DAVID SHINAR – Traffic Safety and Human Behaviour
STOPHER & STECHER – Travel Survey Methods Quality
and Future Directions
HENSHER & BUTTON (eds.) – Handbooks in
Transport
FULLER & SANTOS – Human Factors for Highway
Engineers
GAUDRY & LASSARE (eds.) – Structural Road Accident
Models
DAGANZO – Fundamentals of Transportation and Traffic
Operations
SECOND EDITION
United Kingdom – North America – Japan
India – Malaysia – China
First edition 2004
Second edition 2009
Reprints and permission service
Contact:
[email protected]
No part of this book may be reproduced, stored in a retrieval
system, transmitted in any
form or by any means electronic, mechanical, photocopying,
recording or otherwise
without either the prior written permission of the publisher or a
licence permitting
restricted copying issued in the UK by The Copyright Licensing
Agency and in the USA
by The Copyright Clearance Center. No responsibility is accepted
for the accuracy of
nformation contained in the text, illustrations or advertisements.
The opinions expressed
in these chapters are not necessarily those of the Editor or the
publisher.
British Library Cataloguing in Publication Data
A catalogue record for this book is available from the British
Library
ISBN: 978-1-84855-250-0
Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . xi
1.1. Purpose of the Handbook of Road Safety Measures . . . . .
. . . . . . . . . . . . . . 3
1.2. Which questions does the book answer? . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . 5
1.3. Structure of the book . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.4. Science and politics in road safety. . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . 8
2. Literature Survey and Meta-Analysis. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . 15
2.1. Systematic literature search. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . 15
2.2. Criteria for study inclusion . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . 19
2.3. Study classification . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.4. The use of meta-analysis to summarise study results . . . . .
