CHAPTER 4

RESULTS AND
DISCUSSION

69
4.0 INTRODUCTION

This chapter reports the findings of this research obtained with the help of
previously discussed data analysis techniques applied on the data collected
from a sample of 392 HR professionals through online questionnaire. This
chapter takes the thesis forward by providing a justification of the results with
reference to the previous researches.

The penultimate sections in the chapter present the results in the following
order: Exploratory Factor Analysis, Confirmatory Factor Analysis,
Discriminant Analysis and one-way ANOVA. Further, the results are
discussed on the basis of the e-HRM dimensions one at a time.

4.1 EXPLORATORY FACTOR ANALYSIS (EFA)

Exploratory factor analysis was used to identify the seven determinants of e-

HRM. Initially, 40 items were taken, based on the literature. Due to the
combination of low inter-item correlations and low communalities (<0.40),
two items were removed. The Kaiser-Meyer-Olkin measure of sampling
adequacy was found to be 0.904, which suggests that the responses given with
the sample were adequate (Table 4.1) The same table shows that the Bartlett‟s
Test Of Sphericity is significant (p = 0.000). This suggests that correlation
matrix is not an identity matrix.

Table 4.1 KMO and Bartlett's Test

Kaiser-Meyer-Olkin Test of Adequacy

Kaiser-Meyer-Olkin Measure
.904
of Sampling Adequacy.

Bartlett's Test of Sphericity

7828.180
Approx. Chi-Square

Df (Degree of Freedom) 780

Sig. (Significance)
.000

70
None of the items had cross loading, but 5 items were not loaded on any of the
factors, hence a total of 33 items were finally considered for the study which
loaded on seven factors (Table 4.2).

Table 4.2. Rotated Component Matrix

Variables Component

1 2 3 4 5 6 7
VAR00001 -.568
VAR00002 .653
VAR00003 .701
VAR00004 -.598
VAR00005 .694
VAR00006
VAR00007 .725
VAR00008
VAR00009 .616
VAR00010
VAR00011
VAR00012 .699
VAR00013 .605
VAR00014 .832
VAR00015 .525
VAR00016 .754
VAR00017 .584
VAR00018 .626
VAR00019 .520
VAR00020 .604
VAR00021 .770
VAR00022 .715

71
 VAR00023 .763
VAR00024 .716
VAR00025 .828
VAR00026 .701
VAR00027 .742
VAR00028 .817
VAR00029 .721
VAR00030 .790
VAR00031 .678
VAR00032 .818
VAR00033 .800
VAR00034 .844
VAR00035 .782
VAR00036 .689
VAR00037 .594
VAR00038

The variance explained by each variable of e-HRM is given in the following

table (Table 4.3). This table also shows that the seven factor solution
explained 63% of the cumulative variance.

Table 4.3. Total Variance Explained

Factors Variance Explained (%)

( Cumulative % 63.329)
HRR 14.738
OC 10.489
EC 7.915
SC 7.699
UC 7.197
PEOU 6.211
PU 4.858

72
4.2 CONFIRMATORY FACTOR ANALYSIS (CFA)

The measurement of the validity of the variables is important before analyzing

their relationships i.e. to check if the variables measure appropriately what
they are intended to measure. For doing so, this study investigated the loadings
of the items for unidimensionality, composite reliability (CR) which gives the
internal consistency value of latent constructs (Henseler et al., 2009) and
average variance extracted (AVE) to assess the convergent validity of the
model as follows (Fornell and Larcker, 1981):

 CR = Sum of SRW Estimate Squared / (Sum of SRW Estimate

Squared + Sum of Error Variance)
 AVE = Sum of SRW Estimate Squared / Number of Variable Items

All the 33 items have confirmed a loading value above .50 which is the
threshold value (Hair et al., 2010). For each latent construct, CR > 0.6 and
AVE > 0.5 (Table 4.4) i.e. satisfying the threshold values (Bagozzi, 2011),
except UC and PU for which AVE <0.5 but near 0.5. Hence, the results can be
generalized for all but these two characteristics. Thus, the measurement model
was found to be reliable and holding convergent validity.

Table 4.4. Construct Reliability and Average Variance Extracted (e-HRM)

S. No. Factors C.R. A.V.E.

1 Role of HR (HRR) 0.972 0.573

2 Organizational Characteristics (OC) 0.958 0.544

3 Environmental Characteristics (EC) 0.976 0.557

4 System Characteristics (SC) 0.906 0.569

5 User Characteristics (UC) 0.897 0.400

6 Perceived Ease of Use (PEOU) 0.651 0.530

7 Perceived Usefulness (PU) 0.665 0.380

73
Figure 4.1 shows the path diagram for e-HRM model (by AMOS).

Note: RoleHR=HRR; OrgChar=OC; EnvChar=EC; SysChar=SC; UChar=UC

Figure 4.1: e-HRM Model: Path Diagram

74
The absolute fit statistics showed that the value of chi-square is 903.644 with
degrees of freedom, Df = 474 which was significant (p = 0.000). Model fit
statistics are shown in Table 4.5 which show that the proposed model was fit
for the obtained data. The discriminant validity of the model is ascertained as
the correlation among the constructs was below the threshold limit i.e. less
than 0.85 (Kenny, 2016). The 33 observed variables loaded to their
corresponding construct and also, no cross-loadings of variables was observed.
The factor loading of each of the 33 variables was more than 0.5.The factors
which had factor loading less than 0.5 were removed. Hence,
unidimensionality of the model was established (Nazim and Ahmad, 2013).

