Exploring housing patterns and dynamics in low demand neighbourhoods using Geographically Weighted Regression

School of Environment& Development

Exploring the Renewal of Housing and Neighbourhood

Markets Using Geographically Weighted Regression

Geocomputation Conference, Maynooth, Ireland.

Graham Squires & Richard Kingston

University of Manchester

September 2007


Overview

1. Link to PhD Research

2. Manchester Case Study Context

3. Choice of Variables

4. Results – Regression Model Statistics

5. Results – GWR Model Statistics

6. What has GWR Added?

7. Future Application of GWR

8. Conclusion



• Overview of PhD

• Understanding renewal of housing in low value Neighbourhoods

• Understanding spatial patterns and dynamics of housing and neighbourhood change

• Using geospatial systems and local Information systems to track housing and neighbourhood change

• Generating recommendations for housing policy such as HMR (Housing Market Renewal)


• How does GWR fit into the Research

– Integrating local spatial weighting of statistics

– Add improved foresight as to what variables are influencing house price based on local influence

– Assess whether local information system data can be used in exploring spatial relationships

– Use this understanding to provide policy recommendations



2. Manchester Case Study Context


Regeneration History

• Former Industrial Manufacturing City – 19th Century Cotton Trade

• Industrial Re-structuring since 1970s– Manchester Ship Canal 3rd Largest in 1963– Unable to contain large container ships - port closure in 1982– Industrially restructuring since decline of manufacturing base

• Regeneration– Commonwealth Games (e.g. Sport City in East)– Commercial Areas in Centre (e.g. Arndale, Betham Tower)– Regenerating Residential Areas of Deprivation (e.g. Hulme)


Demographics for City of Manchester

• Population– Total resident population for Manchester of 392,819

(Census, 2001)– 9.2% decline since 1991 census– Current trend of population recovery

• Employment– High Unemployment compared to Greater Manchester Metropolitan

Area (Districts), National and Regional Averages.

• Housing Profile– 2004 Average price for all property types is at £109,426 – 2004 Tenure = 62% Private & 38% Social– 2004 Property Type Sold = Detached 4%; Semi-Detached 25%;

Terraced 43%; Flat 28%– 2004 Void rate of 5%


3. Choice of Variables• How were they Chosen

– Use of data in TNC (Tracking Neighbourhood Change)

– Domains: Crime, Education, Housing, Income

– SOA : Boundary from Census & smallest geography using all available variables. Mean population of 1500 people per SOA (Minimum of 1000)

– One year 2004 (Aggregating monthly or quarterly data)

• What Variables were chosen

– Dependent Variable (Overall Average House Prices)

– Independent Variables (23 Variables)


Dependent Variable:

Overall Average House Price – Average £


Independent / Exploratory Variables:

13 AB_2004 ASBO Count

14 BR_2004 Burglary Rate Rate

15 VC_2004 Vehicle Crime Rate

16 HPR2004 Housing Benefit %

17 PHR2004 Private Housing Benefit %

18 OSAL2004 Overall Sales Count Count

19 DPER2004 Detached Sales % %

20 TPER2004 Terraced Sales % %

21 FPER2004 Flat Sales % %

22 NPER2004 New Build Sales % %

23 SPER2004 Semi-Detached Sales % %

1 IB_2004 Incapacity Benefit %

2 IS_2004 Income Support %

3 JS_2004 Job Seekers Allowance %

4 GC_2004 GCSE Pass A* - C %

5 EN_2004 KS2 (Primary) English %

6 MA_2004 KS2 (Primary) Maths %

7 SC_2004 KS2 (Primary) Science %

8 TNP_2004 Private Tenure %

9 TNC_2004 Council Tenure %

10 TNR_2004 RSL Tenure %

11 TO3_2004 Turnover 3 Times or More %

12 VPL_2004 Long-Term Void %



• t-Values chosen rather than Parameters

– Overcome problems of combination of %, rates and counts that cannot be compared in house price values

