The Impact of Negative Reviews on Online Search and Purchase Decisions

Abstract

Despite evidence indicating the significant influence of online reviews on purchase decisions, even after considering a product's average rating ( Vana and Lambrecht 2021), the underlying factors behind this effect and the broader impact of reviews on consumer decision making remain uncertain. This study uses clickstream data from a major online retailer to explore how negative reviews affect consumer search and purchase decisions. Leveraging exogenous variation created by the display of online reviews sorted by recency, the authors find that negative reviews significantly reduce a product's purchase probability because they (1) contrast with the often-high average product rating, (2) decrease the probability that consumers continue browsing for information about the focal product, (3) increase the probability of visiting the page of substitute products, and (4) increase the probability of viewing reviews about substitute products. Importantly, these effects apply to utilitarian products but not hedonic products and when reviews pertain to product functionality or customer service but not to taste-related factors. The authors estimate a product's vulnerability to negative reviews along two dimensions—purchase and search probability for substitutes—and display these effects on a two-dimensional map.

Keywords

customer reviews online shopping information search elasticity of demand topic modeling

Online retailers commonly post customer ratings and reviews on their websites. Online shoppers frequently refer to reviews for guidance at multiple stages of their online journey to obtain information and to evaluate, compare, and decide among various purchase options (Mudambi and Schuff 2010). Several studies have shown that reviews and ratings are regarded by consumers as highly credible inputs in purchase decisions (Brown et al. 2005; Chen and Xie 2008; Floyd et al. 2014; Kozinets et al. 2010; Liu 2006), and consumers often trust reviews more than advertising (Cheong and Morrison 2008; Hung and Li 2007).

When consumers shop online and browse through product listings, they are typically presented with only a limited amount of information, such as the brand name, price, number of reviews, and average product rating. However, since this limited information is often not enough for them to make an informed purchasing decision, they seek out more detailed information by clicking through to individual product pages (Fung and Lee 1999). On these product pages, they find a wealth of additional information that can help them make a more informed decision. For example, they can read individual customer reviews, see detailed product descriptions, view high-quality product images, and even watch product videos.

While online shoppers clearly take product page information into account, previous research has focused on the impact on choice of aggregate measures, for example, the average product rating, rather than on the role of single reviews (see, e.g., Babić Rosario et al. 2016). One exception to this, as demonstrated in a recent article by Vana and Lambrecht (2021), is that an individual online review can have a significant impact on purchase probabilities, even after accounting for the average rating of the product. This effect is moderated by the position in which the review is displayed, how much it contrasts with the average rating, and how much variance exists in ratings.

Our study builds on this line of research by focusing on the impact of negative reviews on online consumer decisions. We do so for several reasons. In psychology, the concept of negativity bias has been widely demonstrated. Bad emotions, bad experiences, and bad feedback have greater impact than good ones (e.g., Baumeister et al. 2001). Negative information tends to influence evaluations more strongly than positive information (Ito et al. 1998), and distinguishable motivational systems have been shown to underlie the assessment of negative and positive stimuli (Cacioppo, Gardner, and Berntson 1997). When subjects are asked to process multiple pieces of information with different valence about the same item or person (e.g., Fiske 1980; Klein 1996; Skowronski and Carlston 1989), negative information has a more significant impact on their final impression. In fact, negativity bias has been found to exist not only in humans but even in animals (Rozin and Royzman 2001). In the choice domain, negative reviews significantly impact a consumer's propensity to choose the item due to an increased perception of risk (Angie et al. 2011), especially when accompanied by pictures (Xie et al. 2011). Additionally, recent work has shown an increase in the frequency of fake reviews, particularly of positive reviews that are used to increase the product's average rating and number of reviews (He, Hollenbeck, and Proserpio 2022; Salminen et al. 2022). Facing this situation, consumers are likely to place more weight on the frequently genuine negative reviews.

We study the effects of negative reviews at three stages of the buying process. First, at the initial information stage, consumers collect information such as the average product rating and the number of available reviews before visiting the product page. From this prior information, usually displayed on a category page, consumers form an initial impression that likely moderates the relevance of negative reviews in the next decision steps. Second, at the product information stage, consumers collect more detailed information about products by visiting their pages. We test whether the presence of one or more negative reviews—in place of positive reviews—affects these information search decisions. Third, we evaluate whether these mechanisms jointly translate to stronger or weaker effects of negative reviews at the purchase stage and, if so, whether this effect is moderated by the type of product and the content of negative reviews.

To empirically study the effects of negative reviews, we use a data set on consumer decisions at a large online retailer offering a broad selection of products in the technology and home-and-garden categories in the United Kingdom. This particularly rich data set tracks multiple steps of the buying process, allowing us to study how the replacement of one or more positive reviews with the respective number of negative reviews affects search and purchase decisions.

Our identification strategy relies on the retailer's “newest first” policy to display customer reviews. When a new review is posted, the oldest review on the product page is relegated to the second page of reviews that most consumers do not visit. As different consumers see the product page at different times, that is, before versus after the arrival of the newest review and respective relegation of the oldest to the second page, this variation in reviews creates a quasi-natural experimental setting that can be leveraged for our research purposes. Specifically, we compare decisions by consumers who arrived at the product page, scrolled down to the bottom of the product page, where up to five reviews are displayed, and found n negative reviews and 5 − n positive reviews, with decisions by consumers who did likewise but instead saw n′ ≠ n negative reviews and 5 − n′ positive reviews. A negative review is defined as a review with three stars or fewer, out of five stars, and a review is considered positive when it has four or five stars.

Our results show that for consumers who scrolled down to browse product reviews, the probability of purchasing a product decreases by 41.8% when one negative review is present compared with when no negative review is displayed. As these consumers represent about one-fifth of all consumers visiting the product page (the remaining four-fifths do not scroll down to read reviews), this effect approximately represents an 8.4% drop in the overall demand. Moreover, a single negative review on the product page increases the probability of further browsing for substitute products by 9.7%. Consumers who view more than one negative review, in place of positive ones, are even more likely to search for substitutes.

We also discover insights into the circumstances under which reviews have the most impact. Specifically, we find that negative reviews have a greater (lesser) effect on the purchase decisions of utilitarian (hedonic) products, consistent with the findings of Sen and Lerman's (2007) experimental study. This difference may be explained by the fact that preferences for hedonic products are more diverse, as personal taste plays a more significant role, while preferences for utilitarian products are more uniform. Hence, consumers can learn more from previous user experiences in utilitarian product categories (Feick and Higie 1992). Motivated by these findings, we take a closer look at the text of the reviews. Using latent Dirichlet allocation (LDA)—a topic modeling approach—we find that negative reviews have a stronger effect on the purchase decision when they describe problems related to the product's functionality or to customer service. In contrast, negative reviews do not have a significant impact when they relate to matters of taste, such as product design and colors.

The closest work to ours is that of Vana and Lambrecht (2021), who quantify the impact of individual online reviews on purchase decisions. Similarly to our work, the authors focus on the decisions of consumers who scroll down to the review section on the product page. Their approach also benefits from the same variation in the display of reviews. However, their article diverges from ours in that they primarily focus on positive reviews and on measuring how the impact of individual reviews on purchase decisions changes with the review's position (first to fifth) on the product page. In contrast, we study the effect of negative reviews and further explore the factors that lead consumers to react strongly to these reviews. We do so by looking at information search behavior alongside purchase decisions, and by analyzing the text of the reviews to identify the conditions where reviews matter the most.

Our research contributes to the literature on online word of mouth (WOM) and customer reviews. The relationship between online WOM and sales has been the subject of various articles and three meta-analyses (Babić Rosario et al. 2016; Floyd et al. 2014; You, Vadakkepatt, and Joshi 2015). Research has shown that reviews are considered by consumers to be compelling sources of information. For example, Chevalier and Mavzlin (2006) find that an additional one-star review negatively impacts the sales rank of a listed book using data from Amazon and Barnes & Noble, and Sun (2012) shows that a higher standard deviation of historical ratings improves a book's sales rank on Amazon when the product's average rating was low. Wu et al. (2015) analyze how consumers update their restaurant preferences based on online reviews depending on their similarity to the review writer, and Liu, Lee, and Srinivasan (2019) use a supervised deep learning algorithm to investigate how the content of reviews affects sales.

We add to this literature by quantifying the impact of negative customer reviews on product demand, formation of consideration sets, and information search decisions. To the best of our knowledge, our research is the first to document the effect of individual reviews on both information search and purchase decisions. Our findings advance the debate on whether consumers use a simultaneous search framework when browsing for products—setting a predefined number of options to look at before starting search decisions (De los Santos, Hortacsu, and Wildenbeest 2012)—or if they carry out search in a sequential way—when learned information about searched products can affect the consideration set (Honka 2014; Honka and Chintagunta 2016; Kim, Albuquerque, and Bronnenberg 2010). By demonstrating that negative reviews prompt consumers to intensify their search for alternative products, our study suggests that consumers tend to search sequentially. We also extend the growing body of work on how consideration sets are formed and how marketing efforts influence the range of products that consumers are willing to consider (Allenby and Ginter 1995; Demuynck and Seel 2018; Eliaz and Spiegler 2011; Manzini and Mariotti 2018; Van Nierop et al. 2010) by quantifying the effect of negative reviews on the size of the search set.

Our research has substantial implications for practitioners. To illustrate the managerial usefulness of our findings, we provide a visualization for the product's exposure to negative reviews on a vulnerability map. For the set of consumers who scroll down to the review area, a two-dimensional map displays the effect of one additional negative review (replacing a positive review) on (1) the product's purchase probability and (2) on the probability that consumers will visit the page of substitutes after viewing the focal product’s reviews. With the growing importance of customer reviews in online shopping, we propose the concept of elasticity of category sales to negative reviews: the percentage change in the sales of a product category due to a 1% increase in negative reviews accessible on the product pages of that category. We apply this metric to a diverse set of product categories and find that this elasticity is greater for categories where consumers browse for more products before purchase.

The rest of the article is organized as follows. In the next section, we present the conceptual framework on the role of negative reviews over the purchase process. We subsequently introduce our research design and identification strategy. We then discuss results and managerial implications. The last section summarizes and provides future research directions.

Conceptual Setting

In this section, we introduce the proposed framework for understanding the role played by negative reviews on search and purchase decisions in an online environment. Figure 1 displays the sequence of stages of the online consumer journey studied in this article.

