# RECORD 4d32787a617d281fb36974fdcd341ff661cf2fe5 # TITLE: Housing market differentiation: the cases of Yenimahalle and Çankaya in Ankara # YEAR: 2015.0 VENUE: nan DOI: 10.3846/1648715x.2014.1000429 # AUTHORS: Leyla Alkan # PAGES: 14 CANDIDATE TABLE PAGES (1-based): [4, 8, 9, 12] # ABSTRACT: This study aims to investigate housing market differentiation, drawing upon the results of case studies of the Cankaya and Yenimahalle districts that adopt a set of statistical techniques. As a first step, a cluster analysis is carried out to identify whether identifiable clusters of housing attributes exist on the basis of neighborhoods. Next, a Multiple Discriminant Analysis (MDA) is applied to investigate the differences between clusters, and to understand which housing attributes contribute most to submarket separation. Finally, a hedonic price analysis is conducted for each cluster and for the overall market to identify price differences in the housing market. The results of the study support the hypothesis that the housing market is segmented in Yenimahalle and Cankaya, and that location is the main determining factor in this segmented structure of different house values. The study also reveals that within this segmented structure, each cluster has its own dynamics, and that the price formation in each cluster is dependent on different variables. ===== PAGE 1 (score 10, numeric-paren 1, table-words 0, var-words 9) ===== Copyright © 2015 Vilnius Gediminas Technical University (VGTU) Press http://www.tandfonline.com/TSPM INTERNATIONAL JOURNAL OF STRATEGIC PROPERTY MANAGEMENT ISSN 1648-715X / eISSN 1648-9179 2015  Volume  19(1): 13–26 doi:10.3846/1648715X.2014.1000429 * Corresponding author. E–mail: leylaalkan@gazi.edu.tr HOUSING MARKET DIFFERENTIATION: THE CASES OF YENIMAHALLE AND ÇANKAYA IN ANKARA Leyla ALKAN a, * a Department of City and Regional Planning, Faculty of Architecture, Gazi University, Eti Mah. Yükseliş Sok. No: 5, Maltepe, Ankara, Turkey Received 30 March 2013; accepted 24 January 2014 ABSTRACT. This study aims to investigate housing market differentiation, drawing upon the results of case studies of the Çankaya and Yenimahalle districts that adopt a set of statistical techniques. As a first step, a cluster analysis is carried out to identify whether identifiable clusters of housing attrib­ utes exist on the basis of neighborhoods. Next, a Multiple Discriminant Analysis (MDA) is applied to investigate the differences between clusters, and to understand which housing attributes contribute most to submarket separation. Finally, a hedonic price analysis is conducted for each cluster and for the overall market to identify price differences in the housing market. The results of the study support the hypothesis that the housing market is segmented in Yenimahalle and Çankaya, and that location is the main determining factor in this segmented structure of different house values. The study also reveals that within this segmented structure, each cluster has its own dynamics, and that the price formation in each cluster is dependent on different variables. KEYWORDS: Segmentation; Discriminant; Cluster; Hedonic; Submarkets 1. INTRODUCTION Following Ankara’s promotion to the status of cap­ ital in 1923, large-scale migration from Turkey’s rural areas began increasing the population of the city. By the 1930s, Ankara had become a compact city, arranged around a single dominant core; and as the old housing stock next to the city center began to fill up, the new arrivals began setting up homes on open farmland. The rapid urban growth of the 1950s in the wake of industrialization was unexpected, and sourcing adequate housing stock to meet the housing demand of successive waves of immigration was difficult. As a result, newcom­ ers to the city began to construct the illegal set­ tlements that would eventually encircle the city (Uzun 2005). In 1965, the Condominium Law