1# RECORD 4d32787a617d281fb36974fdcd341ff661cf2fe52# TITLE: Housing market differentiation: the cases of Yenimahalle and Çankaya in Ankara3# YEAR: 2015.0 VENUE: nan DOI: 10.3846/1648715x.2014.10004294# AUTHORS: Leyla Alkan5# PAGES: 14 CANDIDATE TABLE PAGES (1-based): [4, 8, 9, 12]6# 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.789===== PAGE 1 (score 10, numeric-paren 1, table-words 0, var-words 9) =====1011Copyright © 2015 Vilnius Gediminas Technical University (VGTU) Press12http://www.tandfonline.com/TSPM13INTERNATIONAL JOURNAL OF STRATEGIC PROPERTY MANAGEMENT14ISSN 1648-715X / eISSN 1648-9179152015 Volume 19(1): 13–2616doi:10.3846/1648715X.2014.100042917* Corresponding author. E–mail: leylaalkan@gazi.edu.tr18HOUSING MARKET DIFFERENTIATION: THE CASES OF 19YENIMAHALLE AND ÇANKAYA IN ANKARA 20Leyla ALKAN a, *21a Department of City and Regional Planning, Faculty of Architecture, Gazi University, Eti Mah. Yükseliş 22Sok. No: 5, Maltepe, Ankara, Turkey23Received 30 March 2013; accepted 24 January 201424ABSTRACT. This study aims to investigate housing market differentiation, drawing upon the results 25of case studies of the Çankaya and Yenimahalle districts that adopt a set of statistical techniques. As 26a first step, a cluster analysis is carried out to identify whether identifiable clusters of housing attrib27utes exist on the basis of neighborhoods. Next, a Multiple Discriminant Analysis (MDA) is applied to 28investigate the differences between clusters, and to understand which housing attributes contribute 29most to submarket separation. Finally, a hedonic price analysis is conducted for each cluster and for 30the overall market to identify price differences in the housing market. The results of the study support 31the hypothesis that the housing market is segmented in Yenimahalle and Çankaya, and that location 32is the main determining factor in this segmented structure of different house values. The study also 33reveals that within this segmented structure, each cluster has its own dynamics, and that the price 34formation in each cluster is dependent on different variables.35KEYWORDS: Segmentation; Discriminant; Cluster; Hedonic; Submarkets361. INTRODUCTION37Following Ankara’s promotion to the status of cap38ital in 1923, large-scale migration from Turkey’s 39rural areas began increasing the population of the 40city. By the 1930s, Ankara had become a compact 41city, arranged around a single dominant core; and 42as the old housing stock next to the city center 43began to fill up, the new arrivals began setting up 44homes on open farmland. The rapid urban growth 45of the 1950s in the wake of industrialization was 46unexpected, and sourcing adequate housing stock 47to meet the housing demand of successive waves 48of immigration was difficult. As a result, newcom49ers to the city began to construct the illegal set50tlements that would eventually encircle the city 51(Uzun 2005). In 1965, the Condominium Law was 52enacted, easing the construction of multi-story 53apartment houses with only a small amount of 54capital. This brought individual contractors and 55small entrepreneurs into the construction market, 56and as a result, a build-and-sell system of housing 57production became widespread throughout the city. 58A high-rise residential pattern began to emerge, 59especially in the central neighborhoods, which wit60nessed an increase in density after higher build61ings were permitted in the already built-up areas. 62The high-income groups, on the other hand, began 63moving out of the center, establishing new suburbs 64after the rapid increase in private car ownership 65witnessed in the 1970s (Uzun 2005; Onder 2000). 66Ankara has experienced a number of different but 67simultaneous processes since the 1980s, including 68suburbanization and the expansion of the city to69wards the west, alongside a transformation of the 70inner-city residential areas as a result of the in71crease in squatter settlements (Uzun 2005). 