Similar to the earlier research, stage three alternative tests were undertaken in order to assess discriminant validity of the five sub-dimensions: first, a test for the confidence interval of the factors’ correlations; second, a test that compares the factor’s shared variance with other factors in the model relative to the AVE by each factor (Fornell & Larcker, 1981); the third test for discriminant validity involved the evaluation of competing factor models by means of chi- square difference tests and the evaluation of fit indices.

238

Sub

dimension

У

Item

X

Squared

Multiple

Correlation

Corrected

Item-to-Total

Correlation

Average

Inter-Item

Correlation

Cronbach’s Alpha

Construct

Reliability

AVE

Construct

Reliability

AVE

Ability

.93

Ab7 Ab8 Ab9 AblO Abl 1

.85

.81

.82

.86

.82

.72

.66

.71

.73

.67

.80

.76

.81

.80

.78

.70

.92

92

.70

.91

.67

Integrity/

Honesty

.90

In2

In3

In4

In5

In6

In7

In9

InlO

.82

.83

.86

.85

.87

.73

.81

.72

.67

.68

.73

.72

.76

.53

.65

.52

.80

.80

.84

.81

.84

.69

.78

.70

.65

.94

.94

.66

Benevolence

.64

Bel

Be2

Be3

Веб

.78

.75

.75

.63

.60

.56

.57

.40

.70

.63

.67

.53

.52

.81

.82

.53

Willingness to rely

.73

Wil

Wi4

Wi5

Wi8

.88

.85

.85

.86

.77

.73

.72

.75

.83

.81

.81

.82

.74

.92

.92

.74

Willingness to depend

.86

Wi2

Wi6

Wi7

.71

.83

.84

.50

.68

.70

.64

.70

.73

.62

.83

.84

.63

Table 28: Psychometric Properties of the eWOM Trust Scale (Sample 4)

Notes: у = Completely standardized second-order loading; X = Completely standardized first-order loading; AVE = Average variance extracted.

Table 29 presents the correlations, the squared correlations, as well as the upper bound of the correlations’ 95 percent confidence intervals between the factors. To illustrate, the upper bound of the confidence interval was .63 for the benevolence and integrity/honesty correlation, .72 for the willingness to rely and ability correlation, and .61 for the willingness to depend and benevolence correlation. The highest upper bound of the confidence interval was achieved for the correlation between the ability and the integrity/honesty dimensions with .86. The two upper bounds of the correlation interval for willingness to depend and integrity/honesty (.81) and willingness to depend and ability (.83) were only slightly lower. However, no factor pairing was found to be perfectly correlated, as none of the identified confidence intervals included 1.0. This result supported the discriminability of the five factors. To further evaluate discriminant validity for each possible factor pairing, the factors’ shared variance was compared with the AVE of the involved factors. It has been suggested that if the square of the parameter estimates between two constructs is less than the average variance extracted of the two constructs, discriminant validity is supported (Fornell & Larcker, 1981). In nine out of ten possible factor pairings, this test led to desirable results, as the squared correlation between those factors was judged lower than the respective AVEs. However, the second test revealed some problems with discriminant validity, as the shared variance between ability and integrity/honesty (.69) was equal to the AVE of the integrity/honesty dimension. A similar pattern was also detected in one sample during the reliability stage. Therefore, while the dimensionality of the remaining factors was without question, the discriminability of these two factors was potentially at risk.

Table 29: Correlations among the Sub-dimensions (Sample 4)

Integrity/

Honesty

Ability

Benevolence

Willingness to rely

Willingness to depend

Integrity/

Honesty

0.69^{a}

Ability

0.83^{b}

0.86^{c}

0.69^{d}

0.70^{a}

Benevolence

0.58^{b}

0.63^{c}

0.33^{d}

0.60^{b}

0.65^{c}

0.36^{d}

0.53^{a}

Willingness to rely

0.65^{b}

0.70^{c}

0.43^{d}

0.67^{b}

0.72^{c}

0.45^{d}

0.47^{b}

0.53^{c}

0.22^{d}

0.74^{a}

Willingness to depend

0.77^{b}

0.81^{c}

0.60^{d}

0.80^{b}

0.83^{c}

0.64^{d}

0.55^{b}

0.61^{c}

0.31^{d}

0.63^{b}

0.68^{c}

0.39^{d}

0.63^{a}

Notes: ^{a} = Average variance extracted (AVE); ^{b} = Pearson correlation coefficient (r); ^{c }confidence interval of r; ^{d} = Squared correlation (r^{2}); n = 526.

