Does Airbnb have an impact on hotel room prices in Istanbul?
Yagmur Unsal
METU Department of EconomicsJanuary 2020
About
The sharing economy is now playing a game-changing role in the accommodation industry. Airbnb, one of the largest sharing platforms for rental private properties, was founded in 2008, previously named AirBed & Breakfast. The company has emerged as an alternative to the traditional hotel industry by connecting the two-sided demands, people who are searching for cheap accommodation, and people who are renting their private rooms/homes. Since its foundation, Airbnb has experienced so much growth that it has more listings than any hotel group in the world (Farronato and Fradkin, 2018).
This disruptive innovation in the accommodation industry has significant implications on the traditional hotel industry since Airbnb offers additional attributes to travelers that do not exist in traditional accommodation options like a cozy place to stay, central location, or flexible booking. Even if the general opinion about the impact of Airbnb on the hotel industry is negative, that is the existence of Airbnb affects hotel sales’ performance or its price negatively, the researches made for different regions show that the effects are indeed more complicated than that. According to the research which analyzed the Airbnb’s impact on hotel sales growth for San Francisco, the impact of Airbnb differentiates according to the hotel segments (Blal, Singal, & Templin, 2018). The low-end and midscale segmented hotels are the ones which affected negatively and are in danger of being replaced by the offers of sharing platforms. On the other hand, the luxury and upper-scaled hotels are affected positively with the ‘enlarging the pie’ argument, which is increasing Airbnb supply is contributing to the traditional hotel industry. Besides, the impacts are altering according to the different customer segments. While Airbnb is used as a substitute for a hotel by a group of travelers, some are preferring to use Airbnb on purpose due to the cultural experience and local sights that Airbnb offers rather than traditional hotels. Several researches conducted on this subject reveal that the existence of Airbnb offerings harm the ability of hotels to increase their room prices during periods of peak demand due to its limited capacity in contrast with the flexible Airbnb supply. However, studies also declare that Airbnb has time and geographic heterogeneity (Zervas et al., 2017; Farronato and Fradkin, 2018). In other words, the effect of Airbnb on incumbent firms and consumers differs over time and by region.
In this research, we analyzed the existence of Airbnb on hotel room prices for Istanbul. Our primary assumption prior to this research is the negative effect of Airbnb on hotel prices; that is, the Airbnb supply decreases the hotel prices in Istanbul. There is a consistency between the results of our research and the researches conducted for other countries in this regard. While expecting a negative relationship between Airbnb supply and hotel prices prior to the research, the results showed us that this is not a one-way relationship. Aside from the fact that the results are more complex than a negative correlation, our regression analysis has been demonstrated that even in Istanbul, effects differ from region to region.
While conducting research, we used daily compiled Airbnb and hotel data for Istanbul. Therefore, we were unable to conduct a time-series analysis or a fully detailed report on the effect of Airbnb on accomodation industry for Istanbul. In order to analyze the trends over time or investigate the Airbnb effect on hotel sales’ performance, more detailed and long-term data is needed.
We conducted the research including the regression analyses by RStudio and built this website where we published the results along with the mappings and correlation/descriptive tables by R Markdown. The packages used throughout the process are as follows:
Data
Data Source
We collected data from three different sources. The Airbnb data came from insideairbnb.com, a website which provides scraped data on the property rental marketplace company Airbnb. The price, rating, latitude and longitude data of Airbnb listings are all available in their scraped data report. The hotel data came from two different sources: Name, number of rooms, zipcodes, latitude and longitude data of hotels are kindly provided by Delta Check, a global accommodation reference database; name, price and zipcode data of hotels in Istanbul is scraped from hotels.com. The two seperate hotel data is merged by hotel names and zipcodes. Afterwards, Airbnb data is merged with the hotel data by using the coordinates (latitude and longitude) of both.
