Professional Documents
Culture Documents
Data Analysis and Interpretation
Data Analysis and Interpretation
CHAPTER 4
DATA ANALYSIS AND INTERPRETATION
“If we have data, let’s look at data. If all we have are opinions, let’s go with mine.”
— Jim Barksdale
This chapter investigates the data for analysis and interpretation for achieving the
research objectives. The chapter comprises of 4 subsections. Section 1 displays the
general analysis -respondent’s profile, tools used for HR analytics, application of HR
analytics and challenges in implementation of HR analytics. Section 2 displays the
common method variance study to ascertain the reliability of the data. The study was
made by three techniques – Harman single factor test, Common latent factor, and
Comprehensive CFA marker approach. Section 3 displays Factor analysis and
Section 4 briefs the Measurement Model, Path Analysis, Testing the moderator and
mediation impact on the study and PLS predict. The output and interpretations are
followed with each study made to achieve the objectives.
The empirical study was made through a structured questionnaire. The survey method
helped us in measuring the association of constructs with the adoption of HR analytics
and its consequences on strategic workforce management. The questionnaire was
prepared on google forms and data was collected online. Questionnaire was sent to
520 respondents across India. 492 respondents reverted with filled questionnaires and
478 were selected for the study after removing incomplete questionnaires.
A) Respondents profile
The study appreciates the responses from 478 respondents, among them 326 (62.8%)
were females and 152 (31.8%) were males. The number of female respondents were
more as we encountered more females in HR department in IT industry in India
(Brandl, Mayrhofer and Reichel, 2008). The 266 (55.6%) respondents were less than
40 years of age and 212 (44.4%) were 40 years and above. The majority of
66
Chapter 4
Gender
Age
Education
67
Chapter 4
Designation
Type of organisation
68
Chapter 4
FREQUENCY PERCENT
C) Application of HR Analytics
Daily HR receive good number of data related to employees. They hold a major
responsibility of analysing this data and generate meaningful reports covering all
major functionalities. To ascertain the results, we asked multiple response questions
to our respondents about the application of HR analytics in their tasks. The results
were analysed on dichotomous analysis by counting ‗yes‘ which is shown in Table 4.3
which shows HR are making most use of HR analytics in HR planning (78%)
followed by Training and Development (75%), Recruitment (61%), Quality
Improvement (45%), Performance Evaluation (44%), Employee satisfaction (42%),
Employee Empowerment (41%), Career Management (31%), Employee Retention
(26%), Knowledge Management (25%). The least applied areas are Employee
69
Chapter 4
Application of HR Analytics
400
350
300
250
200
150
100
50
0
FREQUENCY PERCENT
70
Chapter 4
Every new concept, technology or innovation invites some challenges. The case with
HR analytics adoption is similar as the HR practitioners found themselves struggling
with the numbers, systems and technology (Roger, 1983). Table 4.4 shows the mean
values and standard deviation of the challenges in implementation of HR analytics.
MEAN STANDARD
CHALLENGES IN IMPLEMENTATION OF HR ANALYTICS
DEVIATION
Low analytical skills in HR department 2.19 0.577
Time constraints is there into proper implementation of talent
2.47 0.684
acquisition strategies
Accuracy in organizational challenges 2.23 0.610
Data collection is problematic 2.34 0.691
Faces difficulties in data security 2.21 0.741
The quality of data is not appropriate for analytics 2.30 0.718
Teamwork environment issues 2.31 0.668
It has high investment cost 2.35 0.639
It has high operational cost 2.31 0.655
All mean values are more than 2, which depicts HRs working in IT industry in India
agreed that these challenges are there in implementation of HR analytics. The HR
professionals considers time constraint into proper implementation of talent
acquisition strategies as the biggest challenge (Mean = 2.47) followed by investment
cost (Mean = 2.35) and problems in data collection (Mean = 2.34).
The mean and standard deviation of each statement of the construct along with the
general questions are calculated. Based on each statement of the construct a code is
assigned. All the antecedents – general self-efficacy (GSE), quantitative self-efficacy
(QSE), effort expectancy (EE), performance expectancy (PE), social influence (SI),
tool availability (TA) and data availability (DA) are scaled on seven-point Likert
scale. All the segments of strategic workforce management – task performance (TP),
empowerment practice (EP), strategic participation (SP) and training and development
(TND) are scale on five-point Likert scale.
71
Chapter 4
General Self-Efficacy
In 2001, Ajzen stated in his studies that the perceived behavioural control is different
form the self-efficacy. The use of any technology is not related to what a person has
done in past or willing to do in future (Compeau and Higgins, 1995; Ajzen, 2001;
Shahbaz et.al., 2019). The factor is related to attitude, acceptance to change and
willingness to find solutions (Fred, 2016). Davenport (2013) stated that adopting
analytics is like adopting technology. This as factor is studied and results are in the
Table 4.5. We include seven statements, and their mean and standard deviation values
are obtained. The results shows that GSE5 have maximum mean with 5.79, followed
by GSE4 (5.76), GSE6 (5.69), GSE3 (5.58), GSE7 (5.56), GSE2(5.46) and GSE1
(5.41).
Code Standard
Statements Mean
Assigned Deviation
HR Analytics is easy to use. GSE1 5,41 1,23
HR Analytics is convenient to use. GSE2 5,46 1,29
I can use HR Analytics without much effort. GSE3 5,58 1,30
I can solve most problems if I invest the necessary effort
GSE4 5,76 1,20
related to HR Analytics.
When I am confronted with a problem, I can usually find
GSE5 5,79 1,11
several solutions through HR analytics.
If I am in trouble, I can usually think of a solution. GSE6 5,69 1,25
I can usually handle whatever comes my way through HR
GSE7 5,56 1,28
Analytics.
The complexities of the adopting the new technology and the need to widen the scope
of knowledge, quantitative self-efficacy is included to study as an antecedent of HR
analytics adoption. The studies have shown that attitudes are influenced by knowledge
and self-efficacy is dependent on the belief that a person can perform (Ajzen and
Fishbein, 2000). Quantitative self-efficacy as factor to adoption of HR analytics is
studied and results are in the Table 4.6. We include five statements, and their mean
72
Chapter 4
and standard deviation values are obtained. The results shows that QSE3 has
maximum mean of 5.63 followed by QSE4 (5.61), QSE5 (5.50), QSE2 (5.32), QSE1
(5.26) and QSE6 (5.24).
Effort Expectancy
73
Chapter 4
Performance Expectancy:
Johnston and Warkentin (2010) and Venkatesh et al. (2012) stated that every individual
has its own career plan and aspirations, where they seek if adoption can improve their job
performance. The individual will be inclined to adopt any new technology only if they see
a better performance or improved job quality. Although the new concept might not be
easy to use, resulting in more learning sessions by the users. This may give negative
results as all individuals might not think alike and refuse for upcoming challenges which
lead us to study performance expectancy as a factor of adoption of HR Analytics and the
results are in Table 4.8. The results shows that PE2 has maximum mean of 5.71 followed
by PE3 (5.63), PE1 (5.48) and PE4 (5.47).
Tool Availability
To deploy new technology as practice it‘s important that the organization make use of
appropriate system, software, and tools. The efforts are required to provide necessary
skills sets to work on these tools to analysis the data, visualize and make feasible
decisions. Carlson and Kavanagh (2011) postulated that new technique to viewing
data has changed the organization‘s evaluation process for human capital. The right
application of knowledge and tools has become a necessity to make best use of
analytics on HR data. The studies have claimed that we are short of skilled analyst in
HR domain, and we need to train the HR professionals. Tool availability was studied
as a factor of adoption of HR Analytics. The results are shown in Table 4.9. The
results shows that TA5 has maximum mean of 5.73 followed by TA4 (5.71), TA3
(5.68), TA2 (5.51) and TA1 (5.35).
74
Chapter 4
Data Availability:
75
Chapter 4
Social Influence
Individual thoughts, actions and decisions are easily be influenced by society and social
circle. The adoption of HR analytics can be influenced by peer group, social media,
society and annual reports of the organizations. If the HR practitioners considers the
benefits, the adoption will be convenient but adversely if they reject than it will
negatively impact the process. The desire to work on new tool can be just an attraction
gained from the buzz created by the media but that necessary does not mean that it will be
a success for all (Johnston, 2006). Social influence was studied as a factor of adoption of
HR Analytics and results are shown in Table 4.11. The results state that SI5 has
maximum mean of 5.58 followed by SI4 (5.56), SI3(5.51), SI2(5.48) and SI1(5.25).
Adoption of HR analytics
The adoption of new technique often faces restriction from its users, organisers of
policies, finance section and management delays. HR and team leaders have to take
diligent steps to make adoption a smooth process. Keeping the up and downs into the
consideration we had studied the level of adoption of HR analytics. Table 4.12 shows
the results where LOA4 has maximum mean of 5.73 followed by LOA3 (5.69), LOA5
(5.62), LOA2 (5.55) and LOA1 (5.16).
