Causes and Behavioral Evolution of Negative Electronic Word-of-Mouth Communication: Considering the Mediating Role of User Involvement and the Moderating Role of User Self-Construal

Abstract

The purpose of this study is to develop a framework to examine the intrinsic driving paths of negative electronic word-of-word. In this paper, the “contextual” factor—user involvement, and the “individual” factor—user self-construal were selected to consider their influence on the model path. Data were collected using online questionnaires; then, the model and hypotheses were tested using structured equation model (SEM) software. The research results showed that, firstly, negative online shopping experiences positively influence eWOM motivations, and user involvement partially mediated the relationship between negative product quality, negative online shopping platform environments and negative eWOM motivations; secondly, there was a significant correlation between negative eWOM motivations and eWOM behavior, and the personality traits of the communicators themselves influenced the path of “negative eWOM motivation-negative eWOM behavior”. In addition, the user’s self-construal has a significant moderating effect on the “eWOM motivation–immediate eWOM behavior” path under immediate eWOM behavior.

1. Introduction

WOM is a type of communication between consumers concerning a product, service or company that occurs between a communicator and a recipient, who perceives the information received as noncommercial [1]. These interpersonal communications provide information beyond what is provided by the company and that involuntarily influences an individual’s decision. Since the senders are independent from the market, this gives them increased credibility [2]. This independence makes WOM a more reliable and credible medium. Today’s new form of online WOM communication is known as electronic word-of-mouth. One of the most comprehensive conceptions of eWOM was proposed by Litvin et al., who described it as all of the informal communication via the internet addressed to consumers and related to the use or characteristics of goods or services or the sellers thereof [3]. The advantage is that the platform is open to all consumers and anyone can see another’s comments on the platform and share their opinions with other users. In the past, the only information channels for consumers to learn about products or services were friends and family, but now they are more likely to learn through eWOM [4]. At present, there are three major types of IWOM platforms: the first is a shopping platform’s own post-purchase evaluation system, which mainly evaluates the products, logistics and after-sales services related to completed transactions; the second is the dissemination of eWOM information through virtual community platforms, where a group of like-minded users communicate and share eWOM information through community websites; the third are life service websites that provide users with information and preferential activities on dining, entertainment and other life services. Disgruntled consumers can damage the image of a brand, even if it was previously recognized as offering high-quality goods [5]. Negative eWOM communications can adversely affect user attitudes and purchase intentions as well as a company’s brand image. This can lead to undesirable long-term outcomes, such as brand dilution, fluctuations in stock returns and the overall erosion of a company’s value [6]. Therefore, companies need to pay attention to online electronic word-of-mouth communication. These organizations should have the ability to react to negative eWOM during the exposure phase, as negative eWOM can also be amplified through social media channels.

Previous studies have found that eWOM has a significant impact on consumer decision making in the restaurant, hospitality and travel industries [7,8,9]. Studies have examined which experiences motivate customers to actively participate in eWOM and the impact of eWOM on consumer behavior. Based on this, this paper focused on the intrinsic driving mechanisms of negative IWOM. This paper innovatively considered the impact on the model from both the contextual and individual dimensions. As mentioned above, the study of eWOM communication is a comprehensive research proposition that integrates factors such as consumer behavior, communication and psychology. The generation of negative eWOM is a complex process that combines contextual factors and individual factors. However, previous studies have rarely considered the impact of these two factors on negative word-of-mouth communication as a whole. Finally, this paper selected the two dimensions of the “contextual” factor—user involvement, and the “individual” factor—user self-construal to consider their influence on the model path. The selection of these two types of variables not only improved the construction of the two models, but also broadened ideas for future research.

This study focused on the antecedents and consequences of negative eWOM motivation based on the above discussion. The first part focuses on how negative online shopping experiences stimulate the generation of negative eWOM motivation and whether the contextual factor mediates this process. The second part explores how negative eWOM motivations further evolve into word-of-mouth behavior and whether individual factor plays a moderating role in this transformation.

