1# RECORD 0f2b62072edfd32a439d1606541bbbb8547810e42# TITLE: Willingness to pay for accessibility under the conditions of residential segregation3# YEAR: 2014.0 VENUE: nan DOI: 10.3846/1648715x.2013.8643424# AUTHORS: Marko Kryvobokov; L. Bouzouina5# PAGES: 15 CANDIDATE TABLE PAGES (1-based): [6, 7, 8, 10, 14, 15]6# ABSTRACT: The hypothesis that different income groups have different willingness to pay for accessibility to the city centre is based on the standard monocentric model. This hypothesis is empirically tested with accessibility attributes in a hedonic model of apartment prices in the suburbs of the city of Lyon, France. The conditions of residential segregation are described, and apartment prices in the poor and the rich suburbs are analysed with regression techniques. Travel times to two urban centres are accounted for, as well as centrality and accessibility integral indexes. We found that in the selected areas the hypothesis is true. Spatial differences between the estimates for accessibility measures are significant. In more socially problematic areas, the willingness to pay for better accessibility is higher.789===== PAGE 1 (score 13, numeric-paren 1, table-words 0, var-words 12) =====1011Copyright © 2014 Vilnius Gediminas Technical University (VGTU) Press Technika12http://www.tandfonline.com/TSPM13INTERNATIONAL JOURNAL OF STRATEGIC PROPERTY MANAGEMENT14ISSN 1648-715X print / ISSN 1648-9179 online152014 Volume 18(2): 101–11516doi:10.3846/1648715X.2013.86434217* Corresponding author. E-mail: k_mark@ukr.net18or mark.kryvobokov@let.ish-lyon.cnrs.fr19WILLINGNESS TO PAY FOR ACCESSIBILITY UNDER 20THE CONDITIONS OF RESIDENTIAL SEGREGATION21Marko KRYVOBOKOV a,*, Louafi BOUZOUINA 22a Laboratory of Transport Economics (LET), 14, avenue Berthelot, F-69363 Lyon Cedex 07, France 23b Laboratory of Transport Economics (LET), ENTPE, 3, rue Maurice Audin, F-69518 Vaulx-en-Velin, France24Received 29 September 2012; accepted 16 January 201325ABSTRACT. The hypothesis that different income groups have different willingness to pay for acces-26sibility to the city centre is based on the standard monocentric model. This hypothesis is empirically 27tested with accessibility attributes in a hedonic model of apartment prices in the suburbs of the city 28of Lyon, France. The conditions of residential segregation are described, and apartment prices in the 29poor and the rich suburbs are analysed with regression techniques. Travel times to two urban centres 30are accounted for, as well as centrality and accessibility integral indexes. We found that in the selected 31areas the hypothesis is true. Spatial differences between the estimates for accessibility measures are 32significant. In more socially problematic areas, the willingness to pay for better accessibility is higher.33KEYWORDS: Accessibility; Bid-rent function; Hedonic regression; Apartment price; Residential seg-34regation35REFERENCE to this paper should be made as follows: Kryvobokov, M.; Bouzouina, L. 2014. Willing-36ness to pay for accessibility under the conditions of residential segregation, International Journal of 37Strategic Property Management 18(2): 101–115.381. INTRODUCTION39Economic literature about urban segregation ex-40perienced rapid development since the first work 41of Kain (1968) on the difficulties of employment 42access for the inhabitants of American ghettos. 43Numerous case studies dealing with the question 44of neighbourhood effect test the impact of social 45polarisation or accessibility on unemployment, iso-46lation, crime or school results in problematic dis-47tricts of different cities and countries (Case, Katz 481991; Hoxby 2000; Bayer et al. 2008; Galster et al. 492008). 50In France, the economic studies of segregation 51are more recent and mainly dated by the years 522000s (Fitoussi, Savidan 2003; Maurin 2004; 53Buisson, Mignot 2005). With several exceptions 54analysing geographic and ethnic origins (Pan Ké 55Shon 2010), this literature focuses on the socio-56economic dimensions to demonstrate the increas-57ing inequalities between rich and poor neighbour-58hoods and to measure their negative consequences 59on the economic, human and social capital of poor 60inhabitants, whose significant share is suffered 61from high unemployment (Bouabdallah et al. 2002; 62Choffel, Delattre 2003; Gaschet, Gaussier 2005; 63Gobillon, Selod 2004; Korsu, Wenglenski 2010). 