Household Mobility and House Values A Perceptual Approach

Household Mobility and House Values: A Perceptual Approach to Modelling Accessibility to Urban Services François Des Rosiers, Marius Thériault , Florent Joerin, Paul Villeneuve & Murtaza Haider MCRI-ILUTE 2 nd International PROCESSUS Colloquium Toronto, Canada, June 12 -15, 2005 Research funded by

Introduction: Context and objective (1) 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 The impact of accessibility on house prices is empirically tested applying hedonic modelling to some 952 single-family houses sold in Quebec City between 1993 and 1996 Two sets of accessibility indices are compared: the first one is based on simulated travel times to nearest amenities aggregated through factor analysis (PCA) q Second one rests on perceived accessibility indices obtained via a fuzzy logic approach applied to observed trip patterns derived from the 2001 QMA O-D survey q A centrality index is also developed and used concurrently to perceived accessibility indices

Introduction: Context and objective (2) The overall objective of this paper is to test whether perceptual indices of accessibility are actually internalized in housing prices This rests upon the assumption that the perception of space is not homogeneous and that households will adjust their willingness to pay for an improved access to urban amenities to their needs, preferences and profile Secondary objectives are: èTesting for the marginal contribution to value of centrality as opposed to accessibility èTesting for the relative superiority of “subjective”, over “objective” accessibility indices in their ability to explain house price differences

Measuring accessibility: Time vs. distance (1) Traditional urban models are currently based on the centrality concept (distance decay function) and on accessibility to the CBD (monocentric model) As cities grow in complexity though and turn polycentric, mere Euclidean distance 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

Measuring accessibility: Time vs. distance (2) Hence the usefulness of travel surveys to analyze commuting patterns and accessibility to jobs and houses: è Levinson (1996 – Washington, DC): Commuting durations remain stable in spite of suburbanization of jobs and rising total trip volume è Helling (1996 - Atlanta): Accessibility do not affect everyone similarly while gravity indices only provide partial information Srour et al. (2002 - Dallas-Fort Worth): Specific accessibility-tojob indices tend to perform better that overall measurements while impacting positively on residential land values More sensitive, GIS-derived, measurements of actual road distances and travel times can improve our understanding of mobility behaviour (Thériault et al. 1999 a & 1999 b, Vandersmissen et al. 2003 & 2004)

Designing “objective” accessibility indices using factor analysis (PCA) (1) Step 1: a distance and trip duration modelling procedure is applied to the Quebec Metropolitan Area (QMA) street network using the Trans. CAD transportation-oriented GIS software The resulting regional network is composed of 29 035 street segments - acting as directional links - and 20 262 nodes acting as street intersections - (Des Rosiers et al. 2000) Each property, residential and non residential, can be easily located in the regional GIS which serves as the basic processing device for this study Since speed limits, one-ways and impedance (crossing time) are known for all street segments, distances and access times from each home to any service centre may be computed

Designing “objective” accessibility indices using factor analysis (PCA) (2) Step 2: 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 3: PCA - extract 2 principal components using Varimax rotation è Access_Factor 1 : access to nearest regional-level services (42% of variance) è Access_Factor 2 : access to nearest local-level services (34% of variance) q PCA successfully used by Des Rosiers et al. , 2000 q Mutually independent factors help control multicollinearity Step 4: Factor scores are substituted for access attributes

PCA-derived accessibility indices Access_Factor 1 Regional-level services Access_Factor 2 Local-level services

Integrating QMA’s O-D survey trip patterns Conducted from mid-September to mid-December 2001 by the Ministry of Transport of Quebec (MTQ) and the Quebec City Transit Authority (RTC) Involves 68 121 persons ( 27 839 households) and 174 243 weekday trips (Monday to Friday) Homes and activity places are located using street addresses Each person is characterized by age, gender, occupation and ownership of a car driver licence Various household types are formed (lone person, childless couple, two-parent family, lone-parent family, and other households) Trip purpose, transportation mode as well as departure time, origin and destination of each trip are known Car-based trips represent 73, 3% of all trips; of these, 29 602 originate from Quebec City residential areas