Figure 1.
Framework for Evaluating the Impact of Negative Reviews on Consumer Choice.

Prior Information

Before evaluating individual reviews, potential buyers usually consider the product's average customer rating. This value is typically defined on a continuous scale of one to five and is an important piece of prior information that impacts a consumer's impression of the product (e.g., Chen and Xie 2008; Floyd et al. 2014). When consumers encounter a high average customer rating for a product, it creates a positive impression and leads to more clicks to the product page, with an optimistic expectation about the remaining information. Once the consumer clicks on the product page and scrolls down to the review area, positive reviews will confirm the initial belief of quality, enhancing the overall impression of the product and increasing the probability of purchase. Conversely, encountering negative reviews for a product with a high average customer rating will contradict the consumer's initial favorable belief, leading to a worse overall impression of the product. The contrast between the prior perception of quality (average rating) and the individual review scores is likely to increase the salience of the reviews and affect the decisions to collect further information (e.g., reading more reviews) and to purchase the focal product. Therefore, we expect that negative reviews will have a greater impact on search and choice decisions when the prior information is more favorable.

Effects of Negative Reviews on Search Decisions

By updating the perception of a product's quality, online reviews influence consumer decisions to gather information before the purchase decision (Bigne, Chatzipanagiotou, and Ruiz 2020; Zhang et al. 2014). Negative reviews, in particular, have the potential to degrade the consumers’ perception of the focal product and prompt them to seek out substitutes. In our context, we expect the presence of a negative review to extend the duration of search because the additional information on the product page reduces the perceived utility of the product being evaluated. The finding would be consistent with sequential search frameworks (e.g., Kim, Albuquerque, and Bronnenberg 2010, 2016), suggesting that consumers tend to explore products one at a time, gathering information as they go. In contrast, simultaneous search models assume that consumers sample a fixed number of products and do not add further options based on information collected during the search process (e.g., De los Santos, Hortacsu, and Wildenbeest 2012).

Moreover, when faced with negative reviews that do not agree with a high average rating, consumers have evidence that it is useful to update prior beliefs by collecting more information. This is likely to make them proceed with the search and peruse reviews of alternative products. In essence, viewing a negative review is expected to prompt consumers to expand their consideration set, indirectly lowering the likelihood of the focal product being the final choice.

Moderating Effect of Product Type and Review Topic

Previous literature has established that consumers apply different information-processing strategies depending on product type and that there are distinct hedonic and utilitarian shopping value dimensions (Babin, Darden, and Griffin 1994; Voss, Spangenberg, and Grohmann 2003). Hence, it is likely that the relevance of negative reviews differs by product type. For example, when searching for hedonic products, López and de Maya (2012) observe that consumers in a negative mood show attitudes and purchase intentions consistent with the valence of the review. However, they find that consumers in a positive mood find arguments (and counterarguments) to compensate for negative information and remain unaffected by negative reviews. Consumers often attribute the negative content of the review to the reviewer's own personal taste in a hedonic context. In contrast, for utilitarian products, consumers attribute the reviewer's opinion to product features, rendering negative reviews more instructive. In a related vein, a study by Sen and Lerman (2007) investigating the influence of negative information on consumer choice detects a negativity bias solely in utilitarian product categories. Hence, we expect to find that negative reviews matter more for utilitarian products than for hedonic ones.

The topic of the negative review may also influence its effect on choice. While consumers could potentially disagree on a product’s taste-related attributes (Markopoulos and Clemons 2013) (e.g., aspects of design, color, shape, material, or size), they might easily agree on attributes related to its functionality (statements such as the product “does not turn on,” “breaks apart,” “is not powerful”) or service quality (e.g., delivery, customer support, guarantee). Learning about the product's mismatch with a previous consumer suggested by a low rating on taste-related attributes does not eliminate the possibility that it could still be a good match for the consumer contemplating the purchase. Indeed, Sun (2012) shows that a greater variance in historical product ratings can increase sales for products with a low average rating. Conversely, learning that a product fell short on a functional attribute is likely to prompt doubts about whether it will fulfill its core purpose. Accordingly, we posit that negative reviews are of greater consequence when their topic relates to product functionality but are less relevant when they describe taste-related aspects.

Effects of the Evaluation of Negative Reviews on Purchase Decisions

In summary, we anticipate that negative reviews will reduce the likelihood of a purchase. This effect is expected to be stronger when prior information—in the form of average consumer rating—is in contrast to the negative review. It is also expected to be more pronounced as the number of alternative products considered increases due to the presence of negative reviews. Furthermore, the impact of negative reviews is expected to be greater when the review content and product type relate to vertically differentiated components.

Data and Identification

In this section, we first provide details about the data set used in our empirical setting. We then give an intuitive explanation of our identification strategy, which includes model-free evidence on how negative reviews impact consumer decisions.

Data

We use data on consumer search and purchase decisions at a large online retailer based in the United Kingdom, covering the period between February 1 and March 31, 2015. The data set includes clickstream information on individual visits to web pages of home-and-garden and technology products, with each product being defined as a stockkeeping unit. The retailer sells both utilitarian products—such as microwaves and juicers in the technology department and mops and bath mats in the home-and-garden department—and hedonic products, such as televisions and consoles as well as pet toys and beach huts. The products are classified by the online retailer into 627 categories. We observe 31,283 product pages that were visited by at least one consumer during the analysis period, of which 14,890 product pages had at least one review. For each product, we observe reviews, prices, product category and brand, and page visits and purchases during the two months.¹

For our analysis, we define a search session as a visit that starts when a consumer first opens a product page and ends either with a purchase or with the last observed click in that category. Altogether, in our data set, 121,391 unique consumers carried out 217,177 search sessions. They visited 400,126 product pages during these sessions, viewed product reviews 74,383 times, and made a purchase in 9,502 of the cases; 64% of the consumers clicked on exactly one product page, 18% clicked on two product pages, and the rest clicked on three or more product pages.² Assuming that clicking on a product page implies consideration, 1.85 is the average consideration set size in the data.

In terms of reviews, there were 575,084 product reviews with a respective rating available on the website. Products were rated on a one-to-five discrete scale represented by stars, with the following distribution: 57% of the reviews had a five-star rating, 26% had a four-star rating, 6% had a three-star rating, 3% had a two-star rating, and 8% had a one-star rating. We classify a review as negative if it has three stars or fewer.

While browsing a category page, the product's price, average rating, and number of reviews are readily visible without the need to access the specific product page. Upon clicking on the product page, consumers are presented with two distinct sections (Figure 2). The product description section is immediately visible when the page is opened. It repeats the information from the category page and provides further details about the item, such as technical specifications, size, and color. In contrast, individual reviews are not immediately visible upon arrival at the product page; consumers must scroll down to view them. We denote the area allocated to reviews on the product page as the review area. The number of reviews displayed on the product page is either equal to the total available reviews or capped at five if the product has at least five reviews. By default, the reviews on the product page are ordered by recency.

Figure 2.
Illustration of the Category Page, Product Page, the Review Area on the Product Page, and Additional Pages of Reviews.

After scrolling down and accessing the review area, the consumer has the option to continue browsing for additional reviews on separate pages, each containing up to 20 reviews. Consumers can also choose to sort reviews by rating and helpfulness after they have scrolled down to the review area on the product page. Overall, consumers decided to browse the second page of reviews 13,614 times and sorted the reviews on 1,106 occasions. In other words, about four out of five consumers chose to only view the reviews on the product page.

We highlight that this setup—reviews ordered by recency and accessed by scrolling down—is typical among retailers. For example, in the United States, reviews on Amazon are displayed at the bottom of the product page and are sorted according to either recency or relevance. On Target.com, reviews are sorted by recency as the default option. Walmart.com shows the most helpful reviews from verified purchases by default, but consumers can sort the reviews by recency. In France, Fnac.fr requires consumers to click on a link to obtain the customer reviews for a product, after which the reviews are sorted by recency as well. At Tesco.com in the United Kingdom, consumers see the newest reviews first, close to the bottom of the product page.

The unit of analysis is a consumer–product pair, that is, a product page visit done at a given time by a unique consumer. For our analysis, we filter the data on two conditions. First, we only include observations where consumers scrolled down to the review section, which is measured by an indicator variable in our data. Second, we retain observations only when the product had at least one customer review at the time of the visit. The resulting data set contains 50,211 search sessions and 68,960 consumer–product observations from 37,032 consumers.³ In this data set, 1,775 purchases were made.

Table 1 presents the descriptive statistics. In our data, half of the consumers viewed at least one negative review on the product page. On average, consumers encountered .84 negative reviews after scrolling to the review area. The mean number of reviews during the consumer visits in our analysis is 122.5, while the average rating is 4.21. The average price is £88.5, with £118.9 for technology products and £74.7 for home-and-garden products.

Table 1.
Descriptive Statistics.

Mean SD Min Max

Percentage of product pages with negative reviews^a 51 50 0 100

# Negative reviews on product page^a .84 1.04 0 5

# Reviews per product^b 122.5 248.9 1 4,761

Average rating per product^b 4.21 .63 1 5

Price per product (£)^b 88.5 115.5 .01 4,199.9

Number of products browsed per consumer^b 2.21 2.17 1 43

^a
Products for which at least one consumer scrolled to the reviews on the product page.

^bConsumer–product pairs where the consumer scrolled to the reviews on the product page.

Identification Strategy

During the two months of our data collection period, the online retailer updated the product pages daily with the publication of reviews. New reviews took the top spot on the review area, while older reviews moved further down or were relegated to subsequent review pages. This sequential arrival of reviews, without managerial influence, creates exogenous variation in the location of reviews on the retailer's website at the time of a consumer’s arrival.⁴ Our identification strategy centers on comparing consumers who arrived at the product page, scrolled down to the review area, and, from the total of N reviews there, found n negative reviews (and N − n positive reviews) with consumers who also scrolled down and saw n′ ≠ n negative reviews (and N − n′ positive reviews).⁵ We also measure the impact of seeing at least one negative review by comparing consumers who scrolled down to the review area when there was at least one negative review with consumers who arrived when there was no negative review, either because none was there or because the negative review got relegated to the second review page that the consumer did not browse. This method has the nature of a regression discontinuity approach, allowing us to compare outcomes between periods with and without the presence of additional negative reviews. For an additional discussion of the identification using this exogenous shock, please see Vana and Lambrecht (2021).