Table 4.5. Model Fit Statistics (e-HRM)

Fit Index Recommended Observed Value Overall Model

Value Fit

CMIN/degrees of < = 5.0 1.906 Good fit

freedom

GFI >= 0.90 0.878 Acceptable

AGFI >= 0.80 0.855 Good fit

NFI >= 0.90 0.833 Acceptable

CFI >= 0.90 0.912 Good fit

RMSEA <=0.08 0.048 Good fit

The findings showed that all the determinants are positively and significantly
related to e-HRM except two variables (standardization and interactive e-
HRM) of system characteristics which are significant but negatively related to
e-HRM and user characteristics which do not have higher significance. This
may be due to the samples taken from software industry where users are
technically sound and their experience and skills do not allow them to use a
system which is lacking high technology. PU is surprisingly showing less
significance but both PEOU and PU have a positive impact on e-HRM. The
results are consistent with early and recent studies which suggest that
75
organization, HR and the quality of system: all play a protagonist role towards
e-HRM implementation and adoption (Tansley et al., 2001; Ruël et al, 2004,
Lin, 2011; Bondarouk et al., 2017). These results are further discussed in
detail with respect to the previous researchers' views in 'Discussion' sub-
section.

4.3 MULTIPLE DISCRIMINANT ANALYSIS (MDA)

(i) e-HRM and engagement levels:

The number of discriminant functions is equal to the number of discriminating

variables, if number of groups is more than variables else it is 1 less than the
number of levels in the grouping variable. In this study, employee engagement
has three levels and discriminating variables are seven, so two functions are
there.

The Eigen values (Table 4.6) exhibit, for the first function, the magnitudes of
the eigen values is 1.088 and the percentage of variance explained by this
function is 93.6 which is the discriminating ability of this function. High value
is recommended for a strong function. Similarly, for the second function, the
eigen value associated is .077 which accounts for 6.6 percentage of explained
variance. Percentage (%) of variance specifies the percentage of variance
explained by the variables in the function. This is calculated as the proportion
of the function‟s eigen value to the sum of all the eigen values. A large value
of canonical correlation shows high discriminating ability of the function.

Table 4.6. Eigen Values (e-HRM)

Eigen %of Cumulative Canonical

Function
Value Variance % Correlation

1 1.088a 93.4 93.4 .722

2 .077a 6.6 100.0 .268

a. First 2 canonical discriminant functions were used in the analysis

Wilks' Lambda were significant which means that the group means vary
significantly. This shows the proportion of total variance in the discriminant

76
scores which is not explained by the differences among the groups. A small
value is recommended for apparently different group means. Chi square's high
value shows that the functions differ significantly from each other. Table 4.7
on the next page shows the values for WIlk's Lambda, Chi-Square, etc. In this
table, under 'Test of Functions', '1 through 2' indicate that no function has been
omitted and '2' means when the first function is removed.(Bajpai, 2011). In
both cases, p = 0.000 i.e. p-value is significant or both fuctions taken together
as well as the second function alone contribute significantly to the difference
among groups.

Table 4.8 shows the standardized canonical discriminant function coefficients

which are used to calculate the discriminant score.

Table 4.7. Wilk's Lambda (e-HRM)

Test of Wilk's
Chi-square df Sig.
Function(s) Lambda

1through 2 .445 312.944 14 .000

2 .928 28.771 6 .000

Table 4.8. Standardized Canonical Discriminant Function Coefficients (e-

HRM)

Variables Function 1 Function 2

HRR .263 .357

OC -.236 .256

EC -.075 -.140

SC .898 -.117

UC -.026 -.479

PEOU -.038 .509

PU .010 .433

77
Table 4.9 gives structure matrix which shows the structured correlations
(canonical loading or discriminant loading) of the discriminant functions. It
represents the correlations between the observed discriminating variables and
the dimensions created with the unobserved discriminant functions. The
correlation value is directly proportional to the importance of the
corresponding predictor (Bajpai, 2011).

Table 4.9. Structure Matrix (e-HRM)

Variables Function 1 Function 2

SC .917* .012

OC -.406* .294

HRR .276* .154

PEOU .044 .625*

UC -.102 -.589*

PU .068 .536*

EC -.085 -.276*
Pooled within-groups correlations between
discriminating variables and standardized canonical
discriminant functions
Variables ordered by absolute size of correlation
within function.
*. Largest absolute correlation between each variable
and any discriminant function

Table 4.10 on the next page shows the functions at group centroids. These are
the values of group means for each function calculated by putting the variable
means for each group in the discriminating equation. 'Low' group has the
higher value on Function 2 (predominantly associated with PEOU, PU, EC
and UC). Low engaged employees consider higher PEOU, PU, EC and UC.

78
Table 4.10.Functions at Group Centroids (e-HRM)

Level Function 1 Function 2

Low -1.065 .011

Medium .967 -.325

High 1.080 .465

Unstandardized canonical discriminant functions
evaluated at group means
The percentage of original cases correctly classified is 68.1 which is
satisfactory, shown in Table 4.11 below and hence this analysis can be relied
on.

Table 4.11. Classifiation Resultsa (e-HRM)

Predicted Group Membership

In terms of Level Total
Low Medium High

Low 159 13 19 191

Original Count Medium 14 66 41 121

High 8 30 42 80

Low 83.2 6.8 9.9 100.0

Percentage (%) Medium 11.6 54.5 33.9 100.0

High 10.0 37.5 52.5 100.0

a. 68.1% of original grouped cases correctly classified.