64Most of these studies confirm the hypothesis that 65the unemployment of poor residents is partially 66caused by poor accessibility to jobs (especially to 67low-skilled jobs) and high poverty rates. The two 68main reasons of poor employment accessibility in 69the Paris region are location in remote neighbour-70hoods with few jobs and low car-ownership rate 71(Korsu, Wenglenski 2010). In their analysis of 72the Lyon, Marseille, and Lille urban agglomera-73tions, Mignot et al. (2009) highlight the point that 74though some recent studies report little inequal-75ity in daily mobility since households have access 76to cars, at the same time inequality increases for 77those without car. 78The growing interest of academic research to 79residential segregation in France can be explained 80by a complicated reality including violence and 81b828384===== PAGE 6 (score 35, numeric-paren 0, table-words 7, var-words 19) =====858610687M. Kryvobokov, L. Bouzouina88The described four suburbs are the examples of 89the uneven distribution of social housing stock. In 90the western suburbs, this percentage is lower than 910.20, especially in Sainte-Foy-lès-Lyons, while in 92the eastern banlieues half of population lives in 93social housing (Table 1). 94Income disparities and social contrasts between 95the selected suburbs clearly demonstrate that in 96terms of attractiveness and prestige, “bourgeois” 97western areas are far surpassing their eastern 98counterparts. Moreover, both Vaulx-en-Velin and 99Vénissieux have an image of well-known high-100poverty banlieues. The following quotation from 101Korsu and Wenglenski (2010) about the public 102view of problematic areas in general can be ex-103ploited to describe both: “many people associate 104such neighbourhoods with violence, criminality, 105unwillingness to work, alcohol and drug addiction, 106etc.” (ibid: 2285).107In respect to public transport, the socialist gov-108ernance of the Lyon agglomeration tries to favour 109eastern suburbs. The most expensive investments 110in public transport, such as the extensions of met-111ro and tramway lines, take place to the east from 112Lyon, including Vénissieux and Vaulx-en-Velin, 113while in the western suburbs these types of pub-114lic transports do not exist. In the eyes of many 115Lyonnais, the relative underdevelopment of public 116transport in western suburbs is to same degree 117a deliberate policy supporting their higher stand-118ing. A good illustration of the Not-In-My-BackYard 119principle of the wealthy population is their protest 120(though unsuccessful) in the 1990s against prolon-121gation of the metro line directly linking Vénissieux 122with north-western Lyon. 123We use the data on about 10,000 apartment 124sales selected randomly from all sales in the cen-125tral part of the Lyon metropolitan area in the 126period 1997–2008. These data on sale prices and 127apartment attributes were provided by Perval, who 128collects information about real estate transactions 129in France. After deleting observations with incom-130plete data, 4,362 apartments remained. The defini-131tion of variables and descriptive statistics are given 132in Table 2. It contains information about transac-133tions, as well as on apartment attributes and loca-134tion attributes. It does not include the number of 135rooms, because the cardinal discrete variables for 136them are highly correlated with apartment area 137and are not significant if included in the hedonic 138model. Many observations contain no data about 139the number of parking places, number of cellars, 140and quality of view; therefore the specific dummy 141variables are created. In the sample, 28% of the 142apartments were new, sold in a primary market. 143The location variables in Table 2 include the 144percentages of high-, middle-, and low-income 145households, and dummies for location in one of 146four ad hoc districts. The four ad hoc districts are 147quite large, but relatively homogenous territories, 148divided by water frontiers and the boundaries of 149the urbanised area. District 1 includes the most 150central part of Lyon between the Rhône and the 151Saône. District 2 is an urbanised area on the left 152bank of the Rhône. District 3 is an urbanised area 153on the right bank of the Saône. District 4 is the 154less urbanised territory. 155Table 2 contains the following accessibility 156measures: travel times by car, travel times by pub-157lic transport, tertiary employment centrality index, 158and tertiary employment accessibility index. The 159origin-destination (O-D) matrix of travel times is 160estimated for morning peak with the MOSART 161transportation model developed with the VISUM 162software (Bonnafous et al. 2011). It is a four-step 163model, which forecasts transportation demand. We 164account for the travel times to the Lyon City Hall 165and to Bellecour, which is the main public transport 166junction located in the central part of Lyon. Both 167City Hall and Bellecour are among the fifteen (sub)168centres identified with a regression of service em-169ployment density on travel time (Kryvobokov 2010), 170Table 1. Socio-demographic description of the four suburbs171Attribute172Sainte-Foy-lès-Lyons173Oullins174Vénissieux175Vaulx-en-Velin176Population17722,61617826,42617957,56618040,241181Households1829,20818312,03218421,96518514,293186Persons in household1872.461882.201892.621902.82191Cars to households ratio1921.251931.011940.951950.87196Cars to population ratio1970.511980.451990.372000.31201Percentage of high-income households2020.172030.142040.112050.10206Percentage of low-income households2070.212080.252090.332100.37211Percentage unemployed2120.072130.092140.182150.21216Percentage of households in social housing2170.072180.152190.502200.51221222223===== PAGE 7 (score 39, numeric-paren 0, table-words 4, var-words 22) =====224225107226Willingness to pay for accessibility under the conditions of residential segregation227similarly to McDonald and Prather (1994). The cen-228trality and accessibility indexes are integral meas-229ures, which take account of tertiary employment. 