A study on the perception of mental benefits, hedonic value, and online trust impact on electronic loyalty: A view of customer’s anxiety

and perceived mental benefits. 3.3. QUALITATIVE RESEARCH FOR SCALE ADJUSTMENT Qualitative research is a form of research that exploits the consumer's inner thoughts. Information can be gathered through observations, in-depth interview or focus groups. With the research topic, the author uses narrative research method by in-depth interviews with 11 e-commerce businesses (directors, managers, university lecturers), and the phenomenological research by creating a discussion group with 07 people including experts (lecturers, directors, managers, and customers); who make online purchases. In-depth interviews and focus group discussions will help complete the research model and complete the measurement scale as well as the observed items. Group discussions or in-depth interviews were conducted during July 2018 through group discussions with the duration of 90 to 120 minutes, the location can be done at cafes with separate rooms to discuss, or the self- study room of the university. Documents and questionnaires are sent in advance to the interviewee via email and mail. The results of the group discussion and in-depth interview room have come up with ideas that overlap with the content that builds the perceived mental benefit and electronic loyalty. In addition, the discussion group complements the anxiety, online trust and perceived hedonic value. 3.3.1. Perceived Hedonic Value Scale The perceived hedonic value has been extensively studied to better understand the consumer consumption process (Jones et al., 2006; Kronrod and Danziger, 2013). The value of perceptual pleasure is more empirical (Henry et al., 2004). In terms of consumer experience, the hedonic aspect of a consumer experience is related to the uniqueness of the product or service or emotional connection it evokes in consumers (Overby and Lee, 2006)... This study uses a perceived hedonic value consisting of 04 observed items of Lee and Wu (2017), which is applied from Babin et al. (1994) and Carpenter (2008), specifically this scale is designed for e-commerce context. The observed items were adjusted in qualitative research and used a 5-point Likert scale to measure, where 1 = Strongly disagree, and 5 = Strongly agree. 3.3.2. Online Trust Scale Trust is a multifaceted concept that combines cognitive, emotional and behavioral aspects (Lewis and Weigert, 1985). Trust has been extensively studied in many fields, but each discipline has its understanding of the concept and the different ways to operate it. Previous studies provided essential insights into the elements of online trust. However, because trust research in the online context is relatively new, some consistent issues exist across most studies (Wang and Emurian, 2005). In the literature, researchers used both unidimensional scales (Mayer et al., 1995) and multidimensional scales (McKnight et al., 2002) to measure customer trust. The multidimensional scale can reflect the different sources of belief, while the unidimensional scale reflects the assessment of trust as a whole construct. This study uses a unidimensional scale to assess the overall trust structure of customers. Five observation items are used from the scale of Liu and Tang (2018) to measure three components of the ability, benevolence, and integrity of an online business unit... The online trust scale used a 5- point Likert scale to measure, where 1 = Strongly disagree, and 5 = Strongly agree. 3.3.3. Anxiety Freud defines anxiety as a feeling state, an emotional state that includes feelings of fear, stress, worry and anxiety accompanied by physiological excitement (Freud, 2013). In line with Darwin's evolutionary perspective, Freud found that anxiety is an adaptive way of promoting behavior that helps individuals deal with threatening situations and anxiety is common in most disorders. Psychosis. When measuring anxiety, Krug et al. (1976) emphasized the importance of distinguishing between anxiety as an emotional state and anxiety as a personality trait. Ngoc's study (2018) used 14 items including seven items that measure the psychological symptoms of anxiety: anxiety, irritability, stress, fears, fatigue, poorness, insomnia. Seven items are measuring body symptoms: pain, cardiovascular symptoms, respiratory, digestive, urinary kidneys, plant nerves, and other symptoms. 