Similarly, the results of the discriminant analysis for examining the

classification of engagement levels based on each individual determinant of e-
HRM are shown in the following fashion:

79
(ii) HRR and engagement levels: Tables 4.12-4.17

Table 4.12. Eigen Values (HRR)

Eigen %of Cumulative Canonical

Function
Value Variance % Correlation

1 .102a 76.4 76.4 .305

2 .032a 23.6 100.0 .175

a. First 2 canonical discriminant functions were used in the analysis

Table 4.13. Wilk's Lambda (HRR)

Test of Wilk's
Chi-square df Sig.
Function(s) Lambda

1 through 2 .879 49.470 20 .000

2 .969 11.997 9 .213

Table 4.14. Standardized Canonical Discriminant Function Coefficients

(HRR)

Variables Function 1 Function 2

HRR9 .147 .606

HRR2 .355 -.003

HRR7 .124 -.116

HRR8 .161 -.110

HRR10 -.519 .581

HRR6 -.632 .702

HRR1 .432 -.004

HRR5 .155 -.360

HRR4 .806 -.203

HRR3 -.148 -.260

80
Table 4.15. Structure Matrix (HRR)

Variables Function 1 Function 2

HRR9 .724* .308

HRR2 .478* .343

HRR7 .472* .462

HRR8 .461* .212

HRR10 .265* .211

HRR6 .143 .696*

HRR1 .480 .688*

HRR5 .101 .623*

HRR4 .305 .337*

HRR3 .275 .297*

Pooled within-groups correlations between
discriminating variables and standardized canonical
discriminant functions
Variables ordered by absolute size of correlation
within function.
*. Largest absolute correlation between each variable
and any discriminant function

Table 4.16.Functions at Group Centroids (HRR)

Level Function 1 Function 2

Low .286 -.088

Medium -.098 .260

High -.536 -.183

Unstandardized canonical discriminant functions
evaluated at group means

81
Table 4.17. Classifiation Resultsa (HRR)

Predicted Group Membership

In terms of Level Total
Low Medium High

Low 85 46 60 191

Original Count Medium 31 49 41 121

High 14 18 48 80

Low 44.5 24.1 31.4 100.0

Percentage (%) Medium 25.6 40.5 33.9 100.0

High 17.5 22.5 60.0 100.0

a. 46.4% of original grouped cases correctly classified.

The tables above give the Eigen values, Wilk's Lambda, shows the
standardized canonical discriminant function coefficients, the canonical
structure of the discriminant functions and the group means for each function
calculated by putting the variable means for each group in the discriminating
equation. The values are found to be satisfactory. Also, the percentage of
original cases correctly classified is 46.4 which is satisfactory and hence this
analysis can be relied on. The results clearly show that HRR is able to
discriminate between the levels of employee engagement.

(iii) OC and engagement levels: Tables 4.18-4.23

Table 4.18. Eigen Values (OC)

Eigen %of Cumulative Canonical

Function
Value Variance % Correlation

1 .432a 91.4 91.4 .549

2 .040a 8.6 100.0 .197

a. First 2 canonical discriminant functions were used in the analysis

82
Table 4.19. Wilk's Lambda (OC)

Test of Wilk's
Chi-square df Sig.
Function(s) Lambda

1 through 2 .671 154.221 10 .000

2 .961 15.350 4 .004

Table 4.20. Standardized Canonical Discriminant Function Coefficients (OC)

Variables Function 1 Function 2

OC1 .752 .478

OC2 .477 .486

OC3 -.253 .021

OC4 .027 -.004

OC5 -.434 .830

Table 4.21. Structure Matrix (OC)

Variables Function 1 Function 2

OC1 .852* .160

OC5 -.615 .753*

OC2 .211 .603*

OC3 .048 .358*

OC4 .161 .354*

Pooled within-groups correlations between
discriminating variables and standardized canonical
discriminant functions
Variables ordered by absolute size of correlation
within function.
*. Largest absolute correlation between each variable
and any discriminant function

83
Table 4.22.Functions at Group Centroids (OC)

Level Function 1 Function 2

Low -.657 .042

Medium .472 -.263

High .855 .297

Unstandardized canonical discriminant functions
evaluated at group means

Table 4.23. Classifiation Resultsa (OC)

Predicted Group Membership

In terms of Level Total
Low Medium High

Low 142 22 27 191

Original Count Medium 40 45 36 121

High 15 24 41 80

Low 74.3 11.5 14.1 100.0

Percentage (%) Medium 33.1 37.2 29.8 100.0

High 18.8 30.0 51.3 100.0

a. 58.2% of original grouped cases correctly classified.

The tables above give the Eigen values, Wilk's Lambda, shows the
standardized canonical discriminant function coefficients, the canonical
structure of the discriminant functions and the group means for each function
calculated by putting the variable means for each group in the discriminating
equation. The values are found to be satisfactory. Also, the percentage of
original cases correctly classified is 58.2 which is satisfactory and hence this
analysis can be relied on. The results clearly show that OC is able to
discriminate between the levels of employee engagement.

84
(iv) SC and engagement levels: Tables 4.24-4.29

Table 4.24. Eigen Values (SC)

Eigen %of Cumulative Canonical

Function
Value Variance % Correlation

1 .213a 90.9 90.9 .419

2 .021a 9.1 100.0 .145

a. First 2 canonical discriminant functions were used in the analysis

Table 4.25. Wilk's Lambda (SC)

Test of Wilk's
Chi-square Df Sig.
Function(s) Lambda

1 through 2 .807 83.061 10 .000

2 .979 8.211 4 .084

Table 4.26. Standardized Canonical Discriminant Function Coefficients (SC)

Variables Function 1 Function 2

SC1 .334 .045

SC2 -.236 -.622

SC3 .972 -.113

SC4 .262 .579

SC5 -.020 .808

These tables give the Eigen values, Wilk's Lambda, shows the standardized
canonical discriminant function coefficients, the canonical structure of the
discriminant functions and the group means for each function calculated by
putting the variable means for each group in the discriminating equation. The
values are found to be satisfactory. Also, the percentage of original cases
correctly classified is 53.1 which is satisfactory and hence this analysis can be

85
relied on. The results clearly show that SC is able to discriminate between the
levels of employee engagement.

