g.

' '22!“

w"

.51

 

<

a:

a:

Sun.

.1.

x

24%. m.

 

_. .
"2%...
..

(. .uyvar EMF.“ ... .
t... .11.. 1.: 2
.

a... .32. .
. . If. .5:
mfiéwvik
......2.. _ .1
....2...u222u.2; 2..... .

i
3...

2.~.

2.“

3.39:}:

 

z...— . .

 

1
2004
r’ ’ n LIBRARY
JU’U-v' W57 Michigan State
University

This is to certify that the
thesis entitled

ASSESSMENT OF CONSTRUCTION WORKERS OCCUPATION. 2L
SAFETY COMPETENCIES USING SIGNAL DETECTION THEORY

Presented by
BHAVIN PATEL

has been accepted towards fulfillment
of the requirements for the

MS. Construction Management
Program

gww

Major Professor” 5 Signature

9/22/03

Date

 

 

 

MSU is an Affirmative Action/Equal Opportunity Institution

Ac-I-O-.-l-l-D-O-l-Q-O-0-O-C-‘" - .

‘ . PLACE IN RETURN BOX to remove this checkout from your record.
“ . To AVOID FINE return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE

 

1A?

I

 

‘—

If“? 1,“ TH!"

.. A ‘ ‘_‘ “L1

 

 

 

 

 

 

 

 

 

 

 

 

6/01 c:/ClRC/DateDue.p65-p. 15

 

ASSESSMENT OF CONSTRUCTION WORKERS OCCUPATIONAL SAFETY
COMPETENCIES USING SIGNAL DETECTION THEORY

By

Bhavin J. Patel

A THESIS

Submitted to
Michigan State University
In partial fulfillment ofthe requirements
for the degree of

MASTER OF SCIENCE

Construction Management

.luly 2003

ABSTRACT

ASSESSMENT OF CONSTRUCTION WORKERS OCCUPATIONAL SAFETY
COMPETENCIES USING SIGNAL DETECTION THEORY
By
Bhavin J. Patel
Construction accidents in general and fall accidents in particular are of major concern in
construction, as many lives are lost and business suffers. Despite the contribution of
construction accident causation models. accidents occur. Most of the models that have
been developed stress identification of the underlying causes of accidents and have sided
with either management or the'workers fault. None of these models could adequately
explain the process of construction accident due to their dynamic nature.

A new approach to understand construction accidents has been proposed by
Howell et al (2002) based on the work of Rasmussen (1997). One of the aspects of this
model is focused on worker training to identify the hazard zone (unsafe condition)
beyond which work is no longer safe. The main goal of this research was to deyelop a
methodology by which worker sensitivity to unsafe conditions and risk orientation (the
tendency ofa worker to work in a condition despite knowing it is unsafe) can be assessed
prior to prescribing a training program. This research prOposes such methodology based
on signal detection theory (SDT). which is used in the manufacturing industry to detect
defective component. Application of SDT in the construction industry to determine
sensitivity and risk orientation of ironworkers has been explained in this thesis. This
would help to design guidelines on how to enhance construction workers' training and

also their abilities to identify by themselves the boundary beyond which work is not safe.

DEDICATION

This thesis is dedicated to my mom and dad for their continued support and love.

iii

ACKNOWLEDGMENTS

I am thankful to my advisor Dr. Tariq Abdelhamid for guidance. patience.
assistance and support throughout this study. I am also thankful to Dr. Mohammad Najafi
and Dr. Robert Von Bemuth for their help during this research.

I would like to thank Mr. Art Ellul for his kind help and cooperation for the
survey of the ironworkers.

V I would like to express my thanks to my mom and dad for their great support.
love, and encouragement.

Lastly. thanks to all my lab friends here in the department. my roommates and all

my old friends who have always supported. encouraged me and helped me whenever I

asked.

TABLE OF CONTENTS

 

 

LIST OF TABLE .................................................................................. vii
LIST OF FIGURES .............................................................................. viii
CHAPTERI- INTRODUCTION ............................................................... 01
1.1 MOTIvATION .................................................................................. 02
1.2 PROBLEM AREA ..................................................................................................... .06
1.3 GOAL AND OBJECTIvES ........................................................................................... 09
1.4 PROPOSAL OvERvIEw .............................................................................................. 10
CHAPTER 2 - BACKGROUNDZ. BACKGROUND - - 12
2.1 OVERvIEw OF ACCIDENT CAUSATION MODELS ....................................................... 13
2.2 RASMUSSEN‘S THEORY OF COGNITIVE SYSTEM ENGINEERING ................................ 20
2.3 SIGNAL DETECTIONTHEORY ................................................................................... 25
2.4'ROC CLTRVE ............................................................................................................ 32
CHAPTER 3 - METHODOLOGY3. METHODS----- - - - ......... - ...... 35
3.1 INTRODL‘CTION ........................................................................................................ 36
3.2 SDT AND ITNSAIE AND SAFE CONSTRL'CTION CONDITIONS (OBJECTIVE 1) ............ 37
3.3 SURVEY DEvELOPNIENT AND ITS ANALYSIS USING SDT ......................................... 39 .
3.4 ANALYSIS wI'I‘II ROC .............................................................................................. 42
CHAPTER 4 - SURVEY RESULTS AND ANALYSIS OF DATA ........................... 44
4.1 INTRODUCTIO\ ........................................................................................................ 45
4.2 DATA COLLECTION .................................................................................................. 45
4.3. SENSITIvITv AND RISK ORIENTATION or THE lRONWORkERS BY SDT ................... 46
4.3.1 Relation between t/' gm and (1' my ................................................................... 54
4.4 DATA ANALYSIS ...................................................................................................... 56
4.4.] Average sensitivity and Risk Orientation oflromI'or/I'erx ............................... 5 6
4. 4.2 Distribution (1711' and [Ii-mm" ............................................................................ (St)
4.5 REGRESSION ANALYSIS ............................................................................... f ............ 61
4. 5. 1 Regression A na/_t:s'i.s'_ fOr Age ............................................................................ 63

4.5.2 Regression Analysis/0r Years of Experience .................................................. 66

 

 

 

 

 

 

 

4.5.3 Multiple Regression Analysis ........................................................................... 68

4.6 SUMMARY ................................................................................................................ 70
CHAPTER 5 - SUMMARY AND CONCLUSION - 71
5.1 CONCLUSIONS .......................................................................................................... 72
5.2 LIMITATIONS OF THIS RESEARCH .............................................................................. 74
5.3 AREAS OF FUTURE RESEARCH .................................................................................. 74
5.4 CONTRIBUTIONS OF THE RESEARCH ......................................................................... 75
APPENDIX A-..-- - - - - - _ _ _ - _ 76
APPENDIX B ' - . - 83
APPENDIX C - - 93
APPENDIX D------ - - _ -- - -- -- -- 96
APPENDIX E - .fi 98
APPENDIX F ........... - - - ............. - 101
REFERENCES .................................... _ ................. _ ........................ 103

 

vi

LIST OF TABLES

TABLE 1.1 DISTRIBUTION OF ACCIDENTS IN PROJECT BY TYPE FROM 1997- 2001 ............ 04
TABLE 1.2 DISTRIBUTION OF ACCIDENTS BY NATURE OF CONSTRUCTION ........................ 04

TABLE 2.1 THE FOUR OUTCOMES OF SIGNAL DETECTION THEORY (WICRENS. 1992). 26

TABLE 2.2 THE FOUR PROBABILITIES ........................................................................... ‘ ..... 30
TABLE 3.12 SAMPLE SURvEY ANALYSIS RESULTS ................................................................ 41
Table 4.] RESULT OF THE SURVERY OF THE IRONWORKERS ..................... 47
TABLE 4.2 MATRIX SHOWING THE RESPONSES FOR WORKER wl ...................................... 51
TABLE 4.3 RESULTS OF ANALYSIS OF SURVEY BY STANDARD SDT AND ROC CURVE ..... 52
TABLE 4. 4 SUMMARY OF RESPONSE TO EACH QUESTION ................................................. 53
TABLE 4.5 SUMMARY OF RESULTS FROM SDT AND ROC CURVE ..................................... 57
TABLE 4.5 SUMMARY OF RESL‘LTS FROM SDT AND ROC CURVE (CONTINUED) ............... 58
TABLE 4.6 REGRESSION ANALYSES FOR YEARS OF EXPERIENCE ....................................... 68
TABLE 4.7 ML'LTIPLE REGRFSSIO‘N ANALYSIS RESULTS .................................................... 69

Vii

LIST OF FIGURES

FIGURE 1.1 CAUSE OF CONSTRUCTION ACCIDENTS INVESTIGATED BY OSHA .................. 03
FIGURE 1.2 DISTRIBUTION OF HEIGHT OF CONSTRUCTION FALL ACCIDENTS FROM 1997-
2001. ........................................................................................................................... 05
FIGURE 1.3 THREE ZONES OF RISK .................................................................................... 09
FIGURE 2.1: THE FIVE FACTORS IN THE ACCIDENT SEQUENCE .......................................... 15
FIGURE 2.2 MIGRATION OF WORK TOWARDS LOSS OF CONTROL ...................................... 23
FIGURE 2.3 DISTRIBUTION OF DETECTION THEORY ........................................................... 27
FIGURE 2.4 THEORETICAL REPRESENTATION OF THE ROC CURVE (WICKENS. 1992) ....... 33
FIGURE 2.5 THE ROC CURVE ON PROBABILITY PAPER (WICKENS. I992) ......................... 33

FIGURE 2.6 EXAMPLE OF MEASURE OF P(A) (WICRENS. 1992) ......................................... 34

FIGURE 3.1 THE SDT MATRIX FOR DETECTION OF UNSAFE CONDITIONS IN CONSTRUCTION

................................................................................................................................... 38
FIGURE 4.1 RELATION BETWEEN SENSITIVITY BY SDT & SENSITIVITY BY ROC 55
FIGURE 4.2 SCATTER PLOT BETVIEEN SENSITIVITY BY SDT & SENSITIVITY BY ROC ....... 55
FIGURE 4.3 DISTRIBUTION OF DETECTION THEORY (WICRENS. I992) .............................. 59
FIGURE 4.4 DISTRII3L'TIO\ OF TIIE SURVEY RESULTS 60.
FIGURE 4. 5 AGE OF IR().\'\\‘OR1\'ER vs. SENSITIVITY USING STANDARD SDT ................... 64
FIGURE 4.6 AGE OF IRONVVORRER VS. SENSITIVITY USING ROC .......................... . ............ 65
FIGURE 4.7 AGE OF IRONVVORRER VS. BETA CURRENT ...................................................... 66
FIGURE 4.8 YEARS OF EXPERIENCE OF IRONWORKER VS. SENSITIVITY BY SD'I‘ ............... 67
FIGURE 4. 9 YEARS OF EXPERIENCE OF IRONWORKER VS. SENSITIVITY BY ROC .............. 67
FIGURE 4. 10 YEARS OF EXPERIENCE OF IRONWORKER VS. BEL-222222.22. .................................. 67

viii

Chapter 1
INTRODUCTION

1. INTRODUCTION

1.1 Motivation

Over the past three decades. numerous organizations and researchers have focused on
investigating construction accidents. The literature on construction safety reveals that
much research effort has been directed-at examining accident records to categorize the
most common types of accidents that'occur to a Specific trade, and how these accidents
happen (MacCollum. 1990; La Bette. I990; Rietze. 1990; Pullman, 1984‘; Goldsmith.
1987; Davies and Tomasin. I990: Culver et al., 1990; Helander, 1991; Culver et al..

1992; Peyton and Rubio. I991: Hinze. 1997).

, Asthe leading cause of most injuries and fatalities in construction, fall accidents
have received much attention. In fact. the Occupational Safety and Health Administration
(OSHA) considers reducing falls as a Strategic goal for the organization for the next five
years. OSHA investigated 7.543 construction related accidents from January 1990
through October 2001. and found that falls accounted for 34.6% of the injuries (Huang
and Hinze. 2003). From the statistical analysis it was found that the proportion of falls
has increased (in past 12 years: the average proportion of falls was 34.1 0/0 during the
years before 1996 and increased to 38.4% in the following years (Huang and Hinze

2003). Figure 1.1 shows the breakdown (in percentage) of construction accidents causes.

To examine what time of the day. or month relates to accidents. 3 study conducted
by Huang and Hinze (2003) concluded that most accidents were reported in the month of
July with 820 accidents. However. February. with 493 accidents. was the month with the

least accidents. Analysis also Showed that in winter (December to February) the average

proportion of fall accidents and all accidents per month are 7.6 and 6.6% respectively,
while in summer (June to August) the average proportion of fall accidents and all

accidents are 9.1 and 10.3 %.

Stuck By
24%

 

Figure 1.1 Cause of Construction Accidents Investigated by OSHA
(Source: Huang and Hinze 2003)
The study conducted by Huang and Hinze (2003) from the available data from
OSHA also showed that fall accidents occur more frequently on certain types of projects.
beginning with new construction followed by renovation. maintenance and finally
demolition work. Table 1.1 shows the breakdown of the count and percentage of fall
accidents and all accidents from I997 to 2001. It can be seen from the table that projects
involving commercial buildings and single family or duplex dwellings account for nearly
half of the fall accidents from 1997 to 2001 (Huang and Hinze. 2003). Statistics also

showed that 60% of falls occurred in new construction or additions (Table 1.2).

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

, Falls All Accidents
Project Count Percent Count Percent

Commercial building 404 33.3 715 22.8
Other building 212 17.4 412 13.1
Single family or duplex dwelling 211 17.4 503 16
Multifamily dwelling 113 f 9.3 183 5.8

anufacturing plant 79 6.5 168 5.3
Tower. tank. storage elevator 71 5.8 103 3.3
Bridge 28 2.3 94 3
Other heavy construction 21 1.7 94 3
Highway. road. Street 16 1.3 381 12.1
Sewer/water treatmengalant 14 1 .2 76 2.4
Power plant 13 1.] 33 1.1
Power line 10 0.8 116 3.7
Contractor'SJ'ard/facility 5 0.4 42 1.3
Pipeline 4 0.3 91 2.9
Shoreline development. dam. reservoir 4 0.3 24 0.8
Refinery 3 i 0.2 21 0.7
Excavation. landtill 2 0.2 63 2
Subtotal ' 1210. 100 3119 100
Not known 5 23
Total 1215 3142 I

 

 

 

 

 

 

Table 1.1 Distribution of Accidents in Project by Type from 1997- 2001
(Source: Huang and Hinze, 2003)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

I Type ofConstruction Falls All Accidents
I ' effort .

Count Percent Count Percent
New project or new addition 721 59.3 1.640 I 52.2
Alteration or rehabilitation I 219 18 565 18 I
Maintenance or repair J 189 15.6 531 16.9 I
Demolition i 41 3.4 101 3.2
Other I 41 3.4 283 l 9 I
Subtotal 1.211 100 3.1201 100
Not Known 4 L . 2 22 I
Total 1.2157 3.142 I I

 

 

 

 

Table 1.2 Distribution of Accidents by Nature of Construction
(Source: Huang andHinze, 2003)

In the same study conducted by Huang and Hinze (2003). it was determined that
from 2,741 accidents reported by OSHA, 81% of them occurred while workers were
working on the first to third floor of buildings and the average height of the fall was
almost 37 feet. The distribution of fall heights is Shown in Figure 1.2. From the figure it

could be concluded that 70% of fall accidents occur at 30 feet or less.

 

>200ft

150-200 ft

 

100-15011

 

80-100ft

 

60-80fi’

EUEDD

50-60fi

 

 

40-50ft

 

 

 

30-40 ft

F—p

 

 

 

 

 

 

lo-mifi l

 

 

 

 

 

I

I

I

. I

20-30fi I - l " I
l

I

 

 

 

0-10ft

..__ ~_—. ——
_-—d

592 1092‘ 1596 2092 2596 3002
Figure 1.2 Distribution of Height of Construction Fall Accidents from 1997-2001
(Source: Huang and Hinze, 2003)
This study also determined that falls generally resulted due to misjudgment of
workers about the hazardous Situations. lack of Personal Protective Equipments (PPE) or

insufficient safety protection.

