Joint Effect of Different Dimensions of eWOM on Product Sales

Abstract

Understanding the antecedents and consequences of electronic word-of-mouth (eWOM) is of significant theoretical value and great interest to both consumers and firms. To reconcile the disagreement in the relationship between eWOM and product sales, we investigate the joint effect of different dimensions of eWOM on the sales of durable goods (i.e., cars). Our study is based on eWOM collected from Autohome, a leading online review platform of automobiles in China. Our empirical results show that the volume and valence of eWOM positively affect car sales, and variance negatively affects car sales when the volume of eWOM is low. Further analyses of the joint effect show that the positive impact of eWOM volume on sales is stronger when the valence is lower and the variance is higher. This research contributes to the literature on eWOM and provides helpful suggestions for firms by shedding light on the joint effect of different dimensions of eWOM in the context of the Chinese automobile industry.

1. Introduction

The advent of the Internet, especially Web 2.0, has provided a platform for consumers to communicate with each other, enabling consumers to make comments and share experiences after purchasing products [1,2]. These online reviews are widely known as electronic word-of-mouth (eWOM) [3]. As a complement and substitute for offline word-of-mouth and business-to-customer communication channels [4,5,6], eWOM conveys information about product quality and value. Previous evidence demonstrates that customers can reduce uncertainty and risk by browsing eWOM and make informed purchasing decisions based on their peers’ experiences and their preferences [7,8,9]. Furthermore, previous studies suggest that eWOM is more informative and credible than traditional advertisement information [10,11]. Hence, consumers concern themselves more with their peers’ consumption patterns and cognitive insights [12,13]. According to the report of Bright Local, most consumers read online reviews before making purchase decisions [14].

A large body of studies was conducted to investigate the effect of eWOM on sales. However, the findings of previous studies are mixed and inconsistent. While ample evidence demonstrates the positive relationship between eWOM and product sales [4,15,16,17,18,19,20], some scholars argue that eWOM does not affect sales [11,21,22,23,24]. The disagreement in the academic community is probably due to the categories of goods in previous studies, such as books, movies, video games, hotels, etc. Another reasonable explanation is that the majority of extant research neglects the joint effect of different dimensions of eWOM and the combined influence of the eWOM and product attributes.

Based on a dataset collected from Autohome, one of the most popular car review websites in China, this study aims to reconcile the disagreement in the extant literature. Specifically, we decompose eWOM into valence, volume, and variance and investigate their joint effects on sales performance. The empirical result of this research provides evidence that (1) the volume and valence of eWOM positively affects car sales; (2) a higher variance of eWOM ratings for a car leads to lower sales if and only if the volume of eWOM is low; (3) the positive impact of eWOM volume on sales is stronger when the valence is lower and the variance is higher. Our study makes a contribution to the literature by shedding light on the joint effect of different eWOM dimensions on sales in the context of durable products (i.e., cars). Our findings also provide practical managerial implications for firms/brands in the management of eWOM and product marketing.

The remainder of this paper proceeds as follows. We present a brief review of the academic literature on eWOM, then deduce our hypotheses in Section 2. Section 3 elaborates on the data, variables, and model specification. Then, we describe the descriptive statistics and empirical results in Section 4. Finally, we conclude our findings and discuss the managerial implications and limitations of this research in Section 5.

2. Literature Review and Hypotheses

2.1. Literature Review

Lilien et al. [25] suggest that consumers make purchasing decisions based on the awareness and the cognition of a product. Accordingly, searching and collecting information about a product is an essential process that can help reduce uncertainty risk prior to purchasing [9]. As a supplement and substitute for traditional product information, such as product characteristics and offline WOM, eWOM plays an increasingly important role in marketing thanks to the advance of the Internet [26].

Chintagunta et al. [19] decompose eWOM into three dimensions: valence, volume, and variance. Valence is the average rating and represents the average customer satisfaction. Volume is usually gauged by the total number of online reviews or comments for a particular product. Variance represents the degree of disagreement or heterogeneity among customers’ evaluations.

Prior studies provide mixed findings with respect to the effects of these dimensions of eWOM. While some studies demonstrate a positive effect of eWOM volume on sales [3,7,11,16,22,27,28,29,30], others find no evidence supporting this conclusion [8,19]. A further study conducted by Ho-Dac et al. [31] finds that the effect of eWOM volume on sales varies by product type.