= Upper bound of 95%

Evidence for discriminant validity of the five sub-dimensions was provided by the assessment of the chi-square difference statistics and fit indices of several competing models. Here, ten alternative measurement models were proposed with variable factor combinations. Due to considerable correlations among the ability, integrity/honesty, and willingness to depend dimensions (.83 between Ability and integrity/honesty; .80 between ability and willingness to depend; and .77 between integrity/honesty and willingness to depend), they were combined in a single factor which is hypothesized to coexist with the benevolence and the willingness to rely as first-order factors in a second-order model (Model 3). Additionally, five four-factor models (Model 4a-4e) were estimated, where two first-order constructs were combined due to their considerable intercorrelation while treating the remaining factors separately: Model 4a: ability and integrity/honesty (r = .83); Model 4b: ability and willingness to depend (r = .80); Model 4c: integrity/honesty and willingness to depend (r = .77); Model 4d: integrity/honesty and willingness to rely (r = .65); and Model 4e: willingness to rely and willingness to depend (r = .63). Also a one-factor model (Model 1) where eWOM trust was hypothesized as a unidimensional construct and a null model (Model N) that assumed no systematic relationship within the data were specified, along with the advocated second-order construct consisting of five separate sub-dimensions (Model 5). As shown in Table 30, Model 5 outperformed the other conceptualizations by having the lowest chi-square statistic (x^{2} = 774.84, df = 247, p < .001) and also met levels of satisfactory model fit (RMSEA = .06, NFI = .92, NNFI = .94, CFI = .95, RMR = .11, SRMR = .06, GFI = .89, AGFI = .86). This research was also able to demonstrate that, when comparing the five factor model to alternative models, chi-square difference statistics indicate a significantly better fit.

A final assessment of the discriminant validity among the five sub-dimensions of eWOM trust was applied using confirmatory factor analysis for the comparison of alternative factor pairings. For each possible pair, a two-factor model of which the assumption that the two factors are separate from each other formed the basis, was compared to a one-factor model that handled the two factors as a unidimensional construct. Another frame of reference was the null model, for which it was hypothesized that the items belonging to the two factors did not have any statistically relevant relationship.

As shown in Table 31, the two-factor models generally were superior to the single factor models. For instance, the two-factor model for integrity/honesty and ability appeared to have a chi-square statistic (x^{2} = 184.43; df = 64) which was considerably lower compared to the one- factor alternative (x^{2} = 627.37; df = 65), considering a common underlying construct for both dimensions. Accordingly, the chi-square difference test was significant (x^{2}Diff = 442.94; df Diff = 1; p < .001), which guided this research together with evident improvements in the fit indices to the acceptance that data could be best described by the more complex two-factor structure. As such a pattern was consistent throughout all possible pairings, considerable support in favor of the hypothesized dimensionality of the first-order constructs was gained.

241

Model

Chi

Square

4f

P

Chi Square Difference

Model Fit Indices

Competing

Models

Chi Square Difference

if

Difference

Sign.

Chi Square /

if

RMSEA

NFI

NNFI

CFI

RMR

SRMR

GFI

A GFI

Null Model (N)

9,934.83

276

36.00

n.a.

n.a.

n.a.

n.a.

n.a.

n.a.

n.a.

n.a.

