College Quality of Life Concept Scale

Quality of College Life (QL)

As mentioned, there are generally two popular research trends on quality of life: (1) research on factors affecting quality of life, and

(2) research measures itself. Accordingly, the way to measure quality of life also varies depending on the research trend. According to Vaez et al. (2004) and Zullig et al. (2009), quality of life is a multidimensional concept. However, most studies measuring quality of life in each area often ask specific questions related to its most important components. This is a fairly common unidimensional structure measurement trend and is applied in many aspects of life (Nguyen et al., 2012; Sirgy et al., 2007; Testa & Simonson, 1996).

Quality of university life is a part of overall quality of life, defined as the satisfaction and happiness of learners with their educational experiences during their time studying and living at school (Nguyen et al., 2012; Sirgy et al., 2007). Therefore, the author uses and adjusts the unidimensional scale of Nguyen et al. (2012) and Sirgy et al. (2007). To ensure content validity (covering the content of the concept) as well as safety in assessing the reliability of the scale (Cronbach's Alpha), this scale from three observed variables of two studies Sirgy et al. (2007) and Nguyen et al. (2012), was used by the author to use the empirical research method to adjust into four observed variables. The scale of quality of university life is as follows:

Table 3.6. Conceptual scale of University Quality of Life

Symbol

Observed variable (final)
QL1	How satisfied are you with the academic environment and life in general at school?
QL2	How satisfied are your friends and classmates (that you know) with the academic environment and life in general at school?
QL3	How happy are you with your studies at school?
QL4	How happy are your friends and classmates (that you know) with their schoolwork?

Maybe you are interested!

(Source: Authors adapted from Sirgy et al. (2007) and Nguyen et al. (2012))

3.2.2.2. Formation and adjustment of the scale

The process of forming and adjusting the scale is specifically as follows:

i) Based on theoretical research to select the measurement scale, the author discussed with experienced researchers [in the same research topic] to check and adjust the measurement scale (if any) for the concepts, so the author had the first draft (English version), which was then translated into Vietnamese with the consultation of translators with expertise in the fields of economics and business.

ii) Next, in order to adjust the first draft scale, the author continued to discuss one-on-one with experienced researchers who are also university administrators to check the content of the observed variables in Vietnamese language, to see if they are suitable for the context of Vietnamese higher education or not. After recording the comments, the author continued to adjust the draft scale to prepare for the group discussion step.

iii) Next, the author organizes a group discussion with the research subjects (students) to complete the final draft scale. This step is important because the students will be the ones to answer the author's survey, and at the same time, a group discussion with the research subjects will be better in building the scale because there is high interaction (Nguyen Dinh Tho, 2013). The author sends the draft scale to the student group that was available in step ii), each student is asked to read it carefully, then take turns giving their opinions and discussing with each other as well as with the author about unclear content. At the end of this step, the author and the scientific instructor, based on the opinions of the discussion, continue to adjust to make the content of the scales easy to understand, clear, and convey the content that needs to be measured.

Thus, at the end of the process of forming and adjusting this scale, the author has the final draft of the concept scale to move to the preliminary evaluation stage. The adjustment results through the steps of empirical research, one-on-one discussion and group discussion mentioned above and the complete Survey Questionnaire are presented by the author.

in Appendix 3.1; list of experienced researchers, and student group discussion in Appendix 3.2.

3.2.3. Preliminary assessment of the scale

3.2.3.1. Sample selection

The conceptual scales of the research were preliminarily tested by the author before conducting the official research. This testing was done through the preliminary quantitative research step. According to Hair et al. (2006), to use the factor analysis method, the minimum sample size must be 50, preferably 100, and the observation/measurement variable ratio is 5:1. With 4 independent variables, 2 dependent variables and 60 observed variables in the scale, based on the sample size calculation method mentioned above, using the convenience sampling method of students at the University of Economics Ho Chi Minh City, the author surveyed 450 copies, of which 422 copies met the requirements.

3.2.3.2. Cronbach's Alpha reliability coefficient

The first tool used to preliminarily test the above scales is to assess reliability through the variable-total correlation coefficient and Cronbach's Alpha coefficient. If an observed variable has a variable-total correlation coefficient ≥ 0.30, then that variable meets the requirements. In addition, if Cronbach's Alpha ≥ 0.60, the scale is reliable (Nunnally, 1994; Nguyen Dinh Tho, 2013). However, the Cronbach's Alpha coefficient is not the higher the better. According to Nguyen Dinh Tho (2013), if Cronbach's Alpha is too large (≥ 0.95), many variables in the scale are not different from each other, they measure the same content of the research concept, causing overlap in measurement.

