TravelOAC: development of travel geodemographic classifications for England and Wales based on open data

Presented at GISRUK2015, University of Leeds, Wed 15th April 2015

Transcript

1. Dr  Nick  Bearman,  CGeog  (GIS)   Geographic  Data  Science  Lab   Department  of  Geography  and  Planning   TravelOAC:  development  of  travel   geodemographic  classifica9ons  for  England   and  Wales  based  on  open  data   TwiBer:  @nickbearmanuk   "Cyclists  at  red  2"  by  heb@Wikimedia  Commons  (mail)  -­‐  Own  work.  Licensed  under  CC  BY-­‐SA  3.0  via  Wikimedia  Commons  -­‐  hBp://commons.wikimedia.org/ wiki/File:Cyclists_at_red_2.jpg#/media/File:Cyclists_at_red_2.jpg   epSos.de,  hBps://www.flickr.com/photos/epsos/5591761716/  
2. Developing  a  geodemographic   classifica^on  for  travel   •  Travel   •  Geodemographics   – Variables   – Rou^ng   •  Classifica^on  &  Clusters   •  Findings   •  The  Future  
3. Background     •  Travel  is  vital   •  Many  different  factors  influence  our  choice  of   method  of  travel   •  Travel  choice  is     important   –  CO2  emissions   –  Conges^on   –  Cost  /  Time   –  Availability   –  Infrastructure  development   •  This  analysis  is  possible  due  to  big  data  analysis   and  2011  Census   "High  Five  Interchange".  Licensed  under  CC  BY  2.0  via  Wikimedia  Commons  -­‐  hBp://commons.wikimedia.org/wiki/ File:High_Five_Interchange.jpg#/media/File:High_Five_Interchange.jpg  
4. Geodemographics   •  Classifica^on  of  people  by  where  they  live   •  2001  OAC  and  2011  OAC  started  development   of  open  geodemographics   •  Openness  allows  development  of  targeted   geodemographics  and  applying   geodemographcis  to  custom  data  sets:   – Internet  (Riddlesden  and  Singleton,  2014)   – Retail  (Dolega  and  Singleton,  2014)   – Consumer  Data  (many  examples)   – Transport   Riddlesden  and  Singleton,  2014,  “Broadband  Speed  Equity:  A  New  Digital  Divide?”  Applied  Geography  52  (August):  25–33.  doi:10.1016/j.apgeog.2014.04.008.   Dolega  and  Singleton,  2014,  E-­‐Resillience  of  Bri^sh  retail  centres,hBp://geographicdatascience.com/talk/2014/12/18/regional-­‐studies/  
5. Variable  Selec^on   Domains   Concepts   Variable   Census  table  used   Demography   Gender   Gender   KS101  Usual  resident   popula^on   Age   Age  groups   KS102EW  Age   structure   Social  Class   Na^onal  Sta^s^cs  socio-­‐ economic  class   KS102EW  NS-­‐SeC   Transport   Travel  to  work   Mode  of  usual  travel  to  work   QS701EW  Method  of   travel  to  work   Ease  of  access   to  car   Car  ownership   KS404EW  Car  or  van   availability   Ease  of  access   to  public   transport   Distance  to  closest  bus/tram/ train/ferry/airport  stop   NA  (distance   calculated  from   NaPTAN  data)  
6. Distance  to  closest  transport  stop   •  NaPTAN  –  Na^onal  Public  Transport  Access   Node  database   •  Could  use  straight  line     distance  (as  the  crow  flies)   •  But  for  bus,  tram  &  rail   this  makes  liBle  sense   – Walking  routes  are  more     representa^ve  of  reality   – (Not  for  airport  or  ferry)   Walking  route  modeled  at  hBp://www.rou^no.org/  on  20150302,  Router:  Rou^no  |  Geo  Data:  ©  OpenStreetMap  contributors  |  Tiles:  ©  OpenStreetMap  