. . . . . . . . . . . . . 20
2.5. Does a weighted mean estimate of effect make sense? . . . . .
. . . . . . . . . . . . 25
2.6. Developing accident modification functions . . . . . . . . . .
. . . . . . . . . . . . . . . . 30
2.7. Specification of accident or injury severity . . . . . . . . .
. . . . . . . . . . . . . . . . . . 32
2.8. Updated estimates of effect: Revision of the book . . . . . .
. . . . . . . . . . . . . . 33
3. Factors Contributing to Road Accidents. . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . 35
3.1. A simple conceptual framework. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 35
3.2. The scope of the road accident problem worldwide . . . . . . .
. . . . . . . . . . . . 37
3.3. Incomplete reporting in official road accident statistics. . .
. . . . . . . . . . . . . 47
3.4. Exposure: Traffic volume. . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . 53
3.5. Accident rates for different types of exposure. . . . . . . .
. . . . . . . . . . . . . . . . . 56
3.6. The mixture of road users . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . 57
3.7. A survey of some risk factors for accident involvement. . . .
. . . . . . . . . . . . 59
3.8. A survey of risk factors for injury severity . . . . . . . . .
. . . . . . . . . . . . . . . . . . 67
3.9. Assessing the relative importance of risk factors. . . . . . .
. . . . . . . . . . . . . . . 69
v
4.1. Random and systematic variation in accident counts. . . . . .
. . . . . . . . . . . . 81
4.2. The use of accident rates to measure safety. . . . . . . . . .
. . . . . . . . . . . . . . . . . 86
4.3. Explaining road accidents – the concept of cause . . . . . . .
. . . . . . . . . . . . . . 87
4.4. Road accidents as a self-regulatory problem. . . . . . . . . .
. . . . . . . . . . . . . . . . 93
5. Assessing the Quality of Evaluation Studies. . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . 99
5.1. The concept of study quality. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . 99
5.2. Assessing study quality. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . 99
5.3. The importance of study quality: Some illustrations. . . . . .
. . . . . . . . . . . . . 106
5.4. The treatment of study quality in meta-analysis . . . . . . .
. . . . . . . . . . . . . . . 113
5.5. Can the findings of road safety evaluation studies be
accounted for in
theoretical terms?. . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . 113
6. The Contribution of Research to Road Safety Policy-Making. . . .
. . . . . . . . . . . . 117
6.1. An idealised model of the policy-making process . . . . . . .
. . . . . . . . . . . . . . 117
6.2. The applicability of cost–benefit analysis. . . . . . . . . .
. . . . . . . . . . . . . . . . . . . 119
6.3. Monetary valuation of road safety in different countries. . .
. . . . . . . . . . . . 124
6.4. Current monetary valuations of impacts of road safety
measures
in Norway . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . 125
6.5. The preventability of road accident fatalities and injuries. .
. . . . . . . . . . . . 127
6.6. Vision Zero . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
130
1. Road Design and Road Equipment. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . 145
1.0. Introduction and overview of 20 measures . . . . . . . . . . .
. . . . . . . . . . . . . . . . 145
1.1. Cycle lanes and tracks. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . 155
1.2. Motorways. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
1.3. Bypasses. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
169
1.6. Roundabouts. . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
1.9. Grade-separated junctions. . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . 199
1.11. Cross-section improvements. . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . 212
vi
1.14. Reconstruction and rehabilitation of roads. . . . . . . . . .
. . . . . . . . . . . . . . . . . 248
1.15. Guardrails and crash cushions . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . 251
1.16. Game accident measures. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . 258
1.17. Horizontal curve treatments . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . 268
1.18. Road lighting. . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
272
1.20. Rest stops and service areas . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . 287
References . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
2.1. Resurfacing of roads. . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . 339
2.2. Treatment of unevenness and rut depth of the road surface. . .
. . . . . . . . . 344
2.3. Improving road surface friction . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . 348
2.4. Bright road surfaces. . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . 358
2.5. Landslide protection measures . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . 360
2.6. Winter maintenance of roads . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . 363
2.7. Winter maintenance of pavements, footpaths, cycle paths
and
other public areas. . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . 373
2.9. Traffic control at roadwork sites . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . 380
References . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 385
3.1. Area-wide traffic calming. . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . 403
3.2. Environmental streets . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . 408
3.3. Pedestrian streets . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 412
3.5. Access control. . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 419
3.6. Priority control. . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423
3.9. Traffic signal control at junctions . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . 433
3.10. Signalised pedestrian crossings. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . 440
3.11. Speed limits. . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
445
3.12. Speed-reducing devices . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . 452
3.13. Road markings. . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 458
vii
3.16. One-way streets. . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 479
3.18. Bus lanes and bus stop design. . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . 487
3.19. Dynamic route guidance. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . 492
3.20. Variable message signs . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . 495
3.21. Protecting railway–highway level crossings. . . . . . . . . .
. . . . . . . . . . . . . . . . . 499
3.22. Environmental zones. . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . 504
4.0. Introduction and overview of 29 measures. . . . . . . . . . .
. . . . . . . . . . . . . . . . 543
4.1. Tyre tread depth . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . 550
4.2. Studded tyres . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 554
4.4. High-mounted stop lamps. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . 564
4.5. Daytime running lights for cars. . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . 567
4.6. Daytime running lights for mopeds and motorcycles. . . . . . .
. . . . . . . . . . . 571
4.7. Improving vehicle headlights . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . 574
4.8. Reflective materials and protective clothing. . . . . . . . .
. . . . . . . . . . . . . . . . . 582
4.9. Steering, suspension and vehicle stability . . . . . . . . . .
. . . . . . . . . . . . . . . . . . 586
4.10. Bicycle helmets. . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 591
4.11. Motorcycle helmets. . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . 596
4.13. Child restraints. . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 609
4.15. Seat belts in buses and trucks . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . 624
4.16. Vehicle crashworthiness. . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . 627
4.18. Intelligent cruise control . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . 639
4.19. Regulating vehicle mass (weight). . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . 642
4.20. Regulating automobile engine capacity (motor power) and
top speed. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 649
motorcycles. . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . 656
4.24. Moped and motorcycle equipment. . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . 668
4.25. Bicycle safety equipment. . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . 671
4.26. Safety standards for trailers and caravans . . . . . . . . .
. . . . . . . . . . . . . . . . . . 676
viii
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . 690
5.0. Introduction and overview of four measures. . . . . . . . . .
. . . . . . . . . . . . . . . . 733
5.1. Vehicle safety standards. . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . 737
5.2. Periodic motor vehicle inspections. . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . 742
5.3. Roadside vehicle inspections. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . 749
5.4. Garage regulation and inspections. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 753
References . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 755
6.0. Introduction and overview of 12 measures. . . . . . . . . . .
. . . . . . . . . . . . . . . . 759
6.1. Driving licence age limits . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . 763
6.2. Health requirements for drivers. . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . 771
6.3. Driver performance standards. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . 779
6.4. Basic driver training. . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . 785
6.5. The driving test. . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 793
6.6. Training and testing of moped and motorcycle riders. . . . . .
. . . . . . . . . . . 797
6.7. Training and testing of professional drivers. . . . . . . . .
. . . . . . . . . . . . . . . . . 802
6.8. Graduated driving licences (GDLs) . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . 806
6.9. Motivation and incentive systems in the work place . . . . . .
. . . . . . . . . . . . 815
6.10. Regulation of driving and rest hours. . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . 817
6.11. Safety standards for emergency driving. . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . 827
6.12. Safety standards for school transport . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . 833
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . 839
7.0. Introduction and overview of three measures. . . . . . . . . .
. . . . . . . . . . . . . . . 859
7.1. Education of pre-school children (0–6 years) . . . . . . . . .
. . . . . . . . . . . . . . . . 862
7.2. Education in schools (6–18 years old). . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . 865
7.3. Road user information and campaigns. . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . 867
References . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 873
8.0. Introduction and overview of 13 measures. . . . . . . . . . .
. . . . . . . . . . . . . . . . 879
8.1. Stationary and manual speed enforcement. . . . . . . . . . . .
. . . . . . . . . . . . . . . 885
8.2. Automatic speed enforcement: Speed cameras. . . . . . . . . .
. . . . . . . . . . . . . . 889
ix
8.4. Patrolling. . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
899
8.7. Fixed penalties. . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 913
8.8. DUI legislation . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 916
8.9. DUI enforcement . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . 930
8.12. Fines and imprisonment . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . 945
8.13. Motor vehicle insurance . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . 949
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . 955
9.1. Emergency medical services. . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . 983
9.2. Rescue helicopters . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . 990
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 998
10.0. Introduction and overview of 13 measures. . . . . . . . . . .
. . . . . . . . . . . . . . . . 1005
10.1. Organisational measures . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . 1012
10.3. Quantified road safety targets and road safety programmes . .
. . . . . . . . . 1020
10.4. Safe community programmes. . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . 1023
10.5. Exposure control. . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . 1026
10.7. Road plans and road construction. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . 1039
10.8. Road safety audits and inspections. . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 1043
10.9. Motor vehicle taxation. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . 1048
10.10. Road pricing. . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1053
10.12. Road traffic legislation. . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . 1069
10.13. Regulating commercial transport . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . 1075
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . 1079
Definitions of Technical Terms. . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . 1095
List of Abbreviations. . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1115
PREFACE
The second, revised edition of The Handbook of Road
Safety Measures, first published
by Elsevier Science in 2004, gives a systematic overview of current
knowledge regarding
the effects of road safety measures. The book gives
state-of-the-art summaries of
current knowledge regarding the effects of 128 road safety
measures. Since 2004, the
introduction part and 65 chapters have been revised and 5 chapters
have been added.
Easily accessible knowledge on how to prevent traffic injury is in
increasing demand, as the
number of people killed or injured in road accidents continues to
grow on a global basis.
It is hoped that this book may serve as a reference manual for road
safety professionals
in every country. The 2004 edition of the book was published in
Spanish in 2006.
The book is based on the Norwegian edition of the book, first
published in 1982 and
continuously updated and expanded since 2001. Work on this book
started as far back
as 1980. During the whole period from 1980 until now, the endeavour
to develop and
update the book has been funded by the Norwegian Ministry of
Transport and
Communications and the Norwegian Public Roads Administration. In
recent years, the
Swedish Road Administration has been an important contributor as
well. The Institute
of Transport Economics (TØI) would like to thank these institutions
for their financial
support and their long-term commitment to this research effort.
Without the original
Norwegian edition, the current English version could never have
been produced.
The present edition is the result of the coordinated effort of
Chief Research Officer
Rune Elvik and researchers Alena Høye, Truls Vaa and Michael
Sørensen – all
belonging to the Institute of Transport Economics. The final
preparation of the
manuscript for publication was made by Unni Wettergreen. The points
of view
expressed in the book are those of the authors and do not
necessarily reflect the
positions of the funding agencies. Errors and omissions, if any,
are the sole
responsibility of the authors.
BACKGROUND AND GUIDE TO READERS
1.1 PURPOSE OF THE H ANDBOOK OF ROAD
S AFETY M EASURES
As the title of this book is Handbook of Road Safety
Measures, most readers will
perhaps expect a handbook to give instructions or advice concerning
its main topic, but
not all readers will expect the same kind of instructions or
advice. It is therefore
appropriate to start the book by describing its background and
purpose.
Although this book is called a ‘handbook’, it does not provide any
instructions or
advice of a general nature with respect to how best to design or
implement road safety
measures. The term ‘handbook’ rather denotes a reference manual, a
catalogue or an
encyclopaedia of road safety measures.
Why is this book written and what is its main purpose? The book is
written in order to
summarise and present in an easily accessible form what is
currently known about the
effects of road safety measures. A road safety measure is any
technical device or
programme that has improving road safety as the only objective or
at least one of its
stated objectives. Road safety measures may be directed at any
element of the road
system: patterns of land use, the road itself, road furniture,
traffic control devices,
motor vehicles, police enforcement and road users and their
behaviour.
This book takes a broad view of what constitutes a road safety
measure. It is not
limited to a particular class of safety programmes, but tries to
cover everything that is
intended to improve road safety. A total of 128 road safety
measures are included.
Improving road safety is, unfortunately, not a concept that has a
standard scientific
definition. In this book, it refers to a reduction in the expected
number of accidents, a
The Handbook of Road Safety Measures
Copyright r 2009 by Emerald Group Publishing
Limited
All rights of reproduction in any form reserved
ISBN: 978-1-84855-250-0
reduction in accident or injury severity or a reduction in the rate
of accidents or injuries
per kilometre of travel.
The main purpose of the book is to describe, as objectively as
possible, the effects of
road safety measures on road safety. Some road safety measures
influence not only
road safety but also the ease of travel and the quality of the
environment. Ease of travel
is a broad concept that includes aspects such as accessibility (the
availability of a
certain destination for travel), out-of-pocket expenses (like motor
vehicle operating
costs) and travel time. In this book, the term mobility is used to
denote the ease of
travel in terms of accessibility, cost and travel time.
Environmental impacts of road
safety measures refer primarily to impacts on traffic noise and air
pollution, but in
some cases, other impacts are briefly mentioned, for example,
impacts on the working
conditions of professional drivers.
Some of the terms that have been used to describe the contents of
this book, such as
‘current knowledge’ and ‘objective description’, require a more
extensive discussion.
This will be undertaken in later chapters of Part I (in particular,
Chapters 4 and 5).
Before describing the main questions, the book tries to answer, its
structure and the
role of research in promoting road safety, what this book is not
intended to be needs to
be explained.
