Modelling the Impact of Accessibility Modelling the Impact of Accessibility to Services on House Prices:to Services on House Prices:A Comparative Analysis of A Comparative Analysis of

Two Methodological ApproachesTwo Methodological Approaches

François Des Rosiers, Marius Thériault & Yan Kestens

European Real Estate Society10th Annual Meeting, June 11-13, 2003

Research funded by

This paper is an attempt to bridge the gap between, on the one hand, the mobility behaviour of households and their perception of accessibility to urban amenities and, on the other hand, house price dynamics as captured through hedonic modellingIt consists of an empirical test of the impact of accessibility on house prices, whereby hedonic modelling is applied to some 952 single-family houses sold in Quebec City between 1993 and 1996Two accessibility measures are compared: the former measure is based on simulated travel times to nearest amenities aggregated through factor analysis (PCA)The latter rests on perceived accessibility indices obtained via a fuzzy logic approach applied to observed trips patterns derived from the 2001 QMA O-D survey

Introduction

Our hypothesis is that different people having a heterogeneous perception of space, they will adjust their willingness to pay for additional centrality/accessibility when choosing their home location depending on their needs and preferences The main objective of this paper is to test whether perceptual indices of accessibility are actually internalized in housing pricesSecondary objectives are:Testing for the marginal contribution to value of access to

various amenities: work places, schools, shops, groceries, health care centres, restaurants, leisure places

Testing for the differential impact of accessibility among types of individuals and households

Testing for the way life cycle and income impact upon the perception of accessibility and is translated into house prices

Hypothesis and objectives

Traditional urban models are currently based on the centrality concept (distance decay function) and on accessibility to the CBD (monocentric model)

McMillen’s (2003 – Chicago): decades of urban sprawl in North American cities did not weaken the prominence of the centrality concept

Impact of proximity and accessibility to services on property values: Guntermann and Colwell 1983, Colwell, Gujral and Coley 1985, Colwell 1990, Grieson and White 1989, Sirpal 1994, So et al. 1997, Smersh and Smith 2000, Des Rosiers et al. (1996, 2001 & 2003 – Quebec City)

Not all authors though agree on the actual influence of accessibility upon house prices and residential mobility (McGreal et al. 1999 – Belfast & Bordeaux)

Accessibility and House Values (1)

Polycentric cities: mere Euclidean distances to the CBD falls short of integrating all relevant aspects of accessibility (Jackson 1979, Dubin and Sung 1987, Niedercorn and Ammari 1987, Hoch and Waddell 1993)

Despite use of minimum travel time and walking distance (Bateman et al. 2001), the faulty specification of accessibility descriptors may explain rather poor performances

Travel surveys, commuting patterns and accessibility to jobs and houses:Levinson (1996 – Washington, DC): suburbanization of jobs

maintains stability in commuting durations despite rising congestion and increasing work and non-work trip making and length

Helling (1996 - Atlanta): Effect of residential car accessibility to jobs on the quantity and nature of travel by men and women - Accessibility do not affect everyone while gravity indices only provide partial information

Srour et al. (2002 - Dallas-Fort Worth): Apply both general and specific accessibility indices to the modelling of residential and commercial markets - While common accessibility measures do not perform that well, job accessibility indices impact positively on residential land values

Accessibility and House Values (2)

Database: hedonic modelling applied to 952 single-family houses sold in Quebec City between 1993 and 1996 - sale prices range from $50 000 to $460 000 (Can.$)

High variance on prices: use of a multiplicative functional form (ln of sale price – ln SP)

Three steps:Model 1: Ln SP = f [Property SSpecifics, IInflation, TTaxation]Model 2: Ln SP = f [SS, II, TT, PCAPCA of travel times to nearest amenities]Model 3: Ln SP = f [SS, II, TT, PAIPAI : Perceived Accessibility Indices]

Phone survey among buyers revealed that accessibility to services, jobs, schools, highways and transit networks was an important criteria for choosing new neighbourhoods:Model 3a : Ln SP = f [SS, II, TT, PAIPAI * Buyer’s AgeAge]Model 3b : Ln SP = f [SS, II, TT, PAIPAI ** Buyer’s IncomeIncome]

Database & Modelling Approach

Step 1: compute 15 travel times (car and walking) to the nearest local & regional amenities : primary & high schools, colleges, universities; regional, neighbourhood & local shopping centres; CBD