Table 4.27. Structure Matrix (SC)

Variables Function 1 Function 2

SC3 .931* -.243

SC5 .066 .861*

SC4 .173 .719*

SC1 .218 .373*

SC2 .093 .250*

Pooled within-groups correlations between
discriminating variables and standardized canonical
discriminant functions
Variables ordered by absolute size of correlation
within function.
*. Largest absolute correlation between each variable
and any discriminant function

Table 4.28.Functions at Group Centroids (SC)

Level Function 1 Function 2

Low .471 .010

Medium -.484 .155

High -.392 -.260

Unstandardized canonical discriminant functions
evaluated at group means

86
Table 4.29. Classifiation Resultsa (SC)

Predicted Group Membership

In terms of Level Total
Low Medium High

Low 133 16 42 191

Original Count Medium 44 47 30 121

High 27 25 28 80

Low 69.6 8.4 22.0 100.0

Percentage (%) Medium 36.4 38.8 24.8 100.0

High 33.8 31.3 35.0 100.0

a. 53.1% of original grouped cases correctly classified.

(v) UC and engagement levels: Tables 4.30-4.35

The tables below give the Eigen values, Wilk's Lambda, shows the
standardized canonical discriminant function coefficients, the canonical
structure of the discriminant functions and the group means for each function
calculated by putting the variable means for each group in the discriminating
equation. The values are fund to be satisfactory. Also, the percentage of
original cases correctly classified is 44.6 which is satisfactory and hence this
analysis can be relied on. The results clearly show that UC is able to
discriminate between the levels of employee engagement.

Table 4.30. Eigen Values (UC)

Eigen %of Cumulative Canonical

Function
Value Variance % Correlation

1 .065a 80.3 80.3 .247

2 .016a 19.7 100.0 .125

a. First 2 canonical discriminant functions were used in the analysis

87
Table 4.31. Wilk's Lambda (UC)

Test of Wilk's
Chi-square df Sig.
Function(s) Lambda

1 through 2 .924 30.461 8 .000

2 .984 6.127 3 .106

Table 4.32. Standardized Canonical Discriminant Function Coefficients (UC)

Variables Function 1 Function 2

UC1 -.070 .732

UC2 .866 -.170

UC3 .153 .459

UC4 -.200 .405

Table 4.33. Structure Matrix (UC)

Variables Function 1 Function 2

UC2 .963* .161

UC3 .722* .314

UC1 -.096 .852*

UC4 -.239 .640*

Pooled within-groups correlations between
discriminating variables and standardized canonical
discriminant functions
Variables ordered by absolute size of correlation
within function.
*. Largest absolute correlation between each variable
and any discriminant function

88
Table 4.34.Functions at Group Centroids (UC)

Level Function 1 Function 2

Low -.177 .095

Medium -.043 -.187

High .487 .057

Unstandardized canonical discriminant functions
evaluated at group means

Table 4.35. Classifiation Resultsa (UC)

Predicted Group Membership

In terms of Level Total
Low Medium High

Low 90 57 44 191

Original Count Medium 45 47 29 121

High 20 22 38 80

Low 47.1 29.8 23.0 100.0

Percentage (%) Medium 37.2 38.8 24.0 100.0

High 25.0 27.5 47.5 100.0

a. 44.6% of original grouped cases correctly classified.

(vi) EC and engagement levels: Tables 4.36-4.41

The tables above give the Eigen values, Wilk's Lambda, shows the
standardized canonical discriminant function coefficients, the canonical
structure of the discriminant functions and the group means for each function
calculated by putting the variable means for each group in the discriminating
equation. The values are found to be satisfactory. Also, the percentage of
original cases correctly classified is 61.7 which is good and hence this analysis
can be relied on. The results clearly show that EC is able to discriminate
between the levels of employee engagement.

89
Table 4.36. Eigen Values (EC)

Eigen %of Cumulative Canonical

Function
Value Variance % Correlation

1 .659a 95.0 95.0 .630

2 .034a 5.0 100.0 .182

a. First 2 canonical discriminant functions were used in the analysis

Table 4.37. Wilk's Lambda (EC)

Test of Wilk's
Chi-square df Sig.
Function(s) Lambda

1 through 2 .583 208.983 10 .000

2 .967 13.107 4 .011

Table 4.38. Standardized Canonical Discriminant Function Coefficients (EC)

Variables Function 1 Function 2

EC1 -.376 -.548

EC2 -.001 .104

EC3 .028 .932

EC4 .147 -.371

EC5 1.017 -.082

90
Table 4.39. Structure Matrix (EC)

Variables Function 1 Function 2

EC5 .927* -.185

EC4 .068* -.001

EC3 .031 .751*

EC1 -.124 -.448*

EC2 .039 .383*

Pooled within-groups correlations between
discriminating variables and standardized canonical
discriminant functions
Variables ordered by absolute size of correlation
within function.
*. Largest absolute correlation between each variable
and any discriminant function

Table 4.40.Functions at Group Centroids (EC)

Level Function 1 Function 2

Low -.825 .020

Medium .687 -.228

High .930 .297

Unstandardized canonical discriminant functions
evaluated at group means

91
Table 4.41. Classifiation Resultsa (EC)

Predicted Group Membership

In terms of Level Total
Low Medium High

Low 151 20 20 191

Original Count Medium 26 51 44 121

High 13 27 40 80

Low 79.1 10.5 10.5 100.0

Percentage (%) Medium 21.5 42.1 36.4 100.0

High 16.3 33.8 50.0 100.0

a. 61.7% of original grouped cases correctly classified.