This book is not a technical design handbook. It does not tell
readers how to design a
junction or how to build a car. This book does not offer a
prescription for road safety
policy. It does not tell readers which road safety measures ought
to be taken, nor does
it instruct policymakers in how to set priorities for the provision
of road safety. Section
1.4 outlines how the line separating road safety research from
road safety policymaking
is understood in the book.
This book does not tell you how to do road safety research;
however, it tries to assess
systematically the quality of current knowledge about the effects
of road safety
measures. In doing so, this book of course invokes widely accepted
standards of
technical rigour and quality in applied research. However,
assessing the quality of what
is known is not the same thing as instructing researchers about how
to improve
knowledge.
This book does not tell readers how to set up an accident recording
system or how to
investigate accidents, but discusses the concept of accident
causation and briefly
summarises what is known about factors that contribute to road
accidents. Although
this presentation may perhaps give readers some ideas about what
they should be
looking for when trying to find out why road accidents happen, it
is highly deficient in
acting a guide as to how best to investigate and record road
accidents.
4 The Handbook of Road Safety Measures
Some readers may take exception to the consistent use of the word
‘accident’ in the
book, preferring perhaps other words like crash or unintentional
injury event (Langley
1988). Hopefully, these readers will not be deterred from using the
book. Some of the
arguments for not using the word ‘accident’ are, we believe, based
on misunderstand-
ing. For example, it has been argued that the word ‘accident’ has
traditionally been
used to represent events that occur at random, and which are
therefore unpreventable.
This point of view is both correct and incorrect. It is correct in
that there is an element
of randomness in accident occurrence. However, the occurrence of
accidents is never
entirely random. Young male drivers are systematically
over-involved in road
accidents. The gender and age of drivers involved in road accidents
are, therefore,
not entirely a matter of chance. On the contrary, the occurrence of
a specific road
accident is random in the absolute sense that if it could have been
accurately predicted,
it would not have happened (assuming that accidents are not
deliberate; that nobody
wants to become involved in an accident).
Part of the nature of random events is that the precise time and
place of their occurrence,
as well as the precise nature of their impacts, are unpredictable.
But unpredictability in
this sense does not necessarily imply un-preventability. To
illustrate this, imagine a 100-
km-long road, chopped up into 100 consecutive 1-km sections. The
number of accidents
recorded on each 1-km section is counted, and the distribution of
accident counts among
the 100 sections is found to closely follow the Poisson probability
law, which means that
accident occurrence in these 100 road sections is random in the
sense that it is not
statistically possible to identify one road section that has a
higher expected number of
accidents than any other road section. Yet it hardly follows from
this observation that
the accidents occurring along the 100-km road cannot be prevented.
Suppose, for
example, that all drivers using the road slowed down by 10 km per
hour. It is very likely
that there would then be fewer accidents. Or, suppose road lighting
is installed along the
road. Again, it is likely that there would be a reduction in the
number of accidents.
‘Accident’ is the right word for a road crash, precisely because it
connotes randomness.
It is a matter of fact that there is a large, but not always
dominant, element of
randomness in accident occurrence. It is, however, a serious
misunderstanding to
suggest that randomness as such implies that accidents cannot be
prevented.
1.2 WHICH QUESTIONS DOES THE BOOK ANSWER?
This book provides answers to the following questions:
Which measures can be used to reduce the number of traffic
accidents or the
severity of injury in such accidents?
Part I: 1. Background and Guide to Readers 5
Which accident problems and types of injury are affected by
the different measures? What effects on accidents and
injuries do the various road safety measures have
according to international research? What effects do the
measures have on mobility and the environment? What are the
costs of road safety measures? Is it possible to make
cost–benefit evaluations of the measures? Which measures
give the greatest benefits for traffic safety seen in relation to
the
cost of the measures?
Not all these questions are equally easy to answer, and it is not
always possible to give a
precise or conclusive answer. For example, the effect of a measure
on accidents may
vary from place to place, depending on the design of the measure,
the number of
accidents at the spot, any other measures that have been
implemented, etc. As a result,
different studies of the same measure may provide different
conclusions. An attempt
has been made to identify sources of variation in study findings
and to try to form as
homogeneous groups as possible when presenting estimates of the
effects of measures
on road safety. This will be discussed more detail in Chapter
2.
1.3 STRUCTURE OF THE BOOK
The book consists of three parts, each of which can be read
independently. The
chapters in each part are also designed to be read
independently.
Part I describes the purpose of the book and its structure,
the method used in surveying
and analysing the literature the book is based on, factors
contributing to road
accidents, basic concepts of road safety research, the quality of
road safety evaluation
research and scientific approaches to planning and
policymaking.
Part II describes road safety measures in 10 different areas.
Within each area, a number
of different types of measures are described in individual
sections. The 10 areas are
1. Road design and road equipment (20 measures)
2. Road maintenance (9 measures)
3. Traffic control (21 measures)
4. Vehicle design and protective devices (29 measures)
5. Vehicle and garage inspection (4 measures)
6. Driver training and regulation of professional drivers (12
measures)
7. Public education and information (4 measures)
8. Police enforcement and sanctions (13 measures)
6 The Handbook of Road Safety Measures
10. General purpose policy instruments (14 measures).
Part III contains a glossary of words, symbols and
abbreviations, which are used in the
book and a subject index.
In Part II, each chapter and each of the sections within each
chapter has been written
following the same structure. The first section in each chapter
gives an overview of the
amount of research available and summaries of the effects on
accidents, environment
and mobility, as well as an overview of costs and cost–benefit
analyses. The sections
that described specific types of road safety measures all consist
of the same subsections,
a short description of which is given in the following.
Problem and objective. This section describes the road safety
problem, which the
measure is designed to solve or reduce. A road safety problem can
be described in terms
of a high number of accidents, a high accident rate or a high
proportion of serious
injuries. For example, it is widely seen as a problem that
pedestrians and cyclists are
more often involved in injury accidents per kilometre travelled in
traffic than car
occupants, and that they tend to be more seriously injured than car
occupants when
involved in an accident. As far as possible, the size of the road
safety problem which
each measure is intended to affect is shown by means of accident
figures or estimates of
risk. However, not all road safety problems can be described
exhaustively in numerical
terms only. This applies, for example, to the feeling of insecurity
that some road users
experience.
Many road safety measures are intended to tackle local problems,
having a fairly
clearly limited scope in time and space. However, this does not
apply to all measures.
Some measures are directed towards more general problems, which may
affect all road
users and all places. In such cases, it is difficult to state
precisely the number and nature
of accidents which these measures are designed to affect. For some
road safety
measures, the concept of ‘target accidents’ is thus somewhat ill
defined (Hauer 1997).
Description of the measure. This section gives information
concerning the design of a
road safety measure and its intended function. Detailed technical
descriptions are not
given. Illustrations showing the measure are given in some
cases.
Effect on accidents. This section deals with the effects on
accidents, or on the severity of
injury in accidents, which have been found in research. Whenever
possible, effects are
stated in terms of the percentage change of the number of accidents
or injuries
attributable to a certain measure. All estimates of effect
presented in this book are
uncertain. The most important sources of such uncertainty are
identified for each
Part I: 1. Background and Guide to Readers 7
measure. Statistical uncertainty is stated in terms of a 95%
confidence interval for the
estimate of effect. For measures where no studies have been found
that quantify effects
on road safety, the effect is described in other ways.
Effect on mobility. In addition to the effect on accidents and
injuries, many road safety
measures also have effects on mobility. These impacts are briefly
described, but not in
as great detail as safety effects.
Effects on the environment. Effects on the environment are briefly
described. Such
effects include traffic noise and air pollution in a wide sense.
Major incursions into the
landscape and changes in land use should also be regarded as
important environmental
effects.
