Predicting Construction Labor Productivity Based On Implementation Levels of Human Resource Management Practices

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 13

Predicting Construction Labor Productivity Based on

Implementation Levels of Human Resource


Management Practices
Argaw Tarekegn Gurmu, Ph.D. 1; and Citra S. Ongkowijoyo, Ph.D. 2
Downloaded from ascelibrary.org by University of Waterloo on 12/30/19. Copyright ASCE. For personal use only; all rights reserved.

Abstract: The prediction of the odds of achieving higher or lower productivity as compared to some baseline productivity is one of the
important steps to consider while analyzing labor productivity in construction projects. The objective of this research is to build a logistic
regression model that can be used to estimate the productivity of building projects based on the levels of planning or implementation of human
resource management practices. Quantitative data were collected from 39 contractors who worked on multistory building projects completed
between 2011 and 2016. Correlation analysis was carried out and the associations between productivity, human resource management (HRM)
practices, company profiles, and project characteristics were investigated. Logistic regression analysis was conducted to develop the
probability-based labor productivity prediction model. Project delay is found to be negatively correlated with HRM practices, whereas com-
pany size is positively associated with HRM practices. A scoring tool to measure the levels of HRM practice implementation on building
projects was developed. On that basis, a logistic regression model of HRM practices and productivity was built. This study contributes to the
body of knowledge by proposing a tool that can be used to assess the odds of having high productivity based on the implementation levels of
HRM practices on a certain building project. DOI: 10.1061/(ASCE)CO.1943-7862.0001775. © 2019 American Society of Civil Engineers.
Author keywords: Construction labor productivity; Building project; Human resource management practices.

Introduction The level of implementation of construction management prac-


tices varies from company to company, and a lower level of prac-
The construction industry in most countries accounts for the high- tices is associated with lower productivity (Gurmu et al. 2016).
est share of employment. In the context of Australia, the analysis of Thus, to increase labor productivity, it is important to implement
data obtained from Australian Bureau of Statistics (ABS) showed a higher level of HRM practices. Hence, a tool that can be used to
that, in November 2018, more than 1.1 million workers were in- assess whether the implementation level of HRM practices is low or
volved in construction activities (ABS 2019). The construction in- high has paramount significance. Furthermore, predicting the pro-
dustry comprised about 9.1% of the total workforce in Australia, ductivity associated with a certain level of implementation of HRM
and it is exceeded by only two other sectors, health care and social practices is vital. For instance, if the productivity value that is pre-
assistance (13.2%) and retail trade (10.2%), in terms of the size of dicted based on the implementation levels of HRM practices is low,
the labor force (ABS 2019). Besides the number of workers, the a construction project manager will take necessary remedial ac-
construction industry has higher wage rates since various allow- tions. HRM practice scoring/measuring tools for industrial and in-
ances are added to the base rate. For instance, the Australian Fair frastructure construction projects were developed in North America
Work Ombudsman allows the inclusion of an industry allowance, (CII 2013a, b). However, since construction projects have their own
tool and employee protection allowance, underground allowance, uniqueness, it was found that these tools cannot directly be used for
and others in payments to workers involved in construction projects building projects in Australia because of differences in project lo-
(Fair Work Ombudsman 2019). Hence, the labor cost in construc- cations, scopes, and the relative importance of the practices (Gurmu
tion projects is high. In addition, project delays due to the loss of and Aibinu 2017).
productivity can lead to extra labor costs. Nonetheless, proper man- Additionally, a model that can be used to predict the productiv-
agement of human resources could save unnecessary expenses ity associated with a certain level of implementation of HRM prac-
that arise as a result of a low level of productivity. Hence, it is es- tices has not been developed by researchers. Thus, the users of
sential to investigate the best human resource management (HRM) existing scoring tools do not have a comprehensive model that can
practices for managing construction workers and enhancing labor be used to predict the productivity level of their projects based on
productivity in construction projects. the measured values of HRM practices. Earlier studies affirmed that
logistic regression analysis has various advantages, which include
1
Lecturer in Construction Management, School of Architecture and its capability to develop a model that can integrate probability, pro-
Built Environment, Deakin Univ., Geelong, VIC 3220, Australia ductivity, and construction management practices (Gurmu 2018).
(corresponding author). ORCID: https://orcid.org/0000-0003-0750-4191. While predicting the productivity of building projects based on
Email: argaw.gurmu@deakin.edu.au the level of implementation of HRM practices is crucial, this issue
2
Associate Research Fellow, School of Architecture and Built Environ-
has gained little attention in research.
ment, Deakin Univ., Geelong, VIC 3220, Australia.
Note. This manuscript was submitted on February 21, 2019; approved
Previous researchers developed construction labor produc-
on July 30, 2019; published online on December 26, 2019. Discussion per- tivity prediction methods, such as a polynomial regression
iod open until May 26, 2020; separate discussions must be submitted for model (Srinavin and Mohamed 2003), a linear/multiple regres-
individual papers. This paper is part of the Journal of Construction En- sion model (Zhao et al. 2009; Thomas and Završki 1999), expert
gineering and Management, © ASCE, ISSN 0733-9364. systems (Fayek and Oduba 2005), and artificial intelligence

© ASCE 04019115-1 J. Constr. Eng. Manage.

J. Constr. Eng. Manage., 2020, 146(3): 04019115


(El-Gohary et al. 2017; Lu et al. 2000). However, previous studies labor productivity (Gurmu and Aibinu 2018). Hiyassat et al. (2016)
did not measure the HRM practices quantitatively and, on that ba- found that besides financial rewards, moral incentives such as ver-
sis, did not develop a probability-based model for predicting con- bal encouragement can increase worker productivity. Fagbenle et al.
struction labor productivity. In the real world, the achievement of (2004) compared the performances of construction projects that
an estimated level of labor productivity is uncertain because labor used nonfinancial incentives and those that did not. The study re-
productivity is affected by many factors that might not be consid- vealed that the productive time of bricklayers who received non-
ered in existing models. Thus, it is essential to associate estimated financial incentives improved significantly.
productivity with probability. As discussed in several previous studies on the effects of finan-
Following the previously mentioned knowledge gap, this re- cial and nonfinancial incentives on construction workers, financial
search aims to develop and validate a logistic regression model that and nonfinancial incentive schemes that include better payment
can be used for predicting the labor productivity factor (PF) based systems, appreciation, and job recognition have the potential to
on a score of HRM practices. In this paper, productivity refers to boost the morale of construction workers, which can ultimately
labor productivity at a project level and is defined as the ratio of lead to enhanced productivity.
output to input (Hasan et al. 2018). The output is the cost of a build-
Downloaded from ascelibrary.org by University of Waterloo on 12/30/19. Copyright ASCE. For personal use only; all rights reserved.

ing project in Australian dollars (AUD), and the input is the time
spent in completing the project. The PF is the ratio of actual pro- Career Development Plan and Social Activities
ductivity to planned productivity (Nasir et al. 2015). HRM practices such as social activities and career development
plans can play a significant role in motivating and building team
Literature Review spirit among construction workers. According to Hewage et al.
(2011), construction productivity may not be improved by only
Review of the relevant literature shows that financial incentive pro- working hard and fast. The authors suggested that the social skills
grams, nonfinancial incentive programs, career development plans, of workers should also be developed. Further, Kazaz and Ulubeyli
social activities, employee training, crew composition, stability of (2007) stated that social activity is one of the sociopsychological
organizational structure, clear delegation of responsibilities, exit in- factors that affect labor productivity. Promotion opportunities
terviews, and skill assessment are some of the HRM practices that also have been found to be the most important motivating factor
could enhance productivity in construction projects. These practi- enhancing labor productivity as perceived by Iranian CEOs
ces are briefly elaborated in subsequent subsections. (Ghoddousi et al. 2015). Similarly, Doloi (2007) suggested that
the prospect of promotion and social status are some of the impor-
Financial and Nonfinancial Incentive Programs tant human resource–related project attributes influencing labor
productivity.
Previous studies confirmed that there is a positive correlation be- Additionaly, Kim et al. (2015) identified social factors influ-
tween reward system and employee performance and satisfaction. encing productivity that include social security insurance, par-
Ling et al. (2018) found that the provision of transport and mobile ticipation in decision making, company policy, social recognition,
phone allowances led to the satisfaction of project managers. Sim- social life opportunities, job autonomy, self-development oppor-
ilarly, Lim and Ling (2012) revealed that contractors’ professionals tunities, social status, cultural differences, and job discretion. Ac-
who were given transport, meal, and mobile phone allowances, cordingly, a well-organized human resource development program
dental benefits, and overtime pay had higher job satisfaction. is an important strategy for the success of firms since human
According to Kazaz and Ulubeyli (2007), factors that increase capital plays a significant role in construction projects (Tabassi
employee motivation include incentive remuneration, equal pay-
et al. 2012).
ment for workers performing similar tasks, and the adequacy of
payments.
On the other hand, Durdyev and Mbachu (2018) concluded Employee Training
that lack of financial incentives is one of the factors that influence
labor productivity in residential projects in Cambodia. Ohueri et al. Providing appropriate training for employees can help to enhance
(2018) mentioned that financial incentives are one of the most their skills, which will subsequently increase their performance. A
significant motivation strategies that can positively influence con- study conducted by Wang et al. (2010) showed that training for
struction workers’ productivity. Similarly, Ghodrati et al. (2018) craftsmen increased labor productivity by 5%, reduced turnover
revealed that management strategies, such as communication and by 10%, and decreased absenteeism by 2.5%. Allmon et al. (2000)
incentive programs, have a strong positive relationship with labor suggested that training employees to be multiskilled has the poten-
productivity. Furthermore, Phua (2012) compared HRM practices tial to increase labor productivity in construction. Importantly,
in Australian and Hong Kong construction firms and found that activity-oriented training is found to be one of the practices that
Australian construction workers preferred individual-based remu- can drive labor productivity in construction projects (Rojas and
neration compared to Hong Kong workers. Albattah et al. (2016) Aramvareekul 2003). Further, Rojas and Aramvareekul (2003) rec-
confirmed that extrinsic rewards, such as wages, family-friendly ommended that construction workers attend site-specific training or
benefits, and job security, were among the top preferences for induction before commencing any activity on a site.
US construction workers during the period 2008–2014. Nonethe- Similarly, Jergeas (2009) proposed that investment for training
less, dissimilar pay scales for workers performing comparable jobs apprentices and site supervisors is important for enhancing the pro-
on the same project can negatively influence their productivity. ductivity performance of construction projects. This study is also
According to Chih et al. (2016), workers who perceived organiza- supported by the recommendation made by Srour et al. (2017) that
tional injustice, such as unequal payments, can experience psycho- investing in the workforce through training could help to increase
logical strain, which can lead to a reduced commitment to the employee retention and reduce absenteeism. Therefore, training is
organization. an essential practice as it reduces construction errors or reworks and
The availability of nonfinancial incentives for construction helps construction workers to perform their tasks to the required
workers is also one of the essential HRM practices that can enhance quality standard.