(vii) PEOU and engagement levels: Tables 4.42-4.47

The tables below give the Eigen values, Wilk's Lambda, shows the
standardized canonical discriminant function coefficients, the canonical
structure of the discriminant functions and the group means for each function
calculated by putting the variable means for each group in the discriminating
equation. The values are found to be satisfactory. Also, the percentage of
original cases correctly classified is 53.8 which is satisfactory and hence this
analysis can be relied on. The results clearly show that PEOU is able to
discriminate between the levels of employee engagement.

Table 4.42. Eigen Values (PEOU)

Eigen %of Cumulative Canonical

Function
Value Variance % Correlation

1 .093a 88.2 88.2 .291

2 .012a 11.8 100.0 .111

a. First 2 canonical discriminant functions were used in the analysis

92
Table 4.43. Wilk's Lambda (PEOU)

Test of Wilk's
Chi-square df Sig.
Function(s) Lambda

1 through 2 .904 39.248 4 .000

2 .988 4.807 1 .028

Table 4.44. Standardized Canonical Discriminant Function Coefficients

(PEOU)

Variables Function 1 Function 2

PEOU1 .911 .412

PEOU2 -.422 .907

Table 4.45. Structure Matrix (PEOU)

Variables Function 1 Function 2

PEOU1 .906* .422

PEOU2 -.412 .911*

Pooled within-groups correlations between
discriminating variables and standardized canonical
discriminant functions
Variables ordered by absolute size of correlation
within function.
*. Largest absolute correlation between each variable
and any discriminant function

93
Table 4.46.Functions at Group Centroids (PEOU)

Level Function 1 Function 2

Low -.306 .021

Medium .227 -.144

High .387 .168

Unstandardized canonical discriminant functions
evaluated at group means

Table 4.47. Classifiation Resultsa (PEOU)

Predicted Group Membership

In terms of Level Total
Low Medium High

Low 128 23 40 191

Original Count Medium 35 48 38 121

High 18 27 35 80

Low 67.0 12.0 20.9 100.0

Percentage (%) Medium 28.9 39.7 31.4 100.0

High 22.5 33.8 43.8 100.0

a. 53.8% of original grouped cases correctly classified.

(viii) PU and engagement levels: Tables 4.48-4.53

The tables below give the Eigen values, Wilk's Lambda, shows the
standardized canonical discriminant function coefficients, the canonical
structure of the discriminant functions and the group means for each function
calculated by putting the variable means for each group in the discriminating
equation. The values are found to be satisfactory. Also, the percentage of
original cases correctly classified is 66.8 which is good and hence this analysis
can be relied on. The results clearly show that PU is able to discriminate
between the levels of employee engagement.

94
Table 4.48. Eigen Values (PU)

Eigen %of Cumulative Canonical

Function
Value Variance % Correlation

1 .981a 99.6 99.6 .704

2 .004a .4 100.0 .061

a. First 2 canonical discriminant functions were used in the analysis

Table 4.49. Wilk's Lambda (PU)

Test of Wilk's
Chi-square df Sig.
Function(s) Lambda

1 through 2 .503 267.114 4 .000

2 .996 1.466 1 .226

Table 4.50. Standardized Canonical Discriminant Function Coefficients (PU)

Variables Function 1 Function 2

PU1 -.306 .952

PU2 .961 .279

Table 4.51. Structure Matrix (PU)

Variables Function 1 Function 2

PU2 .952* .306

PU1 -.279 .960*

Pooled within-groups correlations between
discriminating variables and standardized canonical
discriminant functions
Variables ordered by absolute size of correlation
within function.
*. Largest absolute correlation between each variable
and any discriminant function

95
Table 4.52.Functions at Group Centroids (PU)

Level Function 1 Function 2

Low 1.011 -.003

Medium -1.017 -.066

High -.875 .108

Unstandardized canonical discriminant functions
evaluated at group means

Table 4.53. Classifiation Resultsa (PU)

Predicted Group Membership

In terms of Level Total
Low Medium High

Low 156 10 25 191

Original Count Medium 11 72 38 121

High 11 35 34 80

Low 81.7 5.2 13.1 100.0

Percentage (%) Medium 9.1 59.5 31.4 100.0

High 13.8 43.8 42.5 100.0

a. 66.8% of original grouped cases correctly classified.

Figure 4.2 shows the graph representing the distribution of respondents over
three categories of engagement. As it is visible from this figure, the low
engaged employees tend to be at the more negative end of horizontal
dimension (Function 1) and medium/ high engaged employees tend to be at
the opposite end. On vertical dimension (Function 2), the results are not clear
but the high engaged employees tend to be higher as compared to medium
engaged employees.

96
 Figure 4.2: Graph of individuals on discriminant dimensions
Source: Author's findings

The results show that proportion of low engaged employees is much higher
than highly engaged ones which is well supported by the reports of
NASSCOM (Nasscom, 2015). Also, employees standing beneath the umbrella
of medium engaged level are higher than those highly engaged, but less than
low engaged. They are the ones have the potential to reach higher engagement
level if proper measures are taken.

Both the discriminant functions are significant (p<0.05) and hence, both the
functions are able to classify between the segments. Results show that 'system
characteristics' contribute most in differentiating between group 1 (low) and
group 2 (medium) / group 3 (high) to the employee engagement, followed by
the 'role of HR manager'. It suggests that employees are possibly more lured to
the characteristics of the e-HRM system being used in the firm than the HR
efforts. These findings are in line with the results reviewed and suggested by
earlier researches (Bondarouk et al., 2017). The factors adapted from TAM i.e.