was enacted, easing the construction of multi-story apartment houses with only a small amount of capital. This brought individual contractors and small entrepreneurs into the construction market, and as a result, a build-and-sell system of housing production became widespread throughout the city. A high-rise residential pattern began to emerge, especially in the central neighborhoods, which wit­ nessed an increase in density after higher build­ ings were permitted in the already built-up areas. The high-income groups, on the other hand, began moving out of the center, establishing new suburbs after the rapid increase in private car ownership witnessed in the 1970s (Uzun 2005; Onder 2000). Ankara has experienced a number of different but simultaneous processes since the 1980s, including suburbanization and the expansion of the city to­ wards the west, alongside a transformation of the inner-city residential areas as a result of the in­ crease in squatter settlements (Uzun 2005). This historical development has resulted in a segmentary urban pattern in Ankara, supported by deficiencies in the housing finance system in Turkey that bears little resemblance to the well- functioning systems found in other countries. In Turkey, home-ownership among the different so­ cio-economic groups is limited, and this has led low-income groups to create their own irregular mechanisms in order to become home-owners, alongside the regular methods of housing provi­ sion. As a result of this, discussions of socio-eco­ ===== PAGE 4 (score 21, numeric-paren 0, table-words 3, var-words 13) ===== 16 L. Alkan dicated that housing prices were significantly af­ fected by spatial determinants such as distance to the CBD, sub-centers, main transportation arter­ ies and the coast. This literature review has revealed that there is general consensus among researchers regarding the existence of submarkets; however, views tend to differ on how submarkets should be specified (Adair et al. 1996). With the emergence of hedonic price analyses and a number of different analytical methods, the identification and proper characteri­ zation of submarkets has gained critical importance (Goodman, Thibodeau 1998). This study examines house prices in Çankaya and Yenimahalle to un­ derstand the level of segmentation in the housing market. Following the methods of Keskin (2008) and Ozus et al. (2007), administrative boundaries are used to define each neighborhood in Çankaya and Yenimahalle; however, in order to define the boundaries of submarkets, the neighborhoods are grouped according to the set of techniques used in the study of Bourassa et al. (1999). As such, this study attempts to combine different statisti­ cal techniques in an analysis of the definition and construction of housing submarkets in Ankara. 3. METHODOLOGY AND DATA The data used in the analysis was collected from one of the largest real estate websites (Housing asked prices in Ankara 2011), providing informa­ tion about dwelling units on sale between February 2011 and October 2011 in the different neighbor­ hoods of Çankaya and Yenimahalle. The sample comprised 1,733 dwelling units, spread across 30 neighborhoods in Çankaya and 15 in Yenimahalle. For each neighborhood, data was garnered on the asked prices and the different attributes of each dwelling unit. Table 1 presents a description of the variables in the data. In the statistical analyses conducted here, cat­ egorical variables with more than two levels could not be entered directly into the model. To address this problem, the process of dummy coding was used to transform categorical variables with k lev­ els into k-1 variables. The analyses were run by en­ tering each new two-level variable into the model. For example, the Heating System, with four levels, was transformed into three dichotomous variables that contained the same information as the single categorical variable. In Table 