2. Literature Review and Hypothesis

2.1. Negative Online Shopping Experience and User Involvement

According to Zaichkowsky’s research, involvement is the degree of self-concern caused by the attributes of a product or service, which affects the perceived quality of the product and brand preference [10]. Matthes et al. argued that user involvement is the degree to which users perceive a product to be relevant to them based on their needs, goals or interests [11]. It describes a consumer’s subjective perception of the product or service in question. Andrews found that there are two main types of factor that influence user involvement based on the elaboration likelihood model; one is based on personal needs, goals and characteristics, such as personal values, cultural background, cognitive needs and perceptions of information and personal relevance; the other is based on context and decision making, such as the consumption environment and product usage experience [12]. Specifically, if the negative shopping experience is due to negative product quality, this means that consumers are likely to be dissatisfied with being cheated; then, consumers are likely to collect information to prove that the product quality is substandard, and consumer involvement is increased. If consumers find that the platform kills familiarity in the process of online shopping, that is, for the same goods or services, the old users see that the price is much more expensive than do the new users, this will inevitably lead to the dissatisfaction of the old users, and they may make complaints and find other ways to defend their rights and interests. Or perhaps there is a lack of information regarding the goods displayed on the platform and users will inevitably make more search efforts to obtain more information. If an item logistically cannot be delivered at the agreed upon time, or even in the case of lost pieces, then consumers have to spend more time and energy on logistics or negotiate with the merchant, and the degree of involvement increases. If consumers do not receive effective help and timely solutions when they encounter problems after purchase, this will greatly reduce the consumer’s goodwill towards the brand, especially for expensive goods, which will affect the consumer’s involvement. In summary, the following hypotheses were proposed for the study:

Hypothesis 1 (H1). 

User involvement has a positive correlation with negative online shopping experience.

H1a. 

User involvement has a positive correlation with negative product quality.

H1b. 

User involvement has a positive correlation with negative after-sales service.

H1c. 

User involvement has a positive correlation with negative online shopping platform environments.

H1d. 

User involvement has a positive correlation with negative logistics services.

2.2. The Mediating Role of User Involvement

Some studies have used user involvement as an independent variable to explore its effect on user attitudes and behaviors. Lai-Ying Leong et al. found that user involvement was an important factor influencing hotel booking intentions [13]. More studies have used user involvement as a mediating and moderating variable. Tuu and Olsen argue that engagement plays a mediating role in satisfaction and repurchase loyalty [14]. Product involvement moderates the relationship between eWOM and purchase intention [15]. User involvement moderates the relationship between composite attitudes and eWOM behavior [16].

This study attempted to further explore the role of involvement as an intermediate variable in the evolutionary path from shopping failure to emotional feedback. Due to the virtual nature of the online shopping process, users are unable to visually experience the goods they want to purchase and, therefore, can only use the information provided by the merchant as the basis for their purchase decision evaluation. In such a transaction process, cognitive dissonance is likely to occur. Unpleasant shopping experiences can be divided into process failures and outcome failures, both of which can seriously affect consumers’ shopping experience and trigger emotional changes [17]. Consumers’ emotions directly reflect their satisfaction with the shopping process [18]. eWOM motivation arises when users are unhappy with their online shopping experience and increased involvement also leads to an increased incentive to spread negative word of mouth [19]. Zeelenberg and Pieter found that negative consumer emotions associated with negative online purchases, such as regret and disappointment, showed a significant positive correlation with negative word-of-mouth communication, and this relationship was strengthened when user involvement increased [20].

Sundaram et al. argue that “consumer experience and motivation are closely related during word-of-mouth communication.” They suggested that negative word-of-mouth communication motivations include helping others, relieving emotions, retaliating against businesses and seeking information [21]. Wetzer et al. summarize that negative online word-of-mouth communication motivations include psychological comfort, emotional venting, seeking help, bringing people closer together, self-indulgence, self-presentation, warning other users and retaliating against businesses [22]. The potential motivations for consumers to engage in eWOM vary depending on the nature of the consumer experience. Consumers engage in eWOM after a negative online shopping experience primarily to prevent others from experiencing the problems they encountered, to relieve their own anger, anxiety and frustration, retaliate against the company associated with the negative consumer experience, and obtain advice on how to resolve the problem. In summary, the following hypotheses were proposed:

Hypothesis 2 (H2). 

Negative eWOM motivation has a positive correlation with user involvement.

Hypothesis 3 (H3). 

User involvement mediates the relationship between negative online shopping experience and negative eWOM motivation.

H3a. 

User involvement mediates the relationship between product quality and negative eWOM motivation.

H3b. 

User involvement mediates the relationship between after-sales service and negative eWOM motivation.

H3c. 

User involvement mediates the relationship between platform environment and negative eWOM motivation.