97
PEOU and PU contribute most in differentiating between group 2 and 3 which
is a little surprising because these factors play a little role in discriminating
between low engaged and medium/highly engaged towards employee
engagement. However, the results are in line with recent study (Sivapragasam
and Raya, 2017; SHRM, 2017; Bondarouk et al, 2017). The employees who
are not engaged do not consider usability to be very important contributing
factor and the system being easy to use falls on negative side, possibly because
the employees from IT industry are well conversed with technology and ease
of use may be perceived as technically not so high. These factors have already
been suggested not to be very significant (Heikkilä and Smale, 2011).

4.4 ONE-WAY ANOVA

ANOVA Table (Table 4.54) shows the output of the ANOVA analysis. The
results suggest that there is a statistically significant difference between the
group means for employee engagement. There is a statistically significant
difference in the mean employee engagement between the three companies
taken for study. To know about which specific companies differed, Tukey Post
Hoc test results are observed in the Multiple Comparisons table (Table 4.55).
It is observed that there is a statistically significant difference (p=0.000) in
employee engagement between Company 1 (TCS) and Company 2 (Infosys),
as well as between Company 1 and Company 3 (Wipro) and also, between
Company 2 and Company 3. This means that there was a statistically
significant difference between all the 3 companies as determined by one-way
ANOVA (F(2,293) = 575.792, p = .000) and Tukey Post Hoc test.

Table 4.54. ANOVA

Sum of Mean
Engagement Squares Df Square F Sig.

Between
31253.596 2 15626.798 575.792 .000
Groups

Within Groups 7951.918 293 27.140

Total 39205.514 295

98
Table 4.55. Post Hoc Tests: Multiple Comparisons

Dependent Variable: Engagement

Tukey HSD

Mean 95% Confidence Interval

(I) (J) Std.
Difference Sig. Lower Upper
Company Company Error
(I-J) Bound Bound

1 2 -13.224* .736 .000 -14.96 -11.49

3 -25.717* .758 .000 -27.50 -23.93

2 1 13.224* .736 .000 11.49 14.96

3 -12.493* .734 .000 -14.22 -10.76

3 1 25.717* .758 .000 23.93 27.50

2 12.493* .734 .000 10.76 14.22

*. The mean difference is significant at the 0.05 level.

Note: Company 1= Tata Consultancy Services, Company 2= Infosys
Technologies Limited and Company 3= Wipro Technologies

4.5 DISCUSSION

In this study, our research was limited to general e-HRM research and
therefore, it did not cover the literature available for specific e-HRM areas like
e-recruitment. e-selection, e-learning and so on. Moreover, the factors
identified in this study are factors of e-HRM which may cover factors
pertaining to one or all of the adoption, implementation, use and effectiveness
of e-HRM.

The findings of this research suggest that e-HRM is affected by the factors
which can majorly be classified categorically as human, organization, system
and environment (HOSE). For this study, these factors have been extended to
form seven dimensions of e-HRM. 'Role of HR' and 'User Characteristics'
emerge from human category of factors; 'System Characteristics' along with
99
'Perceived Usefulness' and 'Perceived Ease Of Use' belong to system category
of factors; 'Organizational Characteristics' form organization category and
environment category takes the form of 'Environmental Characteristics'.
Earlier, some authors have described a similar framework, known as TOP
framework ( T- Technology, O- Organization and P- People factors)
(Bonadarouk et al., 2017). Technology factors are similar to category of
system factors; People factors correspond to human category of factors, but
this framework lacks the environmental factors. The results and findings
pertaining to the seven factors are discussed below with reference to previous
findings.

Role of HR (HRR)

This consists of a wide spectrum of ten sub-factors: Change agent, Motivator,

Mentor, Leader, Communicator, Analyst, Administrator, Advisor, Managerial
support and Training and Development. All the sub-factors act proportionately
towards effective implementation of e-HRM. HRR plays an important role in
classifying between group 2 and 3.

Authors suggest that lack of top management support is like a hindrance in the
effectiveness and successful adoption of e-HRM (Panayotopoulou et al., 2007;
Bondarouk et al., 2017). HR professionals sometimes believe computerization
as expensive and unhelpful in their own careers. Thus, they are unable to
justify its cost against benefits. HR managers are the ones who act as leader,
motivator, initiator, change agent, etc. and so they account for engaging the
employees to a great extent. These findings are also supported by recent
researches (Deshwal, 2015; SHRM, 2017; Bondarouk et al., 2017). There
should be a shared vision among the HR managers and IT professionals
(Tansley and Newell, 2007). They should act as efficient communicator and
advisor. Lack of proper training and failing to identify the needs of
development can act as barrier towards adoption of e-HRM and lead to
negative attitude of employees towards e-HRM (Martin and Reddington,
2010). According to previous research findings, proper administration,
communication between HR and other functions, provision of training and soft
skills of HR managers who support e-HRM like leadership, initiating power,

100
training skills etc. are found to be important determinants of effective e-HRM
(Tansley and Watson, 2000; Hustad and Munkvold, 2005; Florkowski and
Olivas-Lujan, 2006; Panayotopoulou et al., 2007; Reddick, 2009 Martin and
Reddington, 2010, Bondarouk et al., 2017).

Organizational Characteristics (OC)

This dimension takes care of sub-factors like organization's age, size,

structure, innovative culture, strategies and policies. Organization size and
structure is considered as a factor for e-HRM adoption by most of the
researchers. They found that it has a significant contribution to the same
(Tansley and Watson, 2000; Teo et al., 2007, Strohmeier and Kabst, 2009).
while adoption is more widespread among large organizations. Strohmeier and
Kabst (2009) describe larger companies as earlier adopters, successful
adoption is more widespread among small organizations (Chapman and
Webster, 2003). Organization age also plays a role towards implementing e-
HRM (Haines and Lafleur, 2008; Olivas-Lujan and Florkowski, 2010). The
strategies and policies of an organization are found to be positively significant
towards e-HRM adoption and implementation. Employees are supposed to feel
comfortable in using a new system if they are well aware of and well
communicated about the policies, goals and strategies being adopted for
achieving the goals (Ruël et al., 2004). They should not be kept in dark about
the reasons of adopting a new technology. Lack of proper policies
implementation can be negative towards e-HRM acceptance by employees and
therefore, ineffective e-HRM.