Various research studies have found that fall accidents typically occur due to
faulty equipment, inadequate fall protection, floor openings, aerial lifts. steel or concrete
erection. roofing and/or placing reinforcement. These causes are typically classified
under unsafe conditions or unsafe acts and often are due to organizational problems.
Identification of root causes to find effective corrective actions could prevent these

injuries/fatalities.

1.2 Problem Area

Notwithstanding the progress and improvements made in the safety record, construction
work remains hazardous work. The National Safety Council (NSC) reported that. in
‘2001, construction accounted for 6% of the United States' workforce but claimed a
disproportionate 23 % of all occupational fatalities and 10.5 % of all occupational injuries
(Injury Facts. 2002). Moreover. the NSC estimated that. in 2001. 15% ofthe $145 billion

spent on occupational injuries. was spent on construction cases.

Accidents in general. and fall accidents in particular. are Of major concern in the
construction industry. as many lives are lost and business suffers. Much is known about
accidents through investigations that provide for the "what" and "how" questions.
Despite its necessity as a phase. accident investigation seldom addresses the factors that
contributed to the accident causation. i.e.. ll'HI' the accident occurred. Allan St John
(2003) mentioned that fall prevention is far more effective than fall protection. which
Often involves personal protective equipment and training (Huang and Hinze-2003).
Brown (1995) has argued convincingly that accident investigation techniques should be
firmly based on theories of accident causation and human error. which would result in a

better understanding Of the relation between the “antecedent human behavior" and the

accident at a level enabling the root causes of the accident to be determined.
Consequently, prevention efforts could be directed at the root causes of accidents and not

at symptoms. leading to more potent prevention efforts.

Myriad accident causation models have been proposed over the years. These
models provide many explanations for the occurrence of injuries and fatalities to
industrial workers. Models are classified into different categories such as management
models, behavior models. human factor models, system models. epidemiological models.
decision models. etc. (Heinrich. 1980). Most of the models Stress identification of the
underlying causes of accidents and have Sided with either management or with the
workers. In general. the overall objective of these models is to provide tools for better

industrial accident prevention programs.

Construction accident causation models based on variants of the above models
have been introduced in the literature by a handful of researchers (Abdelhamid and
Everett, 2000: and Suraji et al. 2001). Despite the contributions of these construction
accident causation models in understanding the accident process. none adequately explain
I the underlying causes of construction accidents. For researchers. many topics related to

falls still need to be investigated in great detail.

A new approach to understanding construction accidents has been proposed by
Howell et al. (2002) based on the work of Rasmussen (1997). The model suggested
recognizes that organizational and individual pressures push people to work in hazardous
situations. These pressures defeat efforts to enforce safe work rules. specifically in a

changing work environment such as in construction. Therefore. this approach emphasizes

the need to train workers to be conscious of hazardous work environments and engage the
work with better planning and appropriate protection in a way very similar to how fire

fighters engage hazardous situations.

The original model as proposed by Rasmussen is Shown in Figure 1.3. As shown,
Rasmussen divided the work environment into three zones. Zone I. which is the region
enclosed by the “Boundary of Unconditionally Safe Behavior", “Organizational
Boundary to Economic Failure“. and “Individual Boundary to Unacceptable Work load”.

is considered the safe zone.

Rasmussen states that due to economic or workload pressures, workers will Shift
their work along the workload and/or cost gradients, respectively. So as long as workers
remain within the safe zone. work activities can be safely performed. Current safety
regulations and management practice are directed at keeping the workers in the safe zone.
Rasmussen suggests that enlarging the safe zone through proper planning of operations

will make the work safer.

The zone encompassed by the “Boundary of Unconditionally Safe Behavior" and
the “Irreversible LOSS of Control Boundary" is Zone 11 or the hazard zone. Workers
working in the hazard zone are considered to be working at the edge (pushing their luck).
Rasmussen believes that. despite regulatory or supervisory efforts. workers will move to
the hazard zone for many reasons. He suggested. contrary to current conventional
wisdom. that the only effective way to counter these tendencies to work in the hazard
zone would be to make visible the boundary beyond which work is no longer safe and

teach worker to recognize the boundary.

Irreversible Loss of

Control Boundary Boundary of
Unconditionally

/ Safe Behavior

Loss of
Control Zone Organizational
Boundary to
Economic Failure
Workload Gradient
Hazard
Zone Safe
Zone
Cost Gradient
‘/

IndividuaI‘Boundary to
Unacceptable Workload

Figure 1.3 Three Zones of Risk
(Source: Howell et al., 2002)

The third and final zone in Rasmussen‘s model is the loss of control zone. in
which accidents occur and control is lost. leading to injuries and/or fatalities. He
proposed that workers should be educated on and trained in how to recover from situation
in whichcontrol is lost. This is very similar to instructing drivers in how to respond to

slips on icy roads.

1.3 Coal and Objectives
The acceptance and effectiveness of Rasmussen‘s approach remains a question that only
future research can answer. A number of techniques exist for operations planning that

could help to enlarge Zone 1. Virtual reality and simulation techniques could be used to

train workers in regaining lost control. Teaching workers to recognize that they have
stepped into the hazard zone appears to be achievable through intensified and directed
training. However, this focus on worker training assumes that workers will always recall
what constitutes a safe or unsafe Situation as well as respond to perceived or actual risks

in the same manner.

The main goal of this research was to develop a methodology by which workers’
sensitivity to unsafe conditions and risk orientation (the tendency of a worker to work in
an unsafe condition despite knowing it exists) can be assessed prior to prescribing a
training program. Due to the high rate of occurrences, fall accidents were considered as

case examples. To arrive at this goal, the following objectives were proposed:

1) Develop a technique to assess the sensitivity and risk orientation of workers to

unsafe conditions.

2) Design and conduct a survey to determine the sensitivity and risk orientation of

workers at risk of fall accidents.

13.4 Proposal Overview

This. research report is comprised of five chapters. Chapter 1 provides a general
introduction to the state of safety in construction and the motivation behind this research.
The goal and Objectives of the research are also presented in this chapter. Chapter 2
provides a background on different accident causation models and also introduces signal
detection theory. which will be used extensively in this research. Chapter 3 outlines the

methods used to achieve the research Objectives. Chapter 4 discusses in detail the results

10

achieved using the methods developed in chapter 3. This is followed by chapter 5. which

contains summary. conclusions and contributions of the research.

Appendix A contains the questionnaire and interview format developed for
surveying ironworkers based on OSHA standards and the case study from NIOSH. This
questionnaire was developed to determine the response of ironworkers to unsafe
conditions. Appendix B contains the results of the survey. Appendix C contains the
results of the analysis of the data for the ironworkers using SDT and ROC. Appendix D
contains a normalized SDT table. Appendix E has distribution plots of d' and Bcurrent-

followed by Appendix F. with results from the multiple regression analysis.

11

Chapter 2
BACKGROUND

2. BACKGROUND I

For many years, reducing injuries and accidents has been a prime focus of government
organizations such as the Occupational Safety and Health Administration (OSHA) and
the National Institute for Occupational Safety and Health (NIOSH). Research efforts have
focused on developing accident causation models to unearth root causes of occupational
accidents. In this chapter. an overview of the different accident causation models and

theories is provided.

2.1 Overview of Accident Causation Models

The American industrial accident prevention movement started in 1892 when the safety
department of .loliet works of the Illinois steel company was formed. This was followed
by formation of the National Safety Council in 1913 (Zeller, 1986). Industrial safety or
the concept Of safety Started with a common objective in mind: the desire to reduce
injuries and to save lives and properties. With this Objective in mind. a series of theorems

was developed in the 19305 to define and explain accidents.

One Ofthese theorems is that proposed by Heinrich in his 10 axioms on industrial
safety, which helped many researchers to understand the accident process for the first
time. The first and most famous axiom stated that: “The occurrence of an injury
invariably results from a completed sequence Of factors. the last one of these being the
accident itself. The accident in turn is invariably caused or permitted directly by the

unsafe act Ofa person and/or a mechanical or physical hazard." (Heinrich et al.. 1980).

This axiom was the foundation for developing the “Domino Theory" which.

13

suggested that, to reduce injuries. fatalities. and property damage. the factors leading to

an accident must be prevented. Heinrich proposed the following five dominoes (see

Fig.2.1):

Ancestry and social environment: According to Heinrich, factors like recklessness.
Stubbomness and avariciousness are inherent, and the environment in which one is

brought up also may develop undesirable traits.

Fault of person: Fault or errors of person are due to a violent temper, nervousness.
ignorance of safe practices. etc.. which are inherent factors. These could lead to unsafe

acts or the existence of mechanical or physical hazards

Unsafe act and/or mechanical or physical hazard: Heinrich believes unsafe acts
performed by a worker or the existence Of mechanical or physical hazard directly leads to
accidents. These unsafe acts could be Starting machinery without waming. removal Of

safeguards, etc.

Accident: According to Heinrich. an accident is an unplanned event that leads to an

injury, which is due to an unsafe act.

Injury: Fractures. lacerations. etc. are injuries that result directly from accidents.
According to Heinrich. these factors are sequentially dependent and. if this sequence is

interrupted by eliminating one factor. the occurrence Of injury may be prevented.

Heinrich also defined accident prevention as “an integrated program. a series of
coordinated activities. directed to the control of unsafe personal performance and unsafe

mechanical conditions. and based on certain knowledge. attitudes. and abilities."

l4

(Heinrich et al., 1980). Until recently, the Domino Theory was universally accepted as

the real description of the accident process (Heinrich et al. 1980).

 

q.
g}

 
   

 

 

 

 

 

II .
s-
.8
3
.""|
2

Ac CI D E N '5;

 

’0
0

e‘
.-
2
I.“
2

§

5
5
J
S
0

O.
M

 

 

 

 

Figure 2.1: The Five Factors in the Accident Sequence

(Source: Heinrich. Petersen, Ross 1980)

Heinrich’s views were criticized for oversimplifying the control of human
behaVior’I in causing accidents and for some Statistics he gave regarding the contribution
of unsafe acts versus unsafe conditions (Zeller, 1986). Nevertheless. his work was the _
foundation for many others. Over the past thirty years the domino theory has been
updated with an emphasis on management as a primary cause in accidents. and the
resulting models were labeled as management models or updated domino models. Other
models have evolved separate from the domino theory but were Still based on Hclnrich‘s
work. These models are classified into different categories such as behavior models.

human factors models. system models, epidemiological models. decision models. etc.

(Heinrich, 1980).

15

Management models hold management responsible for causing accidents. and the
models introduced try to identify failures in the management system. Examples of these
models are the Updated Domino Sequence (Bird. 1974), the Adams Updated Sequence
(Adams. 1976). the Weaver Updated Dominoes (Weaver, 1971), and the Energy Release
model (Zabetakis. 1975). Two other accident causation models that are management
based the Stair Step model (Douglas and Crowe 1976) and the Multiple Causation

(Petersen 1971 ).

Human error theories are best captured in behavior models and human factor
models. Behavior models picture workers as being the main cause of accidents. This
approach Studies the tendency of humans to make errors under various Situations and
environmental conditions. with the blame mostly falling on the human (unsafe)
characteristics only. AS defined by Rigby (1970). human error is “any one set of human
actions that exceed some limit of acceptability." Many researchers have devoted great
time and effort to defining and categorizing human error (e.g.. Rook et a.. 1966; Recht.
1970: Norman. 1981; Petersen. 1982: McClay. 1989: DeJoy. 1990: Reason. 1990;

Wagenaar et al.. 1990; and O'Hare et al.. 1994).

The foundation of most behavior models is the accident proneness theory I lx'lumb.
1995). This theory assume that there exist permanent characteristics in a person that
make him or her more likely to have an accident. The theory was supported by the
simple fact that when considering population accident Statistics. the majority ol- people
have no accidents. a relatively small percentage have one accident. and a very small
percentage have multiple accidents. Therefore. this small group must possess personal

characteristics that make them more prone to accidents (International Labor Organization

16

1983). Other theories and behavior models include the Goals Freedom Alertness Theory
(Kerr. 1957), the Life Change Unit Theory (Alkov. 1972), and the Motivation Reward ,
Satisfaction Model (Peterson. 1982). For other behavioral models see Krause et al.
(1984). Hoyos and Zimolong (1988). Dwyer and Raftery (1991). Friend and Kohn

(1992). and Krause and Russell (1994).

The human factors approach holds that human error is the main cause of
accidents. However. the blame does not fall on the human unsafe characteristics alone.
but also on the design of the workplace and tasks that do not consider human limitations
and may have harmful effects. Therefore. these models study the effect of a particular
situation or environment on human performance, and the limitations humans have in
performing tasks are also addressed. Cooper and Volard (1978) States: environment and
human characteristics (both physical and psychological) as factors that contribute to
accidents and to human error. They have also briefly discussed the concept Of overload.
which is when an individual is subjected to more than he or she can handle (Peterson

31975). These ideas are common to the field of human factors engineering. Examples of
, human factor models include the Ferrel theory (Ferrel 1977). the Peterson model

(Peterson 1982). the McClay model (McClay 1989). and the Deloy model (DeJoy' 1990).

A system model recognizes the strong interaction between individuals. their tools
and machines. and their general work environment. Examples of such models are the
Firenze Model (Firenze. 1971 ) and the Ball model (Ball. 1973). Other examples are also
covered in Roland and Moriarty (1990). and Vincoli (1993). Epidemiological models
came about after the safety research community considered an accidentyto be an

epidemic. Epidemiology is the search for causes associated between diseases or other

17

biologic processes and specific environmental experiences. In 1961. Suchman proposed
an epidemiological model that suggests that an accident phenomenon is an “unexpected,
unavoidable, and unintentional act resulting from the interaction of host, agent, and
environmental factors within Situations which involve risk taking and perception of
danger" (Suchman. 1961). Surry developed a decision model based on the

epidemiological model of Suchman (Surry. 1974).

Based on the above-mentioned models a fishbone model was proposed by
Nishishma (1989) for understanding the process of accidents. According to his model the
four factors, which generated unsafe behaviors and unsafe states are: 1) human. 2)
equipment, 3) work and. 4) management (Suraji et al., 2001). Reason (1990) proposed the
-tripod model. which represent the interconnection between accident, unsafe acts and
resident pathogens. In his Study. the resident pathogens are latent failure such as error.

violation or technical failure.

Construction accident causation models based on variants of the above models
have been introduced in the literature by a handful of researchers. McClay' (1989)
'Iidentified hazards. human actions. and work overload as the three key elements of an
accident. The Study conducted by Whittington et al (1992) stated that poor management
decision-making and inadequate management control are major contributors to accidents.
Hinze (1996). in another study. stated risk of accidents might be generated by workers‘
distraction. caused by physical or mental distractiOn. This theory Of his is known as the
distraction theory. In this study of distraction he attributes accidents to production
pressures or other stress factors that distract workers from hazards and increase the

probability of accidents. Hinze et a1 (1998) developed a coding system that would

18

facilitate the categorization of injuries and fatalities. They believed if accidents were
categorized carefully, this categorization would provide a viable basis for implementing
on effective accident prevention program. Based on the OSHA causation code they
further classified it for modification. For example, fall accidents were coded in two,
categories: 1) fall‘from elevation and 2) fall from ground level. Stuck by accidents were
coded in to three categories I) stuck by equipment. 2) stuck by falling material and 3.)

stuck by material (other then falling) (Hinze et al., 1998).

In another Study by Suraji et al. (2001), a model was proposed which highlighted
the underlying complex interaction of factors in the causation process. This model
explains the constraints and responses of various parties involved in design and
construction, which might lead to an accident. The Accident Root Causes Tracing Model
by Abdelhamid and Everett (2000) identifies three general root causes—management
deficiencies. training. and workers' attitudes. The Constraints-response” model (Suraji et
al, 2001) suggests that project conditions and/or management decisions may result in an
inappropriate selection response on the part of workers. leading to an accident. Another
Study conducted by Mohamed (2002) explains the relationship between safe climate and
safe work behavior in construction site environments. A model was developed based on a
hypothesis that safe work behaviors are consequences Of the existing safety climate.
which in turn is determined by five independent sets of factors identified as management.
safety, risk, work pressure and competence. In 2002. Toole identified eight root causes
for construction accidents: lack of proper training; lack of safety equipment: deficient
enforcement of safety: unsafe equipment. methods. or conditions: poor safety attitude:

and isolated deviations from prescribed behavior.

