1# RECORD 8674ce1937729ab9e1f1084511ae7a29ecdbc6352# TITLE: The Role of Subjective Perceptions and Objective Measurements of the Urban Environment in Explaining House Prices in Greater London: A Multi-Scale Urban Morphology Analysis3# YEAR: 2023.0 VENUE: ISPRS Int. J. Geo Inf. DOI: 10.3390/ijgi120602494# AUTHORS: Sijie Yang; K. Krenz; Waishan Qiu; Wen-Jing Li5# PAGES: 21 CANDIDATE TABLE PAGES (1-based): [9, 13, 14, 15, 16]6# ABSTRACT: House prices have long been closely related to the built environment of cities, yet whether the subjective perception (SP) of these environments has a differing effect on prices at multiple urban scales is unclear. This study sheds light on the impact of people’s SP of the urban environment on house prices in a multi-scale urban morphology analysis. We trained a machine learning (ML) model to predict people’s SP of the urban environment around properties across Greater London with survey response data from an online survey evaluating people’s SP of street view image (SVI) and linked this to house price data. This information was used to construct a hedonic price model (HPM) and to evaluate the association between SP and house price data in a series of linear regression models controlling location information and urban morphological characteristics such as street network centralities at multiple urban scales, quantified using space syntax (SS) methods. The findings show that SP influences house prices, but this influence differs depending on the urban scale of analysis. Particularly, a sense of ‘enclosure’ and ‘comfort’ are important factors influencing house price variation. This study contributes by introducing SP of the urban environment as a new dimension into the traditional HPM and by exploring the economic impact of SP on the house price market at multiple urban scales.789===== PAGE 1 (score 0, numeric-paren 0, table-words 0, var-words 0) =====10116.2123.09913The Role of Subjective Perceptions14and Objective Measurements of the15Urban Environment in Explaining16House Prices in Greater London: A17Multi-Scale Urban Morphology18Analysis19Sijie Yang, Kimon Krenz, Waishan Qiu and Wenjing Li20Article21https://doi.org/10.3390/ijgi12060249222324===== PAGE 9 (score 12, numeric-paren 0, table-words 3, var-words 4) =====2526ISPRS Int. J. Geo-Inf. 2023, 12, 249278 of 2028models. Backpropagation and its application to neural networks were first proposed and29further developed by Rumelhart et al. [86]. This multilayer perceptual model can cope with30arbitrarily complex patterns for classification and provides sufficient multi-dimensional31function mapping. As shown in Figure 5, in this study, the view indexes of 30 physical32elements in a single SVI, e.g., Vsky, Vwall, Vtree, constitute the input layer. After the function33calculation in the hidden layer, the output layer yields the scores of this image on eight34subjective perceptual attributes, e.g., Simageability, Senclosure, Shuman scale. The BP neural net-35work model was used as a classifier to rate each SP variable of each SVI as an integer score36ranging from 1–5 based on the information from the SVI segmentation. The data for the37pre-trained model are derived from the results of the online perception survey based on38300 pre-determined SVIs and view indexes extracted from these images. Eighty percent of39the data are used as a training set to build the BP neural network model, while 20% are40used as a testing set to assess the model’s accuracy.41 42Figure 5. Algorithms diagram of Backprop Neural Network.43Accuracy, Precision, Recall, and F1 Score evaluate the prediction performance of44a model. The best-performing model was then used to predict the SP scores for all4570,059 SVIs. The streetscape perception scores within 1 km of a property were averaged to46represent the quality of the street environment in the property’s neighborhood. However,47it is acknowledged that what constitutes a neighborhood can vary in size and scale based48on