H3d. 

User involvement mediates the relationship between logistics service and negative eWOM motivation.

The study proposed a conceptual framework, as shown in Figure 1.

2.3. Negative eWOM Motivations and Behavior

Motivation is the behavioral driver that drives people to achieve their goals. The theory of reasoned action suggests that people will search for relevant information and predict the consequences before deciding whether to carry out a certain behavior and, therefore, human behavior is rational [23]. Ajzen introduced perceived control variables based on the above theory and concluded that the three factors that determine behavioral intentions include behavioral attitudes, subjective norms and perceived behavioral control [24]. Thereafter, the technology acceptance model, proposed by Davis, based on the theory of rational behavior and the theory of planned behavior, further introduced perceived usefulness and perceived ease of use to systematically explain the relationship between behavioral willingness and behavior itself. According to the theory of technology acceptance model proposed by Davis, the strength of behavioral intention directly affects the occurrence or absence of behavior [25]. Therefore, the motivation of user word-of-mouth communication directly determines word-of-mouth behavior. When users perceive eWOM as a way to alleviate negative emotions, specific motivations drive their behavior. In studying customers’ eWOM behavior in a hotel setting, Yaou Hu found that venting and financial incentives were prominent predictors of negative eWOM behavior [26]. In negative emotional states, tourists’ eWOM motivations include blowing off steam, seeking revenge, reminding others and seeking help [9]. Consumers often reduce their stress and anxiety after a negative shopping experience by posting negative eWOM [27]. In a study on consumer posting reviews in virtual communities, it was noted that the research showed that negative eWOM motivations had a significant moderating effect on service remediation and consumer satisfaction. Higher levels of service redress independently increase positive consumer responses [28]. Negative eWOM communication is subject to a combination of factors such as the degree of damage of the negative online shopping experience and user personality traits, and its transmission mechanism is not invariant. Berger and Schwartz further distinguished eWOM communication behaviors into ongoing and immediate word-of-mouth based on duration [29]. In summary, study 2 proposed the following hypotheses:

Hypothesis 4 (H4). 

Negative eWOM motivation has a positive correlation with immediate eWOM behavior.

Hypothesis 5 (H5). 

Negative eWOM motivation has a positive correlation with ongoing eWOM behavior.

2.4. The Moderating Role of User Self-Construal

Self-construal represents a set of thoughts, feelings and behaviors towards the relationship between the self and others and the distinction between the self and others [30]. It mainly reflects the differences in the meaning of “self” under the influence of Eastern and Western cultures. This paper focused on the moderating role of user self-construal in the evolution of word-of-mouth communication motivation on behavior, drawing on Markus’ typical classification of self-structure types, namely, independent self-construal and dependent self-construal [31]. Independent self-structures are usually independent of the social context, emphasizing the individual’s inner abilities, thoughts and feelings; pursuing uniqueness and self-expression; and focusing on inner attributes and the achievement of personal goals. They usually engage in communication in a straightforward manner. In contrast, dependent self-constructors are “flexible and versatile” selves related to social contexts. They usually define themselves based on external attributes, such as social status or relationships. In addition, they try to adapt and find a sense of belonging and participate in communication indirectly. Independent self-construal consumers are more likely to spread WOM by their own related motivations such as self-enhancement, while interdependent self-construal consumers are more likely to spread WOM with the motive of helping others [32].

The altered psychological states resulting from negative online shopping experiences impact differently on individuals due to differences in self-construal. Studies have shown that interdependent individuals respond more strongly to the post-purchase intention and word of mouth than independent individuals [33]. Wu et al. found that independent self-construal can moderate the effect of electronic word-of-mouth fragmentation on consumer order decisions [34]. Lee et al. explored how consumers’ self-construal can affect consumers’ eWOM behavior through two cognitive factors in the context of a social networking site. In addition, self-construal theory has also been widely used in various contexts, such as advertising [35], organizational behavior [36] and brand connection [37]. The virtual nature of the online world blurs people’s otherwise real-life personality traits and people may exhibit very different personality traits online than they do in reality. Such changes urgently require experiments to analyze the relationship between negative motivations and eWOM behavior on the Internet. In summary, study 2 proposed the following hypotheses:

H6a. 

Independent self-construal moderates the relationship between negative eWOM motivation and immediate eWOM behavior.

H6b. 

Independent self-construal moderates the relationship between negative eWOM motivation and ongoing eWOM behavior.