A key assumption underlying our analysis is that consumers are not aware of the individual ratings until they scroll down to the review area. As a test of this assumption, we ran a regression explaining the scrolling decision as a function of the mean of the individual review ratings. Our findings indicate that the decision to scroll down is not significantly explained by the individual review ratings in the review area (Web Appendix, Table W1).

Using the exogenous variation in the number of negative reviews seen by consumers, we examine how their decisions are affected during a search session, such as whether they decide to buy the product and whether they visit pages of alternative products. Additionally, we explore other related consumer decisions, including whether they choose to visit the second page of reviews, sort the focal product’s reviews, and scroll down to the review area of substitute products.⁶

Table 2 displays model-free evidence, conditional on the number of negative reviews observed by consumers arriving at the product page. Each column shows the number and the percentage of consumers who decided to take the respective action. Without additional control variables, and with the caveat that the number of observations is limited when considering the presence of four or five negative reviews, we observe that from five positive reviews (i.e., zero negative reviews) to five negative reviews (i.e., zero positive reviews), there is a significant decrease in the rate of purchase. There is also a decline in the willingness to visit the second page of reviews about the focal product, a decrease in the rate of sorting reviews, an increase in the likelihood of visiting the page of a competing product, and a rise in the willingness to scroll to the review area on the substitute's page.

Table 2.
Model-Free Evidence of the Effect of Negative Reviews on Consumer Choices.

No. of Negative Reviews No. of Arriving Consumers No. of Consumers Deciding to ...

Purchase Product Open Second Review Page Sort the Reviews Search for Substitute Scroll to Reviews of Substitute

0 33,761 998
(3.0%) 5,959
(17.7%) 549
(1.6%) 13,777
(40.8%) 5,883
(42.7%)

1 20,472 518
(2.5%) 3,976
(19.4%) 288
(1.4%) 8,929
(43.6%) 4,062
(45.5%)

2 8,909 181
(2.0%) 1,683
(18.9%) 113
(1.3%) 4,018
(45.1%) 1,998
(49.6%)

3 4,190 63
(1.5%) 754
(18.0%) 48
(1.1%) 1,951
(46.6%) 982
(50.2%)

4 1,229 15
(1.2%) 161
(13.1%) 11
(.9%) 609
(49.6%) 309
(50.7%)

5 399 0
(.0%) 40
(10.0%) 2
(.5%) 189
(47.4%) 77
(40.7%)

Notes: The table displays the number of consumers who decided to take a certain action after viewing the review section, using the sample of consumers who scrolled down on the product page. In parentheses are the percentages of consumers taking the respective decision conditional on the number of negative reviews.

Methodology

We start our methodology description with the model on purchase outcomes. We then move backward and address the different stages in the decision process, which can explain why negative reviews affect product purchase. We also describe the text analysis method we use to study how the review topic moderates the effect of negative reviews.

We chose to evaluate the impact of negative reviews using separate regressions. Although this approach is at odds with recent work on search and choice modeling using integrated frameworks in which the utility of search and choice depend on the same primitives (e.g., Ursu 2018; Ursu, Wang, and Chintagunta 2020), we opted for a reduced-form, staged approach for the following reasons. First, most previous approaches focused on studying only two consumer decisions, whereas our study considers several search decisions and the purchase outcome. We also take into account the moderating factor of prior information and product type. This complexity could make an integrated model lengthy to explore and difficult to estimate. Second, our objective is not to conduct counterfactual analyses but to provide inputs to managers about the sensitivity of each stage of the purchase process with respect to the presence of negative reviews. Consequently, we are first and foremost interested in the effect of negative reviews on the way consumers search for, and react to information.

Effect of Negative Reviews on Product Purchase

To establish the effect of negative reviews on product purchase, we define the dependent variable as the yes/no decision of consumer i regarding the purchase of product j. For the purchase decision, we formulate the model
$Purchas e_{ij} = α_{1} + β_{1} R_{ij} + {x^{'}}_{ij} γ_{1} + {c^{'}}_{i} δ_{1} + μ_{1 j} + η_{1 i} + ϵ_{1 ij},$
(1)
where R_ij is the number of negative reviews on the page of product j that are displayed to consumers i at the time of visit. Note that the website shows the same reviews to different consumers provided that they browse the product reviews before the arrival of an additional review.⁷ Therefore, the coefficient β₁ measures the effect of viewing one additional negative review, in place of a positive review, on the probability of purchase. Consequently, our empirical analysis starts by establishing the overall effect of an individual negative review on the purchase probability of the focal product.

In Equation 1, x_ij is a vector of product page characteristics at the time of the visit by consumer i, including product j's average rating, log number of reviews, and log price as well as the position of the newest negative review on its product page. The term c_i is a vector of variables describing the consumer's search and purchase behavior until viewing the page of product j, including the view rank of j (i.e., how many products consumer i viewed previously) and a dummy variable indicating whether the consumer purchased a product from the same category during our data analysis window. Furthermore, μ_1j is a product-specific fixed effect, η_1i stands for a day fixed effect corresponding to the time of the consumer visit, and ε_1ij is an independent and identically distributed error term. Standard errors are clustered at the product level. For the purchase model, we restrict our attention to consumers who do not open the second review page or sort the reviews, allowing us to compare groups of consumers that are as homogeneous as possible.⁸

We use two alternative definitions for the term R_ij to gain comprehensive insights into the effect of negative reviews and, simultaneously, as a robustness check. In an alternative formulation, R_ij takes the value of 1 if the product page includes at least one negative review when visited by consumer i, and 0 otherwise. In a third formulation, we define four dummy variables, $R_{ij}^{1}$ , $R_{ij}^{2}$ , $R_{ij}^{3}$ , and $R_{ij}^{4}$ , with $R_{ij}^{n} = 1$ if and only if the number of negative reviews is exactly n, for n = 1, 2, and 3, and $R_{ij}^{4} = 1$ if and only if the number of negative reviews is 4 or 5. The base category is zero negative reviews. Note that the interpretation of β₁ depends on which alternative version is used for R_ij. It can capture the marginal effect of the number of reviews, the overall effect of the presence of a negative review on the product page, or the effect of viewing an exact number of negative reviews.

Prior Information

To assess the impact of prior information on consumer decision making, we examine whether products with higher average ratings and more reviews are more frequently selected for further consideration by consumers. Specifically, we model the decision to click on a product page as a function of the information provided on the category page and estimate the model
$\begin{aligned} Visit s_{jt} = & a_{0} + a_{1} \times Avgratin g_{jt} + a_{2} \times Review s_{jt} + a_{3} Pric e_{jt} \\ + \sum_{k = 1}^{K} b_{k} I^{C} (Categor y_{j} = k) + \sum_{l = 1}^{L} b_{l} I^{B} (Bran d_{j} = l) \\ + \sum_{m = 1}^{M} d_{m} I^{D} (Da y_{t} = m) + e_{jt}, \end{aligned}$
(2)
where Visits_jt indicates the log number of clicks on product j on day t, Avgrating_jt is the average consumer rating, Reviews_jt is the log number of reviews, and Price_jt is the log price of product j on day t. The terms I^C, I^B, and I^D are indicator variables for category, brand, and day, and e_jt is an error term. We estimate Equation 2 using product-day observations where at least one visit occurred.

In addition, we divide the sample into two sets using a median split based on the average rating and separately run the purchase decision regression proposed in Equation 1. This approach identifies whether the impact of a negative review on product purchase varies depending on the prior information about the product rating. On the one hand, pages of products with high ratings are likely to be visited more often. On the other hand, the presence of a negative review on the page of a higher-rated product creates a stark contrast between the average rating and the star rating of the negative review. This could lead to a stronger impact of the negative review on the purchase of products with higher average ratings.

Effect of Negative Reviews on Information Search

To examine the impact of negative reviews on outcomes beyond purchase, we employ the regression framework of Equation 1, modifying the dependent variable as follows while keeping the same independent variables.

We explain whether the consumer decided to visit a second page of reviews about product j using
$Ope n_{ij} = α_{2} + β_{2} R_{ij} + {x^{'}}_{ij} γ_{2} + {c^{'}}_{i} δ_{2} + μ_{2 j} + η_{2 i} + ϵ_{2 ij},$
(3)
where Open_ij is a dummy variable indicating whether the consumer opened the second review page of the focal product.

The decision to sort reviews about the focal product is modeled as
$Sor t_{ij} = α_{3} + β_{3} R_{ij} + {x^{'}}_{ij} γ_{3} + {c^{'}}_{i} δ_{3} + μ_{3 j} + η_{3 i} + ϵ_{3 ij},$
(4)
where Sort_ij is an indicator for the sorting decision of individual reviews by any possible criterion. In Equations 3 and 4, β₂ and β₃ measure whether negative reviews increase (or decrease) the consumer's willingness to view further information about the focal product.

Next, we regress a dummy variable, Search_ij, indicating whether consumer i visited the page of at least one competitor after visiting the page of product j, on the same set of covariates as before:
$Searc h_{ij} = α_{4} + β_{4} R_{ij} + {x^{'}}_{ij} γ_{4} + {c^{'}}_{i} δ_{4} + μ_{4 j} + η_{4 i} + ϵ_{4 ij},$
(5)
where β₄ measures the effect of negative reviews on further visits to competing product pages. To keep the sample as homogeneous as possible, we exclude consumers who opened the second page of reviews or sorted the reviews.⁹

Last, we seek to understand whether the consumer who, immediately after viewing the reviews of product j, visited the page of a competitor product j′ ≠ j and scrolled down to the review area of that competitor. Hence, our dependent variable is the yes/no decision of scrolling down to the review area on the page of j′. We then run a regression where the dependent variable is Scroll_ij′, which is an indicator variable taking the value 1 if the consumer scrolled down to the reviews of j′, conditional on visiting j′ after viewing the reviews of j, and 0 otherwise:
$Scrol l_{i j^{'}} = α_{5} + β_{5} R_{ij} + {x^{'}}_{ij} γ_{5} + {c^{'}}_{i} δ_{5} + μ_{5 j} + η_{5 i} + ϵ_{5 ij},$
(6)
where β₅ measures the effect of negative reviews on the willingness to view reviews of the competitor product. These approaches consider all products. However, we also conduct separate analyses for hedonic and utilitarian products to evaluate if the type of product has a notable influence on the effect of negative reviews.