17 For each interview item, there are scores from 0 to 4 that best suit the status of the object: 0 - none; 1-light; 2 - Moderate; 3 - heavy; 4 - very heavy. Psychological test scores range from 0 to 56. The higher the score, the greater the anxiety level. The anxiety level is calculated as follows: ≤ 17 points: Mild anxiety; 18 - 24: Moderate anxiety; ≥ 25: Severe anxiety. Based on previous research as well as qualitative research results, this study measures anxiety as an emotional state when shopping online and uses six items to measure Hamilton's anxiety symptoms (1959), this scale was also applied by Ngoc (2018); At the same time, the study also calculated the anxiety level based on 06 observed variables with psychological test scores from 0 to 24. The classification results were recalculated based on the percentage of anxiety scores and scales, which is shown in table 3.2. Qualitative research results also adjusted and changed perceptions about the measurement aspects of anxiety symptoms. The context of the scales is all about the everyday worries in online shopping, e-commerce in Vietnam. The variables measure customer anxiety in using a 5-point scale to measure, where 0 - none; 1-light; 2 - Moderate; 3 - heavy; 4 - very heavy. Table 3.2. The classification of anxiety levels based on a scale of 14 and 6 observed items 14 items (0 - 56) 6 items (0 - 24) Mild anxiety <=17 <=7,29 Moderate anxiety 18 - 24 7,3 - 10,7 Severe anxiety >=25 >=10,71 3.4. QUANTITATIVE RESEARCH 3.4.1. Questionnaire The Likert scale is often used to measure a set of statements of a concept (Likert, 1932). The measure of the concept is the sum of each statement. Theoretically, the Likert scale is ordinal and measures the level of agreement of the research object. That is, the five points vary from full disagree to full agreement. Therefore, the Likert scale allows a range of options from "Strongly Disagree" to "Strongly Agree." This gives the respondent the ability to make useful distinctions between attitudes. The anxiety scale is based on Hamilton's (1959) study so it will be 0 - no; 1- mild; 2 - moderate; 3 - heavy; 4 - cumbersome. According to each group of factors constituting the perceived components of online shoppers with general information presented first, then it goes to each in-depth exploration question about that element. The nominal scale is a scale in which the number is used to classify; it has no meaning in quantity and uses to describe the sample (occupation, gender ...) (Nguyen Dinh Tho, 2011). The form of an ordinal scale is a type of scale in which the number is used to compare order; it has no meaning in quantity (Nguyen Dinh Tho, 2011). Ph.D. student uses this scale to organize the sample characteristics (age, income ...). 3.4.2. Sample size According to Hair et al. (2010), the size of samples must be at least five times the number of observed variables to ensure accuracy. Tabachnick and Fidell (2007) suggest that the sample sizes will yield corresponding results: 50 is very poor, 100 is poor, 200 is quite good, 300 is good, 500 is very good, and 1000 is excellent. The study used 917 observations in the analysis after eliminating unregulated observations or insufficient confidence. 3.4.3. Sampling method To achieve the research objectives of the topic, the sampling technique in this study is to choose non-probability samples, quota sampling techniques and purpose (judgemental). Because the research is in-depth in the field of e-commerce, the purpose of sampling is to select the survey subjects who understand the research well and can expand the sample quickly (Neuman, 2002). ). Specifically, with the highest e-commerce index 2018 of 82.1 and TP. Ho Chi Minh City is located in the south of Vietnam with a population of over 10 million people. The people living here mainly migrate from other provinces of Vietnam. Therefore, the characteristics of the population in the city. Ho Chi Minh may represent the country's population characteristics, the number of observations in the city. Ho Chi Minh accounted for about 4/10 observations in the sample. Also, Hanoi city has a second-ranking index of 79.8, so it will survey about 3/10 of the sample number. Other cities and 18 provinces of Hai Phong (index of 54.9), Da Nang (54.1) and Binh Duong (50.4) will allocate each province and city one-tenth respectively in the sample size. Survey subjects are individual consumer customers who have purchased online in Vietnam. 3.4.4. Data analyze method To achieve the proposed research objectives, the study uses SPSS, AMOS, and Smart PLS software to process quantitative data collected from preliminary and official surveys. In particular, SPSS software for descriptive statistics, exploratory factor analysis (EFA) and reliability of scales in preliminary research; AMOS software is used to carry out tests in developing a perceived benefit and electronic loyalty. Finally, Smart PLS software will be used to analyze research models and test hypotheses. 