7. •  Street  network  -­‐  OpenStreetMap  &  Rou^no   – Walk  to  nearest  stop   Input:   •  Origin  -­‐  Des^na^on  -­‐  Mode   Output:  Text  file   •  Distance   •  Route   Walking  route  modeled  at  hBp://www.rou^no.org/  on  20150302,  Router:  Rou^no  |  Geo  Data:  ©  OpenStreetMap  contributors  |  Tiles:  ©  OpenStreetMap  
8. Rou^ng  Analysis   •  Run  ^me  for  this  data  analysis  (24  hours)     – for  181,408  routes  (each  OA  centroid)   – 5  transport  methods   •  Use  of  R  to  generate  and     manage  data   •  Lots  of  big  data     opportuni^es   Walking  route  modeled  at  hBp://www.rou^no.org/  on  20150302,  Router:  Rou^no  |  Geo  Data:  ©  OpenStreetMap  contributors  |  Tiles:  ©  OpenStreetMap  
9. Classifica^on   •  Variables     •  Clustergram  (Galili,  2010;     Schonlau,  2002,  2004)   •  8  clusters   •  K-­‐means  clustering   – Classifica^on   Galili,  A.T.  (2010).  Clustergram:  visualiza^on  and  diagnos^cs  for  cluster  analysis  (R  code).   Schonlau,  M.  (2002).  The  clustergram:  A  graph  for  visualizing  hierarchical  and  nonhierarchical  cluster  analyses.  Stata  J.  2,  391–402.   Schonlau,  M.  (2004).  Visualizing  non-­‐hierarchical  and  hierarchical  cluster  analyses  with  clustergrams.  Comput.  Stat.  19,  95–111.  
10. Clusters   •  8  clusters   •  Dis^nc^ve  spa^al   paBerns  and  data   paBerns  
11. •  Cartogram  to  show   classifica^on  by   OA     Clusters  
12. *For  distance,  posi^ve  values  are  higher  distances  than  average,     and  nega^ve  values  are  closer  than  average.     #1.  Higher  managerial,  administra^ve  and  professional  occupa^ons     2.  Lower  managerial,  administra^ve  and  professional  occupa^ons     3.  Intermediate  occupa^ons   4.  Small  employers  and  own  account  workers     5.  Lower  supervisory  and  technical  occupa^ons   6.  Semi-­‐rou^ne  occupa^ons   7.  Rou^ne  occupa^ons   8.  Never  worked  and  long-­‐term  unemployed   •      Cartogram  generated  using  Scapetoad,     hBp://scapetoad.choros.ch     Clusters  
13. *For  distance,  posi^ve  values  are  higher  distances  than  average,     and  nega^ve  values  are  closer  than  average.     #1.  Higher  managerial,  administra^ve  and  professional  occupa^ons     2.  Lower  managerial,  administra^ve  and  professional  occupa^ons     3.  Intermediate  occupa^ons   4.  Small  employers  and  own  account  workers     5.  Lower  supervisory  and  technical  occupa^ons   6.  Semi-­‐rou^ne  occupa^ons   7.  Rou^ne  occupa^ons   8.  Never  worked  and  long-­‐term  unemployed   Clusters  
14. *For  distance,  posi^ve  values  are  higher  distances  than  average,     and  nega^ve  values  are  closer  than  average.     #1.  Higher  managerial,  administra^ve  and  professional  occupa^ons     2.  Lower  managerial,  administra^ve  and  professional  occupa^ons     3.  Intermediate  occupa^ons   4.  Small  employers  and  own  account  workers     5.  Lower  supervisory  and  technical  occupa^ons   6.  Semi-­‐rou^ne  occupa^ons   7.  Rou^ne  occupa^ons   8.  Never  worked  and  long-­‐term  unemployed   Clusters  
