1# RECORD 13a61953f01e5f8f119d4fa1872eb725032525262# TITLE: The Reflection of Income Segregation and Accessibility Cleavages in Sydney's House Prices3# YEAR: 2021.0 VENUE: ISPRS Int. J. Geo Inf. DOI: 10.31235/osf.io/2psk54# AUTHORS: M. Ng; J. Roper; C. Lee; C. Pettit5# PAGES: 25 CANDIDATE TABLE PAGES (1-based): [13, 19]6# ABSTRACT: This paper explores the relationships between accessibility, income segregation, and house prices in the Greater Sydney Area. Sydney is found to have reasonably even employment accessibility, reflecting the increasingly polycentric nature of the modern city; however, it also shows considerable income segregation and variance in property prices between different parts of the city. Entropy is used to examine the diversity and mixing of different income groups. Finally, hedonic price models using ordinary least square and geographically-weighted regression techniques to show the differing effects of employment accessibility on house prices in different parts of the city. The results show that accessibility has small to negative effects on prices in the most valuable areas, suggesting that other effects such as recreational access and employment type/quality may be important to investigate further in this context.789===== PAGE 1 (score 2, numeric-paren 0, table-words 0, var-words 2) =====1011The Reflection of Income Segregation and Accessibility12Cleavages in Sydney’s House Prices13Author:14Ng, MKM; Roper, J; Lee, CL; Pettit, C15Publication details:16ISPRS International Journal of Geo Information17v. 1118Chapter No. 7192220-9964 (ISSN)20Publication Date:212022-07-0122Publisher DOI:23https://doi.org/10.3390/ijgi1107041324License:25https://creativecommons.org/licenses/by/4.0/26Link to license to see what you are allowed to do with this resource.27Downloaded from http://hdl.handle.net/1959.4/unsworks_80905 in https://28unsworks.unsw.edu.au on 2026-10-20293031===== PAGE 13 (score 10, numeric-paren 0, table-words 2, var-words 4) =====3233ISPRS Int. J. Geo-Inf. 2022, 11, 4133412 of 2435being unaccounted for by all other independent variables within the model. Remedial36measures to address the arising issues with multicollinearity include variable exclusions,37principal component analyses, or step-wise regression applications [82]. This paper opted38to exclude collinear variable pairs with an observed VIF scores greater than 10 (VIF ≥10).396. Results Objective 1: Income Distributions in Greater Sydney40As the first objective, the spatial distribution of income groups within the Greater41Sydney area was calculated. Entropy statistics (H) were used to quantify the spatial42diversity between these income groups. These values ranged from 0.00 to a maximum of431.10 (Hmax)—meaning, in areas where H = 1.10, an equal distribution of all three income44groups is observed. The global mean entropy value in Greater Sydney was calculated to be450.862, with a minimum and maximum of 0.00 and 1.09, respectively.46Table 3 quantifies the correlation coefficients between all income groups and their47respective entropy values. What is first noticeable is the negative relationship between48the distribution of low-income groups in Sydney with their middle-(R = −0.22) and high-49(R = −0.56) income counterparts. This relationship is also seen with respect to entropy50values (R = −0.22). These findings come in stark contrast to those values associated51with high- and middle-income earners, with both groups displaying similar correlation52coefficients to the diversity index. These findings indicate that Sydney’s income distribution53appears to be highly asymmetric. Where it was initially thought that both high- and low-54income groups would both display exiguous mixing (i.e., low entropy scores in both high-55and low-income groups), this generalisation can only be extended to the latter. The results56rather show that high-income groups in Sydney appear relatively more integrated at the57metropolitan level, whereas low-income earners look to be concentrated in locations where58those outside this income bracket remain unlikely to reside in.59Table 3. Global correlation coefficients between income groups in Greater Sydney.60Low Income61Middle Income62High Income63Entropy64Low Income651.00066−0.21867−0.56168−0.21969Middle Income70−0.218711.000720.198730.56574High Income75−0.561760.198771.000780.62979Entropy80−0.219810.565820.629831.00084This more cursory evaluation does hint towards a possible degree of socioeconomic85segregation in Greater Sydney; however, a disaggregated view of income distributions86is needed to investigate whether the prominence of low-income groups is a consistent87mainstay of areas of low diversity. Figure 7 illustrates the spatial distribution of entropy88and all income groups in view of this. It is evident from this visualisation that the spread89of entropy values shows a relatively clear delineation between low- and high-diversity90areas, with lower entropy