Moderating Effect of Review Topic

We implemented the LDA (Blei, Ng, and Jordan 2003) approach to categorize the negative reviews based on their content in order to study their heterogeneous effects on purchase decisions. The LDA method has previously been successfully applied in marketing research (Tirunillai and Tellis 2014).

We used the natural language processing toolkit NLTK to clean the data, Python's gensim library to create the dictionary of features through which the model is trained, and Google Colab to execute our LDA topic modeling task. To train the algorithm, we used the 25 most recent reviews of all products in our data set regardless of how often consumers scrolled to or viewed these reviews. Although this choice was primarily driven by the task’s computational intensity, it also ensured that we did not bias the training sample by giving excessive weight to the reviews of popular products. This approach also ensured that we used newer reviews, which are more relevant and more often displayed to the consumers. We applied the standard filtering process in which irrelevant terms such as newline characters, single quotes, and stop words were removed from the text of the reviews.¹⁰ Positive reviews (i.e., reviews with ratings higher than three stars) were excluded from the training set. The optimal number of topics was selected using the highest coherence score.

We used the trained algorithm to classify the topic of every negative review. For each review, contributing topics were identified, and among these topics, the dominant one was selected. The dominant topic was defined as the one with the highest percentage contribution to the overall text of the review. We were able to identify four distinct topics among the 11,452 unique negative reviews consumers viewed in our data.¹¹ These three major topics jointly represent 97.1% of the content within the negative reviews. To derive our labels, we took the common approach (Zhong and Schweidel 2020) to examine the most relevant words in the unlabeled categories. In the first category, functionality, we encountered frequent words such as “make,” “well,” “work,” “value,” and “smell.” In the second category, customer service, we noticed frequent words such as “order,” “line,” “phone,” “delivery,” “wait,” “deliver,” “arrive,” “service,” and “available.” In the third category, matters of taste, we observed the words “look,” “cheap,” “poor,” “light,” “nice,” “wood,” “plastic,” “hard,” “feel,” “flimsy,” and “thin” among the most frequently appearing ones.

Overall, we found 19.8% of the negative reviews to have functionality as the dominant topic, 56.7% to have customer service as the dominant topic, and 20.6% to have matters of taste as the dominant topic. The rest of the reviews (2.9%) predominantly addressed other topics and concerns. Negative reviews in the functionality topic typically note that the product failed to meet the expectations of one or more of its core purposes. An example of this topic is a review that included “product … producing a piece of toast that isn’t actually fully toasted.” Negative reviews in the customer service category talk about the product being shipped late, poor customer service, or unexpected stockouts. An example is a review with the text “could not find my order number, no delivery, reordered, phone call, no delivery, phone call to head office … no delivery, they canceled my order.” The matters of taste topic consists of negative reviews that point to the disappointing design or look of the product. An example text of this group contains the following: “the oak effect frame is quite disappointing. It looks cheap because of the laminated covering.” As an illustration, we present the full text of the ten most typical reviews by topic in Tables W2, W3, and W4 of the Web Appendix.

Our strategy for estimating the effect of negative reviews according to topic is similar to the main approach with Equation 1, but instead of having one variable indicating the number of negative reviews, we include one such variable for each of the four categories of reviews. By doing so, we measure the marginal effect of an additional negative review on a particular topic. We also run a second specification in which we use a dummy for the existence of a negative review of a certain topic on the product page.

Results

Product Purchase

Table 3 displays the results related to the purchase probability of product j, for consumers who scrolled down to the review area. As previously described, we present the results for three different models. First, we consider the impact of the number of negative reviews on the product page. Next, we define the effect of a negative review by looking at the presence of at least one negative review on the product page. Last, we quantify the separate effects of having one, two, three, and four or five negative reviews on the product page. For ease of interpretation, the scale of the choice probability is defined between 0 and 100, and the coefficients can be interpreted as percentage point changes. We use the same rescaling for all other binary dependent variables.

Table 3.
Estimates for the Product Purchase Decision.

Model 1 Model 2 Model 3

# Negative reviews −.627* (.162)

# Negative reviews > 0 −1.165* (.358)

# Negative reviews = 1 −.933 (.371)

# Negative reviews = 2 −1.362* (.461)

# Negative reviews = 3 −2.037* (.542)

# Negative reviews = 4 or 5 −2.354* (.781)

Product page information

# Reviews (log) .145 (.293) .085 (.292) .136 (.293)

Average rating .383 (.411) .301 (.428) .278 (.427)

Price (log £) −.588 (1.039) −.609 (1.040) −.598 (1.040)

Newest negative review at position 2 .119 (.279) .335 (.302) .252 (.300)

Newest negative review at position 3 .409 (.366) .745* (.399) .576 (.404)

Newest negative review at position 4 −.144 (.372) .285 (.420) .052 (.440)

Newest negative review at position 5 .580 (.524) .858 (.346) .482 (.544)

Search and previous choice

Second viewed product −.381* (.203) −.388* (.203) −.382* (.203)

Third viewed product −.435* (.260) −.442* (.260) −.436* (.260)

Fourth or later viewed product −.545 (.216) −.547 (.216) −.545 (.216)

Previous purchase in category 24.626* (6.544) 24.691* (6.547) 24.648* (6.546)

Product fixed effects Yes Yes Yes

Day fixed effects Yes Yes Yes

R² .160 .160 .160

# Observations 55,860 55,860 55,860

p < .1. p < .05. p < .01.

Notes: The models present the estimated coefficients from Equation 1 using alternative definitions for the number of negative reviews on the product page at the time of consumer visit (R_ij). The dependent variable is an indicator of product purchase. In Column 1, the effect of negative reviews is measured with the total number of negative reviews appearing on the product page. In Column 2, the negative review effect is measured with a dummy variable indicating whether there is at least one negative review on the product page. In Column 3, dummy variables indicate the number of reviews on the product page. Robust standard errors clustered at the product level are in parentheses.

Looking at the results concerning the main effects of a negative review on purchase behavior, the estimates in Column 1 indicate that an additional negative review—in place of a positive review—on the product page lowers the probability of purchase by .63 percentage points on average, which translates to an average decrease of 26.87% in purchase probability. The results in Column 2 suggest that consumers who visited the product page when a negative review was present have a 1.16 percentage point lower probability of buying the product compared with those who visited the product page at a time when no negative review was present, translating to a 41.80% decrease in purchase probability in our data. The findings presented in Column 3 indicate that the presence of one, two, three, or more negative reviews on the product page results in a significant decrease in the purchase probability, by .93, 1.36, 2.03, or 2.35 percentage points, respectively, when compared with a product page without any negative reviews. These results are in line with the model-free evidence section (Table 2).

In terms of session characteristics, we see that later-searched products are less likely to be purchased and customers who purchased in the category before are more likely to buy again. Although the average rating, number of reviews, and price have the expected signs, they are not statistically significant, which can be explained by the inclusion of product fixed effects. We find no evidence for the position effect of the most recent negative review once the number of negative reviews is accounted for.

Prior Information

To evaluate the impact of prior information on the effect of negative reviews, we first evaluate if consumers are more likely to click on product pages with a higher average customer rating and with more reviews. Table 4 shows the results. We find that the effects of these two variables are positive and significant. We obtain similar estimates without and with brand fixed effects (Columns 1 and 2, respectively), suggesting that the effects of the reviews on clicks hold both across and within brands. Hence, as consumers come to the product page and arrive at the negative review section, they are more likely to have seen positive prior information about the product: better average rating and higher popularity as signaled by the number of reviews.

Table 4.
Estimates for the Number of Daily Visitors as a Function of Information on the Category Page.

(1) (2)

Log Daily Visits Log Daily Visits

Average rating .013* (.003) .014* (.004)

# Reviews (log) .046* (.002) .042* (.002)

Price (log £) −.027* (.005) −.036* (.005)

Brand fixed effects No Yes

Category fixed effects Yes Yes

Day fixed effects Yes Yes

R² .142 .172

# Observations 198,537 198,537

p < .1. p < .05. p < .01.

Notes: The dependent variable is the logarithm of the product's daily number of visits. Only products that have at least one review are included. In parentheses are robust standard errors clustered at the product level.

Next, to see whether prior information has a moderating role in the effect of negative reviews on product purchase, we reestimate Equation 1 separately using products below and above the median average rating. In light of our proposed framework, we expect that negative reviews are more impactful when the average rating is higher. Compared with the coefficient of negative reviews in the previous section (of −.627), we observe that the effect of negative reviews is about half (−.341) and less significant for below-median rating products, while it is almost 50% higher for above-median rating products (−.967) (see Table 5).¹²

Table 5.
Estimates for the Product Purchase Decision for Products with Low Versus High Average Rating.

Products with Low Average Rating Products with High Average Rating

# Negative reviews −.341* (.190) −.967*** (.368)

Product page controls Yes Yes

Consumer controls Yes Yes

Product fixed effects Yes Yes

Day fixed effects Yes Yes

R² .169 .166

# Observations 27,534 28,326

p < .1. p < .05. p < .01.

Notes: The table shows estimated coefficients from Equation 1, using products with either a lower-than-median (first column) or higher-than-median (second column) average rating. The dependent variable is an indicator of product purchase. Product and consumer control variables are included in both models. Consumers who opened the second review page or sorted the reviews are excluded from both models. In parentheses are robust standard errors clustered at the product level.

Information Search Decisions

In this section, we discuss the results regarding different search process decisions, using the estimates from the model with separate effects of one, two, three, or more negative reviews.¹³ In the first column of Table 6, we see that the greater the number of negative reviews, the less likely the consumer will open the second review page. The second column shows that compared with a product page that has only positive reviews, a page with one or more negative reviews is more likely to deter the consumer from sorting the reviews (by rating, helpfulness, or other criteria). These findings suggest that the consumer's motivation to seek further information about the product decreases as the number of negative reviews increases.

Table 6.
Effects of Negative Reviews on Product and Information Search Decisions.