H7a. 

Interdependent self-construal moderates the relationship between negative eWOM motivation and immediate eWOM behavior.

H7b. 

Interdependent self-construal moderates the relationship between negative eWOM motivation and ongoing eWOM behavior.

The study proposed a conceptual framework, as shown in Figure 2.

3. Methodology

3.1. Measures and Questionnaire

We conducted a literature review and a discussion of variables in order to operationalize the constructs and create a questionnaire. As shown in Appendix A, we designed a structured questionnaire with two parts. The first part concerned personal details, including gender, age, occupation and education; the second part consisted of 35 questions on a 7-point Likert scale ranging from 1 (strongly disagree) to 7 (strongly agree). As shown in

Table 1. Sample Demographics.

Characteristics		Frequency	Percent
Gender	Male	173	54.40%
	Female	145	45.60%
Age	≤20	35	11.01%
	21–30	120	37.74%
	31–40	58	18.24%
	41–50	64	20.13%
	≥51	41	12.89%
Education	High school	57	17.92%
	College	136	42.77%
	Undergraduate	100	31.45%
	Master	14	4.40%
	Doctor	11	3.46%
Age of online shopping	≤1 y	12	3.77%
	1–2 y	59	18.55%
	3–5 y	164	51.57%
	5–10 y	71	22.33%
	≥10y	12	3.77%
Job	Student	38	11.95%
	Teacher	39	12.26%
	Civil Servant	32	10.06%
	Self-employed	130	40.88%
	Employees	68	21.38%
	Jobless/freelance	11	3.46%
Monthly revenue	0–3000	47	14.78%
	3000–6000	141	44.34%
	6000–1000	97	30.50%
	10,000–15,000	21	6.60%
	≥15,000	12	3.77%

, a total of 318 valid respondents’ responses were collected; 145 of them were female (52.9%) and 173 were male (47.1%). The majority of the respondents (319 or 76.3%) were 21–30 years old. A total of 293 (73.7%) had earned a bachelor’s degree, while 25 (12.2%) had earned a master’s degree.

3.2. Reliability and Validity

In this study, reliability was assessed using the CITC index and Cronbach’s alpha (α) coefficient. The overall Cronbach’s alpha coefficient of the questionnaire in this paper was 0.792 and the standardized Cronbach’s alpha was 0.793, which proves that the overall reliability of the questionnaire was good. For each item, the CITC was greater than 0.7 and the α was greater than 0.7, which are well above the suggested thresholds [38]. We used the average variance extracted (AVE) and composite reliability (CR) to assess the convergent validity, as suggested by Fornell and Larcker [38]. As shown in

Table 2. Cronbach’s α, CR, AVE and factor loadings.

		CITC	α	CR	AVE	Loadings
Product Quality (A)	A1	0.759	0.887	0.857	0.666	0.791
	A2	0.863				0.852
	A3	0.721				0.804
After-sales Service (B)	B1	0.780	0.920	0.912	0.777	0.803
	B2	0.885				0.967
	B3	0.849				0.867
Platform Environment (C)	C1	0.770	0.896	0.872	0.696	0.774
	C2	0.870				0.914
	C3	0.746				0.810
Logistics Service (D)	D1	0.741	0.890	0.844	0.643	0.802
	D2	0.871				0.794
	D3	0.748				0.811
User Involvement (E)	J1	0.728	0.887	0.820	0.603	0.786
	J2	0.869				0.724
	J3	0.745				0.817
Negative eWOM Motivation (F)	F1	0.797	0.935	0.916	0.686	0.863
	F2	0.817				0.817
	F3	0.832				0.724
	F4	0.850				0.813
	F5	0.835				0.913
Self-construal type (G)	G1	0.863	0.967	0.966	0.740	0.856
	G2	0.856				0.846
	G3	0.829				0.820
	G4	0.816				0.826
	G5	0.891				0.901
	G6	0.842				0.886
	G7	0.867				0.890
	G8	0.806				0.811
	G9	0.832				0.857
	G10	0.893				0.903
Negative eWOM behavior (H)	H1	0.849	0.949	0.919	0.655	0.791
	H2	0.859				0.857
	H3	0.841				0.783
	H4	0.822				0.821
	H5	0.895				0.803
	H6	0.802				0.799

, for each item, the CR was higher than 0.8 and the AVE was greater than 0.5, which are well above the suggested thresholds. The discriminant validity was assessed by comparing the correlations among the latent variables with the square root of the AVE. As shown in

Table 3. Discriminant validity.

