University of Amsterdam
Faculty of Economics and Business
Masters Program of Business Economics
Organization Economics Track
To tr@st or not to tr@st:
The relevance of trust signals in online markets
across the world
Claudio Pérez Esté
Student ID: 10622977
Extract
This piece of work aims to address the following question: Do trust signals based on feedback mechanisms in electronic markets have greater relevance in countries with higher levels of interpersonal trust? I analyze the transactions of 3 homogeneous goods in eBay sites for 7 countries with different levels of interpersonal trust, and the impact that feedback from previous transactions has on seller’s subsequent sales. My estimates suggest that trust signals from sellers, particularly positive reviews from customers, have a relevant influence in sales, and that there are significant differences on sales determinants depending on product information.
Introduction
Trust, as defined by the Oxford dictionary (2008), is the “Firm belief in the reliability, truth, or ability of someone or something”. As a key element in human interactions, trust is a concept of upmost relevance in markets, transactions and business in general. Through repeated interaction, effective delivery on commitments and adherence to consented rules, trading parties can develop trust between them, and significant benefits may be obtained from this trusting environment: increased business, lower transaction costs, flexibility, etc. This developed trust may also serve an additional and important purpose: signaling legitimacy and quality to other participants in the market.
When evaluating the possibility of engaging in business with a new party, information about the history and past performance of this party usually carries significant value in the decision making process. This information might be provided directly by the new party, but this option will be subject to an expected bias towards highlighting positive elements of this information, and downplaying negative ones. An alternative source for this important information is to learn the experience of others who have already conducted business with the party in question. If someone known and trusted provided information about their experience with the evaluated party, this information will most probably influence the final decision. But, more often than not, this information from another trusted party is not available, particularly if we look at today’s highly interconnected and impersonal electronic markets. Focusing on the many (fundamentally unknown) potential sellers and buyers in this market, some natural questions emerge: how can information about the quality and legitimacy of a seller be made available to buyers? Will buyers use this information to make their decisions? Is the experience of other unknown buyers relevant? Does this information generate trust?
In an attempt to provide this valuable information, electronic market providers like eBay have included feedback mechanisms, where transacting parties can evaluate their experience with each other. The resulting feedback information is then made available to anyone interested in learning about the history of a potential trading party. At first glance, this information seems to be very valuable to anyone interested in engaging in a transaction with an otherwise unknown party. But in reality, many factors come into play when assessing the validity of this feedback information provided by others. One of the most salient factors is
the cultural background of person(s) that looks at this information: what if this person comes from a country were people don’t trust each other? Does the opinion of others factor in the decision to trust a trading party?
Aiming at addressing this interesting issue, I coin the following research question: Do
trust signals based on feedback mechanisms in electronic markets have greater relevance in countries with higher levels of interpersonal trust? To this end I analyze the
transactions of 3 homogeneous goods in eBay sites for 7 countries with different levels of interpersonal trust, and the impact that feedback from previous transactions has on sellers subsequent sales. Previous research in the subject of trust and feedback mechanisms in electronic markets was mainly focused on single or few country samples, so by extending the geographical scope I expect to contribute to the literature on the influence of cultural factors in transactions within these type of markets.
Literature Review
The impact of cultural elements on economic exchange, and particularly the relevance of trust, is analyzed by Guiso, Sapienza and Zingales on their paper “Cultural biases in economic exchange” (2004). They propose that a lack of trust, originating from objective information but more importantly, from cultural biases, is a relevant cause of economic losses. Authors use a sample of surveys from European citizens ranging from 1970 to 1995, which included questions about how much people trusted others from specific nationalities (including their fellow countrymen). Using this data panel and applying fixed-effect country characteristics, they isolate the relative trust between a pair of countries, and find evidence that only a portion of this (relative) trust is explained by objective information (geographical distance, common borders, language and legal system similarities) and that subjective stereotypes (measured using historical conflicts between nations, religion and genetic distance) play an important role in the determination of trust. Regarding the effect of these elements of trust on economic exchange, authors analyze 3 sets of data: trade, portfolio investments, and foreign investment. Their results suggest that trust plays a significant role in all 3 forms of economic exchange, even after controlling for the objective elements of trust and for differences in information, and that this effect is stronger for differentiated goods compared to commodities. Even though this study relies on survey data, which could be biased by the individual interpretation of survey questions, the authors conduct a parallel survey to test their selected measure of trust, and find it to be robust.
In the specific realm of electronic markets, and focusing on customer feedback as a trust signaling mechanism, Melnik & Alm (2002) directly address the question: Does a seller’s ecommerce reputation matter? The authors examine how seller’s reputation signals influence buyer’s decisions to bid on items sold through eBay auctions. By studying the sales of a homogeneous good (U.S. Gold coin) they aim to isolate the effect of reputation signals, as this homogeneity creates a “commodity” effect, meaning product differences cease to be a determinant in the willingness of a buyer to bid on a product from one seller or another. In turn, available reputation signals from sellers should become a key element in this selection process. Current electronic market platforms (eBay, Amazon, etc), usually provide similar feedback mechanisms that allow customers to rate their transactions with a particular seller,
and this information is then available to others who may be considering doing business with this seller. This rating may take three distinct values: positive, neutral or negative. The authors hypothesize that there is a positive and significant relationship between a seller’s overall rating (positive minus negative reviews) and the selling price. They also expect that an increase in negative reviews would have a negative impact on this final price.
Their empirical results show that the seller's overall rating has a positive and statistically significant impact on the price, but this impact is small. Negative reviews are also found to have a negative impact on price, but this effect is also small. Similar results using a larger sample, extending it to more sales periods (as they study only a sales period of three weeks) and adding additional products and markets to the study, would increase the external validity of these results. But, despite the limitations of the study and the small impact found, the statistical significance of their results provide an important hint: online reputation is (relatively) relevant to buyers in this e-commerce market.
