Sie sind auf Seite 1von 6

NARRATIVE REVIEW: APPLICATION OF

STATISTICAL TOOLS IN CONSTRUCTION


SAFETY RESEARCH
Marc Jill M. Dimapilis1
1
Institute of Civil Engineering, University of the Philippines – Diliman,
Quezon City, Philippines
Received Date: December 14, 2014

Abstract
This paper aims to determine the significance of statistical methods and probability concepts in
construction safety research and to assess how they are applied in data analysis. Ten journal articles
about construction safety were reviewed and their corresponding use of statistics concepts were
determined. Also, these articles were compared on how they applied statistics in dealing with their
research problems. Qualitatively speaking, results showed that statistics has played a big role in
data analysis of the reviewed articles. All of them employed statistics although at varying levels.
Lastly, this review tackled how each tool can be used in a specific situation especially in
construction safety research.
Keywords: Construction Safety, Data analysis, Statistics, Probability,

Introduction
In a study conducted by Choudhry et al. (2008), construction is the most accident-prone
industry [1]. As such, there have been numerous efforts to promote safety and mitigate
risks in construction sites and one of them is through research. Construction safety
research has played a big role in determining principal factors in improving the safety
climate in workplaces and how it has contributed in increasing productivity of the workers.
In analyzing the gathered data in researches conducted regarding construction safety,
statistical tools are necessary to develop accurate results and appropriate decisions. Also, it
gives credibility to the research methodology used.
In this paper, the researcher aims to determine the significance of statistical tools and
probability in the field of construction safety and to assess how researchers apply them in
data analysis.

Results and Discussion


Ten journal articles about construction safety were analyzed and their corresponding use of
statistics concepts were determined. Then, these articles were compared (similarities and
differences) on how they approached their research problems on the basis of the statistical
tools used.