72This historical development has resulted in a 73segmentary urban pattern in Ankara, supported 74by deficiencies in the housing finance system in 75Turkey that bears little resemblance to the well-76functioning systems found in other countries. In 77Turkey, home-ownership among the different so78cio-economic groups is limited, and this has led 79low-income groups to create their own irregular 80mechanisms in order to become home-owners, 81alongside the regular methods of housing provi82sion. As a result of this, discussions of socio-eco838485===== PAGE 4 (score 21, numeric-paren 0, table-words 3, var-words 13) =====86871688L. Alkan89dicated that housing prices were significantly af90fected by spatial determinants such as distance to 91the CBD, sub-centers, main transportation arter92ies and the coast.93This literature review has revealed that there 94is general consensus among researchers regarding 95the existence of submarkets; however, views tend 96to differ on how submarkets should be specified 97(Adair et al. 1996). With the emergence of hedonic 98price analyses and a number of different analytical 99methods, the identification and proper characteri100zation of submarkets has gained critical importance 101(Goodman, Thibodeau 1998). This study examines 102house prices in Çankaya and Yenimahalle to un103derstand the level of segmentation in the housing 104market. Following the methods of Keskin (2008) 105and Ozus et al. (2007), administrative boundaries 106are used to define each neighborhood in Çankaya 107and Yenimahalle; however, in order to define the 108boundaries of submarkets, the neighborhoods are 109grouped according to the set of techniques used 110in the study of Bourassa et al. (1999). As such, 111this study attempts to combine different statisti112cal techniques in an analysis of the definition and 113construction of housing submarkets in Ankara. 1143. METHODOLOGY AND DATA115The data used in the analysis was collected from 116one of the largest real estate websites (Housing 117asked prices in Ankara 2011), providing informa118tion about dwelling units on sale between February 1192011 and October 2011 in the different neighbor120hoods of Çankaya and Yenimahalle. The sample 121comprised 1,733 dwelling units, spread across 30 122neighborhoods in Çankaya and 15 in Yenimahalle. 123For each neighborhood, data was garnered on the 124asked prices and the different attributes of each 125dwelling unit. Table 1 presents a description of the 126variables in the data. 127In the statistical analyses conducted here, cat128egorical variables with more than two levels could 129not be entered directly into the model. To address 130this problem, the process of dummy coding was 131used to transform categorical variables with k lev132els into k-1 variables. The analyses were run by en133tering each new two-level variable into the model. 134For example, the Heating System, with four levels, 135was transformed into three dichotomous variables 136that contained the same information as the single 137categorical variable. In Table 1, Balcony, Garden, 138Stove, Boilers and Central ere entered into the 139Table 1. Description of variables140Variable141Description142Price143Price of the dwelling unit144Age145Age of the building in which the unit is located146Floor area147Floor area of the dwelling (m2)148Rooms149Number of rooms in the dwelling unit150Balcony151Dummy equal to 1 if there is a balcony152Bathrooms153Number of bathrooms in the dwelling unit154Garden155Dummy equal to 1 if there is a garden156Story157Story on which the unit is located158Number of stories159Total number of the stories in the apartment unit160Stove161Dummy equal to 1 if heating system is a solid-fuel stove162Boilers163Dummy equal to 1 if heating system is powered by a wall hung gas boiler 164Central165Dummy equal to 1 if heating system is a common heating system for the entire block166Descriptive statistics167Mean168Median169Mode170Std. Dev.171Min.172Max.173Price174178437.27175138000176125000177120478.82178200001791350000180Age18110.11182618321849.76185018651187Floor area188141.6718912519010019161.3119242193550194Rooms1953.26196319731981.0019912007201Balcony2020.95203120412050.2020602071208Bathrooms2091.43210121112120.6421312144215Garden2160.95217121812190.2022002211222Story2232.48224222532262.67227–222825229Number of stories2305.11231423242332.96234223528236Stove2370.01238023902400.1124102421243Boilers2440.84245124612470.3624802491250Central2510.14252025302540.3525502561257258259===== PAGE 5 (score 