areas concentrated in the Central and Western suburbs of Greater91Sydney, such as suburbs around Fairfield, Merrylands, Auburn, and Mount Druitt. In these92areas, the diversity index is estimated to range between 0.39 and 0.55; and they also have93the highest proportions of low-income earners in Greater Sydney (Figure 8). Here, low-94income groups constitute between 63 and 77 per cent of the population—often, with less95than 2 per cent belonging to high-income groups (Figure 8). The findings echo back to96previous work by Lee et al. [83], which discussed the divide of the Sydney’s demographic97and socioeconomic distribution along a ‘latte line’.9899100===== PAGE 14 (score 4, numeric-paren 0, table-words 1, var-words 2) =====101102ISPRS Int. J. Geo-Inf. 2022, 11, 41310313 of 24104Figure 7. Entropy and income group distribution in 2016.105Figure 8. Entropy values against low-income groups in Greater Sydney.106When considered in line with the distribution of high-income groups, it is worth107pondering on the reason these areas do not exhibit similar segregation seen with low-income108areas. From the results displayed in Table 3, the distribution of high-income groups has109already shown to be negatively correlated with the lowest earners (R = −0.57) in Sydney,110which alludes to the spatial division between both groups. However, it is interesting to111note that, despite this disparity, high-income groups also occupy those areas with the112highest entropy values. One explanation is due to the asymmetrical size of the different113groups—as the high income group represents only 12 per cent of Sydney’s population114overall, even areas where they are heavily over-represented will still show considerable115116117===== PAGE 19 (score 50, numeric-paren 5, table-words 9, var-words 18) =====118119ISPRS Int. J. Geo-Inf. 2022, 11, 41312018 of 24121level, there is a relative rise of 0.3 per cent of house prices in areas that have increased122accessibility. It is worth noting that the presence of middle-income groups, as defined in123this study, does not have the same level of predictive potential. This is likely due to their124larger aggregation of income into a single group; and resulting relatively even distribution125throughout all of Greater Sydney.126Table 4. Coefficient values derived from the OLS regression of variables to house prices.127Residuals:128Min1291Q130Median1313Q132Max133−3.08558134−0.14556135−0.018441360.116331372.40827138Coefficients:139Estimate140Std. Error141t-Value142Pr (>|t|)143(Intercept)1441.33 × 1011455.69 × 10−2146233.182147<2.00 × 10−6148***149Bedrooms1501.42 × 10−11511.58 × 10−315290.152153<2.00 × 10−6154***155Bathrooms1564.89 × 10−21571.48 × 10−315833.016159<2.00 × 10−6160***161Distance to Sydney162−1.45 × 10−51631.69 × 10−7164−85.66165<2.00 × 10−6166***167Distance to Secondary City1687.07 × 10−71691.74 × 10−71704.0651714.82 × 10−5172***173Distance to Beach174−2.44 × 10−61752.24 × 10−7176−10.916177<2.00 × 10−6178***179Distance to Bus Interchanges1802.91 × 10−61814.83 × 10−71826.0311831.65 × 10−9184***185Distance to Highway1867.50 × 10−61874.44 × 10−718816.883189<2.00 × 10−6190***191Distance to Primary School1921.02 × 10−51933.26 × 10−61943.1181950.00182196**197Distance to University1981.12 × 10−61994.27 × 10−72002.6172010.00889202**203Distance to Swimming Pool204−4.61 × 10−62059.32 × 10−7206−4.9522077.40 × 10−7208***209Distance to Shopping Centre2101.11 × 10−52117.50 × 10−721214.774213<2.00 × 10−6214***215Distance to Sport Centre2169.00 × 10−62177.27 × 10−721812.378219<2.00 × 10−6220***221Population Above 65 (%)2225.08 × 10−32232.17 × 10−422423.355225<2.00 × 10−6226***227Crime Rate (%)228−4.35 × 10−22291.62 × 10−2230−2.6812310.00735232**233Low Income (%)234−1.87 × 10−12355.75 × 10−2236−3.2442370.00118238**239Middle Income (%)240−1.02 × 10−12415.80 × 10−2242−1.7582430.0788244.245High Income (%)2462.512475.97 × 10−224841.983249<2.00 × 10−6250***251Accessibility Score2522.73 × 10−12538.68 × 10−22543.1422550.00168256**257Significance Codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘.’ 0.1; Number of data points: 32,050; Residual standard error: 0.2455 on25832,049 degrees of freedom; Multiple R-squared: 0.8103, Adjusted R-squared: 0.8102; F-statistic: 7604 on 18 and25932,049 DF, p-value: <2.2 × 10−16; Residual sum of squares: 1931.751; AIC: 952.1797; AICc: 952.2059.2608.3. GWR Results261Whilst the OLS model provides a relatively sound representation of house price262variances at the metropolitan level, local effects also need to be considered with respect263to the spatial non-stationarity of property. These local effects are considered within the264subsequent GWR model. The GWR model specified here shows an improvement in fit,265with an adjusted R2 value of 0.844 and a lowered residual sum of squares value of 1569.266The spatial distribution of the local R2 values, along with the coefficient estimates for267accessibility, high- and low-income groups, is