(1) (2) (3) (4)

Opened Second Review Page Sorted the Reviews Searched for Substitute Scrolled to Reviews of Substitute

# Negative reviews = 1 −.652
(.883) −1.070*
(.293) 3.873*
(1.090) .012
(.019)

# Negative reviews = 2 −2.369
(1.022) −.977*
(.349) 5.243***
(1.378) .038*
(.022)

# Negative reviews = 3 −4.155*
(1.474) −1.415*
(.446) 7.128*
(1.789) .059
(.027)

# Negative reviews = 4 or 5
−9.466*
(1.989) −1.439
(.722) 11.101*
(2.897) .172*
(.044)

Product page controls Yes Yes Yes Yes

Consumer controls Yes Yes Yes Yes

Product fixed effects Yes Yes Yes Yes

Day fixed effects Yes Yes Yes Yes

R² .191 .102 .191 .292

# Observations 68,960 68,960 55,860 23,565

p < .1. p < .05. p < .01.

Notes: Columns 1 and 2 show estimates using the sample of consumers who scrolled down to the reviews. Consumers who opened the second review page or sorted the reviews are excluded from the substitute search model (Column 3). The coefficients for review browsing of competitor products are estimated on a subsample of consumers who searched for a substitute after visiting the focal product’s page (Column 4). The dependent variables are indicators of the respective consumer decisions. In parentheses are robust standard errors clustered at the product level.

The third column of Table 6 reveals that the more negative reviews about the focal product a consumer observes, the more willing the consumer becomes to visit the page of a competing product. In fact, even a single negative review on the product page is enough to have a positive and significant impact on the decision to explore alternative items. Consumers who viewed a product page with one negative review (with the rest of reviews being positive) have a 3.87 percentage point larger probability of visiting the pages of competitor products than consumers who did not face any negative reviews, which translates to a 9.65% increase in search probability. Finding additional negative reviews instead of positive reviews has a cumulative effect on the decision to search for an alternative, with the effect almost doubling in the case of three negative reviews out of five.

As an additional analysis of search behavior, we consider how the search decision depends on the stage of search, in this case the number of product pages visited previously to arriving at the focal product’s page. After splitting our sample into early (when no more than two product pages were previously clicked on) versus late (when at least three products were previously clicked on) arrivals to the product page and reestimating Equation 5, we find that negative reviews significantly motivated consumers to search for competitors when the product was visited early on. However, they did not significantly impact consumer search decisions for late visits. The respective coefficients are shown in the Web Appendix, Table W5.

The results in the last column of the Table 6 suggest that the presence of negative reviews on the focal product's page influences the decision to scroll to the review area of the competitor product. As the number of negative reviews increases on the page, consumers become more willing to explore the reviews of substitutes. This suggests that they become more cautious and spend more time evaluating suitable alternatives when they encounter negative information, in order to avoid choosing a product that might disappoint.

Moderating Effects of Product Type and Review Topic

Product type as moderator

To examine the potential differential effects of negative reviews across utilitarian and hedonic product categories, we conduct separate analyses for each category. Product categories are classified as a function of how they are perceived by consumers (Hirschman and Holbrook 1982). Hedonic products are those that provide a sensory experience of aesthetic or sensual pleasure, such as headphones or home decoration, while utilitarian products are primarily practical in nature, like washing machines or printer ink. As a result of their differing natures, hedonic products tend to be more distinct in their horizontally differentiated dimensions, while utilitarian products have a stronger vertically differentiated dimension.

The results in Table 7 show that negative reviews have a significant impact on reducing the likelihood of purchase for utilitarian products but not for hedonic products.¹⁴ This finding suggests that consumers of utilitarian products, which are characterized by vertically differentiated attributes and experiential qualities such as durability and product/service quality, benefit from access to other users’ experiences with the product. This is not the case for hedonic products, where horizontal differentiation and taste differences are more prevalent.

Table 7.
Estimates for the Product Purchase Decision for Utilitarian and Hedonic Product Categories.

Utilitarian Products Hedonic Products

(1) (2) (3) (4) (5) (6)

# Negative reviews −.742*
(.190) −.319
(.316)

# Negative reviews > 0 −1.506*
(.424) −.349
(.708)

# Negative reviews = 1 −1.252*
(.440) −.168
(.735)

# Negative reviews = 2 −1.734*
(.542) −.464
(.903)

# Negative reviews = 3 −2.371*
(.633) −1.340
(1.054)

# Negative reviews = 4 or 5 −2.811*
(.955) −1.056
(1.006)

R² .168 .168 .168 .141 .141 .141

# Observations 40,724 40,724 40,724 15,136 15,136 15,136

p < .1. p < .05. p < .01.

Notes: The table shows estimated coefficients from Equation 1 using alternative definitions for the number of negative reviews on the product page at the time of consumer visit (R_ij). The dependent variable is an indicator of product purchase. Product, consumer, and day fixed effects are included in all models. In parentheses are robust standard errors clustered at the product level.

Review topic as moderator

In this section, we present the results of the previously described LDA approach applied for categorizing the negative reviews according to their text. By doing so, we seek to deepen our understanding of what type of review content is particularly responsible for the decrease in purchase likelihood.

Our results for the product purchase decision (Table 8) indicate that negative reviews describing functionality issues or problems related to customer service have a significant negative impact on purchase probability. In contrast, reviews concerning taste-based issues have no influence on purchasing decisions. This finding is consistent with the discussion of product type, as taste-based attributes are more generally considered.

Table 8.
Estimates for the Product Purchase Decision as a Function of the Dominant Topic of the Review.

(1) (2)

Product Purchase Product Purchase

# Negative reviews

on functionality −.815* (.257)

on customer service −.687* (.213)

on matters of taste −.078 (.331)

on other topics −.957* (.532)

# Negative reviews > 0

on functionality −.819 (.337)

on customer service −1.039* (.324)

on matters of taste .029 (.429)

on other topics −1.073* (.589)

Product page controls Yes Yes

Consumer controls Yes Yes

Product fixed effects Yes Yes

Day fixed effects Yes Yes

R² .160 .160

# Observations 55,860 55,860

p < .1. p < .05. p < .01.

Notes: The dependent variable is an indicator of product purchase. Column 1 shows results from a model in which the independent variables include the total number of negative reviews seen on the product page from a given review topic. Column 2 shows results from a model in which dummy variables indicate whether on the product page there is at least one negative review from a given topic. In parentheses are robust standard errors clustered at the product level.

To ensure that our findings are not solely driven by differences in star ratings between reviews in different topic categories, we examine whether reviews on taste-related topics have significantly higher star ratings than reviews in the other three topic categories. We find that this is not the case. Negative reviews concerning matters of taste have, on average, similar ratings (1.76) to negative reviews on functionality (1.65), while their frequency is almost the same (Web Appendix, Table W7). Since negative reviews related to functionality were found to significantly decrease purchase probability, we can conclude that our estimates are not solely a result of correlation between star rating and review topic. Instead, our findings suggest that the topic of the review does indeed have an impact on purchase behavior.¹⁵

Combining moderators: Prior information and product type

Finally, we look at the interaction effect between prior information and type of product. To do so, we divide the products into above- and below-median groups with respect to (1) the number of reviews and (2) the average customer rating, resulting in four quadrants of products. Next, we run the purchase regression in each quadrant, separately for hedonic and utilitarian products (Table 9). We find that for utilitarian products, negative reviews have a significantly negative impact on purchase probability, particularly when the product has a high average rating based on a relatively low number of reviews. In this scenario, we observe a decrease of 2.9 percentage points in purchase probability. However, when the product has a high average rating and a high number of reviews, the negative impact of negative reviews is halved, resulting in a decrease of 1.4 percentage points in purchase probability. In contrast, for hedonic products where taste and horizontal differentiation are more important, the effect of negative reviews becomes insignificant.

Table 9.
Estimates for the Product Purchase Decision by Prior Information and Product Type.

Variable: No. of Negative Reviews High No. of Reviews and High Average Rating Low No. of Reviews and High Average Rating High No. of Reviews and Low Average Rating Low No. of Reviews and Low Average Rating

Hedonic products 1.46 (1.00) 1.63 (2.04) −.51 (.56) −.17 (.46)

Utilitarian products −1.40* (.48) −2.91 (1.29) .45* (.28) .20 (.50)

p < .1. p < .05. p < .01.

Notes: The table shows estimated coefficients from Equation 1 for hedonic versus utilitarian products using product subsamples according to the four quadrants based on number of reviews and average rating. The dependent variable is an indicator of product purchase. Product and consumer control variables are included in each regression. Consumers who opened the second review page or sorted the reviews are excluded from each regression. In parentheses are robust standard errors clustered at the product level.

Robustness Tests

To ensure that our results are robust and not driven by how we have selected our sample or chosen the model specifications, we carried out a variety of robustness checks. Supporting tables are shown in the Web Appendix.

We have derived our main results using linear probability models. Since it is possible that the results could be dependent on using this specific choice of functional form, we reestimated Equations 1 and 5 using a logistic regression. We observe that negative reviews continue to have a significant impact on purchase and search behavior in the logit specifications (Web Appendix, Table W8). The respective coefficients measuring the impact of negative reviews on consumer behavior are all negative and significant at the 1% level.

Another possibility is an unobserved (to the researcher) correlation between the consumer's purchase and product search decisions. In other words, the purchase and search equations could have a contemporaneous error correlation. If the correlation of errors is significant but assumed to be zero, the ordinary least squares estimation with each equation separately estimated might be biased (Rao 1974). To investigate whether there is significant correlation between the errors of the purchase and search equations, we conducted a seemingly unrelated regression (SUR) analysis that allows for correlated errors (Zellner 1962). We find that the correlation between the residuals is not significant (−.046), and we obtain substantively identical coefficients using the SUR approach as with estimating the equations independently (Web Appendix, Table W9). We conclude that our findings are not biased due to contemporaneous correlation in the error terms.

One of the concerns when identifying the effect of online ratings on demand is the possibility that parameter estimates may be biased due to spurious correlation, that is, when critical reviews are simply evidence of another event occurring at the same time that the review is written (e.g., delivery delays, stockouts). In these cases, the decline in sales may have occurred even without the existence of the negative review, as consumers may have learned about the issue through other sources. Previous research has encountered similar challenges when measuring the effects of reviews on sales. For example, Reinstein and Snyder (2005) warn of spurious correlation between expert reviews and movie ticket sales induced by an underlying correlation in unobservable quality signals. Chintagunta, Gopinath, and Venkataraman (2010) address the same issue concerning movie sales and reviews, and Anderson and Magruder (2012) account for such correlation in the relationship between average Yelp ratings and sales. We test for spurious correlation by examining whether a forthcoming negative review, that is, one that is not yet on the product page but will be there within a week, affects product purchase. If the drop in demand can be attributed to other reasons than the discovery of the negative customer review, a forthcoming negative review would correlate negatively with purchase and positively with searching for substitutes. We find no evidence for any of these effects (Web Appendix, Table W10), suggesting that spurious correlation is not a concern in our data set.