15, numeric-paren 0, table-words 5, var-words 5) =====26026117262Housing market differentiation: the cases of Yenimahalle and Çankaya in Ankara263analysis in the form of dummy variables, indicat264ing the presence or absence of each characteristic 265in the dwelling unit. 266Following Bourassa et al. (1999), this study at267tempts to apply a set of techniques to investigate 268housing submarkets in a three-stage analysis. In 269the first stage of the study, a cluster analysis was 270carried out to form homogenous groups of differ271ent neighborhoods in Çankaya and Yenimahalle, 272which is a multivariate method that is used to 273classify data into a number of different groups on 274the basis of a set of measured variables (Cornish 2752007). The groups created in the cluster analysis 276are not unique, since they depend upon the vari277ables used and how cluster membership is defined 278(Chan 2005b). To analyze the different types of 279clusters in this study, an agglomerative hierarchi280cal method was chosen, within which are a num281ber of different methods used to determine which 282clusters should be merged at each stage (Cornish 2832007). This study adopts the between-groups link284age method with squared Euclidean Distance as 285the distance or similarity measure. In this method, 286the distance between two clusters is calculated as 287the average distance between all pairs of subjects 288in the two clusters (Cornish 2007). After deciding 289upon the most appropriate procedure and method 290for the analysis, the optimum number of clusters 291(k) in the sample needs to be defined. In this stage, 292a cluster analysis is applied, and the optimum 293number of clusters is chosen by examining the ag294glomeration schedule generated by the results, af295ter which the hierarchical cluster analysis is rerun 296with the selected number of clusters. 297In the second stage of the study, a discriminant 298analysis was used to determine the main charac299teristics of dwelling units that differed across the 300clusters identified in the first stage. A discriminant 301analysis (DA) is a statistical technique used for dif302ferentiating between groups when the independent 303variables are quantitative (Chan 2005a). If the in304dependent variable has more than two groups, a 305Multiple Discriminant Analysis (MDA) is applied 306rather than a simple Discriminant Analysis. An 307MDA comprises three different steps: 1) The esti308mation of the coefficients of the variables; 2) the 309calculation of the discriminant score of each case; 310and 3) the classification of each case, taking into 311account the discriminant scores (Leksrisakul, Ev312ans 2005). A discriminant function is conducted at 313the end of an MDA, which is similar to a regression 314function. The unstandardized discriminant coeffi315cient (analogous to the “b”s in the regression func316tion) maximizes the distance between the means of 317the dependent variable. That is, the discriminant 318function maximizes the distance between clusters. 319Good predictors tend to have large weights (Burns, 320Burns 2008). 321In the third stage of the study, hedonic price 322models were created to estimate the housing char323acteristics that define the price structure in each 324segment and the overall data. Such methods have 325proven to be an important means of analyzing com326modities with various characteristics (Goodman, 327Thibodeau 1998), and are useful tools for explain328ing the determinants of housing prices by taking 329into consideration structural, locational and envi330ronmental attributes. Hedonic models are impor331tant tools for understanding housing segmentation 332by focusing housing price changes (Keskin 2008). 