3.5. PRELIMINARY RESEARCH FOR RELIABILITY EVALUATION The preliminary research program for evaluation of scales was conducted using quantitative research methods. The study used convenient sampling methods; the total number of votes issued was 260; the total number of a valid questionnaire put into the analysis was 252. The preliminary research program conducted from October 2018 to December 2018. Respondents are individual customers who have purchased online on e-commerce sites in Vietnam. Survey subjects are mainly pupils, students, lecturers and office workers. The preliminary survey program was conducted by a direct approach and hand-filled questionnaire distribution. Cronbach’s alpha coefficient is most widely used to measure the reliability of the scale (internal consistency). Cronbach's alpha greater than 0.7 is the lower limit (Hair et al., 2010). According to the results of assessing the reliability of the scale, Cronbach's alpha is the reliability coefficient of the total scale of all scales is more significant than 0.7. Moreover, when testing each measurement item, the correlation coefficient - the total is more significant than 0.3, so all items meet the requirements (Nunnally, 1978). Measurements that have been rated to meet the reliability requirements will be included in the evaluation of the scale value by EFA. Firstly, to consider the possible data conditions for EFA analysis, it is necessary to rely on the Barlett test and KMO test (Kaiser-Meyer-Olkin). We have suitable EFA analysis because Bartlett test has p = 0.00 <5% and KMO = 0.928 ≥ 0.5. Second, to evaluate the value of the scale, three important attributes should be considered in the EFA analysis results: (1) the number of factors extracted, (2) factor weight, and (3) total variance deduction. The number of extracted elements: Eigenvalue is used to determine the number of extracted factors, the number of extracted elements is ten factors when the Eigenvalue factor in the 10th factor is 1,075 > 1 (Nguyen Dinh Tho, 2013). Factor load factor: is greater than or equal to 0.5 which is an acceptable value. Total variance Extracted: this factor shows how many percent of the measurement variables are extracted. The total variance extracted by ten factors is 68,105> 50%, meaning that the standard part is larger than the individual part and the error. 3.6. CONCLUSION OF SCALE USED FOR OFFICIAL RESEARCH After conducting a preliminary survey of 260 online customers using the convenient sampling method, 252 enough votes were included in the analysis. Testing the scale using Cronbach’s Alpha coefficients and discovering factor analysis was performed to evaluate the reliability and value of the scale. As a result, the scales that meet the reliability and validity are aggregated to conduct a formal study. The calibrated scales after the preliminary study are detailed in Table 3.3. Table 3.3. Summary of the official scale Code Content Perceived Shopping Enjoyment (PEB) PEB1 Shopping online makes me feel alive in my own world PEB2 When I am in a depressed mood, shopping online will help me feel better PEB3 Shopping online helps me satisfy myself 19 Code Content PEB4 Shopping online is an adventure PEB5 For me, online shopping is a way to reduce stress PEB6 Compared to other things I can do, online shopping time is enjoyable. Perceived Discreet Shopping (PDB) PDB1 Online shopping ensures privacy on the buying process PDB2 I do not mind not buying anything after asking for product information on e-commerce. PDB3 When shopping online, I do not feel embarrassed if I buy sensitive goods/services. PDB4 I feel free to search for product/service information when shopping online without anyone knowing. PDB5 I do not feel afraid to buy discounted products/services on e-commerce sites. Perceived Social Interaction (PSB) PSB1 I feel connected to other people when shopping online PSB2 I feel there is an exchange of information about products/services when I share my shopping experience with other consumers when shopping online. PSB3 Shopping online is a great way to develop friendships with other internet shoppers. PSB4 I enjoy viewing and interacting with other customers' opinions about products/services that I plan to buy Perceived Control (PCB) PCB1 The e-commerce site allows me to control my online shopping process PCB2 I have the option to participate, express myself and leave my own mark when shopping online PCB3 In addition to shopping, I also participate in the entire consumer experience by collaborating, posting ideas, and/or participating in a part of product creation. PCB4 The e-commerce site designs everything that offers ideas to connect and chooses various shopping options according to my wishes. Perceived Hedonic Value (HV) HV1 Shopping on this e-commerce site is a pleasure. HV2 Shopping on this e-commerce site, I not only buy products but also feel happy. HV3 Shopping on this e-commerce site makes me feel that the product search is entertaining. HV4 Shopping on this e-commerce site allows me to forget all my troubles and discomfort. Online Trust (OT) OT1 This electronic seller is honest. OT2 This electronics seller is very interested