individual preferences. Descriptive statistics for the SP scores of the street environment49around each property are presented in Table 2. In Figure 6, one can see that the SVIs are seg-50mented according to physical features to obtain visual indices of different physical features.51Subsequently, the subjective streetscape perception scores are predicted by the ML model.52Table 2. Descriptive statistics of all variables.53Variable54Description55Count56Mean57Std.Dev.58Min59Max60Data Source61PRICE62£/m2, dependent variable6349,603646793.59653427.2866117.776791,866.9568LR-PPD data69House Structure attribute70H1_FLARA71Total floor area (m2)7249,6037391.257457.15756.26764373.0077EPCs data78H2_INSUP79House insulation80performance8149,603822.58831.63841.00855.0086H3_LIGTP87House lighting performance8849,603893.69901.52911.00925.0093H4_HOTWP94House hot water performance9549,603963.80970.89981.00995.00100H5_CO2EM101House CO2 Emission10249,6031032.001041.62105−1.4010666.00107108109===== PAGE 10 (score 24, numeric-paren 0, table-words 1, var-words 10) =====110111ISPRS Int. J. Geo-Inf. 2023, 12, 2491129 of 20113Table 2. Cont.114Variable115Description116Count117Mean118Std.Dev.119Min120Max121Data Source122Description123Values124Count125Percent126Avg.Price127Avg.Area128Data Source129H6_FLLEV130Floor level1311: Low13215,95913332.17%1346908.7613565.03136EPCs data1372: Mid13832,09313964.70%1406698.90141105.081423: High14315511443.13%1457571.0314674.85147H7_PROTY148Property149type1501: Detached15134641526.98%1536123.85154174.601552: Flat15624,84115750.08%1587539.4115967.131603: Semi161796116216.05%1635683.22164113.491654: Terrace16613,33716726.89%1686244.62169101.24170H8_MENER171Main energy172source1731: Electricity17446261759.33%1766602.9617763.061782: Gas/LPG17938,16718076.94%1816409.8818297.831833: Oil/Coal1841011850.20%1866684.66187100.981884: Others189670919013.53%1919111.5219273.11193Subjective urban perception variable194Count195Mean196Std.Dev.197Min198Max199Data Source200S1_IMBLY201Perceived imageability20249,6032033.052040.362051.002064.16207SVIs, Investi-208gation data,209ML results210S2_ENCLS211Perceived enclosure21249,6032132.862140.462151.502164.65217S3_HMSCL218Perceived human scale21949,6032203.062210.262221.002234.29224S4_TRANS225Perceived transparency22649,6032272.822280.202292.002304.00231S5_CMPLY232Perceived complexity23349,6032343.242350.272361.332374.00238S6_SAFTY239Perceived sense of safety24049,6032413.242420.222431.002444.50245S7_COFRT246Perceived sense of comfort24749,6032483.032490.212502.002514.00252S8_INTST253Perceived level of interest25449,6032553.092560.172572.002584.33259Objective urban perception variable260Count261Mean262Std.Dev.263Min264Max265Data Source266Location Attribute (Urban morphology)267L1_D2CBD268Cost network distance to CBD26949,6032700.302710.152720.022730.81274OS data275L2_POSDT276Postcode District27749,603278/279/280/281/282M1_INT400283Space syntax-Integration[HH]284(R400)28549,60328623.182876.632883.5628968.69290M2_CH400291Space syntax-Choice (R400)29249,603293108.5429459.182950.00296902.55297M3_INT800298Space syntax-Integration[HH]299(R800)30049,60330149.0930220.543033.56304176.85305M4_CH800306Space syntax-Choice (R800)30749,603308788.61309485.013100.003116399.87312M5_INT2000313Space syntax-Integration[HH]314(R2000)31549,603316155.5031788.853183.56319575.72320M6_CH2000321Space syntax-Choice (R2000)32249,60332311,141.623247914.723250.0032647,980.45327M7_INT6000328Space syntax-Integration[HH]329(R6000)33049,603331650.86332459.063333.563342148.27335M8_CH6000336Space syntax-Choice (R6000)33749,603338272,714.0339249,841.43400.003411,552,460342Neighbourhood and Environment Attribute (Urban Functional Property)343F1_DENLS344Density of urban services345(within 1 km)34649,603347341.28348384.773490.003505515.00351OS data352F2_DENWK353Density of workplace (within3541 km)35549,603356337.43357463.933581.003595880.00360F3_DENAT361Density of attraction (within3621 