Finally, the time of review creation could be related to the total number of consumers who visit the product page during similar periods. For example, a new advertising campaign on national television or through another medium could play a role. To explore this relationship, we run a regression that explains the log daily (and weekly) number of visitors to the product page at the time of review creation as a function of the rating of submitted reviews alongside our standard set of control variables. Our results indicate there is no significant correlation between the valence of the created review and the number of visitors to the product page around the time of review creation (Web Appendix, Table W11). In other words, we do not find evidence that other unobserved events that could drive both the sentiment of the review and the number of product page visitors affected our results. This further strengthens our confidence in the causal relationship between individual negative reviews and product demand.

Managerial Implications

Mapping Product and Category Vulnerability to Negative Reviews

Using the estimates from the main model (Column 3 in Table 3), we compute the predicted probability of purchase for each consumer–product pair in our data. We then make a change by adding a negative review to the page of each product and compute the new predicted probability of purchase. With these two values, we estimate the impact of the submission of a negative review among consumers who view the reviews.

Formally, we obtain the purchase probabilities with and without posting an additional negative review using the estimated coefficients through
$p_{ij} = E (Purchas e_{ij} | Dat a_{ij}, {\hat{α}}_{1}^{P}, {\hat{β}}_{1}^{P}, {\hat{γ}}_{1}^{P}, {\hat{δ}}_{1}^{P}, {\hat{μ}}_{1 j}^{P}, {\hat{η}}_{1 i}^{P}})$
and
${\tilde{p}}_{ij} = E (Purchas e_{ij} | {Data}_{ij}^{\sim}, {\hat{α}}_{1}^{P}, {\hat{β}}_{1}^{P}, {\hat{γ}}_{1}^{P}, {\hat{δ}}_{1}^{P}, {\hat{μ}}_{1 j}^{P}, {\hat{η}}_{1 i}^{P}}),$
where Data_ij refers to product characteristics at the time when it was visited by consumer i and ${Data}_{ij}^{\sim}$ contains the same characteristics as the original data except for the addition of a negative review.¹⁶ We consider that the addition of a negative review modifies the product's average rating and total number of reviews. For each product in our data, we calculate the within-product change in purchase probability due to the submission of a negative review by averaging the percentage changes across the N consumers:
$Δ_{ρ j} = \frac{\sum_{i} (\frac{{\tilde{p}}_{ij}}{p_{ij}} - 1) \times 100}{N} .$
In addition, we calculate the effect of a negative review on the probability of searching for further substitutes using
$Δ_{ϱ j} = \frac{\sum_{i} (\frac{{\tilde{ϱ}}_{ij}}{ϱ_{ij}} - 1) \times 100}{N},$
where $ϱ_{ij}$ is the probability that consumer i searches for further substitutes without the additional negative review and ${\tilde{ϱ}}_{ij}$ is the probability that consumer i searches for further substitutes with the additional negative review present on the focal product’s page. We denote the $(Δ_{ρ j}, Δ_{ϱ j})$ values as the product's purchase and search vulnerability to a negative review.

The top panel of Figure 3 shows the vulnerability to a negative review for four product categories as an example, using a sample that consists of products that have no previous negative reviews on the product page at the time of consumer visit. Hence, the negative review is the first and only negative review present on the product's page. The consequences are striking, with the probabilities of purchase declining significantly and the probabilities of searching for substitutes increasing across products in all four categories. Most products are doubly hit by the negative review, with both a reduction of the purchase probability and an increase in the probability of consumers considering alternative products.

Figure 3.
Impact of a Single Negative Review on Purchase and Further Search for Substitutes with No Negative Review on Product Page and One Negative Review Present in the Top Position.

We contrast these findings with the impact of the arrival of a negative review when there was already one negative review in the top position in the review area of the product page, using the same set of products as before. The impact of the arrival of this second negative review is displayed in the bottom panels of Figure 3. The arrival of a second negative review still decreases the probability of purchase, albeit with some variability across product categories. However, the impact on product search is significantly reduced both in terms of average magnitude and variability across products. These findings suggest that the primary and strongest impact of negative reviews on product search is due to the first negative review appearing on the product page, with additional negative reviews having a lesser impact on the likelihood of consumers to browse more.

We note that these changes are calculated for consumers who scroll down to the part of the product page that contains the reviews but who do not browse for further review pages. These consumers generated 18.3% of the revenue during the two months in the analysis in the home-and-garden department and 21.7% in the technology department. Assuming that for the remaining consumers the negative reviews have no effect (as they remain unseen), the overall impact would be about one-fifth of the results presented here. Thus, for most products, the drop in the overall demand, that is, among all consumers regardless of whether they view individual reviews or not, due to a single negative review posted in a product page (previously without negative reviews) would be in the range of 5% to 15%.

Elasticity of Category Sales to a Negative Review

In this section, we propose the metric of elasticity of category sales to negative reviews. To obtain this elasticity, we calculate the percentage change in sales of a product category as a function of a 1% increase in the total number of negative reviews on the product pages within that category, using the coefficients from Column 3 in Table 3.

In other words, we increase the negative reviews available on product pages by 1%, added randomly to product pages within the respective category. We record the implied change in sales at the category level due to the addition of these negative reviews and repeat the entire process 10,000 times for a given product category. Figure 4 presents an example, displaying the resulting distribution of percentage changes in two product categories: vacuum cleaners and curtains. We obtain an elasticity of −15.4 for vacuum cleaners and of −1.9 for curtains.

Figure 4.
Simulated Sales Changes Due to a 1% Increase in the Number of Negative Reviews on Product Pages.

To better understand how the negative review elasticity of sales is related to the product category, we regress the elasticity metrics on product and consumer variables (Web Appendix, Table W12). The independent variables are the mean consideration set size of consumers who visit the category, the mean average product rating in the category, the mean log price in the category, the mean log number of reviews for a product in the category, the mean number of negative reviews on product pages in the category, a dummy for hedonic product, a dummy for technology product, the length of the product description, and an interaction term between hedonic and consideration set size.¹⁷

We find that categories with higher prices and longer product descriptions tend to be more elastic, suggesting that consumers are particularly sensitive to negative reviews when shopping for expensive products and in categories where extensive information is available on the product pages. Additionally, we find that the negative review elasticity of sales has a strong correlation with the average consideration set size in the category, defined as the number of products consumers look at during their online session. This suggests that negative reviews have a greater impact on demand as competition increases (i.e., when consumers consider a larger set of products).

Conclusion

The popularity of online shopping and the significance of user-generated content have been increasing over the years. There are various instances where online sellers request customers not to leave negative reviews and offer their services to resolve any issues in exchange. Our study shows that sellers have a valid reason for doing so as even a single negative review can significantly reduce the likelihood of purchase. The impact on demand of multiple negative reviews is even more substantial.

We show that negative reviews not only affect product purchase, but they also lead consumers to search for substitute products. By increasing the consideration set, this extended search decreases the likelihood of the original product being selected at the end of the consumer journey. When consumers find negative reviews, they are more likely to stop browsing for further reviews about the focal product, suggesting that they are unwilling to spend more time and effort collecting information on that particular item.

We also find that negative reviews have varying impacts on consumer behavior depending on the product category and the content of the reviews. In utilitarian product categories, negative reviews have a much greater effect on purchase likelihood compared with hedonic categories. Additionally, negative reviews that focus on product functionality or customer service are more likely to deter consumers from purchasing, while those that discuss taste-based preferences such as design, color, or material have little to no effect. These findings can provide valuable insights for online sellers looking to improve their product offerings and manage their online reputation.

Using our approach, the effect of a single negative review on purchase probability and competitor search probability can be established at the product level and represented on a two-dimensional vulnerability map. We have introduced a metric to quantify the percentage change in sales caused by a 1% increase in the number of negative reviews in the category. This metric could serve as a reference point for those interested in understanding how online WOM affects sales in their relevant category.

We believe that our findings may motivate future research such as modeling how the design of product pages might change what consumers decide to browse for and eventually purchase. Another potentially fruitful direction could be to study whether and how low ratings influence the arrival frequency of new reviews through decreased purchase rates. Furthermore, the literature might benefit from a better understanding of why and under what circumstances consumers decide to view reviews.

Supplemental Material

sj-pdf-1-mrj-10.1177_00222437231190874 - Supplemental material for The Impact of Negative Reviews on Online Search and Purchase Decisions

Supplemental material, sj-pdf-1-mrj-10.1177_00222437231190874 for The Impact of Negative Reviews on Online Search and Purchase Decisions by Marton Varga and Paulo Albuquerque in Journal of Marketing Research

	Mean	SD	Min	Max
Percentage of product pages with negative reviews^a	51	50	0	100
# Negative reviews on product page^a	.84	1.04	0	5
# Reviews per product^b	122.5	248.9	1	4,761
Average rating per product^b	4.21	.63	1	5
Price per product (£)^b	88.5	115.5	.01	4,199.9
Number of products browsed per consumer^b	2.21	2.17	1	43

No. of Negative Reviews	No. of Arriving Consumers	No. of Consumers Deciding to ...
0	33,761	998 (3.0%)	5,959 (17.7%)	549 (1.6%)	13,777 (40.8%)	5,883 (42.7%)
1	20,472	518 (2.5%)	3,976 (19.4%)	288 (1.4%)	8,929 (43.6%)	4,062 (45.5%)
2	8,909	181 (2.0%)	1,683 (18.9%)	113 (1.3%)	4,018 (45.1%)	1,998 (49.6%)
3	4,190	63 (1.5%)	754 (18.0%)	48 (1.1%)	1,951 (46.6%)	982 (50.2%)
4	1,229	15 (1.2%)	161 (13.1%)	11 (.9%)	609 (49.6%)	309 (50.7%)
5	399	0 (.0%)	40 (10.0%)	2 (.5%)	189 (47.4%)	77 (40.7%)

