Comparación de las medias de metilación global

There have been two phases of data collecting processes performed: pilot test and the main study. The processes and the results of the pilot test and the main study are explained later on in the separate parts, Chapter 5 and Chapter 6. Briefly, a Pilot Test was intended to determine whether the items in the questionnaire have met the ap-propriate research standards or measurement evaluating procedures. This implies that before using the instrument or questionnaire, we need to ensure that indica-tors that we are using to measure a concept can work in an accurate and consistent manner. This prerequisite calls for validity (accuracy) and reliability (consistency) tests.

There are two forms of validity tests that are frequently mentioned in the re-search literatures: external and internal validity. The external validity of rere-search

findings is the data’s ability to be generalized across persons, settings, and times;

while internal validity confirms the ability of a research instrument to measure what it is purported to measure (Cooper & Schindler, 2008). Validity is the extent to which a construct in the questionnaire is able to measure what is supposed to measure (Hair Jr. et al., 2007, p. 246). According to Ghozali (2006), there are three methods can be applied in measuring validity, first, either by correlating the item score with the total score of a construct or variable that more known as internal consistency reliability. One way to accomplish this technique is by looking at Cronbach’s alpha output in the Correlated Item - Total Correlation column. Second way is by using Pearson Bivariate Correlation to see the correlation between each indicator score and the total score of the construct. Under the same items and concept, the result of Bivariate Correlation analysis generates similar result as we could find at Cron-bach’s alpha output in the Correlated Item - Total Correlation column since they implement the similar objective. However, both the first and second techniques can only be applied uni-dimensional concept. In the context of multi-dimensional con-cept, the third method, Confirmatory Factor Analysis test is needed. In validating all items under investigation, I used two different types of validity testing methods, which are the first one: Cronbach’s alpha and Confirmatory Factor Analysis. The arguments of selecting these two approaches are presented in the following expla-nations.

As I have mentioned previously, the attitude towards green food products con-struct in this study is adopted from Tanner & Kast (2003). The attitude towards green food products is constituted by six dimensions or multi-dimensions: environ-mental protection, genetically engineered foods, fair trade, regional product, health, and taste. Notwithstanding that Tanner & Kast did not particularly elucidate a com-prehensive explanation about this, I assumed that direction of causality for attitude construct is from indicators or measures to construct. The direction signifies that changes in the indicators will cause changes in the underlying construct. One ex-ample is if in the first time, consumers did not know, but after some times, they found that green foods are perceived healthier, this expression could be expected will change their attitude towards green food products. In general, I suppose that indicators are mutually exclusive and all have an influence on the attitude con-struct. The pattern and characteristics of this relationship refers to what Bolen &

Lennox (1991) have described as “composite latent construct model”. This notion was echoed by Jarvis et al. ( 2003, p. 201), who noted that if the flow of direction goes from indicators to construct, this type of model is defined as “formative” model. For this model, internal consistency reliability is not the appropriate standard for evalu-ating the adequacy of the measures. Hence, I used Confirmatory Factor Analysis for the validation of items under all six dimensions. On the top of that, in identifying

or extracting the number of underlying factors or dimensions, I used the Principal Component Analysis technique. For this purpose, I employed the most common used approaches in assisting in the decision concerning the number of factors to retain: Kaiser-Meyer-Olkin Measure of Sampling Adequacy (thereafter, it is called

“KMO”) and the Barlett’s Test of Sphericity value. The Kaiser-Meyer-Olkin test is a measure of sampling adequacy that compares the magnitudes of the calculated correlation coefficients to the magnitudes of the partial correlation coefficients (Pett et al., 2003, p. 77).

The KMO criterion ranges from 0 to 1, with small value indicating that the sum of the squared correlation coefficients is small relative to the sum of the squared partial correlation coefficients and therefore a factor analysis may be unwise. Oth-erwise, the larger value of KMO is more acceptable and appropriate to execute fur-ther analysis: factor analysis. When evaluating the size of the overall KMO, Kaiser (1974) suggests using the following criteria for these values: 1) Above .90 is “mar-velous”, 2) In the .80s is “meritorious”, 3) In the .70s is “middling”, 4) Less than .60 is “mediocre”, “miserable”, or “unacceptable” (in Pett et al., 2003, p. 35). Neverthe-less, Ghozali (2006) argued that the value expected of KMO should be>.50 in order to establish factor analysis.

Once the number of factors has been determined, the next stage is to interpret them. To assist in this process, the factors need to be ’rotated’. There are two main techniques of rotation: Orthogonal (e.g., Varimax, Quartimax, Equamax) or Oblique (e.g., Direct Oblimin and Promax) factor solutions (Pallant, 2007, p. 183). This study used the Varimax method, which attempts to minimize the number of variables that have high loadings on each factor. The goal of Varimax is to simplify the columns of the unrotated factor-loading matrix. To accomplish this goal, Varimax maximizes the variances of the loadings within the factors while also maximizing differences between the high and low loadings on a particular factor (Pallant, 2007, p. 141).

Using this rule, only factors with an eigenvalue of≥1 are retained for further evalu-ation. The eigenvalue of a factor depicts the amount of the total variance explained by that factor (Pallant, 2007, p. 183). For the further analysis, Comrey & Lee (1992) have provided some guidelines for assessing factor loadings. Table 4.3 summarizes the guidelines. Comrey & Lee argue that factor loadings with “very good” to “ex-cellent” category can be sufficiently helpful in explaining about the factor.

However, unlike the attitude, the items of subjective norms, perceived behav-ioral control, perceived difficulty, purchase intention and actual behavior were val-idated through the second method: by looking at the Correlated Item - Total Corre-lation in the Cronbach’s alpha output. This analysis can be undertaken by looking at the values of corrected item - total correlation (r-value) with the r-table. I used r-value and r-table comparison as the standard of eliminating the items. More

com-TABLE4.3: Scale of variable-factor correlations

Orthogonal Factor loading Percentage of Variance Category

.32 10 Poor

.45 20 Fair

.55 30 Good

.63 40 Very Good

.71 50 Excellent

Source: Comrey & Lee ( 1992, p. 243)

prehensive explanations for this process are discussed in the Chapter 5.