15. *For  distance,  posi^ve  values  are  higher  distances  than  average,     and  nega^ve  values  are  closer  than  average.     #1.  Higher  managerial,  administra^ve  and  professional  occupa^ons     2.  Lower  managerial,  administra^ve  and  professional  occupa^ons     3.  Intermediate  occupa^ons   4.  Small  employers  and  own  account  workers     5.  Lower  supervisory  and  technical  occupa^ons   6.  Semi-­‐rou^ne  occupa^ons   7.  Rou^ne  occupa^ons   8.  Never  worked  and  long-­‐term  unemployed   Clusters  
16. *For  distance,  posi^ve  values  are  higher  distances  than  average,     and  nega^ve  values  are  closer  than  average.     #1.  Higher  managerial,  administra^ve  and  professional  occupa^ons     2.  Lower  managerial,  administra^ve  and  professional  occupa^ons     3.  Intermediate  occupa^ons   4.  Small  employers  and  own  account  workers     5.  Lower  supervisory  and  technical  occupa^ons   6.  Semi-­‐rou^ne  occupa^ons   7.  Rou^ne  occupa^ons   8.  Never  worked  and  long-­‐term  unemployed   Clusters  
17. *For  distance,  posi^ve  values  are  higher  distances  than  average,     and  nega^ve  values  are  closer  than  average.     #1.  Higher  managerial,  administra^ve  and  professional  occupa^ons     2.  Lower  managerial,  administra^ve  and  professional  occupa^ons     3.  Intermediate  occupa^ons   4.  Small  employers  and  own  account  workers     5.  Lower  supervisory  and  technical  occupa^ons   6.  Semi-­‐rou^ne  occupa^ons   7.  Rou^ne  occupa^ons   8.  Never  worked  and  long-­‐term  unemployed   Clusters  
18. Clusters   *For  distance,  posi^ve  values  are  higher  distances  than  average,     and  nega^ve  values  are  closer  than  average.     #1.  Higher  managerial,  administra^ve  and  professional  occupa^ons     2.  Lower  managerial,  administra^ve  and  professional  occupa^ons     3.  Intermediate  occupa^ons   4.  Small  employers  and  own  account  workers     5.  Lower  supervisory  and  technical  occupa^ons   6.  Semi-­‐rou^ne  occupa^ons   7.  Rou^ne  occupa^ons   8.  Never  worked  and  long-­‐term  unemployed  
19. Findings   •  Income  (based  on  NS-­‐SeC)  is  important  factor   •  As  is  gender  (related  to  income)   •  Both  related  to  SES,  but  very     limited  understanding  of  the     mechanisms  behind  SES   •  Classifica^on  –  speckly,  so     perhaps  transport  has  limited     impact  on  loca^on?  
20. What  the  results  are  useful  for   •  Understanding  transport  use  and  access   •  Do  the  two  factors  match?   •  Jus^fica^on  for  development  of  new  sta^ons  /   services   •  Applica^on  could  be     applied  to  more  refined     data  (e.g.  ^cket  sales,     usage  surveys,  etc.)  s^ll     using  the  rou^ng  element   "KingsCrossDevelopmentModel".  Licensed  under  CC  BY-­‐SA  2.0  via     Wikimedia  Commons  -­‐  hBp://commons.wikimedia.org/wiki/ File:KingsCrossDevelopmentModel.jpg#/media/   File:KingsCrossDevelopmentModel.jpg  
21. Future  developments   •  Transport  specific  geodemographic  could  be   developed   •  Extra  processing  power  allows  na^onal   analysis  of  transport  &   rou^ng  to  be  done     •  Rou^ng  allows  more   accurate  picture  of   accessibility  to  be   generated   Walking  route  modeled  at  hBp://www.rou^no.org/  on  20150302,  Router:  Rou^no  |  Geo  Data:  ©  OpenStreetMap  contributors  |  Tiles:  ©  OpenStreetMap  
22. Ques^ons?   Dr  Nick  Bearman,  CGeog  (GIS)   TwiBer:  @nickbearmanuk   Geographic  Data  Science  Lab   Department  of  Geography  and  Planning