	Model 1	Model 2	Model 3
# Negative reviews	−.627***	(.162)
# Negative reviews > 0			−1.165***	(.358)
# Negative reviews = 1					−.933**	(.371)
# Negative reviews = 2					−1.362***	(.461)
# Negative reviews = 3					−2.037***	(.542)
# Negative reviews = 4 or 5					−2.354***	(.781)
Product page information
# Reviews (log)	.145	(.293)	.085	(.292)	.136	(.293)
Average rating	.383	(.411)	.301	(.428)	.278	(.427)
Price (log £)	−.588	(1.039)	−.609	(1.040)	−.598	(1.040)
Newest negative review at position 2	.119	(.279)	.335	(.302)	.252	(.300)
Newest negative review at position 3	.409	(.366)	.745*	(.399)	.576	(.404)
Newest negative review at position 4	−.144	(.372)	.285	(.420)	.052	(.440)
Newest negative review at position 5	.580	(.524)	.858	(.346)	.482	(.544)
Search and previous choice
Second viewed product	−.381*	(.203)	−.388*	(.203)	−.382*	(.203)
Third viewed product	−.435*	(.260)	−.442*	(.260)	−.436*	(.260)
Fourth or later viewed product	−.545**	(.216)	−.547**	(.216)	−.545**	(.216)
Previous purchase in category	24.626***	(6.544)	24.691***	(6.547)	24.648***	(6.546)
Product fixed effects	Yes	Yes	Yes
Day fixed effects	Yes	Yes	Yes
R²	.160	.160	.160
# Observations	55,860	55,860	55,860

	(1)	(2)
Average rating	.013*** (.003)	.014*** (.004)
# Reviews (log)	.046*** (.002)	.042*** (.002)
Price (log £)	−.027*** (.005)	−.036*** (.005)
Brand fixed effects	No	Yes
Category fixed effects	Yes	Yes
Day fixed effects	Yes	Yes
R²	.142	.172
# Observations	198,537	198,537

	Products with Low Average Rating	Products with High Average Rating
# Negative reviews	−.341* (.190)	−.967*** (.368)
Product page controls	Yes	Yes
Consumer controls	Yes	Yes
Product fixed effects	Yes	Yes
Day fixed effects	Yes	Yes
R²	.169	.166
# Observations	27,534	28,326

	(1)	(2)	(3)	(4)
# Negative reviews = 1	−.652 (.883)	−1.070*** (.293)	3.873*** (1.090)	.012 (.019)
# Negative reviews = 2	−2.369** (1.022)	−.977*** (.349)	5.243*** (1.378)	.038* (.022)
# Negative reviews = 3	−4.155*** (1.474)	−1.415*** (.446)	7.128*** (1.789)	.059** (.027)
# Negative reviews = 4 or 5	−9.466*** (1.989)	−1.439** (.722)	11.101*** (2.897)	.172*** (.044)
Product page controls	Yes	Yes	Yes	Yes
Consumer controls	Yes	Yes	Yes	Yes
Product fixed effects	Yes	Yes	Yes	Yes
Day fixed effects	Yes	Yes	Yes	Yes
R²	.191	.102	.191	.292
# Observations	68,960	68,960	55,860	23,565

	Utilitarian Products	Hedonic Products
# Negative reviews	−.742*** (.190)			−.319 (.316)
# Negative reviews > 0		−1.506*** (.424)			−.349 (.708)
# Negative reviews = 1			−1.252*** (.440)			−.168 (.735)
# Negative reviews = 2			−1.734*** (.542)			−.464 (.903)
# Negative reviews = 3			−2.371*** (.633)			−1.340 (1.054)
# Negative reviews = 4 or 5			−2.811*** (.955)			−1.056 (1.006)
R²	.168	.168	.168	.141	.141	.141
# Observations	40,724	40,724	40,724	15,136	15,136	15,136

	(1)	(2)
# Negative reviews
on functionality	−.815***	(.257)
on customer service	−.687***	(.213)
on matters of taste	−.078	(.331)
on other topics	−.957*	(.532)
# Negative reviews > 0
on functionality			−.819**	(.337)
on customer service			−1.039***	(.324)
on matters of taste			.029	(.429)
on other topics			−1.073*	(.589)
Product page controls	Yes	Yes
Consumer controls	Yes	Yes
Product fixed effects	Yes	Yes
Day fixed effects	Yes	Yes
R²	.160	.160
# Observations	55,860	55,860

Variable: No. of Negative Reviews	High No. of Reviews and High Average Rating	Low No. of Reviews and High Average Rating	High No. of Reviews and Low Average Rating	Low No. of Reviews and Low Average Rating
Hedonic products	1.46	(1.00)	1.63	(2.04)	−.51	(.56)	−.17	(.46)
Utilitarian products	−1.40***	(.48)	−2.91**	(1.29)	.45*	(.28)	.20	(.50)

Footnotes

Acknowledgments

The authors express their gratitude to the participants of the 8th Workshop on Consumer Search and Switching Cost and the 39th Annual ISMS Marketing Science Conference for their valuable feedback. Additionally, they would like to extend a special thanks to the Wharton Customer Analytics Initiative and its partners for providing the data.

Coeditor

Sachin Gupta

Associate Editor

Randolph Bucklin

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Paulo Albuquerque

Notes

References

Allenby

Greg M.

Ginter

James L.

(1995), “The Effects of In-Store Displays and Feature Advertising on Consideration Sets,” International Journal of Research in Marketing, 12 (1), 67–80.

Anderson

Michael

Magruder

Jeremy

(2012), “Learning from the Crowd: Regression Discontinuity Estimates of the Effects of an Online Review Database,” Economic Journal, 122 (563), 957–89.

Angie

Amanda D.

Connelly

Shane

Waples

Ethan P.

Kligyte

Vykinta

(2011), “The Influence of Discrete Emotions on Judgement and Decision-Making: A Meta-Analytic Review,” Cognition & Emotion, 25 (8), 1393–422.

Babić Rosario

Ana

Sotgiu

Francesca

De Valck

Kristine

Bijmolt

Tammo H.A.

(2016), “The Effect of Electronic Word of Mouth on Sales: A Meta-Analytic Review of Platform, Product, and Metric Factors,” Journal of Marketing Research, 53 (3), 297–318.

Babin

Barry J.

Darden

William E.

Griffin

Mitch

(1994), “Work and/or Fun: Measuring Hedonic and Utilitarian Shopping Value,” Journal of Consumer Research, 20 (4), 644–56.

Baumeister

Roy F.

Bratslavsky

Ellen

Finkenauer

Catrin

Vohs

Kathleen D.

(2001), “Bad Is Stronger Than Good,” Review of General Psychology, 5 (4), 323–70.

Bigne

Enrique

Chatzipanagiotou

Kalliopi

Ruiz

Carla

(2020), “Pictorial Content, Sequence of Conflicting Online Reviews and Consumer Decision-Making: The Stimulus-Organism-Response Model Revisited,” Journal of Business Research, 115, 403–16.

Blei

David M.

Andrew Y.

Jordan

Michael I.

(2003), “Latent Dirichlet Allocation,” Journal of Machine Learning Research, 3, 993–1022.

Brown

Tom J.

Barry

Thomas E.

Dacin

Peter A.

Gunst

Richard F.

(2005), “Spreading the Word: Investigating Antecedents of Consumers’ Positive Word-of-Mouth Intentions and Behaviors in a Retailing Context,” Journal of the Academy of Marketing Science, 33 (2), 123–38.

10.

Cacioppo

John T.

Gardner

Wendi L.

Berntson

Gary G.

(1997), “Beyond Bipolar Conceptualizations and Measures: The Case of Attitude and Evaluative Space,” Personality and Social Psychology Review, 1 (1), 3–25.

11.

Chen

Yubo

Xie

Jinhng

(2008), “Online Consumer Review: Word-of-Mouth as a New Element of Marketing Communication Mix,” Management Science, 54 (3), 477–91.

12.

Cheong

Hvuk Jun

Morrison

Margaret A.

(2008), “Consumers' Reliance on Product Information and Recommendations Found in UGC,” Journal of Interactive Advertising, 8 (2), 38–49.

13.

Chevalier

Judith A.

Mayzlin

Dina

(2006), “The Effects of Word of Mouth on Sales: Online Book Reviews,” Journal of Marketing Research, 43 (3), 345–54.

14.

Chintagunta

Pradeep Κ.

Gopinath

Shyam

Venkataraman

Sriram

(2010), “The Effects of Online User Reviews on Movie Box Office Performance: Accounting for Sequential Rollout and Aggregation Across Local Markets,” Marketing Science, 29 (5), 944–57.

15.

De los Santos

Babur

Hortacsu

Ali

Wildenbeest

Matthijs E.

(2012), “Search Model of Consumer Search Using Data on Web Browsing and Purchasing Behavior,” The American Economic Review, 102 (6), 2955–80.

16.

Demuynck

Thomas

Seel

Christian

(2018), “Revealed Preference with Limited Consideration,” American Economic Journal: Microeconomics, 10 (1), 102–31.

17.

Eliaz

Kfir

Spiegler

Ran

(2011), “Consideration Sets and Competitive Marketing,” Review of Economic Studies, 78 (1), 235–62.

18.

Feick

Lawrence

Higie

Robin A.

(1992), “The Effects of Preference Heterogeneity and Source Characteristics on Ad Processing and Judgements About Endorsers,” Journal of Advertising, 21 (2), 9–24.

19.

Fiske

Susan T.

(1980), “Attention and Weight in Person Perception: The Impact of Negative and Extreme Behavior,” Journal of Personality and Social Psychology, 38 (6), 889–906.

20.

Floyd

Kristopher

Freling

Ryan

Alhoqail

Saad

Cho

Hyun Young

Freling

Traci

(2014), “How Online Product Reviews Affect Retail Sales: A Meta-Analysis,” Journal of Retailing, 90 (2), 217–32.

21.

Fung

Raymond

Lee

Matthew

(1999), “EC-Trust (Trust in Electronic Commerce): Exploring the Antecedent Factors,” AMCIS Proceedings 1999, 179, 517–19.

22.

Sherry

Hollenbeck

Brett

Proserpio

Davide

(2022), “The Market for Fake Reviews,” Marketing Science, 41 (5), 896–921.

23.

Hirschman

Elizabeth C.

Holbrook

Morris B.

(1982), “Hedonic Consumption: Emerging Concepts, Methods and Propositions,” Journal of Marketing, 46 (3), 92–101.

