1 s2.0 S0925753509002161 Main

Download as pdf or txt
Download as pdf or txt
You are on page 1of 9

Safety Science 48 (2010) 436–444

Contents lists available at ScienceDirect

Safety Science
journal homepage: www.elsevier.com/locate/ssci

Use of association rules to explore cause–effect relationships in occupational


accidents in the Taiwan construction industry
Ching-Wu Cheng a,b,*, Chen-Chung Lin a,c, Sou-Sen Leu a
a
National Taiwan University of Science and Technology, 43 Keelung Road, Section 4, Taipei 106, Taiwan
b
Ming Chi University of Technology, 84 Gungjuan Rd., Taishan, Taipei 243, Taiwan
c
Institute of Occupational Safety and Health, No.99, Lane 407, Hengke Rd., Sijhih City, Taipei 221, Taiwan

a r t i c l e i n f o a b s t r a c t

Article history: Most models of the incidence of occupational accidents in the construction industry are composed of
Received 2 March 2009 multiple factors. Although statistical techniques can be used to infer cause-and-effect relationships
Received in revised form 20 November 2009 among these factors, the large number of factors involved and the complexity of the relationships among
Accepted 11 December 2009
them make it difficult for managers to identify potential hazards in construction projects and thus
develop effective safety procedures. This study addresses this problem by using the association-rule
method of data mining in performing an analysis of 1347 accidents in the Taiwan construction industry
Keywords:
during the period 2000–2007. The association-rule method enables potential cause-and-effect relation-
Construction industry
Occupational accident
ships to be identified among the many factors that play a role in occupational accidents in the construc-
Data mining tion industry. The study finds that such accidents tend to occur when certain combinations of hazards are
Association rule present – especially working in high places without protective measures, loss of balance when in motion,
Safety management failure to use protective equipment, insufficient experience, and injurious contact with unstable struc-
tures. These hazards are especially evident in small enterprises with less than 10 persons. The results
can help management to formulate effective safety policies with regard to management shortcomings
and staff training.
Ó 2009 Elsevier Ltd. All rights reserved.

1. Introduction be much higher than the fatality rate for all industries (0.0657
deaths per thousand persons in 2005) (see Fig. 1).
Various combinations of factors and circumstances influence Many factors have been proposed to explain this phenomenon.
the incidence of occupational accidents in any industry, and sev- These have included (among others): (i) the inherently hazardous
eral models of the causative factors of occupational accidents have nature of construction projects; (ii) personnel factors; (iii) environ-
been devised – most of which assume that accidents occur under mental and equipment factors; (iv) project factors; (v) manage-
certain specific circumstances that can be controlled (Sawacha ment factors; and (vi) time factors (time of accident, month, etc.).
et al., 1999; Abdelhamid and Everett, 2000; Tam et al., 2004; In view of the fact that virtually all models of work accidents in-
Mitropoulos et al., 2005; Martin et al., 2006; Choudhry and Fang, volve combinations of multiple factors, it is likely that no single
2008). For example, the so-called domino theory’ proposed by factor can be identified to provide a complete explanation for the
Heinrich (1959) emphasised that most occupational accidents are high incidence of serious work accidents in the Taiwanese con-
due to only three main factors – unsafe acts, unsafe conditions, struction industry. The best that can be done in these circum-
and unsafe equipment – all of which are controllable. stances is to use modern information technology to identify key
Nevertheless, despite this rather optimistic view of the prevent- associations between relevant factors among the mass of compli-
ability of occupational accidents, the Yearbook of Labor Statistics cated data on occupational accidents.
published by the Taiwanese Council of Labor Affairs has reported Against this background, the present study undertakes a careful
that the rate of fatalities in the construction industry (for example, classification and encoding of a large number (1347) of occupa-
0.29 deaths per thousand employed persons in 2005) continues to tional accident and fatality reports in Taiwan. These raw data are
then subjected to association-rule analysis to determine the levels
of association among the various influencing factors. The results of
these analyses can provide management with evidence of probable
* Corresponding author. Address: National Taiwan University of Science and cause-and-effect relationships, thus facilitating the establishment
Technology, 43 Keelung Road, Section 4, Taipei 106, Taiwan. Tel.: +886 2
of a safer working environment in the construction industry of Tai-
29089899x4658; fax: +886 2 29041914.
E-mail address: [email protected] (C.-W. Cheng). wan and other countries.

0925-7535/$ - see front matter Ó 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ssci.2009.12.005
C.-W. Cheng et al. / Safety Science 48 (2010) 436–444 437

0.40
0.3686 All industries
Construction
0.35
0.3227

0.2925 0.29
0.30
Fatalities per 1000 0.2681

0.25 0.271

0.2415 0.2363
0.20

0.15
0.1123

0.10 0.0862
0.0736
0.0657
0.0969
0.05 0.0773
0.0595 0.0581

0.00
1999 2000 2001 2002 2003 2004 2005 2006
Year

Fig. 1. Fatalities per 1000 persons (construction industry and all industries), 1999–2006 (Source: Yearbook of Labor Statistics, Taiwanese Council of Labor Affairs).