So, if reputation seems to be relevant in online markets, are the current mechanisms that collect and distribute this reputation information effective? This question is addressed by an experimental investigation conducted by Bolton, Katok & Ockenfels (2004). Their experiment aims to compare three distinct markets:
1) A “strangers” market, were trading parties (seller and buyer) don’t know each other, and the buyer has no information on the seller’s transaction history.
2) A “feedback” market where the buyer does not regularly interact with the same sellers, but is aware of each seller’s transaction history and delivery on commitments to other customers. This experimental market would mimic the common electronic markets available nowadays (eBay, Amazon, etc) where a feedback mechanism is the method of choice to signal legitimacy to other players in the market
3) A “partners” market, where the same buyer and seller interact repeatedly, thus developing a direct relationship between them. This setup is aimed to serve as a representation of the more traditional markets
The authors hypothesize that buyers will be equally influenced by information about the reputation of sellers, regardless of its source. In that sense, the feedback and partners markets would have very similar amounts of effective trade, as feedback from other buyers would be as valuable as the buyer’s own. By using an experimental design where volunteers acted as buyers and sellers under the three mentioned market structures, and were rewarded with
monetary benefits based on their performance, these authors measured trade, trust and trustworthiness in each of the markets to test their hypothesis.
Their findings actually reject this hypothesis. There is much more effective trade in the traditional partners market than in the feedback market. But both of these markets evidence greater levels of effective trade than the strangers market. So, in this sense, reputation information appears to be valuable towards trade, but there are differences depending on its source. Authors argue that these differences are explained by the finding that a buyer’s own past experience with sellers is what drives trust, so in a feedback market a trustworthy seller with good reviews would find that his reputation signals are not effective to induce trade with a new buyer, if this buyer has previous bad experiences with other sellers. Nevertheless, seller reputation information provided by other buyers does have value, particularly if buyers are made aware of the average behavior of the entire market. These findings are based on a sample of 140+ volunteers, mainly students acting as buyers and sellers, so external validity is limited. Nevertheless the experimental protocol is properly designed, so the main findings provide valid insights towards the effectiveness of online reputation mechanisms. A particular result from this experiment is that in a feedback market, negative reviews seem to have greater impact on decreasing sales, relative to the effect positive reviews have on increasing them. Also, the experiment shows that recent feedback has greater relevance to new buyers in their decisions, when compared to older reviews of the same seller.
The relevance of recent and negative feedback has also been reported in field studies, of which I would highlight the research done by Cabral & Hortacsu (2010). In their study “The dynamics of seller reputation: evidence from EBay”, Cabral and Hortacsu also study the relevance of eBay reputation mechanism. Using a similar homogeneous goods argument to Melnik & Alm (2002), that aims to exclude product differences as an explanatory variable for the choice of seller, these authors analyze eBay auctions of three products over a six-month period, using both panel data and cross section analysis.
While their cross section regressions fail to produce any results of significance, the use of a data panel does provide some very interesting insights: when a seller receives a negative review, subsequent sales are consistently reduced across the sample, and this impact is found to be statistically significant. These results show that buyers in this market do account for the reviews of others when evaluating a potential transaction with a seller, particularly if this review is negative in nature. They also find that after receiving the first
negative review, a second review of this same nature usually comes faster, and this second negative review also lowers subsequent sales, but not as significantly as the first one received.
These panel data results of Cabral & Hortacsu rely on some strong assumptions: (a) the frequency of buyer feedback is a good estimation for the total amount of sales a particular seller completes (b) the nature of the feedback (positive, neutral, negative) is an effective reflection of a buyer’s true level of satisfaction with a seller; (c) the probability of a buyer providing feedback is the same before and after a seller receives negative feedback (d) “neutral” feedback is considered “negative” feedback. Authors provide statistical tests that suggest the likelihood of feedback is not correlated with several characteristics of the seller, thus increasing the validity of their strategy of using feedback histories as proxies for transactions histories. Regarding neutral feedback, authors base their assumption on a subjective investigation of online chat rooms of eBay users, discussing their perceptions over the 3 possible feedback options: positive, neutral, negative. They claim most people associate “neutral” with a “negative” perception. Authors also found similarities of the distribution on negative and neutral feedback on their data. These convenient assumptions serve their results, but despite these arguable views on the data, quite common in research using this type of information, their results suggest there is and impact in sales from customer feedback, and that there is a difference in relevance between negative and positive reviews when buyers evaluate sellers.
A common observation across the reviewed research is that existing online feedback mechanisms are manipulable. Sellers with bad reputations might exit the market and come back under a new name, thus “erasing” their bad history, without buyers being able to notice. If, as research suggests, reputation in these markets is relevant, and negative feedback has a significant impact on future sales, seller will have the incentive to perform this kind of action in the event of receiving bad reviews. They might also be inclined to create false positive reviews to mislead customers. These considerations have been addressed by Brown & Morgan (2006) on their paper “Reputation in online markets: some negative feedback”. Authors found that an actual market to buy and sell feedback exists on sites like eBay, where users disguise an offer of positive feedback under a sale of a very low value “item”. To study this phenomenon, authors collected a sample of listings in the market for feedback over a six-month period, and found a significant amount of listings (6000+) offering this fabricated reputation reviews in return for a small amount of money. More than 75% of these
listings resulted into completed sales. This could mean that a portion of the feedback in these markets is actually false. Nevertheless electronic market providers have taken steps to reduce these forged reputation elements, for example allowing detailed reviews and information about the actual product that was transacted to be visible to interested parties. This is relevant in the sense that potential buyers could discriminate between reviews, and fine tune their perception regarding the reputation of a seller based on a detailed review of the available feedback, instead of an overall view of the total reviews of different natures. In theory this should discourage sellers to “buy” their feedback using these false items, and going forward in this investigation I will use the assumption that the reputation of the sellers in my sample is not fabricated.