	Mean	SD	A	B	C	D	E	F	G	H
A	4.797	1.401	0.863
B	4.760	1.363	0.353	0.881
C	4.787	1.389	0.256	0.245	0.861
D	4.867	1.424	0.347	0.452	0.309	0.861
E	4.820	1.457	0.492	0.338	0.367	0.301	0.871
F	4.950	1.434	0.375	0.420	0.474	0.443	0.301	0.876
G	4.941	1.377	0.202	0.267	0.234	0.323	0.421	0.118	0.860
H	4.997	1.350	0.221	0.397	0.331	0.142	0.311	0.224	0.306	0.809

Note: SD: Standard Deviation.

, the square root of the AVE for the construct was higher than its correlation coefficient with other variables, indicating that results of the research were suitable for subsequent structural equation modeling analysis.

4. Results

To evaluate the structural model, this study was conducted in two phases. The first stage used SPSS and AMOS to validate the fit of conceptual model 1 and the significance of the hypothesized paths, and the bootstrap method of bias correction was used to test for the mediating effects of user involvement. The second stage also used SPSS and AMOS to verify the effect of negative eWOM motivation on eWOM behavior and group regression to test the moderating effect of self-construal.

4.1. Structural Model 1

As shown in

Table 4. Fitting Results of Model 1.

Test Statistic	RMR	RMSEA	GFI	RFI	CFI	PGFI	${\mathbf{{\rm X}}}^2/\mathbf{df}$
Result	0.042	0.000	0.97	0.874	0.908	0.547	0.885

, the model fit and values were found to be in the acceptable range, except for the relative fit index (RFI); therefore, the fit of study 1′s model was good.

From

Table 5. Path Coefficient of Model 1.

Hypotheses and Paths	Load Factor	Standard Error	p Value
H1a: User involvement ← Negative product quality	0.401	0.055	*
H1b: User involvement ← Negative after-sale service	−0.130	0.090	0.151
H1c: User involvement ← Negative online shopping platform environment	0.372	0.026	*
H1d: User involvement ← Negative logistics service	−0.005	0.026	0.848
H2: Negative eWOM motivation ← User involvement	0.520	0.033	*
H3a: Negative eWOM motivation ← Negative product quality	0.502	0.021	*
H3b: Negative eWOM motivation ← Negative after-sale service	0.423	0.027	*
H3c: Negative eWOM motivation ← Negative online shopping platform environment	0.277	0.037	*
H3d: Negative eWOM motivation ← Negative logistics service	0.331	0.036	*

Note: * indicate p < 0.001.

, negative product quality and negative online shopping platform environments were positively related to user involvement; therefore, H1a and H1c were verified. This demonstrates that all four negative online shopping experiences were positively related to negative eWOM motivations. User involvement was positively associated with negative eWOM motivations; thus, H2 was supported. By comparing the magnitude of the path coefficients, it is easy to find that product quality had the greatest impact on negative eWOM motivation and product quality had the greatest impact on user involvement, indicating that users are still most concerned about the product quality itself. However, the online shopping platform environment had the smallest path coefficient on the motivation of negative eWOM communication, indicating that its influence is minimal. As shown in

Table 6. Fitting Results of Modified Model 1.

	RMR	RMSEA	GFI	RFI	CFI	PGFI	${\mathbf{{\rm X}}}^2/\mathbf{df}$
Original results	0.042	0.000	0.970	0.874	0.908	0.547	0.885
Modified results	0.044	0.000	0.973	0.887	0.910	0.551	0772

, it was found that the values of AGFI and $\mathsf{\chi }2/\mathrm{df}$ decreased, and the other indexes were significantly optimized. The value of RFI was closer to the ideal value, which was completely within the acceptable range. From

Table 7. Path Coefficient of Modified Model 1.

Hypotheses and Paths	Load Factor	Standard Error	p Value
H1a: User involvement ← Negative product quality	0.474	0.047	$*$
H1c: User involvement ← Negative online shopping platform environment	0.392	0.015	$*$
H2: Negative eWOM motivation ← User involvement	0.531	0.038	$*$
H3a: Negative eWOM motivation ← Negative product quality	0.575	0.006	$*$
H3b: Negative eWOM motivation ← Negative after-sale service	0.323	0.115	0.006
H3c: Negative eWOM motivation ← Negative online shopping platform environment	0.279	0.042	$*$
H3d: Negative eWOM motivation ← Negative logistics service	0.327	0.033	$*$

Note: * indicate p < 0.001, p represents the significance of the interaction between paths. When p < 0.05, the paths are significant.

