精华区文章阅读

发信人: CECMSTA (土木工程系和建设管理系学生科学技术协会), 信区: civil_express
标  题: 23届挑战杯作品资料简介2
发信站: BBS 听涛站 (Mon Apr 25 18:26:20 2005)

【作品名称】  Quantitative research on safety climate in construction companie
s – sampling design and application
【第一作者】  陈扬（建管研02）
    【其他作者】  方东平（清华大学建设管理系教授）
【作品介绍】

QUANTITATIVE RESEARCH ON SAFETY CLIMATE IN CONSTRUCTION COMPANIES – SAMPLING
DESIGN AND APPLICATION
Yang CHEN, Dongping FANG
Department of Construction Management, Tsinghua University, Beijing, 100084
Abstract: The establishment of good safety climate within an organization is e
ssential if high goals of safety performance are to be achieved. To evaluate t
he safety climate in construction companies, a complete questionnaire survey w
as conducted with all sites and employees of Gammon Limited, a well-establishe
d construction company in Hong Kong in 2002. In total, 4,719 valid records wer
e returned from 54 sites of Gammon. This study was a subsequent part of the su
rvey in Hong Kong. Based on the collected data and factor analysis, 87 questio
ns in the original questionnaire are reduced to 40, and a new 9-factor safety
climate factor structure is achieved. The measurement of similarity of two fac
tor structures, and the relationship between percentage of sampling and simila
rity correlation of factor structure are discussed in this paper. A suitable s
ampling percentage (α=50%) for safety climate survey is found and will be app
lied in the subsequent sampling survey in mainland, China.
Keywords: Safety climate; Construction companies; Sampling
INTRODUCTION
Compared with other industries, there is a relatively high accident rate in co
nstruction industry (Fang et al, 2001). How to effectively improve safety perf
ormance and reduce the accident rate has been the focus of researchers. Since
1970s, safety culture and safety climate has attracted more and more attention
s. Safety performance used to be evaluated through accident rate in constructi
on industry. Experience Modification Rating, EMR, which is widely used in indu
strial insurance in U.S., is another indicator for evaluating safety performan
ce. However, the two indicators, accident rate and EMR, are not good enough be
cause they can only evaluate and measure safety performance after the accident
happens, referring to Xie’s (2003) discussions on this topic. It has not bee
n a long time since the concept of safety culture and safety climate came up.
Since that, there has been a movement away from safety measures purely based o
n “lagging indicators,” such as accident rates, toward so-called “leading i
ndicators,” such as measurements of safety climate (Flin et al, 2000). The shift of fo
cus has been driven by the awareness that organizational, managerial, and huma
n factors rather than purely technical failures are prime caused of accidents
(Sherif, 2002). That is, they have been trying to measure and improve the safe
ty performance through safety culture and safety climate. But up to now, the s
afety culture and safety climate hasn’t been given enough attentions in const
ruction industry in our country, and some traditional indicators, such as acci
dent rate and workers’ compensation statistics, are still the major indicator
s for evaluating safety performance of construction companies. Thus, an in-dep
th and systematic study is necessary for the safety culture and safety climate
in the construction companies.
Gammon is a well-established construction company and a market leader in the H
ong Kong construction industry. In 2002, a comprehensive questionnaire survey
was conducted with all sites and employees of Gammon and her subcontractors in
Hong Kong to collect the views, attitudes and believes of safety issues insid
e the company. Occupational Safety & Health Council of Hong Kong (OSHC) and (T
singhua University - Gammon) Construction Safety Research Center (CSRC) were i
nvolved in conducting this survey. Xie (2003) analyzed the colleted 4719 quest
ionnaires and had some significant findings through statistical analysis. Next