1, Balcony, Garden, Stove, Boilers and Central ere entered into the Table 1. Description of variables Variable Description Price Price of the dwelling unit Age Age of the building in which the unit is located Floor area Floor area of the dwelling (m2) Rooms Number of rooms in the dwelling unit Balcony Dummy equal to 1 if there is a balcony Bathrooms Number of bathrooms in the dwelling unit Garden Dummy equal to 1 if there is a garden Story Story on which the unit is located Number of stories Total number of the stories in the apartment unit Stove Dummy equal to 1 if heating system is a solid-fuel stove Boilers Dummy equal to 1 if heating system is powered by a wall hung gas boiler Central Dummy equal to 1 if heating system is a common heating system for the entire block Descriptive statistics Mean Median Mode Std. Dev. Min. Max. Price 178437.27 138000 125000 120478.82 20000 1350000 Age 10.11 6 2 9.76 0 51 Floor area 141.67 125 100 61.31 42 550 Rooms 3.26 3 3 1.00 1 7 Balcony 0.95 1 1 0.20 0 1 Bathrooms 1.43 1 1 0.64 1 4 Garden 0.95 1 1 0.20 0 1 Story 2.48 2 3 2.67 –2 25 Number of stories 5.11 4 4 2.96 2 28 Stove 0.01 0 0 0.11 0 1 Boilers 0.84 1 1 0.36 0 1 Central 0.14 0 0 0.35 0 1 ===== PAGE 5 (score 15, numeric-paren 0, table-words 5, var-words 5) ===== 17 Housing market differentiation: the cases of Yenimahalle and Çankaya in Ankara analysis in the form of dummy variables, indicat­ ing the presence or absence of each characteristic in the dwelling unit. Following Bourassa et al. (1999), this study at­ tempts to apply a set of techniques to investigate housing submarkets in a three-stage analysis. In the first stage of the study, a cluster analysis was carried out to form homogenous groups of differ­ ent neighborhoods in Çankaya and Yenimahalle, which is a multivariate method that is used to classify data into a number of different groups on the basis of a set of measured variables (Cornish 2007). The groups created in the cluster analysis are not unique, since they depend upon the vari­ ables used and how cluster membership is defined (Chan 2005b). To analyze the different types of clusters in this study, an agglomerative hierarchi­ cal method was chosen, within which are a num­ ber of different methods used to determine which clusters should be merged at each stage (Cornish 2007). This study adopts the between-groups link­ age method with squared Euclidean Distance as the distance or similarity measure. In this method, the distance between two clusters is calculated as the average distance between all pairs of subjects in the two clusters (Cornish 2007). After deciding upon the most appropriate procedure and method for the analysis, the optimum number of clusters (k) in the sample needs to be defined. In this stage, a cluster analysis is applied, and the optimum number of clusters is chosen by examining the ag­ glomeration schedule generated by the results, af­ ter which the hierarchical cluster analysis is rerun with the selected number of clusters. In the second stage of the study, a discriminant analysis was used to determine the main charac­ teristics of dwelling units that differed across the clusters identified in the first stage. A discriminant analysis (DA) is a statistical technique used for dif­ ferentiating between groups when the independent variables are quantitative (Chan 2005a). If the in­ dependent variable has more than two groups, a Multiple Discriminant Analysis (MDA) is applied rather than a simple Discriminant Analysis. An MDA comprises three different steps: 1) The esti­ mation of the coefficients of the variables; 2) the calculation of the discriminant score of each case; and 3) the classification of each case, taking into account the discriminant scores (Leksrisakul, Ev­ ans 2005). A discriminant function is conducted at the end of an MDA, which is similar to a regression function. The unstandardized discriminant coeffi­ cient (analogous to the “b”s in the regression func­ tion) maximizes the distance between the means of the dependent variable. That is, the discriminant function maximizes the distance between clusters. Good predictors tend to have large weights (Burns, Burns 2008). In the third stage of the study, hedonic price models were created to estimate the housing char­ acteristics that define the price structure in each segment and the overall data. Such methods have proven to be an important means of analyzing com­ modities with various characteristics (Goodman, Thibodeau 1998), and are useful tools for explain­ ing the determinants of housing prices by taking into consideration structural, locational and envi­ ronmental attributes. Hedonic models are impor­ tant tools for understanding housing segmentation by focusing housing price changes (Keskin 2008). That said, owing to the lack of data on the loca­ tional and environmental attributes of the dwell­ ing units, the regression analysis was conducted focusing only on the structural characteristics of the dwelling units. 4. RESULTS 4.1. Cluster analysis In order to define whether identifiable clusters of housing attributes exist on the basis of neighbor­ hoods, a cluster analysis was conducted as a first step to determine whether certain combinations of physical attributes of dwelling units are common to certain neighborhoods. There are five variables in the data that are available for the classification of neighborhoods in the cluster analysis, informa­ tion on which can be found in Table 2. The clus­ ter analysis was run for 45 neighborhoods, and at the end of the examination of the agglomera­ tion schedule generated by the results nine differ­ ent clusters were identified. Table 3 presents the Table 2. Description of variables in cluster analysis Variable Description A.Price Average price of the dwelling units in the neighborhood A.Age Average age of the buildings in the neighborhood A.Floor Area Average floor area of the dwelling units in the neighborhood (m2) A.Rooms Average number of rooms of the dwellings in the neighborhood A.Number of Stories Average number of the stories in the apartment units in the neighborhood ===== PAGE 8 (score 35, numeric-paren 0, table-words 14, var-words 5) ===== 20 L. Alkan Table 4 provides descriptive statistics for each cluster after fixing each of the 1,733 cases in the two districts to their particular cluster in the data. In Table, it can be seen that age may be a good discriminator for Cluster 1 and Cluster 6, while other clusters score relatively similar results in this category. Table 4 also illustrates that Clus­ ter 5 and Cluster 9 have smaller floor areas, as­ sociated with lower house prices; while Cluster 6 has the most expensive average house prices and the largest floor areas. From this it can be safely assumed that floor area will constitute one of the most important variables affecting house prices. In the table, Cluster 1 has the oldest housing stock with the smallest number of stories, and Cluster 6 has the newest housing units with the highest number of stories, pointing to a significant rela­ tionship between age of the building and the num­ ber of stories. Also, the table illustrates that Clus­ ter 6 seems to be highly differentiated from the other clusters. To test these hypotheses and ex­ amine the segmented housing market within these two districts, Multiple Discriminant and Hedonic Function analyses were conducted as the second and the third steps in the study. 