Regarding cross-country studies about trust in online markets, an often cited paper is “Consumer Trust in an Internet Store: A Cross-Cultural Validation” by Jarvenpaa et. al (1999). The authors aim to identify the determinants of consumer trust in online stores, studying elements like store size and perceived reputation and their relationship to a customer’s willingness to buy. The authors hypothesize that there is a significant difference in the effect of these determinants that is associated to cultural factors, specifically the level of individualism in the customer’s country. To study this hypothesis authors perform online surveys to college students in Australia and Israel (and also report partial results from surveys in Finland). They found that perceived reputation has greater relevance than store size as determinants for sales, and that contrary to their hypothesis, there are no significant differences related to the cultural background of customers. These results might be influenced by their sample selection, mainly college students with similar degrees of internet experience, who might have rather homogeneous perceptions regarding online markets, and their measure of cultural differences associated only to individualism. The authors suggest further studies should be performed including additional cultural dimensions, of which they specifically highlight interpersonal trust.
Given the mentioned literature, I expect to contribute with this study by analyzing a specific measure on interpersonal trust and its relationship to reputational signals, using a sample of 7 countries. To my knowledge no similar studies have been performed, and the extent of the country sample should provide valuable insights on cross-cultural determinants of online sales.
Methodology
As the aim of this research is to analyze the selection criteria of buyers across different countries, focusing on the cultural-specific relevance of seller’s trust signals, my approach to isolate these effects is to study sales of homogeneous products transacted under equal market rules in each country. By employing this approach, (used similarly by Cabral & Hortacsu (2010) and Melnik & Alm (2002)) the product-specific characteristics, as well as the conditions under which these products are transacted, become irrelevant as a cause for the buying criteria differences (as they are all the same). This way the cultural specific aspects involved in the buying choices should be more salient as an explanatory variable.
Market Platform
To achieve the market rules homogeneity, I collect data from the renowned electronic market platform eBay. This company provides localized e-commerce sites in 22 countries, where sellers and buyers interact using the same trading rules. Given data availability, my sample is obtained from the following 7 countries: Canada, Australia, United States, United Kingdom, Germany, Italy and France.
To extract sales data from eBay I employed e-commerce market analytics tool terapeak (www.terapeak.com). This data analysis tool is the first authorized analytics provider of eBay market data, and the leading aggregator of e-commerce data for eBay, and competitor sites Amazon, Yahoo! Japan and Magento.
Product Selection
To secure product homogeneity I focus on the highly standardized consumer electronics industry, selecting products from worldwide-recognized brands, that where highly demanded in each of the sample countries. This criterion is applied to allow for the assumption that these products should be the same regardless on where they are sold. Additionally, the selected products should have relatively high levels of functional complexity and monetary value, with the intent that these characteristics made buyers more mindful about their seller selection criteria, thus reflecting on greater scale their selection priorities and avoiding randomization or simple convenience to play a role. From a sample size standpoint, these products also had to be in demand in each of the sample countries, so I could gather sufficient data on related transactions.
Product #1 - Apple Iphone 5: the latest version of Apple’s best selling Smartphone, this
product is recognized and traded worldwide. Released on September 20, 2012 and with over 9 million units sold in the first three days after release (1) this product is also traded regularly on the eBay sites included in the sample.
Product #2 – Samsung Galaxy Tab 3: A tablet computer manufactured by Samsung
Electronics, belonging to the third generation of the Samsung Galaxy Tab series. It was announced on 24 June 2013 and launched in the US on 7 July 2013. Despite mixed success after its release, the tablet has relevant trading volumes in the eBay sites included in the sample.
An additional element that played a role in the selection of this product, was that the leading product of this segment (tablet computers) in terms of sales volumes through eBay is also manufactured by apple (the Ipad) so, in the intent to add additional variety I included the Samsung galaxy tab instead.
Product #3 GoPro Hero 3+ digital camera: A high-definition digital video camera
designed for extreme conditions. Since its release, GoPro product sales have experienced viral growth, thanks to mass exposure in social media around the world. Currently this company is on the planning stages of an IPO in the United States, expected on June 2014. Despite of being the less known product in the sample, trading volumes are significant on eBay.
Sample statistics –Transactions by product type
Sum of Items Sold
Items sold by product type / country 01/01/2014 – 30/03/2014
Country Apple Iphone 5 Samsung galaxy tab GoPro Hero 3+ Total
AUS 36 307 851 1194 CAN 57 19 68 144 FRA 15 49 134 198 GER 814 958 628 2400 ITA 89 33 125 247 UK 122 681 2866 3669 USA 691 375 7845 8911 Grand Total 1824 2422 12517 16763 1 https://www.apple.com/pr/library/2013/09/23First-‐Weekend-‐iPhone-‐Sales-‐Top-‐Nine-‐Million-‐Sets-‐New-‐Record.html
Data collection procedure
The data collection for this research can be divided into three mayor dimensions: 1) Sellers sales results from the period 01 January 2014 to 31 March 2014
2) Seller’s reputation signals history from the period 01 January 2013 to 31 December 2013
3) Country trust Index for the sample countries
1) Sellers sales results from the period 01 January 2014 to 31 March 2014
As described in the specification for this research, my aim is to determine how do reputation signals from sellers, collected for the period 01 January 2013 to 31 December 2013, influence the subsequent sales performance of these sellers on the first quarter of 2014. To achieve this I first identify the top sellers in each of the sample countries for the mentioned period by performing a query on data analysis tool terapeak. To ensure product homogeneity, I limit the search to only new products. eBay sites are used significantly for the sale of products in highly variable conditions (used and even broken products) so including them in the sample would imply losing the desired homogeneity. Additionally, given the local language differences in some of the sample countries, I use a search operator to include the word “new” in each local language. I.e. for France the search criteria for the Iphone is: “Iphone 5 (new,neuf)”. By including in the search criteria both the words “new” and “neuf” in a parenthesis, the search tool results will include items with either the word “new” or “neuf” in their description. Following this methodology, the search criteria for each of the countries/products/sites (restricted to the period 01/01/2014 to 30/03/2014) is as follows:
Site (Country) Product 1 Product 2 Product 3
eBay.it (Italy) Apple Iphone 5 (new,nuovo)
Samsung Galaxy tab 3 (new, nuovo)
GoPro Hero 3+ (new, nuovo)
eBay.de (Germany) Apple Iphone 5 (new,neu)
Samsung Galaxy tab 3 (new,neu)
GoPro Hero 3+
(new,neu) eBay.fr (France) Apple Iphone 5
(new,neuf)
Samsung Galaxy tab 3 (new,neuf) GoPro Hero 3+ (new,neuf) eBay.com.au (Australia) eBay.co.uk (United Kingdom) eBay.ca (Canada) eBay.com (United States)
Apple Iphone 5 new Samsung Galaxy tab 3 new
GoPro Hero 3+ new
To further secure that the results from the mentioned queries were correct, and in line with the requirements of this research, random individual reviews were conducted for resulting listings for every product in each of the sample countries.