Step 2: PCA - extract 2 principal components using Varimax rotation Factor 1 : access to nearest regional-level services (42% of

variance)

Factor 2 : access to nearest local-level services (34% of variance)

Already used by Des Rosiers et al., 2000 – Quebec City

Mutually independent factors help control multicollinearity

Step 3: Model 2 - Factor scores are substituted for access attributes

Factor Analysis - PCA (Nearest Amenities)

Factor Analysis - PCA (Nearest amenities)

Regional-level services

Local-level services

where :Ai : Raw suitability of residential location i (sum of suitable opportunities)Sij : Suitability index of travelling from residential location i to activity location j : Total number of potential activities at location j

where :Ai* : Accessibility index of residential location i relative to the most suitable place

Modelling Perceptual Accessibility

Model 3 : Accessibility indices were computed for significantly different types of persons and activities using

mjniPSAm

jjiji ,...,1,...,1,

1

niAAAAA

n

ii ,...,1,),...,,max(100

21

*

jP

Perceived Accessibility to Restaurants

0 5 10

kilometres

Restaurants

3 000

1 500

300

Accessibility Index (%)90 to 10080 to 9070 to 8060 to 7050 to 6040 to 5030 to 4020 to 3010 to 200 to 10

C50 : 5,3 min.C90 : 12,6 min.

Analysis of Results (1)Model Summary

Model R R Square Adjusted R Square Std. Error of the Estimate Maximum VIF Residuals Moran’s I

Sig.

1 .858 .736 .731 .17959 3.050 0.2827 .002

2 .884 .782 .777 .16330 3.053 0.2376 .008

3 .874 .763 .758 .17040 3.060 0.2732 .003

ANOVA

Model Sum of Squares df Mean Square F Sig.

1 Regression 83.884 18 4.660 144.487 .000

Residual 30.093 933 .032

Total 113.977 951

2 Regression 89.126 20 4.456 166.951 .000

Residual 24.851 931 .027

Total 113.977 951

3 Regression 86.973 21 4.142 142.631 .000

Residual 27.004 930 .029

Total 113.977 951

All models do perform well in spite of remaining spatial autocorrelation among residuals

Model 2 performs better in all respects

Analysis of Results (2)

Model 1 Model 2 Model 3

B Std. Err

Beta t B Std. Err

Beta t B Std. Err

Beta t

(Constant) 11.68731 .04746 246.3 11.55619 .05250 220.1 11.50028 .05038 228.3

LotSizeSqrMetres .00003 .00002 .031 1.6 .00008 .00002 .078 4.3 .00008 .00002 .080 4.2

Bungalow * LivArea .00235 .00018 .357 13.3 .00231 .00016 .351 14.4 .00228 .00017 .346 13.6

Cottage * LivArea .00249 .00013 .569 19.4 .00250 .00012 .571 21.3 .00247 .00012 .565 20.2

Attached * LivArea .00149 .00027 .101 5.5 .00098 .00025 .067 3.9 .00112 .00026 .076 4.3

AppAge -.00387 .00057 -.138 -6.8 -.00853 .00062 -.303 -13.8 -.00662 .00060 -.235 -11.0

Washrooms .09517 .01252 .144 7.6 .07281 .01151 .110 6.3 .08150 .01197 .124 6.8

Fireplace .05082 .01209 .079 4.2 .04970 .01101 .077 4.5 .05106 .01149 .079 4.4

HardWoodStair .07454 .01623 .096 4.6 .05922 .01489 .076 4.0 .06646 .01544 .085 4.3

HighQualFloor .06689 .01295 .097 5.2 .04912 .01185 .071 4.1 .05814 .01232 .084 4.7

Terrace .12394 .04813 .045 2.6 .10813 .04382 .039 2.5 .10856 .04577 .039 2.4

Brick51FC .04567 .01420 .064 3.2 .03660 .01294 .051 2.8 .04089 .01349 .057 3.0

Clapbord51 -.05414 .01565 -.069 -3.5 -.04675 .01425 -.060 -3.3 -.05210 .01489 -.067 -3.5

SimpAttGarage .13307 .02731 .085 4.9 .11599 .02488 .074 4.7 .12187 .02598 .078 4.7

DoubAttGarage .16945 .03793 .080 4.5 .13446 .03459 .063 3.9 .15802 .03626 .074 4.4