4.2. Multiple discriminant analysis A Multiple Discriminant Analysis (MDA) was used to investigate the differences between clusters, and to identify which attributes of housing contributed most to submarket separation. After excluding the variables with missing values, 1,139 cases were identified as suitable for analysis. The MDA was run using a range of key variables, including Price, Age, Floor Area, Rooms, and Number of Stories. The test of equality of the group means in Table 5 indicates that there are significant group differ­ ences in each of the independent variables (at p < 0.000) in the study. Table 6 provides information on the Eigenval­ ues, Canonical Correlations, Wilks’ Lambda and Coefficient Discriminant Functions. A total of five functions are produced in the discriminant analy­ sis, and both the eigenvalues and canonical cor­ relations are higher for the first function in the table. That is, the first function is statistically more significant than the others, and so the first function will be used for the estimation of the dis­ criminant function. Wilks’ Lambda tests the statistical significance of discriminant functions, illustrating that the first three discriminant functions have significant dis­ criminating power. In Table 6, the unstandardized coefficients are illustrated as canonical discrimi­ nant function coefficients. The table also provides information on standardized canonical function co­ efficients. The standardized discriminant function coefficients (b) indicate the partial contribution of each variable to the discriminate function, control­ ling all other variables in the equation. They can be used to assess the contribution of each variable to the discriminate function, and therefore provide information on the relative importance of each var­ iable (Burns, Burns 2008). The Price of the dwelling is the strongest pre­ dictor in the allocation of housing market clusters, as its “b” is the highest for the first function; this is followed by Floor Area and Number of Stories. In the second function, the strongest predictor is the Number of Stories, while Age is highly relevant for the third function. Only the first three functions are interpreted in this part, in that they explain more of the variance between groups, while Func­ tions 4 and 5 are not significant. The results of the discriminant analysis illustrate that Price, Floor Area, and Number of Stories are more significant variables in the formation of the segmented hous­ ing structure in both Çankaya and Yenimahalle. The unstandardized coefficients in Table 6 are used to create the discriminant function as: D = (0.014 × Age) + (–0.007 × Floor Area) + (–0.222 × Rooms) + (0.016 × Price(,000TL)) + (–0.128 × Number of Stories) – 0.646 Table 7 presents the classification results of the discriminant analysis, in which the rows are the observed categories of the dependent and the columns are the predicted categories. When the prediction is perfect, all cases will lie on the diago­ nal, meaning that the percentage of cases on the Table 5. Tests of equality of group means Wilks’ lambda F df1 df2 Sig. Age 0.942 8.713 8 1140 0.000 Floor area 0.915 13.299 8 1140 0.000 Rooms 0.944 8.429 8 1140 0.000 Price (,000 TL) 0.643 79.030 8 1140 0.000 Number of stories 0.867 21.770 8 1140 0.000 ===== PAGE 9 (score 21, numeric-paren 0, table-words 5, var-words 4) ===== 21 Housing market differentiation: the cases of Yenimahalle and Çankaya in Ankara Table 6. Eigenvalues, canonical correlations, Wilks’ lambda, coefficients discriminant functions Function 1 2 3 4 5 Eigenvalue 0.748 0.156 0.062 0.009 0.004 Canonical correlation 0.654 0.367 0.241 0.094 0.061 Tests of function(s) 1 thr 5 2 thr 5 3 thr 5 4 thr 5 5 Wilks’ lambda 0.460 0.804 0.930 0.987 0.996 Sig. 0.000 0.000 0.000 0.152 0.366 Standardized canonical discriminant function coefficients (b) Age 0.131 –0.241 1.035 0.048 –0.110 Floor area –0.406 –0.230 0.453 –0.958 1.667 Rooms –0.218 –0.063 0.031 1.695 –0.604 Price (,000 TL) 1.443 –0.004 –0.239 –0.147 –0.510 Number of stories –0.369 0.936 0.445 0.178 0.203 Canonical discriminant function coefficients Age 0.014 –0.025 0.108 0.005 –0.011 Floor area –0.007 –0.004 0.007 –0.016 0.028 Rooms –0.222 –0.065 0.031 1.726 –0.615 Price (,000 TL) 0.016 0.000 –0.003 –0.002 –0.006 Number of stories –0.128 0.325 