km)36349,60336427.3636540.803660.00367572.00368F4_D2UDG369Distance to TfL station (km)37049,6033710.753720.583730.003747.26375TfL data376F5_A2UDG377Accessibility to TfL station378(within 3 km)37949,60338015.9438113.043820.0038370.00384F6_D2SCH385Distance to quality school386(km)38749,6033880.563890.403900.003916.94392Ofsted data393F7_A2SCH394Accessibility to quality school395(within 3 km)39649,60339724.9339814.763990.0040083.00401402403===== PAGE 13 (score 17, numeric-paren 0, table-words 4, var-words 5) =====404405ISPRS Int. J. Geo-Inf. 2023, 12, 24940612 of 204073. Results4083.1. Subjective Urban Perception Prediction4093.1.1. Accuracy of the Machine Learning Prediction Model410The prediction performance and model evaluation of the BP neural network model411on the SP scores of the urban street environments are shown in Table 3. The model412performed on average moderate or above for all subjective perceptual variables, with413values greater than 0.5 based on a generally accepted rule of thumb evaluation criteria. The414best-performing perceptual prediction models are models of ‘imageability’ (S1), ‘human415scale’ (S3), ‘transparency’ (S4), ‘sense of comfort’ (S7), and ‘level of interest’ (S8), all with416accuracy greater than 0.7, of which ‘human scale’ has the highest accuracy of 0.795. The417prediction models for ‘enclosure’ and ‘sense of safety’ are less accurate at 0.643 and 0.652,418respectively, while the model for ‘complexity’ has the lowest accuracy at 0.579. The accuracy419of the SP model predictions in this study is relatively high, ranging from 0.58 to 0.80, which420is a significant improvement compared to previous studies, with R2 of Verma, et al.’s421prediction model [98] ranging from 0.20 to 0.66, and R2 of Qiu, Zhang, Liu, Li, Li, Xu and422Huang’s perception prediction model [30] ranging from 0.47 to 0.61.423Table 3. Performance of BP Neural Network predictions.424Perception425Accuracy426Precision427Recall428F1-Score429Criteria430S1. Imageability4310.7104320.7314330.7254340.721435Good436S2. Enclosure4370.6434380.6304390.6664400.629441Moderate442S3. Human Scale4430.7954440.7854450.7804460.779447Good448S4. Transparency4490.7224500.7234510.7324520.724453Good454S5. Complexity4550.5794560.6284570.6134580.613459Moderate460S6. Sense of Safety4610.6524620.6574630.6644640.659465Moderate466S7. Sense of Comfort4670.7204680.7264690.7324700.716471Good472S8. Level of Interest4730.7114740.7374750.7244760.729477Good478Evaluation criteria: >90% very good, 70–90% good, 60–70% moderate, <60% low.479A higher accuracy indicates fewer variations in survey scores, which in turn may imply480that these SP dimensions can be better understood by respondents [5,30]. Imageability (S1),481human scale (S3), transparency (S4), sense of comfort (S7), and level of interest (S8), all482show higher levels of accuracy whereas complexity (S5) showed lower levels of accuracy,483hinting towards a more ambiguous perception of the latter.4843.1.2. Spatial Heterogeneity of Urban Subjective Perception485Figure 7 shows the spatial distribution of SP prediction results, with clear spatial486patterns for some SP variables. For the three SP variables of ‘imageability’ (S1), ‘enclosure’487(S2), and ‘complexity’ (S5), there is a strong tendency for the ratings to cluster geographically488towards the city center, with enclosure being the most pronounced. This phenomenon489can be explained by the high density of buildings in the city center, where the presence of490buildings makes the street space more impressionistic and more enclosed and adds to the491diversity and complexity of the urban space. For the three variables ’human scale’ (S3),492‘transparency’ (S4), and ‘sense of safety’ (S6), the geographical distributions are reversed,493with scores highly concentrated in the periphery of the city. The urban periphery, with494fewer buildings and more greenery, is considered by respondents to have a ‘human scale’495(S3), ‘transparency’ (S4), and a ‘sense of safety’ (S6). The spatial