© ASCE 04019115-2 J. Constr. Eng. Manage.

J. Constr. Eng. Manage., 2020, 146(3): 04019115


Crew Composition, Stability of Organizational proposed a multiple linear regression model for predicting produc-
Structure, and Clear Delegation of Responsibilities tivity in hot and humid environments. The factors considered in the
model were heat tolerance time, work time, and thermal environ-
The composition of the construction crew, maintaining the stability
ment parameter.
of the organizational structure of a project, and delegation of
Fayek and Oduba (2005) built a fuzzy expert model to estimate
responsibilities are some of the HRM practices that affect labor pro-
ductivity. According to Liberda et al. (2003), team spirit and crew labor productivity in rigging pipe and welding pipe activities.
composition are some of the essential HRM practices influencing Poveda and Fayek (2009) showed how fuzzy logic and fuzzy expert
labor productivity in construction projects. Maintaining the stabil- systems could be used to predict and evaluate the performance of
ity of organizational structure was found to be the best practice for construction trade foremen. The authors suggested that the model
enhancing labor productivity in industrial projects (Caldas et al. could assist in measuring the effectiveness of a foreman, monitor
2014). In addition, CII (2013b) suggested that keeping the organi- improvements in that effectiveness over time, and identify areas
zational structure stable, reducing the turnover of key professionals, where training might be required. Tsehayae and Fayek (2016)
planning for unusual staff changes, and preparing a contract docu- also developed and optimized a context-specific fuzzy inference
system–based construction labor productivity model.
Downloaded from ascelibrary.org by University of Waterloo on 12/30/19. Copyright ASCE. For personal use only; all rights reserved.

ment that restricts the frequent replacement of key personnel are


some of the important HRM practices. Ezeldin and Sharara (2006) developed neural networks for
Nasir (2013) found that clear delegation of the responsibilities predicting the productivity rates of form assembly, steel fixing,
of construction workers is the best practice to increase labor pro- and concrete pouring activities in developing countries. Similarly,
ductivity in infrastructure construction projects. Hence, the forma- Heravi and Eslamdoost (2015) proposed an artificial neural network
tion of crews having appropriate skill levels of various trades such (ANN) model for measuring and predicting construction labor pro-
as concrete, masonry, and plastering, maintaining the stability of ductivity on the installation of concrete foundations of gas, steam,
the organizational structure of a project, and clarifying the roles and combined cycle power plant projects in Iran. El-Gohary et al.
and responsibilities of the construction workers can help to improve (2017) introduced a practical and standardized engineering ap-
the labor productivity of building projects. proach to control, predict, and then improve construction labor
productivity. The authors used an ANN to quantify and map the
relationship between construction labor productivity and the rel-
Exit Interviews, Skill Assessment, and Plan for evant influencing factors. According to Yi and Chan (2014) ANNs
Retaining Experienced Personnel can be considered strong prediction modeling techniques that have
Exit interviews and skill assessment are some of the practices that dynamic learning mechanisms with effective recognition capabil-
could improve productivity in construction projects (Gurmu and ities to predict production rates under any specific conditions.
Aibinu 2018). Kazaz and Ulubeyli (2007) found that sociopsycho-
logical factors influencing labor productivity include activities such
as carrying out interviews and understanding the problems of em- Research Methodology
ployees. Further, CII (2013b) concluded that skill assessment and
evaluation as well as exit interviews are the best HRM practices Survey data can be collected using mail, telephone, the Internet,
that could enhance the productivity of infrastructure projects. In personal interview, or group administration, and the survey can
addition, taking corrective actions based on construction workers’ be self-administered, or an interviewer can ask questions and record
performance reviews can also affect the productivity of workers the answers (Fowler and Cosenza 2009). In this research, the quan-
(Dai et al. 2009). titative data were collected using an interview questionnaire survey
Nasir (2013) concluded that assessing construction workers’ whereby the survey was self-administered. The respondents were
skills and checking their previous qualifications and experience be- asked questions face to face and requested to write their answers.
fore commencing construction activities can assist project manag- The approach can increase the response rate, and in this study, it
ers in forming good crews. Moreover, Hong et al. (2012) identified helped to clarify the respondents’ questions.
employee retention strategies such as the preparation of suitable Data regarding productivity and HRM practices were collected
employee appraisal systems, training and development, compensa- from contractors on 39 multistory building projects in the Victorian
tion, and empowerment. Kakar et al. (2015) found that there is a state of Australia using questionnaires. A list of prequalified con-
positive correlation between employee empowerment, compensa- tractors was obtained from the Victorian Government Department
tion, training and development, and retention. The study showed of Treasury and Finance, and the multistory buildings were selected
that the existence of intrinsic and extrinsic rewards can increase from the websites of the contractors (Department of Treasury and
employee retention. Hence, the retention of experienced employees Finance 2016). To be included on the list, contractors were assessed
can help a contractor retain knowledge and skills obtained from a against a range of criteria, including managerial, technical, and
particular construction project. As a result, the contractor’s com- financial capacities. Thus, according to the department’s require-
petitiveness could increase. ments, firms applying for prequalification must demonstrate that
they have the management capability, qualifications, expertise, ex-
perience, and technical and financial capacity to deliver construc-
Construction Labor Productivity Estimating Methods tion projects successfully.
Various modeling techniques have been proposed to estimate labor The questionnaire has two parts. In the first part, project- and
productivity based on influencing factors. These techniques include company-specific data, such as project cost, completion times, and
regression analysis, expert systems, and artificial intelligence companies’ annual turnover, were collected. Based on the gathered
(Yi and Chan 2014). data, the PF of the particular building projects were computed using
Srinavin and Mohamed (2003) developed a polynomial regres- Eq. (1). Here, PF is the ratio of the actual productivity to planned
sion model for estimating productivity for painting, bricklaying, productivity (Gurmu and Aibinu 2017). The PF is used since it is
and excavation under different thermal environments. The model a useful measure for comparing the productivity of construction
considered the nature of the construction task from light activity projects having various characteristics (Nasir 2013; McDonald and
to heavy physical work and clothing ensembles. Zhao et al. (2009) Zack 2004):

© ASCE 04019115-3 J. Constr. Eng. Manage.