The remainder of this paper is organised as follows. The next and condition of materials (27%); and (v) deficiencies in risk man-
section provides a review of the relevant literature. The methodol- agement (84%).
ogy of the empirical study is then described. This is followed by a Tam et al. (2004) characterised the factors influencing safety
presentation of the results. The results are then discussed. The pa- performance in the following stark terms: ‘‘poor safety awareness
per concludes with a summary of the main findings and the impli- of top management”; ‘‘lack of training”; ‘‘poor safety awareness of
cations for practising managers. project managers”; ‘‘reluctance to input resources to safety”; and
(in general) ‘‘reckless operations”.
Several studies have made recommendations regarding factors
2. Literature review that could enhance safety management in the construction indus-
try (HSE, 1988; Williamson et al., 1996; Haslam et al., 2005; Tariq
According to the Labor Safety and Health Act of Taiwan, a major and John, 2000; Kartam et al., 2000; Angela and Inês, 2005). These
occupational accident’ is defined (Article 28, Paragraph 2) as ‘‘an studies have made various recommendations – including improved
accident that causes injuries to three or more persons or causes operating methods and the provision of adequate training and
the death of at least one person at the time it occurs”. As noted safety knowledge for workers and management personnel.
above, the construction industry in Taiwan (and elsewhere) has a With regard to the statistical analysis of construction-industry
very high incidence of such major occupational accidents’. accidents, a recent study by Aksorn and Hadikusumo (2008) used
The literature on construction-industry accidents reveals that factor analysis to explore 16 critical success factors that might be
the factors influencing the incidence of accidents are generally used by management in seeking to achieve project safety. How-
similar in various countries (Cattledge et al., 1996; Kartam et al., ever, given that most models of work accidents involve various
2000; Siu et al., 2003; Tam et al., 2004; Angela and Inês, 2005; combinations of many potential causative factors, it is unlikely that
Macedo and Silva, 2005; Aksorn and Hadikusumo, 2008). Apart a single factor will be identified to provide a full explanation for the
from the inherently dangerous nature of construction work, these high incidence of serious work accidents in the construction indus-
factors include: (i) personnel factors (gender, age, work experience, try. In these circumstances, the present study contends that the ap-
etc.); (ii) environmental and equipment factors (occupational acci- proach that is most likely to fruitful in seeking guidance for
dent type, source of injury, etc.); (iii) project factors (project type, improved safety management is to use modern information tech-
contract amount, etc.); (iv) management factors (company size, nology to identify certain important associations between relevant
hazardous situations, etc.); and (v) time factors (hour, day, and factors among the mass of complicated data on occupational acci-
month of accident). Table 1 summarises some of the relevant liter- dents. In particular, the technique of association-rule analysis’ can
ature on this subject. be used to determine the levels of association among various influ-
Various reviews of safety management in the construction ential factors. The results of these analyses can provide evidence of
industry have revealed that inadequate safety measures and poor probable cause-and-effect relationships.
safety awareness (on the part of both workers and management) The process of association-rule analysis involves the retrieval of
are the major reasons for the high incidence of occupational acci- large amounts of data from an appropriate data warehouse for data
dents in this industry (Sawacha et al., 1999; Tam et al., 2004; An- mining. The aim is to find otherwise obscure patterns (or rules’)
gela and Inês, 2005; Aksorn and Hadikusumo, 2008). For example, hidden in the mass of data. The association-rule method in data
Haslam et al. (2005) identified five key contributing factors to the mining has been successfully used to uncover such patterns in a
high incidence of construction accidents: (i) problems arising from variety of fields – including market basket analysis, product recom-
the workers or work teams (70% of accidents); (ii) workplace issues mendation, web page pre-fetch, gene regulation pathway identifi-
(49%); (iii) shortcomings in equipment, especially personal protec- cation, and medical record analysis (Agrawal and Srikant, 1994;
tion equipment (PPE) (56%); (iv) problems related to the suitability Berry and Linoff, 1997; Liu et al., 2004; Chang, 2007).
438 C.-W. Cheng et al. / Safety Science 48 (2010) 436–444

Table 1
Summary of literature exploring factors influencing incidence of accidents in the construction industry.

Literature source Factors


Worker factors Environmental and equipment factor Project factors Management factors Time factors
Abdelhamid and Everett (2000) H
Chi et al. (2004, 2005) H H H H H
Cattledge et al. (1996) H H H
Arditi et al. (2007) H H
Fabiano et al. (2004) H H H
Huang and Hinze (2003) H H H H
Hinze et al. (1998, 2005) H H H H
Kartam et al. (2000) H H
Kletz (1993) H
Laflamme and Menckel (1995) H H H
McVittie et al. (1997) H
Macedo and Silva (2005) H H H
Siu et al. (2003) H H
Tam et al. (2004) H H H H
Zeng et al. (2007) H H

Note: H Indicates that the study discusses the type of factor in question.

