• 검색 결과가 없습니다.

B. Measurement Model Analysis

3. Validity Analysis

Validity denotes how accurately latent constructs measure their concepts or attributes and how precisely the test tool developed for the measurement reflects the corresponding attribute (Kang, 2002). And, construct validity gets recognition by testing statistically whether an operationalized instrument actually gauges an abstract notion of a construct that it purports to measure (Straub, 1989).

According to Lewis and associates (2005), each of the items in an instrument should be representative of the construct and comprehensively cover all aspects of the construct. Using the construct validity test, this study was to investigate whether the measurement instrument extracted by the researcher condignly estimated the research field. For the construct validity to be confirmed, the measurement variables in the present study were checked in two types of validity: convergent and discriminant validity.

a. Convergent Validity Analysis

First, convergent validity signifies the extent to which multiple items for a potential concept are consistent. In estimating an identical concept, the values

estimated by different measures must have high correlation. To test the convergent validity, confirmatory factor analysis (CFA) was deployed using principal axis factoring and a varimax technique of orthogonal rotation. And the validity testing was identified by factor loadings of the measurement variables.

According to Steenkamp and van Trijip (1991), CFA is a technique that after establishing a specific hypothesis, analyzes and explains observable relationships in data, and regarded as highly useful to test validity of measures estimating a latent concept. There are several indicators in this authorized analysis method.

One of them is Construct Reliability (CR) referring to values of factor loadings, which should equal to or exceed 0.7 (Hair et al., 2010; Song, 2013). Another crucial indicator is Average Variance Extracted (AVE) meaning size or degree of attributes‘ variance, whose over 0.5 value is considered acceptable (Anderson &

Gerbing, 1988; Chin, 1998; Fornell & Larcker, 1981; Hair et al., 2010; Song, 2013). Also, standardized solutions and measurement errors should be consulted as well. Table 10 shows the results of convergent validity analysis.

As displayed in Table 10 below, all the values of factor loading exceed the threshold (0.7), ranging from 0.702 to 0.889. In addition, AVE and CR for all latent variables exceed the recommend value (0.5 and 0.7, respectively). The volume of AVE refers to sufficient values: motivation (0.626), anxiety (0.621), attitude (0.578), perceived ease of use (0.586), perceived usefulness (0.594), perceived interactivity (0.611), and variety (0.734). And the CR scores of individual variables are loaded at 0.9321 for motivation, 0.920 for anxiety, 0.872 for attitude, 0.850 for perceived ease of use, 0.880 for perceived usefulness, 0.886 for perceived interactivity, and lastly 0.892 for variety. Since all of the standardized parameter estimates of all indicators on their constructs achieved remarkably higher than the recommended standards, it can support the evidence that the factor structure of the observed variables is well-constructed. It means that the measurement variables are highly relevant to their respective latent constructs. Therefore, it is concluded that the convergent validity of the measurement model is guaranteed in the constructs.

Latent

Variable Item Std.

Error

Factor Loading

t-value (Critical Ratio)

Motivation AVE = 0.626 CR = 0.921

MOT1 0.107 0.749 16.036

MOT2 0.044 0.739 10.640

MOT3 0.069 0.808 11.108

MOT4 0.036 0.772 12.365

MOT5 0.095 0.871 9.691

MOT6 0.084 0.747 11.166

MOT7 0.059 0.843 9.455

Anxiety AVE = 0.621 CR = 0.920

ANX1 0.054 0.775 12.477

ANX2 0.061 0.842 10.008

ANX3 0.007 0.738 8.275

ANX4 0.079 0.780 10.600

ANX5 0.145 0.746 8.035

ANX6 0.002 0.734 10.682

ANX7 0.064 0.889 9.318

Attitude AVE = 0.578 CR = 0.872

ATT1 0.091 0.716 11.138

ATT2 0.071 0.739 14.166

ATT3 0.049 0.783 11.478

ATT4 0.083 0.808 12.566

ATT5 0.093 0.752 10.280

Perceived Ease of Use

AVE = 0.586 CR = 0.850

PEOU1 0.086 0.748 11.651

PEOU2 Removed

PEOU3 0.123 0.805 13.289

PEOU4 0.105 0.716 14.368

PEOU5 0.159 0.790 12.410

Perceived Usefulness

AVE = 0.594 CR = 0.880

PU1 0.091 0.792 17.108

PU2 0.099 0.757 14.455

PU3 0.117 0.807 10.815

PU4 0.143 0.765 17.811

PU5 0.124 0.731 12.142

Perceived Interactivity

AVE = 0.611 CR = 0.886

PI1 0.104 0.857 13.671

PI2 0.097 0.768 9.195

PI3 0.064 0.702 7.970

PI4 0.136 0.825 11.099

PI5 0.125 0.745 15.786

Variety AVE = 0.734 CR = 0.892

VAR1 0.136 0.868 9.660

VAR2 0.129 0.859 10.222

VAR3 0.095 0.843 9.451

VAR4 Removed

VAR5 Removed

[Table 10] Results of Convergent Validity Analysis

Note. MOT: Motivation; ANX: Anxiety; ATT: Attitude; PEOU: Perceived Ease of Use, PU;

Perceived Usefulness; PI: Perceived Interactivity; VAR: Variety.

b. Discriminant Validity Analysis

Second, discriminant validity signifies one construct should be empirically differentiated from other constructs that may be similar; i.e., the lack of a relationship among latent variables that theoretically should not be related. Thus, it is defined as the distinctive degree to which two similar constructs in their concepts are differentiated (Hair et al., 2010). To this end, the estimates between different constructs must be distinguished, and excessive overlap among different constructs must not exist. In order to demonstrate the discriminant validity, the correlation coefficient (

r

) among latent variables in the horizontal and vertical line should not exceed the square root of AVE (Chin, 1998; Fornell & Larcker, 1981). The bold figures on diagonal of the matrix, shows the square root of AVE, and the bottom shows the correlations. The significance level was set at 0.05 (

p

< 0.05) to test the null hypotheses of no association. Table 11 presents the matrix showing the results of discriminant validity analysis.

Latent

Variable MOT ANX ATT PEOU PU PI VAR

MOT 0.791      

ANX .416** 0.788      

ATT .361** .254** 0.760        

PEOU .288** .284** .245** 0.766      

PU .401** .380** .342** .326** 0.791    

PI .456** .287** .155** .261** .397** 0.782  

VAR .204* .190* .238** .256** .336** .249** 0.857

M 3.29 3.15 3.43 3.58 3.33 3.70 3.02

SD 0.627 0.743 0.641 0.720 0.600 0.622 0.763

[Table 11] Results of Discriminant Validity Analysis

Note. MOT: Motivation; ANX: Anxiety; ATT: Attitude; PEOU: Perceived Ease of Use, PU;

Perceived Usefulness; PI: Perceived Interactivity; VAR: Variety. p<0.05:*, p<0.01:**

As indicated on the matrix, the correlation coefficients (

r

) vary from the lowest at .155 (between perceived interactivity and attention) to the highest at .456 (between perceived interactivity and motivation). The relationships between variety and motivation (.204) and between variety and anxiety (.190) indicate the

p

< 0.05 level of significance; but for two, all the other correlations are highly significant at the

p

< 0.01 level, demonstrating their appropriateness of relationships. In addition, the values on diagonal (the square root of AVE) among observed variables are found to surpass all the correlations (

r

) in any case, implying that the discriminant validity was demonstrated. Hence, it is confirmed that all the latent variables―including motivation, anxiety, attention, perceived ease of use, perceived usefulness, perceived interactivity, variety, and achievement―are well established.

관련 문서