Sample Statistics – Sellers by country
Count of Seller ID
(Top) sellers by product type / country 01/01/2014 – 30/03/2014 Country Apple Iphone
5 Samsung galaxy tab GoPro Hero 3+ Grand Total – top sellers AUS 15 10 10 35 CAN 10 10 10 30 FRA 5 10 10 25 GER 20 20 15 55 ITA 19 13 11 43 UK 20 20 15 55 USA 20 20 15 55 Grand Total 109 103 86 298
2) Seller’s reputation signals history from the period 01 January 2013 to 31 December 2013
After the top sellers for each product were identified in the mentioned query, the next step was to collect their available reputation signals from the year prior to the studied selling period. eBay utilizes a historical seller reputation mechanism where buyers leave an overall feedback rating for a seller (positive, neutral, or negative), based on their buying experience, and in many cases they can also leave detailed seller ratings in 4 areas: item description, communication, shipping time and shipping and handling charges. The detailed seller rating system is based on a one- to five-star scale, five stars being the highest rating, and one star being the lowest rating. Nevertheless, the focus of this data collection would be directed to overall ratings (positive, neutral, or negative), as the specific differences arising from the detailed ratings categories are not relevant to answer the research question (but serve as assurance that reputations are not fabricated).
The feedback ratings would be the information available to buyers with respect to past experiences of other customers with each seller, in each of the eBay platforms, at the time they made their buying decision. These feedback ratings are available up to 12 moths prior to the date on which you look at a seller’s profile. Given that the data collection for this information was performed from April 15, 2014 to April 30, 2014, some of the observed ratings would overlap (or go beyond) the studied sales period, and in that sense, it would be information that was not available for buyers at the time they made their buying decision. In order to isolate the reviews that were effectively available to buyers for the studied period, I collected the feedback from the past 12 months for each seller in the sample, and then individually removed the portion of this feedback that was generated after January 01, 2014. The resulting information obtained from this process for each seller is as follows:
- Positive, neutral and negative reviews received by each seller prior to Dec 31, 2013 - Year of registration in eBay
This second item, related to the experience of the seller, is also considered as a reputation signal. The studied literature commonly mentions that reputations mechanisms in electronic markets are manipulable, particularly because of the fact that a seller that has received negative reviews can exit the market and re-enter with a new identity, without
customers being able to notice. In this sense a seller whose profile was created a long time ago should be perceived as more reputable, when compared to a more recent competitor.
Sample Statistics – Reviews by type
Country Count of Seller ID Sum of total reviews Sum of Positive Sum of Neutral Sum of Negative AUS 35 129,118 128,071 607 440 CAN 30 7,941 7,862 36 43 FRA 25 20,007 19,541 240 226 GER 55 450,127 445,809 2,571 1,747 ITA 43 79,375 78,620 445 310 UK 55 137,681 135,887 973 821 USA 55 406,495 401,234 2,793 2,468 Grand Total 298 1,230,744 1,217,024 7,665 6,055
Sample Statistics – Seller Experience
Average seller experience by product type / country (Years) Country
Apple Iphone
5 Samsung galaxy tab
GoPro Hero 3+ Grand Total – average exp AUS 7.3 2.6 4.9 5.3 CAN 4.7 5.2 6.4 5.4 FRA 4.4 6.1 3.3 4.6 GER 6.6 6.6 7.0 6.7 ITA 7.2 6.0 5.8 6.5 UK 5.3 5.9 4.9 5.4 USA 7.3 8.3 5.0 7.0 Total 6.4 6.1 5.4 6.0
3) Country trust index for the sample countries
In order to identify the interpersonal trust index in each of the sample countries, I turn to the world values survey (http://www.worldvaluessurvey.org) a globally recognized network of social scientists studying changing values and their impact on social and political life. This survey, and the various cultural aspects that are captured on it, have been used in previous economic studies focusing on the relationship between culture, institutions and economic
development, i.e Tabellini (2010), Sapienza, Zingales, & Guiso (2006), Knack & Keefer (1997).
The available values for the sample countries are as follows:
Trust index (% of “yes” answers) World Values Survey Wave 5: 2005-2009
Country/Region
Total Australia Canada France Germany Italy Great Britain United States Most people can be trusted 35.3 45.6 41.8 18.7 33.8 27.5 30 39.1 Need to be very careful 61.4 53.2 55.9 80.8 57.9 66.7 68.3 60.3 No answer 0.9 1.2 0.1 0.4 1.7 2.8 0.2 0.2 Don’t know 2.3 0 2.2 0.1 6.6 3.1 1.5 0 (N) 9.952 1.421 2.164 1.001 2.064 1.012 1.041 1.249
Hypothesis and results
Provided the available market data and the measure of trust from the world values survey, my initial hypothesis is:
“Positive reviews received by a seller will have greater (positive) impact on sales in countries with a higher index of interpersonal trust”
The rationale behind this hypothesis is that the reviews available for each seller come from other customers, so if a buyer’s cultural background is associated to high levels of trust in others, It should mean that the more others demonstrate trust in a seller, through positive reviews, the more it should influence a new buyer granting his trust. On the other hand, for buyers with less trusting cultural backgrounds, other factors will come into play when deciding to choose a seller, and reviews from others should be less relevant.