One Factor Model (1)

2,598.86

252

.001

10.31

.18

.74

.73

.76

.15

.09

.59

.51

(N-l)

7,335.97

24

***

Three Factor Model (3) (A/I/W2; B; Wl)

1,362.67

249

.001

5.47

.11

.86

.87

.89

.11

.07

.77

.72

(N-3)

8,572.16

27

***

(1-3)

1,236.19

3

***

Four Factor Model (4 a) (A/I; B; Wl; W2)

1,230.70

248

.001

4.96

.11

.88 .89

.90 .11 .07 .78 .73

(N-4a)

8,704.13

28

***

(1 -4a)

1,368.16

4

***

(3 -4a)

131.97

1

***

Four Factor Model (4b) (A/W2; I; B; Wl)

981.33

248

.001

3.96

.08

.90

.92

.92

.13

.08

.86

.83

(N-4b)

8,953.50

28

***

(l-4b)

1,617.53

4

***

(3-4b)

381.34

1

***

Four Factor Model (4c) (IAV2; A; B; Wl)

990.32

248

.001

3.99

.08

.90

.91

.92

.13

.08

.85

.82

(N-4c)

8,944.51

28

***

(1 -4c)

1,608.54

4

***

(3-4c)

372.35

1

***

Table 30: Competing Models (Sample 4)

(Continued on next page.)

242

Model „ df Square

P

Chi Square Difference

Model Fit Indices

Competing Chi Square df Models Difference Difference

Sign.

Chi Square/ _{mmA WI mFI CM mIR SRMR GM AGFI}

it

Four Factor Model (4d) 1,849.14 248 (IAVI; I; B; W2)

.001

7.46 .14 .81 .82 .83 .14 .09 .69 .63

(N~4d) 8,085.69 28

***

(l~4d) 749.72 4

***

(3~4d) -486.48 1

n.s.

Four Factor Model (4e) 1,191.77 24 8 (I; A;B; W1AV2)

.001

4.81 .10 .88 .89 .90 .17 .10 .81 .76

(N~4e) 8,743.06 28

***

(Me) 1,407.09 4

***

(3Ae) 170.90 1

***

Five Factor Model (5) 774.84 247

.001

3.14 .06 .92 .94 .95 .11 .07 .89 .86

(N-5) 9,159.99 29

***

(1-5) 1,824.02 5

***

(3-5) 587.83 2

***

(4a-5) 455.86 1

***

(4b-5) 206.49 1

***

(4c-5) 215.48 1

***

(4d-5) 1,074.31 1

***

(4e-5) 416.93 1

***

Table 30 (Cont.): Competing Models (Sample 4)

Notes: A = Ability; I = Integrity/Honesty; В = Benevolence; WR = Willingness to rely; WD = Willingness to depend; *** = p <.001; n.s. = not significant; n.a. = not available.

243

Sub-dimension

Pairing

Chi

Square

Chi Square Difference

Model Fit Indices

Model

4f

P

Competing

Models

Chi Square Difference

if

Difference

Sign.

Chi Square !df

RMSEA

NNFI

CFI

RMR

SRMR

AG FI

Integrity / Ability

Null Model (N)

5,582.59

78

71.57

n.a.

n.a.

n.a.

n.a.

n.a.

n.a.

One Factor Model (1)

627.37

65

.001

(N-l)

4,955.23

13

***

9.65

.16

.88

.90

.08

.06

.70

Two Factor Model (2)

184.43

64

.001

(N-2)

5,398.17

14

***

2.88

.06

.97

.98

.04

.03

.92

(1-2)

442.94

1

***

Integrity / Benevolence

Null Model (N)

4,264.64

66

64.62

n.a.

n.a.

n.a.

n.a.

n.a.

n.a.

One Factor Model (1)

452.41

54

.001

(N-l)

3,812.23

12

***

8.38

.13

.88

.91

.13

.07

.80

Two Factor Model (2)

190.69

53

.001

(N-2)

4,073.95

13

***

3.60

.07

.96

.97

.08

.05

.92

(1-2)

261.72

1

***

Integrity / Willingness to rely

Null Model (N)

4,934.08

66

74.76

n.a.

n.a.

n.a.

n.a.

n.a.

n.a.