3. Methodology et al., 2008). An a priori algorithm was used to derive such associ-
ation rules from the large data set, which was based on some user-
In an attempt to identify underlying rules’ of cause and effect in defined threshold’ values for rule detection (confmin and supmin) –
the data on construction accidents, the present study examined because confidence levels depend mainly on the significance level
1347 reports of occupational accidents and fatalities in the indus- of the analytical results (Agrawal et al., 1993; Agrawal and Srikant,
try from 2000 to 2007, as recorded in the occupational accident 1994; Witten and Frank, 2000; Han and Kamber, 2001; Olafsson
database of the Council of Labor Affairs (Executive Yuan, Taiwan). et al., 2008).
In the pre-processing stage, occupational accidents were classified All collected data were analysed with Statistic Package STATIS-
into categories in accordance with the accident classification rules TICA DATA MINER V8.0.
of the American National Standards Institute (ANSI, 1995) Z.16.2-
1995 (Lortie and Rizzo, 1999). A contingency table was then used 4. Results
to establish the distribution of these categories. Statistical analysis
then involved: (i) correlation coefficient analysis; and (ii) selection 4.1. Basic statistical distribution of occupational accident factors
of major factors using online analytical processing (OLAP).
First, several tests (Chi-square test, goodness-of-fit test, inde- As shown in Table 2, the preliminary statistical results revealed
pendence test, and homogeneity test) were performed to establish that the major factors affecting construction accidents were: (i) the
whether the Pearson value (v2) of observed value and expected va- project type of building construction (51%); (ii) a company size of
lue was significant. The acceptance of the null hypothesis (Ho) was fewer than 10 employees (54%); (iii) worker experience of less than
then determined by observing whether the Pearson value (v2) of 1 month (41%); (iv) accident type (fall or tumble 54%; loss of bal-
the comparison between sample result’ and null hypothesis result’ ance when body in horizontal motion, 51%); (v) unsafe circum-
exceeded 0.05. In addition, based on the acceptance result (for stances, especially working in a high place without a working
example, that Ho is rejected due to the correlation between fac- platform or opening protection (24%); (vi) unsafe acts (mainly
tors), the phi-square coefficient of contingency (u) and Cramer V worker failure to use personal protection equipment) (45%); and
were calculated to analyse the distribution type of the correlation (vii) injurious contact with structures and construction facilities
coefficients between factors. (58%).
In parallel with this, the OLAP method was used to reveal the Fig. 2 shows the distribution of major factors influencing the
distribution of data and thus provide the correlation coefficients incidence of construction-industry occupational accidents. The
of important attributes. OLAP is a database acceleration technique main reason for these accidents was a failure to eliminate hazard-
used for deductive analysis (Goil, 1999). It supports rapid analysis ous conditions in the working environment – especially in compa-
by storing and refining a massive volume of data in a data ware- nies with fewer than 10 employees. With regard to time of
house (Moon et al., 2007). Sensitivity analysis was performed to accidents, the statistical distribution was largely similar to those
establish the relative significance of the many factors. The impor- reported in previous studies (Huang and Hinze, 2003; Macedo
tant attributes were then sorted by establishing boosting classifica- and Silva, 2005). The incidence of construction accidents in the
tion trees with deployment methods. present study was highest during the time periods of 11:00–
To establish the strength of association between variables, the 12:00 h (14%) and 16:00–17:00 h (13%). Accidents peaked between
study employed the knowledge-mining analysis steps recom- July and September (33%). Weekends (Saturdays and Sundays) ac-
mended by Fayyad: (i) data cleaning; (ii) data integration; (iii) data counted for 26% of all accidents. Most workers were aged 30–
selection; (iv) data transformation; (v) data mining; (vi) pattern 49 years (56%), but this is probably not significant because this
evaluation; and (vii) knowledge presentation (Fayyad, 1996; Fay- was the most common age group of employees in the construction
yad and Stolorz, 1997; Voro and Jovic, 2000; Lewis, 2000; Han industry.
and Kamber, 2001). The association-rule’ methods used in the
analysis complied with the concepts proposed by Agrawal et al. 4.2. Correlation analysis of occupational accident factors
(1993). Relevant procedures and threshold values conformed to
the research results obtained by various scholars (Baralis and Psa- Contingency table analysis was used to derive the correlation
ila, 1997; Zhang and Zhang, 2002; Kouris et al., 2005; Olafsson coefficients between variables and their distribution (Olafsson
C.-W. Cheng et al. / Safety Science 48 (2010) 436–444 439

Table 2
Distribution of major factors affecting construction-industry occupational accidents.