J. Constr. Eng. Manage., 2020, 146(3): 04019115


ðactual productivityÞ models using the threefold and fourfold cross-validation techniques
PF ¼ were developed and compared before the final model was selected.
ðplanned productivityÞ
The selection criteria include the overall prediction accuracy and
ðproject costÞ=ðactual project durationÞ
¼ ð1Þ statistical significance of the variable in the model. Fig. 1 shows
ðproject costÞ=ðplanned project durationÞ the flow chart for the process used in the logistic regression model
development and validation.
In the preceding equation, it is assumed that the planned project To validate the logistic regression model, the probabilities of
duration is accurately estimated by construction project managers validation data sets were computed; the PFs of the projects, which
or planners. Furthermore, the data were collected from building were used as validation data sets, were predicted as binary values
projects with fixed-price contracts, and there is no significant varia- (1, 0) and compared with the actual PF; the receiver operating
tion between the actual and planned project costs. characteristic (ROC) curves were drawn and compared with the
In the second part of the questionnaire, data related to the im- reference lines; and area under the ROC curves were computed.
plementation levels of HRM practices were collected using un- According to Hanley and McNeil (1982), the area under curve
weighted scoring tools. The unweighted HRM scoring tools for (AUC) is interpreted as “acceptable” if the graph is close to the
Downloaded from ascelibrary.org by University of Waterloo on 12/30/19. Copyright ASCE. For personal use only; all rights reserved.

building projects (this study) were prepared based on the existing left top boundary, while an AUC of 0.5 and less implies less ac-
tools for engineering projects (CII 2013a). After analyzing the data curacy, and 1 indicates perfect accuracy.
and determining the weights of the levels of implementations of the To check the reliability of the logistic regression model, a boot-
practices, the unweighted scoring tools was converted to the strapping method using 1,000 samples was used. The reason for
weighted scoring tool (refer Table 3 in “Findings and Discussion” choosing 1,000 or more samples in the simulation process is that
section). the method provides more accurate output in bootstrapping
To determine the weights, data regarding the relative importance (Willems and Van Aelst 2005). Finally, sigmoid graphs were drawn
of each practice was collected on a scale of 1–5. The relative im- by computing the probabilities of exceeding the baseline PF using
portance index (RII) of the practices was then computed using the the validated model’s equation.
following equation (El-Gohary and Aziz 2013)
5ðn5 Þ þ 4ðn4 Þ þ 3ðn3 Þ þ 2ðn2 Þ þ n1 Findings and Discussion
RII ¼ ð2Þ
5ðn5 þ n4 þ n3 þ n2 þ n1 Þ
In this section, the findings of the correlation analyses of produc-
where n1 , n2 , n3 , n4 , and n5 represent the number of respondents tivity and HRM practices, the weights of the HRM practices, and
who rated the importance of each HRM practice for improving pro- the logistic regression analysis are presented and discussed.
ductivity as not important, slightly important, somewhat important,
very important, and extremely important, respectively.
Relationships among HRM Practices, Labor
Before running the logistic regression analysis for developing a
Productivity, and Project Characteristics and Company
model, the suitability of the data was checked by computing the
Profiles
number of events per variable (EPV). The rule of thumb for the
sample size for logistic regression analysis states that EPV should As shown in Table 1, productivity and HRM practices are posi-
be greater than 10 (Peduzzi et al. 1996). In this research, the number tively associated (rs ¼ 0.36, p ¼ 0.02 < 0.05), and the relationship
of positive events refers to the number of projects having produc- is statistically significant, which indicates that increasing the levels
tivity factors exceeding the baseline PF; the predictor is the HRM of implementation of HRM practices can improve labor productiv-
practices (one variable); an EPVof 10 is used to develop the logistic ity on building projects. Among the 11 HRM practices, the stability
regression models. The suitability of the survey data was checked of organizational structure has the highest correlation coefficient
using an EPV test, and acceptable results were found. That is, all (rs ¼ 0.47, p ¼ 0.003 < 0.05). This finding shows that project
EPV values are greater than 10 (see Tables 4 and 5 in the “Findings managers who maintain the stability of their organizational struc-
and Discussion” section for the EPV values). Thus, the sample size tures can achieve a higher level of labor productivity than project
is adequate to enable a valid conclusion. managers who do not have a stable project environment. On the
Further, the three measurements of central tendency (mean, basis of the results shown in Table 1, it can also be confirmed that
mode, and median) of the projects’ productivity factors and Z-score all of the HRM practices are interrelated.
for the skewness of the productivity factors’ data (skewness/ The results of the correlation analysis for project delay, project
standard error) were computed, and the baseline PF was fixed (Lurd cost, and HRM practices are presented in Table 2. As shown,
and Lurd 2013; Kim 2013). To determine the baseline PFs that project delay is negatively correlated with HRM practices (rs ¼
satisfy the requirements of logistic regression models, the EPVs −0.30, p ¼ 0.06), which implies that the increment in the imple-
were calculated using various PF values: (1) mean PF, (2) mean mentation levels of HRM practices can reduce delays in building
PF þ 5% × mean PF, (3) mean PF − 5% × mean PF, (4) mean construction projects. On the other hand, project costs and HRM
PF þ 10% × mean PF, and (5) mean PF − 10% × mean PF. After practices are positively associated (rs ¼ 0.51, p ¼ 0.001), which
fixing the cut-off (baseline) productivity factors, the data sets were indicates that projects having a larger scope could also have higher
divided into model building and validation. Random numbers using levels of implementation of HRM practices, and among the 11
Microsoft Excel was assigned to each data point so that the data practices, “stability of organizational structure” has the highest
could be randomly distributed. Based on the random numbers, the correlation coefficient (rs ¼ 0.42, p ¼ 0.01).
data were sorted in ascending order before being split into two. In contrast, the annual turnover of construction companies is
In regression analysis, random sample splitting and K-fold found to be very weakly correlated with HRM practices (rs ¼
cross-validation techniques are applied to divide the data into 0.10, p ¼ 0.65). This indicates that construction companies might
model building and validation data sets (Steyerberg et al. 2001). not change their HRM practices based on the number of projects
In this research, threefold and fourfold cross-validation techniques they carry out. Also, company experience is moderately associated
were used because they produce less bias as compared to the with HRM practices (rs ¼ 0.32, p ¼ 0.05). Similarly, the relation-
sample splitting method (Abou-Assaleh et al. 2014). Alternative ship between company size and HRM practices is significant and

© ASCE 04019115-4 J. Constr. Eng. Manage.

J. Constr. Eng. Manage., 2020, 146(3): 04019115


Downloaded from ascelibrary.org by University of Waterloo on 12/30/19. Copyright ASCE. For personal use only; all rights reserved.

Fig. 1. Flowchart of logistic regression model.

has moderate strength (rs ¼ 0.32, p ¼ 0.045). This implies that LevelðN þ 1Þ ¼ Level N þ RII=5 ð3Þ
larger companies, which have many employees to manage, imple-
ment higher levels of HRM practices and smaller companies adopt where N ranges from 0 to 4; Level 0 = 0; and RII denotes the weight
lower levels. Among the 11 HRM practices, the stability of organi- of a practice.
zational structure has a stronger association with company size
(rs ¼ 0.38, p ¼ 0.02).
Logistic Regression Model of HRM Practices and
Productivity
Weights of HRM Practices
The results of the logistic regression analyses using threefold and
The weights of HRM practices are presented as follows: clear fourfold cross-validation techniques are presented in Tables 4 and 5,
delegation of responsibility = 0.836, stability of organizational respectively. Among the 15 models in the threefold cross validation,
structure = 0.815, crew composition = 0.769, retention plan for 5 are chosen based on their predictive accuracy and statistical sig-
experienced personnel = 0.764, skill assessment and evaluation = nificance for further analysis.
0.738, career development plan = 0.708, financial incentive The model with the highest predictive accuracy (T) is obtained
programs = 0.672, social activities = 0.672, nonfinancial incentive by analyzing the data sets named 21 and 31 (Table 4) and using a
programs = 0.651, and exit interviews = 0.574. The weight of each cut-off PF value of 0.85. The second model (U) is developed using
practice is then proportionally distributed among the five levels of the same data sets as that of Model T but using PF ¼ 0.90 as the
implementations using Eq. (3) (Gurmu 2018), and the weighted cut-off value. The third (V), fourth (W), and fifth (X) models are
scoring tools indicated in Table 3 are prepared. For instance, for built using data sets similar to that of Model T but using cut-off PF
the practice crew composition, RII ¼ 0.769, Level 0 ¼ 0; N ¼ 0, values of 0.95, 1.00, and 1.05, respectively (Table 4). While run-
Level 1 ¼ 0 þ 0.769=5 ¼ 0.154; N ¼ 1, Level 2 ¼ Level 1 þ ning the logistic regression analyses, projects having PFs greater
0.154 ¼ 0.308; N ¼ 2, Level 3 ¼ Level 2 þ 0.154 ¼ 0.461; N ¼ 3, than or equal to the cut-off PF were assigned a binary number
Level 4 ¼ Level 3 þ 0.154 ¼ 0.615; N ¼ 4, and Level 5 ¼ of 1, and projects having PFs of less than the cut-off PF value were
Level 4 þ 0.154 ¼ 0.769 assigned a binary number of 0.

© ASCE 04019115-5 J. Constr. Eng. Manage.

J. Constr. Eng. Manage., 2020, 146(3): 04019115


Table 1. Spearman correlation coefficients for productivity and HRM practices
Variable Parameter PF HRM CC SA ET CD NFIP FIP SoA SOS CDR RPEP EI
PF Coefficient (rs ) 1.000 — — — — — — — — — — — —
Significance — — — — — — — — — — — — —
HRM Coefficient (rs ) 0.364* 1.000 — — — — — — — — — — —
Significance 0.023 — — — — — — — — — — — —
CC Coefficient (rs ) 0.190 0.724** 1.000 — — — — — — — — — —
Significance 0.246 0.000 — — — — — — — — — — —
SA Coefficient (rs ) 0.208 0.659** 0.630** 1.000 — — — — — — — — —
Significance 0.205 0.000 0.000 — — — — — — — — — —
ET Coefficient (rs ) 0.237 0.509** 0.587** 0.399* 1.000 — — — — — — — —
Significance 0.147 0.001 0.000 0.012 — — — — — — — — —
CD Coefficient (rs ) 0.218 0.685** 0.433** 0.487** 0.273 1.000 — — — — — — —
Significance 0.182 0.000 0.006 0.002 0.093 — — — — — — — —
Downloaded from ascelibrary.org by University of Waterloo on 12/30/19. Copyright ASCE. For personal use only; all rights reserved.