illustrated in Figure 13. The GWR model268performance is high everywhere, with at least 67 per cent of house price variances accounted269for across all of Greater Sydney.270A summary of the model’s coefficient estimates is also provided in Table 5. The find-271ings here provide a more granular estimate of the model’s income and accessibility effects—272in which, larger spatial disparities at local levels are more evident than previously indicated273in the OLS model. The model’s estimate coefficients indicate that, generally, improvements274in accessibility garner a 3.3 per cent median increase in house prices. However, disaggre-275gating these results spatially, the models show a clear divide between the East and West276regions of Greater Sydney. The most striking observation from these GWR results is the277difference in direction and magnitude of the effect of accessibility across Greater Sydney.278Generally, there is a negative relationship of prices with accessibility in the far Southeast279and Northeast of Sydney, whilst there is a weak positive relationship in central Sydney.280Further, in Western and outer Sydney, a stronger positive relationship is documented.281These findings suggest that accessibility may not necessarily be an important determinant282of house prices in all areas. There may be several possible explanations for this.283284285===== PAGE 20 (score 23, numeric-paren 0, table-words 1, var-words 21) =====286287ISPRS Int. J. Geo-Inf. 2022, 11, 41328819 of 24289Figure 13. Contour visualisation of income and accessibility coefficients throughout Greater Sydney.290Table 5. Distribution of coefficient values from the GWR model of variables to house prices.291Summary of GWR Coefficient Estimates292Min2931Q294Median2953Q296Max297Intercept2986.22991.3 × 1013001.4 × 1013011.4 × 1013021.8 × 101303Bedrooms3041.1 × 10−13051.2 × 10−13061.4 × 10−13071.6 × 10−13081.9 × 10−1309Bathrooms3102.8 × 10−23114.0 × 10−23125.1 × 10−23135.8 × 10−23148.1 × 10−2315Distance to Sydney316−3.4 × 10−5317−3.0 × 10−5318−2.3 × 10−5319−1.5 × 10−53200.0321Distance to Secondary City322−1.6 × 10−5323−4.7 × 10−6324−5.6 × 10−73254.1 × 10−63260.0327Distance to Beach328−3.8 × 10−5329−8.7 × 10−63302.3 × 10−63318.7 × 10−63320.0333Distance to Bus Interchanges334−3.8 × 10−5335−9.7 × 10−6336−3.2 × 10−63374.4 × 10−63380.0339Distance to Highway340−1.8 × 10−53413.0 × 10−63427.6 × 10−63431.1 × 10−53440.0345Distance to Primary School346−3.7 × 10−53471.3 × 10−53482.7 × 10−53494.5 × 10−53501.0 × 10−4351Distance to University352−2.1 × 10−53532.3 × 10−63546.5 × 10−63551.1 × 10−53560.0357Distance to Swimming Pool358−2.7 × 10−5359−1.3 × 10−5360-4.5 × 10−63616.8 × 10−63620.0363Distance to Shopping Centres364−1.5 × 10−5365−4.0 × 10−73667.2 × 10−63671.8 × 10−53680.0369Distance to Sports Centres370−1.2 × 10−53713.6 × 10−63726.7 × 10−63731.4 × 10−53740.0375Population Above 653761.2 × 10−33772.6 × 10−33783.7 × 10−33795.8 × 10−33808.9 × 10−3381Crime Rate382−3.1 × 10−1383−8.0 × 10−2384−2.1 × 10−23851.0 × 10−13866.8 × 10−1387Low Income (%)388−7.2389−3.8 × 10−1390−2.8 × 10−1391−1.6 × 10−13925.0393Middle Income (%)394−6.6395−3.1 × 10−1396−1.8 × 10−1397−6.7 × 10−23985.4399High Income (%)400−4.54011.54021.74031.84047.7405Accessibility Score406−7.8 × 10−1407−2.8 × 10−14083.3 × 10−44095.7 × 10−14101.1411Diagnostic information: Number of data points: 32,068; Effective number of parameters: 192.2415; Effective412degrees of freedom: 31,875.76; AICc: −5447.286; AIC: −5594.308; Residual sum of squares: 1569.958; R-square413value: 0.8458043; Adjusted R-square value: 0.8448744.414First, the higher attractiveness of homes in Sydney’s more affluent suburbs (e.g., in415Eastern suburbs) appear to be more sensitive to locational attributes such as views, access416to green-spaces and beaches, good schools, and larger houses compared with employment417accessibility. Importantly, these affluent suburbs are characterised by these better amenities418compared with less prosperous suburbs [11]. In comparison, in more segregated areas in the419outer Sydney suburbs, home choice becomes more utilitarian. In these areas, accessibility420to employment becomes a more important factor in determining house prices. This is421likely related to demographic differences seen between these areas. In wealthier areas of422Sydney, accessibility based on journey to work may not be as important for wealthy retirees423that cluster in these areas (e.g., Sutherland Shire, Hornsby-Warringah, and the Northern424Beaches). Another issue may be the further anisotropic nature of Sydney’s public transport425network, resulting in travel times for subgroups of commuters that differ from the road426network based accessibility used in this model. As shown in Figure 14, the Southeastern and427Northeastern suburbs are relatively less served by rail lines or dedicated right-of-way bus428services than other suburbs (in 2016—prior to the opening of new light rail in the Southeast).429