230The former is gravity-based and the latter is a “sub-231jective” measure based on satisfaction thresholds 232and fuzzy logic, for details see Appendix 1. Note 233that in Table 2 the geographical unit of analysis is 234not municipality, but smaller zone; that is why the 235percentages of income groups are lower and higher 236in comparison with Table 1, which represents mu-237nicipalities. The zones are French statistical units 238IRISes (Îlots Regroupés pour l’Information Statis-239tique) close to the tracts in the US.240Of this sample, we select four sub-samples from 241the analysed municipalities (Table 3). The average 242floor area is the highest in Sainte-Foy-lès-Lyons. 243Though this attribute is the lowest in Oullins, it 244becomes the second highest there if calculated 245per person using the average number of persons 246in household in municipality. According to the 247average price per square meter, Sainte-Foy-lès-248Lyons is again the leader. The minimum prices 249in the western suburbs are much higher than in 250the eastern ones, but relatively high average price 251and very high maximum in Vénissieux indicates 252a wide diversity of its housing stock. According to 253the t-test, Vaulx-en-Velin and Vénissieux, unlike 254western suburbs, have equal means for some at-255tributes, e.g. the percentage of low-income house-256holds.257Table 2. Descriptive statistics of apartment attributes258Variable259Description260Mean261Minimum262Maximum263Std. deviation264Price265Transaction price, Euro266122,507.5026710,9652681,120,00026977,200270Year97-Year08271Dummies for year of transaction2720.04-0.13273027412750.20-0.34276Area277Apartment area, square metres27868.94279828030128128.12282Bath1-Bath3283Dummies for number of bathrooms284<0.01–0.93285028612870.05–0.25288ParkUn289Dummy for cases with no data 290about parking places2910.26292029312940.44295Park0-Park3296Dummies for number of parking places297<0.01–0.49298029913000.06–0.50301FloorGr302Dummy for ground floor3030.13304030513060.33307Floor1308Dummy for storey 13090.18310031113120.39313Floor2_4314Dummy for storey 2 to 43150.49316031713180.50319Floor5_8320Dummy for storey 5 to 83210.18322032313240.38325Floor9+326Dummy for storey 9 or more3270.02328032913300.14331Constr<1850-332Constr1992<333Dummies for period of construction3340.03–0.34335033613370.17–0.47338CondGood339Dummy for good state3400.81341034213430.39344CondMed345Dummy for state when some 346maintenance is needed3470.16348034913500.37351CondBad352Dummy for state when renovation 353is needed3540.03355035613570.17358ViewNo359Dummy for cases with no data 360about view3610.60362036313640.49365ViewGood366Dummy for view increasing value3670.38368036913700.48371ViewBad372Dummy for view decreasing value3730.02374037513760.13377Cellar0-Cellar2378Dummies for number of cellars3790.02–0.66380038113820.13–0.47383Garden384Dummy for existence of garden3850.05386038713880.22389Terrace390Dummy for existence of terrace3910.09392039313940.29395NewApartment396Dummy for new apartment3970.28398039914000.45401%HighIncome402Percentage of high-income households40312.554044.3440528.774062.91407%MidIncome408Percentage of middle-income households40957.9641042.7041166.204123.30413%LowIncome414Percentage of low-income households41529.4941610.2441752.124185.78419District1-District4420Dummies for location in districts4210.01–0.64422042314240.12–0.48425TTBellecour, 426TTCityHall 427Travel time to Bellecour, travel time 428to City Hall in minutes42911.00–21.734300.5543129.32–51.024324.99–11.16433Centrality Index434Centrality index43537.04–42.1343612.15–13.45437100.0043815.68–17.23439Accessibility Index440Accessibility index44163.07–64.704420.34–0.65443100.0044425.63–27.47445446447===== PAGE 8 (score 26, numeric-paren 0, table-words 6, var-words 11) =====448449108450M. Kryvobokov, L. Bouzouina451Table 3 gives the average values of accessibility 452measures in the four municipalities. The eastern 453suburbs are located in general a bit farther from 454the two urban centres, but their indexes for public 455transport are higher at the expense of better devel-456oped network. The number of zones in Sainte-Foy-457lès-Lyons, Oullins, Vaulx-en-Velin and Vénissieux 458are 10, 11, 18 and 24 respectively (see Fig. 1).459Besides using the sample of 4,362 observations, 460we extract the 20% most expensive apartments as 461well as the 20% cheapest apartments assuming 462that their buyers belong rather to the rich and 463the poor households respectively. These sub-sam-464ples called hereafter “expensive” and “cheap” are 465used in further analyses as proxies for the non-466geographical submarkets for the rich and the poor, 467because it is difficult to find information about 468the income of the buyers of particular apartments 469in quantity sufficient to create a mass valuation 470model. For these two sub-samples the equality of 471means hypothesis is rejected for price, floor area, 472income levels and all accessibility measures. We 473admit that this division does not correspond to 474the rich and the poor household groups described 475above. 