Hypothesis Testing
The studies “Impact of Green Building Design and Construction on Worker Safety and
Health” by Rajendran et al. (2009), “Relationship between Construction Safety and Quality
Performance” by Wanberg et al. (2013), “Safety Climate in Construction Site
Environments” by Mohamed (2002), “Owner’s Role in Construction Safety: Guidance
Model” by Huang et al. (2006), “Tradesmen Involvement in Health and Safety” by
Maloney (2007), “Relationship Between Unsafe Working Conditions and Workers’
Behavior and Impact of Working Conditions on Injury Severity in U.S. Construction
Industry” by Chi et al. (2013) and “Construction-Safety Best Practices and Relationships to
Safety Performance” by Hinze et al. (2013) all formulated hypothesis in the analysis of
their data and evaluated these using tests appropriate for their situation.
Rajendran et al. (2009) examined if there was a significant difference in recordable
incident rates (RIRs) and lost time case rates (LTCRs) between “green” and “non-green”
projects. First, Kolmogorov-Smirnov (K-S) Test was employed to test for normality of the
distribution of the data. One-sample K-S Test can be modified as a goodness-of-fit test
wherein the sample data is standardized and compared to a specific distribution [2]. This
would make data analysis much simpler since the characteristics of a normal distribution
were already established.
Results showed that the data sets were non-normal. Therefore, z- and t-tests would be
inappropriate. To determine the validity of the null hypothesis, which states that the
medians of two samples and several samples are the same, Mann-Whitney U Test and
Kruskal-Wallis Test were used, respectively. These tests are much like the Analysis of
Variance (ANOVA) test since its parameters are one nominal and one ranked variable only
that they are used for non-normal distributions [3].
In Mohamed’s (2002) study, he evaluated the effects of management commitment,
communication, workers’ involvement, attitudes, competence, work pressure, work
hazards, behavior, and supportive and supervisory environments in attaining a safer
workplace environment. A hypothesis was formulated for each aforementioned variable.
These were tested using Partial Least Squares Structure Equation Modeling (PLS-SEM).
PLS-SEM is a statistical technique that investigates relationships among variables and
determines specific pathways that exist among them (Ketchen, 2013). Correlations
between variables and between safety climate and each variable were then determined.
This method is particularly useful in analyzing data wherein the number of independent
variables are much greater than the number of experimental observations while keeping its
predictive accuracy high (Cramer, 1993).
Another group of researchers, led by Wanberg (2013), assessed the relationship
between construction safety and quality performance. The former consists of two primary
factors namely recordable injury rates and first-aid injury rates. The latter consists of six
parameters which can be categorized into rework and defects. Hypotheses between each
safety and quality metrics were devised and before testing for significance, the data points
were examined for outliers using Cook’s Distance. This method is particularly helpful as it
prevents possible error in judgment of the null hypothesis. Also, it improves accuracy of
results as outliers have big influence on calculations.
Next, linear regressions using Pearson product moment correlation coefficient
(Pearson-r) between each safety and quality metrics were performed to determine the
relationship between the variables being considered. Pearson-r measures the linear
relationship between two variables, producing a value between -1 and 1. A value equal to -
1 implies perfect inverse relationship while a value of 1 indicates perfect direct
proportionality. Meanwhile, a value of 0 indicates that no correlation exists [6]. This
method is very useful in determining linear association between two sets of data, although
it does not speak of causality.
Since there are many independent and dependent variables being considered, the use of
multiple regression analysis or PLS-SEM might be more useful and convenient here.
However, for this case, the predictors, which are the quality performance factors, are
interdependent with each other prohibiting the use of the two aforementioned methods
above in analyzing relationships between these sets of variables.
After performing Pearson-r for each pair of data, its corresponding p-value and
coefficient of determination, r2, were determined. The former is used to evaluate the
hypotheses, by comparing it to a chosen significance level, α, while the latter signifies how
many percent of the dependent variable can be explained by the observed value.
Another group of researchers, led by Maloney (2007), assessed the involvement of
tradesmen in planning for site health and safety. Their involvement is based on three
primary criteria: opportunity, capability and motivation. Laborers and contractor officials
from different construction projects were surveyed using questionnaire forms and their
answers were based on a five-point Likert-type scale. Differences in the means for each
question item between the answers of the two types of respondents were then tested for
significance.
In the research done by Huang et al. (2013), they assessed the role played by the
owners, based on four indices, in construction safety. A guidance model (linear equation)
was the output with the null hypothesis that the coefficient of each index is zero. These
indices were quantified and an analysis of variance (ANOVA) was performed to compare
the safety performance, measured by the total recordable injury rate (TRIR), with the
different levels of indices. ANOVA deals with analysis of more than two numerical
distributions obtained from the experiments wherein more than two treatments were
employed. Distributions are differentiated by characteristic/s known as the factor, while the
number of distributions being considered is referred to as the levels. For this particular
analysis, a one-factor (safety performance based on TRIR) ANOVA with four levels (four
indices) was utilized [9].
Then, to determine the linear relationship of each index with the dependent variable
(safety performance), Spearman Rank Correlation Coefficient was used. This is similar to
Pearson-r only that the latter involves interval or ratio data while the former involves
ordinal or rank data. This was used in this particular case since it does not need the
assumption of normal data distribution and linear relationship between the variables being
considered [6].
After it has been established that the relationship between TRIR and each index is
linear, multiple linear regression was developed. This statistical tool tells us of “how
much” of an independent variable explains the dependent one. This is particularly useful
for obtaining a single relationship involving several predictors. To evaluate the null
hypothesis, ANOVA was again performed and the resulting F-value and the chosen
significance level showed that it must be rejected. This just confirms the validity of the
multilinear equation.
Moreover, in the study conducted by Chi et al. (2013), they focused on three primary
objectives: relationship analysis between two risk factors namely working conditions and
workers’ behavior; impact of these risk factors on accident type and injury severity; and
relationship analysis between the last two aforementioned variables.
For all the relationships, chi-square analysis and Fisher’s exact test were used to
determine if there is significant correlation between the considered variables. Generally,
the null hypothesis is given that there is no relationship between them. Chi-square test is
commonly used to compare sets of observed and expected categorical data. The resulting
value is then compared to a pre-determined chi-square distribution to approximate the p-
value. However, for small sample size (generally less than 10), Fisher’s exact test must be
used. Another difference between the two is that the latter gives the exact p-value, hence,
the name. For paired sets of data, McNemar’s test must be employed [7] [8].
Lastly, the team of Hinze (2008) wanted to determine the effects of construction-safety
strategies in safety performance on the basis of the project’s recordable injury rate (RIR).
First, the researchers evaluated the safety measures which were associated with lower RIRs
by using Mann-Whitney U Test since injury rates follow a non-normal distribution.