Several past studies focused on preventing fall accidents using various tools and
methods. For example a Study proposed by Singh (2000) investigated fall accidents
occurring on low-rise roofs. From his study, he concluded that no single method or rule
of fall prevention would help in preventing falls from low-rise roofs, but stated that
prefabrication was one of the most promising method (Huang and Hinze 2003). Another
study by Duncan and Bennett (1991) reviewed the performance of various fall protection
systems and Stated that both active and passive measures are useful in reducing fall
injuries. Vargas et al. (1996a. 1999b) developed an expert system that would help to
analyze cases of construction falls by using fault-tree analysis and stated that all forms Of

fall protection can be inadequate in different circumstances.

Most of the above mentioned models are theoretical and they lack details about
those factors which make Significant contributions. so it’s hard to follow or implement
these models in real scenarios. Also none of these models consider or address the
organizational and operational factors which may increase the risk of accidents. To
overcome the above-mentioned problem. Rasmussen proposed a model. which helps to

understand the accident process in a more realistic way.

2.2 Rasmussen’s Theory of Cognitive System Engineering

Many accident causation models have been developed. as discussed in the above section
Of this chapter. Despite the contributions of these accident causation models they lack
proper understanding of the accident process. none of these models considers the
dynamic nature of construction work and that accident scenarios differ in how they occur
from site to site. It is not possible to predict every scenario and have rules for each under

the dynamic conditions. SO a new approach is necessary to represent the system behavior

20

one which does not focus on human errors or Violations. mechanical failure or
management, but an approach which understands the mechanism of an accident in an
actual and dynamic work environment (Rasmussen, 1997). The concept of following
preset rules can be applied in a well-structured environment where nothing can go wrong
other then some fixed scenario. but this is not possible under dynamic conditions. To
include this missing dimension. Howell et a1. (2002) proposed a new approach to
understand construction accidents based on Rasmussen’s theory of cognitive system

engineering.

Rasmussen in his theory of “Cognitive System Engineering" argued that there are
no fixed Stop rules for tracing the cause of events. Rather, in a normal case, the analysis
stops when an explanation makes sense from the perspective of the analysts (Howell et

al., 2002). Rasmussen identified Six common perspectives (Rasmussen 94).

Common sense explanation of what happened: Analysis stops when the act or event

that offers reasonable explanations and is familiar to the analyst is identified.

Understanding human behavior: The scientists perspective. This approach seeks to
understand the inner mechanism ofhuman behavior. The stop rule iS to identify any actor
in the flow of accident events that did not maintain control. even though he or she may

not have Started the flow. and then to explore his or her cognitive process.

Evaluating human performance: The reliability analyst‘s perspective. This approach
attempts to predict the effects of likely errors on large system performance. This

approach is very difficult to apply in less Structure scenarios and also is more complex as

humans adapt to the Situation and often push for performance beyond that predicted by

the designer.

Improving performance: The therapist‘s perspective. The availability of a cure
determines when the search for a cause Stops. The bias of the therapist will likely affect
the selection-trainers will see the problems as a lack of training. while psychologists or

safety Officers may see it as a lack of motivation or awareness.

Finding somebody to punish: The attomey‘s perspective. The stop rules are to identify a

person who was in control oftheir behavior. i.e., guilty of the act.

Improving system configuration: The designer‘s perspective. Here the objective is to
find changes in the work system. which will improve its performance. This is tricky
business as the systems are "designed" by a number Of people with different perspectives.

from le islators to machine designers.
g -

The new approach or theory proposed by Rasmussen States that organizational
and individual pressures push people to work in hazardous situations. These pressures
' defeat efforts to enforce safe work rules. specifically in a changing work environment
such as in construction. 'l‘herefore. this approach emphasizes the need to train workers to
be conscious of hazardous work environments and to engage the work with better
planning and appropriate protection. in a way very Similar to how fire fighters engage

hazardous situations.

The framework proposed by Rasmussen. Shown in Figure 2.2. explains more

clearly the relation between the individual and the work environment. In his theory.

5')

~-

Rasmussen stated that the workspace within which the worker can move freely is
bounded by administrative, functional and safety related constraints. These constraints

push workers to work in the hazard zone. beyond which it is no longer safe and accidents

 
     
   
    

,OCCLII'.
Boundary to
functionally acceptable
behavior
Loss of '3 , , .
Control zone Mlgratlon toward Boundary to finanCIal
least effort breakdown
Hazard
Zone
Local C B n'an o e nts
~ row I m v me
acetdents . .
Safe Within space of proper
Zone task performance
\ Management
pressure towards
Boundary of safe efficiency Boundary to
behavior as defined by / Unacceptable workload

safety campaigns

\ \

Figure 2.2 Migration of Work Towards Loss of Control
(Source: Howell et al., 2002).

Rasmussen‘s model leads to a three-Step approach to safety. as shown in Figure

2.2 (Howell et al.. 2002).

Zone 1- The safe Zone. He suggested that one could enlarge this safe zone

through proper planning ofthe operation.

Zone 2 - At the Edge Zone. He suggested that this boundary or zone should be

made visible beyond which work is no longer safe and teach workers how to recognize

IJ
DJ

the boundary. He also suggested teaching workers how to detect and recover when hazard

is released at the edge of control. This could be done through ‘simulators”

Zone 3— Over the edge. This is the zone where accidents occur. so he suggested

designing ways to limit the effect of the hazard once control is lost.

‘

According to Rasmussen. accidents result from a "loss of control" when work
migrates from the boundary of functionally acceptable behavior to the loss Of control
zone. He also believed that the worker him/herself is the best person to judge the
boundaries of safe work. SO instead of forcing workers to follow the rules to stay in the

safe zone, Rasmussen suggested that the workers be trained to:
0 Identify in which zone they are working,
0 Identify hazards.
0 Prevent hazard release. and
0 Recover when hazards are released.

While counterintuitive. Rasmussen's recommendation to train workers to deal
with hazards and recover from scenarios when control is lost recognizes that workers will
frequently work in the hazard zone due to various reasons and pressures. Management
pressure and seeking less effort are realistic examples of what may push workers to the
hazard zone. Rasmussen Still maintains that safety and performance will increase if the

safe zone is enlarged with proper planning.

Rasmussen‘s model explains the process of accidents. The following section of
this chapter will discuss Signal Detection Theory (SDT), which had been applied mostly

in the manufacturing industry to determine the performance of the operator.

2.3 Signai Detection Theory

In the manufacturing industry. quality inspections are performed on products to reject
defective ones. A perfect quality inspection process would be able to identify and reject
all the defective products. This is seldom attained despite the use of sophisticated
equipment to perfoml the inspection instead of using human inspectors. The inspection
problem is also found in other industries or job situations such as a radiologist detecting a
tumor on an X-ray plate. airport security guard detecting weapons. and for the purposes
of this research construction workers detemtining if the work conditions are safe or

unsafe.

The number of defective products. diseases. or weapons. etc. that escape detection
(misses) and non-defective ones that are rejected (false alarms) givesya measure of the
effectiveness of an inspection process. These two measures have also become the basis
for characterizing the sensitivity of the Operator performing inspection. Researchers have ,
dubbed the framework leading to such characterization as “Signal Detection Theory"

(SDT).

SDT is applicable in situations where two discrete states of the world (signal and
noise) cannot be easily discriminated. In such Situations. a human operator (or machine)
is faced with the task Of identifying one of the states. If the State of the world is a Signal.

e.g.. a defective product. the response of the Operator (or machine) is either ‘yeS' the

product is defective (a HIT), or ‘no’ the product is not defective (a MISS). If the state of
the world is noise, e.g.. the product is not defective, the response of the operator (or
machine) is either ‘yes’ the product is defective (a FALSE ALARM), or ‘no’ the product
is not defective (a CORRECT REJECTION). These situations are represented as Shown

in Table 2.1. Clearly. a perfect result would not have any false alam1 or misses. but in

real life this is not possible.

 

State of the World

 

Signal Noise

 

f Yes Hit False Alarm

 

l
| .
Response No Miss Correct Rejection

 

 

 

 

 

Table 2.1 The Four Outcomes of Signal Detection Theory (Wickens, 1992).

‘In a Signal detection task. operators sometimes have response bias and are prone
to say ‘yes‘ more often than they Should. thereby detecting most of the signal but also
producing many false alamls. AS other response could be conservative by saying ’no'
and producing few false alarms but missing many of the Signals (Wickens. 1992).
Depending on the task. a conservative approach with fewer false alarms may be better

than not missing any signals while having many false alarms.

Assuming that a signal indicator or strength has a nomlal distribution. the
information in Table 2.1 could be graphically represented as Shown in Figure 2.3. .\'c.
Shown in Figure 2.3. represents the critical level where an observer decides the nature of
a Signal. In other worlds. Xc represents the "mental" cut-Off the Observe uses to decide

whether to say ‘yes‘ there is a signal (a hit). or ‘no‘ there is noise (correct rejection).

In Figure 2.3. the shaded portion on the left of Xc represents the signals missed by
the observer. The striped portion to the right of Xc represents the signals the observer
incorrectly considered as hits. i.e.. false alarms. The change in the position of Xc
determines the respective proportion of misses to false alarms. For example, if Xc cuts
more into the Signal side. then most responses will be ‘no’ resulting in numerous misses
and fewer hits and false alarms. This strategy is considered conservative. If Xc cuts more
to the left. most responses will be "yes" resulting in fewer misses but more of false alarm.
This indicates a risky Strategy.

Figure 2.3 Distribution of Detection Theory
(Wickens, 1992)

 

[\ /\
No

 

Noise

      
   

Z<i
mi Correct
I— . .
c Rejection
E
C.

P (X S)

 

 

 

v I ‘
Miss / XI- \

False alarm
The tendency of an observer to follow a conservative or risky Strategy is

measured using a parameter termed the response criterion or likelihood ratio. and denoted
Bcumm. This parameter has also been termed the judgment or decision criterion of the

Observer. Mathematically. as Shown in Figure 2.3. Byumm is the ratio of the ordinates

P (X/S) and P (X/N) for a given level of Xc. P (X/S) and P (X/N) represent the
conditional probability of Xe given a Signal and the probability of Xc given noise.

respectively.

Bcumm is calculated using Equation 2.2 (Wickens, 1992).

P(X/S)

,6 current =
P( X / N)

A high value of mem indicates a high number of misses, whereas. a lower one
will generate more false alarms. Because of inter-observer variability with respect to the
choice of Xc. evaluating the results of multiple Observers requires normalization of the
value of Bcumm or comparison to an optimal value. The optimal value of B has been taken
as the value corresponding to a minimum number Of errors, i.e. minimum misses and
false alarms. Mathematically. this vaer is the ratio of the probability of noise. P (N). and

the probability Ofa Signal. P (S). Equation 2.3 (Wickens, 1992) gives this ration.

P(N)
I: -——- ( .
5 0p P(S)

IJ
IJJ
V

After finding the value of mem and Bum. the pair are compared to determine
whether an observer is following a risky or conservative Strategy. If mem is greater than
the value of 0”,“. then Xc is positioned more to the right. resulting in fewer false alarms
and more misses. According to the SDT literature. Observers with such a mental-cutoff
require more evidence to say 'yeS‘ a part is defective or a tumor exists. Under SDT. this

is considered a conservative strategy for operators to adopt because Of the consequences

28

and actions triggered after a false alarm results. such as rejecting a non-defective product
or performing unnecessary medical procedures. When Bcumm is less than 00‘”. then Xc is
positioned more to the left. resulting in more false alarms and fewer misses. Based on
SDT, this indicates that the Observer needs less evidence to say ‘yes‘ a part is defective or

a tumor exists. Therefore. this Strategy is considered a risky Strategy.

Another important measure of an observer‘s performance in Signal detection tasks
is sensitivity to the Signal and the noise. This is measured by the degree of separation
between the means of the two distributions shown in Figure 2.3. denoted as d’. A high
value of d' indicates a high degree of separation and, thus, high Observer sensitivity. Data

from myriad tasks indicate that d’ ranges in value from 0.5 to 2.0.

The value of d' is determined by adding the two values 21 and Z; Shown in Figure
2.3. 21 and Z3 represent the value of the standard normal Variable corresponding to the
probability of a false alarm and probability of a hit. respectively. The values are readily
available from standard tables. The application of SDT will be demonstrated using an

example Ofa typical inspection process.
Example 2.1:

A manufacturer produces DC motors using a process that generates 5% defectives. In
response to increasing customer complaints. the manufacturer institutes a final inspection
system that finds 80% of the defective motors at the expense of falsely rejecting 1% of

the good motors. Determine the sensitivity of the operator and the Strategy adopted.

Solution: From the information given. the following probabilities can be deduced:

P (Noise) = P (product is not defective) = 0.95

P (Signal) = P (product is defective) = 0.05

P (Hit) = 0.80 P (Miss) = l-P (Hit) = 0.20

P(FA)=0.01 P(CR)= l-P(FA)=0.99

Table 2.2 represents the inspection process with its possible outcomes.

 

 

 

 

State of the World
Signal Noise
(Defective Product) (Good Product)
HIT = 80% FALSE ALARM = 1%
Response Yes
“5 moi“, MISS = 20922 CORRECT REJECTION = 9992
defective. ) NO

 

 

 

 

 

 

Table 2.2 The Four Probabilities

Calculation of the sensitivity. i.e.. the value of d' involves the standard normal

values Z; and 2:. Using P (FA) and P (Miss). the values onI and Z; are:

2; = 2.326 and 7.3 = 0.842

d'=Zl +7.3

‘. d' = 2.376 r 0.842 = 3.168.

This indicates a high degree of separation between the signal and the noise

distribution. i.e.. the inspector has high sensitivity.

As indicated by Equation 2.2. calculating Bcumm requires the determination of

P(X/S) and P(X/N). However. Figure 2.3 indicated the following:

P(X/S) = Ordinate corresponding to Z3

P(X/N) = Ordinate corresponding to Z]
Using the tables.

Ordinate corresponding to 23 = 0.28
Ordinate corresponding to z] = 0.027

.'.mem = 0.28 / 0.027 = 10.37-
Using Equation 2.3. Bap, is easily calculated as:

- ,- 0.95
flopr: mitt: 3 16 “I” = 30— : 19

U1

Clearly Bum," < [1m which indicates a risky strategy.

This means that the '

inspector’s cut-Off level. Xc. is positioned more to the left. i.e.. cuts more in the signal

distribution.

The above example illustrates how the sensitivity and risk orientation of a worker

can be determined using SDT. This information sets a benchmark against which the

effectiveness ofnew training can be assessed. Essentially. this information would make it

possible to determine if a worker‘s sensitivity and risk orientation to safe and unsafe

conditions increased. decreased. or remained unchanged. Ultimately. the use of SDI will

result in increasing workers abilities to judge the boundary beyond which work is no

longer safe.

The next section will discuss another approach to representing the analysis of
SDT. the ROC curve. The ROC curve helps to understand the joint effect of sensitivity

and response bias

2.4 ROC Curve

This method of graphical representation is known as the Receiver Operating
Characteristic (ROC ) curve and is helpful to portray the equivalence of sensitivity across
changing levels of bias. i.e.. to understand the joint effects of sensitivity and response
bias on the data from a signal detection analysis experiment (Wickens, 1992). The ROC
curve is plotted on a Single graph using the values of P(Hit) and P(FA) obtained from the
SDT analysis. When the same experiment is repeated several times and each time the
response criterion is changed. a series of different points are produced. When these points
are connected a ROC curve emerges as Shown in Figure 2.4. For more sensitive worker
the ROC curve will be more curved as compared to other workers. This is a theoretical
representation because it is hard to repeat the same experiment to get different points in ,

real life.