One Factor Model (1)

1,205.73

54

.001

(N-l)

3,728.35

12

***

22.33

.24

.71

.76

.19

.13

.51

Two Factor Model (2)

122.89

53

.001

(N-2)

4,811.19

13

***

2.32

.05

.98

.99

.05

.03

.94

0-2)

1,082.84

1

***

Integrity / Willingness to depend

Null Model (N)

4,204.70

55

76.45

n.a.

n.a.

n.a.

n.a.

n.a.

n.a.

One Factor Model (1)

306.68

44

.001

(N-l)

3,898.02

11

***

6.97

.11

.92

.94

.10

.05

.84

Two Factor Model (2)

106.79

43

.001

(N-2)

4,097.92

12

***

2.48

.05

.98

.99

.04

.02

.97

(1-2)

199.90

1

***

Ability / Benevolence

Null Model (N)

2,769.95

36

76.94

n.a.

n.a.

n.a.

n.a.

n.a.

n.a.

One Factor Model (1)

538.88

27

.001

(N-l)

2,231.07

9

***

19.96

.21

.75

.81

.23

.13

.63

Two Factor Model (2)

132.84

26

.001

(N-2)

2,637.11

10

***

5.11

.09

.95

.96

.10

.06

.91

(1-2)

406.04

1

***

Ability /

Willingness to rely

Null Model (N)

3,794.88

36

105.41

n.a.

n.a.

n.a.

n.a.

n.a.

n.a.

One Factor Model (1)

622.38

27

.001

(N-l)

3,172.50

9

***

23.05

.26

.79

.84

.11

.08

.52

Two Factor Model (2)

101.95

26

.001

(N-2)

3,692.93

10

***

3.92

.08

.97

.98

.03

.02

.93

d-2)

520.43

1

***

Table 31: Pairings of Sub-dimensions (Sample 4)

(Continued on next page.)

244

Chi Square Difference

Model Fit Indices

Sub-dimension

Pairing

Model

Chi

Square

if

P

„ . Chi Competing _{e}

Models „. T

Difference

if

Difference

Sign.

Chi Square

/if

RMSEA

NNFI

CFI

RMR

SRMR

AGFI

Ability / Willingness to depend

Null Model (N)

2,798.41

28

99.94

n.a.

n.a.

n.a.

n.a.

n.a.

n.a.

One Factor Model (1)

274.29

20

.001

(N-l)

2,524.13

8

***

13.71

.17

.87

.91

.07

.07

.77

Two Factor Model (2)

53.89

19

.001

(N-2)

2,744.52

9

***

2.84

.06

.98

.99

.03

.02

.95

d-2,

220.40

1

***

Benevolence / Willingness to rely

Null Model (N)

2,345.50

28

83.77

n.a.

n.a.

n.a.

n.a.

n.a.

n.a.

One Factor Model (1)

703.47

20

.001

(N-l,

1,642.03

8

***

35.17

.28

.59

.71

.35

.19

.48

Two Factor Model (2)

85.03

19

.001

(N-2,

2,260.47

9

***

4.48

.08

.96

.97

.10

.06

.93

d-2,

618.44

1

***

Benevolence / Willingness to depend

Null Model (N)

1,522.09

21

72.48

n.a.

n.a.

n.a.

n.a.

n.a.

n.a.

One Factor Model (1)

379.86

14

.001

(N-l,

1,142.23

7

***

27.13

.26

.63

.76

.21

.10

.57

Two Factor Model (2)

60.21

13

.001

_(N-2,

1,461.87

8

***

4.63

.08

.95

.97

.10

.05

.93

d-2,

319.65

1

***

Willingness to rely / Willingness to depend

Null Model (N)

2,415.06

21

115.00

n.a.

n.a.

n.a.

n.a.

n.a.

n.a.

One Factor Model (1)

347.93

14

.001

(N-l,

2,067.13

7

***

24.85

.23

.79

.86

.21

.10

.65

Two Factor Model (2)

37.52

13

.001

_(N-2,

2,377.54

8

***

2.89

.06

.98

.99

.03

.02

.96

d-2,

310.41

1

***

Table 31 (Cont.): Pairings of Sub-dimensions (Sample 4)