476Furthermore, this sampling can be biased as 477some share of expensive apartments could be ex-478pensive just because of their large floor area, but 479cheap in respect to area unit. Some observations in 480the cheap segment, on the contrary, can be cheap 481due to small floor area, but have high price per 482area unit. In the context of the study, however, 483it is logical to assume that the poor are not able 484to buy expensive apartments even if their prices 485per area unit are low; while the rich prefer rather 486bigger accommodations. Therefore we suppose that 487our sampling strategy is less biased than it could 488be if it was based on price per square meter. 489Correlation of apartment price per square me-490ter (deflated to the initial year 1997) with acces-491sibility measures are in line with our hypothesis: 492prices of more expensive apartments are less cor-493related with accessibility. If to consider the sales 494in Sainte-Foy-lès-Lyons and Vénissieux, their cor-495relations behave similarly to the expensive and 496cheap sub-samples respectively. In Oullins and 497Vaulx-en-Velin, the sign of correlation coefficient 498is unstable. To understand the nature of these 499anomalies, econometric analysis is needed. 5004.2. Hedonic regression501Applying the hedonic regression model of price as 502the dependent variable with willingness to pay for 503different attributes as independent variables, we 504continue the tradition proposed by Rosen (1974). If 505the supply of each characteristic is perfectly elas-506tic, hedonic coefficients reveal demand for charac-507teristics, though in most real-world contexts such 508a stringent maintained hypothesis is untenable 509(Malpezzi 2003). Hedonic price model is widely 510used in real estate research. The recent examples 511of its application to housing prices in France are 512Table 3. Apartment sales and accessibility in the four suburbs513Attribute514Sainte-Foy-lès-Lyons515Oullins516Vénissieux517Vaulx-en-Velin518Apartment sales519Number of sales520845211005229752354524Average floor area, m252579526635276752865529Average floor area per average 530number of persons, m253132.153228.653325.653422.9535Average price per m2, Euro5361,6115371,4955381,4645391,171540Minimum price per m2, Euro541675542393543226544307545Maximum price per m2, Euro5463,5485473,4425483,6745492,631550Accessibility by car551TTBellecour, minutes55212.1755310.6255414.4555521.13556TTCityHall, minutes55714.6755814.4655917.6156017.93561Centrality Index56223.8256325.6156426.0256523.73566Accessibility Index56728.2056833.9456930.7757023.10571Accessibility by public transport572TTBellecour, minutes57328.6157429.4257529.7357635.41577TTCityHall, minutes57831.4857933.3358035.0858132.69582Centrality Index58320.8458420.0358523.1358621.74587Accessibility Index58831.6558927.4459038.7259135.48592593594===== PAGE 9 (score 33, numeric-paren 0, table-words 9, var-words 15) =====595596109597Willingness to pay for accessibility under the conditions of residential segregation598Gouriéroux and Lafarrère (2009) and Baltagi and 599Bresson (2011).600There is no strong theoretical basis for substan-601tiating the correct functional form of a hedonic re-602gression (Halvorsen, Pollakowski 1981; Malpezzi 6032003), but empirical tests found that the log-linear 604form has a number of advantages over the linear one 605(see e.g. Follain, Malpezzi 1980). More general and 606flexible form was proposed by Box and Cox (1964), 607but this function is beyond our current study. 608In our specification, the dependent variable is 609the logarithm of Price. Among the independent 610variables described in Table 2, the logarithmic 611transformations of Area, %MidIncome, %High-612Income and travel times are used. The following 613attributes are the default values: Year97, Bath1, 614Park0, FloorGr, Constr1981_1991, CondGood, 615ViewGood, Cellar0 and District1. 616First, an overall Ordinary Least Squares (OLS) 617model is built with 4,362 observations. Separate 618OLS models are created for the expensive and 619cheap apartment price segments; according to the 620t-test of equality of means of regressors, the null 621hypothesis is rejected for the majority of variables, 622including accessibility attributes and apartment 623floor area.624The observations from each of the four selected 625suburbs are exploited with separate OLS and Spa-626tial Error models. In the latter, the error term is 627a function of the residuals in neighbouring areas. 