0.155 0.062 0.070 (Constant) –0.646 –0.666 –2.642 –3.527 –1.245 Table 7. Classification results Clusters Predicted group membership Total 1 2 3 4 5 6 7 8 9 Original Count 1 1 2 0 0 0 1 0 1 0 5 2 15 31 24 65 12 11 7 13 4 182 3 16 4 19 34 0 4 4 8 1 90 4 1 13 43 183 63 2 7 13 29 354 5 2 15 2 82 113 0 3 1 65 283 6 0 0 2 0 0 18 4 2 1 27 7 3 3 1 8 3 5 6 2 6 37 8 3 0 6 2 0 6 2 11 1 31 9 2 11 2 25 20 0 1 2 77 140 % 1 20.0 40.0 0.0 0.0 0.0 20.0 0.0 20.0 0.0 100.0 2 8.2 17.0 13.2 35.7 6.6 6.0 3.8 7.1 2.2 100.0 3 17.8 4.4 21.1 37.8 0.0 4.4 4.4 8.9 1.1 100.0 4 0.3 3.7 12.1 51.7 17.8 0.6 2.0 3.7 8.2 100.0 5 0.7 5.3 0.7 29.0 39.9 0.0 1.1 0.4 23.0 100.0 6 0.0 0.0 7.4 0.0 0.0 66.7 14.8 7.4 3.7 100.0 7 8.1 8.1 2.7 21.6 8.1 13.5 16.2 5.4 16.2 100.0 8 9.7 0.0 19.4 6.5 0.0 19.4 6.5 35.5 3.2 100.0 9 1.4 7.9 1.4 17.9 14.3 0.0 0.7 1.4 55.0 100.0 39.9% of original grouped cases correctly classified. diagonal can be considered as the percentage of correct classifications in the analysis. The overall predictive accuracy of the discriminant function is called the “hit ratio” (Burns, Burns 2008), and Table 7 thus reveals that 39.9% of the total cases are correctly classified. This hit ratio can be ac­ cepted as being satisfactorily valid, in that most researchers accept a hit ratio of greater than 25%, which may be attributed to chance (Burns, Burns 2008). In the table, Cluster 6 is classified with greater accuracy (66.7%) than the other clusters, which could be predicted from the cluster statis­ tics in Table 4. 4.3. Hedonic regression After the cluster and discriminant analyses, a he­ donic regression analysis was performed for each of the submarkets and the overall sample. The ===== PAGE 10 (score 31, numeric-paren 0, table-words 1, var-words 6) ===== 22 L. Alkan Table 8. Results of hedonic regression ANOVA Unstandardized coefficients Standardized coefficients Variables in model (significant at the 0.05 level) F Sig R Square Adjusted R square B Std. Error Beta t Sig. VIF Cluster 1 – – – – – – – – – – Cluster 2 Floor area 52.542 0.000 0.804 0.789 0.769 0.154 0.497 4.992 0.000 5.823 Number of stories 12.081 2.002 0.417 6.035 0.000 2.811 Cluster 3 Age 15.983 0.000 0.738 0.692 1.569 0.752 0.185 2.087 0.042 1.525 Floor area 0.588 0.194 0.482 3.030 0.004 4.936 Story 16.200 5.945 0.456 2.725 0.009 5.467 Cluster 4 Floor area 58.368 0.000 0.687 0.675 0.606 0.105 0.440 5.770 0.000 4.939 Story 6.896 1.713 0.183 4.025 0.000 1.764 Number of stories 7.491 1.668 0.193 4.492 0.000 1.575 Cluster 5 Floor area 22.387 0.000 0.511 0.488 0.379 0.092 0.352 4.115 0.000 3.211 Rooms 13.746 4.150 0.258 3.312 0.001 2.659 Stove –32.285 14.263 –0.112 –2.264 0.025 1.077 Story 3.604 1.184 0.194 3.043 0.003 1.785 Cluster 6 Floor area 6.178 0.003 0.805 0.674 1.445 0.540 0.698 2.675 0.020 4.179 Cluster 7 Floor area 5.330 0.001 0.681 0.553 1.959 0.682 1.021 2.874 0.009 7.911 Number of stories 25.875 9.241 0.627 2.800 0.011 3.137 Cluster 8 Central 7.647 0.001 0.805 0.699 224.649 64.840 0.821 3.465 0.004 3.735 Bathrooms 110.295 50.079 0.403 2.202 0.046 2.228 Story 38.114 15.913 0.850 2.395 0.032 8.373 Cluster 9 Floor area 21.798 0.000 0.651 0.621 1.587 0.201 0.914 7.889 0.000 4.044 Rooms –25.211 9.581 –0.269 –2.632 0.010 3.159 Overall (Constant) 144.394 0.000 0.751 0.746 –70.923 17.987 – –3.943 0.000 – Floor area 0.777 0.063 0.449 12.401 0.000 4.525 Central 29.481 7.078 0.085 4.165 0.000 1.425 Bathrooms 11.400 4.343 0.072 2.625 0.009 2.593 Story 5.938 0.943 0.149 6.297 0.000 1.929 Number of stories 4.493 0.871 0.129 5.159 0.000 2.162 cls1 160.431 25.607 0.111 6.265 0.000 1.094 cls2 54.322 7.410 0.175 7.331 0.000 1.976 cls3 77.309 9.049 0.181 8.543 0.000 1.560 cls4 32.486 6.411 0.139 5.067 0.000 2.617 cls5 –14.037 6.435 –0.057 –2.182 0.029 2.326 cls6 189.974 13.267 0.268 14.320 0.000 1.211 cls7 51.056 11.548 0.084 4.421 0.000 1.255 cls8 159.803 13.398 0.225 11.927 0.000 1.235 ===== PAGE 12 (score 25, numeric-paren 