, OSHC plans to conduct a sampling survey on the safety climate in the constru
ction industry in Hong Kong to draw more general conclusions. Furthermore, the
project “Research on improving safety culture in construction companies in C
hina”, which is supported by Ministry of Construction, is going on. Sampling
survey is also necessary for this project.
However, according to the previous similar literature and research, the theory
of sampling design and application isn’t mature. The selection of sampling m
ethod and percentage of sampling is casual and is not under the direction of u
niform standard. Thus, to address the need, this present study focuses on the
percentage of sampling in the safety climate survey.
LITERATURE REVIEW
Safety culture and safety climate are concepts that derived from organizationa
l culture and are not determinately defined yet. Culture can be seen as a conc
ept that describes the shared corporate values within an organization which in
fluences the attitudes and behaviors of its members. Safety culture is a part
of the overall culture of the organization and is seen as affecting the attitu
des and beliefs of members in terms of health and safety performance (Cooper,
2000). Safety climate is a distinct yet related concept which can be seen as t
he current surface features of safety culture which are discerned from the att
itudes and perceptions of employees (Flin et al., 2000). However, in reality t
he terms are not so clear cut and many writers use the terms safety culture an
d safety climate interchangeably. Safety climate is usually regarded as being
more superficial than safety culture in that it involves the current position
of a company (Glendon and Stanton 2000). Many researchers have assessed safety
climate as an indicator of the overall safety culture of an organization (Gadd 2002).
Actually safety climate, which can be measured through quantitative methods, i
s a “snapshot” of safety culture.
The common method of quantitative research on safety climate is to draw the di
mensions of safety climate first. Dimensions of safety climate are the main fe
atures or levels of safety climate (Glendon and Stanton 2000). A number of att
empts have been made to construct the dimensions of safety climate. These atte
mpts can be divided into two categories. The one is to identify the frequently
used dimensions based on literature review (Sherif, 2002); the other is to ex
tract the dimensions through data of survey questionnaires and certain statist
ical method (Glendon and Litherland, 2001; Zohar, 1980). The limitation of the
first method is the differences between different areas and different industr
ies can not be carefully considered; and the disadvantage of the second method
is too dependent on the statistical method, that is, different statistical me
thod or different parameters may lead to different dimensions.
Factor analysis is the typical method while identifying the dimensions of safe
ty climate (Gadd, 2002). Many in-depth researches in the field of safety clima
te are based on the dimensions of safety climate, which are the factor structu
re from factor analysis (Glendon and Litherland, 2001; Basen, 1998). On the ot
her hand, the method of sampling should be carefully determined before the sur
vey. Generally, a comprehensive survey is seldom used due to the limitation of
time and money. Random sampling and stratified sampling, especially the latte
r, are used in more surveys. According to literature review, most of the resea
rchers in the field of safety climate selected random sampling based on strati
fication. However, how to select the percentage of sampling was seldom reporte
d in their research reports (Diaz and Cabrera 1997; Coyle 1995; Zohar 1980). T
here may be two possible reasons for the phenomenon. The one is the difficulty
of sampling itself; the other is scarcity of the direction of basic theory.
However, the percentage of sampling is very important in the survey and should
be carefully determined. A too high percentage may lead to waste of resources