628The geographically weighted regression method-629ology (GWR) is applied to all available sales to 630solve a regression equation in each observation 631point. For the OLS, a weighted standard error 632test (Schnare, Struyk 1976) shows that the four 633submarkets significantly reduce standard error 634in comparison with the pool model. The t-tests of 635equality of regression coefficients in the four sub-636markets give the following results: for accessibility 637measures, the null hypothesis is rejected in most 638cases, while for some other attributes, e.g. floor 639area, the coefficients are rather stable. 640Each model is run eight times to examine the 641willingness to pay for each of the accessibility at-642tributes in question, i.e. travel times to two urban 643centres and centrality and accessibility indexes, 644while each attribute is measured twice: for car and 645for public transport. Our hypothesis is that the 646richer population evaluates travel time to city cen-647tres differently from the poor. We examine as well 648if this is true in respect to the integral indexes. 649Under the conditions mentioned in Section 2, the 650absolute value of the estimate for the accessibility 651in the hedonic price model should be lower for the 652rich than for the poor. 653Table 4 represents the estimated hedonic coeffi-654cients as well as model fit for OLS and GWR mod-655els. For the OLS models, the maximum variance 656inflationary factor (VIF) is also reported. All the 657significant coefficients of the models with TTBelle-658cour by car are presented in Appendix 2, where the 659estimates of the expensive and cheap sub-samples 660and median GWR results are reported as well. 661The three OLS models support our hypothesis. 662In the expensive segment, the absolute values of 663the estimates are always lower (often more than 664twice) than those in the overall model. In the cheap 665segment, the absolute values of the accessibility 666coefficients are always higher than those in the 667expensive segment. The negative effect of increase 668in travel time to Bellecour by public transport by 6691% leads to price reduction in the cheap segment 670by 0.17%, which is 2.8 times more than the price 671reduction in the expensive segment. 672The four municipalities can also be analysed 673as submarkets, for which individual models are 674created. The same variables except construction 675years and some structural dummies are used in 676the OLS and Spatial Error (with a threshold dis-677tance weighting scheme) models for the submar-678kets. The Lagrange multiplier (LM) tests, applied 679for the OLS models, indicate high significance level 680of spatial errors in each of the four suburbs. The 681LM tests of the spatial lags are significant in Oul-682lins and Vaulx-en-Velin, but due to difficulty in 683estimates interpretation, Spatial Lag model is not 684applied in this study. The estimates for the ac-685cessibility measures are presented in Appendix 3. 686However, this approach is rather useful to study 687the willingness to pay for accessibility measures 688within submarkets than the differences between 689them. Indeed, the estimates are insignificant and 690have “wrong” signs in Oullins and Vaulx-en-Velin, 691where the observations are less dispersed spatially 692than in Sainte-Foy-lès-Lyons and Vénissieux. The 693metro station in Vaulx-en-Velin opened in Octo-694ber 2007, by the end of the study period, and this 695fact can contribute to the “wrong” signs of acces-696sibility variables. The effect of the new station on 697the price dynamics is however beyond the current 698study limits. In Sainte-Foy-lès-Lyons, the coeffi-699cients have “right” signs, but they are insignifi-700cant in most cases. The most intuitive results in 701Vénissieux with its dispersed observations are ob-702tained with the OLS technique, but when spatial 703effect is accounted for, the accessibility measures 704calculated for car lose significance. Thus, within 705Vénissieux public transport has more impact on 706housing prices. 707708709===== PAGE 10 (score 85, numeric-paren 41, table-words 13, var-words 4) =====710711110712M. Kryvobokov, L. Bouzouina713Concerning the geographically weighted results 714in Table 4, the average estimates among all the 715observations are reported as well as the four ex-716tractions from the examined suburbs taken from 717the overall GWR models. The absolute values of 718the average GWR estimates, which are a bit lower 719than the OLS results, increase from municipality 720to municipality in the sequence presented in Table 7214. The only exception is that for the accessibility 722index for car in Oullins, whose coefficient is signifi-723cant for only 56% of observations. In the eastern 724suburbs, the percentage of significant cases is al-725ways higher than 0.90. Vaulx-en-Velin, where this 726percentage is always equal to unity, demonstrates 727the biggest difference from Sainte-Foy-lès-Lyons. 