DoubDetGarage .10959 .03132 .062 3.5 .12144 .02857 .069 4.3 .11030 .02974 .062 3.7

ExcaPool .18383 .02617 .125 7.0 .16487 .02386 .112 6.9 .16491 .02495 .112 6.6

Month93Jan -.00184 .00045 -.070 -4.0 -.00167 .00041 -.063 -4.1 -.00191 .00043 -.072 -4.5

Tax_OvUnzdRate -.25656 .01589 -.292 -16.1 -.14557 .02068 -.166 -7.0 -.25032 .01575 -.285 -15.9

Acces_Factor1 .12485 .00959 .322 13.0

Acces_Factor2 .04177 .00871 .090 4.8

AWork * NoWorkerHld .00287 .00042 .181 6.8

AWork * WorkerHld .00273 .00035 .216 7.7

Centrality Index .00173 .00051 .068 3.4

… Size and Age coefficients are strengthened.Tax rate effect

declines.This suggests

structural spatial links among these

variables and urban form.

Model 1 : All coefficients highly significant

and consistent with expectations.Prominence of age, size and

taxation

Model 2 : Factors 1 and 2

substantially improve

performances.Most other coefficients

unchanged, but…

Model 3 : Journey-to-Work coefficients highly significant

even when controlling for urban centrality.

Perceptual accessibility indices provide a more comprehensive picture of accessibility – more

related to people and less related to closest amenities.

Analysis of Results (3)

Model Accessibility index R Square SE Estimate B Std. Err Beta t VIF

4 ASchool * Family .765 .1678 .00333 .00035 .279 9.6 3.431 Schools ASchool * ChildlessHld .00255 .00032 .220 8.0 3.068

Centrality Index .00146 .00050 .058 2.9 1.572

9 AHealthCare * Family .766 .1673 .00342 .00035 .265 9.9 2.947 Health care AHealthCare *

ChildlessHld .00262 .00033 .199 7.9 2.574

Centrality Index .00124 .00050 .049 2.4 1.618

10 ARestaurant .768 .1668 .00323 .00032 .212 10.1 1.801 Restaurants Centrality Index .00120 .00050 .047 2.4 1.608

Models 4, 9 and 10 : Accessibility to schools and health care facilities for families as well as to restaurants exerts strong influence on prices.

Perceived accessibility indices overcome centrality.

Analysis of Results (4)

Model Accessibility index R Square SE Estimate B Std. Err Beta t VIF

3a AWork * Age34less .757 .1704 .00220 .00037 .155 5.9 2.698

Work places AWork * Age35-44 .00301 .00033 .306 9.0 4.507

Age groups AWork * Age45-54 .00324 .00033 .318 9.7 4.236

AWork * Age55more .00317 .00039 .194 8.1 2.229

3b AWork .771 .1655 .00311 .00037 .179 8.3 1.914

Work places AWork * Income<60K$ -.00111 .00024 -.098 -4.7 1.811

Household AWork * Income60-80K$ -.00060 .00024 -.050 -2.5 1.682

Income level AWork * Income80-100K$ -.00029 .00026 -.021 -1.1 1.544

AWork * income>100K$ .00074 .00025 .060 2.9 1.737

Centrality Index .00192 .00049 .076 3.9 1.582

Model 3a : People aged 35-54 are willing to pay a substantial market premium to locate at a reasonable travel time from their

work place.Model 3b : The higher the household income, the stronger the

propensity to lessen work-trip duration: under an income constraint, households trade-off longer commuting trips for

cheaper land.

The two sub-hypothesis of this research were: [1] Various types of persons experience different constraints and are not

equally willing to travel in order to reach various kinds of activities, meaning that they have an heterogeneous perception of space

[2] Households will adjust their willingness to pay for additional centrality/accessibility when choosing their home location depending on their needs and preferences

Both sub-hypotheses are supported by empirical results, suggesting the accessibility concept might be less straightforwardless straightforward than is usually considered in the literature

The physical, absolute concept of accessibility ought to be complemented by behavioural approachesbehavioural approaches integrating people wills and needs in their valuation of urban space

Considering they are a paramount determinant of location choices and property values, accessibility/centrality issues deserve being further investigated using novel toolsnovel tools, including travel and activity surveys and modelling

Conclusion & Research Agenda