Factor Item %
Project ownership Public project 39
Non-public project 61
Company size Fewer than 10 employees 54
10–29 Employees 25
30–99 Employees 10
100 Employees and above 11
Project type Building construction project 51
Civil engineering project 40
Other project 9
Accident type Fall or tumble 54
Collapse 14
Electric shock 9
Falling object 5
Other 18
State of movement State of movement Body in horizontal motion 51
Body in up or down motion 23
Localized movement at a fixed point 18
Operating construction equipment or tools 8
Source of injury Structures and construction equipment 58
Loading and transport machinery 10
Substances or materials 9
Power machinery 5
Other 18
Work performed at the time of the accident Technical work 15
Odd jobs(temporary work) 19
Form work 11
Assembly of steel structural members 8
Masonry work 5
Electrical 7
Construction machinery operator 6
Other 14
Accident location Ground construction area 13
On scaffolding 13
Close to an edge or opening 10
Roof surface 10
On structural members 8
Restricted space such as tunnel 8
On ladder 4
On upper floor 6
By pipe trench or masonry wall 5
Road surface 6
On work platform 3
Other 14
Worker experience Less than 1 month 41
1–6 months 18
6 months to 1 year 10
More than 1 years 25
Gender Male 96
Female 4
Unsafe acts Failure to use personal protection equipment 45
Failure to use safeguard measures and warnings 41
Incorrect use of personal protection equipment 7
Inattention to footing hazards 5
Improper use of equipment 2
Unsafe condition Working in a high place without use of a work platform or opening protection 24
Use of hazardous methods or procedures 16
Failing to inspect working environment before work 12
No safety measures at nearby edges or openings 9
No safety measures at scaffolding openings 6
Electrical equipment is not properly insulated 6
No safety measures taken on roof 6
Entry restriction markings not placed within scope of hoisting operations 4
Use of inappropriate tools or equipment 3
Other 14

et al., 2008). Because categorical data were used in this study, the A large degree of freedom was found between factors, and v2
phi-square coefficient of contingency and Cramer V were calcu- was normally distributed. Chi-square distribution was therefore
lated to clarify the correlations between the variables (Lyman used for population variance estimation and variance tests. Tests
et al., 1986; Kurtz, 1999; Chi et al., 2004). of independence revealed that v2 > v2ðr1Þðc1Þ; a , which indicated
440 C.-W. Cheng et al. / Safety Science 48 (2010) 436–444

Fig. 2. Important occupational accident factors in construction industry.

that variables came from the same population and were correlated The association-rule analysis was then performed to gain a bet-
with each other. This implies that the null hypothesis should be re- ter understanding of association rules governing the cause–effect
jected. A correlation coefficient analysis between variables was relationships between the various factors.
then performed. The results of this analysis are shown in Table 3.
It can be seen from Table 3 that the p value between the two se- 4.3. Cause–effect relationships between accident factors
lected variables was always above the level of significance. This
implies that the two variables always have significant influence Of a total of 1347 accidents, 54% involved a fall or tumble. As
on the occurrence of accidents in the construction industry. It shown by the OLAP analysis in Table 4, the unsafe circumstances
was discovered from phi value calculations that the accident type for this type of accident were: (i) working in a high place without
and project type (u = 0.777), work content (u = 0. 763), location using a working platform or opening protection (42%); (ii) certain
of accident (u = 1.000), source of injury (u = 1.000), unsafe circum- accident locations, especially on scaffolding (21%), proximity to
stances (u = 1.000), and unsafe acts (u = 0.614) were all highly an edge or opening (18%), or on a roof (16%); (iii) injurious contact
correlated. with structures and construction facilities (scaffolding, stairs, and

Table 3
Correlation coefficient analysis of major factors influencing construction-industry occupational accidents in Taiwan, 2000–2007.

Factor 1 Factor 2 Pearson Chi-square df. p-vales Phi Cramer’s V Contingency coefficient
Project ownership Company size
129.305 3 0.0000 0.310 0.310 0.296
Project type
493.819 16 0.0000 0.605 0.605 0.518
Accident type Company size
120.510 48 0.0000 0.299 0.173 0.276
Project type
813.853 256 0.0000 0.777 0.194 0.614
Work experience
179.439 128 0.0019 0.365 0.129 0.343
Worker age
247.475 175 0.0003 0.429 0.129 0.394
Type of work
455.654 224 0.0000 0.582 0.155 0.503
Work content
784.506 288 0.0000 0.763 0.191 0.607
Accident location
2216.644 304 0.0000 1.000 0.321 0.788
State of motion
268.319 48 0.0000 0.446 0.257 0.407
Source of injury
2526.035 96 0.0000 1.000 0.559 0.808
Unsafe conditions
3681.970 304 0.0000 1.000 0.413 0.856
Unsafe acts
507.800 64 0.0000 0.614 0.307 0.523
C.-W. Cheng et al. / Safety Science 48 (2010) 436–444 441

Table 4
OLAP analysis of major factors influencing construction-industry occupational accidents.