728The ratio of their coefficients is 5.7 for the travel 729time to Bellecour by car. The absolute values of the 730estimates for the western suburbs are, as a rule, 731lower than the GWR average, though Oullins is 732an exception in the three cases of public transport 733(travel times to Bellecour, City Hall and centrality 734index), which can be explained by relative under-735development of public transport at the time, for 736which the data have been collected. In the eastern 737suburbs, the absolute values of the accessibility es-738timates are almost always higher than the overall 739GWR averages. 7405. CONCLUSIONS741The monocentric-based hypothesis that willingness 742to pay for better accessibility for the poor is higher 743than for the rich is empirically supported with a 744hedonic model of apartment prices. Monocentric 745models with travel times to two versions of city 746centre are supplemented with integral centrality 747and accessibility indexes. The results of hedonic 748models with different accessibility measures are 749quite similar. First, the hypothesis is supported 750with the cheap and expensive apartment segments. 751Table 4. Estimates for the accessibility measures 752Model (OLS), 753extraction (GWR)754Accessibility by car755Accessibility by public transport756TTBellecour757TTCityHall758Centrality 759Index760Accessibility 761Index762TTBellecour763TTCityHall764Centrality 765Index766Accessibility 767Index768OLS*769Overall,7704,362 obs., 771adj. R2 = 0.84 – 0.85, 772Max VIF = 6.27773–0.224774(<0.001)775–0.278776(<0.001)7770.0085778(<0.001)7790.0051780(<0.001)781–0.223782(<0.001)783–0.228784(<0.001)7850.0078786(<0.001)7870.0049788(<0.001)789Expensive,790873 obs., 791adj. R2 = 0.84 – 0.86, 792Max VIF = 8.02793–0.093794(<0.001)795–0.129796(<0.001)7970.0042798(<0.001)7990.0024800(<0.001)801–0.062802(<0.001)803–0.083804(<0.001)8050.0028806(<0.001)8070.0011808(0.017)809Cheap,810872 obs., 811adj. R2 = 0.70–0.72, 812Max VIF = 9.18813–0.176814(<0.001)815–0.217816(<0.001)8170.0069818(<0.001)8190.0043820(<0.001)821–0.173822(<0.001)823–0.157824(<0.001)8250.0060826(<0.001)8270.0038828(<0.001)829GWR**830Overall average, 8314,362 obs.,832R2 = 0.87 – 0.88833–0.175834(0.96)835–0.235836(0.97)8370.0071838(0.96)8390.0049840(0.95)841–0.151842(0.93)843–0.159844(0.97)8450.0052846(0.97)8470.0040848(0.91)849Sainte-Foy-lès-Lyons,85084 obs.851–0.106852(0.95)853–0.169854(1.00)8550.0061856(1.00)8570.0025858(0.87)859–0.104860(1.00)861–0.127862(1.00)8630.0044864(1.00)8650.0022866(0.99)867Oullins,868100 obs.869–0.126870(0.73)871–0.173872(0.83)8730.0066874(0.99)8750.0018876(0.56)877–0.202878(1.00)8790.192880(1.00)8810.0088882(1.00)8830.0033884(1.00)885Vénissieux,88697 obs.887–0.169888(0.90)889–0.307890(0.93)8910.0105892(1.00)8930.0049894(1.00)895–0.322896(1.00)897–0.306898(1.00)8990.0125900(1.00)9010.0060902(1.00)903Vaulx-en-Velin,90454 obs.905–0.599906(1.00)907–0.505908(1.00)9090.0133910(1.00)9110.0070912(1.00)913–0.520914(1.00)915–0.398916(1.00)9170.0141918(1.00)9190.0080920(1.00)921Notes: * – significance level is in the parentheses; ** – percentage of cases significant at the 5% level is in the paren-922theses.923924925===== PAGE 11 (score 15, numeric-paren 0, table-words 1, var-words 13) =====926927111928Willingness to pay for accessibility under the conditions of residential segregation929Second, the support is shown with the example of 930the suburbs representing the patterns of residen-931tial segregation: the inhabitants of poor suburbs 932are willing to pay several times more for the acces-933sibility components of apartment price. Moreover, 934spatial differences in perception of accessibility are 935much higher than those between the expensive 936and cheap non-geographical apartment segments.937The spatial dispersion of observations in the 938examined four sub-samples is important, when 939the attributes within sub-sample are examined. 940Within one poor suburb, accessibility measures are 941found significant for public transport, but not for 942car. The GWR methodology allows comparing the 943effect between sub-samples in the overall model. 944In the selected eastern Lyon suburbs we have 945not found, in general, poorer accessibility, espe-946cially when public transport is taken into account. 947What is really different is the willingness to pay 948for accessibility: the more socially problematic is 949the area, the higher is the price for better acces-950sibility. This can be connected with lower car-own-951ership rate and other problems of disadvantaged 952neighbourhoods. From this viewpoint, the prior-953ity of public transport development in the east-954ern suburbs of Lyon looks reasonable as here its 955positive impact is more significantly capitalised in 956housing prices. Thus, investment in transport in-957frastructure in poor areas, which decreases travel 958time and simultaneously increases housing value, 959can be regarded in terms of economic efficiency 960and not only in those of social justice. 