The alternative way of plotting the curves shown in Figure 2.4 is by plotting the
curve on probability paper. as Shown in Figure 2.5 (Wickens. 1992). This representation
has its advantage. as the bowed lines of Figure 2.4 now become straight (Wickens. 1992).
The Value of P(Hit) and P(FA) could be replaced with Z scores. standard value or scores
from the Standard SDT table know as Z(Hit) and Z(FA) respectively. For any given point.

d' (sensitivity) is equal to [1111- Z(FA) (Wickens. 1992).

 

.. Low Beta

0.95 r ---------------

0.70 r ----- 4 Beta =1

0.60 ”"— '"I-

 

Probability

 

0.20 I
High
Beta

 

p—————_——

 

, 1
0.05 0.30 0.40 0.80
Probability of False Alarm

Figure 2.4 Theoretical Representation of the ROC Curve (Wickens, 1992)

 

Z(Hit)

_——

 

 

 

l

l
- L
Z(FA)

Figure 2.5 The ROC Curve on Probability Paper (Wickens, 1992)

It is important to understand the difference between the theoretical representation
Of the ROC curve discussed above and the actual empirical data collected in an SDT
experiment. The representation Shown in Figures 2.4 and 2.5 is continuous and smooth.
while actual data collection provides discrete points. or due to some limitation it might

not be possible to get more then one point. lnisuch circumstances. a measure called P(A).

33

representing the area under the ROC curve. is an alterative which can be used to measure
sensitivity as Shown in Figure 2.5 (Wickens, 1992). This area under the ROC curve
represents the area to the right and below the line segments connecting the lower left and
upper right corners of the curve as shown in Figure 2.6. The area represented by the
formed triangle is A and P(A) represents the sensitivity (d') of the respondent. The area

P(A) is calculated using equation 2.4 (Wickens, 1992).

P(A):

 

P(Hit ) + [l — P(FA )]
2.0

Equation 2.4 (Wickens, 1992) was used to determine the P (A) for the example

discussed in section 2.3 of this chapter. which proddced a value of 0.9. This value Of P

(A) represents the sensitivity (d') for the Operator. The sensitivity calculated using

Standard SDT is 3.16. The difference between the two values of d‘ is caused by the fact

that the ROC curve portrays the joint effects ofresponse bias and sensitivity

 

P (1111) (A)
l
I
l
I

l

P (FA)

 

 

 

Figure 2.6 Example of Measure of P(A) (Wickens. I992)

Chapter 3
METHODOLOGY

35

3. METHODS

The first chapter of this thesis gave an overview of construction fatalities and the
contribution of fall accidents to these fatalities. The problem area, goal and Objectives of
the research were also discussed in this chapter. In Chapter 2, the history and evolution of
accident causation models and theories were discussed. The new approach to accident
causation, as proposed by Rasmussen, was presented. Signal detection theory was also
introduced which will be further utilized in this research. In this chapter, methods to
achieve the objectives of the research will be discussed. The application of signal

detection theory will also be demonstrated.

3.1 Introduction

The main goal of this research was to develop a methodology by which workers‘
sensitivity to unsafe conditions and risk orientation (tendency of a worker to work in a
condition despite knowing its unsafe) can be assessed prior to prescribing a training
program. Due to the high rate Of occurrences. fall accidents were considered as case

examples. To achieve the research goal. the following Objectives were articulated:

1) Developing a technique to assess the sensitivity and risk orientation of workers

to unsafe conditions.

2) Designing and conducting a survey to determine the sensitivity and risk

orientation Of workers at risk of fall accidents.

In the following sections of this chapter. these Objectives will be discussed in detail

36

3.2 SDT and Unsafe and Safe Construction Conditions (Objective 1)

An assessment of the construction workers sensitivity and their risk orientation to unsafe
conditions was performed. after some modifications to the Signal detection theory (SDT).
SDT iS used in this research because of the fact that there is Similarity between the
response of the Operator in identifying defective and non-defective parts and the
construction worker's response in identifying safe and unsafe conditions on Site. Also.
SDT is the only method that will help to determine both the sensitivity and risk
orientation of the construction worker. with some modifications to the theory. Once
performed. this assessment could be used to give guidance to workers on how to enhance
their abilities to identify the boundary beyond which work is no longer safe. Signal
detection theory was discussed in detail in chapter 2. In this section. application and

tailoring of the theory will be presented.

Similar to a detection task in other industries. construction workers are expected
to identify whether the conditions under which they are working are safe. In SDTFthe
State of the world is represented by Signal and noise. From a construction safety
. standpoint. the State of the world is either a "Safe condition" or an “Unsafe condition"

which correspond to the noise and signal States of SDT.

On the one hand. a worker faced with a "safe" condition and asked whether the
condition iS unsafe has one of two possible responses. namely ‘yes‘ the condition is
unsafe (false alarm). or 'no‘ the condition is safe (correct rejection). On the other hand. a
worker faced with an “unsafe" condition and asked whether the condition is unsafe has
one of two possible responses. namely 'yes~ the. condition is unsafe (hit). or ‘no‘ the

condition is safe (miss). Figure 3.1 shows the SDT matrix for these scenarios.

The ideal scenario for a given number of safe and unsafe conditions is for a
worker to correctly identify them all. This may happen for some workers but certainly not
for all. Some workers will incorrectly consider a condition safe while it is unsafe. and
vice versa. Signal detection theory allows the determination of the sensitivity of workers
to unsafe conditions as well as their inclination (bias) to consider a situation as unsafe
while it is not. For construction workers, it is desirable to minimize the number of misses
(considering a condition safe while it is unsafe), at the expense of having more false
alarms (considering a condition unsafe while it is safe). This is because a miss is more

likely to lead to serious injury or death.

 

 

 

 

 

State of the World
Unsafe Safe Condition
Condition (Noise)
Signal) ‘

HIT FALSE ALARM
Response Yes
(Is condition ' y . .
unsafe?) No MISS Correct Rejection

 

 

 

 

 

Figure 3.1 The SDT Matrix for Detection of Unsafe Conditions in Construction

As explained before. worker sensitivity to unsafe and safe conditions as well as
the inclination to regard a condition as a safe or unsafe can be assessed using the SDT
parameters d' and (iwmm High values Of d' indicate high sensitivity in differentiating
between safe and unsafe conditiOns. Conversely. low values of d' indicate that a worker
needs more training to better differentiate between safe and unsafe conditions. Regardless
Of the value Ofd'. the mental cutoff used by a worker to decide the State of a condition is

given by the value of Bum," with respect to Born. However. considering the

38

implementation of SDT is construction. a modification is necessary in interpreting the
values of Bcuncm and Bum. As discussed before. if the value of Bcuncm is greater than that of
Bow, then fewer false alarms and more misses will result; in manufacturing industry. SDT
usual application this strategy is considered conservative. However, in construction, the
cost of a miss could result in a fatality or a serious injury. Therefore. it is a more risky
Strategy to have fewer false alamls and more misses. Similarly. a value of Bcumm smaller
than Bopl, indicating that more false alarms and fewer misses result. will be considered a
risky strategy in STD normal field of use. For construction. this would be considered a

conservative strategy.

Undertaking the assessment of worker sensitivity to unsafe and safe conditions as
well as the inclination to associate a condition with a safe or unsafe requires the
determination of SDT responses (hit. miss. false alarm. and correct rejection) to a number

of safe and unsafe conditions. The second objective of this research addresses this issue.

3.3 Survey Development and Its Analysis Using SDT

Assessing worker performance in detecting unsafe and safe condition in real time is both
dangerous and infeasible. The alternative is to design a survey (questionnaire) that places
the worker in know hypothetical safe and unsafe conditions and asking the worker to
identify whether the condition is safe or unsafe. This survey was developed by referring
to OSHA standards and case Studies of construction fall accidents reported in a NIOSH
report. The survey contains 21 questions involving conditions where fall hazards may

exist. To keep the scope reasonable. the survey was performed on ironworkers only. This

39

choice is primarily based on the fact that ironworkers (see Chapter I) suffer from fall

accidents more than any other trade.

The questionnaire developed for the survey is Shown in Appendix A. For each
question, the worker has to choose from one of three responses: 1) An Unsafe Condition
or 2) A Safe Condition or 3) I Don‘t Know. From the response Of the worker it would be
determined how the worker would react if he/She encounter typical safe or unsafe

conditions.

Based 011 the responses of the worker, the number of hits. misses, false alarms.
and correct rejections were determined and converted to probabilities. This facilitated the
determination of the sensitivity to unsafe conditions and the risk orientation of the

workers.

To illustrate how a response was mapped to a hit. miss. false alarm. or a correct
rejection, 3 sample question is shown below which depicts a safe condition. If the
worker‘s response to this was "An Unsafe Condition" or “I Don‘t Know". then this
indicates that the worker incorrectly considered the condition as unsafe or was not sure of
what it was. i.e.. - a "false alarm". The response "I Don't Know" is considered a miss if i

the condition portrayed by the question was unsafe.

INTERVIEW QUESTIONS

Please choose one of the classifications that fit the following Conditions:

1) Working on a scaffold 8 feet above the lower level without a guardrail system.

All Unsafe Condition

I: A Safe Condition

40

[:1

To illustrate further the type of analysis that was performed based on the response
of the worker to a 30 question survey, with 24 safe condition scenarios and 6 unsafe
condition scenarios. Table 3.1 shows hypothetical responses from one of the workers who

participated in the research.

I Don‘t Know

 

 

 

 

 

 

 

 

 

State of the World
Unsafe Safe Condition
Condition (Noise)
gSignal)
Response Yes Hit = 4 False Alarm = .9
(Is condition _ = 7 . . = 7
unsafe?) No Miss _ Correct RCJCCIIOD -1
Table 3.1 Sample Survey Analysis Results
Note that:

P(Noise) = P(safe conditions) = 24/30 = 80%

P(Signal) = P( unsafe conditions) = 6.2’3 = 209/6.

P( Hit) = 4/6

P(FA) = 3324

Calculation of the sensitivity. i.e.. the value of d'. involves the standard normal values 7,,

P(Miss) = l-P(Hit) = 2/6

P(CR) = l- P(FA) = ill/24

and 22. Using P(FA) and P( Miss). the values onI and Z: are:

Z. = 1.2] and Z; = 0.440

4]

 

and d' = Z. + 23

d' = 1.21 + 0.440 = 1.65.

This is indicates a moderate degree of separation between the signal and noise
distributions. i.e.. the worker has moderate sensitivity. In this scenario. for a perfect score
of 12 “hits‘ and 9 "CR." the value of ideal d'gm = 4.6. and in the worst case scenario of

no “hits" or “CR." the value ofd'gm = - 4.6

Ordinate corresponding to 22 = 0.362

Ordinate corresponding to Z. = 0.194

Using Equation 2.2. BMW... = 0.362 / 0.194 = 1.865

’3

Using Equation .3 Bop. = P (Noise) / P (Signal) = 0.8/0.2 = 4

Clearly Beam... < 0...... which indicates a conservative strategy. Despite that with this

strategy causing the worker will have more false alarms. fewer misses will result.

3.4 Analysis with ROC

The above section discusses how SDT is used to analyze the result of the survey. This
section will discuss how the responses of the workers participating in the research would
be analyzed using ROC. This will be illustrated with the help of the same example as is

found in the previous section.

P (Hit) Z 456 = 0.67 P (Miss) = l-P (Hit) = 0.33

P(FA)=3/24 =0.125 P(CR)=l-P(FA)=0.87S

From Equation 2.4: P (A) = 0.77

As discussed in chapter 2. section 2.4. P (A) represents the sensitivity (d') for the
worker. The sensitivity calculated using standard SDT is 1.65. The difference between
the two values of d' is caused by the fact that the ROC curve portrays the joint effect of

response bias and sensitivity.

It is worth noting that in an ideal scenario with perfect scores of 12 “hits" and 9
“CR”. the value d'RUC = 1 represents an ideal value. and in the worst-case scenario with
no “H its" or “CR" the value of d'RQC = 0. Thus. a score of 0.77 is quite high with respect

to the ideal. This will be further discussed in chapter 4.

43

Chapter 4
SURVEY RESULTS AND ANALYSIS OF DATA

44

4. SURVEY RESULTS AND ANALYSIS OF THE DATA

4.1 Introduction

In this chapter. survey data are presented and analyzed. The analysis of the data uses SDT
calculations to determine the sensitivity and risk orientation of ironworkers. The data is

analyzed following the steps discussed in section 3.3 and 3.4.

4.2 Data Collection

As discussed in chapter 3. construction accidents. and fall accidents in particular. are of
major concern in the construction industry. as many lives are lost and business suffers.
Many construction accident causation models have been developed, but accidents still
occur with fall accidents representing a significant percentage of the whole. Therefore.
this research focuses on understanding the risk orientation of ironworkers to unsafe
condition such that worker specific training could be developed. The ironworkers were

selected because oftheir high risk of fall accidents.

Using Signal Detection Theory (SDT). a survey was developed to investigate how
an ironworker assesses a hypothetical scenario that portrays either an unsafe or a safe
condition. The survey describes 21 scenarios. which are developed based on OSHA fall
protection regulations and steel erection codes and also on fall accident cases reported by
NIOSH (NIOSH. 2000 "case repot on workers‘ death by fall"). The developed survey is

found in Appendix A.

There was no restriction on age. years of experience or any other criteria for the
ironworkers who volunteered to participate in this survey. All the ironworkers were

familiar with the OSHA standards and have been through formal training. All 42

45

ironworkers who participated in this research were members of local 25 (Lansing. MI). In
this research, an effort was made to survey at least 30 ironworkers, as using 30 or more

sample points allows the use of the normal distribution for results.

The ironworkers were asked to select one of three responses to each question on
the survey. The responses were then compared to the correct responses and further

analyzed using SDT to determine the sensitivity and risk orientation of each ironworker.

4.3. Sensitivity and Risk Orientation of the Ironworkers by SDT
The response of each ironworker (Le. a hit. miss. false alarm or correct rejection) is

provided in Table 4.1. Survey details are provided in Appendix B (Table 8.1).

The survey had 21 scenarios with 12 unsafe and 9 safe conditions. If the
ironworker correctly identified an unsafe condition as “An Unsafe Condition" then this
was considered a "HIT" and is represented by "H" in Table 4.1. If he or she incorrectly
identified an unsafe condition as " A Safe Condition“ or as “1 Don’t Know" then it was
considered as a "MISS" and is represented by “M" in Table 4.1. Similarly. a “correct
rejection” represented by “CR" in Table 4.1. results when the ironworker correctly
identified a safe condition. A "false alarm" represented as "FA". was assigned if the
condition was safe but the worker identified it as "An Unsafe Condition" or "I Don‘t

Know”.

46

28.3.58..— 2: be 52.5 2: .3 3:53— .6. use...

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

I I I I I I I I I I I I IIIIs... I I I I :0
E E 5 5 5 E 5 E E E E 5 5 5 E E 5 5 5 5 5 E and
E E E E E E 5 E E E E E E 5 «HA... E E E E E E 20
5 5 5 5 5 5 5 Eu 5 5 5 5 5 5 5 5 5 5 5 5 5 E 20
IzIIzIIIIEEEIEIIIIIIIEEM
E E 5 5 5 E E E E E E 5 5 E E 5 5 E 5 5 5 5 so
SIIIzIzIzzIIIIIIIZEEIZEO
5 E 5 5 5 E 5 E 5 5 5 5 5 5 E 5 5 5 5 5 5 5 z
II2.22.2I22222222....2IIII20
I I I s. I I I I I I I I I I I I I I I I I I :9
5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5-5 5 :m
5 5 5 5 5 E 5 E E E E 5 w..wt~...w.kw.5 5 5 mmm 5 5 Ed
I s. I I I I s. I I I I I I I I .2 .2 I I I .2 I so
I. s. I I s. I I I I I .I I 1:11.112 4211.4. 1:12 I I I rail
.2 s. s. I s. I I I I I I ..._1.... :fl ..Hi.@.fl.-.-.-zl. :14..- I -Wlmzolm
I z z I z ___ z __ _,_ __ __L-._..-.,-_._,..-._._._.._.. .._.-._._...._.%_._..;__ 2 a a -
I I I I I I I I I I I I I I I I I I I I I I «.9
I I I s. I I I I I I -1... I I ......ll....i...l. ....1£..:..1..fl I I "ml
E E E E E E E E E E E E E E <I.. E E E A... E E E .5
E E E E E E 5 E E E E E E E .E E E E E 5 E E No
IIIzIIIIIIIIIIZEIIIzII.0
N? a; a? as a? :3 S: m? z} .2; NE :3 S: a; .3 t, as m: 2, 3 N: .3
9.3—.83 52.0

 

 

 

47

32.5.55 U mug—53:9.— 2: he hot—5 2: he 2.58M 2. «Barr

 

III.2.IIIIIIIIIIIIzIIzzo

 

E 5 5 5 E E 5 5 5 E 5 5 E 5 E 5 5 E E 5 24.