, and a too low one may influence the precision of the survey. Meanwhile, thou
gh there is no in-depth research and exact definition of the degree yet, we ca
n still come to the conclusion that factor analysis is sensitive to the scale
of the data. It is an obvious fact because if we randomly select different sam
pler from the same collectivity and conduct factor analysis we will get very d
ifferent factor structure each time. Thus, an in-depth research should be done
on the percentage of sampling in the safety climate survey.
Before we study the influence of percentage of sampling on factor structure, w
e have to define the degree of similarity between two factor structures. In ot
her words, we should define the similarity correlation of factor structure at
first. A number of methods exist to compare the factors obtained in two separa
te groups when the same set o items are used. One method that is possible is c
onfirmatory factor analysis, in which factor loadings and factor intercorrelat
ions are hypothesized to exist in the same pattern in the second group as in t
he first group. Various indices yield information about how well the hypothesi
zed pattern fits the data obtained from the second group (Schafer 1996). Levin
e (1977) discussed four factor comparison measures. The first one is root mean
square measure, RMS, defined as the square root of the average squared differ
ence of the loadings of the variables. RMS varies from 0 to 2. An RMS of 0 mea
ns a perfect match between samples of both the pattern and the magnitude of fa
ctors in the two samples while 2 indicates all loadings are at unity but differ in sign
between the two samples. But intermediate values are hard to interpret. The s
econd one is coefficient of congruence, CC. Like RMS, CC measures both pattern
and magnitude similarities between samples. However, there is a tendency to g
et a high CC whenever two factors have many variables with the same sign. The
third one is salient variable similarity index, S. The S index will be 0 when
there are no salient loadings, indicating no factor congruence between the two
samples. An S of 1 indicates perfect congruence, and -1 indicates perfect neg
ative congruence. The last one is to directly compute the correlations between
the loadings of a set of variables across two studies. The commonly used corr
elative coefficient is Pearson correlations (Schafer 1996). Generally, Pearson
correlations are indices of vector similarity. That is to say, if one has fou
nd m factors which comprised F1 and n factors in F2, then one can compute the
m x n matrix of correlations, matching each of the factors of F1 with each of those in
F2. Levine (1977) developed the Pearson correlations into one similarity coeff
icient of matrices. All the columns in one matrix are taken as one column, thu
s two matrices can be simplified as two columns. Count the Pearson correlation
of the two columns, and we can get the correlation of the two matrices.
Questionnaire survey is widely used in data collection in safety climate surve
y. Undoubtedly, before the survey, the researcher should consider the capacity
and rationality of the questionnaire. After the survey, based on the collecte
d data, the questionnaire can also be revised and the number of the questions
in the questionnaire may be reduced. The situation of reducing the questions i
n the questionnaire after the survey in recent years (mostly since 1995) is co
ncluded in Table 1.
Table 1 Reducing the number of the questions in the questionnaire
Research team    No. of questions (items)    Reason for reduction
    original    reduced    final
Zohar (1980)    49    9    40    factor loading < 0.49
Coyle, et al. (1995)    30    1    29    factor loading < 0.40
    32    14    18    low factor loadings
Williamson, et al. (1997)    62    30; 5    27    30 items were eliminated bas
ed on their skewed distributions; five items' factor loadings < 0.40
Basen-Engquist, et al. (1998)    18    7    11    3 items were unrelated to th
e other items; 3 items' factor loadings < 0.35; 1 item's factor loading < 0.40