Hotel Data:
- Delta Check (Global Accommodation Reference Database)
- hotels.com
- Hotels’ price data are compiled on 7 November 2019
Airbnb Data:
- insideairbnb.com
- Downloaded on 8 December 2019
- Listings’ are compiled on November 2019
Variables
Hotel variables:
- Price (in turkish lira)
- Number of rooms (proxy for “Hotel Size”)
- Longitude
- Latitude
Airbnb variables:
- Price (in turkish lira)
- Review scores rating
- Longitude
- Latitude
Descriptives
Along with the variables (price, number of rooms, latitude and longitude) directly provided by the data sources for Airbnb and hotels, we also created Airbnb and hotel variables in order to use in regression analysis. These variables are as follows: Hotel supply represents the total number of hotels which are located within one kilometer distance of each hotel and the average number of rooms of these hotels is referred as ‘neighbouring hotel size’ which is used as a proxy for the hotel magnitude in the neighbourhood. By using the same technique, we also created Airbnb variables such as Airbnb supply which stands for the total number of Airbnb offerings located within one kilometer distance of each hotel and the average price and rating of these Airbnb listings, referred as Airbnb price and Airbnb rating respectively. The Airbnb and hotel variables along with their explanations can be found in the tables below.
| Legend | Variable |
|---|---|
| Hotel Size | Number of rooms in the hotel |
| Hotel Supply | Total number of rooms in hotels located within 1 km distance |
| Neighbouring Hotel Size | Average number of rooms in hotels located within 1 km distance |
| Legend | Variable |
|---|---|
| Airbnb Supply | Total number of Airbnb offerings located within 1 km distance |
| Airbnb Price | Average price of Airbnb offerings located within 1 km distance |
| Airbnb Rating | Average rating of Airbnb offerings located within 1 km distance |
Prior to the regression analysis, in order to check whether there is significant relationship between hotels and Airbnb offerings in Istanbul, we draw the correlation graph of Airbnb and hotel supply. According to the graph below, there is significant correlation between Airbnb and hotel supply in Istanbul in some regions. When we analyze the source of this correlation, we found out that the characteristics of Airbnb and hotel supply differentiates in certain regions.
Due to the differentiation in the characteristics of Airbnb and hotel supply in certain regions, by setting Bosphorus and Halic bridge as the breakpoints, we created four regions in Istanbul to be used further in regression analysis. The region in which the hotels are colored with red and referred to as the “Old City” of Istanbul, which is known for its historical places. The northern side of Halic Bridge colored with green is known for its more luxurious and upper-segmented touristic places. The remaining two regions represent other parts of Europe, indicated by purple, and the entire Anatolian region, indicated by blue. While the Anatolian continent can be examined in a single region, three different results have been observed within the European continent.
The correlation matrices and desctiptive tables below show the correlation between Airbnb variables and descriptives of each variable used in regression analysis according to the regions in Istanbul. There are 873 observations in total, that is the total number of hotels, for the inclusive Istanbul regression. Regarding the regional regressions, there are 64 observations in the Anatolian region, 421 observations in the Old City district, 68 and 319 for the rest of the European and the North of Halic region, respectively.
The correlation matrices below demonstrated that the characteristics of Airbnb supply differentiates from region to region. When we observe the correlation between Airbnb supply and average price of Airbnb listings according to different zones, there is significant multicollinearity in the rest of the European region and the Old City district, which colored with purple and red, respectively. However, the direction of the multicollinearity also differs. While we observe positive correlation between Airbnb supply and average Airbnb price in the Old City, there is negative correlation in the rest of the European region. In other words, while increasing Airbnb supply reduces the average Airbnb price in the rest of the European region, it increases the average price in the Old City district. Although these two regions are close geographically, correlations vary significantly. This difference was considered worthy of examination, and the results were obtained by examining with regressions. When we look at the correlation between average Airbnb rating and average Airbnb price, there is significant and negative multicollinearity only in Old City district. Since correlations between Airbnb variables vary regionally, these findings support the method we use when conducting research.