Costs. For the majority of measures, information is given regarding
the cost of the
measure. The information is taken partly from official budgets and
accounts, partly
from research reports and partly from producers or dealers in
safety equipment. Good
estimates of cost have not always been found. The cost figures
presented are usually an
estimate of the average cost for a ‘unit’ of a measure, for
example, 1 km of track for
walking and cycling, one roundabout, one signalised junction, one
seat belt, one set of
ABS brakes, etc. In addition, total costs are presented for
measures whose extent of
usage is sufficiently well known.
Cost–benefit analysis. Examples are given of cost–benefit analysis
of most measures.
It is important to bear in mind that the results of cost–benefit
analyses depend
strongly on the context to which they refer. Monetary valuations of
impacts, which are
a key element of cost–benefit analysis, vary substantially between
countries. As
a rule, one would therefore not expect the results of cost–benefit
analyses made in
one country to apply directly to another country. The context to
which most of the
analyses presented refer is the current situation in Norway.
However, where cost–
benefit analyses have been reported in other countries, they are
quoted. The
applicability of cost–benefit analyses to road safety measures is
discussed in detail in
Chapter 6 of Part I.
1.4 SCIENCE AND POLITICS IN ROAD SAFETY
Road safety research, in particular road safety evaluation
research, is highly applied.
This type of research is carried out mostly to help reduce the
number of road accidents
and the injuries resulting from them. Can science and politics be
kept apart in such a
highly applied field of research? Where is the dividing line
between science and politics
in road safety?
A distinction can be made between three types of issues that arise
in policymaking. The
three types of issues can be stated in the following terms:
Normative: A is a good thing (or the right thing to do).
Empirical: If action B is performed, A will be produced.
Prescriptive: Therefore, we ought to take action B.
Normative issues are about deciding what we think is good or right
and are ultimately
matters of moral judgement. Most people would probably agree that
reducing traffic
injury is a good outcome. Hence, most people would probably also
endorse a policy
objective stating that traffic injuries should be reduced.
Formulating the ideals and objectives that policy should strive to
realise clearly lies
within the realm of politics rather than science. Policy objectives
represent human value
systems and seek to articulate these in an attractive way. Does
this mean that science
has nothing to say about normative issues? No. A scientific
evaluation of the solutions
proposed to normative issues can be made by relying on principles
of logical
consistency. For example, a policy objective stating that every
road user has the right
to safer travel than the average risk faced by road users can be
rejected as logically
inconsistent, since it is impossible for everyone to be safer than
average.
A broader scientific analysis of human value systems belongs to
ethics and moral
philosophy, and is outside the scope of this book. The main topic
of road safety
evaluation research is to determine whether road safety measures
are effective in
improving road safety. This is entirely an empirical issue.
It was stated in Section 1.1 that this book describes, as
objectively as possible, what is
known about the effects of road safety measures, in particular
their effects on road
safety. What does this statement mean? How can any description of
knowledge claim
to be objective? Objectivity is not something that can be
meaningfully measured in
numerical terms. It is, however, an ideal of science to which this
book strives by
seeking to present objective knowledge about the effects of
road safety measures, assessing knowledge according to
standards of validity that are independent of the
content of that knowledge, and depend solely on how it was
produced, and refraining from advocacy.
Let us elaborate on each of these points.
Objective knowledge. In discussing what we mean by scientific
knowledge, epistemology
has traditionally relied on a subjective conception of knowledge,
in which knowledge is
Part I: 1. Background and Guide to Readers 9
defined as justified true belief. Within this framework, knowledge
cannot exist without
a knowing subject. In short, a justified and true statement does
not constitute
knowledge unless someone is aware of the statement and believes
it.
This conception of knowledge lies close to everyday usage of the
term. Hauer, for
example, in discussing the state of knowledge with respect to the
effects of road safety
measures, states (1988, 3): ‘My own critical views about the amount
of factual
knowledge that is available in the field of road safety delivery
rest on years of study. As
I moved from one inquiry to another and began to notice how shallow
are the
foundations of what passes for knowledge, I gradually realized that
ignorance about
the safety repercussions of the many common measures is not the
exception.’ Three
years later, he remarked (Hauer 1991, 135): ‘How little we know
about the safety
consequences of our road design decisions and about the
repercussions of our traffic
control actions is simple to demonstrate. One needs only to ask the
engineer:
‘‘Approximately how many accidents per year do you expect to occur
with design X?’’
While the engineer might venture an opinion, in truth, the arsenal
of knowledge at the
disposal of the North American engineer just does not suffice to
give an answer.’
While conforming both to everyday usage and the traditions of
epistemology, the
subjective concept of knowledge creates a number of difficulties.
Although it makes
sense to say that person A knows more about a subject than person
B, if person A can
pass a more difficult examination about the subject than person B,
it hardly makes
sense to say that the amount of knowledge that is available to the
general public
concerning a subject is determined primarily by how much person A
can remember
when undergoing an examination in the subject.
Karl Popper introduced the concept of objective knowledge (Popper
1979), which he
defines (1979, 73) as ‘the logical content of our theories,
conjectures, guesses’. He adds
that ‘Examples of objective knowledge are theories published in
journals and books
and stored in libraries; discussions of such theories; difficulties
or problems pointed out
in connection with such theories, and so on.’ Knowledge in the
objective sense,
according to Popper (1979, 109), is knowledge without a
knower; it is knowledge
without a knowing subject.
In short, the concept of objective knowledge can be defined as all
results of research,
theoretical or empirical, that are available to the general public
by virtue of being
written or otherwise stored in a medium that is accessible to
anyone who wants to learn
its contents. Knowledge in this sense exists, as pointed out by
Popper, in the shelves of
libraries and archives. This kind of knowledge is objective in the
sense that it exists
irrespective of whether anyone keeps it inside his or her head. It
is, however, not
necessarily objective in the sense that everyone who reads a
certain paper in a journal
10 The Handbook of Road Safety Measures
will find the results reported in the paper convincing and
therefore believe them, as
required according to the subjective conception of knowledge.
This book seeks to develop objective knowledge about the effects of
road safety
measures by relying on an extensive and systematic search of the
literature, described in
detail in Chapter 2, and by summarising this literature by means of
formal techniques
of meta-analysis that minimise the contribution of subjective
factors that are endemic
in traditional, narrative literature surveys.
Assessing the validity of knowledge. Can the results of road safety
evaluation studies be
trusted? Do these studies always show the true effects on road
safety of the measures
that have been evaluated? Regrettably, the answer to these
questions is no. Hauer
(2002, 3) laments: ‘By publishing many biased accounts on a
variety of treatment, all
giving inflated estimates of safety effect, one creates an entirely
incorrect lore about
what is achievable. . . . The publication of incorrect results is
like the release of toxin
into a pristine body of water. It does not take much to make an
entire lake unfit for
drinking. . . . The remedy to knowledge pollution is not reader
education. While it is
useful to educate potential readers to assess critically the
results of safety studies, it is
too much to hope that reader education can undo the damage done by
publishing
poorly done research.’ In this book, a systematic framework has
been used to assess the
validity of the studies that are quoted. This framework applies to
published or at least
written studies, and not to oral communications, personal beliefs,
tacit knowledge or
other forms of subjective knowledge.
Checking studies according to a set of criteria of validity
may be regarded as an overly
restrictive and simplistic way of assessing the validity of
knowledge. Three points can
be made in defence of this approach. First, the set of criteria for
assessing the validity
of evaluation studies are intended as normative criteria,
not as descriptive criteria. All
too often, controversies about research revolve around the contents
of the results,
rather than the methodological rigour of the research, and are
heavily influenced by
vested interests, rather than a disinterested search for the truth
(see Crossen 1994, for
some striking examples of these tendencies).