Previous literature also presents mixed findings with respect to valence. Many studies report a significant positive effect of valence [4,7,15,18,19,20,27,29,32]; however, other researchers challenge this mechanism, as their findings suggest that valence exerts no significant effect on sales [21,22,24,33]. In-depth research by Rosario et al. [34] reveals that valence has an asymmetric effect on product sales.

Variance has not been the focus of extant studies, but the results are also inconsistent. A significant influence of variance is reported by Clemons et al. [20] and Sun [29], whereas Zhu and Zhang [9] identify a significant effect only for less popular games. Chintagunta et al. [19] do not find a significant effect of variance.

The disagreement in the academic community is probably due to the particular categories of goods in previous studies, such as books [4,35,36], movies [11,26,27,33,35], digital cameras [3], software [37], video games [9,38], hotels [39,40,41,42,43], restaurants [44], and fashion products [45]. Prior studies tend to focus on the online sales of nondurable goods or product categories that involve the online consumption of digital products. Durable goods, such as automobiles, are used over a much more extended period; therefore, they have longer inter-purchase intervals and greater product uncertainty and perceived risk than nondurable goods [46]. However, few studies examine the effects of eWOM in the context of the automobile industry. We summarize the major findings of related research in

Table 1. Studies in context of automobile industry.

Study	Method	Data	eWOM Metrics	Key Findings
[30]	three-stage least squares	Sales Data: Automotive News; eWOM Data: Edmunds.com and Consumer Reports (online); Time span: 2000–2007	Volume	Redesigned models of automobiles attract more consumer word of mouth than new models.
[47]	least-squares linear regression	Sales Data: Automotive News; Search Data: Google Trends; eWOM Data: Google’s discussion forum search Time span: 2007–2010	Volume Valence	Adding search trend data to models based on social media data significantly improves predictive accuracy, and the improvement is larger for “value” car brands than “premium” car brands; Predictive models based on search trend data provide predictive accuracy that is comparable, at least, to that of social media data-based predictive models.
[48]	3 levels Binomial response model with random effects	Sales Data: Automotive News; eWOM data: Consumer Reports (online); Time span: 2000–2007	Positive Volume Negative Volume	Reliability/quality, performance/design, and innovativeness play qualitatively different roles in generating word of mouth.
[49]	PVAR model	eWOM Data: Twitter and Facebook; Sales Data: WardsAuto Premium database; Advertising Data: Kantar Media; Time Span: 2009–2015	Volume	Facebook and Twitter have heterogeneous effects on offline vehicle sales. Moreover, FGC is more effective than UGC for influencing offline light vehicle sales.

.

Additionally, the majority of extant research neglects the joint effect of different dimensions of eWOM and the combined influence of eWOM and product attributes [46,50]. Regarding the joint effect of different dimensions of eWOM on product sales, Sun [29] and Kostyra et al. [51] argue that volume and variance can moderate the effect of valence on sales. Regarding the joint effect of eWOM and product attributes on product sales, Zhu and Zhang [9] find that online reviews are more influential for less popular games and games whose players have greater internet experience. Lu et al. [41] finds that eWOM has more impact on sales for hotels with lower rather than higher star ratings. The study by Wang et al. [52] shows that eWOM moderates the effect of the room price and star rating on hotel booking negatively. We summarize the related studies in

Table 2. Studies about the joint effect on sales.