76
Chapter 4
Task Performance
Every change comes with responsibility, challenges, and stress. The adoption of HR
analytics comes with new experiences for its users. Task performance brings a focus
on accomplishment of task or assigned job with efficiency and effectiveness
(Koopmans, 2014). The employees have to focus on both quality and speeding up
with quantity of tasks. This comes with up-to-date knowledge and maintaining
accuracy. Task performance was studied as a consequence of adoption of HR
analytics in reference to strategic workforce management. Table 4.13 shows the
results where TP2 has maximum mean of 4.33 followed by TP4 (4.24), TP6 (4.22),
TP5 (4.21), TP3 (4.15) and TP1 (4.03).
77
Chapter 4
Empowerment practices
HR Analytics aims at right decision making which gives authority to its user to make
some of the decisions based on the insights drawn from the analysis. The employees
can keep a track on the expenses, progress of work, timeline and delivery of project.
In the entire course of project, data and insights helps the employees to stand on the
right side of action (Pare and Tremblay, 2000). This led us to study empowerment
practices as a consequence of HR Analytics as a base of strategic workforce
management. Table 4.14 shows the results for empower practices statements where
EP6 has a maximum mean of 4.14 followed by EP8 (4.13), EP5 (4.09), EP7 (4.09),
EP2 (4.03), EP4 (4.02) and EP1 (3.93).
The insights from data contributed well, in defining the need and areas of training and
development for the employees of the organisation. Waseem et al., 2013 stated that a
78
Chapter 4
systematic training program is a journey which can be achieved through a process and
right check. The efficiency of training is determined through HR Analytics. Thus, we
studied training and development as a strategic workforce management factor
determined as a consequence of HR Analytics. Table 4.15 shows the results of
training and development where TND3 has maximum mean of 4.10 followed by
TND4 (4.08), TND5 (4.08), TND2 (4.07), TND6 (4.05) and TND1 (3.86).
Strategic Participation
79
Chapter 4
Method variance study was performed on the collected data to ensure the data is bias
free. We used constructs from different sources and adopted different scales to avoid
method biasness and have introduced the marker variable in the questionnaire at the
time of collecting data. Marker variables are the indicators of some other variables
which are not related to the study. This is a concept focusing upon reducing or
eliminating common method variance. Williams et.al., (2010) has emphasized on the
unrelated construct in the survey instrument. These unrelated factors can be -
modernity, large city, symbolic patriotism, tolerance for ambiguity, trust, work
engagement, power distance, price consciousness, price premium, professional
identification, friendship, justice, change of attitude, commitment, liking sports,
physical development, financial interest, tenure, job satisfaction, leadership (Williams
et.al., 2010; Simmering, 2015). It was stated by authors that if these unrelated factors
are present in the survey constructs than it will lower the correlations between the
measured and manifested variables (Lindell and Whitney, 2001; William et.al., 2010;
Eichhorn, 2014). In the present study to avoid the possibilities of biasness in the data
further a marker variable – liking for sport (Simmering et.al., 2015), that is, cricket as
a variable was chosen and questions were asked on the likeliness of the game, which
were added in the survey instrument following the comprehensive CFA marker
approach (Williams et.al., 2010).
80
Chapter 4
81
Chapter 4
This marker variable is completely unrelated to the study and has no relation with any of
the construct studies in HR analytics research. Richardson et.al., (2009) defined such
variable as an ideal marker variable. We tested the data using three techniques: Harman
one factor approach, Common latent factor and Comprehensive CFA marker approach.
Firstly, conducted Harman one factor analysis in SPSS. According to this approach we
put all the items of each construct included in the study into factor analysis to determine
the majority of the variance explained by one single general factor (Fuller et al., 2016).
Principal component analysis was employed for single general factor with un-rotated
factor solution. The result in Table 4.17 shows that the first factor estimates the total
variance of 36.6% which is well less than the threshold value of 50% recommended by
(Podsakoff et al., 2003). Moreover, there are 12 resultant factors, with 71% total variance
explained, so one factor is not explaining majority of the variance. This analysis describes
that there is no method biasness exists in the present study.
Second technique which we took in this study is common latent factor (CLF) analysis.
This method is applied in AMOS and model in CLF (Figure 4.3) was created. First we
create latent factor which hold the common variance among all observed variables in
the model. Then establish path from this CLF to all observed variables in the model
constrained equal variance let us say ―a‖ (Table 4.18). This CLF is also constrained
variance as ―1‖. The latent factor is not covaried with other factors in the analysis.
Although other than latent factor others have a covariance. The results of this
technique will be seen in two ways. First, common method variance is the square of
common variance of each path and compare it with the threshold value of 50 percent
(Eichhorn B R, 2014). In the present study the common variance of each path is 0.353
with a significant t-value. The CMV is the square of 0.353 which comes out 0.124
(12.4%) which is much less than threshold value. Secondly, compare the standardized
regression weight of each variable in the model with CLF to without CLF model.
Further, if the difference is more than 0.200 Gaskin J (2021) then consider CMV is
present in the model. While comparing the resultant standardized regression weight,
found that the difference of standardized regression weight of each variable is less
than 0.200 (Table 4.19). Hence no CMV is present in this study.
82
Chapter 4
Relations Estimate S.E. C.R. P Label Relations Estimate S.E. C.R. P Label
GSE3 <--- GSE 1.000 GSE3 <--- CLF .353 .018 19.381 *** a
GSE7 <--- GSE .930 .044 21.068 *** GSE7 <--- CLF .353 .018 19.381 *** a
GSE5 <--- GSE .801 .039 20.597 *** GSE5 <--- CLF .353 .018 19.381 *** a
GSE4 <--- GSE .860 .042 20.245 *** GSE4 <--- CLF .353 .018 19.381 *** a
GSE1 <--- GSE .868 .044 19.567 *** GSE1 <--- CLF .353 .018 19.381 *** a
GSE2 <--- GSE .907 .047 19.224 *** GSE2 <--- CLF .353 .018 19.381 *** a
GSE6 <--- GSE .858 .045 19.107 *** GSE6 <--- CLF .353 .018 19.381 *** a
QSE3 <--- QSE 1.000 QSE3 <--- CLF .353 .018 19.381 *** a
QSE2 <--- QSE .988 .048 20.777 *** QSE2 <--- CLF .353 .018 19.381 *** a
QSE1 <--- QSE .983 .042 23.194 *** QSE1 <--- CLF .353 .018 19.381 *** a
QSE5 <--- QSE .878 .045 19.648 *** QSE5 <--- CLF .353 .018 19.381 *** a
QSE4 <--- QSE .759 .046 16.655 *** QSE4 <--- CLF .353 .018 19.381 *** a
QSE6 <--- QSE .550 .060 9.162 *** QSE6 <--- CLF .353 .018 19.381 *** a
EE3 <--- EE 1.000 EE3 <--- CLF .353 .018 19.381 *** a
EE2 <--- EE 1.152 .063 18.338 *** EE2 <--- CLF .353 .018 19.381 *** a
EE1 <--- EE .930 .059 15.674 *** EE1 <--- CLF .353 .018 19.381 *** a
EE4 <--- EE .798 .057 13.986 *** EE4 <--- CLF .353 .018 19.381 *** a
PE2 <--- PE 1.000 PE2 <--- CLF .353 .018 19.381 *** a
PE3 <--- PE 1.005 .051 19.516 *** PE3 <--- CLF .353 .018 19.381 *** a
PE1 <--- PE .936 .051 18.293 *** PE1 <--- CLF .353 .018 19.381 *** a
PE4 <--- PE .601 .063 9.537 *** PE4 <--- CLF .353 .018 19.381 *** a
SI3 <--- SI 1.000 SI4 <--- CLF .353 .018 19.381 *** a
SI5 <--- SI 1.005 .066 15.213 *** TA4 <--- CLF .353 .018 19.381 *** a
SI2 <--- SI 1.114 .056 19.848 *** TA2 <--- CLF .353 .018 19.381 *** a
SI1 <--- SI 1.039 .054 19.157 *** TA3 <--- CLF .353 .018 19.381 *** a
SI4 <--- SI .993 .052 19.071 *** TA5 <--- CLF .353 .018 19.381 *** a