in meeting my needs and wants OT3 This electronic seller keeps his promises and commitments OT4 This electronic seller is trusted OT5 This electronic salesperson is capable of fulfilling my needs and desires Personal Information Disclosure (PID) PID1 I am willing to submit personal information to e-commerce sites that are not related to shipping (gender, age, income, ..) PID2 I proactively provide even sensitive personal information like my interests to e- commerce sites PID3 I share personal opinion for e-commerce site PID4 I agree to allow the mobile e-commerce app to access my phone contacts or my friends' list on my social networking site. Interaction (INT) INT1 When I like an e-commerce post, I will press the "like" button. INT2 I evaluate and rank articles/products on e-commerce sites 20 Code Content INT3 I participated in sharing some content on the e-commerce site INT4 I accept the time it takes to make comments, even though the e-commerce site does not pay me any compensation. Preference (TOM) TOM1 I use the e-commerce site whenever I need to make a purchase. TOM2 When I intend to buy, this e-commerce site is my first choice. TOM3 Always mention this e-commerce site when friends and relatives need online shopping advice Anxiety (ANX) ANX1 I am worried that the product will not match the description on the e-commerce site ANX2 I feel nervous when ordering new products on e-commerce sites ANX3 I fear my information will be used illegally ANX4 I find it hard to sleep while waiting for the product to arrive ANX5 I could not focus on working while waiting for the product to arrive ANX6 I lost interest in online shopping after receiving the wrong products, or late delivery 21 CHAPTER 4. ANALYZE THE RESEARCH RESULT 4.1. ANALYZE THE RESEARCH SAMPLE A total of 468 questionnaires were collected from 500 questionnaires. After eliminating 51 invalid and unqualified questionnaires (43 questionnaires lacking information; 07 questionnaires selecting more than one answer for the 1-choice-only question type; 01 feeble questionnaire by choosing the same option for most questions), there are 417 valid questionnaires. Combined with 500 results from the online survey, 917 sample survey results were used for analysis and verification. 4.2. ASSESSMENT OF THE PERCEIVED MENTAL BENEFITS CONCEPT The constructs of perceived mental benefits in e-commerce include four potential research variables (PEB, PDB, PCB and PSB). For these constructs, we need to estimate the relationship between the latent variable and its observed variables (outer loadings). All Outer loadings of PEB, PDB, PCB and PSB are higher than the allowed value of 0.708. The observed variable PEB5 has the lowest variable reliability of 0.558 = 0.7472 (outer loading is 0.747), while the PDB1 variable has the highest reliability of 0.852 = 0.9232 (outer loading is 0.923). Therefore, all observed variables for the 04 reflective constructs (reflective constructs) are greater than the minimum allowed values of the outer loadings. The internal consistency reliability of the latent variables is assessed through composite reliability (CR). All CR values are greater than 0.7. From this, we conclude that the latent variables have achieved internal consistency reliability. In addition, the AVE values of the research concepts are > 0.5 (PCB = 0.717; PDB = 0.8; PEB = 0.605 and PSB = 0.723). Therefore, all four research concepts have high convergence. In the end, the research examines the discriminant value among the latent variables in the research. HTMT values for all pairs of variables are studied in a matrix. It can be seen that all HTMT values are much smaller than 0.85. Besides, the lower and upper limit of the 95% confidence level (the difference between the calibration and cumulative) of all groups of research concepts do not include the number 1. This proves the constructs of perceived mental benefits of online customer achieves discriminant value. Thus, the scale of perceived mental benefits includes four latent research variables including PEB, PDB, PSB, PCB. All groups of variables achieved positive values, convergent values and discriminant values. 