 

<u. <r._ <L <n— <..I <u_ <M <..._ «E <M <,._ <,._ {M <«.. <.._ <L <I <.M <..I <.I 201

 

MU MU <.._ MU MU <2 <L MU MU MU MU MU MU MU MU MU <1. MU MU «C 50

 

 

IT

I S— I I I I2 I I I S. I I 2 I E I I I I I Cm

 

<n_ <h_ <n_ MU MU MU <n_ MU MU <..M MU MU <r._ MU <.I (Q <.._ MU <L <.._ Em

 

IIIIIIEIIEIEEIIIIIEIEM

 

MU «Q <.._ MU MU MU <M MU MU <n. MU MU <.._ MU MU MU MU MU E <.._ :0

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

48

 

 

 

 

 

 

 

 

 

I I .2 .2 2 .2 .2 s. .2 s. I I .2 .2 .2 I .2 I s. I lmdl
I I I I I I I I I I mi: I I I I.iml I I I :0
5 5 5 5 E 5 5 5 5 5 5 5 5 5 5 5 E E 5 5 :0
5 5 5 5 E 5 E 5 5 E 5 5 ad 5 E 5 E E E E 20
I I I I I .I I I I I I. .2 I I I I I I I I 1.31,
i... s. .2 s. I I I I I s. I II I I I I .2 -I I I .3
II .2 .2 .2 I I I I I 2 15.. I I i......i...li:. I I I -lIlquH
I I I I I I I I I s. 2 iFLPl...lr-...le I I I I om
I I .2 s. I I I I I I .. I I I .1.,.l his... I I I I Imol
I I I I I I I I I I I -m. .m..f.....i.m .I. ..... I I I 4dr
E E E E E E E E <..I-.fl.lfl.. E .E E- <,......5..<... E E E 3... .
E E E 5 E E E E 5 E 5 5 E E.-fl-.¢fi.3.fl E E Na
I I I I I I I I I I I I I I I 1.2 .2 I I I .o

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

N35 3:5 345 on}, um?» um; on?» mm} an?» mm} mm?» :3 an} on an?» In) 3.) mm} VNJ an}

 

 

23.5.5

 

 

 

To illustrate the additional analysis performed on the collected data for each
worker, the results of worker “W1" are used. As shown in Table 4.1, "wl" had 10 “Hits“
and 5 “False Alarms." This indicates that this particular worker correctly identified
unsafe conditions 10 times out of total 12 possible hits, and that he or she incorrectly
considered 5 out of 9 safe conditions as unsafe. Based on this information. the
corresponding number of hits. misses. false alarms and correct rejections are listed as

shown in Table 4.2.

 

 

 

 

State of the World
Unsafe Condition Safe Condition
(Signal) (Noise)
Y HIT = 10 . FALSE ALARM = 5
Response es
(IS condition unsafe?) No Miss = 2 Correct Rejection = 4

 

 

 

 

 

 

Table 4.2 Matrix Showing the Responses for Worker wl

From Table 4.2. the probability of a hit. a miss. a false alarm and a correct

rejection can be calculated as follows:

Note that:

P(Noise) = P(safe conditions) = 9x21 = 0.43
P(Signal) = Pt unsafe condition) = 12/21 = 0.57
P(Hit) = 101’12 = 0.83 P(Miss) = l-P (Hit) = 0.17

P(FA): 5’9 =0.56 P(CR)=1- P(FA): 0.44

49

After obtaining the probabilities. the sensitivity (i.e.. the value of d') and the risk
orientation (i.e.. Bum...) may be calculated. From the normalized SDT table (Appendix D)
normal devastation values. Z. and 22. are determined based on P (FA) and P (Miss).

respectively.

ForP(FA)=0.56:Z. =-0.15; forP(M)=0.l7223=0.95
C") 'ZZ]+Z2

d' = —0.15 I 0.95 = 0.80

Compared to the ideal d' of 4.6. this d' indicates a low degree of separation
between the signal and noise distribution, i.e.. the worker has low sensitivity. To
determine the risk orientation of the ironworker. the ordinates corresponding to Z. and Z;

are determined from the same normalized SDT table (see appendix D).

Ordinate corresponding to Z. = 0.394

Ordinate corresponding to Z; = 0.253
Using Equation 2.2. mem = 0.253 / 0.394 = 0.64

Using Equation 2.3 Bum = P (Noise) P (Signal) = 043/057 = 0.75

Because Bum... < (1...... this indicates a conservative strategy. With this strategy

worker will have more false alarms and fewer misses.

To determine the sensitivity (d') of the ironworker while taking into account the
effect of the response bias. the ROC curve will be used. The area under the ROC curve is

obtained from the following formula:

50

rvHa)+[1—erAn
20

P(A): (4.1)

 

As an example for using Equation (4.1), consider the response of worker wl.

shown in Table 4.1. Using P (Hit) = 10/12 and P (FA) = 5/9. the

083+[1—056]=O
20 '

64

 

P(A):

The value given by P (A) will be denoted as d'Roc- Again. compared to the ideal
(Time of 1, this d'mc indicate a moderate degree of separation between the signal and
noise distributions. i.e.. the worker has moderate sensitivity. The value of d' obtained by
the ROC curve and SDT is different as the ROC determines the sensitivity of ironworkers

by portraying the effect of the response bias on sensitivity.

Similar calculations to those shown for worker "wl" were performed for all the
ironworkers. The results are summarized in Table 4.3 below. Details for each worker can
be found in Appendix C. Also. Table 4.4 summarizes the number of“Hits". "FA". "(‘R"

and “False Alarm" for each question based on response of all 42 ironworker.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Worker Age E:;::ise::9 d'SDT d. ROC Bcurrent Buptimal
w1 32 . 5 l 0.8 0.64 0.64 0.75
w2 30 j 4 5 l 1.11 0.71 1 0.88 0.75“
w3 22 I 3 l 1.73 0.81 0.85 0 75
m 24 a 9 1.11 . 0.71 0.88 o 75
WS 35 2 5 1 1.11 l 0 69 0.64 0 75.--
WG .24 4 . 1.85 j 079 0.41 4 0.75
w7 21 . 1 I 0.64 l 0.63 . 1.08 i 0 75_.__-_ .
WB 26 3 -0.1 T 049 l 1 1 T o 75
WS 24 45 1.39 T 0.75 0.7 l 075
mo 26 5 1.85 0.79 0.41 r 0 75
w11 30 1 3 1.39 0.75 0.7 l 0.75

 

 

 

 

Table 4.3 Results of Analysis of Survey by Standard SDT and ROC Curve

1J1
h—d

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Worker Age [2:52; eii: e d'snr (1' ROC Bcurrent Boptimal
w12 20 1.5 0.51 0.58 0.7 0.75
w13 31 2 0.23 0.54 0.88 0.75
M4 29 3 0.51 0.58 0.7 0.75
w15 30 ' 7 0.63 0.57 0.51 0.75
w16 36 - 15 1.73 0.81 0.85 0.75
w17 22 i 1 L 0.63 0.57 0.51 0.75
W18 30 3 0.44 0.58 1.1 0.75
w19 25 L 3 0.88 0.67 1 0.75
w20 19 l 2 . 1.11 0.71 0.88 0.75
w21 29 L 2.5 L -0.2 0.46 1.09 0.75
w22 24 L 2.5 0.52 0.6 0.82 0.75
w23 21 L 2 . 0.63 0.57 0.51 0.75
w24 24 1 1 L -01 0.49 1.1 0.75
111125 25 L 6 L 3.06 0.67 0 0.75
w26 . 21 L 1 L -03 0.44 1.25 0.75
w27 L 27 6.5 L 1.85 0.79 0.41 0.75
w28 L 24 - 4.5 l 0.51 L 0.58 0.69 0.75
was 25 l 3 L 1.85 l 0.79 0.42 0.75
wao 31 - 15 L 0.23 0.54 0.89 0.75
w31 23 .' 35 L 1 73 0.81 0.85 0.75
1.1732 ' 32 - 7 L 1.45 L 0.76 1.08 0.75
w33 19 i 1 -0 8 0.36 1.38 0.75
w34 L 29 ; 6 ‘ 218 0.85 1 0.51 0.75
was 27 L 3 L 1.85 L 079 L 0.42 L 0.75
was 21 L 15 L 018 L 0.53 086 l 075
w37 26 L 4 L 0.8 L 0.64 0.64 L 0.75
was 24 L 2 - 0.97 L 0.63 0.42 L 0 75
W39 24 L 5 ' O 97 L 0.68 L 1.32 L 075
mo 26 4 5 5 L 0 L 05 1 1 075
M1 22 4 5 L 0.52 L 0.6 0 82 L 0.75
w42 33 7.5 1.21 068 0.38 L 0.75

 

 

 

 

Table 4.3 Results of Analysis of Survey by Standard SDT and ROC Curve (cont.)

52

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Response
Question .
Numbers Hit Miss FA CR
36 6 0 0
2 0 36 6
3 0 41 1
4 . 41 1 0 0
5 L 40 2 0 0
6 i 36 6 0 ' 0
7 30 12 0 0
8 31 11 0 0
9 35 7 0 0
10 0 l 0 14 28
11 0 0 3 39
12 41 1 0 0
13 L 16 26 0 0
14 0 0 11 31
15 L 29 2 13 0 0
, 16 L 0 0 22 20
17 E 30 12 0 0
1 18 L 0 0 5 37
19 0 1 0 40 L 2
20 i 0 L 0 18 L 24
1 21 38 L 4 0 L 0

 

 

 

 

Table 4. 4 Summary of Response to Each Question

Table 4.4. as discussed before. summarizes the response for each question. The
first column of table 4.4 indicates question numbers from the survey. followed by
columns indicating the number of responses of each type (Hit. Miss. FA. CR) for each

question.

This table will help to determine how the ironworkers as a group interpret each
scenario. meaning there might be a condition or scenario, which is unsafe according to
OSHA standards. but the group as a whole thinks it, is a safe condition. For example, on
question number 13 (see Appendix A). 26 ironworkers out of 42 think the scenario
represents a safe condition but OSHA disagrees. Likewise, in question number l9 (see
Appendix A). 40 ironworkers out of 42 considered that particular scenario is unsafe.
contradicting OSHA standards. So based on this table, feedback could be given to OSHA
on what the workers think about the particular scenarios, and such scenarios could be

addressed in training to change the approach of the workers.
4.3.1 Relation between d' 591- and d' Roc

This section describes the relation between d'gm and d'Roc- As discussed in chapter 3. the
ideal value of d'gm is 4.6 and that of d’Roc is 1. However in the worst-case scenario when ‘
there are no "hits" or "correct rejections" the worst values that d'gm and d'Roc assume are
—4.6 and 0. respectively. Assuming that the relationship between d'gm and d'Rtx‘ is linear.
a theoretical plot is developed as shown in Figure 4.1. In addition. to normalize values of
d'gm, a‘value of d'gm ol‘ —-l.6 (worst case) is considered to be 0% and the value ofd'gm—
of 4.6 (ideal case) is considered as 100%. It is further assumed that d'gnl <1 60%
represents low sensitivity. 60°10: d'gn'r S 80% represents moderate sensitivity. and that

d'sm-> 80% re resents hi yh sensitivitv.
P E .

54

      

 

 

d'ROC 0.80 ._ i l
0.60 — i E
EMODERATE: 5
LOW E 3 i
- 4.6 0.0 0.92 2.76 4.6
00/0 500/0 600/0 800/0 1 000/0
(1. SDT

Figure 4.1 Relation between d' sun and d' Roy

It is important to note that the representation of the relation between d'sln' and
d'Roc in Figure 4.] is only theoretical. To verify whether this assumption is reasonable.

the values of d'gm and d'Roc listed in Table 4.3 were plotted as a scatter plot as shown in

   
   
    
 

Figure 4.2.
Relation between d' SDT and d' ROC
1
y = 0.1156x + 0.5317
0-9 R3 = 0.7176
0.8
d' o
R O
o 0.6
C
0.5 o
o’
0.4
O
-4.6 -2.6 -0.6 1.4 3.4

d‘ by Standard SDT

Figure 4.2 Scatter Plot between d'sm and d'mx-

55

The plot in Figure 4.2 indicates that the theoretical linear representation is a
reasonable approximation of the actual relation. In fact. the high value of the coefficient

of correlation (r = 0.847) provides support that d'gm and (Time are indeed linearly related.

4.4 Data Analysis

Sensitivity and response bias analysis results of ironworkers are summarized in Table 4.5.
The table provides the value of d' obtained for each ironworker using standard SDT and
the ROC curve. Table 4.5 also shows the comparison between beta Current and beta
optimum, which helps to determine the decision making strategy or risk orientation of
each worker. If Bwrrcm is less then Bowmai then such strategy is considered conservative. If
the value ‘of Scum," is greater than BOplIma|~ then it is considered a risky strategy. The

strategy in the last column of Table 4.5.
4.4.1 Average Sensitivity and Risk Orientation of Ironworkers

In this section result from the survey will be discuss in details. First discussing about the
decision-making strategy of the ironworkers. The last column of Table 4.5 shows the
decision-making strategy for each ironworker. The average strategy of the group 01'

ironworker participated in this research was found to be risky.

The risk orientation of each worker was determined by comparing beta [5.1mm and
Bow. lf Bcurrent < Bum then it‘s a conservative strategy and if mem > Bum then it‘s a risky
strategy. There are few individuals with very high Scum,“ value. which shows they are
more risky (e.g. w7. w8. w26. etc). The last column of Table 4.5 shows the sensitivity of

each ironworker. which is determined based on the d'sm 0/o as discussed in section 4.4.1.

