PhD Research Past, Present and Future

1
PhD ResearchPast, Present and Future

• Andy Turner
• http//www.geog.leeds.ac.uk/people/a.turner/

2
Outline

• Introduction
• Past (Geography of Personal Injury Road Accident
  Data in Great Britain since 1992)
• Present (Agent based modelling of daily activity)
• Future

3
Introduction

• Thanks
• I have been a researcher of the Centre for
  Computational Geography at the University of
  Leeds for nearly 10 years
• I began PhD research in 2000
• I abandoned writing up my first topic as a PhD
  thesis in 2006 and started a new PhD topic in
  October
• Spatio-temporal aspects important in both

4
Past

• A Geographical Analysis of Personal Injury Road
  Accident Data in Great Britain since 1992
• http//www.geog.leeds.ac.uk/people/a.turner/projec
  ts/phd/
• http//www.geog.leeds.ac.uk/people/a.turner/resear
  ch/interests/rag/

5
Data

• Based on STATS 19 road accident data
• 3 related tables
• Accident records
• AccID,X,Y,T
• Casualty records
• CasID, AccID, VehID
• Vehicle records
• AccID, VehID
• The data is for personal injury road accidents
• It gets better all the time

6
Aspatial Database Views
7
Pedestrian Casualty Incidence Disaggregated by
Casualty Severity Percentages of All Casualties
and Counts for Child Casualties Aged 11 to 12
Inclusive Disaggregated by Sex (1992 to 1999)
8
Temporal analysis

• What are the 'worst' times on the road
• Worst hours?
• The 50 hours with the most accidents and most
  casualties tended to be in the morning, yet the
  most fatal casualty hours tended to be in the
  evening
• Winter months dominate many views
• Worst days?
• Valentines day 1994-02-14 is favourite
• Interestingly, Valentines day in other years
  appears in the lists quite high up...
• What are the 'worst' days of the week in general?

9
Personal Injury Accident and Casualty Incidence
for 1992 to 1999 Broken Down by Severity and
Aggregated by Day of the Week (DoW) and Ordered
by Fatal Accidents (fatacc)
10
Mapping it out spatially

• Pedestrian Casualty Map
• Looks very like a population density map
• What else can help us develop a model of expected
  incidence rates?
• Road density?
• Junction Density?

11
(No Transcript)
12
(No Transcript)
13
(No Transcript)
14
(No Transcript)
15
Cluster detection

• GAM Excess based on road density
• Regression models, GAM, GEM and STAC
• Little progress in estimating the distribution of
  accidents based on independent variables
• Developing GWS

16
Measuring change over time

• GAM excess clusters in 1993 using accident
  density in 1992 as an expected measure
• This indicates where things might have got worse,
  or where reporting improved, or where damage
  claim experts got busy...

17
Change over time is hard

• The time period has to be comparable
• Is exposure to risk the same?
• No!
• But there is little data that can help...
• Are there the same number of foggy Friday
  evenings from one year to the next?
• Maybe foggy Fridays are not important when it is
  school holidays...
• Assume that the exposure to risk is about the
  same each year...

18
Can you see the pattern of change in the Map?
19
How about in the close up?
20
Clear close?
21
(No Transcript)
22
(No Transcript)
23
How about in a generalised version?
24
(No Transcript)
25
(No Transcript)
26
(No Transcript)
27
(No Transcript)
28
(No Transcript)
29
Filtering
30
So much can be done to visualise these data

• What is interesting? Where? When?
• I plan to continue this work
• BUT...
• There are many other things to do...

31
Present

• Agent Based Models of Daily Activity
• UK large scale
• Individual and household level

A conceptual model of daily activity Start
off with everybody at a home or residence.
Each person is an agent and they choose by random
an available transport mode to work. Move
people from home to work and back again on a 15
minute time step. Each day a new transport
mode is chosen based on the weather and
experience of previous journeys. Try to
settle things down to get an average journey mode
profile for aggregate census areas. Compare
this with 1991 SWS mode of transport data.
32
First attempt at a method Using CAS,
determine the numbers of people in various
occuptations and industries at a fine scale (e.g.
OAs). Use ST Theme Table TT010 from 1991
Census, which has daytme and workplace
populations at ward level. Create a model
with known transport infrastructure. Stage
1 agent model the people in different
occupations. o Move the agents to work,
such that you end up with the right number of
people in different occupations at the
workplaces. o What patterns does this
produce? Stage 2 Attempt to improve the
model using the SWS, which tells you where (in
aggregate) the people living in a given location
go to work. o Are the agents in Stage 1
going to the 'right' locations? If not are they
going to an economically/environmentally/rational
'better' location? Or is the model simply wrong!
Stage 3 Attempt further improvement on the
basis that SWS3 provides high resolution flows by
transport mode. o Are the agents using
the 'right' transport? etc... If there is
reasonable success in getting agents to go to the
right locations using the right transport a)
hurray! b) try changing the transport
infrastructure. Somewhere in all of this,
compare results to a standard stocks based
spatial interaction model.
33
Future

• A completed PhD or the start of a new topic...
• More organised open eResearch
• Publishing my outputs on-line as I go
• Finding time to work on both topics
• Collaboration

• http//www.geog.leeds.ac.uk/people/a.turner/

34
Acknowledgement/Thanks

• Thanks to Ordnance Survey, EDINA, UKDA, MIMAS and
  ONS for providing data
• Thanks to Ian Turton, Andy Evans and Oliver
  Carsten who supervised my abandoned PhD research
• Thanks to Andy Evans and Mark Birkin for
  supervising my on-going PhD research
• Thanks to the University of Leeds, particularly
  the CCG and School of Geography for your help and
  support over the years