3.2.3.3. Exploratory factor analysis - EFA

The second tool for preliminary testing of the scales is the exploratory factor analysis (EFA). EFA is used to assess the discriminant and convergent validity of the scales. The study uses the principal axis factoring method along with promax rotation (which will reflect the data structure more accurately than using

principal components with varimax rotation) to see if the following important properties satisfy the requirements:

- KMO test (Kaiser-Meyer-Olkin measure of sampling adequacy) > 0.50 (Nguyen Dinh Tho, 2013).

- The eigenvalue criterion to determine the number of factors stops at the factor with a minimum eigenvalue of 1 (≥ 1) (Nguyen Dinh Tho, 2013).

- The number of extracted factors is consistent with the initial hypothesis about the quantity (the scale achieves discriminant validity) (Nguyen Dinh Tho, 2013).

- The factor loading of the observed variable on the factor that the variable measures must be high and the loadings on other factors that the variable does not measure must be low (the scale achieves convergent validity). Factor loading ≥ 0.5 or the difference between two weights measuring the same observed variable > 0.3 is an acceptable value, but with a sample size larger than 350, the factor loading coefficient > 0.3 (Nguyen Dinh Tho, 2013).

- Total variance extracted (TVE) shows how much of the variation in the data can be explained by the scale. TVE ≥ 50% means that the common part must be larger than the specific part and error, if TVE ≥ 60% is good (Nguyen Dinh Tho, 2013).

3.2.4. Formal research

3.2.4.1. Official form

As mentioned in the content of Chapter 1 of the thesis, the author determined the scope of the study to be students of universities training in economics and business in Vietnam. Therefore, the official sample surveyed by the author at leading universities in economics and business in Ho Chi Minh City and Hanoi includes: 1) Ho Chi Minh City University of Economics (UEH), 2) University of Finance - Marketing (UFM),

3) University of Economics and Law, Vietnam National University, Ho Chi Minh City (UEL), 4) National Economics University (NEU), and 5) Foreign Trade University (Campus 1 - FTU).

The sampling method for the official data is non-probability, specifically the author uses the standard sampling method combined with convenience, according to criteria with attributes that have the ability to distinguish the research subjects, including: 1) training form (centralized, non-centralized); 2) gender (male, female); and 3) region (Ho Chi Minh City, Hanoi). In order to collect enough necessary observations (that can represent the crowd), the author distributes the standard sample for each criterion to at least 300 [60*5] to test the model and the hypotheses.

Next is the choice of interview format to collect data, according to Nguyen Dinh Tho (2013), face-to-face interview is the type of interview in which the researcher (or interviewer) directly interviews the research subject. This method has many advantages such as being able to stimulate responses, explain questions if they are not understood correctly; thus, the response and completion rate of the questionnaire will be the highest. Therefore, in order to increase the quantity and quality of the collected data, during the period from November to December 2018, the author made an effort to go to each class (surveyed) of the 5 universities mentioned above to directly discuss with the respondents about the purpose and meaning of the research and hoped that they would answer in the most objective way; thanks to that, the number of students participating in the response is quite large (1,520 copies). However, the author understands that errors in data collection are inevitable; Therefore, the author applied data correction techniques such as field correction and center correction to complete and eliminate invalid responses; thus, the final sample size met the requirement of 1,435 copies.

After completing the data editing step, the author continues to perform the data preparation step including data coding, data entry, and data cleaning to have a complete data set for the next steps. At this stage, the author uses SPSS software (version 23) as a processing tool.

3.2.4.2. Confirmatory factor analysis - CFA

Because the scale has been preliminarily evaluated through the Cronbach's Alpha reliability coefficient method and EFA exploratory factor analysis in the preliminary research stage.

set, so the author officially skipped this part (Nguyen Dinh Tho & Nguyen Thi Mai Trang, 2011).

Next, the research concept scales were tested by confirmatory factor analysis (CFA) using AMOS linear structure analysis software (version 20). The CFA method has many outstanding advantages in testing scales, because CFA allows testing the theoretical structure of scales without being biased by measurement errors such as the relationship between a research concept and other concepts (Steenkamp & Van Trijp, 1991). In addition, the convergent and discriminant validity of the scale can be tested through the CFA method without using many different studies (Nguyen Dinh Tho & Nguyen Thi Mai Trang, 2011). Thus, the author used CFA to:

Firstly, measure the suitability of the model (scales) with market information through the Chi-square (CMIN), Chi-square adjusted by degrees of freedom (CMIN/df), GFI (Goodness-of-Fit Index), Comparative Fit Index (CFI), TLI (Tucker and Lewis Index) and RMSEA (Root Mean Square Error Approximation) index. The model is said to be suitable (compatible) with market data when:

- The Chi-square test has a p value > 0.5, Chi-square/df ≤ 2 (Bentler & Bonett, 1980), in some cases CMIN/df can be ≤ 3 (McIver & Carmines, 1981). However, this index also depends on the sample size, the larger the index, the larger it is, meaning it does not reflect the true suitability of the model when the sample size is large Nguyen Dinh Tho and Nguyen Thi Mai Trang (2011).