, the modified model was significant on the remaining paths; that is, the remaining seven hypotheses are all valid.

Since some hypotheses in model 1 did not pass the test, it was planned to modify this model by simplifying it. Hence, the two insignificant paths of “negative after-sales service quality → user involvement degree” and “negative logistics service quality → user involvement degree” in model 1 were deleted.

4.2. Mediation Effects

This paper used the bootstrap method to test for mediating effects, with 2000 replicate samples and a 95% confidence interval.

H3 verified that user involvement had a partial mediating effect between the four negative online shopping experiences and negative eWOM motivation. From

Table 8. Mediation effect test in model 1.

	Effect	Bias Corrected 95% CI
Product Quality → Negative eWOM motivation	Total effect	0.232	0.689
	Indirect effect	0.015	0.292
	Direct effect	0.086	0.577
Online shopping platform environment → eWOM motivation	Total effect	0.315	0.732
	Indirect effect	0.027	0.265
	Direct effect	0.156	0.542

, the 95% CI intervals of the indirect effects of product quality and online shopping platform environment did not include 0, which verified the existence of mediating effects; similarly, the 95% CI intervals of direct effects did not include 0, which also indicated the existence of partial mediating effects, and the results were consistent with the discriminatory results of the path coefficient method. Therefore, the total mediating effect of user involvement in the influence of product quality and online shopping platform environment on negative eWOM motivation was significant.

4.3. Structural Model 2

The results of the fit index of the structural equation model were calculated using Amos software. As shown in

Table 9. Fitting Results of Modified Model 2.

	RMR	RMSEA	GFI	RFI	CFI	PGFI	${\mathbf{{\rm X}}}^2/\mathbf{df}$
Result	0.036	0.000	0.93	0.92	0.912	0.522	1.216

, except for the PGFI index, which was slightly higher than the fit standard, other indexes were within the standard interval. Therefore, the model for study 2 had a high degree of fit.

As shown in

Table 10. Path Coefficient of Model 2.

Hypotheses and Paths	Load Factor	Standard Error	p Value
H4: Negative eWOM motivation→ Immediate eWOM communication	0.553	0.022	$***$
H5: Negative eWOM motivation→ Ongoing eWOM communication	0.401	0.017	$***$

Note: $***$ indicate p < 0.001.

, H4 and H5 are both supported. This shows that there was a significant positive correlation between the motivation of negative online word-of-mouth communication and the two kinds of communication behaviors.

H4 and H5 verify that there was a significant positive correlation between the negative eWOM motivation and the two types of eWOM behavior. Whether it is a short-term transmission form or a lasting transmission form, both are generated by motivational stimulation. From the perspective of behavior categories, the path coefficient of immediate word-of-mouth communication is higher than that of ongoing word-of-mouth communication, indicating that such behavioral intentions are stronger; that is, users are more willing to use online comments and other forms of word-of-mouth communication. The above results show that the relationship between negative eWOM motivation and immediate eWOM behavior is stronger than that of ongoing eWOM behavior, which proves that users are more inclined to spread eWOM through immediate forms, such as posting negative reviews online after generating eWOM motivation.

4.4. Moderation Effects

This paper explored the moderating effect of user self-construal types using grouped regressions and, therefore, examined the moderating effect of self-construal on eWOM motivation under two types of word-of-mouth behaviors separately.

H6 and H7 tested whether self-construal has a moderating effect on eWOM motivation under two types of eWOM behaviors. As shown in

Table 11. Regression results under immediate eWOM behavior.