OC also have an important contribution in discriminating all the groups which

is in line with the previous researches (Robertson-Smith and Markwick, 2009;
Choochote and Chochiang, 2015; Bondarouk et al., 2017).

Environmental Characteristics (EC)

The environmental factors which comprise of culture change, external forces,

competitive actions, regulations and technological development, were taken
care of as a research gap in early studies (Bondarouk et al., 2017). They were
found to have a negative effect on e-HRM implementation. These are the

101
factors which are not studied rigorously and have very little literature
available. These are the factors which affect the organization's decision of e-
HRM adoption in some way or the other but cannot be controlled from within
the organization.

External forces like third party interference, union presence in vendors, etc.
have a negative impact on e-HRM adoption and implementation. If
organizations take a decision under the influence of these factors, e-HRM may
prove unsuccessful in providing benefits (Haines and Lafleur, 2008).
Similarly, rules and regulations imposed by the government and other
governing agencies can also act as a barrier in implementing e-HRM. Change
of culture from one place to another was also found to influence e-HRM
adoption (Olivas-Lujan et al., 2007; Smale and Heikkilä, 2009) in the way
employees look forward to their managers and organization. Countries' culture
vary greatly and so do the decisions for e-HRM implementation. The
competitive forces imposed by the major competitive firms also influence e-
HRM adoption and implementation. The use of one type of e-HRM by a firm
may force its competitors to stick to the same type or move towards updating
to a higher version leading towards resistance by employees (Beulen, 2009;
Smale and Heikkilä, 2009). In a similar fashion, frequent technological
improvements may require frequent upgrading of the system. This in turn,
might make employees feel uncertain about the use of the new system (Olivas-
Lujan et al., 2007; Smale and Heikkilä, 2009; Strohmeier and Kabst, 2009).

The findings suggest that these factors play a more contributing role in
discriminating between medium and high engaged employees as compared to
that between low and medium/high engaged ones. This suggests that measures
taken by the government and other ruling bodies can negatively affect the
successful implementation of e-HRM and thereby less engaging employees.
Language (Tansley et al., 2001; Heikkilä and Smale, 2011) can act as a
barrier. Social influence (Venkatesh et al., 2003; SHRM, 2017), global issues
(SHRM, 2017; Sivapragasam and Raya, 2017) and the change of culture
(Tansley, 2001; Beulen, 2009; Robertson-Smith and Markwick, 2009; SHRM,
2017; Bondarouk et al., 2017) can also act as antagonist in engaging
employees through the use of e-HRM.

102
System Characteristics (SC)

These factors include sub-factors like standardization, reliability, robustness,

interactive and response rate. For e-HRM to deliver expected outcomes, there
are certain features pertaining to the system being adopted which influence the
adoption, implementation and effectiveness of e-HRM. Researchers have
advised in their studies to analyze and identify organizational needs and
accordingly, specify and check for the specifically required technology
characteristics before adopting or upgrading new systems.

Robustness, reliability and response rate are some of the aspects of quality of
e-HRM system which are found to be contributing factors for implementing e-
HRM successfully which are consistent with earlier findings (Ruël et al.,
2007; Bondarouk et al., 2009). Researchers suggest that the system should be
integrated coherently with the HR strategies and it should focus on being
easily accessible, user-friendly, fast, robust and reliable in terms of output.
Only then, the implementation of e-HRM can reach towards attainment of its
goals and provide the expected output well within time (Tansley and Watson,
2000; Ruta, 2009).

Standardization of the system was suggested to be an important factor of e-

HRM adoption (Ruël et al., 2007). Tailoring the system as per the needs are
surprisingly found to be negative towards implementing e-HRM in this
research. The role of standardization of e-HRM does not replicate the expected
results which is justified because if the system is standardized, the intended
output may change subject to the fixed attributes. It might be justified for IT
professionals because if they feel that computerization is overly time
consuming and the output thus obtained is unreliable and not as expected, they
act towards preventing e-HRM adoption (Bondarouk et al., 2017). There are
different goals of e-HRM for different users viz. managers, team leads, IT
professionals, HR professionals, etc. Hence, a multi-stakeholder perspective
needs to be adopted for exploring e-HRM effectiveness in real life (Bondarouk
et al., 2009). For HR professionals to accept new technologies they need to
know how to effectively work with them and become convinced about the
value of new systems. For IT professionals, the system must adhere to their

103
specific needs of understanding the requirements of project, client, etc.
Therefore, the intended output changes with respect to the users. The
standardization, if not done to cater to every user's needs, it might be rejected
by other users. Thus, it shows a negative relation as standardizing the system
as per one type of users would mean non-complying with others.

User friendliness is found to affect e-HRM positively but in the present

research, the feature of e-HRM being interactive doesn't support e-HRM
implementation. This finding can be accounted to the fact that the users of e-
HRM in this research are IT professionals who understand the technological
terms and therefore are comfortable in using the system if it is user friendly
and easy to use without bothering about being interactive. They have an
indifferent attitude towards the use of interactive e-HRM. Moreover, being
interactive means being specific in one or few languages. Previous studies
show that language standardization has a negative effect on the acceptance and
use of e-HRM system and suggested that it is the language capabilities of
employees which decide the side of the continuum of effects (Heikkilä and
Smale, 2011)

User Characteristics (UC)

'User characteristics' which include domain knowledge, technical expertise,

user acceptance and intention to use, are found to have a negative impact on e-
HRM effectiveness. The intended users belong to IT industry and were thus
technically sound which may account for the findings. Managing people is one
of the most essential factors as they are the backbone for adoption of any
technology. Without the right domain knowledge, technical soundness, one
cannot use the system efficiently. In addition, the user should have the
intention to use it for the intended purpose and should be able to accept the
change which the system would bring along.