TA4 <--- TA 1.000 TA1 <--- CLF .353 .018 19.381 *** a
TA2 <--- TA .979 .048 20.208 *** DA4 <--- CLF .353 .018 19.381 *** a
TA3 <--- TA .857 .043 20.039 *** DA2 <--- CLF .353 .018 19.381 *** a
TA5 <--- TA .830 .044 19.021 *** DA3 <--- CLF .353 .018 19.381 *** a
TA1 <--- TA .884 .049 18.152 *** DA1 <--- CLF .353 .018 19.381 *** a
DA4 <--- DA 1.000 LOA4 <--- CLF .353 .018 19.381 *** a
DA2 <--- DA 1.001 .048 20.930 *** LOA3 <--- CLF .353 .018 19.381 *** a
83
Chapter 4
Relations Estimate S.E. C.R. P Label Relations Estimate S.E. C.R. P Label
DA3 <--- DA .969 .050 19.319 *** LOA2 <--- CLF .353 .018 19.381 *** a
DA1 <--- DA .973 .050 19.445 *** LOA1 <--- CLF .353 .018 19.381 *** a
LOA4 <--- LOA 1.000 LOA5 <--- CLF .353 .018 19.381 *** a
LOA3 <--- LOA .964 .054 17.819 *** TP1 <--- CLF .353 .018 19.381 *** a
LOA2 <--- LOA 1.046 .059 17.777 *** TP5 <--- CLF .353 .018 19.381 *** a
LOA1 <--- LOA 1.128 .064 17.688 *** TP6 <--- CLF .353 .018 19.381 *** a
LOA5 <--- LOA .845 .061 13.942 *** TP3 <--- CLF .353 .018 19.381 *** a
TP1 <--- TP 1.000 TP4 <--- CLF .353 .018 19.381 *** a
TP5 <--- TP .894 .070 12.758 *** TP2 <--- CLF .353 .018 19.381 *** a
TP6 <--- TP .835 .063 13.148 *** EP4 <--- CLF .353 .018 19.381 *** a
TP3 <--- TP .854 .068 12.494 *** EP8 <--- CLF .353 .018 19.381 *** a
TP4 <--- TP .833 .068 12.178 *** EP2 <--- CLF .353 .018 19.381 *** a
TP2 <--- TP 1.039 .080 13.038 *** EP6 <--- CLF .353 .018 19.381 *** a
EP4 <--- EP 1.000 EP7 <--- CLF .353 .018 19.381 *** a
EP8 <--- EP .748 .044 16.843 *** EP5 <--- CLF .353 .018 19.381 *** a
EP2 <--- EP .896 .053 16.918 *** EP3 <--- CLF .353 .018 19.381 *** a
EP6 <--- EP .863 .052 16.742 *** EP1 <--- CLF .353 .018 19.381 *** a
EP7 <--- EP .840 .053 15.734 *** TND4 <--- CLF .353 .018 19.381 *** a
EP5 <--- EP .802 .055 14.575 *** TND6 <--- CLF .353 .018 19.381 *** a
EP3 <--- EP .740 .055 13.413 *** TND2 <--- CLF .353 .018 19.381 *** a
EP1 <--- EP .401 .053 7.538 *** TND1 <--- CLF .353 .018 19.381 *** a
TND4 <--- TND 1.000 TND5 <--- CLF .353 .018 19.381 *** a
TND6 <--- TND .946 .046 20.646 *** TND3 <--- CLF .353 .018 19.381 *** a
TND2 <--- TND .815 .043 18.854 *** SP4 <--- CLF .353 .018 19.381 *** a
TND1 <--- TND .598 .041 14.763 *** SP3 <--- CLF .353 .018 19.381 *** a
TND5 <--- TND .536 .042 12.855 *** SP2 <--- CLF .353 .018 19.381 *** a
TND3 <--- TND .446 .042 10.514 *** SP5 <--- CLF .353 .018 19.381 *** a
SP4 <--- SP 1.000 SP1 <--- CLF .353 .018 19.381 *** a
SP3 <--- SP .446 .050 9.007 *** SI3 <--- CLF .353 .018 19.381 *** a
SP2 <--- SP .501 .052 9.644 *** SI5 <--- CLF .353 .018 19.381 *** a
SP5 <--- SP .507 .052 9.688 *** SI2 <--- CLF .353 .018 19.381 *** a
SP1 <--- SP .265 .045 5.892 *** SI1 <--- CLF .353 .018 19.381 *** a
84
Chapter 4
85
Chapter 4
86
Chapter 4
Stage I: Model comparisons. The model fit results of the analyses for each model are
shown in Table 4.20, displaying the values of chi-square, degrees of freedom, and
Comparative Fit Index (CFI). It is revealed from the table that values of CFI in initial
model (Figure 4.4) are shorter than threshold value of the 0.9. The reason of this may
in part reflect the relatively large number (65) of items for the substantive variables
and the resultant constrained parameters in the factor loading matrix. In order to find
out the effect of method marker variable on loadings and to test the null hypothesis
that there is no impact of method marker variable on items comparison of Baseline
Model (Figure 4.5) and Method-C (Figure 4.6) Model were conducted. The
comparison gives in a chi-square difference of 68.4 with four degree of freedom,
which is less than the 0.05 so reject the hypothesis. This indicates that there is an
impact of marker variable on items.
*p<0.05
87
Chapter 4
88
Chapter 4
89
Chapter 4
Figure 4.6: Model C (Where all items have restricted variance “a”)
90
Chapter 4
In the next step, a comparison was computed between the Method-U (Figure 4.7) and
Method-C models to determine if the impact of the method marker variable was equal
for all of the 65 items loading on the substantive indicators. This comparison also
tested the null hypothesis that the method factor loadings are equal on each item. The
result of chi-square difference is 251.36 with 60 degree of freedom. As the p value is
less than 0.05, testing supports the rejection of the hypothesis that method marker
loading is equal on every item, hence different impact on items. In this study,
Method-U Model represents the best model for accounting for marker variance on
substantive indicators. The standardized factor loadings for the Method-U Model are
shown in Table 4.21. The values range from .342 to .900 and significant (p<.05). In
Marker Variable column marker factor loadings from Method-U Model mentioned,
some of the items were statistically significant at the p < .01 level and 0.05 level,
some were insignificant this indicating that the item was contaminated by a source of
method variance captured by the marker variable. These values were ranged from .00
to .236. The square of these loading specifies the percentage of variance in the item
associated with the marker variable. Median amount of marker variance in each
indicator was 0.6%. Finally, the correlations among the substantive latent variables
from the Method-U are ranged from 0.3 to 0.8.
91
Chapter 4
Marker
S. N0. Items GSE QSE EE PE SI TA DA LOA TP EP TND SP
Variable
19 PE3 .847* .028
20 PE1 .814* .027
21 PE4 .503* .193*
22 SI3 .828* .138**
*
23 SI5 .834 .067
24 SI2 .812* .196*
25 SI1 .807* .195*
*
26 SI4 .818 .051
27 TA4 .866* .030
28 TA2 .817* .102**
29 TA3 .815* -.004
*
30 TA5 .790 .089
31 TA1 .778* -.056
32 DA4 .874* .047
33 DA2 .841* .000
34 DA3 .821* .049
35 DA1 .803* -.068
36 LOA4 .818* .040
37 LOA3 .824* .025
38 LOA2 .826* -.073
39 LOA1 .815* .021
40 LOA5 .691* .108**
41 TP1 .762* .011
42 TP5 .734* -.101**
43 TP6 .743* -.067
44 TP3 .728* .015
45 TP4 .701* .083
46 TP2 .697* -.083
47 EP4 .843* .114*
*
48 EP8 .792 .021
49 EP2 .784* .110**
50 EP6 .744* .137**
*
51 EP7 .721 .064
52 EP5 .686* .160*
53 EP3 .664* .041
54 EP1 .494* -.037
55 TND4 .900* .050
56 TND2 .814* .008
57 TND6 .808* .146**
58 TND1 .755* .042
59 TND5 .671* .201*
60 TND3 .342* .030
61 SP4 .829* .190*
62 SP3 .732* .097**
63 SP2 .672* .236*
*
64 SP5 .681 .118**
*
65 SP1 .653 -.004
66 M2 .280a
67 M3 . 130a
68 M4 .070a
92
Chapter 4
93
Chapter 4
The comparison was computed to evaluate the marker variable effects on factor
correlation parameter estimates. Method-R (Figure 4.8) Model known as restricted
model was constructed to perform the test in a method described by Schaubroeck,
Ganster, and Fox (1992) and L.J. Williams et al. (1996). The Method-R model is
precisely same as the Method-U Model with one difference i.e. fixed correlation
parameters of the Method-R model that was obtained in the Baseline Model.
Therefore, the Method-R model contained 65 marker variable factor loadings linking
the marker variables to each indicator of the substantive variables. The correlation
between all the construct remained same correlation of baseline model is mentioned in
Table. The purpose of comparison of the Method-U and Method-R Models is to check
the effect of Marker Variable on correlation among constructs with a null hypothesis
of there is no Impact of Marker Variable on correlation among constructs. The result
of chi-square difference is 2.46 with 64 degrees of freedom. As the p value (1.000) is
more than 0.05, testing fail to reject the null hypothesis. Results of prior to this test
indicated that marker variable effects are significant and denotes significant effects in
the Method-U model. While the result of this comparison determined that the effects
of the marker variable did not significantly bias on factor correlation estimates. Thus,
there was not a significant difference between the Baseline Model factor correlations
and the Method-U factor correlations. Table 4.22 shows a summarized and
comparative view of all correlation among the constructs from initial to Method-R
Model.