Factor 1 Factor 2 Factor 1 \ \ Factor 3 %


Factor 2
Item Sample Item Sample Item Ratio
Private project 827 Fewer than 10 employees Building construction project 351/
732 541 541 65
Fewer than 10 employees 732 Less than 1 month of work Building construction project 201/
experience 549 341 341 59
10–29 Employees 331 Less than 1 month of work Building construction project 69/131
experience 549 131 53
Less than 1 month of work 549 Building construction project Fall or tumble 211/
experience 684 309 309 68
Less than 1 month of work 549 Temporary worker Fall or tumble 32/47
experience 74 47 68
Less than 1 month of work 549 Masonry work Fall or tumble 30/37
experience 71 37 81
Less than 1 month of work 549 Assembly of structural members Fall or tumble 37/53
experience 104 53 70
Building construction project 684 Fewer than 10 employees Fall or tumble 294/
732 413 413 71
Building construction project 684 10–29 Employees Fall or tumble 98/165
331 165 59
Building construction project 684 Structures and construction Fall or tumble 409/
equipment 782 488 488 84
Building construction project 684 Fall or tumble Failure to use personal protection 321/
725 451 equipment 451 71
Fall or tumble 725 Body in horizontal motion Structures and construction equipment 354/
682 388 388 91
Fall or tumble 725 Body in up or down motion Structures and construction equipment 172/
306 195 195 88
Fall or tumble 725 Form work Form assembly (removal) 49/77
143 77 64

openings) (87%); and (iv) certain work procedures at the time of an working on a building construction project (65%). Previous studies
accident, especially hoisting and assembly of steel structural com- have also shown that company size is correlated with the rate of
ponents (10%) and attachment and reinforcing of structural com- occupational accidents (McVittie et al., 1997; Fabiano et al.,
ponents (10%). Unsafe acts mainly consisted of failure to use 2004). These findings suggest that small companies with fewer
personal protection equipment (62%). The main injured part of than 10 employees typically fail to undertake adequate occupa-
the body was the head (67%). tional health-and-safety measures.
It was also found that most accidents occurred in companies The age of workers suffering falls or tumbles was concentrated
with fewer than 10 employees (54%), and when workers had less in the age group of 30–49 years (54%). This result is consistent with
than 1 month of working experience (41%). A large proportion of the findings of Huang and Hinze (2003) and Macedo and Silva
accidents involved workers suffering a fall or tumble (47%) while (2005), who found that persons aged 25–44 years had the highest

Fig. 3. Worker major occupational accidents per 1000 persons by age, 2000–2007.
442 C.-W. Cheng et al. / Safety Science 48 (2010) 436–444

rate of fatal accidents. However, as shown in Fig. 3, if the overall 4.4. Occurrence rules
age of workers involved in the industry is taken into consideration,
it becomes apparent that the highest normalised rate of death and The above analysis investigated the relationship between indi-
injury was among middle-aged and older workers (more than vidual factors and occupational accidents; however, accidents are
55 years of age) (Laflamme and Menckel, 1995; Siu et al., 2003; almost always due to several factors acting in conjunction. To gain
Chi et al., 2004). Given that a large proportion of persons in this a fuller understanding of the cause-and-effect relationships be-
age group engage in temporary work and have lower levels of edu- tween variables, this study applied the association rules module
cation, this finding is consistent with the conclusions of other stud- from StatSoft Statistica Data Miner V8.0 software. The focus of this
ies concerning work hazards faced by temporary workers (Kisner analysis was thus on the cause-and-effect relationships between
and Fosbroke, 1991; Haslam et al., 2005; Fabiano et al., 2004). combinations of circumstances that might influence accident types
With regard to the times at which most falls or tumbles occur, in two project categories: building construction projects (684 data
the most common periods were 10:00–12:00 h (26%), 14:00– items) and civil engineering projects (538 data items). Because the
15:00 h (11%), and 16:00–17:00 h (12%). This indicates that most results generated by association rules can be limited by the chosen
occupational accidents occurred towards the end of a working threshold values (support and confidence level), the association
day or before and after rest periods. This is likely to be related to rules thus obtained can have insufficient explanatory power. The
a more relaxed working mood and/or failure to ensure that protec- present analysis therefore took into consideration the threshold
tive clothing and equipment is properly used at these times. The value settings used in the literature (Coenen et al., 2004; Kouris
occurrence of falls and tumbles was highest between the months et al., 2005; Olafsson et al., 2008) and the analytical goal require-
of June and October (51%). These findings are similar to those of ments of these studies. The threshold values were set as follows:
Huang and Hinze (2003), who found that falls were most common confmin > 80%, correlation between the variables X and Y greater
from May to October, and that accidents tended to occur during the than 50%, and supmin greater than 14%.
periods of 9:00–11:00 h and 13:00–15:00 h. Similarly, Macedo and Analysis yielded 10 association rules for occupational accidents
Silva (2005) found that a relatively high percentage of accidents in building construction projects (see Table 5) and four association
occurred on Mondays and Saturdays, and during the afternoon rules for accidents in civil engineering projects (see Table 6). The
hours of 12:00–16:00 h, which might have been related to fatigue results in Table 5 confirmed that the main accident type was falls
or overtime work. and tumbles. The main reasons for accidents in these projects were

Table 5
Association rules for building construction projects.