333That said, owing to the lack of data on the loca334tional and environmental attributes of the dwell335ing units, the regression analysis was conducted 336focusing only on the structural characteristics of 337the dwelling units.3384. RESULTS3394.1. Cluster analysis340In order to define whether identifiable clusters of 341housing attributes exist on the basis of neighbor342hoods, a cluster analysis was conducted as a first 343step to determine whether certain combinations of 344physical attributes of dwelling units are common 345to certain neighborhoods. There are five variables 346in the data that are available for the classification 347of neighborhoods in the cluster analysis, informa348tion on which can be found in Table 2. The clus349ter analysis was run for 45 neighborhoods, and 350at the end of the examination of the agglomera351tion schedule generated by the results nine differ352ent clusters were identified. Table 3 presents the 353Table 2. Description of variables in cluster analysis354Variable355Description356A.Price357Average price of the dwelling units in the neighborhood358A.Age359Average age of the buildings in the neighborhood360A.Floor Area361Average floor area of the dwelling units in the neighborhood (m2)362A.Rooms363Average number of rooms of the dwellings in the neighborhood364A.Number of Stories365Average number of the stories in the apartment units in the neighborhood366367368===== PAGE 8 (score 35, numeric-paren 0, table-words 14, var-words 5) =====36937020371L. Alkan372Table 4 provides descriptive statistics for each 373cluster after fixing each of the 1,733 cases in the 374two districts to their particular cluster in the data. 375In Table, it can be seen that age may be a good 376discriminator for Cluster 1 and Cluster 6, while 377other clusters score relatively similar results in 378this category. Table 4 also illustrates that Clus379ter 5 and Cluster 9 have smaller floor areas, as380sociated with lower house prices; while Cluster 6 381has the most expensive average house prices and 382the largest floor areas. From this it can be safely 383assumed that floor area will constitute one of the 384most important variables affecting house prices. In 385the table, Cluster 1 has the oldest housing stock 386with the smallest number of stories, and Cluster 3876 has the newest housing units with the highest 388number of stories, pointing to a significant rela389tionship between age of the building and the num390ber of stories. Also, the table illustrates that Clus391ter 6 seems to be highly differentiated from the 392other clusters. To test these hypotheses and ex393amine the segmented housing market within these 394two districts, Multiple Discriminant and Hedonic 395Function analyses were conducted as the second 396and the third steps in the study.3974.2. Multiple discriminant analysis398A Multiple Discriminant Analysis (MDA) was used 399to investigate the differences between clusters, and 400to identify which attributes of housing contributed 401most to submarket separation. After excluding the 402variables with missing values, 1,139 cases were 403identified as suitable for analysis. The MDA was 404run using a range of key variables, including Price, 405Age, Floor Area, Rooms, and Number of Stories. 406The test of equality of the group means in Table 5 407indicates that there are significant group differ408ences in each of the independent variables (at p < 4090.000) in the study. 410Table 6 provides information on the Eigenval411ues, Canonical Correlations, Wilks’ Lambda and 412Coefficient Discriminant Functions. A total of five 413functions are produced in the discriminant analy414sis, and both the eigenvalues and canonical cor415relations are higher for the first function in the 416table. That is, the first function is statistically 417more significant than the others, and so the first 418function will be used for the estimation of the dis419criminant function. 420Wilks’ Lambda tests the statistical significance 421of discriminant functions, illustrating that the first 422three discriminant functions have significant dis423criminating power. In Table 6, the unstandardized 424coefficients are illustrated as canonical discrimi425nant function coefficients. The table also provides 426information on standardized canonical function co427efficients. The standardized discriminant function 428coefficients (b) indicate the partial contribution of 429each variable to the discriminate function, control430ling all other variables in the equation. They can 431be used to assess the contribution of each variable 432to the discriminate function, and therefore provide 433information on the relative importance of each var434iable (Burns, Burns 2008). 