94
Chapter 4
95
Chapter 4
Table 4.22: Summarized View of Correlation Among Constructs in All Models with
Sensitivity Analysis
96
Chapter 4
Stage II: Reliability decomposition. The purpose of the second phase of the study was
to decompose observed values of total reliability associated with the measurement of
the twelve latent constructs into substantive and method portions. Table reveals the
obtained reliability information for the, GSE, QSE, EE, PE, SI, TA, DA, LOA, TP,
SP, EP, TND latent construct and the un-decomposed marker latent construct. The
table depicts the overall reliability values from estimate of the Baseline Model and
values obtained with the decomposition using the Method-U estimates. The values
specify that the measurement of all of the latent constructs obtained adequate overall
97
Chapter 4
reliability ranged from 0.847 to 0.933. Referring Table 4.23, the decomposition value
point out that SP was most affected by method variance, with a method component
value of .026, whereas EP and TND have considerably low 0.010 and 0. 011
respectively. DA with 0.000, TA, LOA and TP with 0.001depecits no or least affected
by method variance. In terms of percentage of the total reliabilities, these method
components accounted for SP (3.03%), QSE (2.62%), and SI (2.39%) high values in
the present study. As compare to low values of DA (0.01%) and LOA (0.09%).
Reliability
Latent Total Substantive Method % Reliability Marker
Variable Reliability Reliability Reliability Variable
GSE 0.933 0.905 0.022 2.34
QSE 0.897 0.863 0.024 2.62
EE 0.867 0.849 0.005 0.53
PE 0.846 0.820 0.008 1.00
SI 0.917 0.880 0.022 2.39
TA 0.907 0.899 0.001 0.16
DA 0.902 0.900 0.000 0.01
LOA 0.896 0.890 0.001 0.09
TP 0.871 0.877 0.001 0.11
EP 0.898 0.881 0.010 1.11
TND 0.899 0.851 0.011 1.17
SP 0.847 0.806 0.026 3.03
Stage III: Sensitivity analysis- Table shows the correlations relating the latent
variables from the all examined models (Initial to Method-S Model). Initial and
Baseline model values which do not include even for method factor loadings are taken
for comparison with those obtained from the Method-C, Method-U Method-R and
Method-S (Figure 4.9) Models. This is evident from Table 4.23, introduction of the
manipulation of the method factor loadings brings only a small corresponding change
in the factor correlations. From the Baseline Model, all substantive factor correlations
were significant. This pattern is continued when method effects were introduced in
the Method-U Model, as the correlations among the latent variables remained
98
Chapter 4
significant, although their values were changed somewhat. When sizes of the method
factor loadings were restricted to values associated with the upper end of the .05
confidence interval in the Method-S model, the correlations between the latent
variables remained to be significant and relatively unchanged.
99
Chapter 4
The Bartlett‘s test was conducted to know the proportion of variance in the variables
and to tests the hypothesis that the correlation matrix is an identity matrix. The results
in Table 4.25 shows the P-value less than 0.05 which shows factors could be useful on
the data and confirming that each item shared some common variance with each item.
Keeping all the indicators, factor analysis was conducted. The extraction was carried
out by principal component analysis which was determined by Kaiser Criterion. The
percentage of variance are obtained from the extraction and rotation sums of squared
loadings. Table 4.24 shows the factor loadings for each variable. The factor loading
was obtained after the method of varimax rotation with Kaiser normalisation was
applied. The results as per extraction method – principal component analysis show
that all components have one component and rotated component matrix gives only
one component for all the factors i.e., GSE, QSE, EE, PE, TA, DA, SI, TP, EP, TND
and SP.
Factor loading shows the correlation coefficient for the variable and factor, thus factor
loading explains the variance by the variable on that factor. The results of factor
loading shows that the component of each factor have values more than 0.5, which
shows that high variance by the variable on the factor.
101
Chapter 4
Total p- Factor
Construct KMO
Variance (%) value Loading
General Self-Efficacy (GSE) 71.17 .930 .000
I am able to use HR Analytics without much effort. .865
(GSE3)
I can usually handle whatever comes my way through .863
HR Analytics. (GSE7)
When I am confronted with a problem, I can usually find
.860
several solutions through HR analytics. (GSE5)
I can solve most problems if I invest the necessary effort
related to HR Analytics. (GSE4) .844
HR Analytics is easy to use. (GSE1)
HR Analytics is convenient to use. (GSE2) .829
If I am in trouble, I can usually think of a solution. .823
(GSE6)
.820
Quantitative Self-Efficacy (QSE) 66 .851 .000
I enjoy working with mathematical and/or statistical .883
measures (QSE3)
I worry about my ability to solve mathematical and/or .867
statistical problems. (QSE2)
I find using mathematical and/or statistical
.850
measurements interesting. (QSE1)
Math and/or statistics are one of my favorite
subjects.(QSE5) .818
I find mathematical and/or statistical measures
challenging.(QSE4) .815
I get nervous when I use mathematics and/or statistics
(QSE6) .611
Effort Expectancy (EE) 71.31 .807 .001
It is easy for me to become skillful at using HR .880
Analytics. (EE3)
Learning to use HR Analytics is easy for me. (EE2) .865
It would be easy for me to become skillful at using HR .843
Analytics. (EE1)
My role related to HR Analytics is clear. (EE4)
.787
Performance Expectancy (PE) 67.46 .796 .001
Using HR Analytics enables me to accomplish tasks .880
more quickly. (PE2)
Using HR Analytics increases my job .877
performance.(PE3)
I would find the use of HR Analytics useful in my job.
.861
(PE1)
The use of HR Analytics is not very visible in my
organization. (PE4 .643
102
Chapter 4
Total p- Factor
Construct KMO
Variance (%) value Loading
helpful in the use of HR Analytics. (SI3) .875
Because of my use of HR Analytics, others in my
organization will see me as a more valuable .871
employee.(SI5)
People who are important to me think that I should use
.866
HR Analytics. (SI2)
People who influence my behavior think that I should
use HR Analytics. (SI1) .861
In general, the organization has supported the use of HR
Analytics. (SI4) .859
Tool Availability (TA) 72.89 .899 .000
I have had a great deal of opportunity to try various HR .890
Analytics applications. (TA4)
My company has invested heavily in HR Analytics tools. .857
(TA2)
Before deciding whether to use any HR Analytics
.856
applications, I am able to properly try them out.(TA3)
I know where I can go to satisfactorily try out various
uses of HR Analytics. (TA5) .838
I have a full array of HR Analytics tools available at
work. (TA1) .827
Data Availability 77.21 .831 .001
My company‘s database has an interface that is .899
compatible with other systems. (DA4)
My organization‘s HR system collects data from all HR .883
interactions. (DA2)
My organization uses the same system/platforms for all
.872
HR activities. (DA3)
My organization‘s database has all the data I need to use
HR Analytics software. (DA1) .861
103
Chapter 4
Total p- Factor
Construct KMO
Variance (%) value Loading
work by use of HR Analytics. (TP3) .728
HR Analytics gives me a focus on the results that I had
to achieve in my work. (TP2) .712
Empowerment Practices (EP) 58.33 .901 .000
HR Analytics helps my work unit by giving liberty to .856
employees in the conduct of their work. (EP4)
My work unit is considered as a strategic division in my .812
organization because of HR Analytics. (EP8)
Employees in my work unit have a lot of autonomy in .804
regard to project management provided through HR
Analytics. (EP2)
.799
Employees in my work unit are regularly consulted in
technological investments decisions because of HR
Analytics. (EP6)
Employees‘ empowerment in my work unit is highly .784
valued with exercising HR Analytics. (EP7)
Employees in my work unit are extensively involved in .755
the recruitment process because of HR Analytics. (EP5)
My supervisor manages our unit‘s budgets through HR .725
Analytics. (EP3)
In my organization, my work unit plays a strategic role .523
through HR Analytics. (EP1)
Training and Development (TND) 66.51 .838 .000
The training efficiency is determined through HR .887
Analytics. (TND4)
HR Analytics determines the financial support you can .840
receive from your employer for further education.
(TND6)
.814
HR Analytics plays a crucial role in conducting Training
programs by external consultants. (TND2)
.814
HR Analytics is used in my organization for determining
systematic techniques used in for specific training needs.
(TND1)
HR Analytics assesses the opportunities you have to .792
engage in training and education activities that are
beyond that needed in your job. (TND5) .739
HR Analytics provides effective communication which
is necessary for delivering quality training. (TND3)
Strategic Participation (SP) 62.19 .802 .000
HR Analytics enables the company to allow employees .855
to submit their suggestions for strategy formation. (SP4)
The circulation of information to all staff at all .796
administration levels are administered through HR
Analytics. (SP3)
.773
HR Analytics enables effective communication system
that supports the integration of strategic planning and
HR planning. (SP2)
HR Analytics enables the company to allow employees .759
to submit their suggestions for strategy implementation.