Second, it is conceded that a set of normative criteria is bound to
be incomplete, in the
sense that it does not exhaust the considerations that are regarded
as relevant in
assessing the validity of studies. Some considerations about study
quality may apply
just to one particular study and are thus not easily stated
in general terms.
Third, while an informal and subjective assessment of the validity
of research can
reflect considerations that are difficult to formalise, it is
nevertheless likely to be subject
to more or less unknown biases. No matter how hard we try to be
objective, there is
Part I: 1. Background and Guide to Readers 11
always a risk that we go by the rule that ‘bad studies are . . .
those whose results
we do not like’ (Rosenthal 1991, 130). By assessing validity in
terms of formally
stated, normative criteria, the role of personal prejudices in the
assessment can be
minimised.
Refraining from advocacy. Suppose an effective remedy for road
accidents is found.
Surely that is good news. Let us apply the remedy at once. Advocacy
in research
reports refers to statements recommending or calling for the use of
specific road safety
measures. To offer policy recommendations is to engage in advocacy.
While advocacy
may be tempting to many researchers (‘Hey look, I’ve found a
wonderful solution to an
important social problem! Please give me some applause’), it is a
temptation that
should be resisted. Let us explain why.
In the first place, advocacy will, at least in the long term,
undermine the confidence in
research. Many road safety measures are controversial. The fact
that a certain road
safety measure is effective does not always mean that people like
it. A researcher who
has repeatedly advocated lower speed limits to improve road safety
will find his
credibility greatly reduced next time he publishes a study that,
once again, concludes
that lowering speed limits is an effective way of improving road
safety.
In the second place, there is nearly always more than one way of
improving road safety.
Treatment A may be effective for a particular accident problem, but
so are treatments
B, C, D, E and F. To choose between these treatments, policymakers
need to know
more than simply the fact that they are all likely to reduce the
number of accidents.
Perhaps costs differ greatly. Perhaps the impacts on mobility and
the environment are
different. Perhaps public opposition is strong to three of the
measures, but not to the
other three. And so on. In short, making road safety policy
involves complex trade-offs
that tend to be overlooked by those who advocate a particular road
safety measure.
In the third place, to advocate something one should really be sure
that it works. If
knowledge is not firmly established, one can get nasty surprises
when introducing a
treatment that was erroneously believed to be effective.
Unfortunately, knowledge
about the effects of road safety measures is not always very firmly
established.
Some readers may object to these arguments by saying that this book
offers covert
policy recommendations by presenting cost–benefit analyses of the
road safety
measures it covers. However, a cost–benefit analysis is not a
policy recommendation.
It is simply a way of showing, in terms of a common scale, the
relative importance of
various impacts of a programme. Trying to identify the practical
implications of a
cost–benefit analysis is not as straightforward as some people
think. It is not the case
that an action should always be adopted if the benefits of that
action are greater than
12 The Handbook of Road Safety Measures
its costs, and should never be adopted if the costs are greater
than benefits. This point is
made in virtually every textbook on cost–benefit analysis.
Moreover, it is not obvious
that road safety policy can or ought to be based slavishly on the
results of cost–benefit
analyses. To determine the weight that cost–benefit analysis should
carry in road safety
policy requires judgements that must be made outside the framework
of cost–benefit
analysis, and are not part of the analysis as such.
Part I: 1. Background and Guide to Readers 13
LITERATURE SURVEY AND META-ANALYSIS
2.1 SYSTEMATIC LITERATURE SEARCH
A comprehensive survey of studies evaluating the effects of road
safety measures has
been made. These studies have been identified by means of a
systematic literature
search. This section describes how the literature search was
done.
The literature search consists of a ‘fixed’ part and a ‘variable’
part. The fixed part is a
comprehensive search for studies in a sample of sources. The
variable part is based on
the results of the fixed part of the search. This approach is
sometimes referred to as the
ancestry approach. The fixed part of the literature search is a
systematic survey of the
following main groups of sources:
Previous Norwegian editions of Handbook of Road
Safety Measures Scientific journals Reports issued by
selected research institutes Conference proceedings from a
sample of regular conferences The library of the Institute
of Transport Economics Bibliographical databases.
The variable part of the literature search comprises references
found in studies that
were retrieved from these sources.
Previous Norwegian editions of Handbook of Road Safety Measures.
Previous editions
of this book have been published in Norwegian and in English. The
previous editions
The Handbook of Road Safety Measures
Copyright r 2009 by Emerald Group Publishing
Limited
All rights of reproduction in any form reserved
ISBN: 978-1-84855-250-0
of the book (Pedersen, Elvik and Berard-Andersen 1982, Elvik,
Vaa and Østvik 1989,
Elvik, Mysen and Vaa 1997, Elvik and Vaa 2004) have been
examined, and we have
tried to obtain studies to which references were made. No studies
that have been
referred to in the earlier editions of the book have been omitted.
Even though the first
edition of the book refers to many studies that by now are
relatively old (over 30 years),
none of these studies have been omitted. There are two main reasons
for this. First, by
keeping old studies, one has the opportunity of finding whether new
and old studies
reach the same conclusions. Second, the research is cumulative.
This means that new
studies are based on and add to the results of older studies, but
attempt to refine,
confirm, falsify, or develop these results by replicating studies
or by applying better
research methods.
Scientific journals. A number of scientific journals has been
hand-searched and relevant
papers have been identified. Table 2.1 shows the journals
that have been searched and
the volumes included for each journal.
The journals that were judged to be the most important have been
examined from
around 1970 or from the first published volume. Less important
journals have been
searched from 1980. Highway Research Record ceased
publication in 1974 and was
replaced by Transportation Research Record .
Reports issued by research institutes. Reports issued by a number
of research
institutions and public agencies in different countries have been
searched. Table 2.2
shows the institutions whose publications have been systematically
surveyed in the
literature search.
Volumes included for the different series of reports issued by
these institutions largely
cover the period for which the report series in question has been
in existence. For
report series that were regarded as less important, only volumes
from after 1980 have
been studied.
Conference proceedings. Every year, or at other fixed intervals, a
number of
international conferences or seminars are held that deal with the
questions of road
safety. Normally, conference proceedings, which contain the
contributions to these
conferences, are published. For conferences that are held
regularly, the proceedings
from conferences in recent years have been searched
systematically. Table 2.3 shows
the conferences concerned.
In addition to these regular conferences, a number of other
conferences are held.
Proceedings of these conferences have been obtained if there was
reason to believe they
might contain relevant papers.
Literature search in the library of the Institute of Transport
Economics. Literature
searches have been made in the library of the Institute of
Transport Economics using
subject words. These searches were done on a supplementary basis,
designed to identify
studies that were not found in the other sources that were searched
systematically.