Anatolian Region
| Airbnb Supply | Airbnb Price | Airbnb Rating | |
|---|---|---|---|
| Airbnb Supply | 1.00 | -0.38 | -0.09 |
| Airbnb Price | -0.38 | 1.00 | 0.33 |
| Airbnb Rating | -0.09 | 0.33 | 1.00 |
| Room price | Number of rooms | Hotel supply | Neighbouring hotel size | Airbnb supply | Airbnb Price | Airbnb Rating | |
|---|---|---|---|---|---|---|---|
| min | 178.00 | 2.08 | 0.69 | 7.00 | 1.10 | 121.00 | 83.62 |
| max | 2599.00 | 5.83 | 2.89 | 320.00 | 7.13 | 857.00 | 100.00 |
| mean | 506.41 | 4.00 | 1.67 | 89.91 | 4.65 | 257.71 | 93.38 |
| std.dev | 434.96 | 1.09 | 0.75 | 59.13 | 1.61 | 137.51 | 3.29 |
The Old City District
| Airbnb Supply | Airbnb Price | Airbnb Rating | |
|---|---|---|---|
| Airbnb Supply | 1.00 | 0.63 | -0.37 |
| Airbnb Price | 0.63 | 1.00 | -0.05 |
| Airbnb Rating | -0.37 | -0.05 | 1.00 |
| Room price | Number of rooms | Hotel supply | Neighbouring hotel size | Airbnb supply | Airbnb Price | Airbnb Rating | |
|---|---|---|---|---|---|---|---|
| min | 133.00 | 1.61 | 1.10 | 15.00 | 4.83 | 236.90 | 86.37 |
| max | 6424.00 | 5.51 | 5.70 | 62.85 | 7.45 | 766.85 | 94.07 |
| mean | 696.85 | 3.35 | 5.18 | 36.17 | 6.75 | 568.02 | 88.02 |
| std.dev | 802.96 | 0.72 | 0.54 | 11.75 | 0.63 | 91.55 | 0.74 |
The Rest of the European Region
| Airbnb Supply | Airbnb Price | Airbnb Rating | |
|---|---|---|---|
| Airbnb Supply | 1.00 | -0.48 | -0.04 |
| Airbnb Price | -0.48 | 1.00 | -0.17 |
| Airbnb Rating | -0.04 | -0.17 | 1.00 |
| Room price | Number of rooms | Hotel supply | Neighbouring hotel size | Airbnb supply | Airbnb Price | Airbnb Rating | |
|---|---|---|---|---|---|---|---|
| min | 143.00 | 2.56 | 0.69 | 18.00 | 2.20 | 152.16 | 76.64 |
| max | 3304.00 | 6.03 | 2.48 | 330.00 | 5.06 | 1437.38 | 100.00 |
| mean | 483.31 | 4.45 | 1.67 | 126.57 | 3.78 | 375.66 | 87.94 |
| std.dev | 444.05 | 0.96 | 0.57 | 74.85 | 0.95 | 250.07 | 4.79 |
The North of Halic
| Airbnb Supply | Airbnb Price | Airbnb Rating | |
|---|---|---|---|
| Airbnb Supply | 1.00 | -0.05 | -0.34 |
| Airbnb Price | -0.05 | 1.00 | 0.15 |
| Airbnb Rating | -0.34 | 0.15 | 1.00 |
| Room price | Number of rooms | Hotel supply | Neighbouring hotel size | Airbnb supply | Airbnb Price | Airbnb Rating | |
|---|---|---|---|---|---|---|---|
| min | 127.00 | 1.39 | 0.69 | 12.50 | 2.71 | 179.02 | 84.32 |
| max | 6048.00 | 6.72 | 5.16 | 259.00 | 8.59 | 1047.64 | 97.92 |
| mean | 807.82 | 3.58 | 4.20 | 64.89 | 7.58 | 360.44 | 91.21 |
| std.dev | 1005.95 | 1.06 | 1.11 | 34.65 | 1.13 | 66.24 | 1.51 |
Istanbul
| Airbnb Supply | Airbnb Price | Airbnb Rating | |
|---|---|---|---|
| Airbnb Supply | 1.00 | 0.10 | -0.01 |
| Airbnb Price | 0.10 | 1.00 | -0.45 |
| Airbnb Rating | -0.01 | -0.45 | 1.00 |
| Room price | Number of rooms | Hotel supply | Neighbouring hotel size | Airbnb supply | Airbnb Price | Airbnb Rating | |
|---|---|---|---|---|---|---|---|
| min | 127.00 | 1.39 | 0.69 | 7.00 | 1.10 | 121.00 | 76.64 |
| max | 6424.00 | 6.72 | 5.70 | 330.00 | 8.59 | 1437.38 | 100.00 |
| mean | 706.31 | 3.57 | 4.29 | 57.61 | 6.67 | 454.40 | 89.57 |
| std.dev | 847.97 | 0.96 | 1.44 | 43.28 | 1.47 | 156.88 | 2.65 |
Model
We will estimate the model explaining room prices of the hotels in Istanbul. While our dependent variable is hotel room prices for one night, we use Airbnb supply, average price and rating of these Airbnb listings, number of rooms in each hotel, hotel supply, neighbouring hotel size and the interaction terms as the explanatory variables.