(SP5) .757
HR Analytics helps in synchronizing company‘s strategy
and HR planning. (SP1)
104
Chapter 4
A) Psychometric Checks
In the study, the questionnaire which was adapted has twelve constructs and sixty-five
items to measure the antecedents of adoption of HR analytics and its consequences on
the strategic workforce management. Hair Jr. et.al., (2010) provided the criterion
where the psychometric checks were assessed by examining the reliability and
validity of measurement model. This check is performed by Partial Least Square
Structural Equation Modelling (PLS-SEM) in SmartPLS software version 3.0. The
study is exploratory in nature and explains the key constructs, therefore, PLS-SEM
was used. The control variables which were formative in nature, were used as
composite. The combination of reflective and formative constructs made the
application of PLS-SEM appropriate as it set off on reflective constructs (Hair et.al.,
2017). Hair et.al., (2017) stated that the PLS-SEM is also called as soft-modelling
where, PLS-SEM can handle complex cause – effect smodels, can study small sample
sizes and highly flexible, thus, confirms the suitability as sixty five items to study
builds complex structure.
B) Measurement Model
After getting positive results from factor analysis, measurement model assessment is the
first step in Partial Least Square Structural Equation Modelling. Measurement model is
performed to compare the theoretical proposed model with the collected data from the
actual sample. We have 12 constructs in total to study, where 7 constructs are antecedents
of adoption of HR analytics, LOA, a construct itself and 4 constructs are its consequence
on strategic workforce management. The assessment for reflective constructs in
measurement model consists of four steps – indicator reliability, internal consistency,
convergent validity, and discriminant validity to step ahead for structural modelling.
Indicator reliability is the first step of measurement model. Alhassan et. al., (2020)
stated that the indicator reliability study expresses the extent, a variable or set of
variables are consistent to its measure. The outer loading – indicator was performed to
measure the indicator reliability. The indicator‘s outer loading shows that the
105
Chapter 4
Original T Statistics
P Values 2,50% 97,50%
Sample (O) (|O/STDEV|)
DA1 <- DA 0,265 22,558 0,000 0,241 0,287
DA2 <- DA 0,302 24,594 0,000 0,281 0,330
DA3 <- DA 0,270 26,360 0,000 0,250 0,291
DA4 <- DA 0,300 28,912 0,000 0,282 0,324
EE1 <- EE 0,290 22,697 0,000 0,268 0,318
EE2 <- EE 0,284 18,111 0,000 0,252 0,315
EE3 <- EE 0,299 23,756 0,000 0,276 0,326
EE4 <- EE 0,312 20,982 0,000 0,286 0,344
EP2 <- EP 0,168 12,724 0,000 0,142 0,195
EP3 <- EP 0,196 13,324 0,000 0,171 0,230
EP4 <- EP 0,158 11,931 0,000 0,128 0,181
EP5 <- EP 0,216 13,760 0,000 0,190 0,252
EP6 <- EP 0,151 10,493 0,000 0,123 0,180
EP7 <- EP 0,184 15,883 0,000 0,161 0,207
EP8 <- EP 0,186 15,484 0,000 0,163 0,211
GSE1 <- GSE 0,154 13,434 0,000 0,131 0,176
GSE2 <- GSE 0,190 14,741 0,000 0,168 0,220
GSE3 <- GSE 0,168 19,486 0,000 0,150 0,184
GSE4 <- GSE 0,190 19,140 0,000 0,172 0,211
GSE5 <- GSE 0,156 18,669 0,000 0,140 0,173
GSE6 <- GSE 0,166 16,401 0,000 0,146 0,186
GSE7 <- GSE 0,162 18,867 0,000 0,145 0,178
LOA1 <- LOA 0,244 27,437 0,000 0,228 0,263
LOA2 <- LOA 0,241 28,940 0,000 0,226 0,258
LOA3 <- LOA 0,247 24,922 0,000 0,228 0,266
LOA4 <- LOA 0,248 27,720 0,000 0,232 0,267
LOA5 <- LOA 0,207 20,888 0,000 0,187 0,226
PE1 <- PE 0,374 27,351 0,000 0,348 0,402
PE2 <- PE 0,370 26,987 0,000 0,344 0,398
PE3 <- PE 0,376 33,806 0,000 0,355 0,400
QSE1 <- QSE 0,242 10,640 0,000 0,196 0,286
QSE2 <- QSE 0,212 8,124 0,000 0,155 0,256
106
Chapter 4
Original T Statistics
P Values 2,50% 97,50%
Sample (O) (|O/STDEV|)
QSE3 <- QSE 0,307 11,036 0,000 0,266 0,377
QSE4 <- QSE 0,151 4,393 0,000 0,067 0,202
QSE5 <- QSE 0,247 7,941 0,000 0,195 0,319
SI1 <- SI 0,235 26,350 0,000 0,219 0,255
SI2 <- SI 0,239 31,321 0,000 0,226 0,256
SI3 <- SI 0,223 28,278 0,000 0,207 0,238
SI4 <- SI 0,214 25,197 0,000 0,198 0,231
SI5 <- SI 0,243 26,755 0,000 0,227 0,263
SP1 <- SP 0,192 7,643 0,000 0,136 0,236
SP2 <- SP 0,212 8,836 0,000 0,155 0,252
SP3 <- SP 0,336 12,021 0,000 0,292 0,402
SP4 <- SP 0,270 15,013 0,000 0,234 0,305
SP5 <- SP 0,251 11,731 0,000 0,210 0,296
TA1 <- TA 0,203 22,387 0,000 0,184 0,220
TA2 <- TA 0,246 27,188 0,000 0,231 0,266
TA3 <- TA 0,238 25,882 0,000 0,221 0,258
TA4 <- TA 0,249 28,459 0,000 0,234 0,269
TA5 <- TA 0,234 28,939 0,000 0,219 0,251
TND1 <- TND 0,221 17,613 0,000 0,197 0,246
TND2 <- TND 0,180 13,249 0,000 0,153 0,207
TND3 <- TND 0,229 13,946 0,000 0,202 0,268
TND4 <- TND 0,210 21,791 0,000 0,191 0,229
TND5 <- TND 0,212 13,873 0,000 0,181 0,242
TND6 <- TND 0,178 13,841 0,000 0,150 0,201
TP1 <- TP 0,243 10,779 0,000 0,204 0,291
TP2 <- TP 0,195 10,419 0,000 0,160 0,234
TP3 <- TP 0,209 12,521 0,000 0,178 0,244
TP4 <- TP 0,217 11,998 0,000 0,186 0,257
TP5 <- TP 0,193 10,750 0,000 0,155 0,226
TP6 <- TP 0,224 12,272 0,000 0,193 0,265
107
Chapter 4
(Nunnally, 1978). Nunnally and Bernstein (1994) stated that in exploratory research
the threshold limit of composite reliability is between 0.6 and 0.7 and for advance
research it is between 0.7 and 0.9. The recommendations for rho A as a measure of
internal consistency with threshold limit of 0.7 is also acceptable (Dijkstra and
Henseler, 2015). The Table 4.27 shows the results of Cronbach alpha to be more than
0.7 as it is a exploratory research, thus the values are acceptable. The rho A values are
also more than 0.7, which shows the collected data is also acceptable as per internal
consistency analysis.
It is the third step in the measurement model study which is evaluated after indicator
reliability study and internal consistency analysis. Hair et.al., (2016) explains that the
convergent analysis is the extent to which a measure positively correlates with the any
other replacement or alternative measure of the same construct. The convergent
validity analysis study focuses on the average variance extracted (AVE) which is the
grand mean of the squared loadings of the indicators from the construct (Hair et.al,
2016). AVE value more than 0.5 is acceptable (Hair et.al., 2019) for approving
108
Chapter 4
convergent validity. The results in Table 4.27 shows the values for AVE for all
constructs are more than 0.5, which shows that more than 50% of variance in the
indicator is explained by its construct. The assure the results further composite
reliability was also performed stating the acceptable values to be more than 0.7 (Hair
et.al., 2013). The Table 4.27 shows all the values for each construct is above 0.7.
Thus, the data satisfies the convergent validity.
Discriminant Validity
109
Chapter 4
Cross loading is another approach to test the discriminant validity of the construct.
This approach is not rigorous but recommended to ensure the collected data surpasses
all criteria. This technique studies the cross loading among each construct. The
indicator for the study is that it should exhibits at least 0.7 loading towards its own
construct and should not contain cross loading towards any other construct which is a
part of measurement model (Hair et.al., 2014). The results in Table 4.29 shows that
the construct value towards its own construct is more than 0.7 and towards other its
less. Therefore, as per the study, the results establish discriminant validity of
constructs in the measurement model.
