and rapes except when very high where it has a negative impact. A positive impact implies that an increase on one of those variable increases the price of an airbnb listing.
Regarding the property crimes, they have a positive impact on crime expect at a certain level, which is the medium level, meaning between the 40% percentile and the 60% percentile, the impact of this type of crime affects negatively the price.
Finally, the burglaries seem to be the only variable having constantly a negative impact on our score.
Some of those results are surprising. For example, we can deduct that robberies are driving the price of an Airbnb up. It might look confusing but, it could be explained by the fact that most high end businesses where most robberies happen are located in “rich” neighbourhoods. This could indicate that it is not the crimes that are affecting the price but the price that is affecting the criminality.
4.2 Conclusion
The value of real estate is affected by many factors, some more subjective than others. This is all the more true in the context of short-term rental with the case of Airbnb. Prices are generally higher and fluctuate more easily depending on demand or the season.
With our model, we have chosen to consider only the specific characteristics of the property in question when determining the general rental price. Thus, we concluded that our model could explain 30 % of the price variation. A more complex model could have given us a better score, but this still reflects the difficulty of evaluating a property.
This model was also used as the basis for the second part of our analysis to determine the impact of crime and more specifically of each type of crime on rental prices.
To respond to this problem, we tried two approaches; one based on crime per 100’000 inhabitants and one based on a score created from the quantiles. The last one gave us a better result which allowed us to explain almost 50% of the price variation. A significant improvement compared to the first model from which we were able to draw interesting conclusions.
Some relationships also surprised us, and we realized that it was sometimes difficult to say what was the cause and what was the consequence. Ex: the positive relationship between high property crime and price.
Overall we have a better understanding of the effects of crime of Airbnbs in Los Angeles. However, these effects need to be taken with a grain of salt and interpreted with caution.