Study	Method	Product Category	Data	eWOM Metrics	Key Findings
[29]	DID approach	Book	Amazon.com and BN.com (two period: January 2009 and May 2009)	Volume Valence Variance	The higher variance would correspond to a higher subsequent demand if and only if the average rating is low.
[9]	DID approach	Video Game	GameSpot (from March 2003 to October 2005)	Volume Valence	Online reviews are more influential for less popular games and games whose players have greater Internet experience.
[41]	DID approach	Hotel	Ctrip and Elong (Hotels in nine major cities in China from December 2009 to November 2011)	Volume Valence Variance	The average rating and variance of customer reviews will have a positive effect on hotel sales; eWOM has more impact on sales for hotels with lower rather than higher star ratings.
[52]	Negative Binomial Regression model	Hotel	Expedia.com (250 hotels in Detroit from 17 April to August 7 2013)	Volume Valence	The online sales performances of hotels with more or better WOM are less likely to be influenced by lower prices or higher star ratings.
[51]	Choice-based conjoint experiment	eBook reader	Survey Data (771 respondents, treatment group included 601 respondents, and the control group included 170 respondents)	Volume Valence Variance	Volume and variance do not affect customers’ choices directly but they moderate the impact of valence on customers’ choices; importance of product attributes decreases in the presence of eWOM
[45]	Structural equation modeling (SEM)	Fashion products	Survey Data (477 Chinese WeChat users in Beijing and Shanghai)	None	Social media usage significantly moderates the relationship between eWOM and the online purchase intention of fashion products
[53]	dynamic panel data (DPD)	tablet computers	312 unique SKUs with sales records spanning 23 weeks from 1 February to 11 July 2012	Volume Valence	Both numerical ratings and sentiment have significant impacts on the sales performance; The influence of overall sentiment on sales is completely mediated by star rating;

.

In summary, a more discerning approach would be to acknowledge that the effect of eWOM varies in different contexts [40]. In this study, we examine the joint effect of different dimensions of eWOM in the context of the Chinese automobile industry.

2.2. Conceptual Framework

Similar to Kostyra et al. [51], we decompose eWOM into the dimensions of valence, volume, and variance to examine them and their interactions in influencing customer choice. Figure 1 depicts our conceptual framework. We assume that the volume and valence of eWOM positively affect consumer decision making, and variance negatively affects consumer decision making. At the same time, the valence of eWOM has a substitution relationship with the volume of eWOM, and the variance will enhance the influence of the volume of eWOM. We detail the explanation for these assumptions below when we develop our hypotheses.

2.3. Development of Hypotheses

According to Clarke and Belk [54], cars are durable goods with high financial risk, and hence, consumers perceive the substantial importance of them. Improper purchasing decisions of cars will subject consumers to large and long-term losses. Prior studies have shown that eWOM can contribute information regarding the quality and value of a product and, therefore, can reduce customers’ choice risk [7,9]. In this study, we will examine the effects of eWOM on car sales from three dimensions, i.e., volume, valence, and variance.

First, the volume of eWOM presents the quantity of information available to the consumer. A large volume of eWOM leads to greater consumer awareness and confidence [3,26], and a reduction in consumer uncertainty [55]. Liu [11] and Cui et al. [7] reveal that the volume of eWOM is a proxy for product popularity. According to the theory of herd behavior [56], consumers frequently select popular brands because they believe that popularity indicates better quality. Hence, the herd effect may drive more potential customers to buy the product if more reviews are available [34]. Second, the valence of eWOM is also indicative of a product’s reputation and expected product quality [11]. High valence reduces the potential consumers’ perceived product uncertainty; hence, consumers prefer products with high valence. Finally, the variance of eWOM represents the degree of disagreement or heterogeneity among customers’ evaluations [29,51]. This disagreement occurs when some consumers rate high while others rate low, and hence, variance is generally associated with uncertainty, which may make customers choose products with low variance. Therefore, we propose the following hypothesis based on the extant literature.

Hypothesis 1a (H1a).

The volume of eWOM is positively related to product sales for automobile industries.

Hypothesis 1b (H1b).

The valence of eWOM is positively related to product sales for automobile industries.

Hypothesis 1c (H1c).

The variance of eWOM is negatively related to product sales for automobile industries.

According to Moe and Schweidel [50], the interpretation of volume and valence can be misleading when each metric is considered separately. Consumers consider the volume and valence of eWOM jointly when they make purchase decisions [21,22,24,33]. We suspect that there is a substitution relationship between volume and valence of eWOM. As cars are expensive durable products, consumers are strongly concerned with quality. The eWOM valence represents previo”s cu’tomers’ opinions about product quality [26]. Low valence increases the potential consumers’ perceived product uncertainty, which makes consumers read more eWOM to infer the true quality of the product. On the contrary, high valence enhances the potential consumers’ confidence, which makes consumers rely less on eWOM when they make purchase decisions. Accordingly, we propose the following hypothesis.

Hypothesis 2 (H2).

The rating will weaken the relationship between the volume of eWOM and product sales for automobile industries.