4.3. ASSESSMENT OF THE ELECTRONIC LOYALTY CONCEPT The construct of electronic loyalty in e-commerce includes three latent research variables (TOM, INT, and PID). For these constructs, we need to estimate the relationship between the latent variable and its observed variables (outer loadings). All Outer loadings of TOM, INT and PID are higher than the allowed value of 0.708. The observed variable INT4 has the lowest variable reliability of 0.592 = 0.772 (outer loading is 0.77), while the variable PID1 has the highest reliability of 0.774 = 0.882 (outer loading is 0.88). Therefore, all observed variables for the 03 reflective constructs (reflective constructs) are greater than the minimum allowed values for the outer loadings. The internal consistency reliability of the latent variables is assessed through composite reliability (CR). All CR values are greater than 0.7. From this, we conclude that the latent variables have achieved internal consistency reliability. In addition, the AVE values of the research concepts are > 0.5 (INT = 0.611; PID = 0.669; and TOM = 0.733). Therefore, all three research concepts have high convergence. Finally, the research examines the discriminant value among the latent variables in the research. Heterotrait - monotrait (HTMT) correlation values for all pairs of variables are studied in a matrix. It can be seen that all values of HTMT are much smaller than 0.85. Besides, the lower and upper limit of the 95% confidence level (the difference between calibration and cumulative) of all groups of research concepts does not include the number 1. This proves that the construct of the electronic loyalty of customers gains discriminant value. Thus, electronic loyalty includes three potential research concepts including PID, INT and TOM. All groups of variables achieved reliable values, convergent values and discriminant values. 22 4.4. THE TEST OF RESEARCH MODEL AND RESEARCH HYPOTHESES 4.4.1. Assess the internal reliability of the research constructs of the research model The research model in e-commerce includes nine research concepts (PEB, PDB, PSB, PCB, OT, HV, TOM, INT, and PID). For this model, we need to estimate the relationship between the study variable and its observed variables (outer loadings). All Outer loadings of PEB, PDB, PSB, PCB, OT, HV, TOM, INT, and PID concepts are higher than the allowed value of 0.708. The observed variable PEB5 has the lowest variable reliability of 0.558 = 0.7472 (outer loading is 0.747), while HV3 variable has the highest reliability of 0.862 = 0.9292 (outer loading is 0.929) . Therefore, all observed variables of the nine reflective constructs are greater than the minimum allowed values for the outer loadings. The internal consistency reliability of the latent variables is assessed through composite reliability (CR). All CR values of the research structures are greater than 0.7. From this, we conclude that the latent variables have achieved internal consistency reliability. In addition, the AVE values of the study variables are > 0.5 (PEB = 0.605; PDB = 0.8, PSB = 0.723; PCB = 0.717; OT = 0.673; HV = 0.791; INT = 0.611; PID = 0.669; and TOM = 0.733). Therefore, all 09 research concepts have high convergence. 4.4.2. Assess the discriminant validity among research constructs in the research model Next, the study examines the discriminant validity among the constructs in the study. HTMT values for all pairs of research structures in a matrix. It can be seen that all values of environmental education are much smaller than the threshold of 0.85. Besides, the low and high thresholds of 95% confidence level (the difference between correction and accumulation) of all the study variables groups do not contain the number 1. This proves that all the constructs in research achieved discriminant validity. Thus, the nine research structures PEB, PDB, PSB, PCB, OT, HV, TOM, INT, and PID all achieved reliability, convergent validity and discriminant validity. 4.4.3. Assess the collinearity of the model We have VIF values of all endogenous variables (shown in the column) and corresponding exogenous variables (shown in rows). Correctly, the study will evaluate the following set (forecast) of the variable. Research on multicollinearity PMB, HV, OT -> ELOY. According to the results of the study, the VIF values are all less than 5, so the collinearity between predictive variables does not occur in the research model... 4.4.4. Coefficient of Determination (R2) The measurement factor most used to evaluate a structural model is the Coefficient of Determination (R2 value). The higher the R2 value, the more accurate the forecast level. Similarly, multiple regression, adjusted R2 values (R2adj) can be used as a standard to avoid bias for complex models. We have ELOY adjusted R2 and R2 coefficients (0,403/0,401) and HV (0,375/0,374) are quite weak and OT (0,511/0,510) are average. However, an R2 value of 0.20 is considered high in areas such as consumer behavior (Hair et al., 2016). Therefore, the relationships in the research model have a consistent level of interpretation of electronic loyalty, online trust, and perceived hedonic value. 4.4.5. Effect size (f2) In addition to evaluating adjusted R2 and R2 values of all endogenous variables, changes in R2 values when a specific exogenous variable is omitted from the model can be used to evaluate whether the ignored variable has a significant impact on an endogenous variable. This measurement is called the Effect size (f2). In the relationship between PMB, HV, OT, and ELOY, the impact coefficient f2 evaluate the contribu