distributions of ‘sense of496comfort’ (S7) and ‘level of interest’ (S8) ratings are scattered across the research region, with497interest scores being the most dispersed, suggesting that people’s perceptions in these two498SP dimensions are more complex.499500501===== PAGE 14 (score 25, numeric-paren 8, table-words 8, var-words 0) =====502503ISPRS Int. J. Geo-Inf. 2023, 12, 24950413 of 20505 506Figure 7. Spatial distribution of properties with SP scores.5073.2. Spatial Hedonic Price Model Result5083.2.1. Correlation of Each Attribute Group with House Prices509Table 4 shows the strength of the four attribute dimensions that make up the HPM in510explaining house prices (p < 0.01). Location attributes, i.e., urban morphology attributes,511are the strongest in explaining house prices (R2 = 0.427), followed by SP scores (R2 = 0.342)512and neighborhood attributes (R2 = 0.339). The weakest influencing attribute group is the513set of house structure attributes (R2 = 0.085).514Table 4. Model performance of each attribute group.515OLS Diagnosis516House Structure Attributes517Location Attributes518(Urban Morphology)519Neighborhood Attributes520(Urban Function)521Subjective Perception Scores522Adjusted R25230.085 ***5240.427 ***5250.342 ***5260.342 ***527Pr. (F-statistic)5280.000 ***5290.000 ***5300.000 ***5310.000 ***532*** p < 0.01 level.5333.2.2. Results of Regression Models with Multi-Scale Urban Morphology534The OLS model results are shown in Table 5, and the impact ranking of each of the535five urban scale models is shown in Figure 8 respectively. The baseline model (Model 0)536537538===== PAGE 15 (score 112, numeric-paren 90, table-words 6, var-words 0) =====539540ISPRS Int. J. Geo-Inf. 2023, 12, 24954114 of 20542uses traditional location attributes, i.e., the property’s network cost distance to the CBD543(L1_D2CBD) and its postcode district (L2_POSDT). The other models (Model 1–4) apply544different scales of integration and choice as location attributes of the HPM. Table 5 shows545the regression model’s performance and diagnostic outcomes. The adjusted R2 indicates546a comparable association with house prices among the five models. When comparing547Model 0 and Model 4, the overall performance of the two models is comparable after548replacing the traditional location attributes with the urban morphology variables of space549syntax (R2 = 0.496, R2 = 0.494). However, the integration variable in Model 4 (M7_INT6000,550M8_CH6000) increases the models standardized coefficients, as shown in Figure 8. Such551a result suggests that integration captures characteristics of the urban environment that552influence house prices that traditional location attributes have been unable to represent.553Table 5. Regression with multi-scale urban morphology: results and diagnosis.554Model 0555Model 1556Model 2557Model 3558Model 4559Location560Attribute561Baseline562(L1, L2)563M1, M2564(R400)565M3, M4566(R800)567M5, M6568(R2000)569M7, M8570(R6000)571Adjusted R25720.496573***5740.482575***5760.484577***5780.486579***5800.494581***582Pr. (F-statistic)5830.000584***5850.000586***5870.000588***5890.000590***5910.000592***593Variable594Coef.595P > t596Coef.597P > t598Coef.599P > t600Coef.601P > t602Coef.603P > t604CONSTANT605***606***607***608***609***610House Structure attribute611H1_FLARA612−0.007613−0.009614**615−0.012616**617−0.014618**619−0.016620***621H2_INSUP622−0.009623**624−0.006625−0.001626−0.0016270.004628H3_LIGTP629−0.039630***631−0.037632***633−0.036634***635−0.036636***637−0.035638***639H4_HOTWP6400.0026410.0026420.0016430.0026440.003645H5_CO2EM6460.054647***6480.059649***6500.059651***6520.06653***6540.059655***656H6_FLLEV6570.068658***6590.062660***6610.063662***6630.062664***6650.066666***667H7_PROTY668−0.026669***670−0.02671***672−0.021673***674−0.021675***676−0.02677***678H8_MENER6790.122680***6810.135682***6830.136684***6850.134686***6870.137688***689Subjective urban perception