In addition to the valence of eWOM, we assume that variance also acts as a moderator of the impact of volume on customers’ purchase decisions. The variance of eWOM represents the degree of disagreement or heterogeneity among customers’ evaluations [29,51]. This disagreement occurs when some consumers rate high while others rate low. The inconsistent opinions will increase the potential consumers’ perceived product uncertainty [57,58]. Under an environment of low information inconsistency, consumers may need to read more eWOM to infer the true quality of the product. Therefore, we propose the following hypothesis.

Hypothesis 3 (H3).

The disagreement among customers’ evaluations will strengthen the relationship between the volume of eWOM and product sales for automobile industries.

3. Methodology

3.1. Data

The data for this study are collected from three publicly available sources: Autohome (https://www.autohome.com.cn, accessed on 10 January 2022), CSMAR (https://www.gtadata.com, accessed on 10 January 2022), and the National Bureau of Statistics of China (NBSC) (http://www.stats.gov.cn, accessed on 10 January 2022).

The eWOM data are crawled from Autohome. We choose Autohome as the source of eWOM data for two important reasons. First, Autohome is a leading online website for automobile consumers in China. According to Alexa.com, a website providing global traffic rank and country-specific ranking service for websites, Autohome is the most popular website for Chinese car buyers; second, the online site of Autohome only permits users who bought cars to post reviews. This mechanism minimizes the probability that carmakers anonymously post online reviews to praise their models or belittle competitors.

The eWOM data set includes more than 1 million customer reviews covering almost all the car models in Chinese auto markets from September 2012 to July 2019. Each customer review includes eight ratings for various dimensions of car quality: space, engine, handling, fuel consumption, comfort, exterior, interior, and value for money. In addition to these numeric ratings, each eWOM also includes the buy date, buy address, dealer, real fuel consumption (L/km), mileage (km), usage purpose, photos, actual price (tax excluded), and text reviews. Furthermore, the customer can add an additional text review if desired. Because the text reviews are written according to the specified template, which is consistent with the multidimensional ratings, we need not use text mining techniques to analyze them. Figure 2 shows a sample of eWOM for Toyota RAV4 from Autohome.

To help customers make a horizontal comparison among various car models, Autohome provides the overall rating and review volume for each model; the overall rating is the average value of multidimensional ratings from all customer reviews of a car model. Figure 3 presents a sample webpage that shows the overall rating and the total number of customer reviews for the Toyota RAV4.

The car sales data are collected from the China Securities Market & Accounting Research (CSMAR) database. The CSMAR is a comprehensive research-oriented database focusing on Chinese Finance and Economy. CSMAR was developed by Shenzhen CSMAR Data Technology Co., Ltd. based on academic research needs, meeting the international professional standards while adapting to China’s features.

The macroeconomic data are collected from the National Bureau of Statistics of China (NBSC). The macroeconomic data include the average resident disposable income, vehicle purchase tax rate, and oil price.

The car sales data are monthly data, and the online WOM data are daily data. These data are integrated monthly, and models with less than one year of sales data are deleted. After screening and cleaning, the dataset spans the time range of January 2013 to October 2017. Finally, we obtain a panel dataset with 14,756 records, including 409 family car models. For each car model, the sales data include at least 12 months and up to 58 months, and the overall average is approximately 26 months (more than 2 years).

3.2. Variables

The dependent variable is car sales. We use the actual sales of a given vehicle in month t, which can be obtained from the CSMAR database, to measure car sales.

The independent variable is eWOM volume. In this research, we use the number of online car reviews in the concurrent month to measure volume.

The moderating variables are the rating and variance of eWOM. Rating is measured as the average value of all ratings for a given car in each month. Variance is measured as the variance of eWOM ratings for a given car in each month.

In addition to the independent and moderating variables, we control variables that may impact car sales. Prices and consumer ratings for dealers, which may impact the consumers’ purchase decision, are accounted for in the model. To control for production diffusion, car ages are included in our model. Moreover, we use the Baidu search index to control for demand shocks (e.g., advertising). Finally, the macroeconomic environment also influences car sales, so macroeconomic data, including average resident disposable income, oil price, and car purchase tax rate, are also included in the research model. The detailed variables are summarized in

Table 3. Definitions of main variables and data sources.