- The GFI, TLI, CFI indexes ≥ 0.9 and RMSEA ≤ 0.08 (Nguyen Dinh Tho & Nguyen Thi Mai Trang, 2011). Note that sometimes the GFI index < 0.9 is also accepted in many previous studies (Hair et al., 2010), specifically this index reaches from 0.8-0.89 which is a reasonable threshold for acceptance (Deeter-Schmelz & Sojka, 2007; Doll et al., 1994; Hong et al., 2017).

Second, calculate the scale evaluation indicators including: i) composite reliability coefficient, ii) total extracted variance, iii) unidimensionality, iv) convergent validity, v) discriminant validity, and vi) theoretical correlation value. Accordingly, the theoretical correlation value indicator will be evaluated in the theoretical model (Anderson & Gerbing, 1988) and indicators 1, 2, 3, 4, and 5 are evaluated in the scale model. These indicators are considered to meet the following requirements:

- The scale meets the reliability requirements when the composite reliability coefficient (ρ C ), and extracted variance (ρ VC ) > 0.5 (Hair et al., 1998).

- The scale achieves unidimensional value when there is no correlation between the errors of the observed variables (Steenkamp & Van Trijp, 1991).

- The scale achieves convergent validity when the standardized weights are all high > 0.5 and statistically significant (p < 0.05; (Anderson & Gerbing, 1988)).

- The scale achieves discriminant validity when the overall correlation coefficient between the concepts is truly different from 1 and is statistically significant (Steenkamp & Van Trijp, 1991).

Parameter estimation in the models is applied by the ML (Maximum Likelihood Estimator) estimation method. The advantage of the distribution in this method is that it deviates less from the multivariate normal distribution, the kurtose and skewnesses are in the range [-1,+1] when testing the distribution of observed variables, therefore, ML is an appropriate estimation method (Muthén & Kaplan, 1985).

3.2.4.3. Structural Equation Modeling - SEM

The author uses the linear structural analysis method to test the research models. According to Nguyen Dinh Tho & Nguyen Thi Mai Trang (2011), the linear structural model is considered to have more advantages than other traditional methods such as multivariate regression because this model can calculate the measurement error in the research model and test the hypothesis (first-generation multivariate analysis methods often assume that the independent variables have no measurement error).

measurement, while in practice errors always appear, so this assumption is unrealistic).

Similar to the CFA method, the SEM model is used to evaluate the suitability of the theoretical model to market data through the following criteria: Chi-square (Chi-square: CMIN), chi-square adjusted by degrees of freedom (CMIN/df), comparative fit index (CFI), Tucker and Lewis index (TLI), RMSEA index. The requirements for these indexes are the same as those in the CFA model mentioned above.

The ML estimation method is also used to estimate the parameters of the model. The estimated (standardized) results of the parameters show the relationship between the variables in the model with the usual required statistical significance level of p < 0.05. In addition, based on this result, we can conclude that the scale of the concepts in the model achieves the theoretical relationship value because "each measure has a relationship with other measures as expected theoretically" (Nguyen Dinh Tho & Nguyen Thi Mai Trang, 2011).

3.2.4.4. Analysis of the role of control variables

According to Nguyen Dinh Tho (2013), a control variable is a variable that we do not focus on studying. We only want to control the extent to which it explains the variation of the dependent variable. Thus, in theory, a control variable is an alternative or complementary explanation (along with the independent variable) for the variation of the dependent variable. In terms of analytical techniques, a control variable is a type of independent variable that we will analyze separately (before or after) with other independent variables to explain.

In this study, in addition to the main objective of testing the research model with nine hypotheses on the relationships between independent variables and dependent variables, the author also tested the controlling role of gender and region on the dependent variable, which is the quality of university life. Because these two control variables are two qualitative variables, the author conducted dummy coding (or matching) with male gender as 1, female gender as 2, and Ho Chi Minh City as 1, Hanoi as 2. These two variables were included in the model.

Comment