56

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

. . Decision
Kirk... £3,533.32. ”35:37:18" N335?“ 3...... making siiiiiiiviiy
Strategy

WI 5 58.7 64 0.64 Conservative Low
w2 4. 5 62.07 71 0. 88 Risky Moderate
w3 3 68.8 81 0.85 Risky Moderate
w4 9 62.07 71 0.88 Risky Moderate
w5 2.5 62.07 69 0.64 Conservative Moderate
w6 4 70.11 79 0.41 Conservative Moderate
w7 1 56.96 63 1.08 Risky Low
w8 3 48.91 49 1.1 Risky Low
w9 4.5 65.11 75 0.7 Conservative Moderate
WW 5 70.11 79 0.41 Conservative Moderate
WI] 3 A. 65.11 75 0.7 Conservative Moderate
w12 1.5 ' 55.54 1 58 0.7 Conservative Low
w13 2 52.5 54 0.88 Risky Low
w14 3 1 55.54 58 0.7 Conservative Low
w15 7 1 56.85 57 0.51 Conservative Low
w16 15 1 68.8 81 0.85 Risky Moderate
wl7 .. 1 g 56.85 1 57 0.51 Conservative Low
wl8 30 r 3 *- 54.78 1 58 1.1 Risky Low
w19 25 3 59.57 1 67 1 Risky Low
w20 19 2 62.07 1 71 0.88 Risky Moderate
w21 29 2.5 A 47.83 1 46 1.09 Risky Low
w22 24; 2.5 1 55.65 i. 60 0.82 Risky Low
w23 21 f 2 56.85 1 57 0.51 Conservative Low
w24 24 E 1 ' 48.91 1 40 1.1 Risky Low
w25 25 . 6 83.26 4' 67 0 Conservative High
w26 21 1 a 46.74 ‘ 44 i 1.25 Risky Low
w27 27 6.5 70.1 1 ; 79 1 0.41 Conservative Moderate
w28 24 , 4.5 55.54 1 58 1 0.69 Conservative Low
w29 25 3 T 70.11 i 79 l 0.42 Conservative Moderate
w30 3t . 1.5 52.5 54 1 0.89 Risky LLow
w31 23 1 3.5 68.8 81 0.85 Risky lModerate
W32 32 1 7 65.76 76 1.08 Risky [Moderate
w33 19 y; 1 41.3 36 1.38 Risky 71.08

 

 

 

 

Table 4.5 Summary of Results from SDT and ROC Curve

' Normalized d' by Standard SDT -‘ (d' by Standard SDT + 4.6.7 9.2)
‘ Normalized d' by ROC ‘ (d' by ROC Ideal d' by ROC) Where ideal d' by ROC T l

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

. . Decision
i:Vorker Age E:‘(ears of Normalized No'rmaltzed Balm." Making Sensitivity
penence d 5m [3] d ROC [4] Strategy
“36 21 1.5 51.96 53.00 0.86 Risky Low
W37 26 4 58.70 64.00 0.64 Conservative Moderate
W38 24 2 60.54 63.00 0.42 Conservative Moderate
W39 24 5 60.54 68.00 1.32 Risky Moderate
W40 26 5.5 50.00 50.00 1 Risky Low
W41 22 4.5 55.65 60.00 0.82 Risky Low
W42 33 7.5 63.15 68 0.38 Conservative Moderate
Average 26.02 3.88 60.01 64.60 0.77 -
Risky Moderate
Standard 4.34 2.66 9.22 . 11.58 0.30
Deviation NA NA
COV [5] 0.17 0.69 0.15 0.18 0.38 NA NA

 

 

Table 4.5 Summary of Results from SDT and ROC Curve (Continued)

As shown in Table 4.4. the average age of the 42 participants was 26 years with
an average 4 years 01‘ experience. The average normalized d'gm and d'Roc are close at
around 60%. indicating a low sensitivity. The average (Scum... indicates a risky strategy.
As indicated by the COV values. there was more variation in Bum...“ values compared to

sensitivity.

It is also worth noting that 5 workers (worker w8. w21. w24. w26 and “33) had
negative sensitivity. Because the sensitivity d' is determined by adding the two normal
deviate values Z. and 2.; (see Figure 4.3). a negative d' results only when the overlap
between the two curves. the signal and the noise. is more than 50% (shown in Figure 4.4).

This could happen in three cases:

3 Normalized d' by Standard SDT (d' by Standard SDT + 4.61’ 9.2)

‘ Normalized d' by ROC = (d' by ROC Ideal d' by ROC) Where ideal d' by ROC = 1

5 COV: coefficient of variation defined as the ratio between standard deviation and average; provided to
give a measure ofthe amount ofvariability relative to the value ofthe average. F

58

1. When P(Miss) is more than P(Hit) in which case Z1 will have high negative
value and d' will be negative.

2. When P(FA) is more than P(CR), that mean Z1 has high negative value.
3. When both the above conditions are true, causing both Z. and Z; to be negative.

Referring to Table 4.1 reveals that in all five cases where d’ is negative more false
alarms were made, which matches case 2 above. This explains why the sensitivity (d') of

those workers is negative. To improve their'sensitivity, worker specific training should

 

  
 

be developed.
N0 4— Yes
‘
Noise Signal
>5
5
‘5
E Correct
ET RCJCCIIOH
P(X‘S) I ,
fl

 

 

 

Miss / Xc \False alami

Figure 4.3 Distribution of Detection Theory (Wickens. 1992)

The following section investigates the distribution of d' and mem obtained fomt

the survey. This will help to dctenninc whether the distributions for d' and (i....,....,,. data

are normally distributed

59

 

 

Si 31
gn Noise

P [(N or S)]/X]

    

False Alarm

 

 

e? ‘ i
/
Correct Rejection

 

Xc

Figure 4.4 Distribution of the Survey Results

The following section investigates the distribution of d‘ and Dam... obtained form
the survey. This will help to detemtine whether the distributions for d' and Bum...“ data

are normally distributed

4.4.2 Distribution of d' and Beam...

To determine whether the data obtained from the survey of 42 ironworkers follow a
normal distribution. normal quantile plots were constructed using the statistical software
Minitab. A quantile graph is plotted with the standard normal (7.) score on the .\ axis '
and the data on the y—axis. lfthe nomtal quintile plot fomts a straight line. then the plot
indicates that the data are nomtally distributed (Moore and McCabc 2002). If there ts any

systematic deviation from a straight line. then that indicates a non nomial distribution.

Using the data in Table 4.3. three quintile plots were constructed to determine the
distribution Ofd'sln. d' Rtx'. and Bum“... All plots exhibited a straight-line conftmting that

the variables follow a nomial distribution. The plots are provided in appendix E.

60

4.5 Regression Analysis

In this section. regression analysis is used to investigate if age or years of experience are
linearly correlated with the sensitivity (d') and risk orientation (Bcuncm) of an ironworker
and also to investigate whether age and years of experience together are linearly
correlated with the sensitivity (d') and risk orientation (Scum...) of an ironworker (by using

multiple regression).

Regression analysis is performed primarily to determine the correlation between a
dependent (response) variable and an independent (predictor) variable. However. unless
an independent variable is controlled and manipulated. a regression model does not imply
that Y necessarily depends on X in a “causal" or “explanatory“ sense. Such a causal
conclusion is only justified in experiments where the independent variable is controlled
and the dependent variable is observed. In this research. the regression analysis is based
on quasi-experiments. i.e.. there was no manipulation of the independent variables.
However. while linear correlation results are reciprocal. meaning the math works

regardless of which variable is labeled independent. causality is not.

In regression analysis. the coefficient of correlation (represented by r) measures
the linear relationship between the response variable and the predictor. The coefficient r
is always a number between —1 and 1. A value of r near 0 indicates a very weak linear
relationship. The strength of the relationship increases as r moves away from 0 toward
either —1 or 1. The extreme values of r = -1 and r = 1 occur only when the points in a
scatter plot lie exactly along a straight line. The value of r can be determined using the

equation 4.2 (Moore and McCabe 2002).

61

of .ra'i- — Z Xi- 2 ii (4.2)
_ (:1 [=1 (=1
r— if. )2 1 n 2 i, n ‘a n 2T
\/InZXi' — (XXI) (cw- (2)7)

L i=1 i=1 L (:1 i=1

 

After r is detemtined. hypothesis testing is typically performed to assess the
significance of the relation between the two variables under investigation. Usually the
null hypothesis is a statement of "not related" or "not effected“, etc. The statement of null
hypothesis is denoted as H..: p = 0. and the statement that will be true if H0 is not true. the

alternative hypothesis and is denoted as H..: p i 0.

To determine whether the null hypothesis is rejected. a test statistic is determined
and computed to a calculate test statistic. The test is designed to check the strength of
evidence against the null hypothesis. The less probable the outcome. the stronger the
evidence that 11.. is false (Moore and McCabe 2002). Assuming the variables have a

bivariate normal distribution. H..: p = 0 versus H..: p 2r; 0 is tested as follows:

If Zeal. > Z “L3,: 11.. is rejected and if Z...“C < Z “.3, H0 cannot be rejected. Z can be

calculated using equation 4.3 (Moore and McCabc 2002).

 

For the purposes of this research. a of 0.05 is used. Hence. Z “.3. = 7. 10052 =

1.96.

Another common method of performing the hypothesis testing is to use the p-
value. The P-value is a statistical quantity that represents the smallest value of a for
which the null hypothesis is rejected. Therefore, the p-value represents the statistical
significance of the altemative hypothesis. In addition, the p-value can also be used to
perform the hypothesis testing itself. If P-value < 0t: H0 is rejected and if P-value > or: H0
cannot be rejected. The p-value can be calculated using the following equation (Moore

and McCabe 2002):

P-value = 2* P (Z > Zea...) V ' (4.4)
4.5.1 Regression Analysis for Age
4.5.1.1 Age vs. d'sm-

The regression plot of at: and d'sm is shown in Figure 4.5. The value of r for this
relation was 0.27. This indicates that the linear relationship between the age of the

ironworkers and their sensitivity is quite low.

- Testing of the null hypothesis that age and d' are not related versus that they are

was conducted as follows using standard SDT:
Using Equation 4.3: Z = 1.78 (note r = 0.27. pa = 0. and n = 42)
The rejection region for 11.. is when Zmlc > Z W; ; Z “.3, = Z (0 1,5,3, = 1.96

..Z“... < Z “L3,. hence 11., cannot be rejected.

63

Using equation 4.4: P-value = 2* P (Z > 1.78) = 0.075 > 01. Hence. for any value

of alpha less than 0.075. the null hypothesis cannot be rejected.

 

 

 

1r Age of lnronworker Vs d' by SDT :

I 1

i 1

1 4 y’ = 0.0544x - 0.4785 1

l 7 R2 = 0.0772 .

O 1

34 o i

F 2‘ ..0000 e i

F . ;

' f 1 3 ° . 0 1

P .. o . 08.
g 0 T . ' I . .f 1 1
15 .20 25 30 35 40

Age of lronworker

 

 

Figure 4. 5 Age of lronworker vs. d' Using Standard SDT

4.5.1.2 Age VS. d'R()('

Figure 4.6 shows a scatter plot for the sensitivity of the ironworkers using the ROC curve
and age. The value of r = 0.256 is close to that obtained for d'sm. This indicates a low
correlation between age and sensitivity. Testing the null hypothesis that age and d' using

ROC are not related versus that they are was conducted as follows:
Using Equation 4.3: Z = 1.64 (note: r = 0.256. p“ = 0. and n = 42): Z103, = 1.96
7...... < Z...3,. hence ll..cannot be rejected.
Using equation 4.4: P-value = 2* P (Z > 1.64) = 0.101> 01. Hence. for any value

of alpha less than 0.101. the null hypothesis cannot be rejected.

64

 

F Age of Ironworker vs d' by ROC 1

y = 0.0069x + 0.4614

 

 

 

 

 

 

R2: .
0.9 4 0066 1
0.81 00:... o o 1
1 . .
. U 0'71 ° , e
1 O 06 _ O O
a: : o .0
1 >. . o:
1 .9 05 -* o 0
: :5 ; O
1 0.4 —
. o
0.3 4
0.2 T T Y 1
15 20 25 30 35 40

Age of lronworker

L____.__ Z2-

Figure 4.6 Age of Ironworker vs. d' Using ROC

4.5.1.3 Regression Analysis for Age vs. Beta Current (Bcumm)

In this case. the response variable is risk orientation (BL-um...) and the predictor is the age
of the ironworker. The scatter plot for the two is shown in Figure 4.7. In this case. a
value of r = 0.156 was found. which again indicates a low correlation between age and
the risk orientation ofthe ironworkers. Testing the null hypothesis that age and 15mm... are

n0t related versus that they are was conducted as follows:
Using Equation 4.3: Z = 0.985 (note: r = 0.156. p0 = 0. and n = 42): Z...3, =—' 1.96
Z...”c < Z...13,. llence H0 cannot be rejected.

Using equation 4.4: P-value = 2* P (Z > 0.985) = 0.3260 > 0: hence. for any value

ofalpha less than 0.3260. the null hypothesis cannot be rejected. This strongly suggests

65

that there is no correlation between the age of the ironworker and Bcumm.

 

 

 

r l
I Age ofIronworker vs Beta Current

1 1.6 1

1 1 y = -0.0107x+ 1.046

I 1.4 .1 . R2 = 0.0245

1 1 °

1 1.2 4 ’

1 ..o i

1 5 1 g o o . o g

1 t 1 ~' 0 o

1 :3 1

1 U 5 o o 0

1 3 O 8 .1 O 3 o o

1 55 f o o o

g 0.6 : ’ ’

I O O O O

1 0 4 a o 0 o o .

1

; 0.2 . . . . . ,
1 15 20 25 30 35 40
1 Age of Ironworker

 

 

Figure 4.7 Age of Ironworker vs. Beta Current

4.5.2 Regression Analysis for Years of Experience

In this section. a similar analysis to that shown in the preceding section is performed to
determine the correlation between the ironworkers' years of experience and the SDT-

derived variables. Figures 4.8-4.10 show the scatter plots for years of experience against

d'SDT- d.R()C~ and BCUUL‘IH-

Table 4.6 shows the results of the regression analysis for experience and the SDT
parameters. The results strongly suggest that there is a correlation between the years of
experience of the ironworker and sensitivity. However. no correlation was found between

the years of experience ofthe ironworker and risk orientation.

66

Years of Experience vs d' by SDT

4 y = 0.1526x + 0.3445
. o R = 0.2295

 

Years of

Figure 4.8 Years of Experience of Ironworker vs. d'sm

 

Years of Experience vs d' by ROC
y = 0.0196x + 0.5638
0.9 s R2 = 0.2038
0.8 2
0.7 —
0.6 e
0.5 -
(l4 .
0.3 . - i
0 5 10 15 20

d' by R()(‘

 

 

Years of Experience

 

FEt—Fre 4. 9 Years of Experience of Ironworker vs. d'mx-

 

,___.____._____ __ ___.

Years of Experiance vs Beta Current

1 6 y = -0.0246x + 0.8642
14~ o , R2=00494
12~: 7
i. ’3 0 ° 1
078‘ o . o
05. ’0...
o 00
04.- 1: {>40 01 1".
02-

o »——-—--—- -

0 5 1O 15 20

Beta Current

A
v

Years of Experience

%

Figure 4. 10 Years of Experience of Ironworker vs. BM..."

67

 

 

 

 

 

 

 

 

 

Re ression 1 H othesis
VaEiables R Z“... P-Value Tesigg Result
Experience and 0.48 3.258 0.0012 H0 is rejected.
d'sor

Experience and 0.45 3.038 0.0024 H0 is rejected.
d'Roc

Experience and 0.22 1.41 1 0.1586 H0 is cannot be
Beumm. rejected.

 

 

Table 4.6 Regression Analyses for Years of Experience

The above section discusses the regression analysis to determine the correlation
between age or years of experience Vs the sensitivity (d') and risk orientation (BMW...) of
an ironworker. Now to determine the combined effect of age and years of experience Vs
the sensitivity (d') and risk orientation (chcm). multiple regression analysis is performed

and is discussed in the following section.
4.5.3 Multiple Regression Analysis

Multiple regression analysis in general is performed to determine the correlation of more
than one variable. In this research. multiple regression is performed to leam more about
the relation between two independent variables. the age and years of experience of
' ironworkers, and two dependent variables. a) the sensitivity (d') and b) risk orientation

(Bcurrent) ofironworkers.

This analysis is perfonned in Microsoft Excel and the tables obtained from the
analysis are provided in appendix F. Based on the value of Adjusted R (Appendix F). Z
values and P-values were determined and are summarized in Table 4.7. The results
strongly suggest that there is a correlation between the age and years of experience

together and the sensitivity of an ironworker. From the analysis in the above section. it

68

was determined that the years of experience of an ironworker alone had a correlation with
the sensitivity of the ironworker. However, the results of multiple regression analysis
suggest that the age and years of experience of an ironworker together affect his/her

sensitivity.

Hence, it could be concluded that the number of years of experience has a strong
correlation with sensitivity. and further. it could also be concluded that the number of
years of experience along with age affectsthe sensitivity of a worker with a 95%
confidence interval. This means that an older worker with more years of experience will

have a higher sensitivity to unsafe conditions.

However, no correlation was found between the age and years of experience of an
ironworker (together) and risk orientation. This means that the risk orientation of an
ironworker does not change with age or with years of experience. Thus. one may
conclude that other variables. such as training or supervision, should be tested to see if

they change ironworkers‘ risk orientation.

 

. . . : vaothesis
V ' acac 1 ' I . '
Regressmn artables R , 7 1 I P ‘ alue 1 Testing Result

1
l
1
1

_ . _.__L..__’ ._._-

Age & Years of j H.. is rejected.