Probability and Probability Distributions


Probability concepts were also used in construction safety research since it act as tools in
predicting an outcome (e.g. occurrence of accidents). Probability distributions are also very
helpful in research because knowing that a set of data follows a specific distribution, its
properties can be used to analyze the sample data. Researches involving probability and
probability distributions were as follows: “Quantitative Method to Model the
Underreporting of Safety Incidents” by Jablonowski (2014) and “Poisson Model of
Construction Incident Occurrence” by Chua & Goh (2004).
The first study, conducted by Jablonowski (2014) specifically aimed to produce a
model regarding safety and reporting phenomena while taking into account the
underreporting of incidents. The researcher used concepts of conditional probability to
determine the probabilities of an incident report (P(R)) and nonreport (P(NR)) given by

P(R) = P(R|I)P(I) + P(R|NI)P(NI) (1)

P(NR) = P(NR|I)P(I) + P(NR|NI)P(NI) (2)

The above equations show examples of conditional probability which deals with
determining the likelihood of occurrence of an event given a certain restriction, denoted by
P(A|B) where A is the likelihood an event happening and B is the given restriction. For the
examples above, the restrictions above are occurrence and non-occurrence of the incident
[6]. This is an important probability concept because there are many instances where we
only want to determine likelihood of an event happening in a targeted sample.
Another research, conducted by Chua & Goh (2004), stipulated that the random
component of the Modified Loss Causation Model (MLCM), the statistical framework of
construction incident occurrence, could be represented by a Poisson distribution. Poisson
distribution is a discrete probability distribution which determines likelihood of occurrence
of a random variable on the basis of the mean number of occurrences of the same variable
in a specified time interval [6]. Determining if the random component of MLCM follows
this distribution is very important because it simplifies data analysis since the properties of
Poisson distribution has already been established. Some of it includes: equal mean and
variance; and “lack of memory” property which means that an occurrence of an event is
independent of when the time interval happened [9].To determine this, the Chi-square
Goodness-of-fit Test was conducted. It is a test to determine the appropriateness of a
specific distribution in modeling a sample data.
To validate the result of Chi-square Goodness-of-fit Test, Dispersion test was also
applied to the same data set. Its importance is to check the variability of the data set and to
determine the value of the coefficient of variation (standard deviation/mean). The latter
must have a value close to 1 since the mean and standard deviation of a Poisson
distribution are equal.

Factor Analysis
The study conducted by Choudhry et al. (2009) entitled “Measuring Safety Climate of a
Construction Company” wanted to determine the factors that would positively affect the
safety climate and performance in construction sites. Among the 31 potential safety factors
that were used in surveying construction companies, factor analysis was used to reduce it
to six significant characteristics. Main advantage of this method is that it focuses the
attention to the principal elements which significantly affect the dependent variable [10].
To test for validity of the results produced by factor analysis, Kaiser-Meyer-Olkin Measure
of Sampling Adequacy (KMO) and Barlett’s Test for Sphericity were conducted. The
former is used to evaluate acceptability of the correlation matrices for factor analysis. The
latter was used to test reliability of large samples (sample size for this particular study is
1,120) wherein values less than 0.05 suggests high probability of that significant
relationships between variables may exist [11].
A scree plot was also graphed and it was found out that a break occurred between the
second and third component, further reducing the factors to two. Principal Component
Analysis (PCA) was utilized in the 22 items of the two safety factors. PCA is a method to
recognize patterns in data and is used in this situation to check for redundancy and
therefore, reduce the said items to a minimum number possible without risking loss of
valuable information [12].

Conclusion
The narrative review on the application of statistical tools in construction safety research
showed that statistics has played a big role in collection and analysis of data and in turn,
making appropriate inferences. All articles that were reviewed used statistical tools in their
data analysis although at varying levels. Also, this review showed the variety of statistical
tools available and how each one can be applied to a specific situation especially in
research problems in construction safety research.