NFIP Coefficient (rs ) 0.056 0.534** 0.371* 0.320* 0.134 0.338* 1.000 — — — — — —
Significance 0.736 0.000 0.020 0.047 0.414 0.035 — — — — — — —
FIP Coefficient (rs ) 0.243 0.710** 0.506** 0.525** 0.307 0.566** 0.573** 1.000 — — — — —
Significance 0.135 0.000 0.001 0.001 0.057 0.000 0.000 — — — — — —
SoA Coefficient (rs ) 0.367* 0.558** 0.290 0.276 0.167 0.528** 0.486** 0.493** 1.000 — — — —
Significance 0.022 0.000 0.073 0.089 0.310 0.001 0.002 0.001 — — — — —
SOS Coefficient (rs ) 0.466** 0.679** 0.314 0.352* 0.420** 0.275 0.140 0.263 0.369* 1.000 — — —
Significance 0.003 0.000 0.051 0.028 0.008 0.090 0.394 0.106 0.021 — — — —
CDR Coefficient (rs ) 0.277 0.584** 0.459** 0.347* 0.214 0.334* 0.018 0.174 0.048 0.491** 1.000 — —
Significance 0.087 0.000 0.003 0.031 0.190 0.038 0.912 0.290 0.770 0.001 — — —
RPEP Coefficient (rs ) 0.199 0.612** 0.246 0.136 0.242 0.323* 0.208 0.308 0.334* 0.569** 0.424** 1.000 —
Significance 0.224 0.000 0.131 0.410 0.138 0.045 0.205 0.056 0.038 0.000 0.007 — —
EI Coefficient (rs ) 0.260 0.672** 0.332* 0.145 0.301 0.522** 0.384* 0.376* 0.430** 0.476** 0.431** 0.452** 1.000
Significance 0.110 0.000 0.039 0.380 0.063 0.001 0.016 0.018 0.006 0.002 0.006 0.004 —
Note: PF = productivity factor; HRM = human resource management practices; CC = crew composition; SA = skill assessment and evaluation; ET = employee
training; CD = career development; NFIP = nonfinancial incentive programs; FIP = financial incentive programs; SoA = social activity; SOS = stability of
organizational structure; CDR = clear delegation of responsibility; RPEP = retention plan for experienced personnel; and EI = exit interview. Correlation is
significant at * p > 0.05, Correlation is significant at ** p > 0.01.

Table 2. Spearman correlation coefficients for project delay, project cost, and HRM practices
Variable Parameter HRM CC SA ET CD NFIP FIP SoA SOS CDR RPEP EI
Project delay Coefficient −0.30 −0.08 −0.10 −0.21 −0.14 −0.08 −0.18 −0.41* −0.47** −0.20 −0.20 −0.26
Significance 0.06 0.62 0.54 0.21 0.40 0.64 0.27 0.01 0.00 0.21 0.23 0.11
Project cost Coefficient 0.50** 0.29 0.41* 0.19 0.31 0.35* 0.38* 0.40* 0.42** 0.24 0.23 0.41**
Significance 0.00 0.07 0.01 0.25 0.05 0.03 0.02 0.01 0.01 0.15 0.16 0.01
Annual turnover Coefficient 0.10 −0.06 0.10 −0.14 0.33 0.02 −0.12 0.02 0.26 0.35 −0.38 0.39
Significance 0.65 0.79 0.65 0.51 0.11 0.92 0.56 0.93 0.20 0.09 0.06 0.05
Company experience Coefficient 0.32 0.14 0.28 −0.02 0.09 0.32* 0.26 0.05 0.27 0.18 0.27 0.24
Significance 0.05 0.40 0.09 0.90 0.60 0.05 0.11 0.77 0.10 0.28 0.10 0.15
Company size Coefficient 0.32* 0.04 0.18 0.01 0.23 0.28 0.28 0.20 0.38* 0.28 0.29 0.25
Significance 0.05 0.83 0.28 0.98 0.17 0.08 0.09 0.23 0.02 0.08 0.08 0.12
Note: HRM = human resource management practices; CC = crew composition; SA = skill assessment and evaluation; ET = employee training; CD = career
development; NFIP = nonfinancial incentive programs; FIP = financial incentive program; SoA = social activity; SOS = stability of organizational structure;
CDR = clear delegation of responsibility; RPEP = retention plan for experienced personnel; and EI = exit interviews; Correlation is significant at * p > 0.05;
Correlation is significant at ** p > 0.01.

The 20 models analyzed using fourfold cross validation are selected as the best models. Model Y has an insignificant constant
shown in Table 5. Among them, the one that is statistically signifi- and AUC.
cant and has better predictive capability (72.4%), Model Y, is Model U has good predictive capability (80.8%); its coeffi-
chosen for further analysis. cient of independent variable (β) is statistically significant (p ¼
Finally, six models (T, U, V, W, X, and Y) are compared to select 0.02 < 0.05), its AUC is greater than 0.5, and its p-values for both
the best HRM practices and productivity model (Table 6). To the variable and constant are statistically significant. Therefore,
choose the most appropriate model among the six, the ROC is Model U is selected as the best model. The cut-off PF of 0.90
drawn, and the AUC for each model is computed using SPSS-24. and Data Sets 21 and 31 were used in logistic regression analysis,
On the basis of the results indicated in Table 6, Model T has the and Data Set 11 was retained for validation purposes. The results of
highest predictive capability (84.6%) and AUC (0.96), but its the logistic regression analysis of the selected model (Model U) are
constant is insignificant (p ¼ 0.18 > 0.05). Thus, Model T cannot provided in Fig. 2.
be the best model. Models V, W, and X have low prediction accu- As shown in Fig. 2, in the Classification Table of Block 0
racies compared to Model U (Table 6); as a result, they were not (scenario without variable), the true negative percentage is 0%,

© ASCE 04019115-6 J. Constr. Eng. Manage.

J. Constr. Eng. Manage., 2020, 146(3): 04019115


Table 3. Weighted scoring tools for HRM practices
Level Description Weight
1. Crew composition
Level 0 Crew formation/composition is not applicable to this building project. 0.000
Level 1 Crew composition is not addressed in this project. 0.154
Level 2 Crew formation is addressed after the commencement of the construction of the building project. 0.308
Level 3 Crew formation is addressed before the commencement of construction works; experience and knowledge of employees, job 0.462
requirements, and location are considered.
Level 4 Continuation of Level 3, plus the performance of the crew is assessed after the completion of each task and corrective measures are taken 0.615
accordingly to improve productivity.
Level 5 Continuation of Level 4, plus the crew formation and its performance is assessed regularly on a daily, weekly, and monthly basis, and 0.769
necessary changes are made.
2. Skill assessment
Downloaded from ascelibrary.org by University of Waterloo on 12/30/19. Copyright ASCE. For personal use only; all rights reserved.

Level 0 Skill assessment is not applicable to this project. 0.000


Level 1 Skill assessment is not addressed. 0.148
Level 2 Skill assessment is addressed after the beginning of the construction of this project. 0.295
Level 3 Skill assessment is addressed before the beginning of construction work; previous experience, skill, and knowledge of workers are 0.443
considered.
Level 4 Continuation of Level 3, plus the assessment is made again after completion of each task, and adjustments are made accordingly to 0.591
improve productivity.
Level 5 Continuation of Level 4, plus the assessment is performed regularly on a daily, weekly, and monthly basis, and necessary changes are 0.738
made and monitored.
3. Employee training
Level 0 Employee training is not applicable to this building project. 0.000
Level 1 Employee training is not addressed. 0.150
Level 2 Employee training is addressed on the job site after the beginning of the project. 0.299
Level 3 Employee training is provided to employees when they begin working for the company, and if needed extra training will occur on the 0.449
job site.
Level 4 Continuation of Level 3, plus worker is certified to work in or supervise specific trade. Before each project, new training in the trade will 0.599
take place if necessary.
Level 5 Continuation of Level 4, plus a worker takes part in training for new technologies that are introduced in that trade annually and 0.749
biannually.
4. Career development
Level 0 Career development is not applicable to this building project. 0.000
Level 1 Career development is not addressed in this project. 0.142
Level 2 The organization does not have a formal career development plan for employees, but management will discuss future plans with them. 0.283
Level 3 The organization has a formal career development plan for employees, but it only addresses short-term career development. 0.425
Level 4 Continuation of Level 3, plus it addresses long-term career development and options. 0.566
Level 5 Continuation of Level 4, plus it addresses expected employee performance and how the performance will affect employee career 0.708
development.
5. Nonfinancial incentive programs
Level 0 Nonfinancial incentive/recognition programs are not applicable to this building project. 0.000
Level 1 Nonfinancial incentive programs are not addressed in this project. 0.130
Level 2 The organization has informal recognition programs that will recognize employees occasionally, but not in a formal manner. 0.261
Level 3 The organization has a formal recognition program that provides recognition on a long-term basis. 0.391
Level 4 Continuation of Level 3, plus it recognizes workers on a regular basis for both positive safety results and good safety behavior. 0.521
Level 5 Continuation of Level 4, plus rewards are given on both a short-term and long-term basis, and employees are recognized by the 0.651
organization’s upper management.
6. Financial incentive programs
Level 0 Financial incentive program is not applicable to this building project. 0.000
Level 1 Financial incentive program is not addressed. 0.134
Level 2 The organization has an informal incentive program that will recognize employees occasionally, but not in a formal manner. 0.269
Level 3 The organization has a formal incentive program that provides incentives on a long-term basis. 0.403
Level 4 Continuation of Level 3, plus it provides a monetary bonus for employees based on their performance. 0.537
Level 5 Continuation of Level 4, plus rewards are given on both a short-term and long-term basis, and they are recognized by the organization’s 0.672
upper management.
7. Social activities
Level 0 Social activities for employees are not applicable. 0.000
Level 1 Social activities for employees are not addressed. 0.134
Level 2 The organization does not formally plan social activities for employees, and there is only one annual organization-wide social activity. 0.269
Level 3 The organization formally plans a social activity once or twice a year, which project managers will attend, along with an annual 0.403
organization-wide social activity.
Level 4 Continuation of Level 3, plus several times throughout the year, which project managers will attend, along with an annual organization- 0.537
wide social activity.
Level 5 Continuation of Level 4, plus monthly events, which project managers will attend and upper management, including the president, will 0.672
attend on a quarterly basis, along with an annual organization-wide social activity.