435The Price of the dwelling is the strongest pre436dictor in the allocation of housing market clusters, 437as its “b” is the highest for the first function; this is 438followed by Floor Area and Number of Stories. In 439the second function, the strongest predictor is the 440Number of Stories, while Age is highly relevant for 441the third function. Only the first three functions 442are interpreted in this part, in that they explain 443more of the variance between groups, while Func444tions 4 and 5 are not significant. The results of the 445discriminant analysis illustrate that Price, Floor 446Area, and Number of Stories are more significant 447variables in the formation of the segmented hous448ing structure in both Çankaya and Yenimahalle. 449The unstandardized coefficients in Table 6 are 450used to create the discriminant function as:451D = (0.014 × Age) + (–0.007 × Floor Area) + (–0.222 452× Rooms) + (0.016 × Price(,000TL)) + (–0.128 × 453Number of Stories) – 0.646454Table 7 presents the classification results of 455the discriminant analysis, in which the rows are 456the observed categories of the dependent and the 457columns are the predicted categories. When the 458prediction is perfect, all cases will lie on the diago459nal, meaning that the percentage of cases on the 460Table 5. Tests of equality of group means461Wilks’ lambda462F463df1464df2465Sig.466Age4670.9424688.713469847011404710.000472Floor area4730.91547413.299475847611404770.000478Rooms4790.9444808.429481848211404830.000484Price (,000 TL)4850.64348679.030487848811404890.000490Number of stories4910.86749221.770493849411404950.000496497498===== PAGE 9 (score 21, numeric-paren 0, table-words 5, var-words 4) =====49950021501Housing market differentiation: the cases of Yenimahalle and Çankaya in Ankara502Table 6. Eigenvalues, canonical correlations, Wilks’ lambda, coefficients discriminant functions503Function50415052506350745085509Eigenvalue5100.7485110.1565120.0625130.0095140.004515Canonical correlation5160.6545170.3675180.2415190.0945200.061521Tests of function(s)5221 thr 55232 thr 55243 thr 55254 thr 55265527Wilks’ lambda5280.4605290.8045300.9305310.9875320.996533Sig.5340.0005350.0005360.0005370.1525380.366539Standardized canonical discriminant function coefficients (b)540Age5410.131542–0.2415431.0355440.048545–0.110546Floor area547–0.406548–0.2305490.453550–0.9585511.667552Rooms553–0.218554–0.0635550.0315561.695557–0.604558Price (,000 TL)5591.443560–0.004561–0.239562–0.147563–0.510564Number of stories565–0.3695660.9365670.4455680.1785690.203570Canonical discriminant function coefficients571Age5720.014573–0.0255740.1085750.005576–0.011577Floor area578–0.007579–0.0045800.007581–0.0165820.028583Rooms584–0.222585–0.0655860.0315871.726588–0.615589Price (,000 TL)5900.0165910.000592–0.003593–0.002594–0.006595Number of stories596–0.1285970.3255980.1555990.0626000.070601(Constant)602–0.646603–0.666604–2.642605–3.527606–1.245607Table 7. Classification results608Clusters609Predicted group membership610Total611161226133614461556166617761886199620Original621Count6221623162426250626062706281629063016310632563326341563531636246376563812639116407641136424643182644364516646464719648346490650465146528653165490655465616571365843659183660636612662766313664296653546665667266815669267082671113672067336741675656762836776678067906802681068206831868446852686168727688768936903691169286933694569566962697669837699870037010702670327040705670627071170817093171097112712117132714257152071607171718271977720140721%722172320.072440.07250.07260.07270.072820.07290.073020.07310.0732100.073327348.273517.073613.273735.77386.67396.07403.87417.17422.2743100.0744374517.87464.474721.174837.87490.07504.47514.47528.97531.1754100.075547560.37573.775812.175951.776017.87610.67622.07633.77648.2765100.076657670.77685.37690.777029.077139.97720.07731.17740.477523.0776100.077767780.07790.07807.47810.07820.078366.778414.87857.47863.7787100.078877898.17908.17912.779221.67938.179413.579516.27965.479716.2798100.079988009.78010.080219.48036.58040.080519.48066.580735.58083.2809100.081098111.48127.98131.481417.981514.38160.08170.78181.481955.0820100.082139.9% of original grouped cases correctly classified.822diagonal can be considered as the percentage of 823correct classifications in the analysis. The overall 824predictive accuracy of the discriminant function 825is called the “hit ratio” (Burns, Burns 2008), and 826Table 7 thus reveals that 39.9% of the total cases 827are correctly classified. This hit ratio can be ac828cepted as being satisfactorily valid, in that most 829researchers accept a hit ratio of greater than 25%, 830which may be attributed to chance (Burns, Burns 8312008). In the table, Cluster 6 is classified with 832greater accuracy (66.7%) than the other clusters, 833which could be predicted from the cluster statis834tics in Table 4. 