Bibliographical databases. Literature searches have been carried
out using several
international bibliographical databases. These are ROADLINE at VTI
(Swedish Road
Table 2.1: Scientific journals surveyed as part of the literature
search
Journal Volumes included
Australian Road Research (ceased publication in 1991) 1970–91
Dansk Vejtidsskrift (Danish Road Journal) 1980–
Ergonomics 1980–
Human Factors 1980–
IATSS Research 1980–
Journal of Risk and Uncertainty 1988–
Journal of Safety Research 1969–
Journal of Traffic Medicine 1974–
Journal of Transport Economics and Policy 1970–
Journal of Transportation Engineering 1970–
Nordic Road and Transport Research 1989–
NTR-nytt (News from Nordic Research) 1992–
Policy Sciences 1980–
Public Roads 1980–
Risk Analysis 1981–
Strassenverkehrstechnik 1980–
Trafikken og Vi 1970–
Transportation Research (series C) 1993–
Traffic Injury Prevention 1999–
Zeitschrift fur Verkehrssicherheit 1970–
literature survey
Beratungsstelle fur Unfallverhuting (BFU, Switzerland) 1980–
Bundesanstalt fur Strassenwesen (BASt, Germany) 1974–
Danmarks Transportforskning (DTF) 2001–
Lunds Tekniske Høgskole (Lund Institute of Technology, Sweden)
1977–
Nordisk Ministerrad (Nordic Council of Ministers, Nordic countries)
1973–
Nordisk Vegteknisk Forbund (NVF, Nordic Road Federation, Nordic
countr ies) 1970–
Organization of Economic Cooperation and Development (OECD)
1970–
Radet for Trafiksikkerhedsforskning (Danish Council for Road Safety
Research, Denmark) 1969–2001
SINTEF Samferdselsteknikk/NTH Samferdselsteknikk (Norwegian
Institute of Technology, Norway) 1975–
Society of Automotive Engineers (SAE, USA) 1980–
Statens vegvesen (Public Roads Administration, Norway) 1980–
Statens Vag- och Trafikinstitut (VTI, Swedish Road and Transport
Research Institute, Sweden) 1975–
SWOV (Institute for Road Safety Research, The Netherlands)
1970–
TØI (Institute of Transport Economics, Norway) 1963–
Transport Research Laboratory (TRL, TRRL, RRL, Great Britain)
1965–
US Department of Transportation (USA) 1980–
US Transportation Research Board (TRB, USA) 1960–
Vejdirektoratet (Public Roads Administration, Denmark) 1980–
Vagverket (National Roads Administration, Sweden 1980–
Table 2.3: Conference proceedings which have been studied as part
of the
literature search
PTRC Summer Annual Conference (now: European Transport Forum,
annual) 1985–
Road Safety in Europe (VTI et al.) (every second year) 1985–
Road Safety on Four Continents (VTI and TRB) (every second year)
1985–
TRB Annual Meeting (annual) 1985–
VTI/TFBs Research Days (annual) 1989–
18 The Handbook of Road Safety Measures
and Transport Research Institute), OECD’s database IRRD, the
database TRANS-
PORT (Silverplatter), Sciencedirect (the online database from
Elsevier), PubMed (of
the US National Library of Medicine) and the Cochrane
Library.
A large number of road safety evaluation studies have been found in
the sources listed
above. Many of these studies refer to other studies, which were
obtained if the
references appeared to be relevant. Relevance was judged according
to study titles and
abstracts (if available). This approach to searching the literature
does not guarantee
100% coverage. We do believe, however, that we have retrieved a
large proportion of
the best road safety evaluation research that has been
published.
2.2 CRITERIA FOR STUDY INCLUSION
The main objective of the literature search was to find studies
that have quantified, or
at least have tried to quantify, the effect of one or more road
safety measures on the
number of accidents, accident rate and the number of injuries or
risk of injuries.
Studies that have evaluated the effects of road safety measures by
relying on proxy
measures for safety, such as conflicts between road users or
changes in road user
behaviour, rather than accidents or injuries, are less relevant.
One reason for this is the
fact that for many forms of behaviour, the relationship to accident
occurrence is
unknown. Another reason is that the ultimate objective of all road
safety measures is to
reduce the expected number of accidents or injury severity.
This does not mean that measurements of road user behaviour, for
example, are not of
interest. On the contrary, they can make a study more valuable by
supplementing
accident records. For example, the validity of a study is greater
if it describes changes
in both speed and accidents – and shows that these changes are
closely related to each
other – than if an otherwise similar study provides information
only on speed or
accidents by itself.
2.3 STUDY CLASSIFICATION
Studies have been classified according to the road safety measure
whose effects they
have evaluated. Some studies have evaluated several measures and
are therefore
included for each of the measures evaluated. However, the majority
of studies evaluate
the effects of just one road safety measure.
It has traditionally been regarded as a strength if a study tried
to evaluate the effects of
a particular road safety measure. However, as far as road safety
policy is concerned,
Part I: 2. Literature Survey and Meta-Analysis 19
several measures are usually combined in one programme. In that
case, it is important
to know not just the effects of each measure that goes into the
programme but the
combined effects of all measures put together. It is not obvious
that the effects of a
road safety programme will be equal to the sum of the effects of
the individual
measures that make up the programme. The effect of a measure will
not necessarily be
the same when it is implemented in combination with other measures,
as when it is
implemented on its own.
Another general limitation of road safety evaluation research is
that it often requires
that the measures are implemented fairly extensively to provide
enough data to
evaluate effects. This means that evaluation research does not
always provide a good
basis for predicting the effects of new measures. Those who develop
new measures
would like to be able to predict the effects of the measures before
they are introduced.
Such prediction is not always possible. In Chapter 5, the
possibility of giving a
theoretical account for the findings of road safety evaluation
research will be discussed.
2.4 THE USE OF META-ANALYSIS TO SUMMARISE STUDY RESULTS
The results of studies that have evaluated the effects on accidents
and injuries of
different measures are summarised by means of meta-analysis,
provided it is applicable.
Meta-analysis is a quantified synthesis of results of several
studies that have evaluated
the same road safety measure stated in the form of a weighted mean
estimate of effect
(Elvik 1999). As a part of the meta-analysis, moderating factors
are investigated that
influence the size of the effect of a road safety measure on
accidents or injuries.
There are a number of textbooks on meta-analysis (Cooper and Hedges
1994, Petitti
2000, Lipsey and Wilson 2001) that describe various techniques
in detail. Here, only the
main elements are described to help readers understand the results
that are presented in
the individual chapters.
Main elements of meta-analysis. The study unit in a meta-analysis
is a result, or an
estimate of effect. An estimate of effect has to be stated as a
precise point estimate in
order to be included in a meta-analysis. If a result is stated
simply as: ‘No statistically
significant changes in the number of accidents were found’, it
cannot be included in a
meta-analysis. Moreover, the standard error of an estimate of
effect has to be known,
at least if results are to be weighted according to their
statistical precision. A single
study can contain more than one result. In such cases, all results,
or the most important
results from studies with a very large number of results, have been
included in the
meta-analyses. Multiple results from the same study have been
treated as statistically
independent, although this assumption may not always be
correct.
20 The Handbook of Road Safety Measures
Study results can be summarised by means of meta-analysis if the
studies
provide at least one numerical estimate of the effect of a
road safety measure, or
provide information that can be used to derive such an estimate and
state the number of accidents on which the estimate of
effect is based or provide
other information that allows the calculation of the statistical
uncertainty of the
effect estimate, such as the confidence interval.
Basics of the log odds method of meta-analysis. The log odds method
of meta-analysis
has been applied throughout (Fleiss 1981, Shadish and Haddock
1994). According to
this method, a weighted mean estimate of effect is calculated on
the basis of the
estimates of effect found in the studies that have been retrieved.
This method of meta-
analysis was chosen because the odds ratio (OR) is the most
commonly found estimate
of effect in road safety evaluation studies. An example of how an
OR is calculated is as
follows: If a study finds that there were 75 accidents on road X
before a measure was
implemented, and 23 accidents afterwards, whereas on a comparison
road, there were
67 before the implementation of the measure on road X and 25
afterwards (no measure
was implemented on the comparison road), the OR is
(23/75)/(25/67) ¼ 0.307/
0.373 ¼ 0.822. This corresponds to an accident reduction
of 17.8% (1 þ 0.822). In
studies that employ multivariate techniques of analysis, effects
are normally stated in
terms of an OR that has been adjusted for confounding.
When applying the log odds method of meta-analysis, a summary
effect is calculated as
the weighted mean of the logarithms of the individual estimates of
effect (ORs).