Varonen and Mattila (2000)    18    5    13    3 items' factor loadings < 0.40
; 2 items were in rejected factors (Because the fourth and fifth factor consis
ted of only one variable, they were rejected.)
Glendon and Litherland (2001)    40    8    32    loaded on two factors

From Table 1, we can conclude:
    “Factor loading < 0.4” is usually used as the borderline of redu
ction, meanwhile every factor consists of at least 3 items;
    Although the number of reduced questions (items) is different, the
final numbers of questions are all below 40.
REDUCTION OF THE QUESTIONS
The 110-item safety climate questionnaire was designed by OSHC and CSRC, and c
omposed of two parts. The first part included 87 questions on safety climate.
It was purposely designed to seek the view of managers, supervisors and workfo
rce on key aspects of safety climate within their organizations. It was develo
ped based on the safety management system in Hong Kong. In 1995, The Hong Kong
government developed a safety management system based on a complete examinati
on on industrial safety in Hong Kong. Fourteen safety management elements were
included in the system: safety policy, organizational structure, safety train
ing, in-house safety rules, program of inspection, program to identify hazard
or risk, investigation of accidents or incidents, emergency preparedness, eval
uation selection and control of subcontractors, safety committees, evaluation
of job hazards, promotion safety and health awareness, program for accident co
ntrol and program to protect workers (OSHB 2002). The respondents could rate t
heir responses to the 87 items on a 5-point Likert scale, from strongly disagree to str
ongly agree. The second part included 23 questions on personal information, in
cluding demographic information, such as age, sex, marital status and educatio
n level, and other personal characteristics, such as direct employer and safet
y knowledge situation. This part of the questionnaire was designed to explore
the relationship between safety climate and personal characteristics. The resp
ondents also needed to tick the most appropriate options for the 23 additional
questions.
4719 questionnaires, of which 3410 were from workers, 193 from clerk staffs, 6
81 from supervisors, 112 from managers and 323 with vacancy in position, were
collected from 54 sites of Gammon in Hong Kong. Factor analysis (Xie, 2003) ex
tracted 15 factors, which construct the dimensions of safety climate in the co
mpany. However, there are two problems of the factor structure:
    Too many factors, failing to exhibit the significant factors;
    All 87 items were kept in the structure, resulting in the scale of
the questionnaire was still too big, and the respondents may feel reluctant t
o finish such a long questionnaire.
Therefore, according to the literature review, the questions are reduced in tw
o ways. At first, all questions that have a factor loading of less than 0.4 on
all factors are discarded, and because the factors after factor 9 consist of
less than 3 variables (questions), they are rejected. Totally, 40 questions ar
e dropped in the first step. Then, considering that the questionnaire will be
used in the safety climate survey in the mainland, 7 questions that may not be
applicable for the mainland are deleted (such as some questions on safety com
mittee meetings). Finally, 40 questions are remained in the final questionnair
e. The corresponding safety climate 9-factor structure is shown in Table 2.
Table 2: safety climate 9-factor structure
factor    Description
1    Safety attitude of employee
2    Safety consultation & Safety training
3    Management commitment
4    Risk taking behaviour
5    Safety resources
6    Implement of safety procedure
7    Improper safety procedure
8    Worker’s involvement
9    Workmate’s influence
THE RELATIONSHIP BETWEEN PERCENTAGE OF SAMPLING AND FACTOR STRUCTURE
Levine’s Pearson correlation (referring to Literature review) is selected as
the similarity index of factor structure. Formula 1 is for calculating the Pea
rson correlation. In formula (1), n equals to 360 (=40x9). Xi and Yi are the v
alues in two compared factor matrices.
                              (1)
The hypothesis of the relationship model is that different percentage of sampl
ing, within the same collectivity and with the same method (including factor e
xtraction method and rotation method), will not bring big changes on factor st
ructure and on the factors’ order.
The procedure is:
1)    Select a sampling percentage, α, conduct random sampling on the 4719 da
ta, and get sample A;
2)    A principal components factor analysis, followed by a varimax rotation i
s conducted to the sample A, and get a factor loading matrix, B’ 40 x 9;
3)    Calculate the Pearson correlation, R, between B’ 40 x 9 and B, which is
get from 4719 complete data;
4)    Repeat step 1)~3) ten times, and calculate the mean of R, Rm;
Change α from 10% to 80%, then repeat step 1)~4), and a relationship between
α and Rm can be get, shown in Figure 1.
As we can see from figure 1, Rm rises with the increase of α, which is consis
tent with our experiences. That is, the bigger percentage of sampling, more si
milar the factor structure with that of the collectivity. However, though at t
he level of α=80%, Rm value is the biggest (0.994), it is obvious that the cu
rve becomes more horizontal between α=50% and α=70%. Thus, it is reasonable
to choose α=50% as the suitable percentage of sampling in safety climate surv
ey.