Interdependent			Independent
Variable	Coefficient	p Value	Variable	Coefficient	p Value
Constant term	0.1491	0.1101	Constant term	−0.0958	0.5071
X	0.9519	0.0000	X	1.0342	0.0000
R2	0.8588		R2	0.8392
F	1198.3171		F	610.7641
Sig.F	0.0000		Sig.F	0.0000

, under immediate eWOM behavior, the significance of the F-test statistic for both regression models was 0, and the model was significant overall. In the model for interdependent self-construal, negative eWOM motivation could explain 85.9% of the immediate eWOM behavior. In the model of independent self-construal, negative eWOM motivation could explain 83.9% of the immediate eWOM behavior. The t-statistic of the moderating model of self-construal under the immediate word of mouth behavior was 1.832, and the concomitant probability was 0.067, which means that the original hypothesis was rejected with a 10% confidence interval and the moderating effect was significant. The moderating effect of different user self-construal on the “motivation–immediate word-of-mouth” model was found to be significant. Therefore, H6a and H7a were supported. In terms of specific personality types, independent self-construal moderates negative eWOM motivation and immediate online word-of-mouth behavior more significantly, because the coefficient of the independent variable X for the interdependent personality model was 0.952, which is smaller than the coefficient of the independent personality model at 1.034. After experiencing an unpleasant shopping experience, these customers are more likely to give quick and timely feedback on their post-purchase experience and vent their emotions by posting online reviews and other forms.

As shown in

Table 12. Regression results under ongoing eWOM behavior.

Interdependent			Independent
Variable	Coefficient	p Value	Variable	Coefficient	p Value
Constant term	0.0119	0.9122	Constant term	−0.1264	0.3946
X	1.0066	0.0000	X	1.0460	0.0000
R2	0.8340		R2	0.8350
F	989.6102		F	591.8913
Sig.F	0.0000		Sig.F	0.0000

, the models of interdependent and independent personality on eWOM motivation pass the significance test. The t-statistic of the moderating model of self-construal on negative eWOM motivation under ongoing electronic word-of-mouth behavior was calculated at 0.0819, and its concomitant probability was 0.413, which means that the original hypothesis was accepted with a 10% confidence interval and the moderating effect was not significant; thus, H6b and H7b were not supported.

5. Discussion

5.1. Summary

Based on the results, the main conclusions of the paper are as follows:

• The results of the empirical analysis showed that negative product quality, negative platform environment, negative logistics and negative after-sales service all had different degrees of influence on the motivation of negative eWOM communication. There is a significant positive correlation between negative product quality and negative platform environment and user involvement.
• User involvement was positively associated with negative eWOM communication motivations and was partially mediated between negative product quality, negative online shopping platform environment and negative eWOM motivation.
• There was a significant positive correlation between negative eWOM motivation and the two kinds of eWOM communication behaviors.
• Under immediate eWOM behavior, the user’s self-construal moderated the relationship between eWOM motivation and immediate eWOM behavior, while this moderating effect is not significant under ongoing eWOM behavior.

Conclusion 1 further confirms that the experience of shopping failure is an important trigger of emotional changes in consumers. It can be inferred that users are still most concerned about the product itself, and that while users choose to shop online and enjoy the favorable price and convenient logistics, they do not lower their expectations of the product and service quality itself. The online shopping platform environment is an important factor affecting consumers’ shopping experience. According to the ELM model, online shopping experiences can be divided into central path cues and edge path cues, central path cues include factors such as community atmosphere environment in addition to product quality itself, and this variable has a significant impact on word-of-mouth communication intention [39]. However, there was no strong statistical evidence that the online shopping platform environment has a significant effect on users’ negative word-of-mouth communication motivations. Therefore, this paper argues that the unpleasantness brought to users by a terrible shopping experience, to some extent, undermines the unpleasantness formed due to the unfriendly online shopping platform environment, as users care most about the product itself. This is also reflected by the fact that the product quality dimension had the highest path coefficient among the four factors. Moreover, we believe that in addition to “hard” services such as logistics services, users are increasingly concerned about “soft” services such as after-sales maintenance, the lack of which is more likely to cause dissatisfaction. Previous studies have investigated the effect of information diversity and valence on influencing the level of user engagement [40], and the effect of the dimensions of negative online shopping experience on user engagement complements previous studies.

Conclusion 2 suggests that increased user engagement leads to increased negative eWOM motivation. This study is consistent with the findings of Cheung et al. [41]. If users are willing to invest personal energy (physically, emotionally and cognitively) into an online shopping experience, they will have a higher propensity to spread word-of-mouth communication about it. In addition, contextual factors in a given environment are important factors influencing the level of user engagement. This finding is similar to that of the study by Mclean et al. [42]. From the perspective of path coefficients, there is a significant positive correlation between product quality dimensions and platform context and user engagement. Product quality remains the primary factor influencing user engagement. The environment of the online shopping platform had the second highest impact on user engagement. A bad online shopping platform environment consumes users’ energy in the shopping process, increases search costs and, thus, affects the generation of negative IWOM communication motivations.