Combining logarithms of ORs yields an unbiased estimate of the
weighted mean effect
of a set of studies. The steps in a log odds meta-analysis
are
calculation of estimates of effect, calculation of
statistical weights and choice of the model of meta-analysis:
Fixed
effects when there is no systematic variation in the estimates of
effect or random
effects when there is systematic variation in the estimates of
effect, calculation of summary estimates of effect, and
confidence intervals: for each summary effect, a 95%
confidence interval is
calculated.
Calculation of estimates of effect. Estimates of effect are
calculated as ORs. Some of the
estimators of effect commonly found in road safety evaluation
studies are listed in
Table 2.4. The list is not exhaustive. Estimates of effect based on
coefficients produced
by multivariate analyses, which have the statistical properties of
ORs, are not as
common, but have increasingly been used in recent studies. The
different estimators of
effect should not be mixed up. Producing summary estimates of
effect in meta-analysis
Part I: 2. Literature Survey and Meta-Analysis 21
based on studies that employ different estimators of effect can be
misleading because
both the statistical properties and the substantive interpretations
of the various
estimators differ. When other estimates of effect other than ORs
are reported, ORs are
calculated as far as possible based on the available
information.
Calculation of statistical weights and choice of model . There
are two methods of
combining estimates of effect in meta-analysis, the fixed effects
model and the random
effects model. The fixed effects model of analysis is based on the
assumption that there
is no systematic variation in effects in the set of studies
considered, that is, all estimates
of effect are samples of the same ‘true’ effect. When there is
systematic variation, or
heterogeneity, in the estimates of effect, the estimates cannot be
regarded as
representing the same ‘true’ effect. In this case, a random effects
model is more
adequate. In a random effects model, an account is taken of
heterogeneity in the results
and an underestimation of the uncertainty of the summary effect is
avoided.
The differences between the fixed effects and the random effects
models can be
summarised as follows: The fixed effects model is adequate only if
there is no
heterogeneity in the results. Otherwise it will assign too much
weight to results with
large statistical weights and the confidence interval of the
summary effect will be
underestimated. The random effects model can be applied whether or
not there is
heterogeneity in the results. When there is significant
heterogeneity, it assigns relatively
less weight to results with large fixed effects weights, and
confidence intervals of
summary effects are larger than that in the fixed effects model.
The less heterogeneity
there is in the estimates of effect, the more similar will be the
results from the two
models.
When applying fixed effects and random effects models in
meta-analysis, they differ
with respect to how the statistical weights are calculated. In the
fixed effects model, the
Table 2.4: Commonly used estimators of effect in road safety
evaluation studies
Name of dependent variable Formal definition
Odds Uat/Ubt
Ratio of odds ratios
[(Uati/Ubti)/(Uaci/Ubci)]/[(Uatj/Ubtj)/(Uacj/Ubcj)]
Ratio of relative risk [Uati/(UatiþUbti)]{[Uatj/(UatjþUbtj)]
Accident rate ratio (Ua/Ta)/(Ub/Tb)
U ¼ number of accidents, T ¼ traffic volume,
exposure to risk, a ¼ after, or with, some measure whose
effect is evaluated, b ¼ before, or
without, some measure whose effect is evaluated, t ¼ test
group, c ¼ comparison group, i ¼ category I,
j ¼ category j.
22 The Handbook of Road Safety Measures
statistical weight of the natural logarithm of each effect estimate
is the inverse of its
variance:
vi ¼ 1
A þ 1
B þ 1
C þ 1
D
where A, B, C and D are the
four numbers that enter the calculation of the estimate
of
effect. In studies that do not use comparison groups, the terms
1/C and 1/D drop out.
The same applies to studies that state the effects of a road safety
measures in terms of
an accident rate ratio. Statistical weights are estimated on the
basis of the recorded
number of accidents. In case of zero accidents, 0.5 is added to all
four (or two) numbers
used in estimating the statistical weight of a result.
In a random effects model, the statistical weights are calculated
as a function of the
fixed effects weights and a measure of the heterogeneity in the
estimates of effect.
The more heterogeneity there is in the results, the more similar
will the statistical
weights of the estimates of effect become, that is estimates based
on large fixed effects
weights will have their weights adjusted more than estimates based
on small fixed
effects weights.
In order to test the amount of heterogeneity in the estimates of
effect, the following test
statistic, Q, is estimated:
wi
where yi is the estimate of
effect i and wi the fixed effects
weight of estimate i . This test
statistic has a w2 distribution with g1 degrees of
freedom, where g is the number of
estimates of effect that have been combined. If this test statistic
is statistically
significant, a random effects model is more adequate than a fixed
effects model. In a
random effects model, the statistical weights are modified to
include a component
reflecting the systematic variation of estimated effects between
cases. This component
is estimated as follows (Shadish and Haddock 1994):
t2 ¼ Q ð g 1Þ C
Part I: 2. Literature Survey and Meta-Analysis 23
Q is the test statistic described earlier, g the
number of estimates and c the following
estimator:
vi ¼ t2 þ vi
The corresponding statistical weight becomes the inverse of the
variance.
Random or fixed effects? Most meta-analyses that are
presented in the book have been
calculated based on a random effects model. Fixed effects models
have been applied
only when too few estimates of effect are available for calculating
a random effects
model. In meta-analyses that have not been updated after 1997, the
fixed effects model
is the most commonly used model.
Summary effects. The weighted summary effect based on a set
of g estimates is
calculated as follows:
y ¼ exp
1
C
C
A
where ‘exp’ is the exponential function (i.e., 2.71828 raised to
the power of the
expression in parenthesis), yi the logarithm of
each estimate of effect and wi the
statistical weight of each estimate of effect.
Confidence intervals. A 95% confidence interval for the weighted
mean estimate of
effect is obtained according to the following expression:
95% confidence interval ðupper=lower limitÞ ¼ exp
P
g
wi
s
2
6
6
6
6
4
3
7
7
7
7
5
The weights in this expression are either the fixed effects weights
or the random effects
weights, depending on the model of analysis adopted.
24 The Handbook of Road Safety Measures
2.5 DOES A WEIGHTED MEAN ESTIMATE OF EFFECT MAKE SENSE?
A concern that many people have about meta-analysis is the
so-called apples and
oranges problem. This refers to the fact that studies that may
differ greatly among
themselves are combined into an overall estimate of the average
effect of a road safety
measure. It is argued that this does not make sense if studies are
very heterogeneous,
for example, with respect to different versions of the measure,
countries or methods
used in the studies.
Fortunately, the relevance of this argument can to some extent be
tested in a meta-
analysis. By doing so, one gains an impression of how meaningful it
is to generalise a
set of findings of evaluation studies in terms of a weighted
average result. A way of
checking whether a weighted mean estimate of effect makes sense is
to prepare a funnel
graph plot. An example of such a graph is shown in Figure
2.1.
The graph shows 94 results of studies that have evaluated the
effects of road lighting on
the number of accidents. The horizontal axis shows the natural
logarithms of the
estimates of effect. Values below 0 mean that the number of
accidents is reduced, the
0
100
200
300
400
500
600
700
-3.0-2.5-2.0-1.5-1.0-0.50.00.51.01.52.0
Effect estimate (natural logarithm; 0 = no effect)
S t a
t i s
t i c a
l w e
i g h
t
( f i x
e d e
f f e c
t s )
Summary effect (fixed effects): -0.194
Arithmetic mean: -0.292
Median: -0.319
Figure 2.1: Funnel graph of studies that have evaluated the effects
of road lighting on the
number of accidents (unspecified severity).
Part I: 2. Literature Survey and Meta-Analysis 25
value 0 means that the number of accidents is unchanged and values
above 0 mean that
the number of accidents increases. The vertical axis shows the
statistical weight (fixed
effects) of the results. The greater the statistical weight, the
more the accidents which
form the basis of a result. The dots indicate the individual
results. Furthermore, three
measures of the main tendency of the results are shown: the median,
the arithmetic
(unweighted) mean and the summary effect that has been calculated
with the fixed
effects model.