 

 

 

. , . 0. 37 1 2.92 0.0030 -
Eyenence vs. d 51,. 4 i 1 J
A e&Yearsof ? 1 . i . 1 .‘ '. f

g . , , 0.404 1 2.68 I 0.0070 1 H‘ '5 “J““d ;
Experience vs. d iii 1. 1 1 l 1
Ae&Yearsof = ,_ 1 . 1 .':' 3 1

g . 1 0.055 1 0.342 0.733 : H‘. '5 “mm" b“ 1
Experience vs. 0......“ L ‘ j rejected. 1

 

 

Table 4.7 Multiple Regression Analysis Results

69

4.6 Summary

In this chapter, survey data was analyzed using Signal Detection Theory to determine the
sensitivity and risk orientation of ironworkers to unsafe condition. The ROC curve also
helped to determine the sensitivity of ironworkers by considering the joint effects of
response bias and sensitivity (i.e. the joint effects of risk orientation and sensitivity). The
results were further analyzed using regression analysis to determine whether the
sensitivity or risk orientation of ironworkers is related to their age or years of experience.
Also, to determine whether age and years of experience together had any correlation,

multiple regression was perfomied.

The objectives stated in chapter 1 were achieved using the methods and
techniques discussed in chapter 3 and chapter 4. Chapter 5 discusses the results and
research contributions and concludes with the research limitations and areas of future

research.

Chapter 5
SUMMARY AND CONCLUSION

71

5. Conclusions and Summary

5.1 Conclusions

Construction accidents and specially falls are of major concern for the construction
industry and the researchers. Despite the contributions of many construction accident
causation models in understanding the accident process. none adequately explain the
underlying causes of construction accidents due to its dynamic nature. To over come this
limitation a new approach to understand construction accidents has been proposed by
Howell et al (2002) based on the work of Rasmussen (1997). The model suggested
recognizes that organizational and individual pressures push people to work in hazardous
situations. These pressures defeat efforts to enforce safe work rules specifically in a
changing work environment such as in construction. Therefore, this approach emphasizes
the need to train workers to be conscious of hazardous work environments and engage the
work with better planning. So the focus of this research was to develop a model to
determine the sensitivity and risk orientation of construction workers. which in turn will

help to design worker specific training.

To achieve this goal a survey was developed based on OSHA standard of fall
protection and from fall cases reported by NIOSH. With the help of this survey the
sensitivity and risk orientation of ironworkers was be determined using SDT. This
research focused on assessment of occupational safety and health competencies of
construction ironworkers. The result of this research suggest that around 95 9o (i.e. 40

out of 42) ironworker who participated in this research have “low" to "moderate"

sensitivity toward unsafe condition. This reveals that most workers lack proper safety and

health knowledge and require additional training.

The tools presented in this study provide may be used to determine the sensitivity
and risk orientation of workers to unsafe conditions. Based on the result of this analysis.
worker-specific could be developed to increase the sensitivity and decrease risky

behavior towards unsafe conditions.

0 In general. based on the analysis performed in this research. the following

conclusions are drawn:

0 This model could be used as a pre-test and post- test after training for

assessing the effect of training.

0 Feed back to OSHA on regulation. if for example a particular scenario is

always missed or considered safe.

0 The whole group of ironworker who participated in this research has risky
strategy. which means they should be trained again to change their risk

orientation.

0 50% ofthe ironworkers who participated in the survey have a risky decision-
making strategy which means they will have more misses then false alarms.
These workers should be trained to change the decision-making strategy from

risky to conservative.

o The average sensitivity ofthe group is moderate when compared to ideal d'.

73

o The sensitivity and response bias data for the ironworkers follow a normal

distribution.

0 ' Regression analysis indicated that sensitivity (d') and risk orientation (Bcumm)

of ironworkers is not linearly correlated to age.

0 Regression analysis indicates a moderate dependency between years of
experience and sensitivity of the ironworkers (r = 0.48. p—value <0.0006).
However no linear correlation was found between years of experience and risk

orientation ofthe ironworkers.

5.2 Limitations of this Research

The questionnaire developed for the survey is not based on any company's safety policy
or training guides and also just focuses on fall protection. It is based on OSHA fall
protection standards and the case studies from NIOSH report only. In this research it is
assumed that the worker would react the same as he or she responded in the survey when
faced with any of the scenarios portrayed by the survey. Based on this assumption the
sensitivity and risk orientation of the ironworker has been determined using Signal

Detection Theory.

5.3 Areas of future Research

Future research should consider larger samples as well as other construction trade to
determine the sensitivity and risk orientation of the workers. Based on the results. SDT
and ROC curve analysis could be performed in a similar way to that performed in this
research. Real-time investigation of how workers respond to safe and unsafe condition is

also important. Another important area of research is that regarding the design of

74

training after the SDT parameters are determined. Effect of injury history and training

frequency should be considered.

5.4 Contributions of the Research
0 A technique to determine the sensitivity and risk orientation of the

construction workers to safe and unsafe condition was developed.

0 A survey allowing the assessment of worker sensitivity and risk orientation to

conditions leading to fall accidents was developed.

0 Signal detection theory was implanted in construction to determine the

sensitivity and risk orientation of workers to unsafe conditions

0 The practical application of Rasmussen theory of accident causation in

construction was enabled.

0 A framework for developing guidelines to design worker specific training

75

APPENDIX A

Consent letter and Survey questionnaire

76

CONSENT LETTER

Subject Consent Form

IRONWORKER OCCUPATIONAL SAFETY KNOWLEDGE

Principal Investigators: Tariq S. Abdelhamid, PhD
Research Assistant: Bhavin Patel

The Construction Management program at Michigan State University is conducting a
research project to assess the occupational safety knowledge of ironworkers. The research
will help in improving the effectiveness of safety training programs. You are being asked
to participate in this project in your capacity as a construction ironworker.

As a participant in this research. you will be asked to complete a Zl-question survey on
occupational safety rules related to fall protection.

Your assistance is voluntary and you may choose to stop assisting at any time during this
project. Your privacy will be protected to the maximum extent allowable by law. Your
company or you will not be identified by name. The estimated time for the survey is 30-
45 minutes. As a participant. you may request a copy of this consent letter for your
records.

If you have any questions about this project. you can do so by contacting Dr. Tariq '
Abdelhamid. Construction Management Program. Michigan State University at (517)
432—6188. Also if you have any question about your rights as a human subject to a
research project. please contact Dr. Ashir Kumar. at University Committee on Research
Involving Human Subjects (UCRIHS). Michigan State University at 517—355-2180
(email: ucrihsatmsuedu; 202 Olds Hall. East Lansing. MI 48824).

I voluntarily agree to participate in this study.

 

Subject Name Occupation Signature Date

 

Witness Name Occupation Signature Date

77

SURVEY QUESTIONNAIRE

MSU Member:

 

MICHIGAN STATE UNIVERSITY

Ironworker Occupational Safety Assessment

Date:

 

Name of the Company:

 

Location of Job Site:

 

Name/Title of Person Interviewed:

 

Construction Experience (In Years):

 

Age:

 

INTERVIEW QUESTIONS

Please choose one of the classifications that fits the follOwing conditions:

1 ) Working on a I30 foot high coupler scaffold designed by the company foreman.

[I An Unsafe Condition

C A Safe Condition
[3 I Don't Know

2) Working on a scaffold 8 feet above the lower level without a guardrail system.
C] An L’nsafe Condition
E A Safe Condition

C I Don't Know

3) Working on the Slh floor of a building where permanent bolting/welding on the l“I
floor hasjust begun.

78

[:I An Unsafe Condition
l: A Safe Condition

|:] I Don’t Know

4) Working on the erection of the 15m floor of a steel structure where the permanent floor
decking has been installed up to the 6‘h floor only.

C] An Unsafe Condition
C A Safe Condition

E I Don't Know

5) Working on a scaffold which is 12 feet above the lower level (where permanent
decking has been installed) without any fall protection.

E] An Unsafe Condition
l: A Safe Condition
E l Don't Know
6) Working on a 3.500-sqft. decking which has an unsecured connection.
[:1 An Unsafe Condition
D A Safe Condition
[3 I Don't Know

7) Working on the second floor of a building. which is provided with perimeter safety
cables. The top cable is fixed at 35 inches from floor level.

E] An Unsafe Condition
C] A Safe Condition

j:] I Don't Know

 

79

8) Working on a 37-foot high platform that is provided with a 3.5 foot high steel railing.
An 8-foot diameter vent stack runs vertically through the center of the platform, with
12-inch annular space between the vent stack and the platform.

I: An Unsafe Condition

D A Safe Condition

C] I Don't Know

9) A 50-inch square opening was created while working on renovation of a flat roof.

 

:1 An Unsafe Condition

[:1 A Safe Condition

C] I Don‘t Know

10‘) You are working on the 1 lm floor ofa building where permanent bolting/welding on
the 9‘h floor has just begun.

[:1 An Unsafe Condition
[:J A Safe Condition
[:3 I Don't Know

1 I) When climbing a portable ladder used for access to an upper landing surface. the side
rail extends 3.5 feet above the upper landing surface.

C] An Unsafe Condition

 

[:j A Safe Condition

D I Don't Know

12) Working on the erection of the 1 floor of a steel structure where the permanent
floor decking has been installed up to the 61h floor only.

 

III}

[:] An Unsafe Condition
D A Safe Condition

D I Don't Know

80

13) Climbing a portable ladder. which is set lfoot out for every 5 feet. as shown in the
figure.

C] An Unsafe Condition

[:3 A Safe Condition

[3 I don’t know

 

 

1 Ft
14) Working on a scaffold 5 feet above the lower level without a guardrail system.

[:1 An Unsafe Condition

[:j A Safe Condition

[3 I Don't Know

15) Bolting a steel member with a co-worker on the second floor of a building while the
3A steel sway bracing rod has not been installed.

D An Unsafe Condition

C] A Safe Condition

C 1 Don't Know

16) Working on a 37 foot high platform that is provided with a 3.5 foot high steel railing.
An 8-foot diameter vent stack runs vertically through the center of the platform. with
6-inch annular space between the vent stack and the
platform.

[:J An Unsafe Condition

 

j: A Safe Condition

[:3 1 Don‘t Know

17) Working on a scaffold with a walkway that is 10 foot long. 12 inches wide and is
extended over its support by 18 inches. ‘

 

C] An Unsafe Condition

[3 A Safe Condition
C] IDon‘tKnow

81

18) You are working on a 3.000-sq.ft. decking which has an unsecured connection.
E] An Unsafe Condition
C] A Safe Condition

[3 I Don‘t Know

19) You are working on the 1 11h floor of a building where permanent bolting/welding on
the 9'h floor has just begun.

I: An Unsafe Condition
D A Safe Condition
[:J 1 Don't Know

20) Climbing a portable ladder used for access to an upper landing surface when the side
rail extends 2 feet above the upper landing surface

S An Unsafe Condition
E A Safe Condition

[:j I Don‘t Know

21) Working on the second floor ofa building. which is provided. with perimeter safety
cables. The top cable is fixed at 42-inch from floor level.

 

[:1 An Unsafe Condition
[:j A Safe Condition

|:| 1 don‘t know

   

Ara-.5

APPENDIX B

Results of Response of Ironworker

h. u .i. rift... toll—1y .SIV

3:525; 3.35m 2:... 5.33— .5.232— _.= 3:5...

 

mmd

mud mod Ndd

mmd

$6 26

$6

mmd

We mud

mod

2.:

mm c

 

o—

N

o. 0

ll v: Ii illlYi . l
l

S. : ..,._ :1

3. c

3. ll) 1!

:—

 

E2;

m::sa;ecg

 

X

x

_-o

 

><

 

=-e

 

a.~.

 

m_~.

 

 

s_~.

 

e_~.

 

m_~e

 

v_~v

 

M—O

 

~_~v

 

__~V

 

=_~.

 

am.

 

 

 

 

 

 

><><><><><><

><><><><><><

 

X

x

 

x

x

 

 

m~v

 

>~V

 

eNV

 

 

m~e

 

 

MN.

 

-v

 

X

x

_~.

CA

 

.20

(n.— mfl—Z =2

 

 

 

10

«E 3:2

 

 

 

3:

£0

a; 8:2

 

 

 

:2

“U

 

tn.—

 

 

mag:

32 19

 

 

 

«H.— mm_ 2

z:

 

h u 98:: w;
annou<

n u man.— .9 mtr
a u ou<

N u an be a..;
.muou<

:Nu

m._ n mun.— ..e P;

ou<

m u mum ..e a;
an n ou<

 

 

m— hwy—563

 

V— how—LOB

 

n— emu—.83

 

~_.ax.e>’

 

_ _ 53.3.5

 

.oz.~.

 

 

86

mos—5.50 aye—am 2: .3 3.33— Econ _.m «Bah

 

3.6

$6

mmwod

mad

vvd

cmd

mmd

mud

£332..

 

I

(”’7

tack.

 

 

:0

 

£0

 

20

 

£9

 

to.

 

2a

 

m5

 

:0

 

20

 

X

:0

 

x

:01

 

Ev

 

oo

 

 

mo

 

 

 

><><><

so

 

 

><><

 

 

A

X

Ill

.5

 

 

mo

92

x ,....ml...

 

1*

l

 

 

 

 

x ~ 0 i

 

X

X

.0

 

m0

<m

32

:1

£0

3:2

z:

 

 

 

 

 

 

 

m.» u mum he 95
mm u eu<

mi N mum he 9; dz .0
NN n ew<

 

 

 

 

 

a. .383 z. 583

 

APPENDIX C

Result ofthe Analysis ofthe Data for Ironworkers Using SDT and ROC

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

WorkersL Probability Zvalues Coordinates 'with .. beta beta
. l ' SDT £32. current optimum
Hit MISS FA CR 21 22 21 22

W] 0.83 0.17 0.56 0.44 -015 0.954 0.39 0.25 0.8 0.64 0.64 0.75
w2 0.75 0.25 0.33 0.67 0.44 0.67 0.36 0.32 1.11 0.71 0.88 0.75
w3 0.83 0.17 0.22 0.78 0.77 0.954 0.3 0.25 1.73 0.81 0.85 0.75
w4 0.75 j0.25 037 0.67 0.44 0.67 0.36 0.32 1.11 0.71 0.88 0.75
w5 0.83 10.17 0.44 0.56 0.15 0.95 0.39 0.25 1.11 0.69 0.64 0.75
w6 0.92 [0.08 0.33 0.67 0.44 1.41 0.36 0.15 1.85 0.79 0.41 0.75
w7 0.58 0.43033 0.67 0.44 0.2 0.36 0.39 0.64 0.63 1.08 0.75
w8 0.75 0.25j0.77 0.23 -077 0.67 0.29 0.32 -01 0.49 1.1 0.75
w9 0.83 0.17j033 0.67 0.44 0.95 0.36 0.25 1.39 0.75 0.7 0.75
w10 0.92 00810.33 0.67 0.44 1.4 0.36 0.15 1.85 0.79 0.41 0.75
WI] 0.83 0.17j0.33;0.67 0.44 0.95 0.36 0.25 1.39 0.75 0.7 0.75
w12 0.83 0.17jt).67j0.33 -04 0.95 0.36 0.25 0.51 0.58 0.7 0.75
w13 0.75 02570671033 -04 0.67 0.23 0.54 0.23 0.54 0.88 0.75
w14 0.83 '0.l7j0.67 0.33 -04 0.95 0.36 0.25 0.51 0.58 0.7 0.75
w15 0.92i(1.()81().78 0.22 -0.8 1.41 0.29 0.15 0.63 0.57 0.51 0.75
w16 0.8310.17j0.2210.78 0.77 0.954 0.3 0.25 1.73 0.81 0.85 0.75
w17 0.92 l(1.t)8j£.78 0.22 -0.8 1.41 0.29 0.15 0.63 0.57 0.51 07.5
W18 0.5 :05 103310.67 0.44 0 0.36, 0.4 0.44 0.58 1.1 0.75
wl9 0.67%().33l1).3310.67 0.444044 0.361036 0.88 0.67 1 0.75
w20 0.75 102510331067 0.44 0.67 0.36 0.32 1.11 0.71 0.88 0.75
w21 0.58 t).42+().6710.33 -04 0.2 0.39 0.36 -02 0.46 1.09 0.75
w22 0.75 0.25-0.56 0.44 4115;067 0.39 0.32 0.52 0.6 0.8: 0.75
w23 0.92 41.084117? 0.2: -077 1.41 0.29 0.15 0.63 0.57 0.51 0.75
w24 0.75 "11.25;().78fi().22 -0.8j0.67 0.299032 -01 0.49 1.1 0.75
w25. 1 ; 0 41.674033 -041 3.5 0.36l 0 3.06 0.67 0.01 0.75
w26 0.67 [0.3331178j022 -().8 | 0.44 02910.36 -03 0.44 1.25 0.75
w27 0.92 10.1)8i().3.;().67 04411.41 0.36:0.15 1.85 0.79 0.41_ 0.75
w28 0.833 M1174 0.7 £3 -04; 1 (1.4 j 0.3 5 11.58 11.694 ____1_1_.j_s_
w29 0.92 101183133067 0.44j 1.41 0.36015 ‘ 0.79 0.4: _ 0.75
w30 0.753112541611133 0.41067 0.30032 0.23 0.54 0.891 0.75
w31 0.83 I0.17 0.22 0.78 _.77 10.954 0.3j0.25 1.73 0.81 _(_)._8_.f_-_._t).75
w32 0.75 51.230.220.78 0.77 . 0.67 0.3 30.32 1.45 0.76 '-‘?§_.-.._1’.75
w33 0.5 : 11.5;078412: -().8 0 0.29 0.4 -().8 0.36 1.38 0.75
w34 0.9211).()8L().2370.78 0.77 1.41 0.3 0.15 2.18 0.85 0.5_1_ 0.75
w35 0.92j11.08j0.33 0.67 0.44 1.41 0.36 0.15 1.85 0.79 042 0.75