© ASCE 04019115-7 J. Constr. Eng. Manage.

J. Constr. Eng. Manage., 2020, 146(3): 04019115


Table 3. (Continued.)
Level Description Weight
8. Stability of organizational structure
Level 0 Maintaining the stability of the organizational structure is not important. 0.000
Level 1 No plans to manage change of key people in contract. 0.163
Level 2 Key professionals are named or define in the contract. 0.326
Level 3 Continuation of Level 2, plus key professionals cannot be changed without notice and prior approval. 0.489
Level 4 Continuation of Level 3, plus there are preapproved designated successors. 0.652
Level 5 Continuation of Level 4, plus contract specifies all professionals of subcontractor and principal contractor teams, along with possible 0.815
successors and right of approval by the other party.
9. Clear delegation of responsibility
Level 0 Clear delegation of responsibility is not applicable to this building project. 0.000
Level 1 There is simple and centralized delegation of responsibility. 0.167
Downloaded from ascelibrary.org by University of Waterloo on 12/30/19. Copyright ASCE. For personal use only; all rights reserved.

Level 2 There is simple and very formal delegation of responsibility. 0.334


Level 3 There is a stable project environment and more formal delegation of responsibility. 0.502
Level 4 The delegation of responsibility is formal, but differing between technical, administrative, and so forth. 0.669
Level 5 There is a formal delegation of authority that is clearly defined for all involved parties. The plan is reviewed periodically and evolves 0.836
when necessary.
10. Retention plan for experienced personnel
Level 0 Retention plan for experienced personnel is not applicable to this building project. 0.000
Level 1 A retention plan is not addressed. 0.153
Level 2 Each project manager is responsible for the retention of his workers. 0.306
Level 3 Incentives such as employee training and others are available but not required. Junior staff have higher pay and preferred hiring status on 0.458
the next project for the same employer.
Level 4 Continuation of Level 3, plus employee training is required for junior staff. The employer makes available a list of opportunities for the 0.611
next project.
Level 5 Continuation of Level 4, plus testing and certification on the site lead to pay increases. Employer meets with individual worker and 0.764
offers job(s) at new project site(s) as per requirements.
11. Exit interviews
Level 0 Exit interview is not applicable. 0.000
Level 1 No exit interview. 0.115
Level 2 There is an exit interview for key personnel only. 0.230
Level 3 Random exit interviews as time allows. 0.345
Level 4 Formal exit interview for all personnel. 0.459
Level 5 Formal exit interview for all personnel and feedback to management about lessons learned and how to improve retention when 0.574
applicable.

Table 4. Alternative HRM practices and productivity models using the true positive percentage is 100%, and overall model accuracy is
threefold cross-validation technique 69.2%. The variable in the equation is insignificant (p ¼ 0.06 >
Baseline 0.05), and the variable not in the equation is significant (p ¼
productivity factor Selection criterion Trial 1a Trial 2b Trial 3c 0.02 < 0.05). The variable not in the equation in Block 0 refers
Cut-off PF1 ¼ 0.85 Predictive accuracy 84.6T 80.8 76.9
to the independent variable or HRM practices. Thus, the model that
Significance of β 0.04 0.25 0.89 incorporates the independent variable or the model in Block 1 is
EPV 31 31 31 more significant than the model in Block 0.
Cut-off PF2 ¼ 0.90 Predictive accuracy 80.8U 65.4 65.4 In Block 1 (scenario with variables), the chi-squared (11.2) of
Significance of β 0.02 0.55 0.87 the omnibus test is statistically significant at a 5% level of signifi-
EPV 26 26 26 cance (p ¼ 0.001 < 0.05), which shows that the model in Block 1
Cut-off PF3 ¼ 0.95 Predictive accuracy 76.9V 61.5 53.8 is better than the null model (model in Block 0). The p-value of
Significance of β 0.02 0.23 0.67 the chi-squared (3.28) of Hosmer and Lemeshow test of goodness
EPV 23 23 23 is greater than the rejection percentage of the null hypothesis
Cut-off PF4 ¼ 1.00 Predictive accuracy 73.1W 61.5 61.5
(p ¼ 0.86 > 0.05). Consequently, the null hypothesis, which states
Significance of β 0.03 0.12 0.24
EPV 21 21 21
that HRM practices and the productivity model are good in fitting
Cut-off PF5 ¼ 1.05 Predictive accuracy 73.1X 76.9 76.9 the data, is accepted.
Significance of β 0.04 0.25 0.17 Lastly, using the coefficients presented in Fig. 2, the equation
EPV 12 12 12 for the logistic regression model of productivity and HRM practices
is constructed as follows:
Note: β = parameter estimate; superscripts T, U, V, W, and X represent the
best model in threefold cross validation.  
Pi
a
Validation data set (1–13 data points), model building data set (14–26 data logðOddsÞ ¼ logitðPi Þ ¼ Ln ¼ 1.5HRMTSi -5.68
points), and model building data set (27–39 data points). 1-Pi
b
Model building data set (1–13 data points), validation data set (14–26 data
points), and model building data set (27–39 data points). where Pi = probability of exceeding PF ¼ 0.90 for ith project; and
c
Model building data set (1–13 data points), model building data set (14–26 HRMTSi = HRM practices total score for ith project. The preced-
data points), and validation data set (27–39 data points). ing equation can be simplified as indicated in Eq. (4)

© ASCE 04019115-8 J. Constr. Eng. Manage.

J. Constr. Eng. Manage., 2020, 146(3): 04019115


Table 5. Alternative HRM practices and productivity models using fourfold cross-validation technique
Baseline productivity factor Selection criteria Trial 4a Trial 5b Trial 6c Trial 7d
Cut-off PF1 ¼ 0.85 Predictive accuracy 86.2 89.7 72.4 80.0
Significance of β 0.07 0.09 0.31 0.86
EPV 31 31 31 31
Cut-off PF2 ¼ 0.90 Predictive accuracy 72.4Y 79.3 58.6 70.0
Significance of β 0.03 0.20 0.35 0.83
EPV 26 26 26 26
Cut-off PF3 ¼ 0.95 Predictive accuracy 69.0 69.0 69.0 56.7
Significance of β 0.03 0.09 0.17 0.55
EPV 23 23 23 23
Cut-off PF4 ¼ 1.00 Predictive accuracy 65.5 69.0 58.6 70.0
Significance of β 0.04 0.04 0.08 0.18
EPV 21 21 21 21
Downloaded from ascelibrary.org by University of Waterloo on 12/30/19. Copyright ASCE. For personal use only; all rights reserved.

Cut-off PF5 ¼ 1.05 Predictive accuracy 75.9 58.6 82.8 60.0


Significance of β 0.06 0.10 0.17 0.13
EPV 12 12 12 12
Note: β = parameter estimate; the superscript Y represents the best model in fourfold cross validation.
a
Validation data set (1–10 data points), model building data set (11–20 data points), model building data set (21–30 data points), and model building data set
(31–39 data points).
b
Model building data set (1–10 data points), validation data set (11–20 data points), model building data set (21–30 data points), and model building data set
(31–39 data points).
c
Model building data set (1–10 data points), model building data set (11–20 data points), validation data set (21–30 data points), and model building data set
(31–39 data points).
d
Model building data set (1–10 data points), model building data set (11–20 data points), model building data set (21–30 data points), and validation data set
(31–39 data points).