Conclusions 3 and 4 show that both immediate and ongoing eWOM behaviors are generated by motivation, which further validates that the construction of this study’s model was consistent with the S–O–R framework. The path coefficients of immediate eWOM behaviors were higher than those of ongoing eWOM behaviors, indicating that users have stronger intentions to engage in immediate eWOM behaviors. This paper focuses on the effect of users’ personality traits on word-of-mouth communication, a study that differs from Yan Li et al.’s approach of using self-construal as a mediator of product scarcity and WOM willingness [43]. The results of the SEM show that negative eWOM motivations significantly affect eWOM behavior. Independent individuals are freer to express themselves on an online platform without being influenced by other voices; therefore, independent personality constructs have a more significant moderating effect on the motivation and immediate eWOM communication, and this finding is similar to that of Song et al. that independent self-construal weakens the negative impact of electronic word-of-mouth dispersion on order decision [44]. Interdependent individuals are more concerned about the social attributes of their behavior, so they are more rational and cautious. This is consistent with Hoffman et al.’s findings that consumers’ interdependent self-construal does not impact their sharing behavior, purchase intention and WOM [33].

5.2. Theoretical Implications

The theoretical contribution of this study was the exploration of the intrinsic driving mechanisms of negative eWOM communication behavior by considering the mediating role of user engagement and the moderating role of self-construal. Since previous studies have only empirically examined which experiences in offline contexts trigger positive eWOM, this paper fills the research gap on the effect of experiences in online contexts on negative eWOM. The SEM results suggest that all four dimensions of negative online shopping experience have an impact on the motivation of negative eWOM communication, while user engagement plays a partly mediating role in the middle. A partial positive relationship between consumer involvement and eWOM intention was identified [45]. If the level of involvement is high, consumers will pay more attention to the specific product and its information. After purchase, consumers are expected to have stronger emotional reactions and thus be more eager to share information about the product [46]. However, our study noticed that negative experiences enhance user involvement by providing more time and effort.

The study also found that the communicator’s own personality traits can shape the negative oral transmission motivation–negative oral transmission behavior path. However, this moderating effect was not always significant for the different types of eWOM behaviors. Compared with dependent personality constructs, individuals with independent self-constructs are more likely to give timely feedback on their post-purchase experiences and vent their emotions after experiencing unpleasant shopping experiences by posting online reviews. This further confirms the effectiveness of the self-construct on social media success.

5.3. Managerial Implications

Consuming in nonphysical scenarios means that users have to pay a greater cost of trust. Therefore, users measure the quality of online shopping in a holistic dimension, and the lack of any aspect can cause an unpleasant shopping experience for users which, in turn, causes a lack of trust. Therefore, online retailers need to pay attention to service recovery after service failure. When merchants provide service remedies that satisfy users, users will not be motivated to spread negative word-of-mouth and, subsequently, they will not engage in negative eWOM communication. They may even become propagators of corporate image building.

Given that eWOM behavior is inconsistent among individuals with different types of self-construal, it is necessary for companies to focus on user traits and introduce big data analysis techniques for user profiling. In the era of big data, users’ purchasing habits, browsing records and evaluation information are clearly presented in the background. Businesses should conduct user profiling based on the information they already have, and carry out targeted and accurate marketing and corporate image maintenance for users. In fact, negative eWOM communication is often an irrational act of unplanned behavior. By establishing a membership mechanism and maintaining a community of users, merchants can provide users with a channel to vent, a space to discuss and a platform to seek help in order to guide word-of-mouth and then take accurate service remedies to reduce the negative impact to a manageable level.

6. Limitations and Future Research

The sample size of this paper was not rich enough to cover adequately the situation of internet users in all regions of China. From the regional scope of respondents’ responses, as shown by the ID at the backend of the questionnaire, it is obvious that the respondents were concentrated in East China, and there were few respondents in Central and Western China. The reason was that people in the eastern region have a higher level of internet participation. The regional distribution tended to reflect the economic strength of the respondents, which is also an important influencing factor on the eWOM of users. In addition to considering user personality traits, future research models could introduce income level, education level and length of online shopping as control variables to enrich the model construction and provide targeted suggestions for enterprise word-of-mouth crisis management.