Fig. 1 The relationship between α and Rm
ONGOING WORKS
Now the sampling survey on the safety climate in construction companies in mai
nland is in preparation, the questionnaires (N=40) are ready and the survey wi
ll start in one or two months later. Meanwhile, a correlative project “Develo
pment of safety climate survey tool” is coming to the end, and the developed
software, an outcome of the project, will be used into the sampling survey (α
=50%) in mainland. Comparisons will be made between the results from the sampl
ing survey of mainland and those from the survey in 2002.
CONCLUSIONS
The safety climate questionnaire survey of Gammon has been successfully done,
with 4719 employees from 54 sites completing the questionnaire. Based on the c
ollected data and factor analysis, 87 questions in the original questionnaire
are reduced to 40, and a new 9-factor safety climate factor structure is achie
ved. The measurement of similarity of two factor structures, and the relations
hip between percentage of sampling and similarity correlation of factor struct
ure are discussed in this paper. A suitable sampling percentage (α=50%) for s
afety climate survey is found and will be applied in the subsequent sampling s
urvey in mainland, China.
This research is a case study and the results come from the data of one compan
y-Gammon, but the methodology of this research might be useful for the similar
survey and research, and the findings and suggestions might also be helpful f
or the construction management and safety practitioners in cultivating safety
culture in the construction industry.
REFERENCES
Basen-Engquist, K., Hudmon, K.S., Tripp, M., and Chamberlain R.. (1998) “Work
site Health and Safety Climate: Scale Development and Effects of a Health Prom
otion Intervention”, Preventive Medicine, 1998, 27(1), 111-119.
Carretta, T.R., Ree, M.J. (1995). "Near identity of cognitive structure in sex
and ethnic groups". Person Individ. Diff., 19(2), 149-155.
Cooper, M.D. (2000). "Towards a model of safety culture". Safety Science, 36,
111-136.
Coyle I.R., Sleeman S.D., and Neil A. (1995). “Safety climate”. Journal of S
afety Research, 26(4), 247-254.
Fang, D.P., Huang, X.Y., and Hinze J. (2001). “Safety management in construct
ion”. Beijing: Publishing House of Irrigation Works and Water-Electricity of
China.
Flin R., Mearns K., O’Connor P., et al. (2000). “Measuring safety climate: i
dentifying the common features”. Safety Science, 34, 177-192.
Gadd, S. (2002). “Safety culture-a review of the literature”. Sheffield: HSL
.
Glendon, A.I., and Stanton N.A. (2000). “Perspectives on safety culture”. Sa
fety Science, 34, 193-214.
Glendon, A.I., and Litherland D.K. (2001). “Safety climate factors, group dif
ferences and safety behavior in road construction”. Safety Science, 39, 157-1
88.
Levine, M.S. (1988). “Canonical analysis and factor comparison.” London: Sag
e Publications.
OSHB. (2002). “Code of practice on safety management.” Occupational Safety a
nd Health Branch, Labour Department, Hong Kong.
Schafer, J., and Leigh, B.C. (1996). "A comparison of factor structures of ado
lescent and adult alcohol effect expectancies". Addictive Behaviors, 21(3), 40
3-408.
Sherif, M. (2002). “Safety climate in construction site environments”. Journ
al of construction engineering and management, September/Octobor, 375-383.
Varonen, U., and Mattila, M. (2000). “The safety climate and its relationship
to safety practices, safety of the work environment and occupational accident
s in eight wood-processing companies”. Accident Analysis and Prevention, 32,
761–769.
Williamson A.M., Feyer A.M., Cairns D., et al. (1997). “The development of a
measure of safety climate:the role of safety perceptions and attitudes.” Safe
ty Science, 25, 15-27.
Xie, F. (2003). “Research on Construction Safety Performance Measurement”. T
hesis for Master Degree. Beijing: Tsinghua Univ.

说明：
1.    纸型与页边距：A4，上2.54、下2.54、左3.17、右3.17
2.    开头四行为加粗宋体五号字，如果是作品正式文档，就把最后的【作品简介】改成
【作品介绍】，把页眉中的“作品简介”改为“作品正式文档”只是作品简介限制字数，
制作类在A4一页纸以内，论文类在两页以内，作品正式文档不限字数。
3.    开头四行中括号与后面的内容中间要空两个格（半角），也就是一个汉字的距离，
如果人名是两个字，中间也要空一个汉字的距离，“本科”与“无07”中间也要空一个汉
字的距离。
4.    正文用宋体五号字，首行缩进两个字，单倍行距。如果是英文用times new roman,
也是单倍行距五号字。
5.    图片的注释用加粗宋体小五号字，“图1”与后面的内容也空出一个汉字的距离。

6.    关于文件的命名，作品正式文档就用作品名作为文件名，作品简介用作品名后加括
号并写上简介二字，例如：音乐舞蹈机器人（简介）

--
     ■清华大学土木工程系和建设管理系学生科学技术协会

※ 来源:·BBS 听涛站 http://tingtao.net·[FROM: 59.66.170.190]

civil_express 版 (精华区)