Table 6. Comparison of selected logistic regression models of HRM practices and productivity
Model’s Overall Model Significance of Significance Constant Significance Model’s data Validation
code accuracy (%) AUC model’s AUC Variable’s β of β (β o ) of β o Cut-off PF set code data set
T 84.6 0.96 0.001 0.99 0.04 −2.657 0.18 0.85 21 þ 31 11(3-fold)
U 80.8 0.90 0.001 1.50 0.02 −5.677 0.04 0.90 21 þ 31 11(3-fold)
V 76.9 0.84 0.002 1.34 0.02 −5.616 0.03 0.95 21 þ 31 11(3-fold)
W 73.1 0.83 0.003 0.92 0.03 −5.060 0.03 1.00 21 þ 31 11(3-fold)
X 73.1 0.83 0.013 1.12 0.04 −4.745 0.04 1.05 21 þ 31 11(3-fold)
Y 72.4 0.66 0.180 0.95 0.03 −3.611 0.07 0.90 24 þ 34 þ 44 14(4-fold)

e1.5HRMTSi -5.68 of HRM practices (HRMTS). The HRMTS is computed using the
Pi ¼ ð4Þ weighted scoring tools (Table 3).
1 þ e1.5HRMTSi -5.68

The reliability test (bootstrapping) results for the HRM practices Practical Application of Logistic Regression Model of
and productivity model is also presented in Fig. 2. The coefficient HRM Practices and Productivity
(1.50) and the constant (−5.68) obtained after bootstrapping
using 1,000 samples were found to be similar to the model’s co- The weighted scoring tools of HRM practices (Table 3) and the
efficient (1.50) and constant (−5.68). Moreover, the bootstrapped logistic regression model [Eq. (4)] can assist a contractor’s project
coefficient and constant were found to be statistically significant manager to plan appropriate HRM practices that can be imple-
(p ¼ 0.01 < 0.05). This confirms that the model in Block 1 is mented on a certain building construction project. The project
reliable. manager can make the necessary decisions to improve project
To validate the logistic regression model, the probability of productivity after assessing the risk of low productivity using the
exceeding PF ¼ 0.90 for the validation data Set 11 (Table 4) scoring tool and the model.
was predicted using the model’s equation [Eq. (4)]; the ROC curve The application of the scoring tool and the logistic regression
was drawn as shown in Fig. 3 and interpreted with respect to the model is illustrated as follow. Suppose a project manager plans
reference line; and the AUC was computed. For a logistic regres- to implement Level 2 for all 11 HRM practices, as shown in Table 7.
sion to be valid, the value of AUC should be greater than 0.5, and Based on the project manager’s plan, the total score of human
the graph should be above the diagonal line (Hanley and McNeil resource management practices (HRMTS) is computed to be 3.18
1982). Accordingly, the ROC curve is greater than the reference (Table 7).
line, indicating that the model is good at prediction. Further, the Alternatively, the project manager can use the following equa-
AUC is calculated to be 0.67, which is higher than the minimum tion to compute the total score of the HRM practices:
acceptable value of 0.5. Thus, the model is valid.
Using Eq. (4), the probabilities of exceeding PF ¼ 0.90 were HRMTS ¼ CCS þ SAS þ ETS þ CDS þ NFIPS þ FIPS
computed and the sigmoid graph indicated in Fig. 4 was plotted
so that users can see the probability that corresponds to a total score þ SOAS þ SOSS þ CDRS þ RPEPS þ EIS ð5Þ

© ASCE 04019115-9 J. Constr. Eng. Manage.

J. Constr. Eng. Manage., 2020, 146(3): 04019115


Classification Table
Predicted
PF Percentage Correct
O b s e rv e d 0 1
0 0 8 0.0
PF
Step 0 1 0 18 100.0
Overall Percentage 69.2
The cut value is 0.50
Variables in the Equation
Step 0 B S. E Wald df significance E x p( B )
Constant 0.811 0.425 3.642 1 0.056 2.25
Variables not in the Equation
Score df significance
Step 0 HRM Practices 9.353 1 .002
Overall Statistics 9.353 1 .002
Downloaded from ascelibrary.org by University of Waterloo on 12/30/19. Copyright ASCE. For personal use only; all rights reserved.

Omnibus Test for Model Coefficients


Chi-square df significance
Step 1 Step 11.210 1 0. 0 0 1
Block 11.210 1 0.001
Model 11.21 0 1 0.001
Hosmer and Lemeshow Test
Step 1 Chi-square df s i gn i fi ca n ce
3.278 7 0.858
Classification Table
Predicted
PF Percentage Correct
Observed 0 1
Step 1 PF 0 4 4 50 .0
1 1 17 9 4.4
Overall Percentage 80.8
The cut value is 0.50
Variables in the Equation
β SE Wald df significance Exp(B)
Step 1 HRM Practices 1.504 .656 5.256 1 .022 4.500
Constant -5.677 2.817 4.062 1 .044 .003
Bootstrap
95% Confidence Interval
β Bias SE significance Lower Upper
Step 1 HRM Practices 1.504 1.676 18.708 0.006 0.779 5.982
Constant -5.677 -6.702 76.568 0.013 -21.42 -2.374
Note: β = parameter estimate; df = degree of freedom; SE = standard error

Fig. 2. Outputs of logistic regression analysis of HRM practices.

where HRMTS = human resource management practices total


score; CCS = crew composition score; SAS = skill assessment and
evaluation score; ETS = employee training score; CDS = career
development score; NFIPS = nonfinancial incentive programs
score; FIPS = financial incentive programs score; SOAS = social
activities score; SOSS = stability of organizational structure score;
CDRS = clear delegation of responsibility score; RPEPS = retention
plan for experienced personnel score; and EIS = exit interview
score.
Using a HRMTS of 3.18 (Table 7), the corresponding probability
of exceeding PF = 0.90 can be read from the sigmoid graph of
the HRM practices (Fig. 5). Accordingly, the probability value is
approximately 26% (Fig. 5). This indicates that the actual produc-
tivity of a particular project could be less than the planned produc-
tivity. In other words, the actual productivity could be low, and the
risk of occurring project delay is high. Thus, the project manager
should plan to implement higher levels of HRM practices and read
the probability from the graph until the desired level of productivity
Fig. 3. ROC curve for validating HRM practices and productivity
is achieved. Following the preceding discussion, it can be con-
model.
cluded that the model can assist the project manager in planning

© ASCE 04019115-10 J. Constr. Eng. Manage.

J. Constr. Eng. Manage., 2020, 146(3): 04019115


Downloaded from ascelibrary.org by University of Waterloo on 12/30/19. Copyright ASCE. For personal use only; all rights reserved.

Fig. 5. Demonstration of use of sigmoid graph of HRM practices.

For models based on expert systems, the rules obtained from


Fig. 4. Probability plot of HRM practices. experts are affected by personal prejudices and attitudes because
of the complex nature of productivity estimation (Yi and Chan
2014). For instance, Fayek and Oduba (2005) developed fuzzy ex-
pert rules based on the opinions of four experts. On the other hand,
suitable HRM practices to implement on a given building construc- the scoring tool proposed in this research helps to reduce the sub-
tion project. jectivity of the inputs to the logistic regression model. The descrip-
The advantages of using the model developed in this research tions for each level of implementation of the HRM practices are
compared to existing models are as follows. According to Yi and provided so that the users of the tool can objectively choose the
Chan (2014), multiple and polynomial regression analyses are types of practices.
generally limited by the number of influencing factors that can
be included and by their ability to measure the combined effect
of influencing factors. For example, Zayed and Halpin (2005) de- Conclusion
veloped a regression model for the prediction of pile construction
productivity based on a few factors. The model proposed by the Based on the analysis carried out in this research, it is found that
authors is composed of two independent variables (factors), auger clear delegation of responsibility, the stability of organizational
height and pile depth. Similarly, Jang et al. (2011) used four structure, and crew composition are the three most important HRM
variables: work characteristic, work technique, work management practices for enhancing labor productivity in building construction
component, and worker component. However, this study adopted projects. Following the correlation analysis, it is also indicated that
a scoring technique in which each of the 11 HRM practices project delay is negatively correlated with HRM practices, which
(independent variables) are scored, aggregated, and used in the implies that the increment in the level of implementation of HRM
prediction of labor productivity. practices could decrease the occurrence of delays in building

Table 7. Illustration for application of HRM practices scoring tool


Implementation levels
HRM practices Description Level Score
Crew composition Crew composition is addressed after commencement of construction of building project 2 0.31
Skill assessment and evaluation Skill assessment is addressed on job site after beginning of project 2 0.30
Employee training Employee training will be addressed on job site after beginning of project 2 0.30
Career development Organization has informal recognition programs to recognize employees, but management 2 0.28
discusses future plans with them informally
Nonfinancial incentive programs Organization has informal recognition programs to recognize employees occasionally, but not 2 0.26
in a formal manner
Financial incentive programs Organization has informal incentive program to provide incentives occasionally, but not in a 2 0.27
formal manner
Social activities Organization does not formally plan social activities for employees, and there is only an 2 0.27
annual organization-wide social activity
Stability of organizational structure Key professionals are named or defined in contract 2 0.33
Clear delegation of responsibility There is simple and very formal delegation of responsibility 2 0.33
Retention plan for experienced personnel Each project manager is responsible for retention of his/her workers 2 0.31
Exit interviews Exit interview is held for key personnel only 2 0.23
Total score 3.18

© ASCE 04019115-11 J. Constr. Eng. Manage.