8354.3. Hedonic regression836After the cluster and discriminant analyses, a he837donic regression analysis was performed for each 838of the submarkets and the overall sample. The 839840841===== PAGE 10 (score 31, numeric-paren 0, table-words 1, var-words 6) =====84284322844L. Alkan845Table 8. Results of hedonic regression846ANOVA847Unstandardized 848coefficients849Standardized 850coefficients851Variables in model (significant at the 0.05 level)852F853Sig854R Square855Adjusted R square B856Std. Error857Beta858t859Sig.860VIF861Cluster 1 –862–863–864–865–866–867–868–869–870–871Cluster 2 Floor area87252.5428730.0008740.8048750.7898760.7698770.1548780.4978794.9928800.0008815.823882Number of stories88312.0818842.0028850.4178866.0358870.0008882.811889Cluster 3 Age 89015.9838910.0008920.7388930.6928941.5698950.7528960.1858972.0878980.0428991.525900Floor area9010.5889020.1949030.4829043.0309050.0049064.936907Story90816.2009095.9459100.4569112.7259120.0099135.467914Cluster 4 Floor area91558.3689160.0009170.6879180.6759190.6069200.1059210.4409225.7709230.0009244.939925Story9266.8969271.7139280.1839294.0259300.0009311.764932Number of stories9337.4919341.6689350.1939364.4929370.0009381.575939Cluster 5 Floor area94022.3879410.0009420.5119430.4889440.3799450.0929460.3529474.1159480.0009493.211950Rooms95113.7469524.1509530.2589543.3129550.0019562.659957Stove958–32.28595914.263960–0.112961–2.2649620.0259631.077964Story9653.6049661.1849670.1949683.0439690.0039701.785971Cluster 6 Floor area9726.1789730.0039740.8059750.6749761.4459770.5409780.6989792.6759800.0209814.179982Cluster 7 Floor area9835.3309840.0019850.6819860.5539871.9599880.6829891.0219902.8749910.0099927.911993Number of stories99425.8759959.2419960.6279972.8009980.0119993.1371000Cluster 8 Central10017.64710020.00110030.80510040.6991005224.649100664.84010070.82110083.46510090.00410103.7351011Bathrooms1012110.295101350.07910140.40310152.20210160.04610172.2281018Story101938.114102015.91310210.85010222.39510230.03210248.3731025Cluster 9 Floor area102621.79810270.00010280.65110290.62110301.58710310.20110320.91410337.88910340.00010354.0441036Rooms1037–25.21110389.5811039–0.2691040–2.63210410.01010423.1591043Overall1044(Constant)1045144.39410460.00010470.75110480.7461049–70.923105017.9871051–1052–3.94310530.0001054–1055Floor area10560.77710570.06310580.449105912.40110600.00010614.5251062Central106329.48110647.07810650.08510664.16510670.00010681.4251069Bathrooms107011.40010714.34310720.07210732.62510740.00910752.5931076Story10775.93810780.94310790.14910806.29710810.00010821.9291083Number of stories10844.49310850.87110860.12910875.15910880.00010892.1621090cls11091160.431109225.60710930.11110946.26510950.00010961.0941097cls2109854.32210997.41011000.17511017.33111020.00011031.9761104cls3110577.30911069.04911070.18111088.54311090.00011101.5601111cls4111232.48611136.41111140.13911155.06711160.00011172.6171118cls51119–14.03711206.4351121–0.0571122–2.18211230.02911242.3261125cls61126189.974112713.26711280.268112914.32011300.00011311.2111132cls7113351.056113411.54811350.08411364.42111370.00011381.2551139cls81140159.803114113.39811420.225114311.92711440.00011451.235114611471148===== PAGE 12 (score 25, numeric-paren 0, table-words 2, var-words 3) =====11491150241151L. Alkan1152have relatively newer housing stock, along with 1153Çamlıca in Yenimahalle. The table further shows 1154that there has been a strong shift towards larg1155er dwelling units, but not increases in building 1156heights in these neighborhoods in Çankaya; where1157as increases in the heights of buildings are more 1158remarkable in Çamlıca, although there has been no 1159significant increase in Floor Area. 1160Table 9. Residential characteristics of neighborhoods1161Cluster1162District1163Neighborhood1164Age1165Floor