24.

Honka

Elisabeth

(2014), “Quantifying Search and Switching Costs in the US Auto Insurance Industry,” RAND Journal of Economics, 45 (4), 847–84.

25.

Honka

Elisabeth

Chintagunta

Pradeep

(2016), “Simultaneous or Sequential? Search Strategies in the U.S. Auto Insurance Industry,” Marketing Science, 36 (1), 21–42.

26.

Hung

Kineta H.

Stella Yivan

(2007), “The Influence of eWOM on Virtual Consumer Communities: Social Capital, Consumer Learning, and Behavioral Outcomes,” Journal of Advertising Research, 47 (4), 485–95.

27.

Ito

Tiffany A.

Larsen

Jeff T.

Smith

N. Kyle

Cacioppo

John T.

(1998), “Negative Information Weighs More Heavily on the Brain: The Negativity Bias in Evaluative Categorizations,” Journal of Personality and Social Psychology, 75 (4), 887–900.

28.

Kim

Jun B.

Albuquerque

Paulo

Bronnenberg

Bart J.

(2010), “Online Demand Under Limited Consumer Search,” Marketing Science, 29 (6), 1001–23.

29.

Kim

Jun B

Albuquerque

Paulo

Bronnenberg

Bart J.

(2016), “The Probit Choice Model Under Sequential Search with an Application to Online Retailing,” Management Science, 63 (11), 3911–29.

30.

Klein

Jill Gabrielle

(1996), “Negativity in Impressions of Presidential Candidates Revisited: The 1992 Election,” Personality and Social Psychology Bulletin, 22 (3), 288–95.

31.

Kozinets

Robert V.

De Valck

Kristine

Wojnicki

Andrea C.

Wilner

Sarah J.S.

(2010), “Networked Narratives: Understanding Word-of-Mouth Marketing in Online Communities,” Journal of Marketing, 74 (2), 71–89.

32.

Liu

Xiao

Lee

Dokyun

Srinivasan

Kannan

(2019), “Large-Scale Cross-Category Analysis of Consumer Review Content on Sales Conversion Leveraging Deep Learning,” Journal of Marketing Research, 56 (6), 918–43.

33.

Liu

Yong

(2006), “Word of Mouth for Movies: Its Dynamics and Impact on Box Office Revenue,” Journal of Marketing, 70 (3), 74–89.

34.

López

Inés López

De Maya

Salvador Ruiz

(2012), “When Hedonic Products Help Regulate my Mood,” Marketing Letters, 23 (3), 701–17.

35.

Manzini

Paola

Mariotti

Marco

(2018), “Competing for Attention: Is the Showiest Also the Best?” Economic Journal, 128 (609), 827–44.

36.

Markopoulos

Panos M.

Clemons

Eric K.

(2013), “Reducing Buyers’ Uncertainty About Taste-Related Product Attributes,” Journal of Management Information Systems, 30 (2), 269–99.

37.

Mudambi

Susan M.

Schuff

David

(2010), “What Makes a Helpful Online Review? A Study of Customer Reviews on Amazon.com,” MIS Quarterly, 34 (1), 185–200.

38.

Rao

Potluri

(1974), “Specification Bias in Seemingly Unrelated Regressions,” in Econometrics and Economic Theory, Sellekaerts

Willy

, ed. Springer, 101–13.

39.

Reinstein

David A.

Snyder

Christopher M.

(2005), “The Influence of Expert Reviews on Consumer Demand for Experience Goods: A Case Study of Movie Critics,” Journal of Industrial Economics, 53 (1), 27–51.

40.

Rozin

Paul

Royzman

Edward B.

(2001), “Negativity Bias, Negativity Dominance, and Contagion,” Personality and Social Psychology Review, 5 (4), 296–320.

41.

Salminen

Joni

Kandpal

Chandrashekhar

Kamel

Ahmed Mohamed

Jung

Soon-gyo

Jansen

Bernard J.

(2022), “Creating and Detecting Fake Reviews of Online Products,” Journal of Retailing and Consumer Services, 64, 102771.

42.

Sen

Shahana

Lerman

Dawn

(2007), “Why Are You Telling Me This? An Examination into Negative Consumer Reviews on the Web,” Journal of Interactive Marketing, 21 (4), 76–94.

43.

Skowronski

John J.

Carlston

Donal E.

(1989), “Negativity and Extremity Biases in Impression Formation: A Review of Explanations,” Psychological Bulletin, 105 (1), 131.

44.

Sun

Monie

(2012), “How Does the Variance of Product Ratings Matter?” Management Science, 58 (4), 696–707.

45.

Tirunillai

Seshadri

Tellis

Gerard J.

(2014), “Mining Marketing Meaning from Online Chatter: Strategic Brand Analysis of Big Data Using Latent Dirichlet Allocation,” Journal of Marketing Research, 51 (4), 463–79.

46.

Ursu

Raluca M.

(2018), “The Power of Rankings: Quantifying the Effect of Rankings on Online Consumer Search and Purchase Decisions,” Marketing Science, 37 (4), 530–52.

47.

Ursu

Raluca M.

Wang

Qingliang

Chintagunta

Pradeep K.

(2020), “Search Duration,” Marketing Science, 39 (5), 849–71.

48.

Vana

Prasad

Lambrecht

Anja

(2021), “The Effect of Individual Online Reviews on Purchase Likelihood,” Marketing Science, 40 (4), 708–30.

49.

Van Nierop

Erjen

Bronnenberg

Bart

Paap

Richard

Wedel

Michel

Franses

Philip Hans

(2010), “Retrieving Unobserved Consideration Sets from Household Panel Data,” Journal of Marketing Research, 47 (1), 63–74.

50.

Voss

Kevin E.

Spangenberg

Erie R.

Grohmann

Bianca

(2003), “Measuring the Hedonic and Utilitarian Dimensions of Consumer Attitude,” Journal of Marketing Research, 40 (3), 310–20.

51.

Chunhua

Che

Hai

Chan

Tat Y.

Xianghua

(2015), “The Economic Value of Online Reviews,” Marketing Science, 34 (34), 739–54.

52.

Xie

Xiao-Fei

Wang

Mei

Zhang

Ruo-Gu

Jie

Qing-Yuan

(2011), “The Role of Emotions in Risk Communication,” Risk Analysis: An International Journal, 31 (3), 450–65.

53.

You

Vadakkepatt

Gautham G.

Joshi

Amit M.

(2015), “A Meta-Analysis of Electronic Word-of-Mouth Elasticity,” Journal of Marketing, 79 (2), 19–39.

54.

Zellner

Arnold

(1962), “An Efficient Method of Estimating Seemingly Unrelated Regressions and Tests for Aggregation Bias,” Journal of the American Statistical Association, 57 (298), 348–68.

55.

Zhang

Kem Z.K.

Zhao

Sesia J.

Cheung

Christy M.K.

Lee

Matthew K.O.

(2014), “Examining the Influence of Online Reviews on Consumers’ Decision-Making: A Heuristic-Systematic Model,” Decision Support Systems, 67, 78–89.

56.

Zhong

Ning

Schweidel

David A.

(2020), “Capturing Changes in Social Media Content: A Multiple Latent Changepoint Topic Model,” Marketing Science, 39 (4), 827–46.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.15 MB

No. of Negative Reviews	No. of Arriving Consumers	No. of Consumers Deciding to ...
No. of Negative Reviews	No. of Arriving Consumers	Purchase Product	Open Second Review Page	Sort the Reviews	Search for Substitute	Scroll to Reviews of Substitute
0	33,761	998 (3.0%)	5,959 (17.7%)	549 (1.6%)	13,777 (40.8%)	5,883 (42.7%)
1	20,472	518 (2.5%)	3,976 (19.4%)	288 (1.4%)	8,929 (43.6%)	4,062 (45.5%)
2	8,909	181 (2.0%)	1,683 (18.9%)	113 (1.3%)	4,018 (45.1%)	1,998 (49.6%)
3	4,190	63 (1.5%)	754 (18.0%)	48 (1.1%)	1,951 (46.6%)	982 (50.2%)
4	1,229	15 (1.2%)	161 (13.1%)	11 (.9%)	609 (49.6%)	309 (50.7%)
5	399	0 (.0%)	40 (10.0%)	2 (.5%)	189 (47.4%)	77 (40.7%)

	(1)	(2)
	Log Daily Visits	Log Daily Visits
Average rating	.013*** (.003)	.014*** (.004)
# Reviews (log)	.046*** (.002)	.042*** (.002)
Price (log £)	−.027*** (.005)	−.036*** (.005)
Brand fixed effects	No	Yes
Category fixed effects	Yes	Yes
Day fixed effects	Yes	Yes
R²	.142	.172
# Observations	198,537	198,537

	(1)	(2)	(3)	(4)
	Opened Second Review Page	Sorted the Reviews	Searched for Substitute	Scrolled to Reviews of Substitute
# Negative reviews = 1	−.652 (.883)	−1.070*** (.293)	3.873*** (1.090)	.012 (.019)
# Negative reviews = 2	−2.369** (1.022)	−.977*** (.349)	5.243*** (1.378)	.038* (.022)
# Negative reviews = 3	−4.155*** (1.474)	−1.415*** (.446)	7.128*** (1.789)	.059** (.027)
# Negative reviews = 4 or 5	−9.466*** (1.989)	−1.439** (.722)	11.101*** (2.897)	.172*** (.044)
Product page controls	Yes	Yes	Yes	Yes
Consumer controls	Yes	Yes	Yes	Yes
Product fixed effects	Yes	Yes	Yes	Yes
Day fixed effects	Yes	Yes	Yes	Yes
R²	.191	.102	.191	.292
# Observations	68,960	68,960	55,860	23,565

	(1)		(2)
	Product Purchase		Product Purchase
# Negative reviews
on functionality	−.815***	(.257)
on customer service	−.687***	(.213)
on matters of taste	−.078	(.331)
on other topics	−.957*	(.532)
# Negative reviews > 0
on functionality			−.819**	(.337)
on customer service			−1.039***	(.324)
on matters of taste			.029	(.429)
on other topics			−1.073*	(.589)
Product page controls	Yes		Yes
Consumer controls	Yes		Yes
Product fixed effects	Yes		Yes
Day fixed effects	Yes		Yes
R²	.160		.160
# Observations	55,860		55,860