J. Constr. Eng. Manage., 2020, 146(3): 04019115


projects. On the other hand, the relationship between company size Chih, Y.-Y., K. Kiazad, D. Cheng, A. Capezio, and S. L. D. Restubog.
and HRM practices is positive and significant. This shows that the 2016. “Does organizational justice matter? Implications for construc-
levels of implementation of HRM practices increase as a company tion workers’ organizational commitment.” J. Manage. Eng. 33 (2):
gets larger. Meanwhile, the project cost is found to be strongly as- 04016043. https://doi.org/10.1061/(ASCE)ME.1943-5479.0000490.
CII (Construction Industry Institute). 2013a. Best productivity practices
sociated with HRM practices. Finally, company experience and
implementation index for industrial projects. Construction Industry
project turnover are positively correlated with HRM practices. Institute Implementation Resource 252-3d. Austin, TX: CII.
During the project-planning phase, the weighted scoring tool CII (Construction Industry Institute). 2013b. Best productivity practices im-
and the logistic regression model can be used to plan appropriate plementation index for infrastructure projects. Construction Industry
HRM practices that could be implemented on a certain multistory Institute Implementation Resource 252-4d. Austin, TX: CII.
building project. Importantly, the model can be used when the Dai, J., P. M. Goodrum, and W. F. Maloney. 2009. “Construction
project manager wants to determine whether the HRM practices craft workers’ perceptions of the factors affecting their productivity.”
adopted on a certain project are associated with high or low pro- J. Constr. Eng. Manage. 135 (3): 217–226. https://doi.org/10.1061
ductivity compared to the baseline productivity factor. Based on the /(ASCE)0733-9364(2009)135:3(217).
predicted probability of exceeding the baseline productivity factor, Department of Treasury and Finance. 2016. “Construction supplier regis-
Downloaded from ascelibrary.org by University of Waterloo on 12/30/19. Copyright ASCE. For personal use only; all rights reserved.

the contractor’s project manager can then implement corrective ac- ter.” Accessed July 27, 2016. http://www.dtf.vic.gov.au.
tions that could increase the likelihood of achieving a higher level Doloi, H. 2007. “Twinning motivation, productivity and management
strategy in construction projects.” Eng. Manage. J. 19 (3): 30–40.
of productivity compared to the baseline.
https://doi.org/10.1080/10429247.2007.11431738.
The findings of this study can also provide useful information Durdyev, S., and J. Mbachu. 2018. “Key constraints to labour productivity
for international contractors who want to carry out construction in residential building projects: Evidence from Cambodia.” Int. J.
works in Australia. Moreover, since the relative importance and Constr. Manage. 18 (5): 385–393. https://doi.org/10.1080/15623599
type of HRM practices, which could enhance productivity in build- .2017.1326301.
ing and engineering projects, could differ, it is suggested that El-Gohary, K. M., and R. F. Aziz. 2013. “Factors influencing construction
future researchers should identify and prioritize those practices that labor productivity in Egypt.” J. Manage. Eng. 30 (1): 1–9. https://doi
are appropriate for increasing productivity in infrastructure and .org/10.1061/(ASCE)ME.1943-5479.0000168.
industrial (engineering) construction projects, particularly in the El-Gohary, K. M., R. F. Aziz, and H. A. Abdel-Khalek. 2017. “Engineering
Australian context. Researchers can use the findings of this study approach using ANN to improve and predict construction labor produc-
for comparison purposes. They can use a similar methodology to tivity under different influences.” J. Constr. Eng. Manage. 143 (8):
04017045. https://doi.org/10.1061/(ASCE)CO.1943-7862.0001340.
develop logistic regression models for predicting the probability
Ezeldin, A. S., and L. M. Sharara. 2006. “Neural networks for estimating
of achieving a desired level of productivity in infrastructure and
the productivity of concreting activities.” J. Constr. Eng. Manage.
industrial projects. Future research could also be conducted by 132 (6): 650–656. https://doi.org/10.1061/(ASCE)0733-9364(2006)
collecting data from unionized and nonunionized projects and 132:6(650).
analyze the types of HRM practices to be implemented to enhance Fagbenle, O. I., A. Y. Adeyemi, and D. A. Adesanya. 2004. “The impact of
productivity in the two categories of projects. Furthermore, sepa- non-financial incentives on bricklayers' productivity in Nigeria.”
rate probability-based regression models can be developed for Constr. Manage. Econ. 22 (9): 899–911.
unionized and nonunionized projects since the labor productivity Fair Work Ombudsman. 2019. “Building and construction general on-site
for unionized and nonunionized projects could be different. award 2010.” Accessed January 24, 2019. http://awardviewer.fwo.gov
.au/award/show/MA000020#P84_2825.
Fayek, A. R., and A. Oduba. 2005. “Predicting industrial construction labor
productivity using fuzzy expert systems.” J. Constr. Eng. Manage.
Data Availability Statement 131 (8): 938–941. https://doi.org/10.1061/(ASCE)0733-9364(2005)
131:8(938).
Data generated or analyzed in this study are available from the Fowler, F. J., Jr., and C. Cosenza. 2009. “Design and evaluation of survey
corresponding author by request. Information about the Journal’s questions.” In The SAGE handbook of applied social research methods,
data-sharing policy can be found here: http://ascelibrary.org/doi/10 375–412. Thousand Oaks, CA: SAGE Publications.
.1061/(ASCE)CO.1943-7862.0001263. Ghoddousi, P., O. Poorafshar, N. Chileshe, and M. R. Hosseini. 2015.
“Labour productivity in Iranian construction projects: Perceptions of
chief executive officers.” Int. J. Productivity Perform. Manage. 64 (6):
References 811–830. https://doi.org/10.1108/IJPPM-10-2013-0169.
Ghodrati, N., T. Wing Yiu, S. Wilkinson, and M. Shahbazpour. 2018.
Abou-Assaleh, T., N. Cercone, V. Keselj, and R. Sweidan. 2014. “N-gram- “Role of management strategies in improving labor productivity in
based detection of new malicious code.” In Proc., 28th Annual Int. general construction projects in New Zealand: Managerial perspective.”
Computer Software and Applications Conf., COMPSAC 2004, 41–42. J. Manage. Eng. 34 (6): 04018035. https://doi.org/10.1061/(ASCE)ME
New York: IEEE. .1943-5479.0000641.
ABS (Australian Bureau of Statistics). 2019. Labour force, Australia, Gurmu, A. T. 2018. “Tools for measuring construction materials manage-
detailed. Catalogue No. 6291.0.55.001. Canberra, Australia: ABS. ment practices and predicting labor productivity in multistory building
Albattah, M., Y. Shan, P. M. Goodrum, and T. R. Taylor. 2016. “Relation- projects.” J. Constr. Eng. Manage. 145 (2): 04018139. https://doi.org
ships between cycles of economic expansion in construction and craft /10.1061/(ASCE)CO.1943-7862.0001611.
workers’ job satisfaction and preferences.” Can. J. Civ. Eng. 44 (1): Gurmu, A. T., and A. A. Aibinu. 2017. “Construction equipment manage-
29–36. ment practices for improving labor productivity in multistory building
Allmon, E., C. T. Haas, J. D. Borcherding, and P. M. Goodrum. 2000. construction projects.” J. Constr. Eng. Manage. 143 (10): 04017081.
“U.S. construction labor productivity trends, 1970–1998.” J. Constr. https://doi.org/10.1061/(ASCE)CO.1943-7862.0001384.
Eng. Manage. 126 (2): 97–104. https://doi.org/10.1061/(ASCE)0733 Gurmu, A. T., and A. A. Aibinu. 2018. “Survey of management practices
-9364(2000)126:2(97). enhancing labor productivity in multi-storey building construction
Caldas, C. H., J.-Y. Kim, C. T. Haas, P. M. Goodrum, and D. Zhang. 2014. projects.” Int. J. Productivity Perform. Manage. 67 (4): 717–735.
“Method to assess the level of implementation of productivity practices https://doi.org/10.1108/IJPPM-02-2017-0032.
on industrial projects.” J. Constr. Eng. Manage. 141 (1): 04014061. Gurmu, A. T., A. A. Aibinu, and C. Toong Khuan. 2016. “A study of best
https://doi.org/10.1061/(ASCE)CO.1943-7862.0000919. management practices for enhancing productivity in building projects:

© ASCE 04019115-12 J. Constr. Eng. Manage.