Variable	Measure	Data Source
Dependent Variable
$Sale{s}_{it}$	Sales of car i in month t.	CSMAR
Independent Variable
$Volum{e}_{it}$	Volume of eWOM of car i in month t.	Autohome
Moderating Variables
$Ratin{g}_{it}$	Average of eWOM ratings of car i in month t.	Autohome
$Varianc{e}_{it}$	Variance of eWOM ratings for car i in month t.	Autohome
Control Variables
$Ag{e}_{it}$	Months from the launch month to month t of car i.	Autohome
$SearchInde{x}_{it}$	Search volume on Baidu for car i in month t.	Baidu
$Pric{e}_{it}$	Average of actual selling prices of car i in month t.	Autohome
$DealerRatin{g}_{it}$	Average of ratings for dealers of car i in month t.	Autohome
$Incom{e}_t$	Average resident disposable income in month t.	NBSC
$Ta{x}_t$	Vehicle purchase tax rate in month t.	NBSC
$PriceOi{l}_t$	Oil price in month t.	NBSC

.

3.3. Model Specification

To test our hypotheses about the relationship between electronic word mouth and car sales, we first develop a multivariate regression model, which is shown in Equation (1), to analyze our data set. All the variables in Equation (1) are defined in Table 3.

$$\begin{array}{cc}Sale{s}_{it}& ={\beta }_0+{\beta }_{1}Volum{e}_{it}+{\beta }_{2}Ratin{g}_{it}+{\beta }_{3}Ag{e}_{it}+{\beta }_{4}Varianc{e}_{it}\\ & +{\beta }_{5}Volum{e}_{it}*Ratin{g}_{it}+{\beta }_{6}Volum{e}_{it}*Varianc{e}_{it}+{\beta }_{7}Pric{e}_{it}+{\beta }_{8}DealerRain{g}_{it}\\ & +{\beta }_{9}SearchInde{x}_{it}+{\beta }_{10}Incom{e}_t+{\beta }_{11}Ta{x}_t+{\beta }_{12}OilPric{e}_t+{\epsilon }_i+{\epsilon }_t+c\end{array}$$

(1)

Prior studies have found that eWOM not only drives retail sales, but is also an outcome of retail sales [5,26]. This dynamic relationship raises concerns regarding the endogeneity between eWOM and sales. According to Gu et al. [3], major factors contributing to the endogeneity problem are product characteristics and demand shocks, which have the potential to affect both product sales and eWOM.

To address the endogeneity problem, we further employ a two-way fixed effect model to control for both product fixed effects and time fixed effects. We include product fixed effects ε_i to control for both observed and unobserved product characteristics and time-specific fixed effects ε_t to control for time-variant factors (e.g., seasonality). In addition, following the approach by Gu et al. [3], we use the Baidu search index as a control variable to control the demand shocks (such as those caused by advertising).

4. Results

4.1. Descriptive Statistics

Descriptive statistics for the main variables in this study are provided in

Table 4. Descriptive statistics.

Variable	Obs#	Mean	Std. Dev.	Min	Max
Sales	14,756	5578.15	6909.74	−173	80,495
Volume	14,756	47.35	59.88	1	823
Rating	14,756	4.21	0.31	1	5
Variance	14,756	0.21	0.15	0	2.84
Price	14,756	13.37	9.13	2.4	79.9
Age	14,756	60.97	48.55	0	214
DealerRating	14,756	4.19	0.39	1	5
SearchIndex	14,756	127,044.7	194,792.5	66	5,823,991
Income	14,756	1841.45	261.44	1347.67	2394.67
Tax	14,756	0.08	0.02	0.05	0.1
OilPrice	14,756	7329.71	944.61	5965	8900.97

. We report the original values to provide a more intuitive description of our data set. As shown in Table 4, the sales volume may be negative numbers due to purchase returns. The mean rating is 4.21, which means that most car consumers are quite satisfied with their cars.

Table 5. Correlation matrix and VIFs of main variables.