0, table-words 2, var-words 3) ===== 24 L. Alkan have relatively newer housing stock, along with Çamlıca in Yenimahalle. The table further shows that there has been a strong shift towards larg­ er dwelling units, but not increases in building heights in these neighborhoods in Çankaya; where­ as increases in the heights of buildings are more remarkable in Çamlıca, although there has been no significant increase in Floor Area. Table 9. Residential characteristics of neighborhoods Cluster District Neighborhood Age Floor area Number of stories Price 1 Çankaya Anıttepe 24.40 160.00 3.80 317 000.00 2 Çankaya Ayrancı 26.82 128.10 5.13 199 188.12 Cevizlidere 4.94 171.86 4.36 211 616.28 Emek 14.83 132.42 3.64 227 833.33 Kırkkonaklar 6.60 184.29 3.81 210 455.88 Yenimahalle Çamlıca 4.50 134.42 10.89 217 263.16 3 Çankaya Bahçeli 29.82 140.72 5.41 265 250.00 Birlikmahallesi 6.93 188.42 4.10 264 188.17 Çankaya 12.71 144.59 6.14 248 676.47 Kavaklıdere 30.67 143.81 4.79 252 452.38 4 Çankaya Balgat 9.76 142.68 3.70 177 347.83 Dikmen 9.14 143.83 3.90 160 119.27 İlker 9.49 139.47 4.13 151 496.09 Keklikpınarı 4.73 149.03 4.43 167 898.57 Öveçler 8.98 144.11 4.20 168 910.96 Yenimahalle Çarşı 7.56 146.17 5.03 172 159.09 5 Çankaya Büyükesat 13.33 124.55 3.73 117 500.00 Cebeci 14.83 108.13 3.90 114 229.17 İncesu 2.14 105.61 4.84 108 173.91 Sağlık 18.00 104.90 4.51 116 979.59 Seyranbağları 11.93 112.42 4.03 114 930.23 Sokullu 16.97 108.91 4.10 113 320.45 Topraklık 3.50 101.67 3.25 109 666.67 Yenimahalle Demet 17.10 116.80 6.09 95 806.67 Ergenekon 9.79 125.96 3.31 104 000.00 Karşıyaka 16.35 111.86 6.71 83 071.43 Pamuklar 3.56 145.98 6.15 119 559.52 Susuz 5.25 135.90 9.60 99 700.00 Yakacık 2.08 137.25 5.00 89 166.67 Yunusemre 18.00 125.83 4.33 80 416.67 6 Çankaya Çukurambar 5.00 196.44 8.56 432 777.78 Yıldızevler 1.85 212.64 9.26 454 107.14 7 Çankaya Gaziosmanpaşa 13.88 167.17 4.98 299 091.67 Yenimahalle Çayyolu 6.55 152.27 9.83 284 204.55 8 Çankaya Hilal 4.47 206.27 4.21 371 409.09 Oran 15.30 176.33 8.13 370 333.33 Sancak 10.40 231.29 3.71 356 428.57 Yenimahalle Yaşamkent 2.60 185.67 9.80 358 333.33 9 Çankaya Küçükesat 22.90 116.82 4.19 139 214.29 Kızılay 30.50 100.00 4.29 144 714.29 Öncebeci 4.81 107.50 4.26 130 783.33 Yenimahalle Gazimahallesi 12.00 107.00 3.20 136 600.00 Kentkoop 6.01 132.17 8.25 142 765.22 Ostim 5.33 127.20 5.90 124 800.00 Ragıp Tüzün 8.36 136.09 3.64 145 727.27 All of the neighborhoods in Cluster 3 are in Çankaya, where there has been a shift to larger housing units in new settlements like Birlikma­ hallesi. Again, there has been no increase in the number of stories in newer buildings, and the he­ donic analysis shows a positive relationship be­ tween the age of the building and the price of the dwelling units for this cluster. The average prices ===== PAGE 13 (score 9, numeric-paren 0, table-words 3, var-words 3) ===== 25 Housing market differentiation: the cases of Yenimahalle and Çankaya in Ankara in Bahçelievler and Kavaklıdere in particular are high, despite the average age of the housing stock in these locations, being the inner, older and more prestigious areas of Ankara. There is a more homogenous pattern in Clus­ ter 4 in terms of Age, Floor Area and Number of Stories. As discussed previously, there is a posi­ tive relationship between Price and the Number of Stories for this cluster, and Table 9 shows that Çarşı in Yenimahalle contain relatively higher apartment buildings. In Cluster 5, there is no significant increase in Floor Area or the Number of Stories in Topraklık and İncesu, where the housing stock is newer than in other parts of Çankaya. In Yenimahalle, the newer housing stock in Pamuklar, Yakacık and Susuz features relatively larger dwelling units, and a greater Number of Stories in Susuz. Cluster 6 contains the most expensive neigh­ borhoods in the Çankaya district – Çukurambar and Yıldızevler. These neighborhoods have similar patterns in terms