J. Constr. Eng. Manage., 2020, 146(3): 04019115


Construction methods perspectives.” Constr. Econ. Build. 16 (3): 1–19. Nasir, H., C. T. Haas, C. H. Caldas, and P. M. Goodrum. 2015. “An
https://doi.org/10.5130/AJCEB.v16i3.4882. integrated productivity-practices implementation index for planning
Hanley, J. A., and B. J. McNeil. 1982. “The meaning and use of the area the execution of infrastructure projects.” J. Infrastruct. Syst. 22 (2):
under a receiver operating characteristic (ROC) curve.” Radiology 04015022. https://doi.org/10.1061/(ASCE)IS.1943-555X.0000275.
143 (1): 29–36. https://doi.org/10.1148/radiology.143.1.7063747. Ohueri, C. C., W. I. Enegbuma, N. H. Wong, K. K. Kuok, and R. Kenley.
Hasan, A., B. Baroudi, A. Elmualim, and R. Rameezdeen. 2018. “Factors 2018. “Labour productivity motivation framework for Iskandar
affecting construction productivity: A 30 year systematic review.” Eng. Malaysia.” Built Environ. Project Asset Manage. 8 (3): 293–304.
Constr. Archit. Manage. 25 (7): 916–937. https://doi.org/10.1108 https://doi.org/10.1108/BEPAM-09-2017-0070.
/ECAM-02-2017-0035. Peduzzi, P., J. Concato, E. Kemper, T. R. Holford, and A. R. Feinstein.
Heravi, G., and E. Eslamdoost. 2015. “Applying artificial neural networks 1996. “A simulation study of the number of events per variable in
for measuring and predicting construction-labor productivity.” J. Constr.
logistic regression analysis.” J. Clin. Epidemiol. 49 (12): 1373–1379.
Eng. Manage. 141 (10): 04015032. https://doi.org/10.1061/(ASCE)CO
https://doi.org/10.1016/S0895-4356(96)00236-3.
.1943-7862.0001006.
Phua, F. T. T. 2012. “Do national cultural differences affect the nature and
Hewage, K. N., A. Gannoruwa, and J. Y. Ruwanpura. 2011. “Current status
of factors leading to team performance of on-site construction profes- characteristics of HRM practices? Evidence from Australian and Hong
Downloaded from ascelibrary.org by University of Waterloo on 12/30/19. Copyright ASCE. For personal use only; all rights reserved.

sionals in Alberta building construction projects.” Can. J. Civ. Eng. Kong construction firms on remuneration and job autonomy.” Constr.
38 (6): 679–689. https://doi.org/10.1139/l11-038. Manage. Econ. 30 (7): 545–556. https://doi.org/10.1080/01446193
Hiyassat, M. A., M. A. Hiyari, and G. J. Sweis. 2016. “Factors affecting .2012.682074.
construction labour productivity: A case study of Jordan.” Int. J. Constr. Poveda, C. A., and A. R. Fayek. 2009. “Predicting and evaluating construc-
Manage. 16 (2): 138–149. https://doi.org/10.1080/15623599.2016 tion trades foremen performance: Fuzzy logic approach.” J. Constr.
.1142266. Eng. Manage. 135 (9): 920–929. https://doi.org/10.1061/(ASCE)CO
Hong, E. N. C., L. Z. Hao, R. Kumar, C. Ramendran, and V. Kadiresan. .1943-7862.0000061.
2012. “An effectiveness of human resource management practices on Rojas, E. M., and P. Aramvareekul. 2003. “Labor productivity drivers
employee retention in institute of higher learning: A regression analy- and opportunities in the construction industry.” J. Manage. Eng. 19 (2):
sis.” Int. J. Bus. Res. Manage. 3 (2): 60–79. 78–82. https://doi.org/10.1061/(ASCE)0742-597X(2003)19:2(78).
Jang, H., K. Kim, J. Kim, and J. Kim. 2011. “Labour productivity model for Srinavin, K., and S. Mohamed. 2003. “Thermal environment and construc-
reinforced concrete construction projects.” Constr. Innovation 11 (1): tion workers’ productivity: Some evidence from Thailand.” Build.
92–113. https://doi.org/10.1108/14714171111104655. Environ. 38 (2): 339–345. https://doi.org/10.1016/S0360-1323(02)
Jergeas, G. 2009. Improving construction productivity on Alberta oil and 00067-7.
gas capital projects. Calgary, AB, Canada: Univ. of Calgary. Srour, F. J., I. Srour, and M. G. Lattouf. 2017. “A survey of absenteeism on
Kakar, P., A. Raziq, and F. Khan. 2015. “Impact of human resource man- construction sites.” Int. J. Manpower 38 (4): 533–547. https://doi.org/10
agement practices on employee retention: A case of banking sector in
.1108/IJM-08-2015-0135.
Quetta Baluchistan.” J. Manage. Info 4 (3): 5–11. https://doi.org/10
Steyerberg, E. W., F. E. Harrell, G. J. Borsboom, M. Eijkemans,
.31580/jmi.v5i1.24.
Y. Vergouwe, and J. D. F. Habbema. 2001. “Internal validation of
Kazaz, A., and S. Ulubeyli. 2007. “Drivers of productivity among construc-
predictive models: Efficiency of some procedures for logistic regression
tion workers: A study in a developing country.” Build. Environ. 42 (5):
2132–2140. https://doi.org/10.1016/j.buildenv.2006.04.020. analysis.” J. Clin. Epidemiol. 54 (8): 774–781. https://doi.org/10.1016
Kim, H.-Y. 2013. “Statistical notes for clinical researchers: Assessing /S0895-4356(01)00341-9.
normal distribution (2) using skewness and kurtosis.” Restor. Dent. Tabassi, A. A., M. Ramli, and A. H. A. Bakar. 2012. “Effects of training
Endodontics 38 (1): 52–54. https://doi.org/10.5395/rde.2013.38.1.52. and motivation practices on teamwork improvement and task efficiency:
Kim, S., J.-D. Kim, Y. Shin, and G.-H. Kim. 2015. “Cultural differences in The case of construction firms.” Int. J. Project Manage. 30 (2): 213–224.
motivation factors influencing the management of foreign laborers in https://doi.org/10.1016/j.ijproman.2011.05.009.
the Korean construction industry.” Int. J. Project Manage. 33 (7): Thomas, H. R., and I. Završki. 1999. “Construction baseline productivity:
1534–1547. https://doi.org/10.1016/j.ijproman.2015.05.002. Theory and practice.” J. Constr. Eng. Manage. 125 (5): 295–303.
Liberda, M., J. Ruwanpura, and G. Jergeas. 2003. “Construction pro- https://doi.org/10.1061/(ASCE)0733-9364(1999)125:5(295).
ductivity improvement: A study of human, management and external Tsehayae, A., and A. R. Fayek. 2016. “Developing and optimizing context-
issues.” In Proc., Construction Research Congress. Reston, VA: ASCE. specific fuzzy inference system-based construction labor productivity
Lim, L. J. W., and F. Y. Y. Ling. 2012. “Human resource practices of models.” J. Constr. Eng. Manage. 142 (7): 04016017. https://doi.org/10
contractors that lead to job satisfaction of professional staff.” Eng. .1061/(ASCE)CO.1943-7862.0001127.
Constr. Archit. Manage. 19 (1): 101–118. https://doi.org/10.1108 Wang, Y., P. M. Goodrum, C. Haas, R. Glover, and S. Vazari. 2010. “Analy-
/09699981211192599. sis of the benefits and costs of construction craft training in the United
Ling, F. Y. Y., Y. Ning, Y. H. Chang, and Z. Zhang. 2018. “Human resource States based on expert perceptions and industry data.” Constr. Manage.
management practices to improve project managers’ job satisfaction.” Econ. 28 (12): 1269–1285. https://doi.org/10.1080/01446193.2010
Eng. Constr. Archit. Manage. 25 (5): 654–669. https://doi.org/10.1108 .524238.
/ECAM-02-2017-0030.
Willems, G., and S. Van Aelst. 2005. “Fast and robust bootstrap for LTS.”
Lu, M., S. Abourizk, and U. H. Hermann. 2000. “Estimating labor
Comput. Stat. Data Anal. 48 (4): 703–715. https://doi.org/10.1016/j
productivity using probability inference neural network.” J. Comput.
.csda.2004.03.018.
Civ. Eng. 14 (4): 241–248. https://doi.org/10.1061/(ASCE)0887-3801
(2000)14:4(241). Yi, W., and A. P. C. Chan. 2014. “Critical review of labor productivity re-
Lurd, A., and M. Lurd. 2013. “Measures of central tendency.” Accessed search in construction journals.” J. Manage. Eng. 30 (2): 214–225.
June 13, 2016. https://statistics.laerd.com. https://doi.org/10.1061/(ASCE)ME.1943-5479.0000194.
McDonald, D. F., and J. G. Zack. 2004. Estimating lost labor productivity Zayed, T. M., and D. W. Halpin. 2005. “Productivity and cost regression
in construction claims. Morgantown, WV: Association for the Advance- models for pile construction.” J. Constr. Eng. Manage. 131 (7): 779–789.
ment of Cost Engineering International. https://doi.org/10.1061/(ASCE)0733-9364(2005)131:7(779).
Nasir, H. 2013. “Best productivity practices implementation index (BPPII) Zhao, J., N. Zhu, and S. Lu. 2009. “Productivity model in hot and humid
for infrastructure projects.” Ph.D. dissertation, Dept. of Civil and Envi- environment based on heat tolerance time analysis.” Build. Environ.
ronmental Engineering, Univ. of Waterloo. 44 (11): 2202–2207. https://doi.org/10.1016/j.buildenv.2009.01.003.

© ASCE 04019115-13 J. Constr. Eng. Manage.

J. Constr. Eng. Manage., 2020, 146(3): 04019115

You might also like