961The standard theoretical model presented in 962the paper does not include location amenities 963other than the proximity to the CBD. Some em-964pirical extension is possible with integral meas-965ures of centrality and accessibility considering 966service employment. In future study, more atten-967tion should be paid to other location amenities and 968their place in the trade-off with accessibility in the 969utility functions of household location choice of so-970cial groups. The techniques of Location Value Re-971sponse Surface and Market Basket Value (McClus-972key et al. 2000; Borst, McCluskey 2007; D’Amato 9732010) can be potentially useful to identify value 974influence centres and to be applied as a stage of 975market segmentation.976ACKNOWLEDGEMENTS977This study is part of project PLAINSUDD spon-978sored through French ANR (number ANR-08-979VD-00). The authors are grateful to Perval and 980Pierre-Yves Péguy for providing data on apart-981ment prices and attributes and to Nicolas Ovtracht 982for calculating the coordinates of apartment blocks 983and travel times with MOSART. The paper has 984been benefited from the discussions with Alain Bon-985nafous, Yves Crozet and Charles Raux as well as 986from valuable comments of anonymous reviewers.987REFERENCES988Alonso, W. 1964. Location and land use. Cambridge, 989MA: Harvard University Press.990Anas, A.; Arnott, R.; Small, K. A. 1998. Urban spatial 991structure, Journal of Economic Literature 36(3): 9921426–1464.993Baltagi, B. H.; Bresson, G. 2011. Maximum likelihood 994estimation and Lagrange multiplier tests for panel 995seemingly unrelated regressions with spatial lag 996and spatial errors: an application to hedonic housing 997prices in Paris, Journal of Urban Economics 69(1): 99824–42. http://dx.doi.org/10.1016/j.jue.2010.08.007 999Bayer, P.; Ross, S. L.; Topa, G. 2008. Place of work and 1000place of residence: informal hiring networks and la-1001bor market outcomes, Journal of Political Economy 1002116(6): 1150–1196. http://dx.doi.org/10.1086/595975 1003Bonnafous, A.; Crozet, Y.; Mercier, A.; Ovtracht, N.; 1004Péguy, P. -Y.; Puech, F. 2011. MOSART et le projet 1005PLAINSUDD: une plate-forme de modélisation et 1006simulation de l’accessibilité pour l’aide à la décision 1007et l’aménagement du territoire, in J.-P. Antoni (Ed.). 1008Modéliser la ville: formes urbaines et politiques de 1009transport. Paris: Economica, 186–210. 1010Borst, R.; McCluskey, W. 2007. Comparative evaluation 1011of the comparable sales method with geostatistical 1012valuation models, Pacific Rim Property Research 1013Journal 13(1): 106–130.1014Bouabdallah, K.; Cavaco, S.; Lesueur, J.-Y. 2002. Re-1015cherche d’emploi, contraintes spatiales et durée de 1016chômage: une analyse microéconométrique, Revue 1017d’Economie Politique 1: 137–157.1018Bouzouina, L. 2008. Ségrégation spatial et dynamique 1019métropolitaines. Thèse de Doctorat en Sciences Eco-1020nomique. Université Lumière Lyon 2.1021Bouzouina, L. 2009 Densité et ségrégation socio-spatiale 1022au sein des villes françaises, in F. Gaschet, C. La-1023cour (Eds.). Métropolisation et ségrégation. Bordeaux: 1024Presses Universitaires de Bordeaux, 120–134,1025Box, G. E. P.; Cox, D. 1964. An analysis of transforma-1026tions, Journal of the Royal Statistical Society, Series 1027B-Statistical Methodology 26(2): 211–252. 1028Buisson, M.-A.; Mignot, D. 2005 Concentration écono-1029mique et ségrégation spatiale. Bruxelles: De Boeck.1030Case, A. C.; Katz, L. F. 1991 The company you keep: the 1031effect of family and neighborhood on disadvantaged 1032youths, National Bureau Economic Research Work-1033ing Papers, 3705.1034Choffel, P.; Delattre, E. 2003. Habiter un quartier défa-1035vorisé: quels effets sur la durée de chômage?, Pre-1036mières synthèses 43.1, Dares.1037Goffette-Nagot, F.; Thomas, I.; Zenou, Y. 2000. Struc-1038ture urbaine et revenus des ménages, in C. Baumont, 1039P. Combes, P.-H. Derycke; H. Jayet (Eds.). Economie 1040géographique: approches théoriques et empiriques. 