110
Chapter 4
111
Chapter 4
112
Chapter 4
113
Chapter 4
C) Structural model
Collinearity Assessment
Collinearity is the first step of assessment for structural modelling. This assessment is
made to assess the collinearity between the predictor constructs established in the
structural model, that is, by examining the correlation between the exogenous
constructs. To execute the assessment, variance inflation factor (VIF) was calculated
which is the reciprocal of tolerance. The threshold limit is of 5, that is, all the values
less than 5 confirms the absence of collinearity concerns in the model (Pallant, 2010;
Hair et.al., 2020). Table 4.31 shows the VIF inner model and the results are less than
5 which represents no presence of collinearity
114
Chapter 4
The VIF for the outer model threshold value should be less than 5 (Hair et.al., 2020)
and the results in Table 4.32 shows that all the of the constructs are less than 5. Thus,
we can assure there is no collinearity issues in the constructs of the structural model.
115
Chapter 4
After ensuring no collinearity in the predictor constructs of the structural model, the
next step is to examine the significance of path coefficients, that is, the hypothesis
relationships. The bootstrapping process was conducted in SmartPLS to assess the
significance of path coefficients. Alhassan et.al., (2020) explained that the
bootstrapping is a non-parametric resampling procedure which focuses on examining
the variability of the sample data other than the parametric assumptions and drawing
two-tailed distribution to test the precision of the estimates. T-statistics are drawn for
direct and indirect effects under the bootstrapping process (Hair et.al., 2016).
After examining the significance of path coefficient, the relevance of path coefficient
was explored. Hair et.al., (2016) has recommended the examination of relevance of
significant relationships with the established constructs. In the study Hair et.al. (2016)
has stated that the significant relationship in the model is not largely influenced by its
size. The PLS-SEM is used with the intention to examine the significant and relevant
relationships between the constructs both endogenous and exogenous. The Figure
4.10 shows the path relationships between the constructs and its relative significance.
The study executes the bootstrapping assessment at 95% confidence interval and 5%
level of significance for two-tailed T-statistics at the value of 1.96 (Hair et.al., 2011).
In the way to establish significance of the path relationship one must ensure that the
confidence interval results of bootstrapping should not have zero as the confidence
interval (Hair et.al., 2016). Referring to the results in the Table 4.33 of bootstrapping
with t-values for the direct effect shows the acceptable results (p-value < 0.05) for the
following hypothesis:
116
Chapter 4
117
Chapter 4
All are hypothesis are accepted since p-value is less than 0.05 and all these factors
verifies the association with adoption of HR adoption. In all the accepted factors QSE
has negative value which shows QSE is influencing adoption of HR analytics
negatively.
The p-value for the association of data availability (DA) on adoption of HR analytics
has negative value (-0.11) and p-value is more than 0.05 and hence H5: Data
Availability is significantly associated to adoption of HR Analytics, is rejected.
GSE -> LOA 0,104 0,105 0,043 2,436 0,015 Accepted (H1)
QSE -> LOA -0,124 -0,122 0,030 4,108 0,000 Accepted (H2)
LOA -> TND 0,450 0,453 0,048 9,342 0,000 Accepted (H8(c))
118
Chapter 4
119
Chapter 4
R square assessment
After assessing the path coefficient and its relevance with all the constructs of the
structural model, the assessment for fitness of good of the model was made. The index
for goodness of fit was calculated which defines the operational solution to the
problem (Tenenhaus et.al., 2005). The index was popularly executed to examine the
global validation of the model using PLS-SEM. Therefore, assessment was made to
ascertain the fitness of the proposed model by identifying Standard Root Mean
Squared Residual (SRMR) (Henseler et.al., 2015). SRMR measures the unweighted
least squares discrepancy, geodesic discrepancy, Chi-square and NIF index. Bailey
et.al., (2017) and Pappas (2017) suggested to calculate SRMR to check the goodness
of fit for the model. If the threshold value is less than 0.08 than the results are
accepted to be good fit (Henseler et al., 2016). Table 4.34 shows the SRMR values
ascertained on the saturated model and estimated model. The results show the values
less than 0.08 indicating goodness of fit of the proposed model.
After assessing the SRMR values, the coefficient of determination (R2) is measured
which represents the combined effect of exogenous latent variables on the
endogenous latent variables (Hair et.al., 2014). R2 values above 0.2 is considered to
be good as Shiva et.al., (2020) stated it to be high explanatory power referring to
behaviour science studies. The results in Table 4.35 shows a high explanatory power
for LOA, EP, TND and TP as values are more than 0.2. The results for SP are lower
than 0.2, hence shows low explanatory power.
Table 4.35: R square
120
Chapter 4
To study the strength of independent variables relationship, F2 test was executed. This
test studied the contribution of each exogenous independent variable on endogenous
dependent variable. Table 4.36 represents the F2 results which explains the
endogenous variables (LOA) that are projection of exogenous variables in the
structural model. The values turning 0.02 shows a small effect, 0.15 shows a medium
effect and 0.35 shows a large effect. Referring to the results in the Table 4.36, EP
(0.270), TND (0.253) and TP (0.255) shows effect between medium to large and SP
(0.172) is close to medium effect.
D) Mediation Analysis
121
Chapter 4
latent construct (strategic workforce management studied through TP, SP, TND and
EP). To execute the study, the bootstrapping process of PLS-SEM was used as it
makes no assumptions with the distribution of variables (Hair et.al., 2016). Table 4.37
shows the specific indirect effect results. On analysing the results, exogenous latent
constructs show acceptable values i.e., p-value < 0.05 for following hypothesis:
H9: Adoption of HRA has a mediating impact between general self-efficacy and
strategic workforce management: To understand the mediation impact of adoption of
HR analytics on general self-efficacy and strategic workforce management, all the
segments of strategic workforce management were studied - H9(a) GSE-LOA-TP
(4.7%), H9(b) GSE-LOA-SP (4%), H9(c) GSE-LOA-TND (4.7%) and H9(d) GSE-
LOA-EP (4.8%).
H10: Adoption of HRA has a mediating impact between quantitative self-efficacy and
strategic workforce management: To understand the mediation impact of adoption of
HR analytics on quantitative self-efficacy and strategic workforce management, all
the segments of strategic workforce management were studied - H10(a) QSE-LOA-
TP (-5.6%), H10(b) QSE-LOA-SP (-4.8%), H10(c) QSE-LOA-TND (-5.6%) and
H10(d) QSE-LOA-EP (-5.7%).
H11: Adoption of HRA has a mediating impact between effort expectancy and
strategic workforce management: To understand the mediation impact of adoption of
HR analytics on effort expectancy and strategic workforce management, all the
segments of strategic workforce management were studied - H11(a) EE-LOA-TP
(7.6%), H11(b) EE-LOA-SP (6.5%), H11(c) EE-LOA-TND (7.6%) and H11(d) EE-
LOA-EP (7.8%).
H12: Adoption of HRA has a mediating impact between performance expectancy and
strategic workforce management: To understand the mediation impact of adoption of
HR analytics on performance expectancy and strategic workforce management, all the
segments of strategic workforce management were studied - H12(a) PE-LOA-TP
(7.9%), H12(b) PE-LOA-SP (6.8%), H12(c) PE-LOA-TND (7.9%) and H12(d) PE-
LOA-EP (8.1%).
122
Chapter 4
H14: Adoption of HRA has a mediating impact between tool availability and strategic
workforce management: To understand the mediation impact of adoption of HR
analytics on tool availability and strategic workforce management, all the segments of
strategic workforce management were studied - H14(a) TA-LOA-TP (11.7%), H14(b)
TA-LOA-SP (10%), H14(c) TA-LOA-TND (11.7%) and H14(d) TA-LOA-EP (12%).
H15: Adoption of HRA has a mediating impact between social influence and strategic
workforce management: To understand the mediation impact of adoption of HR
analytics on social influence and strategic workforce management, all the segments of
strategic workforce management were studied - H15(a) SI-LOA-TP (10.1%), H15(b)
SI-LOA-SP (8.6%), H15(c) SI-LOA-TND (10.1%) and H15(d) SI-LOA-EP(10.3%).
Since the p-value for all these hypotheses is less than 0.05 therefore all these
hypotheses are accepted, i.e., adoption of HR analytics has mediating effect between
GSE and strategic workforce management, QSE and strategic workforce
management, EE and strategic workforce management, PE and strategic workforce
management, TA and strategic workforce management and SI and strategic workforce
management.
The p-value is more than 0.05 for adoption of HRA mediating impact between data
availability and strategic workforce management (H13), hence H13 is rejected.