By studying such funnel graphs, an informed opinion can be formed
of how reasonable
a weighted mean result is. Properties of the distribution of
estimates of effect that are
investigated based on the funnel graph are the modality and
dispersion of the results,
the skewness and the sensitivity to outliers.
Modality and dispersion of the results refers to the shape of
the distribution of estimates
of effect and how many humps or peaks it has. Figure 2.1
shows a unimodal
distribution, that is, a distribution where the data points gather
round a single peak. In
this type of distribution, the weighted summary effect lies close
to the highest peak of
the distribution and thus is representative of the centre of
gravity of the distribution.
A bimodal distribution is one that has two peaks. In this type of
distribution, the
average will often lie between the two peaks and thus will not
really be very
informative. If possible, bimodal distributions should be divided
into two, and an
average should be calculated for each mode.
There may also be distributions with no clear pattern at all,
randomly scattered
distributions. In these types of distributions, the results are
highly dispersed, with no
clear tendency in any direction. An average may then be arbitrary
and any differences
concealed as a result of arbitrary assignment would be important to
highlight. Ideally,
the distribution of the results should not only be unimodal but
also exhibit a systematic
pattern where the results that have the largest statistical weights
are closest to the mean
and results that are further away from the mean have smaller
statistical weights. It is
not always easy to see if the results follow an ideal distribution
or not. There are
statistical methods for investigating the distribution of the
results and for treating
results that are not ideal.
First, heterogeneity can be tested statistically as has been
described earlier, and a
random effects model can be applied that takes into account
heterogeneity. A random
effects model takes into account that there is heterogeneity, but
does not explain it.
Second, there are possibilities for explaining heterogeneity. The
simplest way is to
divide results into sub-groups and to calculate new summary effects
for each of the sub-
groups of results. Results may be grouped, for example, according
to injury severity or
26 The Handbook of Road Safety Measures
variants of the measure. When summary effects differ between
sub-groups, and when
heterogeneity is reduced within the sub-groups, the sorting
variable is likely to have
contributed to the heterogeneity. It is then called a moderator
variable.
Heterogeneity can also be explained by using meta-regression. In
meta-regression
analysis, regression models are developed on study level with the
estimates of effect as
dependent variable and characteristics of the studies as
predictors. Characteristics of
the studies may be the same variables as the sorting variables in
the sub-group analysis
(e.g., type of measure investigated, type of roads, methodological
aspects, and so on).
Thereby, it is possible to investigate which characteristics of the
studies affect the
outcome of the studies, while controlling for several factors at
the same time. One
restriction of meta-regression is that it requires quite large
numbers of estimates of
effect. As a rule of thumb, there should be at least 10 estimates
of effect for each
predictor included in the model. When there are few estimates of
effect, the results may
be arbitrary and highly sensitive to, for example, adding or
omitting predictors or
individual estimates of effect from the analysis.
A third possibility that should be considered in some cases is to
refrain from calculating
a summary effect. When the distribution of estimates of effect is
highly heterogeneous
without showing any signs of unimodality, a summary effect would
not be meaningful.
Indications for such a distribution are results that are highly
different between the fixed
effects and the random effects model and extremely large confidence
intervals in the
random effects model. This is illustrated by a numerical example in
which six estimates
of effect have been generated that have a highly heterogeneous and
non-unimodal
distribution. In this example, the result from the fixed effects
model is a summary effect
of 57% (95% confidence interval [58; 55]), and that
from the random effects
model is a summary effect of þ6% (95% confidence
interval [62; þ195]).
Skewness in a distribution refers to how the data points are
distributed around the
average, that is, how the individual results distribute themselves
around a weighted
average result. Ideally, the distribution should be symmetrical
around (the natural
logarithm of) the summary effect. If a distribution is very skew,
the mean will give a
misleading impression of where the majority of the results lies. An
indication of
skewness is a large difference between the median and the
arithmetic mean of the
distribution. An unskewed distribution will have very similar
median and arithmetic
mean.
Publication bias is one possible source of skewness.
Publication bias means that studies
are more likely to be published when the results are in accordance
with the expectation.
In most accident studies, the expectation is that one will find
accident reductions
following the implementation of a safety measure. Publication bias
leads to a skewed
Part I: 2. Literature Survey and Meta-Analysis 27
distribution of the estimates of effect because there will be fewer
results on its
‘undesired’ side. Moreover, results (also in the desired direction)
are more likely to be
significant when based on large numbers of accidents. Results from
small studies that
find large effects in the expected direction will therefore be
over-represented, and
results from small studies that are unexpected will be
under-represented. In the absence
of publication bias (and other biases), the distribution will be
symmetrical.
When there is publication bias, the summary effect is likely to
show larger accident
reductions than would be the case if all studies had been
published, and none had been
omitted because of insignificant or unexpected results.
A statistical possibility for controlling for publication bias is
the trim and fill method.
This method simulates studies that are assumed to have been
suppressed by a tendency
to publish (large) effects in the expected direction (Christensen
2003). The distribution
of all original estimates of effect and the simulated estimates of
effect that are
generated in a trim and fill analysis is symmetrical around the
peak of the original
distribution. The simulated estimates of effect are mostly effects
from small studies in
the unexpected direction. A summary effect is then calculated based
on all, original and
simulated, estimates of effect. This summary effect is usually less
favourable for the
measure that has been evaluated than the summary effect that is
based on the published
estimates only. The trim and fill method can be applied both to a
fixed effects and a
random effects model of meta-analysis. An example is shown
in Figure 2.2.
Figure 2.2 shows the results of the same studies as in
Figure 2.1. The ‘funnel’-shaped
dotted lines (which are drawn by hand, not fitted to the data)
indicate roughly the outer
limits of the distribution of the estimates of effect. All original
estimates of effect are
located inside these lines and seem to be almost symmetrically
distributed. All the
same, 15 new data points have been generated in the trim and fill
analysis. All of them
show increases in accident numbers on roads with road lighting and
have relatively
small weights. The summary effect that is based on all, original
and simulated, results is
consequently somewhat less favourable (15% accidents on lit roads
compared with
unlit roads) than the summary effect that is based on the original
estimates of effect
only (18% accidents on lit roads compared with unlit roads). The
difference is,
however, not large.
Other biases can produce results that resemble those expected
when there is publication
bias. The most frequent such bias occurs when results that refer to
different accident
severities are combined in one analysis. Many road safety measures
have larger effects
on more severe accidents. One such measure is road lighting; other
measures are
guardrails, roundabouts, electronic stability control, seat belts
and numerous others.
Evaluation studies of such measures are likely to find larger
effects on fatal accidents
28 The Handbook of Road Safety Measures
than on injury accidents. Since there are usually far more injury
accidents than fatal
accidents, the (large) estimates of effect that refer to fatal
accidents will have smaller
statistical weights than the (smaller) estimates of effect that
refer to injury accidents. A
distribution of estimates of effect that refer to a mixture of
fatal and injury accidents
may therefore tempt one to conclude that there is publication bias.
This is illustrated in
Figure 2.3, which consists of the same data points as Figure
2.1. The results that refer
to fatal accidents (black) have small statistical weights and on
the average are more on
the right side of the distribution, that is, show larger accident
reductions than the
results that refer to non-fatal accidents (white).
The trim and fill analysis that has been applied to the data in
Figure 2.2 should
therefore not have been applied to these data. The skewness of the
distribution is more
likely to be due to the mixing of accident severities, and not due
to the publication bias.
In this case, there is an apples and oranges problem and a summary
effect should not
be calculated based on all results.
Sensitivity to outliers denotes how strongly the me