 

Table C .1 Summary for Each Worker Who Participated in the Survey

94

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

lWorkersL Probability Zvalues Coordinates 'nglfhl ngh beta t gem
Hit Miss FA CR 21 22 2] :2 ROC curre" 0P "““m
W36 0.83 0.17 0.78 0.22 —0.8 0.95 0.29 0.25 0.18 0.53 0.86 0.75
W37 7 0.83 0.17 0.56 0.44 -0.15 0.95 0.39 0.25 0.8 0.64 0.64 0.75
W38 0.92 0.08 0.67 0.33 -0.44 1.41 0.36 0.15 0.97 0.63 0.42 0.75
W39 0.58 0.42 0.22 0.78 0.77 0.2 0.3 0.39 0.97 0.68 1.32 0.75
W40 0.67 0.33 0.67 0.33 -0.4 0.4 0.36 0.36 O 0.5 l 0.75
W41 0.75 0.25 l 0.56 0.44 -O.2 0.67 0.39 0.32 0.52 0.6 0.82 0.75
W42 0.92 , 0.08 0.56 0.44 -0.2 1.41 0.39 0.15 1.21 0.68 0.38 0.75

 

Table C .1 Summary for each worker who participated in survey (Continued)

 

 

APPENDIX D

Normalized SDT Table

96

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Normal ,
p Deviation Ordinates
z
0.01 2.326 0.027
0.02 2.054 0.048
0.03 1.881 0.068
0.04 1.751 0.086
0.05 1.645 0.103
0.06 1.555 0.119
0.07 1.476 0.134
0.08 1.405 0.149
0.09 1.341 0.162
0.1 1.282 0.176
0.11 1.227 0.188
0.12 1.175 0.2
0.13 1.126 0.212
0.14 1.08 0.223
0.15 L 1.036 0.233
0.16 a 0.994 0.243
0.17 1 0.954 0.253
0.18 1 0.915 - 0.263
0.19 1 0878 0.272
0.2 1 0.842 0.28 g
0.21 4 0.806 0.288 ‘
0.22 j 0.772 ‘7 0.296
0.23 i 0.739 0.304
0.24 0,706 0.311 1
0.25 0.674 j 0.318 1

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

p 0:37:23" Ordinates
Z
0.26 0.643 0.325
0.27 0.613 0.331
0.28 0.583 0.337
0.29 0.553 0.342
0.3 0.524 0.348
0.31 0.496 0.353
0.32 0.468 0.358
0.33 0.44 0.362
0.34 0.412 0.367
0.35 0.385 0.371
0.36 0.358 0.374
0.37 0.332 0.378
0.38 0.305 0.381
0.39 0.279 0.384
0.4 0.253 0.386
0.41 0.228 0.389
0.42 0.202 0.391
0.43 0.176 0.393
0.44 0.151 0.394
0.45 0.126 0.396
0.46 0.1 0.397
0.47 0.075 0.398
0.48 0.5 0.398
0.49 0.25 0.399
1 0.5 0 0.399

 

 

Table D.I Normal Deviation and Ordinates far Calculating d' and B

97

I- ‘_l’ ‘2'-

 

‘-...... ..
‘

 

APPENDIX E

Distribution Plots of d' and Bum...“

98

 

 

 

d' by SDT vs. 2 Value for d'

 

3.5-
3.0-1
2.5-7
2.0-
11.5-
1.0-
0.5-
0.0-
-O.5*

d by Standard SDT

 

 

 

I l I I

—'1 0 1 2 3
2 value for Standard SDT

 

 

Figure E. 1 Distribution Plot for d' by SDT

 

 

d' by ROC vs. 2 Value for d'

 

0.4a

 

 

 

l I l l l l l l
-2.5-2.0-1.5-1.0 -0.5 0.0 0.5 1.0 1.5

2 value for ROC

 

 

Figure E. 2 Distribution Plot for d' by ROC

99

‘wfi 1L3 .
\

 

 

 

Beta current vs. 2 Value for beta

;.V..‘ .’

 

 

1.5-1

Beta current
i?

_o
‘1"

 

 

 

2 value for beta

 

 

Figure E. 3 Distribution Plot for Beta Current

100

 

APPENDIX F

Result of Multiple Regression

101

 

 

2.9.2.0 Sam 5.» 3:38....m— he 88> Q ow< .8 £92.25. 385 22 be 338m :5:— Mn.— 03....

 

 

 

 

 

 

 

 

 

.35... 8.....- 5...... 528...- 868... So..- ~88... 52.8.... 85.6.5628.»
.58... ms... .86 3.8.... 8.5... 82.. N35... 3.83.- 6w<
£94m. Rm... 6%.. 6862... $.83. 22 Snow... 8.2.... 686....
.\....m.. 6...... $92. .26.. .33 6...... $3 .25.. 2:2-.. 2%. at... .525 25.2.66
.52? :4 .966
$8... 883“ 3. 5.28m
83m... 28.. 3.8... 53:... N 8.6883.
... 3..3....=...1m ... 9.. mm ..e
<>oz<
NV mCOSNEOmDO
.252. SE. Baccam
.583... 225 m 622......
cc _ 2... Beam .3
6&8... a 29:32
mecmzfim 5.38.31
5.56 5222.5

€2.50 Sum .m> oucotoaxm mo 88> 0% ow<

102

00¢ an brew—Sm .9» 3:23an .3 23> an 03. .8.— m_m»_a:< 338— m2 .2. 31.8: :53: N... «EFF

 

 

 

Gmmod vood mmod mmvood SNSd mooN ,cmmood am 56 8:28me .0 88>
3256 cod- oood awood- vommwd ow_.o ovvood wooed ow<
mvwmnd and wmnd mm_mm.o ccdmmfi 00—h wmmEd $2de .8825
o\e¢.mo .25: e\cc.ma 33¢»— o\cma .25: $3 .839— o=_a>-m .Sm . .995 32.5% 356580

mmd 3 :30...
:06 mmnmvd 3. 72.23%
mm_ 5.0 .mNSd cmod cvm. _ .c N 568.3%
m Sassflw .. m2 mm E I

<>OZ<
NV 2232850
awe—.0 .otm gaggm
if... 225m a 8.2.3
38.0 227m a
223 m 29:22

muzmtfim ..BmmoHMem

FDA—P36 >¢<E$5m

 

 

 

 

 

 

 

DOM E ESEmEm .m> vocatoaxm no 88> a. uw<

A.»
0
11

 

ham 3 53.32% .9 3:333H .8 33> Q «.3. .8 £3.52 .3me ..Z .c 3.55: :59: ..u. 2...;

 

 

 

 

 

 

 

 

 

w _ mmd flowed good mownd mmmmod wow: .3 oucotoaxm .0 an>
caged amnod :mod- chemd .mvmd v _ Nmod $25.0 uw<
SK.— omnm. T .omwd mm 510 Socnd mmmc _ .0 E082:—
c\ec.mo .25: e\ec.mo .83.: fiema Sam: {emo 33°1— o=_a>-m .Sm . Eta 3:33:25
. 22:35
3mmd~ Zn 3.0%
Nwmd vw _ fimm cm 3.660%
oooood vwnvwh 00.1“ New _ wd N :oEmeUm
.. agesgm , h. m: mm 3
<>OZ<

NV mcozmiomnO
mmond .otm Emvcmfi
:2... 2.37m a 8.3.3
8mm... 2.3% m
I 8%.: a 2.332

35.35 5.3353—

,FDA—PDO >¢<222m

 

 

 

Ham 3 bEEEm .m> 85598 mo 23» um ow<

104

REFERENCES

105

REFERENCES

Abdelhamid, T. (2003). "Construction Safety and Ergonomics”. Construction research
congress.

Abdelhamid, T. and Everett (2000). “Identifying root causes of construction accidents”.
Journal of Construction Engineering and Management. ASCE, 52-59.

B, Kirwan. (1994), A guide to practical human reliability assessment by Barry Kirwan.

Carlos, A. Vargas (1998). Investigating Construction Falls using Fault Tree analysis and
developing a prototype tool to reduce falls using expert system and computer
assisted instruction method. Ph.D. Diss., Civil Engrg., Ohio State University

Culver, C., and Connolly. C. (1994), “Prevent Fatal Falls in Construction”. Safety and
Health , NSC. Itasca. IL

Culver, C., Hardesty. C.. and Anderson, F. (1993). “Ironworker Fatalities in
construction”. Occupational hazard, 47 pp.

DeJoy, M. (1990). “ Towards comprehensive human factors model of workplace accident
causation." Professional Safety, 35(5), 11-16.

Duncan, C. W._. and Bennett. R., 111 (1991). “Fall protection and debris containment
during construction." Preparing for construction in 2] st century, L.—M. Chang.

ed., ASCE. New York. 97-102.

Heinrich, H.. Petersen. D.. Ross. N. (1980). Industrial Accident Prevention. McGraw-
Hill. New York.4 pp

Hinze. J. (1996). “The distraction theory of accident causation.” Proc., Int. Cont‘. An
Implementation of safety and Health on Constr. Sites. C [B working Commission
W99

Hinze, J.. Pedersen. C .. and Frdley. J. (1998) “ Identifying Root Cause Of Construction
Injuties". Journal of construction engineering and management. ASCl-I. 124 ( 1 ).
67-69 pp.

Holt, A. J. (2001). Principles ofconstruction safety. Blackwell Science, Maiden. Mass

Howell, G. A.. Ballard. 0.. Abdelhamid. T. S.. and Mitropoulos, P. (2002). " Workin 1
near the edge: a new approach to construction safety.“ Proceedings of the I l)’ '

International Group/or Lean Construction Conference, Gramado, Brazil. August
5-8, 2002

Huang. X. Hinze. J. (2003). “ Analysis of Construction Worker Fall Accidents". Journal
‘ of construction engineering and management. ASCE 129 (3), 262—271 pp

106

Johnson, H., Singh, A., and Young, F. (1998) “Fall protection analysis for worker on
residential roofs”. Journal of construction engineering and management. ASCE,
September/October ,41 8-428.

Johnson, M., Singh, A., and Young, R. H. F. (1998). “ Fall protection analysis for worker
on residential roofs”. Journal of Construction Engineering and Management.
ASCE, September/October, 124(5), 418-428.

McCam, M. (2000). “Cause of Roofer Death”. The center to Protect Workers’ Rights.
Silver Spring, MD.

McClay, R. E (1989). “ Toward a more universal model of loss incident causation- Part
11.” Professional Safety, February, 34-39

Mohamed, S.(2002). “ Safety Climate in Construction Site Environments”. Journal of
construction engineering and management, ASCE, 128 (5), 375-3 84 pp.

Moore, D. S., McCabc. G. P. (2002). Introduction to the Practice of Statistics. W.H.
Freeman and Company. New York.

Norman, D.A (1981). “ Categorization of action slips.” Psychological Review, 88, 1-14

National Institute for occupation Safety and Health (NIOSH) 1997. “Construction Safety
and Health”.

NIOSH & NORA (2000). “National Occupational research Agenda (NORA)" November
2000. NORA fact.

NIOSH (2000). “A summary of Surveillance Findings and Investigative Case Report”. 7
Technical Report November 2000. US. Department Of Health and Human
Services

NSC (2000), National Safety Council. “Accident facts publication"

O’Hare, D., Wiggins. M., Batt. R.. and Morrison, D. (1994). “ Cogntive failure analysis
for aircraft accident investigation.” Ergonomics, vol. 37, no 1 1. 1855-1869.

Petersen, D. (1982). Human Error- Reduction and Safety Management. STMP Press.
New York. NY.

Pollack, E., and Chowdhury. R. (2001). “Trends in Work-Related Death and 1njury Rates
among US Construction Workers. 1992-98”. The center to Protect Workers”
Rights. Silver Spring. MD

Rasmussen, J.. Pejtersen, A.M.. Goodstein. I-.P., Cognitive System Engineering. John
Wiley & Sons. lnc. New York, 1994.

107

Rasmussen Jens. (1997) “Risk management in a dynamic society: a modeling problem.”
Safety Science Vol. 27 No.2/3,pp. 183-213, 1997

Reason, J. T. (1990). Human Error. Cambridge University Press, New York

Recht, J. (1970). System Safety Analysis: Error Rated and Cost. National Safety Council,
Chicago. -

Rook, L., Altman, J ., and Swain, A. (1966). Human error Quantification. Sandia

Scott A.Shappell and Douglas A. Wiegmann (2001). “Human factors analysis and

classification system (HF ACS)”. HFES 45th Proceeding
. ‘1’ w

Spear, J. (2002) “Incident investigation: A problem solving prOcess” Professional Safety.
25-30

‘

Suraji, A. Duff. R, and S J. Peckitt. (2001) “Development of Causal. Model of
construction accident causation”. Journal of construction engineering and
management. J uly/ August, 337-344

Toole, M. (2002). “Construction Site Safety Roles”. Journal of construction engineering
and management. May-June, 203-210

Vargas, C., Hadipriono. F.. and Larew, R. (1996a). “ A fault tree model for falls from a
form scaffolding.” L. Alves Dias and R. Coble, eds., Implementation of safety and
health on construction sites, Balkema, Rotterdam, The Netherlands, 291 -301.

Vargas, C., Hadipriono, F ., and Larew, R. (1996b). “ Cause of falls from floor openings
and edges. ” L. Alves Dias and R. Coble, eds, Implementation of safety and
health on construction sites, Balkema, Rotterdam, The Netherlands, 303-313

Wagenaar. W. A., Hudson, PT, and Reason J. T. (1990). “ Cognitive failure and
accidents.” Applied Cognitive Psychology, Vol. 4, 273-294.

Whittington, C., Livingston. A., and Lucas, D.A. (1992). “ Research into management,
organizational and human factors in construction industry.” HSE Contract Res.
Rep. No. 45. Her Majesty's Stationery ofc., London.

Wickens, C. (1992), Engineering Psychology and Human Performance. HarperCollins
Publisher lnc..24pp.

Zeller, D. (1986) “ Heinrich revisited” professional safety. April 40-42

Zeller, D. B. (1986). “Heinrich revisited.” Professional Safety, October. 40-42
Corporation. Albuquerque, New Mexico.

108

   

uiiiiiiiiiiiiiiiiiiI