QSE -> LOA -> TP -0,056 -0,055 0,015 3,851 0,000 -0,088 -0,030 Accepted
QSE -> LOA -> SP -0,048 -0,047 0,011 4,225 0,000 -0,072 -0,028 (H10):
HRA has
QSE -> LOA -> TND -0,056 -0,055 0,014 3,943 0,000 -0,086 -0,031 mediation
QSE -> LOA -> EP -0,057 -0,057 0,015 3,859 0,000 -0,089 -0,031 effect
EE -> LOA -> TP 0,076 0,077 0,022 3,484 0,000 0,037 0,122 Accepted
EE -> LOA -> SP 0,065 0,066 0,020 3,319 0,001 0,032 0,107 (H11):
123
Chapter 4
DA -> LOA -> TP -0,005 -0,004 0,020 0,251 0,802 -0,042 0,035 Rejected
(H13):
DA -> LOA -> SP -0,004 -0,004 0,017 0,249 0,804 -0,037 0,030
HRA has
DA -> LOA -> TND -0,005 -0,005 0,020 0,251 0,802 -0,043 0,034 no
mediation
DA -> LOA -> EP -0,005 -0,005 0,020 0,248 0,804 -0,045 0,035 effect
TA -> LOA -> TP 0,117 0,118 0,030 3,885 0,000 0,065 0,182 Accepted
TA -> LOA -> SP 0,100 0,100 0,025 4,056 0,000 0,057 0,152 (H14):
HRA has
TA -> LOA -> TND 0,117 0,117 0,028 4,103 0,000 0,066 0,177 mediation
TA -> LOA -> EP 0,120 0,120 0,030 4,019 0,000 0,069 0,185 effect
SI -> LOA -> TP 0,101 0,102 0,028 3,573 0,000 0,051 0,162 Accepted
SI -> LOA -> SP 0,086 0,087 0,025 3,479 0,001 0,043 0,138 (H15):
HRA has
SI -> LOA -> TND 0,101 0,102 0,029 3,490 0,000 0,050 0,162 mediation
SI -> LOA -> EP 0,103 0,104 0,028 3,744 0,000 0,054 0,163 effect
124
Chapter 4
have revealed that type of organisation structure has an impact in the adoption of HR
analytics process, together if the training is provided before actual implementation of
analytical tools, have supported the use of action. Keeping the studies in mind, type of
organisation – Large organisation and MSME (micro, small and medium enterprises) and
training of HR analytics are studied as a moderator for adoption of HR analytics towards
strategic workforce management.
The study of moderating variables was introduced in the theoretical framework (Adams,
Nelson and Todd, 1992). The concept of moderation analysis was initiated to predict the
validity of the model and confirms the impact of moderators for adoption of HR analytics
(Agarwal and Prasad, 1998; Venkatesh and Bala, 2008). Chin, Marcolin and Newsted
(2003) conducted a comprehensive study knows as Partial Least Square latent variable
modelling. This approach measues the interaction effects of the moderators on the
structural relationships by using Monte Carlo Simulation study which is a improved
version of previous studies. Henseler and Chin (2010) has provided with their findings on
conducting the potential moderator analysis in PLS-SEM. They also introduced the
product indicator approach and explained the shortcomings of LISREL approach about
analytical errors and incomplete solutions. Henseler and Fassott (2010) has introduced
two stage approach where the moderator analysis and independent variables are
measured. So far two stage approach is considered one of the best technique to study the
moderator analysis.
Work satisfaction, job commitment, strategic participation are more driven by the nature
and type of organisation (Saha and Kumar, 2017). Grossi et.al., (2015) has stated in their
research that the employee‘s performance and acceptance to change in largely dependent
on the organisational culture, type and size as it is driven more with the investment and
efforts required. Kalsi and Kiran (2013) suggested that the quality, coordination, success
and thought process of the employees are formed by the managerial practices an
organisation defines. MSME are presently growing in number in India and many received
support from Indian government under ―Make in India‖ umbrella. These organisations
might start with small, but their investment and right selection of talent have helped them
to come a long way despite the tragic COVID-19 (Banu and Suresh, 2020). The
125
Chapter 4
foundation and culture are ruling to integrate the MSMEs with IT and ITES (Indrakumar,
2020). Adoption of HR analytics was not a cake walk for MSMEs rather it was well
defined strategic move (Saxena et.al., 2020). Large organisations have capital to
investment and strong experiences and connections which helps them to research well
before introducing change. Adoption of analytics was a planned task for them. Strategic
planning and understanding the pros and cons were part of it (Performance Report, 2015).
The culture, support, quality and leadership are driving the change in large enterprises
(KPMG, 2012). Any new approach or technology has to undergo a series of action of test
like usability, adaptability, sustainability and future scope, an organisation performs these
analysis before introducing any change in the organisation. Large organisations being big
in size and extensive in work pressures do execute this analysis to assure their move will
bring results to them (Potter, 2003). Understanding the type of organisation can impact
the success or failure of any process, project or technology adoption and based on these
findings, type of organisation was studied as a in adoption of HR analytics towards
strategic workforce management.
The moderation effects of type of organisation were estimated through the multi-Group
analysis (MGA) techniques. To study the type of organisation as moderator, the sample
was divided into two groups: large organisation and micro, small and medium
organisations. The model was then estimated for the two groups and verified the
significance for both the groups separately for the structural model. The outcomes of the
MGA were studied through Henseler-MGA non-parametric analysis. This technique
studies the differences between the path coefficients of the two groups and this technique
is used to study the group differences in PLS-SEM (Hair et.al., 2017). After that,
Parametric test (Two-tailed test) and Welch-Satterthwaite test were applied to ascertain
where there is any significant difference between adoption of HR analytics across large
organisation and MSME organisation. In addition to this boot strapping was also
performed to analysis the relationships of adoption of HR analytics.
PLS-MGA results in Table 4.38 revealed that EP, SP and TND have no impact of
moderator as p-value is not less than 0.05, which signifies that there is no significant
difference between large organisation and MSME organisation in the above stated
structural relationship, rejecting the moderating effect for EP, SP and TND. TP have
126
Chapter 4
shown a significance with p-value of 0.016, that means there is a significant difference in
large organisation and MSME organisation with task performance and structural
relationships, hence, accepting hypothesis.
The results are further analysed with parametric test (t=2.05; p<0.05). The results in Table
4.39 shows that TP is significantly different at 5% level of difference, defining the
significant difference between large and MSME organisation with TP and structural
relationships. The p-value is more than 0.05 for EP, SP and TND, thus rejecting the
hypothesis.
Executing the Welch Satterthwait Test (t=2.05, p-value<0.05) in Table 4.40 revealed that
except TP (p-value = 0.017), all constructs were insignificant (p-value >0.05) to adoption
of HR analytics towards strategic workforce management, that is, there is no significant
difference between large and MSME organisation with the structural relationships,
rejecting hypothesis. All the three test reveals that for TP there is significant difference
between large and MSME organisation with the structural relationships.
127
Chapter 4
The group level results from PLS MGA, Parametric test and Welch Satterthwait test
shows insignificant effect of type of organisation on EP, SP and TND. The results
show significance with TP, that is, type of organisation has an impact as a moderator
for TP. Referring to the boot strapping study, which is made at individual level in
Table 4.41, shows the path coefficient difference (Large organisation – MSME
organisation) as all positive values for all the four segments of strategic workforce
management, which means, large organisation has more impact than the MSME
organisations. All the t-values are positive, and p-values is less than 0.05, which
shows level of adoption (LOA) impact is more on strategic workforce management.
Individually all are significant, and type of organisation has an impact as a moderator.
Figure 4.11 and 4.12 shows the path coefficient for the moderators – type of
organisation – large organisation and MSME organisation respectively and its
relevant significance. Since we studied strategic workforce management through task
performance, strategic participation, training and development and empowerment
practices, the results found to be significant for all the factors. Therefore, here we
accept the hypothesis H16 as type of organisation has a moderating impact on
adoption of HR Analytics and strategic workforce management.
128
Chapter 4
LOA -> TND 0,461 0,441 0,059 0,467 0,448 0,066 0,061 7,024 7,229 0,000 0,000 0,363 0,598 0,284 0,559
LOA -> EP 0,491 0,432 0,165 0,496 0,440 0,060 0,069 8,142 6,238 0,000 0,000 0,335 0,566 0,163 0,418
LOA -> SP 0,456 0,292 0,02 0,463 0,302 0,060 0,066 7,653 4,389 0,000 0,000 0,325 0,583 0,308 0,553
LOA -> TP 0,551 0,285 0,266 0,554 0,301 0,061 0,092 9,018 3,095 0,000 0,002 0,417 0,656 0,132 0,453
129
Chapter 4
130
Chapter 4
Knowledge sharing and knowledge management is a top priority in the IT and ITES
organisation where learning of new techniques and technology never ends (Noor and
Nasirun, 2015). Many organisations irrespective of large or small or medium, have
started to count on competitive advantage through knowledge management (Pee &
Kankanhali, 2015). These organisations incorporate the effective of knowledge
management if they are involved with the services (Noor and Nasirun, 2015). The
concept of knowledge sharing gets started as soon as an employee joins the
organisation in the form of training. Training is considered important not only in the
joining process, but also when there is a new initiation of project, exist of employee
leads to knowledge transfer and training and when new technology is introduced or
about to introduce (McAdam & Reid, 2000). In case of adoption of HR analytics by
the HRs in IT organisations, have faced the situation where some non-tech resources
got opened to technical work processes. To bring this new work scenario a routine
and quality driven work its important to have knowledge sharing and training. The
studies devise that training can have an impact on adoption of HR analytics towards
strategic workforce management, therefore, investigating HR analytics training as a
moderator in the structural model.