area1166Number of stories Price116711168Çankaya1169Anıttepe117024.401171160.0011723.801173317 000.00117421175Çankaya1176Ayrancı117726.821178128.1011795.131180199 188.121181Cevizlidere11824.941183171.8611844.361185211 616.281186Emek118714.831188132.4211893.641190227 833.331191Kırkkonaklar11926.601193184.2911943.811195210 455.881196Yenimahalle1197Çamlıca11984.501199134.42120010.891201217 263.16120231203Çankaya1204Bahçeli120529.821206140.7212075.411208265 250.001209Birlikmahallesi12106.931211188.4212124.101213264 188.171214Çankaya121512.711216144.5912176.141218248 676.471219Kavaklıdere122030.671221143.8112224.791223252 452.38122441225Çankaya1226Balgat12279.761228142.6812293.701230177 347.831231Dikmen12329.141233143.8312343.901235160 119.271236İlker12379.491238139.4712394.131240151 496.091241Keklikpınarı12424.731243149.0312444.431245167 898.571246Öveçler12478.981248144.1112494.201250168 910.961251Yenimahalle1252Çarşı12537.561254146.1712555.031256172 159.09125751258Çankaya1259Büyükesat126013.331261124.5512623.731263117 500.001264Cebeci126514.831266108.1312673.901268114 229.171269İncesu12702.141271105.6112724.841273108 173.911274Sağlık127518.001276104.9012774.511278116 979.591279Seyranbağları128011.931281112.4212824.031283114 930.231284Sokullu128516.971286108.9112874.101288113 320.451289Topraklık12903.501291101.6712923.251293109 666.671294Yenimahalle1295Demet129617.101297116.8012986.09129995 806.671300Ergenekon13019.791302125.9613033.311304104 000.001305Karşıyaka130616.351307111.8613086.71130983 071.431310Pamuklar13113.561312145.9813136.151314119 559.521315Susuz13165.251317135.9013189.60131999 700.001320Yakacık13212.081322137.2513235.00132489 166.671325Yunusemre132618.001327125.8313284.33132980 416.67133061331Çankaya1332Çukurambar13335.001334196.4413358.561336432 777.781337Yıldızevler13381.851339212.6413409.261341454 107.14134271343Çankaya1344Gaziosmanpaşa134513.881346167.1713474.981348299 091.671349Yenimahalle1350Çayyolu13516.551352152.2713539.831354284 204.55135581356Çankaya1357Hilal13584.471359206.2713604.211361371 409.091362Oran136315.301364176.3313658.131366370 333.331367Sancak136810.401369231.2913703.711371356 428.571372Yenimahalle1373Yaşamkent13742.601375185.6713769.801377358 333.33137891379Çankaya1380Küçükesat138122.901382116.8213834.191384139 214.291385Kızılay138630.501387100.0013884.291389144 714.291390Öncebeci13914.811392107.5013934.261394130 783.331395Yenimahalle1396Gazimahallesi139712.001398107.0013993.201400136 600.001401Kentkoop14026.011403132.1714048.251405142 765.221406Ostim14075.331408127.2014095.901410124 800.001411Ragıp Tüzün14128.361413136.0914143.641415145 727.271416All of the neighborhoods in Cluster 3 are in 1417Çankaya, where there has been a shift to larger 1418housing units in new settlements like Birlikma1419hallesi. Again, there has been no increase in the 1420number of stories in newer buildings, and the he1421donic analysis shows a positive relationship be1422tween the age of the building and the price of the 1423dwelling units for this cluster. The average prices 142414251426===== PAGE 13 (score 9, numeric-paren 0, table-words 3, var-words 3) =====14271428251429Housing market differentiation: the cases of Yenimahalle and Çankaya in Ankara1430in Bahçelievler and Kavaklıdere in particular are 1431high, despite the average age of the housing stock 1432in these locations, being the inner, older and more 1433prestigious areas of Ankara. 1434There is a more homogenous pattern in Clus1435ter 4 in terms of Age, Floor Area and Number of 1436Stories. As discussed previously, there is a posi1437tive relationship between Price and the Number 1438of Stories for this cluster, and Table 9 shows that 1439Çarşı in Yenimahalle contain relatively higher 1440apartment buildings. 1441In Cluster 5, there is no significant increase in 1442Floor Area or the Number of Stories in Topraklık 1443and İncesu, where the housing stock is newer than 1444in other parts of Çankaya. In Yenimahalle, the 1445newer housing stock in Pamuklar, Yakacık and 1446Susuz features relatively larger dwelling units, 1447and a greater Number of Stories in Susuz. 