The IT professionals are supposedly technically sound and knowledgeable.

But, if the employees have high expertise, which may be due to high
experience or vast knowledge, they may resist using new system due to hands-
on experience with the existing system (Olivas-Lujan et al., 2007; Beulen,
2009). Moreover, e-HRM systems take typically long time for development

104
and so, by the time systems were finally up and running they might fail to
represent the latest technology which was expected by highly skilled users.
Overall, users with more developed computer skills seemed to use systems
earlier, but at the same time were generally less positive about doing so
(Haines and Petit, 1997, Tansley et al., 2001, Bondarouk et al., 2017).

UC have a substantive contribution in discriminating between medium and

high engaged employees but the intensity of contribution is less in
discriminating low engaged employees from medium/high engaged
employees. Moreover, these factors lie on negative side which is justified as
the users are technically efficient because they belong to technical industry
and when they are technically sound, they can easily relate the system to their
needs. Finding even a small discrepancy can lead to a negative feedback and
thus less engagement. similarly, earlier studies suggest that if the employees
are more experienced, they may feel the use of a new technology not very
attractive as they are used to the practices being followed for years and hence,
a new introduction can lead to engaging themselves on a lesser extent (Bell et
al., 2006; Heikkilä and Smale, 2011; Bondarouk et al., 2017).

Perceived Ease Of Use (PEOU) and Perceived Usefulness (PU)

These two factors are discussed together because as stated earlier, both of
them are adapted from the same model i.e. TAM (Davis, 1989). As per this
model, the users of a new system or new technology accept it depending upon
how they perceive it in terms of being easy to use or useful. e-HRM is a
technological advancement or a technically advanced system which has to be
accepted by the users for obtaining the maximum benefits. If it is not
perceived by the users to be easy to use, then they are likely to reject it or
unable to utilize it to the fullest extent. Likewise, if the users perceive the
system as not of use, possibly, they would not like to use it. Hence, the
findings of this research say that the perceptions of employees are important
factors of e-HRM adoption and implementation. The findings are in line with
the previous studies which say that higher the perceived usefulness of e-HRM
systems in the minds of the employees, the greater the net benefit of high
levels of e-HRM systems in the organization (Pant et al. 2012).

105
The views and findings of previous researches are contradicting when talking
about PEOU and PU. Some authors found PEOU to be less significant
whereas PU to be more significant factor of e-HRM (Voermans and Van
Veldhoven, 2007) whereas some found it to be other way (Teo et al., 2007).
Some researchers also found out both of them to be significant (Olivas-Lujan
et al., 2007; Pant et al. 2012) whereas a few suggested both of them to be less
significant (Ruël et al., 2004; Ruël et al., 2007).

Discussing about the differences among the top three companies (TCS,
Infosys and Wipro), findings suggest that the engagement is highest in Wipro
and lowest in TCS with Infosys in between. The reason beyond this could be
attributed to the age of the organization; the oldest being the least engaged.
TCS was established in 1965 and both Infosys and Wipro started their
technology business in 1981 (www.tcs.com, 2018; www.infosys.com;
www.wipro.com). Moreover, the technologies and strategies being used for
the employee engagement in the companies might be the major role playing
actors. This is evident from the statement given by Mr. Saurabh Govil, the
Senior Vice President and Global HR Head, Wipro, "We have shifted focus
from hiring to re-skilling and training employees, giving them the opportunity
to continuously learn and grow by the use of newer technologies and take on
higher responsibilities. Our employee engagement scores thus went up by 1%,
employee participation scores by 4% and employee attrition dropped to be
in16% band in the last six quarters." (www.thehindubusinessline.com, 2018).
He also said that Wipro is soon going to introduce a new employee
engagement and participation framework which is currently work in progress
and will be launched very soon.

This clearly indicates that Wipro is moving at a fast rate towards achieving its
goal of bringing attrition rate to 14 percent band and using newer technologies
with continuous up gradation in order to gain this.

However, TCS is still using the same programs like PEEP (Proactive
Employee Engagement Program), PULSE, Hats Off and PROPEL for
engaging employees which it was using in 2008 i.e. a decade ago when I was
working with TCS (www.tcs.com, 2018). It shows that TCS has not updated

106
its employee engagement programs. Although TCS has won Employee
engagement awards in 2016 but those awards were for North America
(www.tcs.com, 2018) and this study is limited to Indian employees.

Similarly, in Infosys, they are using the employee engagement programs but
are not upgrading with today's dynamic environment. It has a web portal
named Bubble for social networking for its employees just like Facebook. It
uses Infosys Television and radio for quick interaction and communication, the
age-old methods (www.infosys.com).

Therefore, the results obtained can be attributed to the use of technology in

these companies. Thus, the results and findings of this research are mostly
found to be in line with the previous studies. The change of domain i.e. the
geographical area and the industry could be accounted the findings which
differ from the earlier results, due to respondents being technically sound and
comfortable.

In the next chapter, recommendations and suggestions for the use of e-HRM in
IT industry for engaging employees are provided. The recommendations are
enlisted keeping in mind the implication of the research findings on different
sections of society viz. budding scholars, industrialists, mangers,
academicians, etc.

107