The moderation effects of HR analytics training were estimated through the multi-
Group analysis (MGA) techniques. To study the HR analytics training as a moderator,
the sample was divided into two groups: with HR analytics training and without HR
analytics training. The model was then estimated for the two groups and verified the
significance for both the groups separately for the structural model. The outcomes of
the MGA were studied through Henseler-MGA non-parametric analysis. This
technique studies the differences between the path coefficients of the two groups and
this technique is used to study the group differences in PLS-SEM (Hair et.al., 2017).
After that, Parametric test (Two-tailed test) and Welch-Satterthwaite test were applied
to ascertain where there is any significant difference between adoption of HR
analytics across with HR analytics training and without training. In addition to this
boot strapping was also performed to analyse the relationships of adoption of HR
analytics on strategic workforce management with training and without training.
131
Chapter 4
Table 4.42: PLS-MGA and Parametric test – With and without Training
Path
Path
Path Coefficients- Path p-
Path Coefficients STDEV STDEV t-Value t-Value p-Value 2.5% 97.5% 2.5% 97.5%
Coefficients diff (With Coefficients Value
Coefficients Mean (With (Without (With (Without (Without (With (With (Without (Without
(Without Training - Mean (With
(With Trg) (Without Trg) Trg) Trg) Trg) Trg) Training) Training) Training) Training)
Trg) Without (With Trg) Trg)
Trg)
Training)
LOA -> EP 0,423 0,539 -0,116 0,426 0,543 0,069 0,051 6,095 10,524 0,000 0,000 0,274 0,55 0,433 0,634
LOA -> SP 0,453 0,358 0,095 0,463 0,363 0,058 0,063 7,802 5,701 0,000 0,000 0,328 0,557 0,23 0,474
LOA -> TND 0,451 0,481 -0,031 0,456 0,484 0,078 0,053 5,794 9,018 0,000 0,000 0,284 0,592 0,366 0,576
LOA -> TP 0,559 0,392 0,167 0,567 0,396 0,061 0,076 9,164 5,144 0,000 0,000 0,417 0,66 0,233 0,533
132
Chapter 4
PLS-MGA results in Table 4.42 revealed that there is no impact on EP, SP, TND and
TP of moderator (HR analytics training) as p-value is not less than 0.05. Hence, there
is no significant relation of training (with and without) with the structural
relationships. The results are further analysed with parametric test (t=2.05; p<0.05).
The results in Table 4.42 shows that all the constructs are insignificant as t is less than
2.05 and p is more than 0.05. According to the study the adoption of HR analytics
does not differ in its application towards strategic workforce management. Executing
the Welch Satterthwait Test (t=2.05, p-value<0.05) in Table 4.42 revealed all
constructs were insignificant to adoption of HR analytics towards strategic workforce
management.
The group level results from PLS MGA, Parametric test and Welch Satterthwait test
shows insignificant effect of HR analytics training on EP, SP, TND and TP. Referring
to the boot strapping study which is made at individual level in Table 4.43, shows the
path coefficient difference (With training – Without training) as positive value for
LOA – SP and LOA – TP, which means that the HR analytics training has more
impact on SP and TP than without training. The path coefficient difference (With
training – Without training) is negative for LOA – EP and LOA – TND, which mean
without HR analytics training has more impact on EP and TND than with training. All
the t-values are positive and p-values are less than 0.05, which is, training has impact
on structural relations with LOA. Individually all are significant and HR analytics
training has an impact as a moderator. Figure 4.13 and 4.14 shows the path coefficient
for the moderators – With and without HR analytics training respectively and its
relevant significance to strategic workforce management which is studied through
task performance, strategic participation, training and development and empowerment
practices. The results found to be significant for all the factors of strategic workforce
management, i.e., p-value < 0.05. Therefore, here we accept H17 which is HR
Analytics training has a moderating impact on adoption of HR analytics and strategic
workforce management.
133
Chapter 4
134
Chapter 4
PLS predict
Ensuring the data meets all the quality standards, that shows data reliability and
validity, builds a base for PLS predict. PLS predict is different from the concepts of
training and holdout samples that are used to estimate and evaluates the model on
prediction grounds. A training sample consists of a section of overall data set which is
used to evaluate the model‘s parameters, that are, path study, loading and indicator
weights. The unused part of training set is known as holdout sample set (Hair et.al,
2018). In order to predict the value, independent construct‘s indicators of cases of
holdout sample are used on the selected dependent construct‘s indicators. This is
applied to the training set to get the prediction (Shmueli et.al., 2016).
Shmueli et.al., (2016) has provided with the knowledge and steps for calculating
predictive performance of the model and stated this technique to be better than
blindfolding. As per the technique, a cross-validation between the training set and
sample set are made and PLS predict is applied for cross-validation. When we apply
the PLS predict the samples are splitted randomly into folds called as subsets. K=10
was considered for getting the sub-groups (Shmueli et.al., 2016; Singh and Bala,
2020). The results in the form of root mean squared error (RMSE), the absolute mean
error and Q2 were analysed. RMSE will hold value and analysed only in the case of
normalised data (Witten et.al., 2016). In case of non-normalised data, MAE is
analysed. MAE studies the average magnitude of the errors in the prediction-set
irrespective of the directions whether over or under. It studies the average absolute
difference between the actual observation and the prediction values, considering the
individual differences have equal weights. The MAPE analysis is an intuitive study
showing predictive errors in form of percentage metric. Thus, it is considered
independent from the manifest variable scaling (Shmueli et.al., 2016). MAPE has
number of limitations as it is not defined with zero value for manifest variables. This
is considered as a drawback, as it is not defined at the composite level and causes
potential singularity problems (Tofallis, 2015).
135
Chapter 4
Sharma et.al., (2019a) in their studies stated the efficacy of the three statistics in PLS-
Sem model were tested and postulated that MAE and RMSE reliably emerged as a
best model balance fit with the predictive powers. On the other, MAPE turns to be
emerged with incorrect specific models. The naïve benchmarks shows that MAE,
MAPE and RMSE are scaled to have higher predictive power with the smaller values
as their absolute levels are difficult to study as their values are dependent upon the
scaling made on dependent constructs resulting in arbitrary threshold. Shmueli et.al.,
(2016) stated the application of simple indicator level average is a naïve benchmark
when considering the training sample. The study is similar to the blindfolding based
Q2 statistic in PLS-SEM (Stone, 1974). This Q2 predict is considered as naïve
benchmark. This makes use of mean values of the variables of training set as
prediction variables from the holdout set. A positive result in Q2 predict ascertains
that the prediction error in the path model is smaller than the naïve benchmark‘s
prediction error.
If the Q2 predict value is zero or less, that indicates that the predictive power of PLS-
SEM analysis could not perform better than the naïve benchmark. With Q2 predict >
0, RMSE for symmetric and MAE for non-symmetric data values are compared with
the naïve LM benchmarks. If the result PLS-SEM < LM , in terms of RMSE or MAE
for none of the indicator, then it is considered as model lacks predictive power. If
PLS-SEM < LM for the minority of the dependent constructs, ascertains that the
model has low predictive power. If PLS-SEM < LM for a majority of indicators, then
it shows medium prediction power. If PLS-SEM < LM for all indicators, then the
model is said to be high predictive power.
136
Chapter 4
PLS LM
MAE Q²_predict MAE Q²_predict
EP5 0,540 0,159 0,528 0,271
EP6 0,563 0,123 0,537 0,245
EP4 0,478 0,162 0,471 0,337
EP2 0,562 0,159 0,546 0,309
EP8 0,508 0,171 0,463 0,300
EP7 0,537 0,132 0,511 0,263
EP3 0,546 0,134 0,512 0,314
SP2 0,544 0,119 0,519 0,195
SP3 0,487 0,149 0,452 0,284
SP5 0,481 0,091 0,455 0,333
SP4 0,613 0,180 0,486 0,388
SP1 0,374 0,074 0,409 0,207
TND2 0,531 0,157 0,506 0,318
TND4 0,540 0,193 0,499 0,371
TND6 0,548 0,156 0,525 0,317
TND1 0,392 0,182 0,429 0,349
TND3 0,481 0,135 0,466 0,180
TND5 0,474 0,156 0,460 0,241
TP5 0,557 0,115 0,548 0,092
TP1 0,413 0,172 0,432 0,248
TP4 0,538 0,152 0,538 0,162
TP3 0,503 0,143 0,514 0,152
TP6 0,492 0,149 0,469 0,214
TP2 0,623 0,143 0,593 0,167
4.5 Summary
This chapter outlined the statistical results on the collected data based on the
objectives of the study. It was found from the data that general self-efficacy,
quantitative self-efficacy, effort expectancy, performance expectancy, tool availability
and social influence has a significant impact on adoption of HR analytics and later
influence the strategic workforce management which was studied through task
137
Chapter 4
138