variable690S1_IMBLY6910.024692***6930.032694***6950.031696***6970.035698***6990.029700***701S2_ENCLS7020.135703***7040.157705***7060.16707***7080.164709***7100.125711***712S3_HMSCL7130.043714***7150.037716***7170.035718***7190.033720***7210.027722***723S4_TRANS724−0.022725***726−0.009727***728−0.009729***730−0.01731***732−0.017733***734S5_CMPLY7350.028736***7370.034738***7390.028740***7410.028742***7430.026744***745S6_SAFTY746−0.039747***748−0.051749***750−0.051751***752−0.052753***754−0.048755***756S7_COFRT7570.112758***7590.124760***7610.126762***7630.123764***7650.112766***767S8_INTST768−0.043769***770−0.044771***772−0.044773***774−0.041775***776−0.037777***778Objective urban perception variable779Location Attribute (Urban Morphology)780L1_D2CBD781−0.098782***783/784/785/786/787L2_POSDT7880.136789***790/791/792/793/794M1_INT400795/7960.115797***798/799/800/801M2_CH400802/803−0.052804***805/806/807/808M3_INT800809/810/8110.217812***813/814/815M4_CH800816/817/818−0.14819***820/821/822M5_INT2000823/824/825/8260.286827***828/829M6_CH2000830/831/832/833−0.185834***835/836M7_INT6000837/838/839/840/8410.324842***843M8_CH6000844/845/846/847/848−0.087849***850851852===== PAGE 16 (score 45, numeric-paren 35, table-words 2, var-words 3) =====853854ISPRS Int. J. Geo-Inf. 2023, 12, 24985515 of 20856Table 5. Cont.857Model 0858Model 1859Model 2860Model 3861Model 4862Neighborhood Attribute (Urban Functional Property)863F1_DENLS8640.135865***8660.111867***8680.088869***8700.087871***8720.102873***874F2_DENWK875−0.137876***877−0.177878***879−0.165880***881−0.164882***883−0.181884***885F3_DENAT8860.419887***8880.486889***8900.479891***8920.456893***8940.437895***896F4_D2UDG897−0.028898***899−0.019900***901−0.018902***903−0.021904***905−0.02906***907F5_A2UDG9080.048909***9100.044911***9120.046913***9140.041915***9160.009917***918F6_D2SCH9190.041920***9210.031922***9230.032924***9250.03926***9270.034928***929F7_A2SCH930−0.071931***932−0.039933***934−0.045935***936−0.067937***938−0.126939***940** p < 0.05 level, *** p < 0.01 level.941 942Figure 8. Impact ranking in models based on absolute standardized coefficients.943For models 1–4, while different scales of urban morphology are used as the location944attribute, no significant change in R2 can be observed. Meanwhile, the density of attractions945(F3_DENAT) exhibits the highest standardized coefficient among all models, implying that946being close to functional opportunities has a greater influence on house prices than being947closer to generic opportunities. Other important functional variables include the density of948workplaces (F2_DENWK) and the density of urban services (F1_DENLS).949The models also show that the effect of people’s SP varies depending on varying950urban morphological scales. At a relatively large scale, i.e., a radius of 6 km (Model 4),951the integration variable (M7_INT6000) can better explain house price variation, than the952SP variables ‘enclosure’ and ‘sense of comfort’ (S2_ENCLS, S7_COFRT). The importance953diminishes as the scale decreases (Model 2, Model 3) and eventually falls below some SP954variables (Model 1). The choice variable has a similar trend in Models 1–3 but is not as955noticeable as the integration value.956In terms of SP attributes, ‘enclosure’ and ‘sense of comfort’ can best explain house price957variation, followed by ‘safety’ and ‘interest’, whereas ‘imageability’, ‘transparency’ and958‘complexity’ are the lowest. This result indicates that when holding other conditions equal,959960961===== PAGE 17 (score 6, numeric-paren 0, table-words 1, var-words 4) =====962963ISPRS Int. J. Geo-Inf. 2023, 12, 24996416 of 20965people’s SP plays a role in shaping the house price market. Particularly areas featuring a966sense of enclosure and which are also comfortable to live in tend to be more valued than967other places.9684. Conclusions969The findings show that SP of the built environment has a significant effect on house970prices. This effect is higher than neighborhood attributes (i.e., urban