Variable	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)
(1)	1.00
(2)	0.61	1.00
(3)	0.14	0.24	1.00
(4)	0.04	−0.03	−0.29	1.00
(5)	0.04	0.03	0.31	−0.17	1.00
(6)	0.20	0.08	−0.27	0.09	0.02	1.00
(7)	0.06	0.12	0.54	−0.20	0.15	−0.06	1.00
(8)	0.34	0.29	0.18	0.01	0.12	−0.03	0.08	1.00
(9)	0.02	−0.04	0.11	0.06	0.02	0.01	0.07	0.11	1.00
(10)	−0.03	−0.01	−0.07	−0.06	−0.01	−0.01	−0.01	−0.09	−0.50	1.00
(11)	−0.01	−0.02	−0.07	−0.06	−0.00	−0.01	0.00	−0.11	−0.60	0.70	1.00
VIF		1.20	1.94	1.13	1.16	1.15	1.47	1.14	1.65	2.05	2.39

Notes: (1): Sales; (2): Volume; (3): Rating; (4): Variance; (5) Price; (6): Age; (7): DealerRating; (8): SearchIndex; (9): Income; (10): Tax; (11): OilPrice.

provides the correlation matrix and variance inflation factor (VIF) values for the main variables in our study. As shown in Table 5, the correlations between each pair of variables are small, except for the correlation between interior and comfort (0.74) and between interior and exterior (0.73). To further test for multicollinearity, we calculate the VIFs for all the independent variables. Table 3 shows that the maximum VIF is below 5, indicating that no strong multicollinearity exists in the data set [44].

4.2. Panel Data Analysis

To test our assumptions, we estimate our research model shown in Equation (1) using a two-way fixed-effect panel data model. To highlight the importance of the joint effect of different dimensions of eWOM, we compare three specifications of our model.

Table 6. Effects of eWOM on car sales.

Variable	Model 1	Model 2	Model 3
Volume	29.406 (0.737) ***	306.719 (10.470) ***	217.519 (14.552) ***
Volume*Rating		−61.674 (2.323) ***	−43.971 (3.068) ***
Volume*Variance			67.809 (7.705) ***
Rating	496.512 (128.714) ***	1109.333 (127.766) ***	951.389 (128.684) ***
Variance	22.119 (174.604)	−240.444 (170.753)	−1002.331 (191.041) ***
Price	−54.956 (18.781) **	−10.079 (18.413)	−14.8406 (18.372)
Age	7.787 (15.595)	45.486 (15.291) **	40.921 (15.259) ***
DealerRating	−155.230 (77.453) **	−106.844 (75.639)	−111.701 (75.440)
SearchIndex	0.002 (0.000) ***	0.001 (0.000) ***	0.001 (0.000) ***
Income	−0.810 (1.128)	−2.280 (1.103) **	−2.133 (1.100) **
Tax	−13,571.47 (1725.394) ***	−12,810.39 (1684.752) ***	−12,075.91 (1682.346) ***
OilPrice	0.441 (0. 043) ***	0.536 (0.042) ***	0.536 (0.042) ***
Intercept	3633.444 (1540.727) **	−84.300 (1510.722)	695.564 (1509.311)
Time fixed effect	Yes	Yes	Yes
Car Model fixed effect	Yes	Yes	Yes
Obs#	14,756	14,756	14,756
R-squared	0.187	0.225	0.229

Notes: Standard errors are shown in parentheses; ***: p < 0.01, **: p < 0.05.

presents our estimation results.

According to Model 1, Model 2, and Model 3, the eWOM volume and rating are positively related to car sales. Therefore, the H1a and H1b are supported. However, H1c is partially supported. Although the coefficient of variance is not significant in Model 1 and Model 2, but it is significantly negative in Model 3, which means that a higher variance of eWOM ratings hurts car sales when the volume of eWOM is lower than 15.

The moderating effects of eWOM rating and eWOM variance are captured by the interaction terms. According to the results of Model 2 and Model 3, the interaction terms between eWOM volume and eWOM rating are significantly negative, indicating that the eWOM rating reduces the effect of eWOM volume on car sales, supporting H2. In Model 3, the interaction terms between eWOM volume and eWOM variance are significantly positive, indicating that eWOM variance enhances the effect of eWOM volume on car sales, supporting H3.

In terms of the control variables, search index, car age, and oil price are positively related to car sales; however, price and tax are negatively related to car sales. The estimated results are showed in

Table 7. Hypotheses test results.

Hypotheses	Test Result
H1a: The volume of eWOM is positively related to product sales for automobile industries.	Supported
H1b: The valence of eWOM is positively related to product sales for automobile industries.	Supported
H1c: The variance of eWOM is negatively related to product sales for automobile industries.	Partially Supported
H2: The rating will weaken the relationship between the volume of eWOM and product sales for automobile industries.	Supported
H3: The disagreement among customers’ evaluations will strengthen the relationship between the volume of eWOM and product sales for automobile industries.	Supported

.