Rule Association rule Support Confidence Correlation


ID (%) (%) (%)
X Y
1 Working in a high place without work platform or opening protection & fall or tumble Failure to use personal
protection equipment 21.345 80.220 55.371
2 Failure to use personal protection equipment Fall or tumble
46.930 84.031 77.337
3 Body in horizontal motion while working and failure to use personal protection equipment Fall or tumble
24.853 87.179 57.325
4 Working in a high place without work platform or opening protection and failure to use Fall or tumble
personal protection equipment 21.345 94.805 55.399
5 Working in a high place without work platform or opening protection Fall or tumble
26.608 95.288 62.011
6 Near opening in work area Fall or tumble
21.053 97.959 55.926
7 Contact with source of injury such as structures or construction equipment and body in Fall or tumble
horizontal motion while working and failure to use personal protection equipment 23.684 95.294 58.506
8 Contact with source of injury such as structures or construction equipment and company Fall or tumble
size of fewer than 10 employees and failure to use personal protection equipment 29.825 95.327 65.665
9 Contact with source of injury such as structures or construction equipment and work Fall or tumble
experience less than 1 month and failure to use personal protection equipment 21.199 97.315 55.935
10 Contact with source of injury such as structures or construction equipment and working in Fall or tumble
a high place without work platform or opening protection 24.123 98.214 59.943

Table 6
Association rules for civil engineering projects.

Rule Association rules Support Confidence Correlation


ID (%) (%) (%)
X Y
1 Working in a high place without work platform or opening protection Fall or
tumble 17.844 89.720 61.871
2 Company size fewer than 10 employees and contact with source of injury such as structures or Fall or
construction equipment tumble 19.145 81.102 60.932
3 Contact with source of injury such as structures or construction equipment and failure to use personal Fall or
protection equipment tumble 19.145 91.964 64.884
4 Working in a high place without work platform or opening protection and contact with source of injury Fall or
such as structures or construction equipment tumble 15.242 95.349 58.949
C.-W. Cheng et al. / Safety Science 48 (2010) 436–444 443