- Model 1
log(room price) ~ log(airbnb supply) + log(number of rooms)
- Model 2
log(room price) ~ log(airbnb supply) + airbnb rating + airbnb price + log(number of rooms) + neighbouring hotel size
- Model 3
log(room price) ~ log(airbnb supply) + airbnb rating + airbnb price + log(number of rooms) + log(hotel supply) + log(airbnb supply)log(number of rooms) + log(hotel supply)log(number of rooms)
Anatolian Region
According to the regression results for the Anatolian region, the number of rooms has significant and positive effect on room prices. 10% increase in the number of rooms, a parameter that represents the hotel size, also increases room prices by 1.7%. That is, when the hotel size increases, its room price also increases by 1.7% for a 10% increase in the magnitude. However, the effect of Airbnb supply on room price is insignificant for the Anatolian region. When we include other Airbnb and hotel parameters, we can see that along with the number of rooms, Airbnb price also positively affects room price.
| Dependent variable: | |||
| Log(room price) | |||
| 1st model | 2nd model | 3rd model | |
| (1) | (2) | (3) | |
| Airbnb supply | -0.001 | 0.057 | -0.152 |
| (0.053) | (0.064) | (0.313) | |
| Airbnb rating | 0.006 | 0.003 | |
| (0.023) | (0.024) | ||
| Airbnb price | 0.001** | 0.001* | |
| (0.001) | (0.001) | ||
| Number of rooms | 0.177** | 0.303*** | 0.050 |
| (0.079) | (0.095) | (0.301) | |
| Neighbouring hotel size | -0.002 | ||
| (0.002) | |||
| Hotel supply | 0.214 | ||
| (0.514) | |||
| Airbnb supply*Number of rooms | 0.072 | ||
| (0.078) | |||
| Hotel supply*Number of rooms | -0.091 | ||
| (0.139) | |||
| Constant | 5.308*** | 3.759* | 4.777** |
| (0.496) | (2.092) | (2.380) | |
| Observations | 64 | 64 | 64 |
| R2 | 0.101 | 0.223 | 0.228 |
| Adjusted R2 | 0.072 | 0.156 | 0.131 |
| Residual Std. Error | 0.583 (df = 61) | 0.556 (df = 58) | 0.564 (df = 56) |
| F Statistic | 3.441** (df = 2; 61) | 3.326** (df = 5; 58) | 2.357** (df = 7; 56) |
| Note: | p<0.1; p<0.05; p<0.01 | ||
The Old City District
The interaction terms we added to the third regression are significant for the Old City region. Therefore, we cannot trust the results of models in which the related parameters are isolated into consideration and must pay attention to the results of the third model. When we observe the results of the last model, we can mention the positive and significant effect of Airbnb supply on room price. The significance of interactions tells us the fact that when the hotel size increases, that is the number of rooms, if the total number of Airbnb listings in the close region is also increases, then the room price increases, but if the hotel supply in the close region increases, then the room price decreases. Based on these results, we can make a deduction which claims a competition between hotels and a supplementary role of Airbnb in the Old City region. However, since we observed significant interaction terms, we cannot be sure about the main effects of the Airbnb and hotel supply on room prices for Old City district.
| Dependent variable: | |||
| Log(room price) | |||
| 1st model | 2nd model | 3rd model | |
| (1) | (2) | (3) | |
| Airbnb supply | 0.297*** | -0.025 | -0.841** |
| (0.057) | (0.110) | (0.409) | |
| Airbnb rating | -0.042 | 0.099 | |
| (0.061) | (0.065) | ||
| Airbnb price | 0.0002 | 0.001* | |
| (0.001) | (0.001) | ||
| Number of rooms | 0.331*** | 0.409*** | -0.246 |
| (0.049) | (0.052) | (0.559) | |
| Neighbouring hotel size | -0.023*** | ||
| (0.006) | |||
| Hotel supply | 1.035** | ||
| (0.437) | |||
| Airbnb supply*Number of rooms | 0.331*** | ||
| (0.118) | |||
| Hotel supply*Number of rooms | -0.317*** | ||
| (0.121) | |||
| Constant | 3.116*** | 9.456 | -3.857 |
| (0.474) | (6.069) | (6.673) | |
| Observations | 421 | 421 | 421 |
| R2 | 0.113 | 0.161 | 0.151 |
| Adjusted R2 | 0.109 | 0.151 | 0.137 |
| Residual Std. Error | 0.673 (df = 418) | 0.657 (df = 415) | 0.662 (df = 413) |
| F Statistic | 26.581*** (df = 2; 418) | 15.950*** (df = 5; 415) | 10.506*** (df = 7; 413) |
| Note: | p<0.1; p<0.05; p<0.01 | ||
The Rest of the European Region
For this region, we cannot refer to a supplementary relationship between Airbnb and hotels since the interactions terms are now insignificant. Therefore, we can trust the models which include interaction variables (airbnb supply, number of rooms, hotel supply) in isolation. When we look at the first model, Airbnb supply and number of rooms positively affects room prices in addition to the average price of Airbnb listings located within one kilometer distance.