It is important to highlight the assumption that the cultural background of the eBay customers in the sample is defined by the country in which they perform the transaction. It is very plausible that a portion of these buyers might be foreigners using eBay in a country different than their own, and that their views regarding trust are different. This could be particularly relevant in countries with a high level of international migrant stock as a percentage of their total population, which, in the case of this sample, the highest ranked is Australia, with 27,7% as of 2013 (2). As no specific data is available regarding the cultural origin of buyers in eBay, I employ the assumption that the trust index obtained from the world values survey accurately reflects the perception of buyers, in each sample country, regarding trust.
My second hypothesis:
“The amount of items sold by a seller’s is positively and significantly correlated to the number of years this seller’s profile has been registered on eBay”
Given the fact that sellers are able to change identities, a long standing profile should signal trust to the buyer, in the sense that this seller has greater amounts of information available regarding his past performance, and has encountered no reason to change his identity. On the contrary, a newer seller who is still building up his reputation, might have
2 United Nations, Department of Economic and Social Affairs (2013). Trends in International Migrant Stock: The 2013 revision (United Nations database, POP/DB/MIG/Stock/Rev.2013)
been selling under a different identity in the past, and changed his identity to avoid the consequences of negative information regarding his performance. So, if there are no product differences, a seller with an older profile should sell more, holding the other sales determinants constant.
Based on the collected data I perform an OLS regression with econometrics software STATA, using the following variables:
Variable type Name Description
Dependent
Variable # Items Sold
Total number of items sold by the seller during the first quarter of 2014
Independent
Variable # Positive Reviews
Total amount of positive reviews received by seller from previous customers up to Dec 31, 2013 Independent
Variable Country trust index
Seller’s country index of interpersonal trust, from World Values Survey
Independent Variable
Years Selling on eBay
Total number of years since seller’s profile creation in eBay
Interaction variable
Positive reviews x country trust index
Interaction term between variables: # of positive reviews & Country trust index
Interaction variable
Years selling on eBay x country trust index
Interaction term between variables: Years selling on eBay & Country trust index
My initial results are summarized in table 1. Column (1) includes all mentioned independent variables. Columns (2) and (3) also include the interaction terms. Column (4) includes all variables and both interaction terms
Table 1
Observations (sellers)= 298
Dependent Variable = Number of items sold
Regressors (1) (2) (3) (4)
0.009 -0.026 0.009 -0.026 # Positive reviews
(0.002)*** (0.013)* (0.002)*** (0.014)* 1.219 -0.537 0.089 -0.933 Country trust Index
(0.663)* (0.700) (1.154) (1.204)
-3.142 -2.247 -10.474 -4.873
Years selling on eBay
(1.919) (1.972) (5.706)* (6.305)
0.095 0.095
Positive reviews x
country trust index (0.037)*** (0.037)**
21.125 7.555
Years selling on eBay x
country trust index (15.397) (16.728)
-6.705 56.456 32.301 70.133
Intercept
(29.821) (28.234)** (45.67) (46.321)
R2 0.5134 0.5586 0.5142 0.5587
Notes: In parentheses are reported the robust standard errors * Significant at the 10% confidence level
** Significant at the 5% confidence level ***Significant at the 1% confidence level
The estimates reported in column (1) show a positive and statistically significant coefficient for the number of positive reviews as a determinant for the number of items sold. A direct interpretation of this coefficient is that for every additional 100 positive reviews, a seller will sell (close to) one additional item (0.0098693 x 100 = 0.98693 items). The statistical significance of this coefficient seems to confirm the relevance of positive reviews, holding constant the trust index of the seller’s country and the experience of the seller in eBay. For the second variable (country trust index) the coefficient is positive as expected, and significant at the 10% confidence level. The interpretation of this coefficient is that a 1% increase in trust between buyers (as citizen of a particular country) would result in a estimated increase in sales of 1,219 items from a particular seller, holding constant his positive reviews and the years he has been selling on eBay.
Regarding the experience of the seller, and contrary to my hypothesis, the coefficient is negative, but without relevant statistical significance. This first regression supports the notion that buyers are influenced by the amount of positive reviews that a seller has received, and by their particular cultural backgrounds (more specifically the interpersonal trust levels in each of their countries). The years of experience a seller has on eBay does not seem to play a direct role as a determinant for the amount of items sold by a seller. Nevertheless, in order to further explore the relationship between positive reviews and trust, and to directly test my first hypothesis, I include the interaction between these 2 terms in regression #2.
The estimates reported in column (2) support the hypothesis of significance for positive reviews as an explanatory variable for the amount of items sold by a seller, and the inclusion of the interaction term has a positive effect on the R2. Both the coefficient of positive reviews and the interaction term with the country trust index are statistically significant, but in this case the individual coefficient of positive reviews is negative. However, as this regression includes an interaction term, it’s the combined effect of both coefficients on the dependent variable that has to be accounted for, to observe the overall effect of a change in the interacted independent variables. In the case of positive reviews, an additional review will have an overall effect on items sold of: -0,0265316 + 0.0958332 x (country trust index). This result means that with a 1% increase in the interpersonal trust index, a positive review received by a seller increases its effect on sales by an additional 0,0693 items (holding the years of experience of the seller constant). In simpler terms, the more buyers’ trust others, the more value a positive review has to a seller. To illustrate this effect, I use the values from the sample. When the country trust index is at the sample low (18.7 - Italy) the slope of this regression line #2 relating the number of items sold to the number of positive reviews is estimated to be 1.7655 (= 0,0265316 + 0.0958332 x 18.7 --holding the years of experience constant). At the sample median (33.8 – Germany) this slope increases to 3.2126. Using the top value of the sample (45.6 – Australia) the slope reaches 4.3434. This means that even though positive reviews from buyers are valuable to sellers in all countries (as they increase sales) they seem to be more effective in countries with higher levels of interpersonal trust. In order to calculate the appropriate standard errors for this combination of parameter estimates, and to account for a (possible) non-linear combination, I apply the Stata command nlcom to calculate these values using the delta method. Results as follows:
ItemsSold Coefficient Std. Err. t P>|t| [95% Conf. Interval] _nl_1 0.0693016 .0233558 2.97 0.003 0.0233352 0.115268
With the statistical significance on the coefficient for this parameter combination (at 1% level), these results support the hypothesis that positive reviews received by a seller will have greater positive impact on sales in countries with higher index of interpersonal trust.