combinations of: (i) failure to install a work platform or protection ing environment; and (ii) the many unsafe acts committed by
when working in a high place; (ii) workers’ horizontal movement; workers themselves.
(iii) failure to use personal protection equipment when at work; As a consequence, the study concludes that the main steps that
(iv) less than 1 month of work experience; (v) injurious contact should be taken to minimise on-the-job-site accidents are: (i) the
with structures and construction facilities on the job site; and promotion of safety management in the working environment;
(vi) working at a company with fewer than 10 employees. These (ii) the implementation of pre-service education and training for
factors were all closely correlated with the occurrence of falls workers; and (iii) an enhanced management emphasis on (and
and tumbles; the level of support was in the range of 20–47, and support for) proper safety procedures.
confidence was in the range of 80–98. The distribution of results The main conclusion of this study is, therefore, that occupa-
in Table 6 was similar to that in Table 5; support was in the range tional injuries are preventable because most occupational acci-
of 15–19, and confidence was 81–95. dents are due to the negligence of workers and/or management.
It is also apparent that the level of safety in the construction indus-
try is largely dependent on whether workers and management re-
5. Discussion
gard it as important. If safety is taken seriously, every single
occupational accident should be investigated to establish its cause.
According to the results of the foregoing analysis, the main rea-
Moreover, a full database of occupational accidents should be
son that workers suffer a fall when working in a high place without
established and analysed to optimise safety procedures in every
a working platform or having other installed protective measures
construction workplace.
is the absence of personal protection equipment. These association
results indicate the potential seriousness of workers being heed-
less of safety rules and not taking their personal safety seriously. Acknowledgments
As shown in Tables 5 and 6, falls and tumbles are mainly attribut-
able to: (i) the horizontal movement of workers (for example, The authors thank the Department of Labor Inspection, Council
when carrying materials or cleaning up); (ii) injurious contact with of Labor Affairs, Taiwan, for providing the original accident reports
structures and construction facilities (for example, scaffolding, and information on the status of relevant public construction pro-
openings, roofs, roof trusses, and work platforms); (iii) unsafe acts jects during the research period. Research funding for this study
(when the structure is unstable or no protective measures are in was provided by the Institute of Occupational Safety and Health,
place); and (iv) absence of personal protection equipment. It is Council of Labor Affairs (IOSH96-S305, IOSH96-S306).
apparent that these factors have a high level of association with
falls. As a consequence, construction companies should ensure that
References
their workers are subject to tighter restraints and more effective
management – such as rewards and disciplinary measures with re- American National Standards Institute (ANSI), 1995. American National Standards
gard to safety rules. for Information Management for Occupational Safety and Health. National
Insufficient working experience (temporary workers and new Safety Council. ANSI 16.2-1995, New York.
Abdelhamid, T.S., Everett, J.G., 2000. Identifying root causes of construction
recruits) and employment in a company with fewer than 10 accidents. Journal of Construction Engineering and Management 126 (1), 52–60.
employees are also closely associated with occupational accidents. Agrawal, R., Imielinski, T., Swami, A., 1993. Mining association rules between sets of
The main cause-and-effect relationships in these circumstances are items in large databases. In: Proceedings of the ACM SIGMOD, pp. 207–216.
Agrawal, R., Srikant, R., 1994. Fast algorithms for mining association rules. In:
between falls and: (i) workers’ unfamiliarity with the working Proceedings of the 1994 International Conference on Very Large Data Bases, pp.
environment; (ii) insufficient awareness of potentially hazardous 487–499.
situations; (iii) insufficient supervision on the part of the employ- Aksorn, T., Hadikusumo, B.H.W., 2008. Critical success factors influencing safety
program performance in Thai construction projects. Safety Science 46, 709–727.
er; and (iv) and a lack of safety-management measures being in Angela, C.M., Inês, L.S., 2005. Analysis of occupational accidents in Portugal between
place. These findings suggest that there is a need for employment 1992 and 2001. Safety Science 43, 269–286.
contracts to specify that: (i) employers are required to implement Arditi, D., Lee, D.-E., Polat, G., 2007. Fatal accidents in nighttime vs. Daytime
highway construction work zones. Journal of Safety Research 38, 399–405.
education and training in worker safety; (ii) safety rules are in
Baralis, E., Psaila, G., 1997. Designing templates for mining association rules. Journal
place; (iii) self-inspection work is implemented; (iv) occupational of Intelligent Information Systems 9, 7–32.
health-and-safety personnel are appointed; and (v) inspection, Berry, M., Linoff, G., 1997. Data Mining Techniques: For Marketing, Sales, and
Customer Support. Wiley, New York.
maintenance, and protective tasks are performed with regard to
Cattledge, G.H., Hendricks, S., Stanevich, R., 1996. Fatal occupational falls in the US
potential hazards. construction industry. Analyses and Prevision 28 (5), 647–654.
These findings are in accordance with the analysis of Tam et al. Chang, C.-L., 2007. A study of applying data mining to early intervention for
(2004), who found that lack of training and poor safety awareness developmentally delayed children. Expert Systems with Applications 33 (2),
407–412.
on the part of project managers were the major causes of high acci- Chi, C.-F., Chang, T.-C., Hung, K.-H., 2004. Significant industry-source of injuries-
dent rates. These authors also argued that avoidance of falls and accident type for occupational fatalities in Taiwan. International Journal of
tumbles required appropriate safeguards when working in high Industrial Ergonomics 34, 77–91.
Chi, C.-F., Chang, T.-C., Ting, H.-I., 2005. Accident patterns and prevention measures
places, the use of working platforms, the installation of safety nets, for fatal occupational falls in the construction industry. Applied Ergonomics 36,
lifelines, and railing, and a requirement to wear protective gear. 391–400.
Choudhry, R.M., Fang, D., 2008. Why operatives engage in unsafe work behavior:
investigating factors on construction sites. Safety Science 46, 566–584.
6. Conclusions Coenen, F., Goulbourne, G., Leng, P., 2004. Tree structures for mining association.
Rules, Data Mining and Knowledge Discovery 8, 25–51.
Fabiano, B., Currò, F., Pastorino, R., 2004. A study of the relationship between
This study has subjected 1347 occupational accident and fatal- occupational injuries and firm size and type in the Italian industry. Safety
ity reports (as recorded by the Taiwan Council of Labor Affairs over Science 42 (2004), 587–600.
Fayyad, U.M., 1996. Data mining and knowledge discovery: making sense out of
the period 2000–2007) to statistical analysis and data mining asso-
data. IEEE Expert 11 (5), 22–23.
ciation rules. The results have shown that both workers and man- Fayyad, U.M., Stolorz, P., 1997. Data mining and KDD: promise and challenges.
agement had insufficient awareness of safety issues and potential Future Generation Computer Systems 13 (2–3), 99–115.
hazards. Most accidents were found to stem from a combination Goil, S. 1999. High performance on-line analytical processing and data mining on
parallel computers. Ph.D. Thesis, Dept. ECE, Northwestern University.
of: (i) the failure of management to implement adequate safety Han, J.W., Kamber, M., 2001. Data Mining: Concepts and Techniques. Morgan
measures to protect workers against potential hazards in the work- Kaufmann, San Francisco.
444 C.-W. Cheng et al. / Safety Science 48 (2010) 436–444