| Dependent variable: | |||
| Log(room price) | |||
| 1st model | 2nd model | 3rd model | |
| (1) | (2) | (3) | |
| Airbnb supply | 0.153** | 0.233*** | 0.403 |
| (0.072) | (0.083) | (0.691) | |
| Airbnb rating | 0.007 | 0.001 | |
| (0.013) | (0.014) | ||
| Airbnb price | 0.0005* | 0.001* | |
| (0.0003) | (0.0003) | ||
| Number of rooms | 0.333*** | 0.269*** | 0.304 |
| (0.071) | (0.099) | (0.318) | |
| Neighbouring hotel size | 0.001 | ||
| (0.001) | |||
| Hotel supply | -0.450 | ||
| (1.127) | |||
| Airbnb supply*Number of rooms | -0.020 | ||
| (0.142) | |||
| Hotel supply*Number of rooms | 0.049 | ||
| (0.225) | |||
| Constant | 3.919*** | 3.008** | 3.590* |
| (0.502) | (1.400) | (1.941) | |
| Observations | 68 | 68 | 68 |
| R2 | 0.254 | 0.301 | 0.314 |
| Adjusted R2 | 0.231 | 0.244 | 0.234 |
| Residual Std. Error | 0.501 (df = 65) | 0.496 (df = 62) | 0.500 (df = 60) |
| F Statistic | 11.062*** (df = 2; 65) | 5.332*** (df = 5; 62) | 3.919*** (df = 7; 60) |
| Note: | p<0.1; p<0.05; p<0.01 | ||
The North of Halic
We observed significant interaction terms for this region again. However, contrary to the situation for the other regions, Airbnb supply has no significant effect on room prices, but the average Airbnb rating has. Therefore, we can conclude that the characteristics of this region highly differentiates from the others. For a unit increase in average, Airbnb ratings around the region increases room prices by 6.5%. The effect of average Airbnb price on room prices is also in the same direction with the ratings, but with different magnitude. Besides, a 10% increase in the hotel size increases room price by 4%.
| Dependent variable: | |||
| Log(room price) | |||
| 1st model | 2nd model | 3rd model | |
| (1) | (2) | (3) | |
| Airbnb supply | 0.010 | 0.020 | -0.292 |
| (0.034) | (0.041) | (0.391) | |
| Airbnb rating | 0.062** | 0.060** | |
| (0.027) | (0.028) | ||
| Airbnb price | 0.001* | 0.001** | |
| (0.001) | (0.001) | ||
| Number of rooms | 0.379*** | 0.399*** | -0.201 |
| (0.037) | (0.038) | (0.373) | |
| Neighbouring hotel size | -0.001 | ||
| (0.001) | |||
| Hotel supply | 0.076 | ||
| (0.408) | |||
| Airbnb supply*Number of rooms | 0.093 | ||
| (0.100) | |||
| Hotel supply*Number of rooms | -0.024 | ||
| (0.101) | |||
| Constant | 4.874*** | -1.171 | 0.839 |
| (0.324) | (2.536) | (2.800) | |
| Observations | 319 | 319 | 319 |
| R2 | 0.266 | 0.291 | 0.301 |
| Adjusted R2 | 0.262 | 0.279 | 0.285 |
| Residual Std. Error | 0.667 (df = 316) | 0.659 (df = 313) | 0.656 (df = 311) |
| F Statistic | 57.400*** (df = 2; 316) | 25.660*** (df = 5; 313) | 19.119*** (df = 7; 311) |
| Note: | p<0.1; p<0.05; p<0.01 | ||
Istanbul
When we conduct the regressions by including all regions, that are all hotels and all Airbnb listings which are located within one kilometer distance in Istanbul, we can see that almost all parameters we included in the analysis are significant. These results are consistent with the region-based regression results of some regions while in contrast with others. In order to understand whether it is more accurate to a do region-based analysis or not, we did the log-likelihood ratio test. The likelihood ratio test showed us that we cannot pool the regions when we analyze the effects of Airbnb on hotel prices for Istanbul and we should estimate the regressions region by region.