Extending this analysis to the other interacted variable, the country level of trust, the first salient element is that the individual coefficient on this variable is not statistically significant. For this specification, the relevant role of trust is in its interaction with the positive reviews variable. The overall impact of a unit increase in trust on items sold will be: -0.5372241 + 0.0958332 x (# of positive reviews). In this case, if a seller has only 1 positive review he will actually sell less items (-0,4413) with a unit increase in trust. The slope of the line relating the number of items sold to the country level of trust becomes positive when a seller has received 6 or more positive reviews, which could mean there is little relevance in the first few positive reviews of a seller in the eyes of buyers. Nevertheless, as evidenced by the calculation of standard errors using the delta method, this effect lacks statistical significance:
ItemsSold Coefficient Std. Err. t P>|t| [95% Conf. Interval] _nl_1 -0.4413909 0.6863033 -0.64 0.521 -1.7921 0.9093181
The estimates reported in column (3) are included to analyze the specific effect of seller experience on sales, and its interaction with the country level of trust. This regression also points to the relevance of positive reviews as a determinant of items sold, as evidenced by the statistically significant coefficient. The specific coefficient on the years of experience of the seller is statistically significant at the 10% confidence level, and negative. This could mean that less experienced sellers actually sell more items, but in order to evaluate the overall impact on items sold the effect of the interaction term with the trust index must be included. Using the delta method to evaluate this combined effect on the number of items sold:
ItemsSold Coefficient Std. Err. t P>|t| [95% Conf. Interval] _nl_1 10.65065 10.20563 1.04 0.298 -9.434988 30.73629
Given that the overall effect of the combination of these coefficients is not statistically significant, there is no evidence that experience has a negative effect on sales. If this were the case, there would be an incentive for experienced sellers to continuously change their identities and reappear as “new” sellers in the market (particularly those who received many negative reviews, as the reviewed literature suggests). On the other hand, starting again with a new identity would mean losing any positive reviews received in the past, and the positive effect on sales that they seem to have, so the benefits of doing this would not be clear even in this hypothetical scenario. If the results showed that newer sellers did sell more, it would likely be a sign of omitted variables reflecting some attractive characteristics of newer sellers (better marketing, competitive prices, etc) but this cannot be concluded from these results. Without statistical significance on the coefficient of seller experience in columns 1&2, and neither from the combined effect of seller experience ant trust in column 3, the results only suggest that the amount of years a seller has been registered on eBay is not a relevant element for a buyer to select a seller.
The estimates reported in column (4) include both interaction terms: between positive reviews and trust, and between seller experience in trust. These estimates also point to the relevance of positive reviews, and their interaction with the trust index, as an explanatory variables for the number of items sold by a seller.
Initial results, summarized on table 1, suggest that my first hypothesis is true (Positive
reviews received by a seller will have greater positive impact on sales in countries with a higher index of interpersonal trust), and that my second hypothesis cannot be confirmed (the amount of items sold by a seller’s is positively and significantly correlated to the number of years this seller’s profile has been registered on eBay).
In order to further attest whether the effects found are evidenced in each of the 3 products in the sample, I will now review them separately. As explained in the methodology section, I selected 3 products that are considered to behave as commodities across the sample countries – Apple Iphone 5, Samsung Galaxy tab 3 and the GoPro digital Camera. Table 1 included consolidated results for all 3 products. By separating them, I expect to observe differences in the sales determinants, and identify if either of this products is particularly influencing the consolidated results.
Product #1 – Apple Iphone 5
Table 2 – Apple Iphone 5Observations (sellers)= 109
Dependent Variable = Number of items sold
Regressors (1) (2) (3) (4) 0.0002 -0.006 0.0002 -0.006 # Positive reviews (0.0001)** (0.003)** (0.0001)** (0.003)** 0.261 -0.098 0.263 0.105 Country trust Index (0.308) (0.300) (0.274) (0.289) 1.737 1.727 1.750 1.766 Years selling on eBay (1.149) (1.154) (2.954) (2.976) 0.020 0.020 Positive reviews x
country trust index (0.0096)** (0.009)**
-0.036 -0.110 Years selling on eBay x country trust index (7.659) (7.701) -4.085 1.736 -4.163 1.499 Intercept (11.076) (10.469) (10.521) (10.65) R2 (Adj) 0.002 0.033 0.027 0.033
Notes: In parentheses are reported the robust standard errors * Significant at the 10% confidence level
** Significant at the 5% confidence level ***Significant at the 1% confidence level
The results for the Apple Iphone 5 show a significant coefficient for the positive reviews received by a seller as a determinant of items sold, and this is consistent with the consolidated results. Nevertheless the value of this coefficient for the selected specifications is very low (i.e in the estimates reported in column 1, a seller will need 5,000 additional positive reviews to increase sales by one item, all else constant). The estimates reported in column (2) are also consistent with the consolidated results, in showing statistical significance for both the coefficient in positive reviews and its interaction with the trust index. Nevertheless the explanatory power of the selected specifications for the variations in sales for the Iphone 5 is virtually non existing, as evidenced by the low level of R2. This
would mean that variations in sales are explained by omitted variables, which will be further discussed in the next section.