of larger floor areas and higher apartment units. Cluster 7 comprises Çayyolu in Yenimahalle and Gaziosmanpaşa in Çankaya, which feature similar mean prices. There is no significant differ­ ence in the average floor area of the dwelling units in these two neighborhoods, although the housing stock in Çayyolu is relatively newer than that of Gaziosmanpaşa. That said, Çayyolu is a subur­ ban residential area of Ankara that was affected by the rapid growth in high-rise apartment build­ ings, as discussed at the beginning of the study. It is important to note that the hedonic analysis highlights a positive relationship between Price and the Number of Stories for this cluster. As discussed previously, the results of the he­ donic analysis highlight no significant relation­ ship between house Prices and the Number of Stories for Cluster 8. All of the neighborhoods in Çankaya in Cluster 8 feature relatively larger dwelling units; however, there has been a shift to a lower Numbers of Stories in newer settlements like Sancak and Hilal in contrast to Oran, where the housing stock is older. There has also been a growing trend of high-rise buildings in Yaşamkent (another suburban settlement in Ankara) in Yeni­ mahalle. Table 9 illustrates that the Öncebeci neighbor­ hood has the newest housing stock in Çankaya in Cluster 9, although there is no significant varia­ tion in Çankaya in terms of Floor Area and the Number of Stories for this cluster. In addition, there is again a trend towards higher apartment buildings in the newer settlements of Yenimahalle in this cluster. Table 9 highlights a shift towards larger dwell­ ing units within most clusters, consistent with the results of the hedonic analysis; and the overall trend in floor area is upwards, both in Çankaya and Yenimahalle. In terms of number of stories within clusters, there is no immediate change in Çankaya; however, the data shows an important increase in higher apartment buildings, especially in the outer neighborhoods of Yenimahalle. That is, the demand side of the housing market indi­ cated in the hedonic price analysis does not appear to have had an observable impact on the supply side of housing in terms of the number of stories. 6. CONCLUSIONS This paper has investigated housing market dif­ ferentiation in Ankara through a case study of the Yenimahalle and Çankaya districts by employing a cluster analysis, a discriminant analysis and a hedonic price analysis. The paper demonstrates that the housing market in Çankaya and Yeni­ mahalle has a segmented structure, with location being the main determining factor of different house prices. The hedonic analysis revealed that analyzing the price structure in the market area as a whole prohibits the development of an un­ derstanding of the housing market, as the price structures of different clusters have their own dy­ namics, consistent with the results of the study by Alkay (2008). The results of this study reveal the demand- side preferences of households in different neigh­ borhoods. Both Çankaya and Yenimahalle have experienced a trend of larger housing units in new residential areas, although there has been no immediate change in Çankaya in terms of the Number of Stories. In contrast, the data shows a converse increase in higher apartments, espe­ cially in the suburban neighborhoods of Yenima­ halle. This trend in growth areas towards higher apartment buildings is inconsistent with the hous­ ing choice of households, as the results of the he­ donic analysis illustrate, highlighting an invisible mechanism that shapes the residential pattern of growth areas in Ankara. The results of the study can be seen as a useful tool for understanding the demand side of the housing market in Çan­ kaya and Yenimahalle, and further studies may be conducted to cover all urban residential areas in Ankara in order to characterize the residential pattern for the city as a whole.