5. Conclusions, Implications and Limitations

5.1. Conclusions

As a supplement and extension of existing eWOM studies that focus on books, movies, hotels, and restaurants, we investigate the influence of eWOM on car sales utilizing car reviews collected from Autohome. The moderating effects of eWOM rating and variance are also investigated. Based on a stepwise regression analysis with panel data, some interesting findings are obtained and summarized as follow.

First, in the context of the Chinese automobile industry, both volume and valence of eWOM are positively related to sales. The results are consistent with most of the prior literature. This means the awareness effect and persuasion effect are effective in the automotive industry. In other words, the large volume and high rating of eWOM are signals of high quality, which reduce the potential buyers’ uncertainty about the car and increase the car sales.

Second, like the findings of Sun [29], we find that the variance of the eWOM rating is not related to sales when being considered separately. However, a higher variance of eWOM ratings for a car leads to lower sales if and only if the volume of eWOM is low. Notably, if a researcher ignores the interaction, he may reach an imprecise conclusion.

Third, contrary to the research results of Kostyra et al. [51], who selected the eBook reader as their subject for their experiment, and Lu et al. [41], who selected the hotels as their research object, we find that the eWOM valence negatively moderates the relationship between eWOM volume and product sales in the context of the automobile industry. It verifies that the joint effect of volume and valence of eWOM is different for different product categories.

Finally, we find that the variance of eWOM ratings strengthens the effect of eWOM volume on car sales, which is consistent with our assumption that inconsistent opinions will increase the potential consumers’ perceived product uncertainty and make them read more eWOM to avoid risk.

This research makes theoretical contributions by enriching the literature on eWOM. First, this study provides a potential explanation of the previous mixed results regarding the relationship between eWOM and product sales. In addition to showing the direct effect of eWOM on sales, our findings further illustrate the complicated mechanism of different dimensions of eWOM when consumers make purchasing decisions. Second, we bridge the gap in previous studies by focusing on a special product category that is representative of durable products. Because there are significant differences in information search behavior between different categories of products in the purchase decision-making process, it is necessary to consider product categories and understand online consumer information search behavior and its impact on retail strategy.

5.2. Managerial Implications

This research provides several valuable managerial implications for the Chinese automobile industry. First of all, our findings reaffirm the importance of eWOM in marketing. Large volume and positive reviews will increase the awareness and publicity of the company and attract more attentions from potential customers. Hence, car marketers should emphasize the importance of eWOM and motivate satisfied customers to share their experience.

Secondly, in general, we may think that high disagreement about a products’ quality hurt sales. However, it is not meaningful for all possible situations. In our study, we find that a higher variance of eWOM ratings for a car leads to lower sales only when the eWOM volume is low. The result suggests that car marketers should invest more to increase sales by stimulating consumers to post eWOM.

Thirdly, the empirical result of the joint effect suggest that firms can make better use of the eWOM volume by conducting eWOM management tactically, specifically, for products that have lower average rating and higher variance of eWOM ratings. The result once more illustrates that motivating the car buyers to post reviews is essential for car marketers.

In summary, the most important thing for eWOM management in the automotive industry is to encourage satisfied consumers to publish eWOM, to give full play to the awareness and persuasion effect of eWOM.

5.3. Limitations and Future Research

Our study still has some limitations. First, making a purchasing decision is a complicated process, and many factors exert combined effects. However, we only concentrate on the joint effect of different dimensions of eWOM on product sales. Future studies may examine the joint effect of eWOM with other factors, such as product features, platform features, characteristics of consumers, and so on.

Second, the available data limit our research to cars, a durable product category. We cannot generalize the results to all product categories. Replications of the study with other product categories would be useful.

Third, many eWOM platforms allow for a written statement to back up the rating. Hence, both quantitative and qualitative eWOM are available to the customers. In our study, we only focus on the quantitative features of eWOM. Thus, extending our analysis with qualitative eWOM would be interesting.

Finally, our research is based on Chinese eWOM data. China, in contrast, is culturally very different from the West. The cultural differences have been shown to impact a wide range of marketing considerations. Hence, the joint effect of eWOM and cultural factors may be addressed in future research.