Despite being the most widely recognized product in the sample, it is also the one with lower sales volumes for the studied period. Given the effective marketing and distribution model used by Apple, most customers will probably buy directly from the store, and the products transacted through eBay (although new and the same as in the store for this sample), might appeal to particular customers by other factors like delivery convenience, savings in shipping costs, price differences due to exchange rate, etc. Nevertheless I lack the data to further analyze this or other determinants of Iphone 5 sales through eBay.
Product #2 – Samsung Galaxy Tab 3
Table 3 – Samsung Galaxy Tab 3Observations (sellers)= 103
Dependent Variable = Number of items sold
Regressors (1) (2) (3) (4) 0.003 0.001 0.003 0.0007 # Positive reviews (0.003) (0.007) (0.003) (0.007) 0.371 0.245 0.713 0.645 Country trust Index (0.342) (0.229) (0.653) (0.596) -0.793 -0.731 1.421 1.970 Years selling on eBay (1.425) (1.416) (4.944) (4.775) 0.008 0.009 Positive reviews x
country trust index (0.019) (0.019)
-6.372 -7.747 Years selling on eBay x country trust index (11.519) (10.929) 4.257 8.140 -7.800 -6.013 Intercept (13.25) (11.396) (26.240) (25.046) R2 0.146 0.147 0.146 0.148
Notes: In parentheses are reported the robust standard errors * Significant at the 10% confidence level
** Significant at the 5% confidence level ***Significant at the 1% confidence level
Despite some increase in R2 when compared to the Iphone 5, the results for the Samsung Galaxy tab still show little evidence of explanatory power for the independent variables, and in this case none of the coefficients are statistically significant. Being also a recognized product, readily available in stores across many countries, explanation of sales
through eBay may point to similar variables as the Iphone 5. These individual results for products 1 & 2 point directly to product 3 as the main driver behind the consolidated results.
Product #3– GoPro Digital Camera
Table 4 – GoPro Digital CameraObservations (sellers)= 86
Dependent Variable = Number of items sold
Regressors (1) (2) (3) (4) 0.011 0.018 0.011 0.019 # Positive reviews (0.000)*** (0.034) (0.000)*** (0.035) 2.182 2.607 0.406 0.681 Country trust Index (2.062) (2.192) (3.320) (3.277) -10.183 -10.587 -23.836 -26.222 Years selling on eBay (6.041)* (6.502) (17.441) (19.109) -0.017 -0.021 Positive reviews x
country trust index (0.087) (0.090)
39.949 45.482 Years selling on eBay x country trust index (53.526) (55.896) 44.189 30.800 103.688 95.512 Intercept (83.307) (78.218) (120.739) (116.41) R2 0.632 0.633 0.633 0.634
Notes: In parentheses are reported the robust standard errors * Significant at the 10% confidence level
** Significant at the 5% confidence level ***Significant at the 1% confidence level
Results for the GoPro digital camera show a substantial increase in R2 for all specifications, and statistically significant coefficients for positive reviews in regressions 1 & 3. These results suggest the GoPro sales are better explained by the selected specifications than any of the other products in the sample, and are the main driver behind the significance in the coefficients of the consolidated results.
Being the least known product in the sample, The GoPro camera is more prone to the famous “market for lemons” failure proposed by Akerlof (1970). With less information available to buyers compared to the other products, a seller’s reputation might be more relevant under this uncertainty, as buyers are always looking to avoid buying a “lemon”. Shapiro (1983) discussed seller reputation as a valid but imperfect mechanism of assuring
product quality, and Rice (2011) proposed that under increased uncertainty, reputation of electronic market seller’s influences buying decisions. Applying these arguments to my results, then when a GoPro seller has many positive reviews, and a buyer doesn’t have much additional information to attest the quality of the product, then these positive reviews might drive the selection of a seller much more than with other products for which more information is available, and for which other factors might come into play (like the Iphone and the Samsung Galaxy). This idea seems to be supported by the statistical significance (at 1% confidence level) of the coefficient of positive reviews columns 1 & 3, and the increased R2 in the specifications, compared to the other products. Nevertheless its important to note that for this product, the trust index does not seem to play a role in defining the number of items sold, either by itself or by its interaction with the number of positive reviews. This could mean that in the absence of product information, buyers rely more on positive reviews regardless of differences in their trust levels.
Discussion
The results observed in this study point to significant differences in the determinants of eBay sales between different products across countries. The consolidated results suggest that trust signals from sellers, particularly positive reviews from other customers, might indeed have influence in sales, and this influence seems to be greater in countries with higher levels of interpersonal trust. On the other hand, a seller’s level of experience shows no evidence as a determinant of sales in any of the defined specifications. When the analysis is performed on a product specific basis, evidence points to the lesser-known product (the GoPro camera) as the driver behind the significance of positive reviews in the consolidated results. These empirical results suggest that the relevance of trust signals from sellers in online markets (particularly positive reviews) might be in fact related to the amount of information available on the product that is transacted.
Another interesting fact evidenced in the sample is that GoPro Camera, is sold much more through eBay than the other two products, which are far more exposed to the general public due to the global nature of their manufacturing companies.
Items sold by product type 01/01/2014 – 30/03/2014 Apple Iphone 5 Samsung galaxy tab GoPro Hero 3+ Grand Total – items sold Grand Total 1824 2422 12517 16763 % of sample 10.88% 14.45% 74.67% 100%
This element needs to be accounted for as it causes a bias in the consolidated results. It might also be evidence that lesser known products, from companies without the global scope of giants like Apple or Samsung, might have higher transaction volumes through eBay given the existence of markets that are not properly covered by their distribution models, and present an opportunity to individual sellers to cover this distribution shortcomings and make a profit. The relationship between sales volumes in electronic market platforms and product information might be another interesting area for future research.
Compared to other cross-country studies, which use samples from 2 or 3 countries, this study uses an ambitious approach by including 7. This fact is expected to increase validity of the results, but also implies many underlying mechanisms particular to each of the countries
and markets are most probably being overlooked, and thus not controlled for. Market size, local product availability, general perception of eBay, customer demographics, etc might be differential factors that are being overlooked, and could provide additional insights on the questions at hand. Also, as this study uses only one measure of interpersonal trust, a natural extension is to include additional measures to further test the proposed hypotheses. The World Values Survey is the most salient source for this information, particularly given the extent on this sample, but including additional measures would add robustness to any results, and would be particularly interesting if they differ between them.