| Dependent variable: | |||
| Log(room price) | |||
| 1st model | 2nd model | 3rd model | |
| (1) | (2) | (3) | |
| Airbnb supply | 0.127*** | 0.109*** | -0.176* |
| (0.017) | (0.019) | (0.098) | |
| Airbnb rating | 0.034*** | 0.021* | |
| (0.009) | (0.011) | ||
| Airbnb price | 0.001*** | 0.001*** | |
| (0.0002) | (0.0002) | ||
| Number of rooms | 0.323*** | 0.366*** | -0.078 |
| (0.026) | (0.028) | (0.104) | |
| Neighbouring hotel size | -0.001* | ||
| (0.001) | |||
| Hotel supply | 0.095 | ||
| (0.106) | |||
| Airbnb supply*Number of rooms | 0.094*** | ||
| (0.025) | |||
| Hotel supply*Number of rooms | -0.050* | ||
| (0.027) | |||
| Constant | 4.221*** | 0.726 | 3.402*** |
| (0.172) | (0.883) | (1.069) | |
| Observations | 873 | 873 | 873 |
| R2 | 0.160 | 0.204 | 0.219 |
| Adjusted R2 | 0.158 | 0.199 | 0.213 |
| Residual Std. Error | 0.666 (df = 870) | 0.650 (df = 867) | 0.644 (df = 865) |
| F Statistic | 82.826*** (df = 2; 870) | 44.325*** (df = 5; 867) | 34.725*** (df = 7; 865) |
| Note: | p<0.1; p<0.05; p<0.01 | ||
Results
Our main findings can be summarized as follows:
10% increase in the number of rooms and Airbnb price increases room prices by 1.7% and 3% respectively in the Anatolian region. In other words, when the hotel size and average price of Airbnb listings located within one kilometer distance increases by 10%, the room price increases by 1.7% and 3% respectively.
Airbnb and hotel parameters other than number of rooms and Airbnb price are insignificant in the Anatolian region.
Airbnb supply has negative effect on room price in the Old City district. When the total number of Airbnb listings located within one kilometer distance increases by 10%, room prices go up by 8.3% in the Old City district.
Airbnb price has significant and positive effect for this region. The magnitude of the effect of Airbnb price on room price is 1.7% for a 10% increase in Airbnb price, the same with the Anatolian region.
If the Airbnb supply within one kilometer distance increases, the room price increases in the Old City region and vice versa, provided that hotel size is increasing along with Airbnb supply.
Provided that the hotel size is increasing, if the hotel supply within one kilometer distance increases, room price decreases. On the other hand, under the same condition, Airbnb supply affecting room price positively in the Old City region.
Airbnb supply, Airbnb price and hotel size have significant and positive effect on room prices in the rest of the European region.
10% increase in the Airbnb supply and hotel size increases room price by 2% and 2,6% in the rest of the European region, respectively.
Surprisingly, Airbnb rating positively affects room prices by 6.4% in the North of Halic contrary to other regions.
The magnitude and the direction of the effect of Airbnb price and hotel size on room price in the North of Halic is approximately the same with the other regions.
In order to test whether these four regions can be pooled and considered as one region in regression analysis or not, we conducted log likelihood ratio rest. We conducted this test on the third model since it is the most advanced model we used in our regression analysis. Our null hypothesis is that we can pool these four models. Our test statistic is 43.2 with 27 degrees of freedom which is higher than the chi-square values under 5% and 10% probability level, 40.11 and 36.74 respectively. Therefore, we reject the null hypothesis and conclude that these four models cannot be pooled and each of them should be considered independently. As a result, regression results for Istanbul which evaluates the individual regions as a whole are misleading and insignificant.