– Can compare like-for-like values

• Using 5 (out of 23) most significant independent variable t-Values

• Note: Analysis is to Support Understanding; Not Necessarily Arguing for Causality


t-Test Dynamics of Indicators

Magnitude of influence on the Overall House Price Model for Manchester SOA 2004

Indicator t-Test

1. GCSE Results % 3.5432. Turnover 3 Times or More % 3.4633. Detached Type % of Overall 3.1204. New Build % of Overall 2.4185. Long-Term Void % -2.3056. Semi Type % of Overall 2.2647. Flat Type % of Overall 2.2108. Burglary Rate (1000 houses) -1.6789. Incapacity Benefit % -1.66510. Job Seekers Allowance % 1.61911. ASBO Counts 1.49312. Science Primary Results % 1.44013. Terraced Type % of Overall 1.24814. English Primary Results % 1.16115. Property in Receipt of Benefit % -1.14716. Maths Primary Results % -0.82617. Vehicle Crime (1000 houses) -0.28818. Council Tenure % 0.20319. RSL Tenure % 0.17820. Income Support % -0.07521. Private Tenure % 0.030

Note: Excluding Overall Sales Count & Private Tenure Housing Benefit % (Not Part of Initial non-GWR City Wide Analysis)

5 Most Significant



• 10 Colour Classifications (Green to Red)

– Bring out greater definition of spatial change

– Darker Green = Greater Positive Relation of Independent Variables with House Prices

– Darker Red = Greater Negative relation of Independent Variables with House Prices

• Interpolation by IDW (Inverse Distance Weighted)

– Interpolated Raster Maps

• Assigning values to unknown points by using values from known points

– IDW: The weight decreases as distance increases from the interpolated points

5. Results – GWR Model Statistics

1. GCSE Pass A*- C – Average % 1. GCSE Pass A*- CCoefficient: GWR Interpolation – 23 Variables

2. Turnover 3 Times or More – Average %2. Turnover 3 Times or MoreCoefficient: GWR Interpolation – 23 Variables

4. New Build Sales – Average % 4. New Build SalesCoefficient: GWR Interpolation – 23 Variables

5. Long Term Void – Average %5. Long Term VoidCoefficient: GWR Interpolation – 23 Variables


6. What has GWR Added?

• What has the GWR results added to the research?

– Better understanding of vulnerability effects

• Proximity to concentrations of positive or negative significant indicators

– Added understanding to how local geographical influence of variables can influence the housing market

– Highlight local significance of Education, Turnover, New Build and Voids on House Price



• What can be done now with the GWR results?

– Apply to other LA (Local Authorities) to compare and contrastresults

– Apply to other LIS (Local Information System) data sets to reveal how robust they are in tracking neighbourhood change

– Improvement and recommendations to policy and LA service delivery at city level

– Attempt to explain model and results to a LA audience



• What could be added to the local information system and/or my GWR model

– Add other years (compare to 2004) and show change in maps

– Health Domain (e.g. Birth Weight, Teenage Conception Rates, Mortality Ratio)

– Transport Domain (e.g. Travel to Work Patterns, Access to Public Transport, % Households without a car)

– Population Domain – Not using as standard Output Areas that dictate population (1500 residents)

– Ethnicity Domain – (e.g. International In & Out Migration, Ethnic Mix and Concentrations)


• GCSE Results averages mirror GWR influence on house price

• Turnover average high in city centre, but GWR turnover t-value in city centre has less of an association with price

• New build has a greater impact on house price outside of city centre despite large geographical area of new build in the city centre.

• Voids low on average in far North, but as voids increase this may have a greater negative effect on property prices

– Surrounding geography is more unstable / Closer proximity to greater void concentrations in the north

• Understanding vulnerability of neighbourhoods from using GWR

• Importance of case study knowledge when using GWR to understand housing and neighbourhood change

• Paradox?: GWR Local Analysis is visually represented by more general maps?

• Other Local Authorities could compare and contrast with the findings

• Further indicators and years could be added to the GWR model (and Local Information Systems) such as Health, Transport, Population, Ethnicity

8. Conclusion


Graham SquiresPlanning and LandscapeSchool of Environment and DevelopmentUniversity of ManchesterHumanities, Arthur Lewis BuildingOxford Road, ManchesterM13 9PL

Email: [email protected]@postgrad.manchester.ac.uk

Richard KingstonPlanning and LandscapeSchool of Environment and DevelopmentUniversity of ManchesterHumanities, Arthur Lewis BuildingOxford Road, ManchesterM13 9PL

Email: [email protected]

Contact Details