Trip duration differences By Type of Person/Household

Modelling “subjective” accessibility indices using fuzzy logic criteria (1) According to Kim and Kwan (2003) accessibility measurement should consider “thresholds on activity participation time and travel time in order to identify a meaningful opportunity set when evaluating space-time accessibility” Suitability indices (Sij ) of travelling from home i to activity j are computed from actual O-D trip durations based on travel time satisfaction thresholds, using the following fuzzy logic criteria: q [1] any travel time < median (C 50) is totally acceptable (1) q [2] a travel time > C 90 is likely to be unsatisfactory (0) q [3] intermediate cases obtained using linear interpolation between 1 and 0

Modelling “subjective” accessibility indices using fuzzy logic criteria (2) Finally, raw and relative accessibility indices are computed as follows: 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 Pj : Total number of potential activities at location j where: Ai* : Accessibility index of residential location i relative to the most suitable place

Travel time satisfaction thresholds by trip purpose and type of person/hhld

Perceived Accessibility to Restaurants C 50 : 5, 3 min. C 90 : 12, 6 min.

Perceived Accessibility to workplace Women C 50 : 5, 3 min. C 90 : 12, 6 min.

Perceived Accessibility to groceries Families C 50 : 5, 3 min. C 90 : 12, 6 min.

Designing an urban centrality index Gravity model of interaction flows (Tiefelsdorf, 2003) used to estimate centrality: where: μij : Expected number of car trips between locations i and j Pi : Total population at residential location i Pj : Total number of potential activities at location j Dij : Travel time by car from residential location i to activity location j (minutes) Resulting 52 801 392 potential trips were reduced to 300 079 pairs of residential-activity places (563 residential grid cells x 533 activity cells) Indices then normalized to 100 by comparison to observed maximum flow

Database & Modelling Approach (1) 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 (mean price: $109 000) Due to a high variance on sale prices (SP), we use a semilog, multiplicative functional form (Dependent var. : ln_SP) Phone survey conducted on sampled houses to get information about buyer’s profile (household structure, age, income, choice factors, etc. ) Survey revealed that accessibility to services, jobs, schools, highways and transit networks was an important criteria for choosing new neighbourhoods

Database & Modelling Approach Three steps: q Model 1: Ln SP = f [Property Specifics, Inflation, Taxation] q Model 2: Ln SP = f [S, I, T, PCA of travel times to nearest PCA amenities] q Models 3 @ 12: Ln SP = f [S, I, T, PAI (Perceived Accessibility PAI Indices), Centrality Index]

Main regression results (1) All models do perform well in spite of better in all respects Model 2 performs remaining spatial autocorrelation among residuals

Main regression results (2) Model 3 : Journey-to-Work coefficients highly significant even when controlling for urban centrality. Perceptual accessibility indices Model 2 highly provide a more comprehensivecoefficients and Age Model 1 : All … Size : Factors picture of accessibility – more significant 1 and 2 are coefficients substantially related to people and less with strengthened. and consistent expectations. related to closest amenities. of age, improve Prominence Tax rate effect size and performances. taxation declines. Most other This suggests coefficients structural spatial unchanged, but… links among these variables and urban form.

Main regression results (3) ØModels 4, 9 and 10 : Accessibility to schools and health care facilities for families as well as to restaurants exerts strong influence on prices ØPerceived accessibility indices far outweigh centrality

Main regression results (4) ØModel 11 : People aged 35 -54 are willing to pay a substantial market premium to locate at a reasonable travel time from their work place ØModel 12 : 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

Conclusion, Limits & Research agenda Both “objective” and “subjective” indices adequately capture the internalization of accessibility into house prices While the former seem to perform better from a merely statistical point of view, perceptual accessibility indices are more efficient at accounting for the heterogeneous behaviour of households with respect to space and time The different nature of centrality and accessibility is well brought out in the fact that both dimensions are shown to affect prices significantly, although perceptual accessibility indices far outweigh the centrality index This questions the prominence of the CBD as a location criterion for households

Conclusion, Limits & Research agenda While most useful a tool, O-D survey is limited by the fact that only week-day trips (Monday to Friday) are available for analysis, which has the side effect of underestimating various kinds of trips, like week-end shopping and leisure ones Much remains to be known about households’ travel as well as house location choice decisions Relating house prices to travel behaviour and accessibility to urban amenities from a space-time perspective, using for instance panel regression techniques, is certainly among the research avenues the are worth exploring