Haslam, R.A., Hide, S.A., Gibb, A.G.F., Gyi, D.E., Pavitt, T., Atkinson, S., Duff, A.R., 2005. Macedo, A.C., Silva, I.L., 2005. Analysis of occupational accidents in Portugal
Contributing factors in construction accidents. Applied Ergonomics 36 (2005), between 1992 and 2001. Safety Science 43, 269–286.
401–415. Martin, A.C., Randy, E.C., Barbara, S., Tom, S., Peregrin, S., 2006. Work-related deaths
Safety Health Executive (HSE),, 1988. Blackspot Construction: a Study of Five Years in Washington State, 1998–2002. Journal of Safety Research 37, 307–319.
Fatal Accidents in the Building and Civil Engineering Industries. HMSO, London. McVittie, D., Banikin, H., Brocklebank, W., 1997. The effects of firm size on injury
Heinrich, H.W., 1959. Industrial Accident Prevention. McGraw-Hill, New York. frequency in construction. Safety Science 27, 19–23.
Hinze, J., Huang, X., Terry, L., 2005. The nature of struck-by accidents. Journal of Mitropoulos, P., Abdelhamid, T.S., Howell, G.A., 2005. Systems model of construction
Construction Engineering and Management 131 (2), 262–268. accident causation. Journal of Construction Engineering and Management 131
Hinze, J., Pedersen, C., Fredley, J., 1998. Identifying root causes of construction (7), 816–825.
injuries. Journal of Construction Engineering and Management 124 (1), 67–71. Moon, S.W., Kim, J.S., Kwon, K.N., 2007. Effectiveness of OLAP-based cost data
Huang, X., Hinze, J., 2003. Analysis of construction worker fall accidents. Journal of management in construction cost estimate. Automation in Construction 16,
Construction Engineering and Management 129, 262–271. 336–344.
Kartam, N.A., Flood, I., Koushki, P., 2000. Construction safety in Kuwait: issues, Olafsson, S., Li, X., Wu, S., 2008. Operations research and data mining. European
procedures, problems, and recommendations. Safety Science 36, 163–184. Journal of Operational Research 187, 1429–1448.
Kisner, S., Fosbroke, D., 1991. Injury hazards in the construction industry. Journal of Sawacha, E., Naoum, S., Fong, D., 1999. Factors affecting safety performance on
Occupational Medicine 36, 137–144. construction sites. International Journal of Project Management 17 (5), 309–
Kletz, T.A., 1993. Accident data – the need for a new look at the sort of data that are 315.
collected and analyzed. Safety Science 16 (3–4), 407–415. Siu, O.L., Phillips, D.R., Leung, T.W., 2003. Age differences in safety attitudes and
Kouris, I.N., Makris, C.H., Tsakalidis, A.K., 2005. Using information retrieval safety performance in Hong Kong construction workers. Journal of Safety
techniques for supporting data mining. Data & Knowledge Engineering 52, Research 34, 199–205.
353–383. Tam, C.M., Zeng, S.X., Deng, Z.M., 2004. Identifying elements of poor construction
Kurtz, N., 1999. Statistical Analysis for the Social Sciences. Allyn & Bacon, MA safety management in China. Safety Science 42, 569–586.
(379pp). Tariq, S.A., John, G.E., 2000. Identifying root causes of construction accidents.
Laflamme, L., Menckel, E., 1995. Aging and occupational accidents: a review of the Journal of Construction Engineering and Management 126 (1), 52–60.
literature of the last three decades. Safety Science 21, 145–161. Voro, A., Jovic, F., 2000. Multiple attribute entropy classification of school-age
Liu, B., Grossman, R., Yanhong, Z., 2004. Mining web pages for data records. IEEE injuries. Accident Analysis and Prevention 32 (3), 445–454.
Intelligent Systems 19 (6), 49–55. Williamson, A.M., Feyer, A.-M., Cairns, D.R., 1996. Industrial differences in accident
Lortie, M., Rizzo, P., 1999. The classification of accident data. Safety Science 31, 31– causation. Safety Science 24, 1–12.
57. Witten, I., Frank, E., 2000. Data Mining: Practical Machine Learning Tools and
Lewis, R.J., 2000. An introduction to classification and regression tree (CART) Techniques. Morgan Kaufmann, New York.
analysis. <http://www.saem.org/download/lewis1.pdf>. Zhang, C., Zhang, S., 2002. Association Rule Mining: Models and Algorithms.
Lyman, O., Ruchard, L., Rexroat, C., Mendenhall, W., 1986. Statistics: A Tool for the Springer, New York.
Social Sciences. PWS-Kent Publishing Company, Boston (379pp). Zeng, S.X. et al., 2007. Towards occupational health and safety systems. Safety
Science. doi:10.1016/j.ssci.2007.08.005.

You might also like