CHAPTER 3 – METHODOLOGY
3.5 Analytical Methodology
3.5.1 Date detection and descriptive statistics
Data collection took place through a paper and an online survey. Two ways of approach mitigates coverage errors and other biases resulting from data collection methods. Using two sample t-test to tested whether have reaction biases problem in different ways of survey. Therefore, it can understand two different ways of survey whether have differential.
This study used frequency distribution and percentage to execute basic information analysis. It can understand the situation of information distribution. It’s conclude gender, age, education degree, career, frequency of consume product categories, frequency of consume and income. Mean and standard deviations are chosen as the descriptive statistics to analyses each answers for the questionnaire. Among mean can generally be used to gather opinion of the respondents. It may also be further enhanced by making a comparison amongst the respondents who favor the same answer. The higher the result is, the more favorable to the answer. And standard deviation may also be used as an alternative to
40
understand the opinion of the respondents. The lower the deviations, the closer it is to the expected response.
3.5.2 Reliability
The reliability is trustworthiness of measures. Also, it is the degree of consistency or stability in the measures. A measure is considered reliable if it would give us the same result over and over again. This study is used Cronbach’s Alpha and composite reliability to test consistency and stability in questionnaire.
3.5.2.1 Construct Reliability
Cronbach’s alpha is between 0 and 1. According to Guielford (1965) suggestion, when Cronbach’s alpha is greater than 0.7, it shows the questionnaire has relative high internal reliability; between 0.7 and 0.35 means medium; and below than 0.35, it has low internal reliability.
3.5.2.2 Composite Reliability, CR
Composite reliability is a reliability index for measure or test potential variables. The function of CR value lies in measuring the composition quality of construct reliability. Fornell and Larcker (1981) mention that CR value is greater than 0.6, it means have good composite reliability.
When CR value is high, it shows between index has relative high internal relatively.
3.5.3 Validity
The validity is defined as testing the distinguishing of measure or the accuracy of behavior. Generally, the index of measuring validityis content
41
validity, convergent validity and discriminant validity.
3.5.3.1 Content Validity
The content validity of a measuring instrument is the extent to which it provides adequate coverage of the investigative questions guiding the study. Content validity emphasize that the breadth of content, the coverage of content and the richness of content (邱皓政, 2006). It means test have reasonable content validity that content comes from theoretical foundation, empirical research, logical reasoning and expert consensus.
3.5.3.2 Convergent Validity
The convergent validity tests are aimed at verifying whether answers from different individuals to question-statements are sufficiently correlated with the respective latent variables.
3.5.3.3 Discriminant Validity
Fornell and Larcker (1981) present a method for assessing thediscriminant validity of two or more factors. Here, a researcher compares the AVE of each construct with the shared variance between constructs. When the discriminant validity is supported, the AVE for each construct is greater than its shared variance with any other construct (Hair, Anderson, Tatham and Black, 1998).
3.5.4 Structural Equation Model, SEM
Structural equation model implies a structure for the co-variances between observed variables. Also, researchers refer to structural equation models as LISREL models. Hair et al. (1998) shows that estimated the
42
index of overall fit are Absolute Fit Measures, Incremental Fit Measures and Parsimonious Fit Measures. This study used AMOS to execute SEM analysis.
3.5.4.1 Absolute Fit Measure (1) Normed Chi-Square, NC
Hayduk (1987) define that NC value below 3, it means good fitness. On the other hand, Bollen (1989) definition that NC value should below 5, it is better for overall fitness.
(2) Goodness of Fit Index, GFI
When GFI value between 0 and 1 that value more close to 1, it means the model is better in fitness. Doll, Xia and Torkzadeh (1994) pointed out if GFI value is greater than 0.9, it means good fitness, but between 0.80 and 0.89 is reasonable fitness.
(3) Adjusted Goodness of Fit Index, AGFI
Both GFI and AGFI have standardization, the value between 0 and 1. If more close to 1, it means the model is better in fitness. Doll et al. (1994) indicated that the value of AGFI is greater than 0.9, it means the model is better in fitness. As if between 0.80 and 0.89 is reasonable fitness.
(4) Standardized Root Mean Square Residual, SRMR
SRMR value is between 0 and 1 that more close to 0 means the model is better in fitness. When the value below 0.08, it means the model is better in fitness(Hu & Bentler, 1999; 邱皓政, 2003).
43
(5) Root Mean Square Error of Approximation, RMSEA
Jarvenpaa, Tractinsky and Vitale (2000) mention that when RMSEA value below 0.08 is acceptable. McDonald and Ho (2002) suggest that 0.05 is good fit and 0.08 is acceptable.3.5.4.2 Incremental Fix Measures (1) Normed Fix Index, NFI
The NFI value is between 0 and 1. Bentler and Bonett (1980) and Doll et al. (1994) pointed out that NFI value have greater than 0.9 means good mode.
(2) Incremental Fix Index, IFI
The IFI value is between 0 and 1. When value is more close to 1, it means have good fitness. If IFI value us equal to 1, it means the data match mode completely. Whether is acceptable in judging mode, IFI value must greater than 0.9 means ideal fitness.
(3) Non-Normed Fix Index, NNFI
At beginning, NNFI was called Tucker-Lewin Index (TLI) for use in Exploratory Factor Analysis, then it explores to SEM. Generally, the NNFI value must greater than 0.9, it means have ideal fitness (Bentler &
Bonett, 1980).
(4) Comparative-Fix Index, CFI
44
The comparative-fix index is provided from Bentler (1990), the purpose is overcome the disadvantage of NFI. According to Bentler (1990) and Hair et al. (1998), suggest that CFI value is greater than 0.9 means acceptable in fitness.
3.5.4.3 Parsimonious Fit Measures
(1) Parsimonious Goodness of Fix Index, PGFI
The PGFI value is between 0 and 1. The value is more bigger means better fitness. Mulaik et al. (1989) claim that have greater than 0.5 is possible.
(2) Parsimonious Normed Fix Index, PNFI
The PNFI is mainly using in compare different mode and purpose on revise NFI. The higher the value is better. 黃芳銘(2007) suggest that the mode fitness greater than 0.5 as standard of pass.
45