1448Cluster 6 contains the most expensive neigh1449borhoods in the Çankaya district – Çukurambar 1450and Yıldızevler. These neighborhoods have similar 1451patterns in terms of larger floor areas and higher 1452apartment units.1453Cluster 7 comprises Çayyolu in Yenimahalle 1454and Gaziosmanpaşa in Çankaya, which feature 1455similar mean prices. There is no significant differ1456ence in the average floor area of the dwelling units 1457in these two neighborhoods, although the housing 1458stock in Çayyolu is relatively newer than that of 1459Gaziosmanpaşa. That said, Çayyolu is a subur1460ban residential area of Ankara that was affected 1461by the rapid growth in high-rise apartment build1462ings, as discussed at the beginning of the study. 1463It is important to note that the hedonic analysis 1464highlights a positive relationship between Price 1465and the Number of Stories for this cluster. 1466As discussed previously, the results of the he1467donic analysis highlight no significant relation1468ship between house Prices and the Number of 1469Stories for Cluster 8. All of the neighborhoods 1470in Çankaya in Cluster 8 feature relatively larger 1471dwelling units; however, there has been a shift to 1472a lower Numbers of Stories in newer settlements 1473like Sancak and Hilal in contrast to Oran, where 1474the housing stock is older. There has also been a 1475growing trend of high-rise buildings in Yaşamkent 1476(another suburban settlement in Ankara) in Yeni1477mahalle.1478Table 9 illustrates that the Öncebeci neighbor1479hood has the newest housing stock in Çankaya in 1480Cluster 9, although there is no significant varia1481tion in Çankaya in terms of Floor Area and the 1482Number of Stories for this cluster. In addition, 1483there is again a trend towards higher apartment 1484buildings in the newer settlements of Yenimahalle 1485in this cluster.1486Table 9 highlights a shift towards larger dwell1487ing units within most clusters, consistent with the 1488results of the hedonic analysis; and the overall 1489trend in floor area is upwards, both in Çankaya 1490and Yenimahalle. In terms of number of stories 1491within clusters, there is no immediate change in 1492Çankaya; however, the data shows an important 1493increase in higher apartment buildings, especially 1494in the outer neighborhoods of Yenimahalle. That 1495is, the demand side of the housing market indi1496cated in the hedonic price analysis does not appear 1497to have had an observable impact on the supply 1498side of housing in terms of the number of stories. 14996. CONCLUSIONS1500This paper has investigated housing market dif1501ferentiation in Ankara through a case study of the 1502Yenimahalle and Çankaya districts by employing 1503a cluster analysis, a discriminant analysis and a 1504hedonic price analysis. The paper demonstrates 1505that the housing market in Çankaya and Yeni1506mahalle has a segmented structure, with location 1507being the main determining factor of different 1508house prices. The hedonic analysis revealed that 1509analyzing the price structure in the market area 1510as a whole prohibits the development of an un1511derstanding of the housing market, as the price 1512structures of different clusters have their own dy1513namics, consistent with the results of the study 1514by Alkay (2008). 1515The results of this study reveal the demand-1516side preferences of households in different neigh1517borhoods. Both Çankaya and Yenimahalle have 1518experienced a trend of larger housing units in 1519new residential areas, although there has been 1520no immediate change in Çankaya in terms of the 1521Number of Stories. In contrast, the data shows 1522a converse increase in higher apartments, espe1523cially in the suburban neighborhoods of Yenima1524halle. This trend in growth areas towards higher 1525apartment buildings is inconsistent with the hous1526ing choice of households, as the results of the he1527donic analysis illustrate, highlighting an invisible 1528mechanism that shapes the residential pattern of 1529growth areas in Ankara. The results of the study 1530can be seen as a useful tool for understanding 1531the demand side of the housing market in Çan1532kaya and Yenimahalle, and further studies may 1533be conducted to cover all urban residential areas 1534in Ankara in order to characterize the residential 1535pattern for the city as a whole. 1536