An important limitation for these results is the existence of other omitted variables, of which the most salient one is the item’s price. This data was not included in the sample simply because it was not available in the data sources used. As evidenced on the reviewed literature, price is a key determinant for customers to select a seller in electronic markets, and good reputations sometimes allow sellers to sell at higher prices. Furthermore, even though the methodology used provides confidence that the items included in the sample are the same, and equal to new items found in stores, many elements might come into play that cause a differentiation in price: varying shipping costs, different return policies, guarantees, volume inventory purchases, taxes, exchange rate differentials, etc. Additional elements that can influence sales include product descriptions by the seller, quality pictures of the sold product, and direct availability of the product in the traditional market.
Another element that limits the external validity of these results is the sample size. The selected sample includes 3 products from the electronics industry, generating industry bias, and a sales sample of only 3 months from 7 different countries. Extending the sample to more products, sales periods and countries would increase the validity of the results found.
Regarding the selection of specifications, I assume a linear relationship between the dependent and independent variables.
Regarding the characteristics of the positive feedback used for this sample, the available information includes reviews for sales of any type of product, not only the specific products included in the sample. My assumption is that positive reviews increase the reputation of the seller, regardless of the nature of the product sold in the transaction that generated them. I find that this assumption is acceptable given the fact that customers on eBay receive information about the overall reviews from a seller, and rarely rely specifically on only the feedback from previous sales of the product they are looking to buy. Also, because the
specific information on the reviews is available to buyers if they chose to review it, sellers should be discouraged to manufacture these reviews.
Conclusion
The determinants of trust in electronic markets, the influence of cultural factors, and their impact on sales is a complex interaction of variables, many of which are yet to be studied. The aim of this investigation was directed to evaluate if trust signals in electronic markets, based on feedback mechanisms, have greater relevance in countries with higher levels of interpersonal trust. While the results suggest that trust signals from sellers, particularly positive reviews from other customers, might indeed have influence in sales, there are significant differences to account for depending on the product characteristics.
In the consolidated analysis, the results suggest the initial hypothesis that positive reviews received by a seller will have greater positive impact on sales in countries with a higher index of interpersonal trust is true. Nevertheless, when products are analyzed separately, this hypothesis cannot be confirmed. Positive reviews seem to explain sales particularly from the least-known product in the sample, and the relationship of positive reviews and trust is not significant in that case. In the case of the product of which there is more information available, the relationship between positive reviews and trust is statistically significant, but the selected specifications have little explanatory power on variations in sales. Regarding the hypothesis that the experience of a seller is a relevant determinant of sales, the results fail to provide evidence to confirm this hypothesis.
Even though the fundamental questions of this study remain unanswered, the results provide valid leads towards future research. A particular area of interest is the fact that product information availability seems to play a significant role in the relevance of trust signals in online markets, and also on trading volumes. Another natural extension of this research is to determine which type of products are influenced by buyer’s cultural background, and which are more dependant on other factors. Also, while there may be several omitted variables, including price data in the proposed specifications will probably lead towards much higher explanatory power, and to a better answer to the questions at hand. Extending this research will serve online sellers to decide how relevant is it to invest in creating and maintaining a good reputation, depending on the type of product they sell and on the cultural background of their customers.
References
Akerlof, G. A. (1970). The market for" lemons": Quality uncertainty and the market mechanism. The quarterly journal of economics, 488-500.
Avery, C., Resnick, P., & Zeckhauser, R. (1999). The market for evaluations. American
Economic Review, 564-584.
Bolton, G. E., Katok, E., & Ockenfels, A. (2004). How effective are electronic reputation mechanisms? An experimental investigation. Management science, 50(11), 1587-1602. Brown, J., & Morgan, J. (2006). Reputation in online markets: some negative feedback. University of California, Berkeley.
Dictionary, O. E. (2008). Oxford English dictionary online. Oxford University Press, Oxford, UK http://www. oed. com.
Edelman, B. (2012). Using internet data for economic research. The Journal of Economic
Perspectives, 189-206.
Guiso, L., Sapienza, P., & Zingales, L. (2004). Cultural biases in economic exchange (No. w11005). National Bureau of Economic Research.
Hortaçsu, Ali, F. Asís Martínez-Jerez, and Jason Douglas. 2009. "The Geography of Trade in Online Transactions: Evidence from eBay and MercadoLibre." American Economic Journal:
Microeconomics, 1(1): 53-74
Jarvenpaa, S. L., Tractinsky, N. and Saarinen, L. (1999), Consumer Trust in an Internet Store: A Cross-Cultural Validation. Journal of Computer-Mediated Communication.
Knack, S., & Keefer, P. (1997). Does social capital have an economic payoff? A cross-country investigation. The Quarterly journal of economics, 1251-1288.
L Cabral, A Hortacsu.(2010).“The dynamics of seller reputation: Evidence from ebay”
Journal of Industrial Economics 58, 54-78
Melnik, M. I., & Alm, J. (2002). Does a seller’s ecommerce reputation matter? Evidence from eBay auctions. The journal of industrial economics, 50(3), 337-349.
Rice, S. C. (2012). Reputation and uncertainty in online markets: an experimental study. Information Systems Research, 23(2), 436-452.
Rutherford, M. W., Buller, P. F., & Stebbins, J. M. (2009). Ethical considerations of the legitimacy lie. Entrepreneurship Theory and Practice, 33(4), 949-964.
Sapienza, P., Zingales, L., & Guiso, L. (2006). Does culture affect economic outcomes? (No. w11999). National Bureau of Economic Research.
Shapiro, C. (1983). Premiums for high quality products as returns to reputations. The
quarterly journal of economics, 659-679.