References
[1] M. R. Choudhry, D. P. Fang, and S. Mohamed, “Developing a model of construction
safety culture”, J. Manage. Eng., Vol. 23, No. 4, pp. 207–212, 2007b.
[2] “Kolmogorov-Smirnov Test”, Available:
http://www.princeton.edu/~achaney/tmve/wiki100k/docs/Kolmogorov-
Smirnov_test.html (Accessed December 2014).
[3] “Kruskal-Wallis Test”, Available:
http://www.biostathandbook.com/kruskalwallis.html (Accessed December 2014).
[4] D. Ketchen Jr., “A Primer on Partial Least Squares Structural Equation Modeling”,
Long Range Planning, Volume 46, Issues 1–2, February–April 2013, Pages 184-185,
2010.
[5] R. Cramer III, “Partial Least Squares (PLS): Its strengths and limitations”,
Perspectives in Drug Discovery and Design, Vol. 1, Issue 2, pp. 269-278, 1993.
[6] R. Walpole et al, Probability and Statistics for Engineers and Scientists 9th Edition,
Pearson Education, Inc., Boston, USA, 2012.
[7] “Chi-Square Test”, Available: http://www2.lv.psu.edu/jxm57/irp/chisquar.html
(Accessed December 2014).
[8] “Chi-square, Fisher’s exact and McNemar’s test”, Available:
http://yatani.jp/teaching/doku.php?id=hcistats:chisquare (Accessed December 2014).
[9] J. Devore, Probability and Statistics for Engineering and the Sciences 8th Edition,
Cengage Learning Boston, USA, 2010.
[10] “Factor Analysis”, Available: http://www.woelfelresearch.com/factorAnalysis.html
(Accessed December 2014).
[11] “Appendix 6: Comments on statistical validity of factor analysis samples”, Available:
www.aliquote.org/articles/tech/multvar/22_Appendix_6.pdf (Accessed December
2014).
[12] “A Tutorial on Principal Components Analysis”, Available:
www.cs.otago.ac.nz/cosc453/student.../principal_components.pdf (Accessed
December 2014).
[13] S. Rajendran et al., “Impact of Green Building Design and Construction on Worker
Safety and Health”, Journal of Construction Engineering and Management, Vol. 135,
No. 10, pp. 1058-1066, 2009.
[14] J. Wanberg et al., “Relationship between Construction Safety and Quality
Performance”, Journal of Construction Engineering and Management, Vol. , No., pp.
1-10, 2013.
[15] S. Mohamed, “Safety Climate in Construction Site Environments”, Journal of
Construction Engineering and Management, Vol. 128, No. 5, pp. 375-384, 2002.
[16] X. Huang et al., “Owner’s Role in Construction Safety: Guidance Model”, Journal of
Construction Engineering and Management, Vol. 132, No. 2, pp. 174-181, 2006.
[17] W. Maloney et al., “Tradesmen Involvement in Health and Safety”, Journal of
Construction Engineering and Management, Vol. 133, No. 4, pp. 207-305, 2007.
[18] S. Chi et al., “Relationship Between Unsafe Working Conditions and Workers’
Behavior and Impact of Working Conditions on Injury Severity in U.S. Construction
Industry”, Journal of Construction Engineering and Management, Vol. 139, No. 7, pp.
826-838, 2013.
[19] J. Hinze et al.,“Construction-Safety Best Practices and Relationships to Safety
Performance”, Journal of Construction Engineering and Management, Vol. 139, No.
10, 2013.
[20] C. Jablonowski, “Quantitative Method to Model the Underreporting of Safety
Incidents”, Journal of Construction Engineering and Management, 2014.
[21] D. K. H. Chua and Y. M. Goh, “Poisson Model of Construction Incident Occurrence”,
Journal of Construction Engineering and Management, Vol. 131, No. 6, pp. 715-722,
2005.
[22] R. Choudhry et al., “Measuring Safety Climate of a Construction Company”, Journal
of Construction Engineering and Management, Vol. 135, No. 9, pp. 890-899, 2009.

Das könnte Ihnen auch gefallen