functions) or house971structure attributes. However, our analysis has shown that location attributes (i.e., urban972morphology) have a higher importance than SP attributes. This suggests that people’s SP973variables of the urban environment have significant impacts on house prices, above some974objective urban environment attributes.975The HPM based on multi-scale urban morphology shows that when comparing urban976morphological attributes with SP attributes, the smaller the scale of urban morphology977(i.e., the more local the measure), the more important is the SP. Moreover, the larger the978scale of urban morphology (i.e., the more global the measure), the more important are the979urban morphological attributes. For the different SP variables, ‘enclosure’ (S2_ENCLS)980and ‘sense of comfort’ (S7_COFRT) have more significant impacts on house prices than981other variables. This study adds the SP dimension to the house price assessment model,982helping to explore the impact of SP on house prices at different scales of urban morphology.983Future related research could refer to such a framework using SVI data and SP surveys984to incorporate people’s SP of the urban streetscape near the property into house price985assessment at multiple urban scales supported by SS techniques.9865. Limitation and Future Work987There are several limitations to this study. First, this study only used visual indices988from SVI segmentation in building an SP prediction model. The prediction accuracy989of this approach may be improved by increasing the quality and precision of SVI data990extraction and expanding the volume of perception survey data: (i) more features, such as991HSL histograms and Blob detection, may be added to complement the physical features992presented by the SVIs; (ii) more field experiences and surveys may be added to the study to993cross-validate the SP predictions and minimize the bias introduced by web image surveys.994Second, in the SP survey, some participants’ understanding of the concepts of SP may not995accurately capture their perception. A better questionnaire design and question setting as996well as a larger sample size could improve this limitation. Future research could involve a997larger more representative sample including urban design professionals to provide a more998accurate and efficient assessment of the street environment in images. In addition, VR999technology and panoramic SVIs can be used in the perception survey sessions to provide1000participants with a more realistic environment and more accurate exploration of their SP.1001Finally, many studies have shown that spatial dependence and non-stationarity violate the1002basic assumptions of OLS regression. Due to spatial correlation effects, OLS regression1003models may be biased in their coefficients and report incorrect significance. More tests for1004OLS regression may be carried out, such as Moran’s I and robust Lagrange multiplier (LM)1005tests, to see if lags and errors due to spatial correlation exist. Given there are spatial lags or1006errors in OLS regression models, other regression models that solve for spatial dependence1007may be introduced for more accurate multi-factor impact studies, including the Kelejian–1008Prucha’s model (SAC) [99] and geographically weighted regression (GWR) [100]. These1009models can effectively manage spatial dependence and non-stationarity-related issues and1010achieve higher R2 values, thereby enhancing the models’ precision and providing stronger1011data support for the paper’s conclusions [51,55].1012Author Contributions: Conceptualization, Sijie Yang, Kimon Krenz, and Waishan Qiu; methodology,1013Sijie Yang, Kimon Krenz, and Wenjing Li; software, Sijie Yang; validation, Sijie Yang; formal analysis,1014Sijie Yang; investigation, Sijie Yang; resources, Wenjing Li; data curation, Sijie Yang; writing—original1015draft preparation, Sijie Yang; writing—review and editing, Kimon Krenz and Waishan Qiu; visualiza-1016