1041Paris: Economica, 276–302.104210431044===== PAGE 14 (score 97, numeric-paren 58, table-words 2, var-words 7) =====104510461141047M. Kryvobokov, L. Bouzouina1048APPENDIX 2 1049OLS and GWR models with TTBellecour by car1050Variable,1051parameter1052OLS1053GWR1054Overall1055Expensive1056Cheap1057Constant10584.552 (<0.001)10597.186 (<0.001)10606.947 (<0.001)10614.2551062TTBellecour1063–0.224 (<0.001)1064–0.093 (<0.001)1065–0.176 (<0.001)1066–0.1361067Year9910680.092 (0.001)10690.090 (0.02)1070–0.087 (0.26)10710.0621072Year0010730.162 (<0.001)10740.157 (<0.001)1075–0.029 (0.71)10760.1361077Year0110780.249 (<0.001)10790.281 (<0.001)10800.204 (0.01)10810.2201082Year0210830.310 (<0.001)10840.330 (<0.001)10850.223 (0.004)10860.2791087Year0310880.474 (<0.001)10890.485 (<0.001)10900.402 (<0.001)10910.4511092Year0410930.633 (<0.001)10940.589 (<0.001)10950.590 (<0.001)10960.6111097Year0510980.801 (<0.001)10990.748 (<0.001)11000.749 (<0.001)11010.7641102Year0611030.933 (<0.001)1104.801 (<0.001)11050.912 (<0.001)11060.8971107Year0711081.011 (<0.001)11090.900 (<0.001)11101.055 (<0.001)11110.9831112Year0811131.003 (<0.001)11140.916 (<0.001)11151.010 (<0.001)11160.9631117Area11180.932 (<0.001)11190.690 (<0.001)11200.383 (<0.001)11210.9371122Bath211230.094 (<0.001)11240.046 (0.001)11250.053 (0.78)11260.0811127ParkUn11280.092 (<0.001)11290.016 (0.48)11300.043 (0.09)11310.0431132Park111330.154 (<0.001)11340.039 (0.20)11350.090 (0.001)11360.0991137Park211380.193 (<0.001)11390.080 (0.01)11400.067 (0.58)11410.1391142Floor111430.067 (<0.001)11440.027 (0.36)11450.078 (0.01)11460.0721147Floor2_411480.104 (<0.001)11490.038 (0.16)11500.103 (<0.001)11510.1181152Floor5_811530.133 (<0.001)11540.073 (0.01)11550.134 (<0.001)11560.1471157Floor9+11580.132 (<0.001)11590.166 (0.002)11600.194 (0.01)11610.1651162Constr1850_19131163–0.092 (<0.001)11640.042 (0.18)1165–0.119 (0.01)1166–0.0831167Constr1914_19471168–0.082 (<0.001)11690.112 (0.001)1170–0.110 (0.01)1171–0.0621172Constr1948_19691173–0.153 (<0.001)11740.008 (0.73)1175–0.122 (0.002)1176–0.1191177Constr1970_19801178–0.123 (<0.001)1179–0.005 (0.83)1180–0.088 (0.03)1181–0.0831182Constr1992<11830.099 (<0.001)11840.099 (<0.001)1185–0.085 (0.12)11860.1041187CondMed1188–0.109 (<0.001)11890.009 (0.63)1190–0.072 (0.001)1191–0.1071192CondBad1193–0.215 (<0.001)1194–0.092 (0.04)1195–0.197 (<0.001)1196–0.2251197ViewNo1198–0.040 (<0.001)1199–0.027 (0.03)1200–0.017 (0.37)1201–0.0351202ViewBad1203–0.098 (0.001)1204<0.001 (0.99)1205–0.002 (0.96)1206–0.0831207Cellar112080.099 (0.001)1209–0.002 (0.88)12100.092 (<0.001)12110.0761212Cellar212130.032 (0.001)12140.001 (0.98)12150.021 (0.84)1216–0.0051217Garden12180.049 (0.02)1219–0.011 (0.73)12200.101 (0.16)12210.0461222Terrace12230.043 (0.01)12240.049 (0.01)12250.037 (0.69)12260.0411227NewApartment12280.046 (0.01)1229–0.067 (0.001)12300.256 (<0.001)12310.0471232District21233–0.109 (<0.001)1234–0.035 (0.01)1235–0.013 (0.64)1236–0.0531237District31238–0.130 (<0.001)1239–0.104 (<0.001)12400.021 (0.54)1241–0.0881242District41243–0.229 (<0.001)1244–0.188 (0.01)12450.093 (0.01)1246–0.1891247%MidIncome12480.671 (<0.001)12490.313 (0.05)12500.558 (0.01)12510.7701252%HighIncome12530.136 (<0.001)12540.09712550.029 (0.61)12560.0281257Adjusted R2 12580.83812590.83912600.70512610.8751262N observations12634,3621264873126587212664,3621267Note: Significance level is in the parentheses.126812691270===== PAGE 15 (score 95, numeric-paren 54, table-words 12, var-words 3) =====127112721151273Willingness to pay for accessibility under the conditions of residential segregation1274APPENDIX 3 1275OLS and Spatial Error models for sumbarkets1276Submarket1277Accessibility by car1278Accessibility by public transport1279TTBellecour1280TTCityHall1281Centrality 1282Index1283Accessibil-1284ity Index1285TTBellecour1286TTCityHall1287Centrality 1288Index1289Accessibil-1290ity Index1291OLS1292Sainte-Foy-lès-Lyons,129384 obs.,1294adj. R2 = 0.83 – 0.84, 1295Max VIF = 3.251296–0.5251297(0.02)1298–0.3811299(0.11)13000.0221301(0.12)13020.0061303(0.06)1304–0.3041305(0.05)1306–0.2031307(0.21)13080.0131309(0.19)13100.0031311(0.13)1312Oullins,1313100 obs.,1314adj. R2 = 0.88 – 0.89, 1315Max VIF = 3.3513160.0981317(0.62)13180.3061319(0.25)1320–0.0081321(0.39)1322–0.0031323(0.33)13240.3071325(0.18)13260.3961327(0.03)1328–0.0201329(0.08)1330–0.0051331(0.06)1332Vénissieux,133397 obs.,1334adj. R2 = 0.78 – 0.85, 1335Max VIF = 5.561336–0.2371337(0.17)1338–0.4781339(0.02)13400.0221341(<0.001)13420.0061343(<0.001)1344–0.9351345(<0.001)1346–0.9111347(<0.001)13480.0611349(<0.001)13500.0131351(<0.001)1352Vaulx-en-Velin,135354 obs.,1354adj. R2 = 0.66 – 0.68, 1355Max VIF = 2.8213561.9701357(0.19)13580.0161359(0.99)1360–0.0141361(0.74)1362–0.0051363(0.69)13640.5591365(0.54)1366–0.3551367(0.67)1368–0.0071369(0.87)1370–0.0041371(0.72)1372Spatial Error1373Sainte-Foy-lès-Lyons,1374R2 = 0.881375–0.5991376(0.01)1377–0.4271378(0.10)13790.0271380(0.10)13810.0071382(0.05)1383–0.4861384(0.01)1385–0.2681386(0.17)13870.0211388(0.10)13890.0051390(0.07)1391Oullins,1392R2 = 0.9113930.1171394(0.59)13950.3111396(0.27)1397–0.0081398(0.43)1399–0.0031400(0.35)14010.3181402(0.18)14030.4221404(0.02)1405–0.0221406(0.06)1407–0.0051408(0.05)1409Vénissieux,1410R2 = 0.87 – 0.8914110.2841412(0.39)1413–0.2601414(0.45)14150.0221416(0.04)14170.0051418(0.11)1419–0.9471420(<0.001)1421–1.1621422(<0.001)14230.0611424(<0.001)14250.0141426(<0.001)1427Vaulx-en-Velin,1428R2 = 0.8414290.1471430(0.94)1431–0.5621432(0.65)14330.0191434(0.70)14350.0031436(0.81)14370.6851438(0.50)14390.4991440(0.57)1441–0.0081442(0.86)1443–0.0051444(0.65)1445Note: Significance level is in the parentheses.1446